1 /* 2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 */ 32 33 #include <linux/mutex.h> 34 #include <linux/mlx5/driver.h> 35 #include <linux/mlx5/vport.h> 36 #include <linux/mlx5/eswitch.h> 37 38 #include "mlx5_core.h" 39 #include "fs_core.h" 40 #include "fs_cmd.h" 41 #include "fs_ft_pool.h" 42 #include "diag/fs_tracepoint.h" 43 44 #define INIT_TREE_NODE_ARRAY_SIZE(...) (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\ 45 sizeof(struct init_tree_node)) 46 47 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\ 48 ...) {.type = FS_TYPE_PRIO,\ 49 .min_ft_level = min_level_val,\ 50 .num_levels = num_levels_val,\ 51 .num_leaf_prios = num_prios_val,\ 52 .caps = caps_val,\ 53 .children = (struct init_tree_node[]) {__VA_ARGS__},\ 54 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \ 55 } 56 57 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\ 58 ADD_PRIO(num_prios_val, 0, num_levels_val, {},\ 59 __VA_ARGS__)\ 60 61 #define ADD_NS(def_miss_act, ...) {.type = FS_TYPE_NAMESPACE, \ 62 .def_miss_action = def_miss_act,\ 63 .children = (struct init_tree_node[]) {__VA_ARGS__},\ 64 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \ 65 } 66 67 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\ 68 sizeof(long)) 69 70 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap)) 71 72 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \ 73 .caps = (long[]) {__VA_ARGS__} } 74 75 #define FS_CHAINING_CAPS FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \ 76 FS_CAP(flow_table_properties_nic_receive.modify_root), \ 77 FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \ 78 FS_CAP(flow_table_properties_nic_receive.flow_table_modify)) 79 80 #define FS_CHAINING_CAPS_EGRESS \ 81 FS_REQUIRED_CAPS( \ 82 FS_CAP(flow_table_properties_nic_transmit.flow_modify_en), \ 83 FS_CAP(flow_table_properties_nic_transmit.modify_root), \ 84 FS_CAP(flow_table_properties_nic_transmit \ 85 .identified_miss_table_mode), \ 86 FS_CAP(flow_table_properties_nic_transmit.flow_table_modify)) 87 88 #define FS_CHAINING_CAPS_RDMA_TX \ 89 FS_REQUIRED_CAPS( \ 90 FS_CAP(flow_table_properties_nic_transmit_rdma.flow_modify_en), \ 91 FS_CAP(flow_table_properties_nic_transmit_rdma.modify_root), \ 92 FS_CAP(flow_table_properties_nic_transmit_rdma \ 93 .identified_miss_table_mode), \ 94 FS_CAP(flow_table_properties_nic_transmit_rdma \ 95 .flow_table_modify)) 96 97 #define LEFTOVERS_NUM_LEVELS 1 98 #define LEFTOVERS_NUM_PRIOS 1 99 100 #define RDMA_RX_COUNTERS_PRIO_NUM_LEVELS 1 101 #define RDMA_TX_COUNTERS_PRIO_NUM_LEVELS 1 102 103 #define BY_PASS_PRIO_NUM_LEVELS 1 104 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\ 105 LEFTOVERS_NUM_PRIOS) 106 107 #define KERNEL_RX_MACSEC_NUM_PRIOS 1 108 #define KERNEL_RX_MACSEC_NUM_LEVELS 2 109 #define KERNEL_RX_MACSEC_MIN_LEVEL (BY_PASS_MIN_LEVEL + KERNEL_RX_MACSEC_NUM_PRIOS) 110 111 #define ETHTOOL_PRIO_NUM_LEVELS 1 112 #define ETHTOOL_NUM_PRIOS 11 113 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS) 114 /* Promiscuous, Vlan, mac, ttc, inner ttc, {UDP/ANY/aRFS/accel/{esp, esp_err}} */ 115 #define KERNEL_NIC_PRIO_NUM_LEVELS 7 116 #define KERNEL_NIC_NUM_PRIOS 1 117 /* One more level for tc */ 118 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1) 119 120 #define KERNEL_NIC_TC_NUM_PRIOS 1 121 #define KERNEL_NIC_TC_NUM_LEVELS 3 122 123 #define ANCHOR_NUM_LEVELS 1 124 #define ANCHOR_NUM_PRIOS 1 125 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1) 126 127 #define OFFLOADS_MAX_FT 2 128 #define OFFLOADS_NUM_PRIOS 2 129 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS) 130 131 #define LAG_PRIO_NUM_LEVELS 1 132 #define LAG_NUM_PRIOS 1 133 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + KERNEL_RX_MACSEC_MIN_LEVEL + 1) 134 135 #define KERNEL_TX_IPSEC_NUM_PRIOS 1 136 #define KERNEL_TX_IPSEC_NUM_LEVELS 1 137 #define KERNEL_TX_IPSEC_MIN_LEVEL (KERNEL_TX_IPSEC_NUM_LEVELS) 138 139 #define KERNEL_TX_MACSEC_NUM_PRIOS 1 140 #define KERNEL_TX_MACSEC_NUM_LEVELS 2 141 #define KERNEL_TX_MACSEC_MIN_LEVEL (KERNEL_TX_IPSEC_MIN_LEVEL + KERNEL_TX_MACSEC_NUM_PRIOS) 142 143 struct node_caps { 144 size_t arr_sz; 145 long *caps; 146 }; 147 148 static struct init_tree_node { 149 enum fs_node_type type; 150 struct init_tree_node *children; 151 int ar_size; 152 struct node_caps caps; 153 int min_ft_level; 154 int num_leaf_prios; 155 int prio; 156 int num_levels; 157 enum mlx5_flow_table_miss_action def_miss_action; 158 } root_fs = { 159 .type = FS_TYPE_NAMESPACE, 160 .ar_size = 8, 161 .children = (struct init_tree_node[]){ 162 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS, 163 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 164 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS, 165 BY_PASS_PRIO_NUM_LEVELS))), 166 ADD_PRIO(0, KERNEL_RX_MACSEC_MIN_LEVEL, 0, FS_CHAINING_CAPS, 167 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 168 ADD_MULTIPLE_PRIO(KERNEL_RX_MACSEC_NUM_PRIOS, 169 KERNEL_RX_MACSEC_NUM_LEVELS))), 170 ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS, 171 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 172 ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS, 173 LAG_PRIO_NUM_LEVELS))), 174 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS, 175 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 176 ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, 177 OFFLOADS_MAX_FT))), 178 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS, 179 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 180 ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS, 181 ETHTOOL_PRIO_NUM_LEVELS))), 182 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {}, 183 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 184 ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS, 185 KERNEL_NIC_TC_NUM_LEVELS), 186 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS, 187 KERNEL_NIC_PRIO_NUM_LEVELS))), 188 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS, 189 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 190 ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, 191 LEFTOVERS_NUM_LEVELS))), 192 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {}, 193 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 194 ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, 195 ANCHOR_NUM_LEVELS))), 196 } 197 }; 198 199 static struct init_tree_node egress_root_fs = { 200 .type = FS_TYPE_NAMESPACE, 201 .ar_size = 3, 202 .children = (struct init_tree_node[]) { 203 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0, 204 FS_CHAINING_CAPS_EGRESS, 205 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 206 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS, 207 BY_PASS_PRIO_NUM_LEVELS))), 208 ADD_PRIO(0, KERNEL_TX_IPSEC_MIN_LEVEL, 0, 209 FS_CHAINING_CAPS_EGRESS, 210 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 211 ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS, 212 KERNEL_TX_IPSEC_NUM_LEVELS))), 213 ADD_PRIO(0, KERNEL_TX_MACSEC_MIN_LEVEL, 0, 214 FS_CHAINING_CAPS_EGRESS, 215 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 216 ADD_MULTIPLE_PRIO(KERNEL_TX_MACSEC_NUM_PRIOS, 217 KERNEL_TX_MACSEC_NUM_LEVELS))), 218 } 219 }; 220 221 enum { 222 RDMA_RX_COUNTERS_PRIO, 223 RDMA_RX_BYPASS_PRIO, 224 RDMA_RX_KERNEL_PRIO, 225 }; 226 227 #define RDMA_RX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_REGULAR_PRIOS 228 #define RDMA_RX_KERNEL_MIN_LEVEL (RDMA_RX_BYPASS_MIN_LEVEL + 1) 229 #define RDMA_RX_COUNTERS_MIN_LEVEL (RDMA_RX_KERNEL_MIN_LEVEL + 2) 230 231 static struct init_tree_node rdma_rx_root_fs = { 232 .type = FS_TYPE_NAMESPACE, 233 .ar_size = 3, 234 .children = (struct init_tree_node[]) { 235 [RDMA_RX_COUNTERS_PRIO] = 236 ADD_PRIO(0, RDMA_RX_COUNTERS_MIN_LEVEL, 0, 237 FS_CHAINING_CAPS, 238 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 239 ADD_MULTIPLE_PRIO(MLX5_RDMA_RX_NUM_COUNTERS_PRIOS, 240 RDMA_RX_COUNTERS_PRIO_NUM_LEVELS))), 241 [RDMA_RX_BYPASS_PRIO] = 242 ADD_PRIO(0, RDMA_RX_BYPASS_MIN_LEVEL, 0, 243 FS_CHAINING_CAPS, 244 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 245 