drivers/net/ethernet/mellanox/mlx5/core/fs_core.c

   1 /*
   2  * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #include <linux/mutex.h>
  34 #include <linux/mlx5/driver.h>
  35 #include <linux/mlx5/vport.h>
  36 #include <linux/mlx5/eswitch.h>
  37 #include <net/devlink.h>
  38
  39 #include "mlx5_core.h"
  40 #include "fs_core.h"
  41 #include "fs_cmd.h"
  42 #include "fs_ft_pool.h"
  43 #include "diag/fs_tracepoint.h"
  44 #include "devlink.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 RDMA_RX_COUNTERS_PRIO_NUM_LEVELS 1
 103 #define RDMA_TX_COUNTERS_PRIO_NUM_LEVELS 1
 104
 105 #define BY_PASS_PRIO_NUM_LEVELS 1
 106 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
 107                            LEFTOVERS_NUM_PRIOS)
 108
 109 #define KERNEL_RX_MACSEC_NUM_PRIOS  1
 110 #define KERNEL_RX_MACSEC_NUM_LEVELS 2
 111 #define KERNEL_RX_MACSEC_MIN_LEVEL (BY_PASS_MIN_LEVEL + KERNEL_RX_MACSEC_NUM_PRIOS)
 112
 113 #define ETHTOOL_PRIO_NUM_LEVELS 1
 114 #define ETHTOOL_NUM_PRIOS 11
 115 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
 116 /* Promiscuous, Vlan, mac, ttc, inner ttc, {UDP/ANY/aRFS/accel/{esp, esp_err}}, IPsec policy,
 117  * IPsec RoCE policy
 118  */
 119 #define KERNEL_NIC_PRIO_NUM_LEVELS 9
 120 #define KERNEL_NIC_NUM_PRIOS 1
 121 /* One more level for tc */
 122 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
 123
 124 #define KERNEL_NIC_TC_NUM_PRIOS  1
 125 #define KERNEL_NIC_TC_NUM_LEVELS 3
 126
 127 #define ANCHOR_NUM_LEVELS 1
 128 #define ANCHOR_NUM_PRIOS 1
 129 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 130
 131 #define OFFLOADS_MAX_FT 2
 132 #define OFFLOADS_NUM_PRIOS 2
 133 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS)
 134
 135 #define LAG_PRIO_NUM_LEVELS 1
 136 #define LAG_NUM_PRIOS 1
 137 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + KERNEL_RX_MACSEC_MIN_LEVEL + 1)
 138
 139 #define KERNEL_TX_IPSEC_NUM_PRIOS  1
 140 #define KERNEL_TX_IPSEC_NUM_LEVELS 3
 141 #define KERNEL_TX_IPSEC_MIN_LEVEL        (KERNEL_TX_IPSEC_NUM_LEVELS)
 142
 143 #define KERNEL_TX_MACSEC_NUM_PRIOS  1
 144 #define KERNEL_TX_MACSEC_NUM_LEVELS 2
 145 #define KERNEL_TX_MACSEC_MIN_LEVEL       (KERNEL_TX_IPSEC_MIN_LEVEL + KERNEL_TX_MACSEC_NUM_PRIOS)
 146
 147 struct node_caps {
 148         size_t  arr_sz;
 149         long    *caps;
 150 };
 151
 152 static struct init_tree_node {
 153         enum fs_node_type       type;
 154         struct init_tree_node *children;
 155         int ar_size;
 156         struct node_caps caps;
 157         int min_ft_level;
 158         int num_leaf_prios;
 159         int prio;
 160         int num_levels;
 161         enum mlx5_flow_table_miss_action def_miss_action;
 162 } root_fs = {
 163         .type = FS_TYPE_NAMESPACE,
 164         .ar_size = 8,
 165           .children = (struct init_tree_node[]){
 166                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 167                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 168                                   ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 169                                                     BY_PASS_PRIO_NUM_LEVELS))),
 170                   ADD_PRIO(0, KERNEL_RX_MACSEC_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 171                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 172                                   ADD_MULTIPLE_PRIO(KERNEL_RX_MACSEC_NUM_PRIOS,
 173                                                     KERNEL_RX_MACSEC_NUM_LEVELS))),
 174                   ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 175                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 176                                   ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 177                                                     LAG_PRIO_NUM_LEVELS))),
 178                   ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 179                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 180                                   ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS,
 181                                                     OFFLOADS_MAX_FT))),
 182                   ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 183                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 184                                   ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 185                                                     ETHTOOL_PRIO_NUM_LEVELS))),
 186                   ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 187                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 188                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS,
 189                                                     KERNEL_NIC_TC_NUM_LEVELS),
 190                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 191                                                     KERNEL_NIC_PRIO_NUM_LEVELS))),
 192                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 193                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 194                                   ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS,
 195                                                     LEFTOVERS_NUM_LEVELS))),
 196                   ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 197                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 198                                   ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS,
 199                                                     ANCHOR_NUM_LEVELS))),
 200         }
 201 };
 202
 203 static struct init_tree_node egress_root_fs = {
 204         .type = FS_TYPE_NAMESPACE,
 205         .ar_size = 3,
 206         .children = (struct init_tree_node[]) {
 207                 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 208                          FS_CHAINING_CAPS_EGRESS,
 209                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 210                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 211                                                   BY_PASS_PRIO_NUM_LEVELS))),
 212                 ADD_PRIO(0, KERNEL_TX_IPSEC_MIN_LEVEL, 0,
 213                          FS_CHAINING_CAPS_EGRESS,
 214                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 215                                 ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS,
 216                                                   KERNEL_TX_IPSEC_NUM_LEVELS))),
 217                 ADD_PRIO(0, KERNEL_TX_MACSEC_MIN_LEVEL, 0,
 218                          FS_CHAINING_CAPS_EGRESS,
 219                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 220                                 ADD_MULTIPLE_PRIO(KERNEL_TX_MACSEC_NUM_PRIOS,
 221                                                   KERNEL_TX_MACSEC_NUM_LEVELS))),
 222         }
 223 };
 224
 225 enum {
 226         RDMA_RX_IPSEC_PRIO,
 227         RDMA_RX_COUNTERS_PRIO,
 228         RDMA_RX_BYPASS_PRIO,
 229         RDMA_RX_KERNEL_PRIO,
 230 };
 231
 232 #define RDMA_RX_IPSEC_NUM_PRIOS 1
 233 #define RDMA_RX_IPSEC_NUM_LEVELS 2
 234 #define RDMA_RX_IPSEC_MIN_LEVEL  (RDMA_RX_IPSEC_NUM_LEVELS)
 235
 236 #define RDMA_RX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_REGULAR_PRIOS
 237 #define RDMA_RX_KERNEL_MIN_LEVEL (RDMA_RX_BYPASS_MIN_LEVEL + 1)
 238 #define RDMA_RX_COUNTERS_MIN_LEVEL (RDMA_RX_KERNEL_MIN_LEVEL + 2)
 239
 240 static struct init_tree_node rdma_rx_root_fs = {
 241         .type = FS_TYPE_NAMESPACE,
 242         .ar_size = 4,
 243         .children = (struct init_tree_node[]) {
 244                 [RDMA_RX_IPSEC_PRIO] =
 245                 ADD_PRIO(0, RDMA_RX_IPSEC_MIN_LEVEL, 0,
 246                          FS_CHAINING_CAPS,
 247                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 248                                 ADD_MULTIPLE_PRIO(RDMA_RX_IPSEC_NUM_PRIOS,
 249                                                   RDMA_RX_IPSEC_NUM_LEVELS))),
 250                 [RDMA_RX_COUNTERS_PRIO] =
 251                 ADD_PRIO(0, RDMA_RX_COUNTERS_MIN_LEVEL, 0,
 252                          FS_CHAINING_CAPS,
 253                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 254                                 ADD_MULTIPLE_PRIO(MLX5_RDMA_RX_NUM_COUNTERS_PRIOS,
 255                                                   RDMA_RX_COUNTERS_PRIO_NUM_LEVELS))),
 256                 [RDMA_RX_BYPASS_PRIO] =
 257                 ADD_PRIO(0, RDMA_RX_BYPASS_MIN_LEVEL, 0,
 258                          FS_CHAINING_CAPS,
 259                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 260                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS,
 261                                                   BY_PASS_PRIO_NUM_LEVELS))),
 262                 [RDMA_RX_KERNEL_PRIO] =
 263                 ADD_PRIO(0, RDMA_RX_KERNEL_MIN_LEVEL, 0,
 264                          FS_CHAINING_CAPS,
 265                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN,
 266                                 ADD_MULTIPLE_PRIO(1, 1))),
 267         }
 268 };
 269
 270 enum {
 271         RDMA_TX_COUNTERS_PRIO,
 272         RDMA_TX_IPSEC_PRIO,
 273         RDMA_TX_BYPASS_PRIO,
 274 };
 275
 276 #define RDMA_TX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_PRIOS
 277 #define RDMA_TX_COUNTERS_MIN_LEVEL (RDMA_TX_BYPASS_MIN_LEVEL + 1)
 278
 279 #define RDMA_TX_IPSEC_NUM_PRIOS 1
 280 #define RDMA_TX_IPSEC_PRIO_NUM_LEVELS 1
 281 #define RDMA_TX_IPSEC_MIN_LEVEL  (RDMA_TX_COUNTERS_MIN_LEVEL + RDMA_TX_IPSEC_NUM_PRIOS)
 282
 283 static struct init_tree_node rdma_tx_root_fs = {
 284         .type = FS_TYPE_NAMESPACE,
 285         .ar_size = 3,
 286         .children = (struct init_tree_node[]) {
 287                 [RDMA_TX_COUNTERS_PRIO] =
 288                 ADD_PRIO(0, RDMA_TX_COUNTERS_MIN_LEVEL, 0,
 289                          FS_CHAINING_CAPS,
 290                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 291                                 ADD_MULTIPLE_PRIO(MLX5_RDMA_TX_NUM_COUNTERS_PRIOS,
 292                                                   RDMA_TX_COUNTERS_PRIO_NUM_LEVELS))),
 293                 [RDMA_TX_IPSEC_PRIO] =
 294                 ADD_PRIO(0, RDMA_TX_IPSEC_MIN_LEVEL, 0,
 295                          FS_CHAINING_CAPS,
 296                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 297                                 ADD_MULTIPLE_PRIO(RDMA_TX_IPSEC_NUM_PRIOS,
 298                                                   RDMA_TX_IPSEC_PRIO_NUM_LEVELS))),
 299
 300                 [RDMA_TX_BYPASS_PRIO] =
 301                 ADD_PRIO(0, RDMA_TX_BYPASS_MIN_LEVEL, 0,
 302                          FS_CHAINING_CAPS_RDMA_TX,
 303                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 304                                 ADD_MULTIPLE_PRIO(RDMA_TX_BYPASS_MIN_LEVEL,
 305                                                   BY_PASS_PRIO_NUM_LEVELS))),
 306         }
 307 };
 308
 309 enum fs_i_lock_class {
 310         FS_LOCK_GRANDPARENT,
 311         FS_LOCK_PARENT,
 312         FS_LOCK_CHILD
 313 };
 314
 315 static const struct rhashtable_params rhash_fte = {
 316         .key_len = sizeof_field(struct fs_fte, val),
 317         .key_offset = offsetof(struct fs_fte, val),
 318         .head_offset = offsetof(struct fs_fte, hash),
 319         .automatic_shrinking = true,
 320         .min_size = 1,
 321 };
 322
 323 static const struct rhashtable_params rhash_fg = {
 324         .key_len = sizeof_field(struct mlx5_flow_group, mask),
 325         .key_offset = offsetof(struct mlx5_flow_group, mask),
 326         .head_offset = offsetof(struct mlx5_flow_group, hash),
 327         .automatic_shrinking = true,
 328         .min_size = 1,
 329
 330 };
 331
 332 static void del_hw_flow_table(struct fs_node *node);
 333 static void del_hw_flow_group(struct fs_node *node);
 334 static void del_hw_fte(struct fs_node *node);
 335 static void del_sw_flow_table(struct fs_node *node);
 336 static void del_sw_flow_group(struct fs_node *node);
 337 static void del_sw_fte(struct fs_node *node);
 338 static void del_sw_prio(struct fs_node *node);
 339 static void del_sw_ns(struct fs_node *node);
 340 /* Delete rule (destination) is special case that
 341  * requires to lock the FTE for all the deletion process.