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS, 246 BY_PASS_PRIO_NUM_LEVELS))), 247 [RDMA_RX_KERNEL_PRIO] = 248 ADD_PRIO(0, RDMA_RX_KERNEL_MIN_LEVEL, 0, 249 FS_CHAINING_CAPS, 250 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN, 251 ADD_MULTIPLE_PRIO(1, 1))), 252 } 253 }; 254 255 enum { 256 RDMA_TX_COUNTERS_PRIO, 257 RDMA_TX_BYPASS_PRIO, 258 }; 259 260 #define RDMA_TX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_PRIOS 261 #define RDMA_TX_COUNTERS_MIN_LEVEL (RDMA_TX_BYPASS_MIN_LEVEL + 1) 262 263 static struct init_tree_node rdma_tx_root_fs = { 264 .type = FS_TYPE_NAMESPACE, 265 .ar_size = 2, 266 .children = (struct init_tree_node[]) { 267 [RDMA_TX_COUNTERS_PRIO] = 268 ADD_PRIO(0, RDMA_TX_COUNTERS_MIN_LEVEL, 0, 269 FS_CHAINING_CAPS, 270 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 271 ADD_MULTIPLE_PRIO(MLX5_RDMA_TX_NUM_COUNTERS_PRIOS, 272 RDMA_TX_COUNTERS_PRIO_NUM_LEVELS))), 273 [RDMA_TX_BYPASS_PRIO] = 274 ADD_PRIO(0, RDMA_TX_BYPASS_MIN_LEVEL, 0, 275 FS_CHAINING_CAPS_RDMA_TX, 276 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 277 ADD_MULTIPLE_PRIO(RDMA_TX_BYPASS_MIN_LEVEL, 278 BY_PASS_PRIO_NUM_LEVELS))), 279 } 280 }; 281 282 enum fs_i_lock_class { 283 FS_LOCK_GRANDPARENT, 284 FS_LOCK_PARENT, 285 FS_LOCK_CHILD 286 }; 287 288 static const struct rhashtable_params rhash_fte = { 289 .key_len = sizeof_field(struct fs_fte, val), 290 .key_offset = offsetof(struct fs_fte, val), 291 .head_offset = offsetof(struct fs_fte, hash), 292 .automatic_shrinking = true, 293 .min_size = 1, 294 }; 295 296 static const struct rhashtable_params rhash_fg = { 297 .key_len = sizeof_field(struct mlx5_flow_group, mask), 298 .key_offset = offsetof(struct mlx5_flow_group, mask), 299 .head_offset = offsetof(struct mlx5_flow_group, hash), 300 .automatic_shrinking = true, 301 .min_size = 1, 302 303 }; 304 305 static void del_hw_flow_table(struct fs_node *node); 306 static void del_hw_flow_group(struct fs_node *node); 307 static void del_hw_fte(struct fs_node *node); 308 static void del_sw_flow_table(struct fs_node *node); 309 static void del_sw_flow_group(struct fs_node *node); 310 static void del_sw_fte(struct fs_node *node); 311 static void del_sw_prio(struct fs_node *node); 312 static void del_sw_ns(struct fs_node *node); 313 /* Delete rule (destination) is special case that 314 * requires to lock the FTE for all the deletion process. 315 */ 316 static void del_sw_hw_rule(struct fs_node *node); 317 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1, 318 struct mlx5_flow_destination *d2); 319 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns); 320 static struct mlx5_flow_rule * 321 find_flow_rule(struct fs_fte *fte, 322 struct mlx5_flow_destination *dest); 323 324 static void tree_init_node(struct fs_node *node, 325 void (*del_hw_func)(struct fs_node *), 326 void (*del_sw_func)(struct fs_node *)) 327 { 328 refcount_set(&node->refcount, 1); 329 INIT_LIST_HEAD(&node->list); 330 INIT_LIST_HEAD(&node->children); 331 init_rwsem(&node->lock); 332 node->del_hw_func = del_hw_func; 333 node->del_sw_func = del_sw_func; 334 node->active = false; 335 } 336 337 static void tree_add_node(struct fs_node *node, struct fs_node *parent) 338 { 339 if (parent) 340 refcount_inc(&parent->refcount); 341 node->parent = parent; 342 343 /* Parent is the root */ 344 if (!parent) 345 node->root = node; 346 else 347 node->root = parent->root; 348 } 349 350 static int tree_get_node(struct fs_node *node) 351 { 352 return refcount_inc_not_zero(&node->refcount); 353 } 354 355 static void nested_down_read_ref_node(struct fs_node *node, 356 enum fs_i_lock_class class) 357 { 358 if (node) { 359 down_read_nested(&node->lock, class); 360 refcount_inc(&node->refcount); 361 } 362 } 363 364 static void nested_down_write_ref_node(struct fs_node *node, 365 enum fs_i_lock_class class) 366 { 367 if (node) { 368 down_write_nested(&node->lock, class); 369 refcount_inc(&node->refcount); 370 } 371 } 372 373 static void down_write_ref_node(struct fs_node *node, bool locked) 374 { 375 if (node) { 