 342  */
 343 static void del_sw_hw_rule(struct fs_node *node);
 344 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 345                                 struct mlx5_flow_destination *d2);
 346 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 347 static struct mlx5_flow_rule *
 348 find_flow_rule(struct fs_fte *fte,
 349                struct mlx5_flow_destination *dest);
 350
 351 static void tree_init_node(struct fs_node *node,
 352                            void (*del_hw_func)(struct fs_node *),
 353                            void (*del_sw_func)(struct fs_node *))
 354 {
 355         refcount_set(&node->refcount, 1);
 356         INIT_LIST_HEAD(&node->list);
 357         INIT_LIST_HEAD(&node->children);
 358         init_rwsem(&node->lock);
 359         node->del_hw_func = del_hw_func;
 360         node->del_sw_func = del_sw_func;
 361         node->active = false;
 362 }
 363
 364 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 365 {
 366         if (parent)
 367                 refcount_inc(&parent->refcount);
 368         node->parent = parent;
 369
 370         /* Parent is the root */
 371         if (!parent)
 372                 node->root = node;
 373         else
 374                 node->root = parent->root;
 375 }
 376
 377 static int tree_get_node(struct fs_node *node)
 378 {
 379         return refcount_inc_not_zero(&node->refcount);
 380 }
 381
 382 static void nested_down_read_ref_node(struct fs_node *node,
 383                                       enum fs_i_lock_class class)
 384 {
 385         if (node) {
 386                 down_read_nested(&node->lock, class);
 387                 refcount_inc(&node->refcount);
 388         }
 389 }
 390
 391 static void nested_down_write_ref_node(struct fs_node *node,
 392                                        enum fs_i_lock_class class)
 393 {
 394         if (node) {
 395                 down_write_nested(&node->lock, class);
 396                 refcount_inc(&node->refcount);
 397         }
 398 }
 399
 400 static void down_write_ref_node(struct fs_node *node, bool locked)
 401 {
 402         if (node) {
 403                 if (!locked)
 404                         down_write(&node->lock);
 405                 refcount_inc(&node->refcount);
 406         }
 407 }
 408
 409 static void up_read_ref_node(struct fs_node *node)
 410 {
 411         refcount_dec(&node->refcount);
 412         up_read(&node->lock);
 413 }
 414
 415 static void up_write_ref_node(struct fs_node *node, bool locked)
 416 {
 417         refcount_dec(&node->refcount);
 418         if (!locked)
 419                 up_write(&node->lock);
 420 }
 421
 422 static void tree_put_node(struct fs_node *node, bool locked)
 423 {
 424         struct fs_node *parent_node = node->parent;
 425
 426         if (refcount_dec_and_test(&node->refcount)) {
 427                 if (node->del_hw_func)
 428                         node->del_hw_func(node);
 429                 if (parent_node) {
 430                         down_write_ref_node(parent_node, locked);
 431                         list_del_init(&node->list);
 432                 }
 433                 node->del_sw_func(node);
 434                 if (parent_node)
 435                         up_write_ref_node(parent_node, locked);
 436                 node = NULL;
 437         }
 438         if (!node && parent_node)
 439                 tree_put_node(parent_node, locked);
 440 }
 441
 442 static int tree_remove_node(struct fs_node *node, bool locked)
 443 {
 444         if (refcount_read(&node->refcount) > 1) {
 445                 refcount_dec(&node->refcount);
 446                 return -EEXIST;
 447         }
 448         tree_put_node(node, locked);
 449         return 0;
 450 }
 451
 452 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 453                                  unsigned int prio)
 454 {
 455         struct fs_prio *iter_prio;
 456
 457         fs_for_each_prio(iter_prio, ns) {
 458                 if (iter_prio->prio == prio)
 459                         return iter_prio;
 460         }
 461
 462         return NULL;
 463 }
 464
 465 static bool is_fwd_next_action(u32 action)
 466 {
 467         return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
 468                          MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
 469 }
 470
 471 static bool is_fwd_dest_type(enum mlx5_flow_destination_type type)
 472 {
 473         return type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM ||
 474                 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE ||
 475                 type == MLX5_FLOW_DESTINATION_TYPE_UPLINK ||
 476                 type == MLX5_FLOW_DESTINATION_TYPE_VPORT ||
 477                 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER ||
 478                 type == MLX5_FLOW_DESTINATION_TYPE_TIR ||
 479                 type == MLX5_FLOW_DESTINATION_TYPE_RANGE ||
 480                 type == MLX5_FLOW_DESTINATION_TYPE_TABLE_TYPE;
 481 }
 482
 483 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 484 {
 485         int i;
 486
 487         for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 488                 if (spec->match_value[i] & ~spec->match_criteria[i]) {
 489                         pr_warn("mlx5_core: match_value differs from match_criteria\n");
 490                         return false;
 491                 }
 492
 493         return true;
 494 }
 495
 496 struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 497 {
 498         struct fs_node *root;
 499         struct mlx5_flow_namespace *ns;
 500
 501         root = node->root;
 502
 503         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 504                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 505                 return NULL;
 506         }
 507
 508         ns = container_of(root, struct mlx5_flow_namespace, node);
 509         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 510 }
 511
 512 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 513 {
 514         struct mlx5_flow_root_namespace *root = find_root(node);
 515
 516         if (root)
 517                 return root->dev->priv.steering;
 518         return NULL;
 519 }
 520
 521 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 522 {
 523         struct mlx5_flow_root_namespace *root = find_root(node);
 524
 525         if (root)
 526                 return root->dev;
 527         return NULL;
 528 }
 529
 530 static void del_sw_ns(struct fs_node *node)
 531 {
 532         kfree(node);
 533 }
 534
 535 static void del_sw_prio(struct fs_node *node)
 536 {
 537         kfree(node);
 538 }
 539
 540 static void del_hw_flow_table(struct fs_node *node)
 541 {
 542         struct mlx5_flow_root_namespace *root;
 543         struct mlx5_flow_table *ft;
 544         struct mlx5_core_dev *dev;
 545         int err;
 546
 547         fs_get_obj(ft, node);
 548         dev = get_dev(&ft->node);
 549         root = find_root(&ft->node);
 550         trace_mlx5_fs_del_ft(ft);
 551
 552         if (node->active) {
 553                 err = root->cmds->destroy_flow_table(root, ft);
 554                 if (err)
 555                         mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 556         }
 557 }
 558
 559 static void del_sw_flow_table(struct fs_node *node)
 560 {
 561         struct mlx5_flow_table *ft;
 562         struct fs_prio *prio;
 563
 564         fs_get_obj(ft, node);
 565
 566         rhltable_destroy(&ft->fgs_hash);
 567         if (ft->node.parent) {
 568                 fs_get_obj(prio, ft->node.parent);
 569                 prio->num_ft--;
 570         }
 571         kfree(ft);
 572 }
 573
 574 static void modify_fte(struct fs_fte *fte)
 575 {
 576         struct mlx5_flow_root_namespace *root;
 577         struct mlx5_flow_table *ft;
 578         struct mlx5_flow_group *fg;
 579         struct mlx5_core_dev *dev;
 580         int err;
 581
 582         fs_get_obj(fg, fte->node.parent);
 583         fs_get_obj(ft, fg->node.parent);
 584         dev = get_dev(&fte->node);
 585
 586         root = find_root(&ft->node);
 587         err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte);
 588         if (err)
 589                 mlx5_core_warn(dev,
 590                                "%s can't del rule fg id=%d fte_index=%d\n",
 591                                __func__, fg->id, fte->index);
 592         fte->modify_mask = 0;
 593 }
 594
 595 static void del_sw_hw_rule(struct fs_node *node)
 596 {
 597         struct mlx5_flow_rule *rule;
 598         struct fs_fte *fte;
 599
 600         fs_get_obj(rule, node);
 601         fs_get_obj(fte, rule->node.parent);
 602         trace_mlx5_fs_del_rule(rule);
 603         if (is_fwd_next_action(rule->sw_action)) {
 604                 mutex_lock(&rule->dest_attr.ft->lock);
 605                 list_del(&rule->next_ft);
 606                 mutex_unlock(&rule->dest_attr.ft->lock);
 607         }
 608
 609         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER) {
 610                 --fte->dests_size;
 611                 fte->modify_mask |=
 612                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 613                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 614                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 615                 goto out;
 616         }
 617
 618         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT) {
 619                 --fte->dests_size;
 620                 fte->modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
 621                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW;
 622                 goto out;
 623         }
 624
 625         if (is_fwd_dest_type(rule->dest_attr.type)) {
 626                 --fte->dests_size;
 627                 --fte->fwd_dests;
 628
 629                 if (!fte->fwd_dests)
 630                         fte->action.action &=
 631                                 ~MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
 632                 fte->modify_mask |=
 633                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 634                 goto out;
 635         }
 636 out:
 637         kfree(rule);
 638 }
 639
 640 static void del_hw_fte(struct fs_node *node)
 641 {
 642         struct mlx5_flow_root_namespace *root;
 643         struct mlx5_flow_table *ft;
 644         struct mlx5_flow_group *fg;
 645         struct mlx5_core_dev *dev;
 646         struct fs_fte *fte;
 647         int err;
 648
 649         fs_get_obj(fte, node);
 650         fs_get_obj(fg, fte->node.parent);
 651         fs_get_obj(ft, fg->node.parent);
 652
 653         trace_mlx5_fs_del_fte(fte);
 654         WARN_ON(fte->dests_size);
 655         dev = get_dev(&ft->node);
 656         root = find_root(&ft->node);
 657         if (node->active) {
 658                 err = root->cmds->delete_fte(root, ft, fte);
 659                 if (err)
 660                         mlx5_core_warn(dev,
 661                                        "flow steering can't delete fte in index %d of flow group id %d\n",
 662                                        fte->index, fg->id);
 663                 node->active = false;
 664         }
 665 }
 666
 667 static void del_sw_fte(struct fs_node *node)
 668 {
 669         struct mlx5_flow_steering *steering = get_steering(node);
 670         struct mlx5_flow_group *fg;
 671         struct fs_fte *fte;
 672         int err;
 673
 674         fs_get_obj(fte, node);
 675         fs_get_obj(fg, fte->node.parent);
 676
 677         err = rhashtable_remove_fast(&fg->ftes_hash,
 678                                      &fte->hash,
 679                                      rhash_fte);
 680         WARN_ON(err);
 681         ida_free(&fg->fte_allocator, fte->index - fg->start_index);
 682         kmem_cache_free(steering->ftes_cache, fte);
 683 }
 684
 685 static void del_hw_flow_group(struct fs_node *node)
 686 {
 687         struct mlx5_flow_root_namespace *root;
 688         struct mlx5_flow_group *fg;
 689         struct mlx5_flow_table *ft;
 690         struct mlx5_core_dev *dev;
 691
 692         fs_get_obj(fg, node);
 693         fs_get_obj(ft, fg->node.parent);
 694         dev = get_dev(&ft->node);
 695         trace_mlx5_fs_del_fg(fg);
 696
 697         root = find_root(&ft->node);
 698         if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
 699                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 700                                fg->id, ft->id);
 701 }
 702
 703 static void del_sw_flow_group(struct fs_node *node)
 704 {
 705         struct mlx5_flow_steering *steering = get_steering(node);
 706         struct mlx5_flow_group *fg;
 707         struct mlx5_flow_table *ft;
 708         int err;
 709
 710         fs_get_obj(fg, node);
 711         fs_get_obj(ft, fg->node.parent);
 712
 713         rhashtable_destroy(&fg->ftes_hash);
 714         ida_destroy(&fg->fte_allocator);
 715         if (ft->autogroup.active &&
 716             fg->max_ftes == ft->autogroup.group_size &&
 717             fg->start_index < ft->autogroup.max_fte)
 718                 ft->autogroup.num_groups--;
 719         err = rhltable_remove(&ft->fgs_hash,
 720                               &fg->hash,
 721                               rhash_fg);
 722         WARN_ON(err);
 723         kmem_cache_free(steering->fgs_cache, fg);
 724 }
 725
 726 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 727 {
 728         int index;
 729         int ret;
 730
 731         index = ida_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, GFP_KERNEL);
 732         if (index < 0)
 733                 return index;
 734
 735         fte->index = index + fg->start_index;
 736         ret = rhashtable_insert_fast(&fg->ftes_hash,
 737                                      &fte->hash,
 738                                      rhash_fte);
 739         if (ret)
 740                 goto err_ida_remove;
 741
 742         tree_add_node(&fte->node, &fg->node);
 743         list_add_tail(&fte->node.list, &fg->node.children);
 744         return 0;
 745
 746 err_ida_remove:
 747         ida_free(&fg->fte_allocator, index);
 748         return ret;
 749 }
 750
 751 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 752                                 const struct mlx5_flow_spec *spec,
 753                                 struct mlx5_flow_act *flow_act)
 754 {
 755         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 756         struct fs_fte *fte;
 757
 758         fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 759         if (!fte)
 760                 return ERR_PTR(-ENOMEM);
 761
 762         memcpy(fte->val, &spec->match_value, sizeof(fte->val));
 763         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 764         fte->action = *flow_act;
 765         fte->flow_context = spec->flow_context;
 766
 767         tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
 768
 769         return fte;
 770 }
 771
 772 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 773                                struct mlx5_flow_group *fg)
 774 {
 775         rhashtable_destroy(&fg->ftes_hash);
 776         kmem_cache_free(steering->fgs_cache, fg);
 777 }
 778
 779 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 780                                                 u8 match_criteria_enable,
 781                                                 const void *match_criteria,
 782                                                 int start_index,
 783                                                 int end_index)
 784 {
 785         struct mlx5_flow_group *fg;
 786         int ret;
 787
 788         fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 789         if (!fg)
 790                 return ERR_PTR(-ENOMEM);
 791
 792         ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 793         if (ret) {
 794                 kmem_cache_free(steering->fgs_cache, fg);
 795                 return ERR_PTR(ret);
 796         }
 797
 798         ida_init(&fg->fte_allocator);
 799         fg->mask.match_criteria_enable = match_criteria_enable;
 800         memcpy(&fg->mask.match_criteria, match_criteria,
 801                sizeof(fg->mask.match_criteria));
 802         fg->node.type =  FS_TYPE_FLOW_GROUP;
 803         fg->start_index = start_index;
 804         fg->max_ftes = end_index - start_index + 1;
 805
 806         return fg;
 807 }
 808
 809 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 810                                                        u8 match_criteria_enable,
 811                                                        const void *match_criteria,
 812                                                        int start_index,
 813                                                        int end_index,
 814                                                        struct list_head *prev)
 815 {
 816         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 817         struct mlx5_flow_group *fg;
 818         int ret;
 819
 820         fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 821                               start_index, end_index);
 822         if (IS_ERR(fg))
 823                 return fg;
 824
 825         /* initialize refcnt, add to parent list */
 826         ret = rhltable_insert(&ft->fgs_hash,
 827                               &fg->hash,
 828                               rhash_fg);
 829         if (ret) {
 830                 dealloc_flow_group(steering, fg);
 831                 return ERR_PTR(ret);
 832         }
 833
 834         tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 835         tree_add_node(&fg->node, &ft->node);
 836         /* Add node to group list */
 837         list_add(&fg->node.list, prev);
 838         atomic_inc(&ft->node.version);
 839
 840         return fg;
 841 }
 842
 843 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport,
 844                                                 enum fs_flow_table_type table_type,
 845                                                 enum fs_flow_table_op_mod op_mod,
 846                                                 u32 flags)
 847 {
 848         struct mlx5_flow_table *ft;
 849         int ret;
 850
 851         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 852         if (!ft)
 853                 return ERR_PTR(-ENOMEM);
 854
 855         ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 856         if (ret) {
 857                 kfree(ft);
 858                 return ERR_PTR(ret);
 859         }
 860
 861         ft->level = level;
 862         ft->node.type = FS_TYPE_FLOW_TABLE;
 863         ft->op_mod = op_mod;
 864         ft->type = table_type;
 865         ft->vport = vport;
 866         ft->flags = flags;
 867         INIT_LIST_HEAD(&ft->fwd_rules);
 868         mutex_init(&ft->lock);
 869
 870         return ft;
 871 }
 872
 873 /* If reverse is false, then we search for the first flow table in the
 874  * root sub-tree from start(closest from right), else we search for the
 875  * last flow table in the root sub-tree till start(closest from left).
 876  */
 877 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 878                                                          struct list_head *start,
 879                                                          bool reverse)
 880 {
 881 #define list_advance_entry(pos, reverse)                \
 882         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 883
 884 #define list_for_each_advance_continue(pos, head, reverse)      \
 885         for (pos = list_advance_entry(pos, reverse);            \
 886              &pos->list != (head);                              \
 887              pos = list_advance_entry(pos, reverse))
 888
 889         struct fs_node *iter = list_entry(start, struct fs_node, list);
 890         struct mlx5_flow_table *ft = NULL;
 891
 892         if (!root)
 893                 return NULL;
 894
 895         list_for_each_advance_continue(iter, &root->children, reverse) {
 896                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 897                         fs_get_obj(ft, iter);
 898                         return ft;
 899                 }
 900                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 901                 if (ft)
 902                         return ft;
 903         }
 904
 905         return ft;
 906 }
 907
 908 static struct fs_node *find_prio_chains_parent(struct fs_node *parent,
 909                                                struct fs_node **child)
 910 {
 911         struct fs_node *node = NULL;
 912
 913         while (parent && parent->type != FS_TYPE_PRIO_CHAINS) {
 914                 node = parent;
 915                 parent = parent->parent;
 916         }
 917
 918         if (child)
 919                 *child = node;
 920
 921         return parent;
 922 }
 923
 924 /* If reverse is false then return the first flow table next to the passed node
 925  * in the tree, else return the last flow table before the node in the tree.
 926  * If skip is true, skip the flow tables in the same prio_chains prio.
 927  */
 928 static struct mlx5_flow_table *find_closest_ft(struct fs_node *node, bool reverse,
 929                                                bool skip)
 930 {
 931         struct fs_node *prio_chains_parent = NULL;
 932         struct mlx5_flow_table *ft = NULL;
 933         struct fs_node *curr_node;
 934         struct fs_node *parent;
 935
 936         if (skip)
 937                 prio_chains_parent = find_prio_chains_parent(node, NULL);
 938         parent = node->parent;
 939         curr_node = node;
 940         while (!ft && parent) {
 941                 if (parent != prio_chains_parent)
 942                         ft = find_closest_ft_recursive(parent, &curr_node->list,
 943                                                        reverse);
 944                 curr_node = parent;
 945                 parent = curr_node->parent;
 946         }
 947         return ft;
 948 }
 949
 950 /* Assuming all the tree is locked by mutex chain lock */
 951 static struct mlx5_flow_table *find_next_chained_ft(struct fs_node *node)
 952 {
 953         return find_closest_ft(node, false, true);
 954 }
 955
 956 /* Assuming all the tree is locked by mutex chain lock */
 957 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_node *node)
 958 {
 959         return find_closest_ft(node, true, true);
 960 }
 961
 962 static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft,
 963                                                 struct mlx5_flow_act *flow_act)
 964 {
 965         struct fs_prio *prio;
 966         bool next_ns;
 967
 968         next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
 969         fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent);
 970
 971         return find_next_chained_ft(&prio->node);
 972 }
 973
 974 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 975                                struct fs_prio *prio,
 976                                struct mlx5_flow_table *ft)
 977 {
 978         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 979         struct mlx5_flow_table *iter;
 980         int err;
 981
 982         fs_for_each_ft(iter, prio) {
 983                 err = root->cmds->modify_flow_table(root, iter, ft);
 984                 if (err) {
 985                         mlx5_core_err(dev,
 986                                       "Failed to modify flow table id %d, type %d, err %d\n",
 987                                       iter->id, iter->type, err);
 988                         /* The driver is out of sync with the FW */
 989                         return err;
 990                 }
 991         }
 992         return 0;
 993 }
 994
 995 static struct mlx5_flow_table *find_closet_ft_prio_chains(struct fs_node *node,
 996                                                           struct fs_node *parent,
 997                                                           struct fs_node **child,
 998                                                           bool reverse)
 999 {
1000         struct mlx5_flow_table *ft;
1001
1002         ft = find_closest_ft(node, reverse, false);
1003
1004         if (ft && parent == find_prio_chains_parent(&ft->node, child))
1005                 return ft;
1006
1007         return NULL;
1008 }
1009
1010 /* Connect flow tables from previous priority of prio to ft */
1011 static int connect_prev_fts(struct mlx5_core_dev *dev,
1012                             struct mlx5_flow_table *ft,
1013                             struct fs_prio *prio)
1014 {
1015         struct fs_node *prio_parent, *parent = NULL, *child, *node;
1016         struct mlx5_flow_table *prev_ft;
1017         int err = 0;
1018
1019         prio_parent = find_prio_chains_parent(&prio->node, &child);
1020
1021         /* return directly if not under the first sub ns of prio_chains prio */
1022         if (prio_parent && !list_is_first(&child->list, &prio_parent->children))
1023                 return 0;
1024
1025         prev_ft = find_prev_chained_ft(&prio->node);
1026         while (prev_ft) {
1027                 struct fs_prio *prev_prio;
1028
1029                 fs_get_obj(prev_prio, prev_ft->node.parent);
1030                 err = connect_fts_in_prio(dev, prev_prio, ft);
1031                 if (err)
1032                         break;
1033
1034                 if (!parent) {
1035                         parent = find_prio_chains_parent(&prev_prio->node, &child);
1036                         if (!parent)
1037                                 break;
1038                 }
1039
1040                 node = child;
1041                 prev_ft = find_closet_ft_prio_chains(node, parent, &child, true);
1042         }
1043         return err;
1044 }
1045
1046 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
1047                                  *prio)
1048 {
1049         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
1050         struct mlx5_ft_underlay_qp *uqp;
1051         int min_level = INT_MAX;
1052         int err = 0;
1053         u32 qpn;
1054
1055         if (root->root_ft)
1056                 min_level = root->root_ft->level;
1057
1058         if (ft->level >= min_level)
1059                 return 0;
1060
1061         if (list_empty(&root->underlay_qpns)) {
1062                 /* Don't set any QPN (zero) in case QPN list is empty */
1063                 qpn = 0;
1064                 err = root->cmds->update_root_ft(root, ft, qpn, false);
1065         } else {
1066                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
1067                         qpn = uqp->qpn;
1068                         err = root->cmds->update_root_ft(root, ft,
1069                                                          qpn, false);
1070                         if (err)
1071                                 break;
1072                 }
1073         }
1074
1075         if (err)
1076                 mlx5_core_warn(root->dev,
1077                                "Update root flow table of id(%u) qpn(%d) failed\n",
1078                                ft->id, qpn);
1079         else
1080                 root->root_ft = ft;
1081
1082         return err;
1083 }
1084
1085 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
1086                                          struct mlx5_flow_destination *dest)
1087 {
1088         struct mlx5_flow_root_namespace *root;
1089         struct mlx5_flow_table *ft;
1090         struct mlx5_flow_group *fg;
1091         struct fs_fte *fte;
1092         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1093         int err = 0;
1094
1095         fs_get_obj(fte, rule->node.parent);
1096         if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1097                 return -EINVAL;
1098         down_write_ref_node(&fte->node, false);
1099         fs_get_obj(fg, fte->node.parent);
1100         fs_get_obj(ft, fg->node.parent);
1101
1102         memcpy(&rule->dest_attr, dest, sizeof(*dest));
1103         root = find_root(&ft->node);
1104         err = root->cmds->update_fte(root, ft, fg,
1105                                      modify_mask, fte);
1106         up_write_ref_node(&fte->node, false);
1107
1108         return err;
1109 }
1110
1111 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
1112                                  struct mlx5_flow_destination *new_dest,
1113                                  struct mlx5_flow_destination *old_dest)
1114 {
1115         int i;
1116
1117         if (!old_dest) {
1118                 if (handle->num_rules != 1)
1119                         return -EINVAL;
1120                 return _mlx5_modify_rule_destination(handle->rule[0],
1121                                                      new_dest);
1122         }
1123
1124         for (i = 0; i < handle->num_rules; i++) {
1125                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
1126                         return _mlx5_modify_rule_destination(handle->rule[i],
1127                                                              new_dest);
1128         }
1129
1130         return -EINVAL;
1131 }
1132
1133 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
1134 static int connect_fwd_rules(struct mlx5_core_dev *dev,
1135                              struct mlx5_flow_table *new_next_ft,
1136                              struct mlx5_flow_table *old_next_ft)
1137 {
1138         struct mlx5_flow_destination dest = {};
1139         struct mlx5_flow_rule *iter;
1140         int err = 0;
1141
1142         /* new_next_ft and old_next_ft could be NULL only
1143          * when we create/destroy the anchor flow table.
1144          */
1145         if (!new_next_ft || !old_next_ft)
1146                 return 0;
1147
1148         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1149         dest.ft = new_next_ft;
1150
1151         mutex_lock(&old_next_ft->lock);
1152         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
1153         mutex_unlock(&old_next_ft->lock);
1154         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
1155                 if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) &&
1156                     iter->ft->ns == new_next_ft->ns)
1157                         continue;
1158
1159                 err = _mlx5_modify_rule_destination(iter, &dest);
1160                 if (err)
1161                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
1162                                new_next_ft->id);
1163         }
1164         return 0;
1165 }
1166
1167 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
1168                               struct fs_prio *prio)
1169 {
1170         struct mlx5_flow_table *next_ft, *first_ft;
1171         int err = 0;
1172
1173         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
1174
1175         first_ft = list_first_entry_or_null(&prio->node.children,
1176                                             struct mlx5_flow_table, node.list);
1177         if (!first_ft || first_ft->level > ft->level) {
1178                 err = connect_prev_fts(dev, ft, prio);
1179                 if (err)
1180                         return err;
1181
1182                 next_ft = first_ft ? first_ft : find_next_chained_ft(&prio->node);
1183                 err = connect_fwd_rules(dev, ft, next_ft);
1184                 if (err)
1185                         return err;
1186         }
1187
1188         if (MLX5_CAP_FLOWTABLE(dev,
1189                                flow_table_properties_nic_receive.modify_root))
1190                 err = update_root_ft_create(ft, prio);
1191         return err;
1192 }
1193
1194 static void list_add_flow_table(struct mlx5_flow_table *ft,
1195                                 struct fs_prio *prio)
1196 {
1197         struct list_head *prev = &prio->node.children;
1198         struct mlx5_flow_table *iter;
1199
1200         fs_for_each_ft(iter, prio) {
1201                 if (iter->level > ft->level)
1202                         break;
1203                 prev = &iter->node.list;
1204         }
1205         list_add(&ft->node.list, prev);
1206 }
1207
1208 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1209                                                         struct mlx5_flow_table_attr *ft_attr,
1210                                                         enum fs_flow_table_op_mod op_mod,
1211                                                         u16 vport)
1212 {
1213         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1214         bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED;
1215         struct mlx5_flow_table *next_ft;
1216         struct fs_prio *fs_prio = NULL;
1217         struct mlx5_flow_table *ft;
1218         int err;
1219
1220         if (!root) {
1221                 pr_err("mlx5: flow steering failed to find root of namespace\n");
1222                 return ERR_PTR(-ENODEV);
1223         }
1224
1225         mutex_lock(&root->chain_lock);
1226         fs_prio = find_prio(ns, ft_attr->prio);
1227         if (!fs_prio) {
1228                 err = -EINVAL;
1229                 goto unlock_root;
1230         }
1231         if (!unmanaged) {
1232                 /* The level is related to the
1233                  * priority level range.
1234                  */
1235                 if (ft_attr->level >= fs_prio->num_levels) {
1236                         err = -ENOSPC;
1237                         goto unlock_root;
1238                 }
1239
1240                 ft_attr->level += fs_prio->start_level;
1241         }
1242
1243         /* The level is related to the
1244          * priority level range.