376 if (!locked) 377 down_write(&node->lock); 378 refcount_inc(&node->refcount); 379 } 380 } 381 382 static void up_read_ref_node(struct fs_node *node) 383 { 384 refcount_dec(&node->refcount); 385 up_read(&node->lock); 386 } 387 388 static void up_write_ref_node(struct fs_node *node, bool locked) 389 { 390 refcount_dec(&node->refcount); 391 if (!locked) 392 up_write(&node->lock); 393 } 394 395 static void tree_put_node(struct fs_node *node, bool locked) 396 { 397 struct fs_node *parent_node = node->parent; 398 399 if (refcount_dec_and_test(&node->refcount)) { 400 if (node->del_hw_func) 401 node->del_hw_func(node); 402 if (parent_node) { 403 down_write_ref_node(parent_node, locked); 404 list_del_init(&node->list); 405 } 406 node->del_sw_func(node); 407 if (parent_node) 408 up_write_ref_node(parent_node, locked); 409 node = NULL; 410 } 411 if (!node && parent_node) 412 tree_put_node(parent_node, locked); 413 } 414 415 static int tree_remove_node(struct fs_node *node, bool locked) 416 { 417 if (refcount_read(&node->refcount) > 1) { 418 refcount_dec(&node->refcount); 419 return -EEXIST; 420 } 421 tree_put_node(node, locked); 422 return 0; 423 } 424 425 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns, 426 unsigned int prio) 427 { 428 struct fs_prio *iter_prio; 429 430 fs_for_each_prio(iter_prio, ns) { 431 if (iter_prio->prio == prio) 432 return iter_prio; 433 } 434 435 return NULL; 436 } 437 438 static bool is_fwd_next_action(u32 action) 439 { 440 return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO | 441 MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS); 442 } 443 444 static bool is_fwd_dest_type(enum mlx5_flow_destination_type type) 445 { 446 return type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM || 447 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE || 448 type == MLX5_FLOW_DESTINATION_TYPE_UPLINK || 449 type == MLX5_FLOW_DESTINATION_TYPE_VPORT || 450 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER || 451 type == MLX5_FLOW_DESTINATION_TYPE_TIR; 452 } 453 454 static bool check_valid_spec(const struct mlx5_flow_spec *spec) 455 { 456 int i; 457 458 for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++) 459 if (spec->match_value[i] & ~spec->match_criteria[i]) { 460 pr_warn("mlx5_core: match_value differs from match_criteria\n"); 461 return false; 462 } 463 464 return true; 465 } 466 467 struct mlx5_flow_root_namespace *find_root(struct fs_node *node) 468 { 469 struct fs_node *root; 470 struct mlx5_flow_namespace *ns; 471 472 root = node->root; 473 474 if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) { 475 pr_warn("mlx5: flow steering node is not in tree or garbaged\n"); 476 return NULL; 477 } 478 479 ns = container_of(root, struct mlx5_flow_namespace, node); 480 return container_of(ns, struct mlx5_flow_root_namespace, ns); 481 } 482 483 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node) 484 { 485 struct mlx5_flow_root_namespace *root = find_root(node); 486 487 if (root) 488 return root->dev->priv.steering; 489 return NULL; 490 } 491 492 static inline struct mlx5_core_dev *get_dev(struct fs_node *node) 493 { 494 struct mlx5_flow_root_namespace *root = find_root(node); 495 496 if (root) 497 return root->dev; 498 return NULL; 499 } 500 501 static void del_sw_ns(struct fs_node *node) 502 { 503 kfree(node); 504 } 505 506 static void del_sw_prio(struct fs_node *node) 507 { 508 kfree(node); 509 } 510 511 static void del_hw_flow_table(struct fs_node *node) 512 { 513 struct mlx5_flow_root_namespace *root; 514 struct mlx5_flow_table *ft; 515 struct mlx5_core_dev *dev; 516 int err; 517 518 fs_get_obj(ft, node); 519 dev = get_dev(&ft->node); 520 root = find_root(&ft->node); 521 trace_mlx5_fs_del_ft(ft); 522 523 if (node->active) { 524 err = root->cmds->destroy_flow_table(root, ft); 525 if (err) 526 mlx5_core_warn(dev, "flow steering can't destroy ft\n"); 527 } 528 } 529 530 static void del_sw_flow_table(struct fs_node *node) 531 { 532 struct mlx5_flow_table *ft; 533 struct fs_prio *prio; 