1245          */
1246         ft = alloc_flow_table(ft_attr->level,
1247                               vport,
1248                               root->table_type,
1249                               op_mod, ft_attr->flags);
1250         if (IS_ERR(ft)) {
1251                 err = PTR_ERR(ft);
1252                 goto unlock_root;
1253         }
1254
1255         tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1256         next_ft = unmanaged ? ft_attr->next_ft :
1257                               find_next_chained_ft(&fs_prio->node);
1258         ft->def_miss_action = ns->def_miss_action;
1259         ft->ns = ns;
1260         err = root->cmds->create_flow_table(root, ft, ft_attr, next_ft);
1261         if (err)
1262                 goto free_ft;
1263
1264         if (!unmanaged) {
1265                 err = connect_flow_table(root->dev, ft, fs_prio);
1266                 if (err)
1267                         goto destroy_ft;
1268         }
1269
1270         ft->node.active = true;
1271         down_write_ref_node(&fs_prio->node, false);
1272         if (!unmanaged) {
1273                 tree_add_node(&ft->node, &fs_prio->node);
1274                 list_add_flow_table(ft, fs_prio);
1275         } else {
1276                 ft->node.root = fs_prio->node.root;
1277         }
1278         fs_prio->num_ft++;
1279         up_write_ref_node(&fs_prio->node, false);
1280         mutex_unlock(&root->chain_lock);
1281         trace_mlx5_fs_add_ft(ft);
1282         return ft;
1283 destroy_ft:
1284         root->cmds->destroy_flow_table(root, ft);
1285 free_ft:
1286         rhltable_destroy(&ft->fgs_hash);
1287         kfree(ft);
1288 unlock_root:
1289         mutex_unlock(&root->chain_lock);
1290         return ERR_PTR(err);
1291 }
1292
1293 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1294                                                struct mlx5_flow_table_attr *ft_attr)
1295 {
1296         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1297 }
1298 EXPORT_SYMBOL(mlx5_create_flow_table);
1299
1300 u32 mlx5_flow_table_id(struct mlx5_flow_table *ft)
1301 {
1302         return ft->id;
1303 }
1304 EXPORT_SYMBOL(mlx5_flow_table_id);
1305
1306 struct mlx5_flow_table *
1307 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1308                              struct mlx5_flow_table_attr *ft_attr, u16 vport)
1309 {
1310         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1311 }
1312
1313 struct mlx5_flow_table*
1314 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1315                                  int prio, u32 level)
1316 {
1317         struct mlx5_flow_table_attr ft_attr = {};
1318
1319         ft_attr.level = level;
1320         ft_attr.prio  = prio;
1321         ft_attr.max_fte = 1;
1322
1323         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1324 }
1325 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1326
1327 #define MAX_FLOW_GROUP_SIZE BIT(24)
1328 struct mlx5_flow_table*
1329 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1330                                     struct mlx5_flow_table_attr *ft_attr)
1331 {
1332         int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1333         int max_num_groups = ft_attr->autogroup.max_num_groups;
1334         struct mlx5_flow_table *ft;
1335         int autogroups_max_fte;
1336
1337         ft = mlx5_create_flow_table(ns, ft_attr);
1338         if (IS_ERR(ft))
1339                 return ft;
1340
1341         autogroups_max_fte = ft->max_fte - num_reserved_entries;
1342         if (max_num_groups > autogroups_max_fte)
1343                 goto err_validate;
1344         if (num_reserved_entries > ft->max_fte)
1345                 goto err_validate;
1346
1347         /* Align the number of groups according to the largest group size */
1348         if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE)
1349                 max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1;
1350
1351         ft->autogroup.active = true;
1352         ft->autogroup.required_groups = max_num_groups;
1353         ft->autogroup.max_fte = autogroups_max_fte;
1354         /* We save place for flow groups in addition to max types */
1355         ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1356
1357         return ft;
1358
1359 err_validate:
1360         mlx5_destroy_flow_table(ft);
1361         return ERR_PTR(-ENOSPC);
1362 }
1363 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1364
1365 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1366                                                u32 *fg_in)
1367 {
1368         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1369         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1370                                             fg_in, match_criteria);
1371         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1372                                             fg_in,
1373                                             match_criteria_enable);
1374         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1375                                    start_flow_index);
1376         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1377                                  end_flow_index);
1378         struct mlx5_flow_group *fg;
1379         int err;
1380
1381         if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1382                 return ERR_PTR(-EPERM);
1383
1384         down_write_ref_node(&ft->node, false);
1385         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1386                                      start_index, end_index,
1387                                      ft->node.children.prev);
1388         up_write_ref_node(&ft->node, false);
1389         if (IS_ERR(fg))
1390                 return fg;
1391
1392         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1393         if (err) {
1394                 tree_put_node(&fg->node, false);
1395                 return ERR_PTR(err);
1396         }
1397         trace_mlx5_fs_add_fg(fg);
1398         fg->node.active = true;
1399
1400         return fg;
1401 }
1402 EXPORT_SYMBOL(mlx5_create_flow_group);
1403
1404 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1405 {
1406         struct mlx5_flow_rule *rule;
1407
1408         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1409         if (!rule)
1410                 return NULL;
1411
1412         INIT_LIST_HEAD(&rule->next_ft);
1413         rule->node.type = FS_TYPE_FLOW_DEST;
1414         if (dest)
1415                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1416         else
1417                 rule->dest_attr.type = MLX5_FLOW_DESTINATION_TYPE_NONE;
1418
1419         return rule;
1420 }
1421
1422 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1423 {
1424         struct mlx5_flow_handle *handle;
1425
1426         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1427         if (!handle)
1428                 return NULL;
1429
1430         handle->num_rules = num_rules;
1431
1432         return handle;
1433 }
1434
1435 static void destroy_flow_handle(struct fs_fte *fte,
1436                                 struct mlx5_flow_handle *handle,
1437                                 struct mlx5_flow_destination *dest,
1438                                 int i)
1439 {
1440         for (; --i >= 0;) {
1441                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1442                         fte->dests_size--;
1443                         list_del(&handle->rule[i]->node.list);
1444                         kfree(handle->rule[i]);
1445                 }
1446         }
1447         kfree(handle);
1448 }
1449
1450 static struct mlx5_flow_handle *
1451 create_flow_handle(struct fs_fte *fte,
1452                    struct mlx5_flow_destination *dest,
1453                    int dest_num,
1454                    int *modify_mask,
1455                    bool *new_rule)
1456 {
1457         struct mlx5_flow_handle *handle;
1458         struct mlx5_flow_rule *rule = NULL;
1459         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1460         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1461         int type;
1462         int i = 0;
1463
1464         handle = alloc_handle((dest_num) ? dest_num : 1);
1465         if (!handle)
1466                 return ERR_PTR(-ENOMEM);
1467
1468         do {
1469                 if (dest) {
1470                         rule = find_flow_rule(fte, dest + i);
1471                         if (rule) {
1472                                 refcount_inc(&rule->node.refcount);
1473                                 goto rule_found;
1474                         }
1475                 }
1476
1477                 *new_rule = true;
1478                 rule = alloc_rule(dest + i);
1479                 if (!rule)
1480                         goto free_rules;
1481
1482                 /* Add dest to dests list- we need flow tables to be in the
1483                  * end of the list for forward to next prio rules.
1484                  */
1485                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1486                 if (dest &&
1487                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1488                         list_add(&rule->node.list, &fte->node.children);
1489                 else
1490                         list_add_tail(&rule->node.list, &fte->node.children);
1491                 if (dest) {
1492                         fte->dests_size++;
1493
1494                         if (is_fwd_dest_type(dest[i].type))
1495                                 fte->fwd_dests++;
1496
1497                         type = dest[i].type ==
1498                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1499                         *modify_mask |= type ? count : dst;
1500                 }
1501 rule_found:
1502                 handle->rule[i] = rule;
1503         } while (++i < dest_num);
1504
1505         return handle;
1506
1507 free_rules:
1508         destroy_flow_handle(fte, handle, dest, i);
1509         return ERR_PTR(-ENOMEM);
1510 }
1511
1512 /* fte should not be deleted while calling this function */
1513 static struct mlx5_flow_handle *
1514 add_rule_fte(struct fs_fte *fte,
1515              struct mlx5_flow_group *fg,
1516              struct mlx5_flow_destination *dest,
1517              int dest_num,
1518              bool update_action)
1519 {
1520         struct mlx5_flow_root_namespace *root;
1521         struct mlx5_flow_handle *handle;
1522         struct mlx5_flow_table *ft;
1523         int modify_mask = 0;
1524         int err;
1525         bool new_rule = false;
1526
1527         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1528                                     &new_rule);
1529         if (IS_ERR(handle) || !new_rule)
1530                 goto out;
1531
1532         if (update_action)
1533                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1534
1535         fs_get_obj(ft, fg->node.parent);
1536         root = find_root(&fg->node);
1537         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1538                 err = root->cmds->create_fte(root, ft, fg, fte);
1539         else
1540                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1541         if (err)
1542                 goto free_handle;
1543
1544         fte->node.active = true;
1545         fte->status |= FS_FTE_STATUS_EXISTING;
1546         atomic_inc(&fg->node.version);
1547
1548 out:
1549         return handle;
1550
1551 free_handle:
1552         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1553         return ERR_PTR(err);
1554 }
1555
1556 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1557                                                      const struct mlx5_flow_spec *spec)
1558 {
1559         struct list_head *prev = &ft->node.children;
1560         u32 max_fte = ft->autogroup.max_fte;
1561         unsigned int candidate_index = 0;
1562         unsigned int group_size = 0;
1563         struct mlx5_flow_group *fg;
1564
1565         if (!ft->autogroup.active)
1566                 return ERR_PTR(-ENOENT);
1567
1568         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1569                 group_size = ft->autogroup.group_size;
1570
1571         /*  max_fte == ft->autogroup.max_types */
1572         if (group_size == 0)
1573                 group_size = 1;
1574
1575         /* sorted by start_index */
1576         fs_for_each_fg(fg, ft) {
1577                 if (candidate_index + group_size > fg->start_index)
1578                         candidate_index = fg->start_index + fg->max_ftes;
1579                 else
1580                         break;
1581                 prev = &fg->node.list;
1582         }
1583
1584         if (candidate_index + group_size > max_fte)
1585                 return ERR_PTR(-ENOSPC);
1586
1587         fg = alloc_insert_flow_group(ft,
1588                                      spec->match_criteria_enable,
1589                                      spec->match_criteria,
1590                                      candidate_index,
1591                                      candidate_index + group_size - 1,
1592                                      prev);
1593         if (IS_ERR(fg))
1594                 goto out;
1595
1596         if (group_size == ft->autogroup.group_size)
1597                 ft->autogroup.num_groups++;
1598
1599 out:
1600         return fg;
1601 }
1602
1603 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1604                                   struct mlx5_flow_group *fg)
1605 {
1606         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1607         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1608         void *match_criteria_addr;
1609         u8 src_esw_owner_mask_on;
1610         void *misc;
1611         int err;
1612         u32 *in;
1613
1614         in = kvzalloc(inlen, GFP_KERNEL);
1615         if (!in)
1616                 return -ENOMEM;
1617
1618         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1619                  fg->mask.match_criteria_enable);
1620         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1621         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1622                  fg->max_ftes - 1);
1623
1624         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1625                             misc_parameters);
1626         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1627                                          source_eswitch_owner_vhca_id);
1628         MLX5_SET(create_flow_group_in, in,
1629                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1630
1631         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1632                                            in, match_criteria);
1633         memcpy(match_criteria_addr, fg->mask.match_criteria,
1634                sizeof(fg->mask.match_criteria));
1635
1636         err = root->cmds->create_flow_group(root, ft, in, fg);
1637         if (!err) {
1638                 fg->node.active = true;
1639                 trace_mlx5_fs_add_fg(fg);
1640         }
1641
1642         kvfree(in);
1643         return err;
1644 }
1645
1646 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1647                                 struct mlx5_flow_destination *d2)
1648 {
1649         if (d1->type == d2->type) {
1650                 if (((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT ||
1651                       d1->type == MLX5_FLOW_DESTINATION_TYPE_UPLINK) &&
1652                      d1->vport.num == d2->vport.num &&
1653                      d1->vport.flags == d2->vport.flags &&
1654                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1655                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1656                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1657                       (d1->vport.pkt_reformat->id ==
1658                        d2->vport.pkt_reformat->id) : true)) ||
1659                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1660                      d1->ft == d2->ft) ||
1661                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1662                      d1->tir_num == d2->tir_num) ||
1663                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1664                      d1->ft_num == d2->ft_num) ||
1665                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER &&
1666                      d1->sampler_id == d2->sampler_id) ||
1667                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_RANGE &&
1668                      d1->range.field == d2->range.field &&
1669                      d1->range.hit_ft == d2->range.hit_ft &&
1670                      d1->range.miss_ft == d2->range.miss_ft &&
1671                      d1->range.min == d2->range.min &&
1672                      d1->range.max == d2->range.max))
1673                         return true;
1674         }
1675
1676         return false;
1677 }
1678
1679 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1680                                              struct mlx5_flow_destination *dest)
1681 {
1682         struct mlx5_flow_rule *rule;
1683
1684         list_for_each_entry(rule, &fte->node.children, node.list) {
1685                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1686                         return rule;
1687         }
1688         return NULL;
1689 }
1690
1691 static bool check_conflicting_actions_vlan(const struct mlx5_fs_vlan *vlan0,
1692                                            const struct mlx5_fs_vlan *vlan1)
1693 {
1694         return vlan0->ethtype != vlan1->ethtype ||
1695                vlan0->vid != vlan1->vid ||
1696                vlan0->prio != vlan1->prio;
1697 }
1698
1699 static bool check_conflicting_actions(const struct mlx5_flow_act *act1,
1700                                       const struct mlx5_flow_act *act2)
1701 {
1702         u32 action1 = act1->action;
1703         u32 action2 = act2->action;
1704         u32 xored_actions;
1705
1706         xored_actions = action1 ^ action2;
1707
1708         /* if one rule only wants to count, it's ok */
1709         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1710             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1711                 return false;
1712
1713         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1714                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1715                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1716                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1717                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1718                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1719                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1720                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1721                 return true;
1722
1723         if (action1 & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT &&
1724             act1->pkt_reformat != act2->pkt_reformat)
1725                 return true;
1726
1727         if (action1 & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR &&
1728             act1->modify_hdr != act2->modify_hdr)
1729                 return true;
1730
1731         if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH &&
1732             check_conflicting_actions_vlan(&act1->vlan[0], &act2->vlan[0]))
1733                 return true;
1734
1735         if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2 &&
1736             check_conflicting_actions_vlan(&act1->vlan[1], &act2->vlan[1]))
1737                 return true;
1738
1739         return false;
1740 }
1741
1742 static int check_conflicting_ftes(struct fs_fte *fte,
1743                                   const struct mlx5_flow_context *flow_context,
1744                                   const struct mlx5_flow_act *flow_act)
1745 {
1746         if (check_conflicting_actions(flow_act, &fte->action)) {
1747                 mlx5_core_warn(get_dev(&fte->node),
1748                                "Found two FTEs with conflicting actions\n");
1749                 return -EEXIST;
1750         }
1751
1752         if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1753             fte->flow_context.flow_tag != flow_context->flow_tag) {
1754                 mlx5_core_warn(get_dev(&fte->node),
1755                                "FTE flow tag %u already exists with different flow tag %u\n",
1756                                fte->flow_context.flow_tag,
1757                                flow_context->flow_tag);
1758                 return -EEXIST;
1759         }
1760
1761         return 0;
1762 }
1763
1764 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1765                                             const struct mlx5_flow_spec *spec,
1766                                             struct mlx5_flow_act *flow_act,
1767                                             struct mlx5_flow_destination *dest,
1768                                             int dest_num,
1769                                             struct fs_fte *fte)
1770 {
1771         struct mlx5_flow_handle *handle;
1772         int old_action;
1773         int i;
1774         int ret;
1775
1776         ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1777         if (ret)
1778                 return ERR_PTR(ret);
1779
1780         old_action = fte->action.action;
1781         fte->action.action |= flow_act->action;
1782         handle = add_rule_fte(fte, fg, dest, dest_num,
1783                               old_action != flow_act->action);
1784         if (IS_ERR(handle)) {
1785                 fte->action.action = old_action;
1786                 return handle;
1787         }
1788         trace_mlx5_fs_set_fte(fte, false);
1789
1790         for (i = 0; i < handle->num_rules; i++) {
1791                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1792                         tree_add_node(&handle->rule[i]->node, &fte->node);
1793                         trace_mlx5_fs_add_rule(handle->rule[i]);
1794                 }
1795         }
1796         return handle;
1797 }
1798
1799 static bool counter_is_valid(u32 action)
1800 {
1801         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1802                           MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1803                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1804 }
1805
1806 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1807                           struct mlx5_flow_act *flow_act,
1808                           struct mlx5_flow_table *ft)
1809 {
1810         bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1811         u32 action = flow_act->action;
1812
1813         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1814                 return counter_is_valid(action);
1815
1816         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1817                 return true;
1818
1819         if (ignore_level) {
1820                 if (ft->type != FS_FT_FDB &&
1821                     ft->type != FS_FT_NIC_RX &&
1822                     ft->type != FS_FT_NIC_TX)
1823                         return false;
1824
1825                 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1826                     ft->type != dest->ft->type)
1827                         return false;
1828         }
1829
1830         if (!dest || ((dest->type ==
1831             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1832             (dest->ft->level <= ft->level && !ignore_level)))
1833                 return false;
1834         return true;
1835 }
1836
1837 struct match_list {
1838         struct list_head        list;
1839         struct mlx5_flow_group *g;
1840 };
1841
1842 static void free_match_list(struct match_list *head, bool ft_locked)
1843 {
1844         struct match_list *iter, *match_tmp;
1845
1846         list_for_each_entry_safe(iter, match_tmp, &head->list,
1847                                  list) {
1848                 tree_put_node(&iter->g->node, ft_locked);
1849                 list_del(&iter->list);
1850                 kfree(iter);
1851         }
1852 }
1853
1854 static int build_match_list(struct match_list *match_head,
1855                             struct mlx5_flow_table *ft,
1856                             const struct mlx5_flow_spec *spec,
1857                             struct mlx5_flow_group *fg,
1858                             bool ft_locked)
1859 {
1860         struct rhlist_head *tmp, *list;
1861         struct mlx5_flow_group *g;
1862
1863         rcu_read_lock();
1864         INIT_LIST_HEAD(&match_head->list);
1865         /* Collect all fgs which has a matching match_criteria */
1866         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1867         /* RCU is atomic, we can't execute FW commands here */
1868         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1869                 struct match_list *curr_match;
1870
1871                 if (fg && fg != g)
1872                         continue;
1873
1874                 if (unlikely(!tree_get_node(&g->node)))
1875                         continue;
1876
1877                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1878                 if (!curr_match) {
1879                         rcu_read_unlock();
1880                         free_match_list(match_head, ft_locked);
1881                         return -ENOMEM;
1882                 }
1883                 curr_match->g = g;
1884                 list_add_tail(&curr_match->list, &match_head->list);
1885         }
1886         rcu_read_unlock();
1887         return 0;
1888 }
1889
1890 static u64 matched_fgs_get_version(struct list_head *match_head)
1891 {
1892         struct match_list *iter;
1893         u64 version = 0;
1894
1895         list_for_each_entry(iter, match_head, list)
1896                 version += (u64)atomic_read(&iter->g->node.version);
1897         return version;
1898 }
1899
1900 static struct fs_fte *
1901 lookup_fte_locked(struct mlx5_flow_group *g,
1902                   const u32 *match_value,
1903                   bool take_write)
1904 {
1905         struct fs_fte *fte_tmp;
1906
1907         if (take_write)
1908                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1909         else
1910                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1911         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1912                                          rhash_fte);
1913         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1914                 fte_tmp = NULL;
1915                 goto out;
1916         }
1917         if (!fte_tmp->node.active) {
1918                 tree_put_node(&fte_tmp->node, false);
1919                 fte_tmp = NULL;
1920                 goto out;
1921         }
1922
1923         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1924 out:
1925         if (take_write)
1926                 up_write_ref_node(&g->node, false);
1927         else
1928                 up_read_ref_node(&g->node);
1929         return fte_tmp;
1930 }
1931
1932 static struct mlx5_flow_handle *
1933 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1934                        struct list_head *match_head,
1935                        const struct mlx5_flow_spec *spec,
1936                        struct mlx5_flow_act *flow_act,
1937                        struct mlx5_flow_destination *dest,
1938                        int dest_num,
1939                        int ft_version)
1940 {
1941         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1942         struct mlx5_flow_group *g;
1943         struct mlx5_flow_handle *rule;
1944         struct match_list *iter;
1945         bool take_write = false;
1946         struct fs_fte *fte;
1947         u64  version = 0;
1948         int err;
1949
1950         fte = alloc_fte(ft, spec, flow_act);
1951         if (IS_ERR(fte))
1952                 return  ERR_PTR(-ENOMEM);
1953
1954 search_again_locked:
1955         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1956                 goto skip_search;
1957         version = matched_fgs_get_version(match_head);
1958         /* Try to find an fte with identical match value and attempt update its
1959          * action.
1960          */
1961         list_for_each_entry(iter, match_head, list) {
1962                 struct fs_fte *fte_tmp;
1963
1964                 g = iter->g;
1965                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1966                 if (!fte_tmp)
1967                         continue;
1968                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1969                 /* No error check needed here, because insert_fte() is not called */
1970                 up_write_ref_node(&fte_tmp->node, false);
1971                 tree_put_node(&fte_tmp->node, false);
1972                 kmem_cache_free(steering->ftes_cache, fte);
1973                 return rule;
1974         }
1975
1976 skip_search:
1977         /* No group with matching fte found, or we skipped the search.
1978          * Try to add a new fte to any matching fg.
1979          */
1980
1981         /* Check the ft version, for case that new flow group
1982          * was added while the fgs weren't locked
1983          */
1984         if (atomic_read(&ft->node.version) != ft_version) {
1985                 rule = ERR_PTR(-EAGAIN);
1986                 goto out;
1987         }
1988
1989         /* Check the fgs version. If version have changed it could be that an
1990          * FTE with the same match value was added while the fgs weren't
1991          * locked.
1992          */
1993         if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1994             version != matched_fgs_get_version(match_head)) {
1995                 take_write = true;
1996                 goto search_again_locked;
1997         }
1998
1999         list_for_each_entry(iter, match_head, list) {
2000                 g = iter->g;
2001
2002                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
2003
2004                 if (!g->node.active) {
2005                         up_write_ref_node(&g->node, false);
2006                         continue;
2007                 }
2008
2009                 err = insert_fte(g, fte);
2010                 if (err) {
2011                         up_write_ref_node(&g->node, false);
2012                         if (err == -ENOSPC)
2013                                 continue;
2014                         kmem_cache_free(steering->ftes_cache, fte);
2015                         return ERR_PTR(err);
2016                 }
2017
2018                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
2019                 up_write_ref_node(&g->node, false);
2020                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
2021                 up_write_ref_node(&fte->node, false);
2022                 if (IS_ERR(rule))
2023                         tree_put_node(&fte->node, false);
2024                 return rule;
2025         }
2026         rule = ERR_PTR(-ENOENT);
2027 out:
2028         kmem_cache_free(steering->ftes_cache, fte);
2029         return rule;
2030 }
2031
2032 static struct mlx5_flow_handle *
2033 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
2034                      const struct mlx5_flow_spec *spec,
2035                      struct mlx5_flow_act *flow_act,
2036                      struct mlx5_flow_destination *dest,
2037                      int dest_num)
2038
2039 {
2040         struct mlx5_flow_steering *steering = get_steering(&ft->node);
2041         struct mlx5_flow_handle *rule;
2042         struct match_list match_head;
2043         struct mlx5_flow_group *g;
2044         bool take_write = false;
2045         struct fs_fte *fte;
2046         int version;
2047         int err;
2048         int i;
2049
2050         if (!check_valid_spec(spec))
2051                 return ERR_PTR(-EINVAL);
2052
2053         if (flow_act->fg && ft->autogroup.active)
2054                 return ERR_PTR(-EINVAL);
2055
2056         if (dest && dest_num <= 0)
2057                 return ERR_PTR(-EINVAL);
2058
2059         for (i = 0; i < dest_num; i++) {
2060                 if (!dest_is_valid(&dest[i], flow_act, ft))
2061                         return ERR_PTR(-EINVAL);
2062         }
2063         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
2064 search_again_locked:
2065         version = atomic_read(&ft->node.version);
2066
2067         /* Collect all fgs which has a matching match_criteria */
2068         err = build_match_list(&match_head, ft, spec, flow_act->fg, take_write);
2069         if (err) {
2070                 if (take_write)
2071                         up_write_ref_node(&ft->node, false);
2072                 else
2073                         up_read_ref_node(&ft->node);
2074                 return ERR_PTR(err);
2075         }
2076
2077         if (!take_write)
2078                 up_read_ref_node(&ft->node);
2079
2080         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
2081                                       dest_num, version);
2082         free_match_list(&match_head, take_write);
2083         if (!IS_ERR(rule) ||
2084             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
2085                 if (take_write)
2086                         up_write_ref_node(&ft->node, false);
2087                 return rule;
2088         }
2089
2090         if (!take_write) {
2091                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
2092                 take_write = true;
2093         }
2094
2095         if (PTR_ERR(rule) == -EAGAIN ||
2096             version != atomic_read(&ft->node.version))
2097                 goto search_again_locked;
2098
2099         g = alloc_auto_flow_group(ft, spec);
2100         if (IS_ERR(g)) {
2101                 rule = ERR_CAST(g);
2102                 up_write_ref_node(&ft->node, false);
2103                 return rule;
2104         }
2105
2106         fte = alloc_fte(ft, spec, flow_act);
2107         if (IS_ERR(fte)) {
2108                 up_write_ref_node(&ft->node, false);
2109                 err = PTR_ERR(fte);
2110                 goto err_alloc_fte;
2111         }
2112
2113         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
2114         up_write_ref_node(&ft->node, false);
2115
2116         err = create_auto_flow_group(ft, g);
2117         if (err)
2118                 goto err_release_fg;
2119
2120         err = insert_fte(g, fte);
2121         if (err)
2122                 goto err_release_fg;
2123
2124         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
2125         up_write_ref_node(&g->node, false);
2126         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
2127         up_write_ref_node(&fte->node, false);
2128         if (IS_ERR(rule))
2129                 tree_put_node(&fte->node, false);
2130         tree_put_node(&g->node, false);
2131         return rule;
2132
2133 err_release_fg:
2134         up_write_ref_node(&g->node, false);
2135         kmem_cache_free(steering->ftes_cache, fte);
2136 err_alloc_fte:
2137         tree_put_node(&g->node, false);
2138         return ERR_PTR(err);
2139 }
2140
2141 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
2142 {
2143         return ((ft->type == FS_FT_NIC_RX) &&
2144                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
2145 }
2146
2147 struct mlx5_flow_handle *
2148 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
2149                     const struct mlx5_flow_spec *spec,
2150                     struct mlx5_flow_act *flow_act,
2151                     struct mlx5_flow_destination *dest,
2152                     int num_dest)
2153 {
2154         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2155         static const struct mlx5_flow_spec zero_spec = {};
2156         struct mlx5_flow_destination *gen_dest = NULL;
2157         struct mlx5_flow_table *next_ft = NULL;
2158         struct mlx5_flow_handle *handle = NULL;
2159         u32 sw_action = flow_act->action;
2160         int i;
2161
2162         if (!spec)
2163                 spec = &zero_spec;
2164
2165         if (!is_fwd_next_action(sw_action))
2166                 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2167
2168         if (!fwd_next_prio_supported(ft))
2169                 return ERR_PTR(-EOPNOTSUPP);
2170
2171         mutex_lock(&root->chain_lock);
2172         next_ft = find_next_fwd_ft(ft, flow_act);
2173         if (!next_ft) {
2174                 handle = ERR_PTR(-EOPNOTSUPP);
2175                 goto unlock;
2176         }
2177
2178         gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
2179                            GFP_KERNEL);
2180         if (!gen_dest) {
2181                 handle = ERR_PTR(-ENOMEM);
2182                 goto unlock;
2183         }
2184         for (i = 0; i < num_dest; i++)
2185                 gen_dest[i] = dest[i];
2186         gen_dest[i].type =
2187                 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
2188         gen_dest[i].ft = next_ft;
2189         dest = gen_dest;
2190         num_dest++;
2191         flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
2192                               MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
2193         flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2194         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2195         if (IS_ERR(handle))
2196                 goto unlock;
2197
2198         if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
2199                 mutex_lock(&next_ft->lock);
2200                 list_add(&handle->rule[num_dest - 1]->next_ft,
2201                          &next_ft->fwd_rules);
2202                 mutex_unlock(&next_ft->lock);
2203                 handle->rule[num_dest - 1]->sw_action = sw_action;
2204                 handle->rule[num_dest - 1]->ft = ft;
2205         }
2206 unlock:
2207         mutex_unlock(&root->chain_lock);
2208         kfree(gen_dest);
2209         return handle;
2210 }
2211 EXPORT_SYMBOL(mlx5_add_flow_rules);
2212
2213 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2214 {
2215         struct fs_fte *fte;
2216         int i;
2217
2218         /* In order to consolidate the HW changes we lock the FTE for other
2219          * changes, and increase its refcount, in order not to perform the
2220          * "del" functions of the FTE. Will handle them here.
2221          * The removal of the rules is done under locked FTE.
2222          * After removing all the handle's rules, if there are remaining
2223          * rules, it means we just need to modify the FTE in FW, and
2224          * unlock/decrease the refcount we increased before.
2225          * Otherwise, it means the FTE should be deleted. First delete the
2226          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2227          * the FTE, which will handle the last decrease of the refcount, as
2228          * well as required handling of its parent.
2229          */
2230         fs_get_obj(fte, handle->rule[0]->node.parent);
2231         down_write_ref_node(&fte->node, false);
2232         for (i = handle->num_rules - 1; i >= 0; i--)
2233                 tree_remove_node(&handle->rule[i]->node, true);
2234         if (list_empty(&fte->node.children)) {
2235                 fte->node.del_hw_func(&fte->node);
2236                 /* Avoid double call to del_hw_fte */
2237                 fte->node.del_hw_func = NULL;
2238                 up_write_ref_node(&fte->node, false);
2239                 tree_put_node(&fte->node, false);
2240         } else if (fte->dests_size) {
2241                 if (fte->modify_mask)
2242                         modify_fte(fte);
2243                 up_write_ref_node(&fte->node, false);
2244         } else {
2245                 up_write_ref_node(&fte->node, false);
2246         }
2247         kfree(handle);
2248 }
2249 EXPORT_SYMBOL(mlx5_del_flow_rules);
2250
2251 /* Assuming prio->node.children(flow tables) is sorted by level */
2252 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2253 {
2254         struct fs_node *prio_parent, *child;
2255         struct fs_prio *prio;
2256
2257         fs_get_obj(prio, ft->node.parent);
2258
2259         if (!list_is_last(&ft->node.list, &prio->node.children))
2260                 return list_next_entry(ft, node.list);
2261
2262         prio_parent = find_prio_chains_parent(&prio->node, &child);
2263
2264         if (prio_parent && list_is_first(&child->list, &prio_parent->children))
2265                 return find_closest_ft(&prio->node, false, false);
2266
2267         return find_next_chained_ft(&prio->node);
2268 }
2269
2270 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2271 {
2272         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2273         struct mlx5_ft_underlay_qp *uqp;
2274         struct mlx5_flow_table *new_root_ft = NULL;
2275         int err = 0;
2276         u32 qpn;
2277
2278         if (root->root_ft != ft)
2279                 return 0;
2280
2281         new_root_ft = find_next_ft(ft);
2282         if (!new_root_ft) {
2283                 root->root_ft = NULL;
2284                 return 0;
2285         }
2286
2287         if (list_empty(&root->underlay_qpns)) {
2288                 /* Don't set any QPN (zero) in case QPN list is empty */
2289                 qpn = 0;
2290                 err = root->cmds->update_root_ft(root, new_root_ft,
2291                                                  qpn, false);
2292         } else {
2293                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2294                         qpn = uqp->qpn;
2295                         err = root->cmds->update_root_ft(root,
2296                                                          new_root_ft, qpn,
2297                                                          false);
2298                         if (err)
2299                                 break;
2300                 }
2301         }
2302
2303         if (err)
2304                 mlx5_core_warn(root->dev,
2305                                "Update root flow table of id(%u) qpn(%d) failed\n",
2306                                ft->id, qpn);
2307         else
2308                 root->root_ft = new_root_ft;
2309
2310         return 0;
2311 }
2312
2313 /* Connect flow table from previous priority to
2314  * the next flow table.