534 535 fs_get_obj(ft, node); 536 537 rhltable_destroy(&ft->fgs_hash); 538 if (ft->node.parent) { 539 fs_get_obj(prio, ft->node.parent); 540 prio->num_ft--; 541 } 542 kfree(ft); 543 } 544 545 static void modify_fte(struct fs_fte *fte) 546 { 547 struct mlx5_flow_root_namespace *root; 548 struct mlx5_flow_table *ft; 549 struct mlx5_flow_group *fg; 550 struct mlx5_core_dev *dev; 551 int err; 552 553 fs_get_obj(fg, fte->node.parent); 554 fs_get_obj(ft, fg->node.parent); 555 dev = get_dev(&fte->node); 556 557 root = find_root(&ft->node); 558 err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte); 559 if (err) 560 mlx5_core_warn(dev, 561 "%s can't del rule fg id=%d fte_index=%d\n", 562 __func__, fg->id, fte->index); 563 fte->modify_mask = 0; 564 } 565 566 static void del_sw_hw_rule(struct fs_node *node) 567 { 568 struct mlx5_flow_rule *rule; 569 struct fs_fte *fte; 570 571 fs_get_obj(rule, node); 572 fs_get_obj(fte, rule->node.parent); 573 trace_mlx5_fs_del_rule(rule); 574 if (is_fwd_next_action(rule->sw_action)) { 575 mutex_lock(&rule->dest_attr.ft->lock); 576 list_del(&rule->next_ft); 577 mutex_unlock(&rule->dest_attr.ft->lock); 578 } 579 580 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER) { 581 --fte->dests_size; 582 fte->modify_mask |= 583 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) | 584 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS); 585 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT; 586 goto out; 587 } 588 589 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT) { 590 --fte->dests_size; 591 fte->modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION); 592 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW; 593 goto out; 594 } 595 596 if (is_fwd_dest_type(rule->dest_attr.type)) { 597 --fte->dests_size; 598 --fte->fwd_dests; 599 600 if (!fte->fwd_dests) 601 fte->action.action &= 602 ~MLX5_FLOW_CONTEXT_ACTION_FWD_DEST; 603 fte->modify_mask |= 604 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST); 605 goto out; 606 } 607 out: 608 kfree(rule); 609 } 610 611 static void del_hw_fte(struct fs_node *node) 612 { 613 struct mlx5_flow_root_namespace *root; 614 struct mlx5_flow_table *ft; 615 struct mlx5_flow_group *fg; 616 struct mlx5_core_dev *dev; 617 struct fs_fte *fte; 618 int err; 619 620 fs_get_obj(fte, node); 621 fs_get_obj(fg, fte->node.parent); 622 fs_get_obj(ft, fg->node.parent); 623 624 trace_mlx5_fs_del_fte(fte); 625 WARN_ON(fte->dests_size); 626 dev = get_dev(&ft->node); 627 root = find_root(&ft->node); 628 if (node->active) { 629 err = root->cmds->delete_fte(root, ft, fte); 630 if (err) 631 mlx5_core_warn(dev, 632 "flow steering can't delete fte in index %d of flow group id %d\n", 633 fte->index, fg->id); 634 node->active = false; 635 } 636 } 637 638 static void del_sw_fte(struct fs_node *node) 639 { 640 struct mlx5_flow_steering *steering = get_steering(node); 641 struct mlx5_flow_group *fg; 642 struct fs_fte *fte; 643 int err; 644 645 fs_get_obj(fte, node); 646 fs_get_obj(fg, fte->node.parent); 647 648 err = rhashtable_remove_fast(&fg->ftes_hash, 649 &fte->hash, 650 rhash_fte); 651 WARN_ON(err); 652 ida_free(&fg->fte_allocator, fte->index - fg->start_index); 653 kmem_cache_free(steering->ftes_cache, fte); 654 } 655 656 static void del_hw_flow_group(struct fs_node *node) 657 { 658 struct mlx5_flow_root_namespace *root; 659 struct mlx5_flow_group *fg; 660 struct mlx5_flow_table *ft; 661 struct mlx5_core_dev *dev; 662 663 fs_get_obj(fg, node); 664 fs_get_obj(ft, fg->node.parent); 665 dev = get_dev(&ft->node); 666 trace_mlx5_fs_del_fg(fg); 667 668 root = find_root(&ft->node); 669 if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg)) 670 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n", 671 fg->id, ft->id); 672 } 673 674 static void del_sw_flow_group(struct fs_node *node) 675 { 676 struct mlx5_flow_steering *steering = get_steering(node); 677 struct mlx5_flow_group *fg; 678 struct mlx5_flow_table *ft; 679 int err; 680 681 fs_get_obj(fg, node); 682 fs_get_obj(ft, fg->node.parent); 