2315  */
2316 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2317 {
2318         struct mlx5_core_dev *dev = get_dev(&ft->node);
2319         struct mlx5_flow_table *next_ft;
2320         struct fs_prio *prio;
2321         int err = 0;
2322
2323         err = update_root_ft_destroy(ft);
2324         if (err)
2325                 return err;
2326
2327         fs_get_obj(prio, ft->node.parent);
2328         if  (!(list_first_entry(&prio->node.children,
2329                                 struct mlx5_flow_table,
2330                                 node.list) == ft))
2331                 return 0;
2332
2333         next_ft = find_next_ft(ft);
2334         err = connect_fwd_rules(dev, next_ft, ft);
2335         if (err)
2336                 return err;
2337
2338         err = connect_prev_fts(dev, next_ft, prio);
2339         if (err)
2340                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2341                                ft->id);
2342         return err;
2343 }
2344
2345 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2346 {
2347         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2348         int err = 0;
2349
2350         mutex_lock(&root->chain_lock);
2351         if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2352                 err = disconnect_flow_table(ft);
2353         if (err) {
2354                 mutex_unlock(&root->chain_lock);
2355                 return err;
2356         }
2357         if (tree_remove_node(&ft->node, false))
2358                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2359                                ft->id);
2360         mutex_unlock(&root->chain_lock);
2361
2362         return err;
2363 }
2364 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2365
2366 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2367 {
2368         if (tree_remove_node(&fg->node, false))
2369                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2370                                fg->id);
2371 }
2372 EXPORT_SYMBOL(mlx5_destroy_flow_group);
2373
2374 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2375                                                 int n)
2376 {
2377         struct mlx5_flow_steering *steering = dev->priv.steering;
2378
2379         if (!steering || !steering->fdb_sub_ns)
2380                 return NULL;
2381
2382         return steering->fdb_sub_ns[n];
2383 }
2384 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2385
2386 static bool is_nic_rx_ns(enum mlx5_flow_namespace_type type)
2387 {
2388         switch (type) {
2389         case MLX5_FLOW_NAMESPACE_BYPASS:
2390         case MLX5_FLOW_NAMESPACE_KERNEL_RX_MACSEC:
2391         case MLX5_FLOW_NAMESPACE_LAG:
2392         case MLX5_FLOW_NAMESPACE_OFFLOADS:
2393         case MLX5_FLOW_NAMESPACE_ETHTOOL:
2394         case MLX5_FLOW_NAMESPACE_KERNEL:
2395         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
2396         case MLX5_FLOW_NAMESPACE_ANCHOR:
2397                 return true;
2398         default:
2399                 return false;
2400         }
2401 }
2402
2403 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2404                                                     enum mlx5_flow_namespace_type type)
2405 {
2406         struct mlx5_flow_steering *steering = dev->priv.steering;
2407         struct mlx5_flow_root_namespace *root_ns;
2408         int prio = 0;
2409         struct fs_prio *fs_prio;
2410         struct mlx5_flow_namespace *ns;
2411
2412         if (!steering)
2413                 return NULL;
2414
2415         switch (type) {
2416         case MLX5_FLOW_NAMESPACE_FDB:
2417                 if (steering->fdb_root_ns)
2418                         return &steering->fdb_root_ns->ns;
2419                 return NULL;
2420         case MLX5_FLOW_NAMESPACE_PORT_SEL:
2421                 if (steering->port_sel_root_ns)
2422                         return &steering->port_sel_root_ns->ns;
2423                 return NULL;
2424         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2425                 if (steering->sniffer_rx_root_ns)
2426                         return &steering->sniffer_rx_root_ns->ns;
2427                 return NULL;
2428         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2429                 if (steering->sniffer_tx_root_ns)
2430                         return &steering->sniffer_tx_root_ns->ns;
2431                 return NULL;
2432         case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
2433                 root_ns = steering->fdb_root_ns;
2434                 prio =  FDB_BYPASS_PATH;
2435                 break;
2436         case MLX5_FLOW_NAMESPACE_EGRESS:
2437         case MLX5_FLOW_NAMESPACE_EGRESS_IPSEC:
2438         case MLX5_FLOW_NAMESPACE_EGRESS_MACSEC:
2439                 root_ns = steering->egress_root_ns;
2440                 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2441                 break;
2442         case MLX5_FLOW_NAMESPACE_RDMA_RX:
2443                 root_ns = steering->rdma_rx_root_ns;
2444                 prio = RDMA_RX_BYPASS_PRIO;
2445                 break;
2446         case MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL:
2447                 root_ns = steering->rdma_rx_root_ns;
2448                 prio = RDMA_RX_KERNEL_PRIO;
2449                 break;
2450         case MLX5_FLOW_NAMESPACE_RDMA_TX:
2451                 root_ns = steering->rdma_tx_root_ns;
2452                 break;
2453         case MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS:
2454                 root_ns = steering->rdma_rx_root_ns;
2455                 prio = RDMA_RX_COUNTERS_PRIO;
2456                 break;
2457         case MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS:
2458                 root_ns = steering->rdma_tx_root_ns;
2459                 prio = RDMA_TX_COUNTERS_PRIO;
2460                 break;
2461         case MLX5_FLOW_NAMESPACE_RDMA_RX_IPSEC:
2462                 root_ns = steering->rdma_rx_root_ns;
2463                 prio = RDMA_RX_IPSEC_PRIO;
2464                 break;
2465         case MLX5_FLOW_NAMESPACE_RDMA_TX_IPSEC:
2466                 root_ns = steering->rdma_tx_root_ns;
2467                 prio = RDMA_TX_IPSEC_PRIO;
2468                 break;
2469         default: /* Must be NIC RX */
2470                 WARN_ON(!is_nic_rx_ns(type));
2471                 root_ns = steering->root_ns;
2472                 prio = type;
2473                 break;
2474         }
2475
2476         if (!root_ns)
2477                 return NULL;
2478
2479         fs_prio = find_prio(&root_ns->ns, prio);
2480         if (!fs_prio)
2481                 return NULL;
2482
2483         ns = list_first_entry(&fs_prio->node.children,
2484                               typeof(*ns),
2485                               node.list);
2486
2487         return ns;
2488 }
2489 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2490
2491 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2492                                                               enum mlx5_flow_namespace_type type,
2493                                                               int vport)
2494 {
2495         struct mlx5_flow_steering *steering = dev->priv.steering;
2496
2497         if (!steering)
2498                 return NULL;
2499
2500         switch (type) {
2501         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2502                 if (vport >= steering->esw_egress_acl_vports)
2503                         return NULL;
2504                 if (steering->esw_egress_root_ns &&
2505                     steering->esw_egress_root_ns[vport])
2506                         return &steering->esw_egress_root_ns[vport]->ns;
2507                 else
2508                         return NULL;
2509         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2510                 if (vport >= steering->esw_ingress_acl_vports)
2511                         return NULL;
2512                 if (steering->esw_ingress_root_ns &&
2513                     steering->esw_ingress_root_ns[vport])
2514                         return &steering->esw_ingress_root_ns[vport]->ns;
2515                 else
2516                         return NULL;
2517         default:
2518                 return NULL;
2519         }
2520 }
2521
2522 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2523                                        unsigned int prio,
2524                                        int num_levels,
2525                                        enum fs_node_type type)
2526 {
2527         struct fs_prio *fs_prio;
2528
2529         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2530         if (!fs_prio)
2531                 return ERR_PTR(-ENOMEM);
2532
2533         fs_prio->node.type = type;
2534         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2535         tree_add_node(&fs_prio->node, &ns->node);
2536         fs_prio->num_levels = num_levels;
2537         fs_prio->prio = prio;
2538         list_add_tail(&fs_prio->node.list, &ns->node.children);
2539
2540         return fs_prio;
2541 }
2542
2543 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2544                                               unsigned int prio,
2545                                               int num_levels)
2546 {
2547         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2548 }
2549
2550 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2551                                       unsigned int prio, int num_levels)
2552 {
2553         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2554 }
2555
2556 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2557                                                      *ns)
2558 {
2559         ns->node.type = FS_TYPE_NAMESPACE;
2560
2561         return ns;
2562 }
2563
2564 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2565                                                        int def_miss_act)
2566 {
2567         struct mlx5_flow_namespace      *ns;
2568
2569         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2570         if (!ns)
2571                 return ERR_PTR(-ENOMEM);
2572
2573         fs_init_namespace(ns);
2574         ns->def_miss_action = def_miss_act;
2575         tree_init_node(&ns->node, NULL, del_sw_ns);
2576         tree_add_node(&ns->node, &prio->node);
2577         list_add_tail(&ns->node.list, &prio->node.children);
2578
2579         return ns;
2580 }
2581
2582 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2583                              struct init_tree_node *prio_metadata)
2584 {
2585         struct fs_prio *fs_prio;
2586         int i;
2587
2588         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2589                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2590                 if (IS_ERR(fs_prio))
2591                         return PTR_ERR(fs_prio);
2592         }
2593         return 0;
2594 }
2595
2596 #define FLOW_TABLE_BIT_SZ 1
2597 #define GET_FLOW_TABLE_CAP(dev, offset) \
2598         ((be32_to_cpu(*((__be32 *)(dev->caps.hca[MLX5_CAP_FLOW_TABLE]->cur) +   \
2599                         offset / 32)) >>                                        \
2600           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2601 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2602 {
2603         int i;
2604
2605         for (i = 0; i < caps->arr_sz; i++) {
2606                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2607                         return false;
2608         }
2609         return true;
2610 }
2611
2612 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2613                                     struct init_tree_node *init_node,
2614                                     struct fs_node *fs_parent_node,
2615                                     struct init_tree_node *init_parent_node,
2616                                     int prio)
2617 {
2618         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2619                                               flow_table_properties_nic_receive.
2620                                               max_ft_level);
2621         struct mlx5_flow_namespace *fs_ns;
2622         struct fs_prio *fs_prio;
2623         struct fs_node *base;
2624         int i;
2625         int err;
2626
2627         if (init_node->type == FS_TYPE_PRIO) {
2628                 if ((init_node->min_ft_level > max_ft_level) ||
2629                     !has_required_caps(steering->dev, &init_node->caps))
2630                         return 0;
2631
2632                 fs_get_obj(fs_ns, fs_parent_node);
2633                 if (init_node->num_leaf_prios)
2634                         return create_leaf_prios(fs_ns, prio, init_node);
2635                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2636                 if (IS_ERR(fs_prio))
2637                         return PTR_ERR(fs_prio);
2638                 base = &fs_prio->node;
2639         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2640                 fs_get_obj(fs_prio, fs_parent_node);
2641                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2642                 if (IS_ERR(fs_ns))
2643                         return PTR_ERR(fs_ns);
2644                 base = &fs_ns->node;
2645         } else {
2646                 return -EINVAL;
2647         }
2648         prio = 0;
2649         for (i = 0; i < init_node->ar_size; i++) {
2650                 err = init_root_tree_recursive(steering, &init_node->children[i],
2651                                                base, init_node, prio);
2652                 if (err)
2653                         return err;
2654                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2655                     init_node->children[i].num_leaf_prios) {
2656                         prio += init_node->children[i].num_leaf_prios;
2657                 }
2658         }
2659
2660         return 0;
2661 }
2662
2663 static int init_root_tree(struct mlx5_flow_steering *steering,
2664                           struct init_tree_node *init_node,
2665                           struct fs_node *fs_parent_node)
2666 {
2667         int err;
2668         int i;
2669
2670         for (i = 0; i < init_node->ar_size; i++) {
2671                 err = init_root_tree_recursive(steering, &init_node->children[i],
2672                                                fs_parent_node,
2673                                                init_node, i);
2674                 if (err)
2675                         return err;
2676         }
2677         return 0;
2678 }
2679
2680 static void del_sw_root_ns(struct fs_node *node)
2681 {
2682         struct mlx5_flow_root_namespace *root_ns;
2683         struct mlx5_flow_namespace *ns;
2684
2685         fs_get_obj(ns, node);
2686         root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2687         mutex_destroy(&root_ns->chain_lock);
2688         kfree(node);
2689 }
2690
2691 static struct mlx5_flow_root_namespace
2692 *create_root_ns(struct mlx5_flow_steering *steering,
2693                 enum fs_flow_table_type table_type)
2694 {
2695         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2696         struct mlx5_flow_root_namespace *root_ns;
2697         struct mlx5_flow_namespace *ns;
2698
2699         /* Create the root namespace */
2700         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2701         if (!root_ns)
2702                 return NULL;
2703
2704         root_ns->dev = steering->dev;
2705         root_ns->table_type = table_type;
2706         root_ns->cmds = cmds;
2707
2708         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2709
2710         ns = &root_ns->ns;
2711         fs_init_namespace(ns);
2712         mutex_init(&root_ns->chain_lock);
2713         tree_init_node(&ns->node, NULL, del_sw_root_ns);
2714         tree_add_node(&ns->node, NULL);
2715
2716         return root_ns;
2717 }
2718
2719 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2720
2721 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2722 {
2723         struct fs_prio *prio;
2724
2725         fs_for_each_prio(prio, ns) {
2726                  /* This updates prio start_level and num_levels */
2727                 set_prio_attrs_in_prio(prio, acc_level);
2728                 acc_level += prio->num_levels;
2729         }
2730         return acc_level;
2731 }
2732
2733 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2734 {
2735         struct mlx5_flow_namespace *ns;
2736         int acc_level_ns = acc_level;
2737
2738         prio->start_level = acc_level;
2739         fs_for_each_ns(ns, prio) {
2740                 /* This updates start_level and num_levels of ns's priority descendants */
2741                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2742
2743                 /* If this a prio with chains, and we can jump from one chain
2744                  * (namespace) to another, so we accumulate the levels
2745                  */
2746                 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2747                         acc_level = acc_level_ns;
2748         }
2749
2750         if (!prio->num_levels)
2751                 prio->num_levels = acc_level_ns - prio->start_level;
2752         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2753 }
2754
2755 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2756 {
2757         struct mlx5_flow_namespace *ns = &root_ns->ns;
2758         struct fs_prio *prio;
2759         int start_level = 0;
2760
2761         fs_for_each_prio(prio, ns) {
2762                 set_prio_attrs_in_prio(prio, start_level);
2763                 start_level += prio->num_levels;
2764         }
2765 }
2766
2767 #define ANCHOR_PRIO 0
2768 #define ANCHOR_SIZE 1
2769 #define ANCHOR_LEVEL 0
2770 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2771 {
2772         struct mlx5_flow_namespace *ns = NULL;
2773         struct mlx5_flow_table_attr ft_attr = {};
2774         struct mlx5_flow_table *ft;
2775
2776         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2777         if (WARN_ON(!ns))
2778                 return -EINVAL;
2779
2780         ft_attr.max_fte = ANCHOR_SIZE;
2781         ft_attr.level   = ANCHOR_LEVEL;
2782         ft_attr.prio    = ANCHOR_PRIO;
2783
2784         ft = mlx5_create_flow_table(ns, &ft_attr);
2785         if (IS_ERR(ft)) {
2786                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2787                 return PTR_ERR(ft);
2788         }
2789         return 0;
2790 }
2791
2792 static int init_root_ns(struct mlx5_flow_steering *steering)
2793 {
2794         int err;
2795
2796         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2797         if (!steering->root_ns)
2798                 return -ENOMEM;
2799
2800         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2801         if (err)
2802                 goto out_err;
2803
2804         set_prio_attrs(steering->root_ns);
2805         err = create_anchor_flow_table(steering);
2806         if (err)
2807                 goto out_err;
2808
2809         return 0;
2810
2811 out_err:
2812         cleanup_root_ns(steering->root_ns);
2813         steering->root_ns = NULL;
2814         return err;
2815 }
2816
2817 static void clean_tree(struct fs_node *node)
2818 {
2819         if (node) {
2820                 struct fs_node *iter;
2821                 struct fs_node *temp;
2822
2823                 tree_get_node(node);
2824                 list_for_each_entry_safe(iter, temp, &node->children, list)
2825                         clean_tree(iter);
2826                 tree_put_node(node, false);
2827                 tree_remove_node(node, false);
2828         }
2829 }
2830
2831 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2832 {
2833         if (!root_ns)
2834                 return;
2835
2836         clean_tree(&root_ns->ns.node);
2837 }
2838
2839 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2840 {
2841         struct fs_prio *prio;
2842
2843         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2844         if (!steering->sniffer_tx_root_ns)
2845                 return -ENOMEM;
2846
2847         /* Create single prio */
2848         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2849         return PTR_ERR_OR_ZERO(prio);
2850 }
2851
2852 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2853 {
2854         struct fs_prio *prio;
2855
2856         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2857         if (!steering->sniffer_rx_root_ns)
2858                 return -ENOMEM;
2859
2860         /* Create single prio */
2861         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2862         return PTR_ERR_OR_ZERO(prio);
2863 }
2864
2865 #define PORT_SEL_NUM_LEVELS 3
2866 static int init_port_sel_root_ns(struct mlx5_flow_steering *steering)
2867 {
2868         struct fs_prio *prio;
2869
2870         steering->port_sel_root_ns = create_root_ns(steering, FS_FT_PORT_SEL);
2871         if (!steering->port_sel_root_ns)
2872                 return -ENOMEM;
2873
2874         /* Create single prio */
2875         prio = fs_create_prio(&steering->port_sel_root_ns->ns, 0,
2876                               PORT_SEL_NUM_LEVELS);
2877         return PTR_ERR_OR_ZERO(prio);
2878 }
2879
2880 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2881 {
2882         int err;
2883
2884         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2885         if (!steering->rdma_rx_root_ns)
2886                 return -ENOMEM;
2887
2888         err = init_root_tree(steering, &rdma_rx_root_fs,
2889                              &steering->rdma_rx_root_ns->ns.node);
2890         if (err)
2891                 goto out_err;
2892
2893         set_prio_attrs(steering->rdma_rx_root_ns);
2894
2895         return 0;
2896
2897 out_err:
2898         cleanup_root_ns(steering->rdma_rx_root_ns);
2899         steering->rdma_rx_root_ns = NULL;
2900         return err;
2901 }
2902
2903 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2904 {
2905         int err;
2906
2907         steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2908         if (!steering->rdma_tx_root_ns)
2909                 return -ENOMEM;
2910
2911         err = init_root_tree(steering, &rdma_tx_root_fs,
2912                              &steering->rdma_tx_root_ns->ns.node);
2913         if (err)
2914                 goto out_err;
2915
2916         set_prio_attrs(steering->rdma_tx_root_ns);
2917
2918         return 0;
2919
2920 out_err:
2921         cleanup_root_ns(steering->rdma_tx_root_ns);
2922         steering->rdma_tx_root_ns = NULL;
2923         return err;
2924 }
2925
2926 /* FT and tc chains are stored in the same array so we can re-use the
2927  * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2928  * When creating a new ns for each chain store it in the first available slot.
2929  * Assume tc chains are created and stored first and only then the FT chain.
2930  */
2931 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2932                                         struct mlx5_flow_namespace *ns)
2933 {
2934         int chain = 0;
2935
2936         while (steering->fdb_sub_ns[chain])
2937                 ++chain;
2938
2939         steering->fdb_sub_ns[chain] = ns;
2940 }
2941
2942 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2943                                         struct fs_prio *maj_prio)
2944 {
2945         struct mlx5_flow_namespace *ns;
2946         struct fs_prio *min_prio;
2947         int prio;
2948
2949         ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2950         if (IS_ERR(ns))
2951                 return PTR_ERR(ns);
2952
2953         for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2954                 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2955                 if (IS_ERR(min_prio))
2956                         return PTR_ERR(min_prio);
2957         }
2958
2959         store_fdb_sub_ns_prio_chain(steering, ns);
2960
2961         return 0;
2962 }
2963
2964 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2965                              int fs_prio,
2966                              int chains)
2967 {
2968         struct fs_prio *maj_prio;
2969         int levels;
2970         int chain;
2971         int err;
2972
2973         levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2974         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2975                                           fs_prio,
2976                                           levels);
2977         if (IS_ERR(maj_prio))
2978                 return PTR_ERR(maj_prio);
2979
2980         for (chain = 0; chain < chains; chain++) {
2981                 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2982                 if (err)
2983                         return err;
2984         }
2985
2986         return 0;
2987 }
2988
2989 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2990 {
2991         int err;
2992
2993         steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2994                                        sizeof(*steering->fdb_sub_ns),
2995                                        GFP_KERNEL);
2996         if (!steering->fdb_sub_ns)
2997                 return -ENOMEM;
2998
2999         err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
3000         if (err)
3001                 return err;
3002
3003         err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
3004         if (err)
3005                 return err;
3006
3007         return 0;
3008 }
3009
3010 static int create_fdb_bypass(struct mlx5_flow_steering *steering)
3011 {
3012         struct mlx5_flow_namespace *ns;
3013         struct fs_prio *prio;
3014         int i;
3015
3016         prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH, 0);
3017         if (IS_ERR(prio))
3018                 return PTR_ERR(prio);
3019
3020         ns = fs_create_namespace(prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
3021         if (IS_ERR(ns))
3022                 return PTR_ERR(ns);
3023
3024         for (i = 0; i < MLX5_BY_PASS_NUM_REGULAR_PRIOS; i++) {
3025                 prio = fs_create_prio(ns, i, 1);
3026                 if (IS_ERR(prio))
3027                         return PTR_ERR(prio);
3028         }
3029         return 0;
3030 }
3031
3032 static void cleanup_fdb_root_ns(struct mlx5_flow_steering *steering)
3033 {
3034         cleanup_root_ns(steering->fdb_root_ns);
3035         steering->fdb_root_ns = NULL;
3036         kfree(steering->fdb_sub_ns);
3037         steering->fdb_sub_ns = NULL;
3038 }
3039
3040 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
3041 {
3042         struct fs_prio *maj_prio;
3043         int err;
3044
3045         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
3046         if (!steering->fdb_root_ns)
3047                 return -ENOMEM;
3048
3049         err = create_fdb_bypass(steering);
3050         if (err)
3051                 goto out_err;
3052
3053         err = create_fdb_fast_path(steering);
3054         if (err)
3055                 goto out_err;
3056
3057         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1);
3058         if (IS_ERR(maj_prio)) {
3059                 err = PTR_ERR(maj_prio);
3060                 goto out_err;
3061         }
3062
3063         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 4);
3064         if (IS_ERR(maj_prio)) {
3065                 err = PTR_ERR(maj_prio);
3066                 goto out_err;
3067         }
3068
3069         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
3070         if (IS_ERR(maj_prio)) {
3071                 err = PTR_ERR(maj_prio);
3072                 goto out_err;
3073         }
3074
3075         /* We put this priority last, knowing that nothing will get here
3076          * unless explicitly forwarded to. This is possible because the
3077          * slow path tables have catch all rules and nothing gets passed
3078          * those tables.
3079          */
3080         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
3081         if (IS_ERR(maj_prio)) {
3082                 err = PTR_ERR(maj_prio);
3083                 goto out_err;
3084         }
3085
3086         set_prio_attrs(steering->fdb_root_ns);
3087         return 0;
3088
3089 out_err:
3090         cleanup_fdb_root_ns(steering);
3091         return err;
3092 }
3093
3094 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
3095 {
3096         struct fs_prio *prio;
3097
3098         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
3099         if (!steering->esw_egress_root_ns[vport])
3100                 return -ENOMEM;
3101
3102         /* create 1 prio*/
3103         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
3104         return PTR_ERR_OR_ZERO(prio);
3105 }
3106
3107 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
3108 {
3109         struct fs_prio *prio;
3110
3111         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
3112         if (!steering->esw_ingress_root_ns[vport])
3113                 return -ENOMEM;
3114
3115         /* create 1 prio*/
3116         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
3117         return PTR_ERR_OR_ZERO(prio);
3118 }
3119
3120 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports)
3121 {
3122         struct mlx5_flow_steering *steering = dev->priv.steering;
3123         int err;
3124         int i;
3125
3126         steering->esw_egress_root_ns =
3127                         kcalloc(total_vports,
3128                                 sizeof(*steering->esw_egress_root_ns),
3129                                 GFP_KERNEL);
3130         if (!steering->esw_egress_root_ns)
3131                 return -ENOMEM;
3132
3133         for (i = 0; i < total_vports; i++) {
3134                 err = init_egress_acl_root_ns(steering, i);
3135                 if (err)
3136                         goto cleanup_root_ns;
3137         }
3138         steering->esw_egress_acl_vports = total_vports;
3139         return 0;
3140
3141 cleanup_root_ns:
3142         for (i--; i >= 0; i--)
3143                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
3144         kfree(steering->esw_egress_root_ns);
3145         steering->esw_egress_root_ns = NULL;
3146         return err;
3147 }
3148
3149 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev)
3150 {
3151         struct mlx5_flow_steering *steering = dev->priv.steering;
3152         int i;
3153
3154         if (!steering->esw_egress_root_ns)
3155                 return;
3156
3157         for (i = 0; i < steering->esw_egress_acl_vports; i++)
3158                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
3159
3160         kfree(steering->esw_egress_root_ns);
3161         steering->esw_egress_root_ns = NULL;
3162 }
3163
3164 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports)
3165 {
3166         struct mlx5_flow_steering *steering = dev->priv.steering;
3167         int err;
3168         int i;
3169
3170         steering->esw_ingress_root_ns =
3171                         kcalloc(total_vports,
3172                                 sizeof(*steering->esw_ingress_root_ns),
3173                                 GFP_KERNEL);
3174         if (!steering->esw_ingress_root_ns)
3175                 return -ENOMEM;
3176
3177         for (i = 0; i < total_vports; i++) {
3178                 err = init_ingress_acl_root_ns(steering, i);
3179                 if (err)
3180                         goto cleanup_root_ns;
3181         }
3182         steering->esw_ingress_acl_vports = total_vports;
3183         return 0;
3184
3185 cleanup_root_ns:
3186         for (i--; i >= 0; i--)
3187                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
3188         kfree(steering->esw_ingress_root_ns);
3189         steering->esw_ingress_root_ns = NULL;
3190         return err;
3191 }
3192
3193 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev)
3194 {
3195         struct mlx5_flow_steering *steering = dev->priv.steering;
3196         int i;
3197
3198         if (!steering->esw_ingress_root_ns)
3199                 return;
3200
3201         for (i = 0; i < steering->esw_ingress_acl_vports; i++)
3202                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
3203
3204         kfree(steering->esw_ingress_root_ns);
3205         steering->esw_ingress_root_ns = NULL;
3206 }
3207
3208 u32 mlx5_fs_get_capabilities(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type type)
3209 {
3210         struct mlx5_flow_root_namespace *root;
3211         struct mlx5_flow_namespace *ns;
3212
3213         ns = mlx5_get_flow_namespace(dev, type);
3214         if (!ns)
3215                 return 0;
3216
3217         root = find_root(&ns->node);
3218         if (!root)
3219                 return 0;
3220
3221         return root->cmds->get_capabilities(root, root->table_type);
3222 }
3223
3224 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
3225 {
3226         int err;
3227
3228         steering->egress_root_ns = create_root_ns(steering,
3229                                                   FS_FT_NIC_TX);
3230         if (!steering->egress_root_ns)
3231                 return -ENOMEM;
3232
3233         err = init_root_tree(steering, &egress_root_fs,
3234                              &steering->egress_root_ns->ns.node);
3235         if (err)
3236                 goto cleanup;
3237         set_prio_attrs(steering->egress_root_ns);
3238         return 0;
3239 cleanup:
3240         cleanup_root_ns(steering->egress_root_ns);
3241         steering->egress_root_ns = NULL;
3242         return err;
3243 }
3244
3245 static int mlx5_fs_mode_validate(struct devlink *devlink, u32 id,
3246                                  union devlink_param_value val,
3247                                  struct netlink_ext_ack *extack)
3248 {
3249         struct mlx5_core_dev *dev = devlink_priv(devlink);
3250         char *value = val.vstr;
3251         int err = 0;
3252
3253         if (!strcmp(value, "dmfs")) {
3254                 return 0;
3255         } else if (!strcmp(value, "smfs")) {
3256                 u8 eswitch_mode;
3257                 bool smfs_cap;
3258
3259                 eswitch_mode = mlx5_eswitch_mode(dev);
3260                 smfs_cap = mlx5_fs_dr_is_supported(dev);
3261
3262                 if (!smfs_cap) {
3263                         err = -EOPNOTSUPP;
3264                         NL_SET_ERR_MSG_MOD(extack,