683 684 rhashtable_destroy(&fg->ftes_hash); 685 ida_destroy(&fg->fte_allocator); 686 if (ft->autogroup.active && 687 fg->max_ftes == ft->autogroup.group_size && 688 fg->start_index < ft->autogroup.max_fte) 689 ft->autogroup.num_groups--; 690 err = rhltable_remove(&ft->fgs_hash, 691 &fg->hash, 692 rhash_fg); 693 WARN_ON(err); 694 kmem_cache_free(steering->fgs_cache, fg); 695 } 696 697 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte) 698 { 699 int index; 700 int ret; 701 702 index = ida_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, GFP_KERNEL); 703 if (index < 0) 704 return index; 705 706 fte->index = index + fg->start_index; 707 ret = rhashtable_insert_fast(&fg->ftes_hash, 708 &fte->hash, 709 rhash_fte); 710 if (ret) 711 goto err_ida_remove; 712 713 tree_add_node(&fte->node, &fg->node); 714 list_add_tail(&fte->node.list, &fg->node.children); 715 return 0; 716 717 err_ida_remove: 718 ida_free(&fg->fte_allocator, index); 719 return ret; 720 } 721 722 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft, 723 const struct mlx5_flow_spec *spec, 724 struct mlx5_flow_act *flow_act) 725 { 726 struct mlx5_flow_steering *steering = get_steering(&ft->node); 727 struct fs_fte *fte; 728 729 fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL); 730 if (!fte) 731 return ERR_PTR(-ENOMEM); 732 733 memcpy(fte->val, &spec->match_value, sizeof(fte->val)); 734 fte->node.type = FS_TYPE_FLOW_ENTRY; 735 fte->action = *flow_act; 736 fte->flow_context = spec->flow_context; 737 738 tree_init_node(&fte->node, del_hw_fte, del_sw_fte); 739 740 return fte; 741 } 742 743 static void dealloc_flow_group(struct mlx5_flow_steering *steering, 744 struct mlx5_flow_group *fg) 745 { 746 rhashtable_destroy(&fg->ftes_hash); 747 kmem_cache_free(steering->fgs_cache, fg); 748 } 749 750 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering, 751 u8 match_criteria_enable, 752 const void *match_criteria, 753 int start_index, 754 int end_index) 755 { 756 struct mlx5_flow_group *fg; 757 int ret; 758 759 fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL); 760 if (!fg) 761 return ERR_PTR(-ENOMEM); 762 763 ret = rhashtable_init(&fg->ftes_hash, &rhash_fte); 764 if (ret) { 765 kmem_cache_free(steering->fgs_cache, fg); 766 return ERR_PTR(ret); 767 } 768 769 ida_init(&fg->fte_allocator); 770 fg->mask.match_criteria_enable = match_criteria_enable; 771 memcpy(&fg->mask.match_criteria, match_criteria, 772 sizeof(fg->mask.match_criteria)); 773 fg->node.type = FS_TYPE_FLOW_GROUP; 774 fg->start_index = start_index; 775 fg->max_ftes = end_index - start_index + 1; 776 777 return fg; 778 } 779 780 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft, 781 u8 match_criteria_enable, 782 const void *match_criteria, 783 int start_index, 784 int end_index, 785 struct list_head *prev) 786 { 787 struct mlx5_flow_steering *steering = get_steering(&ft->node); 788 struct mlx5_flow_group *fg; 789 int ret; 790 791 fg = alloc_flow_group(steering, match_criteria_enable, match_criteria, 792 start_index, end_index); 793 if (IS_ERR(fg)) 794 return fg; 795 796 /* initialize refcnt, add to parent list */ 797 ret = rhltable_insert(&ft->fgs_hash, 798 &fg->hash, 799 rhash_fg); 800 if (ret) { 801 dealloc_flow_group(steering, fg); 802 return ERR_PTR(ret); 803 } 804 805 tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group); 806 tree_add_node(&fg->node, &ft->node); 807 /* Add node to group list */ 808 list_add(&fg->node.list, prev); 809 atomic_inc(&ft->node.version); 810 811 return fg; 812 } 813 814 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, 815 enum fs_flow_table_type table_type, 816 enum fs_flow_table_op_mod op_mod, 817 u32 flags) 818 { 819 struct mlx5_flow_table *ft; 820 int ret; 821 822 ft = kzalloc(sizeof(*ft), GFP_KERNEL); 823 if (!ft) 824 return ERR_PTR(-ENOMEM); 825 826 ret = rhltable_init(&ft->fgs_hash, &rhash_fg); 827 if (ret) { 828 kfree(ft); 829 return ERR_PTR(ret); 830 } 831 832 ft->level = level; 833 ft->node.type = FS_TYPE_FLOW_TABLE; 834 ft->op_mod = op_mod; 835 ft->type = table_type; 836 ft->vport = vport; 837 ft->flags = flags; 838 INIT_LIST_HEAD(&ft->fwd_rules); 839 mutex_init(&ft->lock); 840 841 return ft; 842 } 843 844 /* If reverse is false, then we search for the first flow table in the 845 * root sub-tree from start(closest from right), else we search for the 846 * last flow table in the root sub-tree till start(closest from left). 