3265                                            "Software managed steering is not supported by current device");
3266                 }
3267
3268                 else if (eswitch_mode == MLX5_ESWITCH_OFFLOADS) {
3269                         NL_SET_ERR_MSG_MOD(extack,
3270                                            "Software managed steering is not supported when eswitch offloads enabled.");
3271                         err = -EOPNOTSUPP;
3272                 }
3273         } else {
3274                 NL_SET_ERR_MSG_MOD(extack,
3275                                    "Bad parameter: supported values are [\"dmfs\", \"smfs\"]");
3276                 err = -EINVAL;
3277         }
3278
3279         return err;
3280 }
3281
3282 static int mlx5_fs_mode_set(struct devlink *devlink, u32 id,
3283                             struct devlink_param_gset_ctx *ctx)
3284 {
3285         struct mlx5_core_dev *dev = devlink_priv(devlink);
3286         enum mlx5_flow_steering_mode mode;
3287
3288         if (!strcmp(ctx->val.vstr, "smfs"))
3289                 mode = MLX5_FLOW_STEERING_MODE_SMFS;
3290         else
3291                 mode = MLX5_FLOW_STEERING_MODE_DMFS;
3292         dev->priv.steering->mode = mode;
3293
3294         return 0;
3295 }
3296
3297 static int mlx5_fs_mode_get(struct devlink *devlink, u32 id,
3298                             struct devlink_param_gset_ctx *ctx)
3299 {
3300         struct mlx5_core_dev *dev = devlink_priv(devlink);
3301
3302         if (dev->priv.steering->mode == MLX5_FLOW_STEERING_MODE_SMFS)
3303                 strcpy(ctx->val.vstr, "smfs");
3304         else
3305                 strcpy(ctx->val.vstr, "dmfs");
3306         return 0;
3307 }
3308
3309 static const struct devlink_param mlx5_fs_params[] = {
3310         DEVLINK_PARAM_DRIVER(MLX5_DEVLINK_PARAM_ID_FLOW_STEERING_MODE,
3311                              "flow_steering_mode", DEVLINK_PARAM_TYPE_STRING,
3312                              BIT(DEVLINK_PARAM_CMODE_RUNTIME),
3313                              mlx5_fs_mode_get, mlx5_fs_mode_set,
3314                              mlx5_fs_mode_validate),
3315 };
3316
3317 void mlx5_fs_core_cleanup(struct mlx5_core_dev *dev)
3318 {
3319         struct mlx5_flow_steering *steering = dev->priv.steering;
3320
3321         cleanup_root_ns(steering->root_ns);
3322         cleanup_fdb_root_ns(steering);
3323         cleanup_root_ns(steering->port_sel_root_ns);
3324         cleanup_root_ns(steering->sniffer_rx_root_ns);
3325         cleanup_root_ns(steering->sniffer_tx_root_ns);
3326         cleanup_root_ns(steering->rdma_rx_root_ns);
3327         cleanup_root_ns(steering->rdma_tx_root_ns);
3328         cleanup_root_ns(steering->egress_root_ns);
3329
3330         devl_params_unregister(priv_to_devlink(dev), mlx5_fs_params,
3331                                ARRAY_SIZE(mlx5_fs_params));
3332 }
3333
3334 int mlx5_fs_core_init(struct mlx5_core_dev *dev)
3335 {
3336         struct mlx5_flow_steering *steering = dev->priv.steering;
3337         int err;
3338
3339         err = devl_params_register(priv_to_devlink(dev), mlx5_fs_params,
3340                                    ARRAY_SIZE(mlx5_fs_params));
3341         if (err)
3342                 return err;
3343
3344         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
3345               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
3346              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
3347               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
3348             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
3349                 err = init_root_ns(steering);
3350                 if (err)
3351                         goto err;
3352         }
3353
3354         if (MLX5_ESWITCH_MANAGER(dev)) {
3355                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
3356                         err = init_fdb_root_ns(steering);
3357                         if (err)
3358                                 goto err;
3359                 }
3360         }
3361
3362         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
3363                 err = init_sniffer_rx_root_ns(steering);
3364                 if (err)
3365                         goto err;
3366         }
3367
3368         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
3369                 err = init_sniffer_tx_root_ns(steering);
3370                 if (err)
3371                         goto err;
3372         }
3373
3374         if (MLX5_CAP_FLOWTABLE_PORT_SELECTION(dev, ft_support)) {
3375                 err = init_port_sel_root_ns(steering);
3376                 if (err)
3377                         goto err;
3378         }
3379
3380         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
3381             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3382                 err = init_rdma_rx_root_ns(steering);
3383                 if (err)
3384                         goto err;
3385         }
3386
3387         if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3388                 err = init_rdma_tx_root_ns(steering);
3389                 if (err)
3390                         goto err;
3391         }
3392
3393         if (MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3394                 err = init_egress_root_ns(steering);
3395                 if (err)
3396                         goto err;
3397         }
3398
3399         return 0;
3400
3401 err:
3402         mlx5_fs_core_cleanup(dev);
3403         return err;
3404 }
3405
3406 void mlx5_fs_core_free(struct mlx5_core_dev *dev)
3407 {
3408         struct mlx5_flow_steering *steering = dev->priv.steering;
3409
3410         kmem_cache_destroy(steering->ftes_cache);
3411         kmem_cache_destroy(steering->fgs_cache);
3412         kfree(steering);
3413         mlx5_ft_pool_destroy(dev);
3414         mlx5_cleanup_fc_stats(dev);
3415 }
3416
3417 int mlx5_fs_core_alloc(struct mlx5_core_dev *dev)
3418 {
3419         struct mlx5_flow_steering *steering;
3420         int err = 0;
3421
3422         err = mlx5_init_fc_stats(dev);
3423         if (err)
3424                 return err;
3425
3426         err = mlx5_ft_pool_init(dev);
3427         if (err)
3428                 goto err;
3429
3430         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
3431         if (!steering) {
3432                 err = -ENOMEM;
3433                 goto err;
3434         }
3435
3436         steering->dev = dev;
3437         dev->priv.steering = steering;
3438
3439         if (mlx5_fs_dr_is_supported(dev))
3440                 steering->mode = MLX5_FLOW_STEERING_MODE_SMFS;
3441         else
3442                 steering->mode = MLX5_FLOW_STEERING_MODE_DMFS;
3443
3444         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
3445                                                 sizeof(struct mlx5_flow_group), 0,
3446                                                 0, NULL);
3447         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
3448                                                  0, NULL);
3449         if (!steering->ftes_cache || !steering->fgs_cache) {
3450                 err = -ENOMEM;
3451                 goto err;
3452         }
3453
3454         return 0;
3455
3456 err:
3457         mlx5_fs_core_free(dev);
3458         return err;
3459 }
3460
3461 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3462 {
3463         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3464         struct mlx5_ft_underlay_qp *new_uqp;
3465         int err = 0;
3466
3467         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3468         if (!new_uqp)
3469                 return -ENOMEM;
3470
3471         mutex_lock(&root->chain_lock);
3472
3473         if (!root->root_ft) {
3474                 err = -EINVAL;
3475                 goto update_ft_fail;
3476         }
3477
3478         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3479                                          false);
3480         if (err) {
3481                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3482                                underlay_qpn, err);
3483                 goto update_ft_fail;
3484         }
3485
3486         new_uqp->qpn = underlay_qpn;
3487         list_add_tail(&new_uqp->list, &root->underlay_qpns);
3488
3489         mutex_unlock(&root->chain_lock);
3490
3491         return 0;
3492
3493 update_ft_fail:
3494         mutex_unlock(&root->chain_lock);
3495         kfree(new_uqp);
3496         return err;
3497 }
3498 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3499
3500 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3501 {
3502         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3503         struct mlx5_ft_underlay_qp *uqp;
3504         bool found = false;
3505         int err = 0;
3506
3507         mutex_lock(&root->chain_lock);
3508         list_for_each_entry(uqp, &root->underlay_qpns, list) {
3509                 if (uqp->qpn == underlay_qpn) {
3510                         found = true;
3511                         break;
3512                 }
3513         }
3514
3515         if (!found) {
3516                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3517                                underlay_qpn);
3518                 err = -EINVAL;
3519                 goto out;
3520         }
3521
3522         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3523                                          true);
3524         if (err)
3525                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3526                                underlay_qpn, err);
3527
3528         list_del(&uqp->list);
3529         mutex_unlock(&root->chain_lock);
3530         kfree(uqp);
3531
3532         return 0;
3533
3534 out:
3535         mutex_unlock(&root->chain_lock);
3536         return err;
3537 }
3538 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3539
3540 static struct mlx5_flow_root_namespace
3541 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3542 {
3543         struct mlx5_flow_namespace *ns;
3544
3545         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3546             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3547                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3548         else
3549                 ns = mlx5_get_flow_namespace(dev, ns_type);
3550         if (!ns)
3551                 return NULL;
3552
3553         return find_root(&ns->node);
3554 }
3555
3556 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3557                                                  u8 ns_type, u8 num_actions,
3558                                                  void *modify_actions)
3559 {
3560         struct mlx5_flow_root_namespace *root;
3561         struct mlx5_modify_hdr *modify_hdr;
3562         int err;
3563
3564         root = get_root_namespace(dev, ns_type);
3565         if (!root)
3566                 return ERR_PTR(-EOPNOTSUPP);
3567
3568         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3569         if (!modify_hdr)
3570                 return ERR_PTR(-ENOMEM);
3571
3572         modify_hdr->ns_type = ns_type;
3573         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3574                                               modify_actions, modify_hdr);
3575         if (err) {
3576                 kfree(modify_hdr);
3577                 return ERR_PTR(err);
3578         }
3579
3580         return modify_hdr;
3581 }
3582 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3583
3584 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3585                                 struct mlx5_modify_hdr *modify_hdr)
3586 {
3587         struct mlx5_flow_root_namespace *root;
3588
3589         root = get_root_namespace(dev, modify_hdr->ns_type);
3590         if (WARN_ON(!root))
3591                 return;
3592         root->cmds->modify_header_dealloc(root, modify_hdr);
3593         kfree(modify_hdr);
3594 }
3595 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3596
3597 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3598                                                      struct mlx5_pkt_reformat_params *params,
3599                                                      enum mlx5_flow_namespace_type ns_type)
3600 {
3601         struct mlx5_pkt_reformat *pkt_reformat;
3602         struct mlx5_flow_root_namespace *root;
3603         int err;
3604
3605         root = get_root_namespace(dev, ns_type);
3606         if (!root)
3607                 return ERR_PTR(-EOPNOTSUPP);
3608
3609         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3610         if (!pkt_reformat)
3611                 return ERR_PTR(-ENOMEM);
3612
3613         pkt_reformat->ns_type = ns_type;
3614         pkt_reformat->reformat_type = params->type;
3615         err = root->cmds->packet_reformat_alloc(root, params, ns_type,
3616                                                 pkt_reformat);
3617         if (err) {
3618                 kfree(pkt_reformat);
3619                 return ERR_PTR(err);
3620         }
3621
3622         return pkt_reformat;
3623 }
3624 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3625
3626 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3627                                   struct mlx5_pkt_reformat *pkt_reformat)
3628 {
3629         struct mlx5_flow_root_namespace *root;
3630
3631         root = get_root_namespace(dev, pkt_reformat->ns_type);
3632         if (WARN_ON(!root))
3633                 return;
3634         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3635         kfree(pkt_reformat);
3636 }
3637 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3638
3639 int mlx5_get_match_definer_id(struct mlx5_flow_definer *definer)
3640 {
3641         return definer->id;
3642 }
3643
3644 struct mlx5_flow_definer *
3645 mlx5_create_match_definer(struct mlx5_core_dev *dev,
3646                           enum mlx5_flow_namespace_type ns_type, u16 format_id,
3647                           u32 *match_mask)
3648 {
3649         struct mlx5_flow_root_namespace *root;
3650         struct mlx5_flow_definer *definer;
3651         int id;
3652
3653         root = get_root_namespace(dev, ns_type);
3654         if (!root)
3655                 return ERR_PTR(-EOPNOTSUPP);
3656
3657         definer = kzalloc(sizeof(*definer), GFP_KERNEL);
3658         if (!definer)
3659                 return ERR_PTR(-ENOMEM);
3660
3661         definer->ns_type = ns_type;
3662         id = root->cmds->create_match_definer(root, format_id, match_mask);
3663         if (id < 0) {
3664                 mlx5_core_warn(root->dev, "Failed to create match definer (%d)\n", id);
3665                 kfree(definer);
3666                 return ERR_PTR(id);
3667         }
3668         definer->id = id;
3669         return definer;
3670 }
3671
3672 void mlx5_destroy_match_definer(struct mlx5_core_dev *dev,
3673                                 struct mlx5_flow_definer *definer)
3674 {
3675         struct mlx5_flow_root_namespace *root;
3676
3677         root = get_root_namespace(dev, definer->ns_type);
3678         if (WARN_ON(!root))
3679                 return;
3680
3681         root->cmds->destroy_match_definer(root, definer->id);
3682         kfree(definer);
3683 }
3684
3685 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3686                                  struct mlx5_flow_root_namespace *peer_ns,
3687                                  u16 peer_vhca_id)
3688 {
3689         if (peer_ns && ns->mode != peer_ns->mode) {
3690                 mlx5_core_err(ns->dev,
3691                               "Can't peer namespace of different steering mode\n");
3692                 return -EINVAL;
3693         }
3694
3695         return ns->cmds->set_peer(ns, peer_ns, peer_vhca_id);
3696 }
3697
3698 /* This function should be called only at init stage of the namespace.
3699  * It is not safe to call this function while steering operations
3700  * are executed in the namespace.
3701  */
3702 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3703                                  enum mlx5_flow_steering_mode mode)
3704 {
3705         struct mlx5_flow_root_namespace *root;
3706         const struct mlx5_flow_cmds *cmds;
3707         int err;
3708
3709         root = find_root(&ns->node);
3710         if (&root->ns != ns)
3711         /* Can't set cmds to non root namespace */
3712                 return -EINVAL;
3713
3714         if (root->table_type != FS_FT_FDB)
3715                 return -EOPNOTSUPP;
3716
3717         if (root->mode == mode)
3718                 return 0;
3719
3720         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3721                 cmds = mlx5_fs_cmd_get_dr_cmds();
3722         else
3723                 cmds = mlx5_fs_cmd_get_fw_cmds();
3724         if (!cmds)
3725                 return -EOPNOTSUPP;
3726
3727         err = cmds->create_ns(root);
3728         if (err) {
3729                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3730                               err);
3731                 return err;
3732         }
3733
3734         root->cmds->destroy_ns(root);
3735         root->cmds = cmds;
3736         root->mode = mode;
3737
3738         return 0;
3739 }