847 */ 848 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node *root, 849 struct list_head *start, 850 bool reverse) 851 { 852 #define list_advance_entry(pos, reverse) \ 853 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list)) 854 855 #define list_for_each_advance_continue(pos, head, reverse) \ 856 for (pos = list_advance_entry(pos, reverse); \ 857 &pos->list != (head); \ 858 pos = list_advance_entry(pos, reverse)) 859 860 struct fs_node *iter = list_entry(start, struct fs_node, list); 861 struct mlx5_flow_table *ft = NULL; 862 863 if (!root || root->type == FS_TYPE_PRIO_CHAINS) 864 return NULL; 865 866 list_for_each_advance_continue(iter, &root->children, reverse) { 867 if (iter->type == FS_TYPE_FLOW_TABLE) { 868 fs_get_obj(ft, iter); 869 return ft; 870 } 871 ft = find_closest_ft_recursive(iter, &iter->children, reverse); 872 if (ft) 873 return ft; 874 } 875 876 return ft; 877 } 878 879 /* If reverse is false then return the first flow table in next priority of 880 * prio in the tree, else return the last flow table in the previous priority 881 * of prio in the tree. 882 */ 883 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse) 884 { 885 struct mlx5_flow_table *ft = NULL; 886 struct fs_node *curr_node; 887 struct fs_node *parent; 888 889 parent = prio->node.parent; 890 curr_node = &prio->node; 891 while (!ft && parent) { 892 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse); 893 curr_node = parent; 894 parent = curr_node->parent; 895 } 896 return ft; 897 } 898 899 /* Assuming all the tree is locked by mutex chain lock */ 900 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio) 901 { 902 return find_closest_ft(prio, false); 903 } 904 905 /* Assuming all the tree is locked by mutex chain lock */ 906 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio) 907 { 908 return find_closest_ft(prio, true); 909 } 910 911 static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft, 912 struct mlx5_flow_act *flow_act) 913 { 914 struct fs_prio *prio; 915 bool next_ns; 916 917 next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS; 918 fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent); 919 920 return find_next_chained_ft(prio); 921 } 922 923 static int connect_fts_in_prio(struct mlx5_core_dev *dev, 924 struct fs_prio *prio, 925 struct mlx5_flow_table *ft) 926 { 927 struct mlx5_flow_root_namespace *root = find_root(&prio->node); 928 struct mlx5_flow_table *iter; 929 int err; 930 931 fs_for_each_ft(iter, prio) { 932 err = root->cmds->modify_flow_table(root, iter, ft); 933 if (err) { 934 mlx5_core_err(dev, 935 "Failed to modify flow table id %d, type %d, err %d\n", 936 iter->id, iter->type, err); 937 /* The driver is out of sync with the FW */ 938 return err; 939 } 940 } 941 return 0; 942 } 943 944 /* Connect flow tables from previous priority of prio to ft */ 945 static int connect_prev_fts(struct mlx5_core_dev *dev, 946 struct mlx5_flow_table *ft, 947 struct fs_prio *prio) 948 { 949 struct mlx5_flow_table *prev_ft; 950 951 prev_ft = find_prev_chained_ft(prio); 952 if (prev_ft) { 953 struct fs_prio *prev_prio; 954 955 fs_get_obj(prev_prio, prev_ft->node.parent); 956 return connect_fts_in_prio(dev, prev_prio, ft); 957 } 958 return 0; 959 } 960 961 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio 962 *prio) 963 { 964 struct mlx5_flow_root_namespace *root = find_root(&prio->node); 965 struct mlx5_ft_underlay_qp *uqp; 966 int min_level = INT_MAX; 967 int err = 0; 968 u32 qpn; 969 970 if (root->root_ft) 971 min_level = root->root_ft->level; 972 973 if (ft->level >= min_level) 974 return 0; 975 976 if (list_empty(&root->underlay_qpns)) { 977 /* Don't set any QPN (zero) in case QPN list is empty */ 978 qpn = 0; 979 err = root->cmds->update_root_ft(root, ft, qpn, false); 980 } else { 981 list_for_each_entry(uqp, &root->underlay_qpns, list) { 982 qpn = uqp->qpn; 983 err = root->cmds->update_root_ft(root, ft, 984 qpn, false); 985 if (err) 986 break; 987 } 988 } 989 990 if (err) 991 mlx5_core_warn(root->dev, 992 "Update root flow table of id(%u) qpn(%d) failed\n", 993 ft->id, qpn); 994 else 995 root->root_ft = ft; 996 997 return err; 998 } 999 1000 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule, 1001 struct mlx5_flow_destination *dest) 1002 { 1003 struct mlx5_flow_root_namespace *root; 1004 struct mlx5_flow_table *ft; 1005 struct mlx5_flow_group *fg; 1006 struct fs_fte *fte; 1007 int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST); 1008 int err = 0; 1009 1010 fs_get_obj(fte, rule->node.parent); 1011 if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)) 1012 return -EINVAL; 1013 down_write_ref_node(&fte->node, false); 1014 fs_get_obj(fg, fte->node.parent); 1015 fs_get_obj(ft, fg->node.parent); 1016 1017 memcpy(&rule->dest_attr, dest, sizeof(*dest)); 1018 root = find_root(&ft->node); 1019 err = root->cmds->update_fte(root, ft, fg, 1020 modify_mask, fte); 1021 up_write_ref_node(&fte->node, false); 1022 1023 return err; 1024 } 1025 1026 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle, 1027 struct mlx5_flow_destination *new_dest, 1028 struct mlx5_flow_destination *old_dest) 1029 { 1030 int i; 1031 1032 if (!old_dest) { 1033 if (handle->num_rules != 1) 1034 return -EINVAL; 1035 return _mlx5_modify_rule_destination(handle->rule[0], 1036 new_dest); 1037 } 1038 1039 for (i = 0; i < handle->num_rules; i++) { 1040 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr)) 1041 return _mlx5_modify_rule_destination(handle->rule[i], 1042 new_dest); 1043 } 1044 1045 return -EINVAL; 1046 } 1047 1048 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */ 1049 static int connect_fwd_rules(struct mlx5_core_dev *dev, 1050 struct mlx5_flow_table *new_next_ft, 1051 struct mlx5_flow_table *old_next_ft) 1052 { 1053 struct mlx5_flow_destination dest = {}; 1054 struct mlx5_flow_rule *iter; 1055 int err = 0; 1056 1057 /* new_next_ft and old_next_ft could be NULL only 1058 * when we create/destroy the anchor flow table. 1059 */ 1060 if (!new_next_ft || !old_next_ft) 1061 return 0; 1062 1063 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE; 1064 dest.ft = new_next_ft; 1065 1066 mutex_lock(&old_next_ft->lock); 1067 list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules); 1068 mutex_unlock(&old_next_ft->lock); 1069 list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) { 1070 if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) && 1071 iter->ft->ns == new_next_ft->ns) 1072 continue; 1073 1074 err = _mlx5_modify_rule_destination(iter, &dest); 1075 if (err) 1076 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n", 1077 new_next_ft->id); 1078 } 1079 return 0; 1080 } 1081 1082 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft, 1083 struct fs_prio *prio) 1084 { 1085 struct mlx5_flow_table *next_ft, *first_ft; 1086 int err = 0; 1087 1088 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */ 1089 1090 first_ft = list_first_entry_or_null(&prio->node.children, 1091 struct mlx5_flow_table, node.list); 1092 if (!first_ft || first_ft->level > ft->level) { 1093 err = connect_prev_fts(dev, ft, prio); 1094 if (err) 1095 return err; 1096 1097 next_ft = first_ft ? first_ft : find_next_chained_ft(prio); 1098 err = connect_fwd_rules(dev, ft, next_ft); 1099 if (err) 1100 return err; 1101 } 1102 1103 if (MLX5_CAP_FLOWTABLE(dev, 1104 flow_table_properties_nic_receive.modify_root)) 1105 err = update_root_ft_create(ft, prio); 1106 return err; 1107 } 1108 1109 static void list_add_flow_table(struct mlx5_flow_table *ft, 1110 struct fs_prio *prio) 1111 { 1112 struct list_head *prev = &prio->node.children; 1113 struct mlx5_flow_table *iter; 1114 1115 fs_for_each_ft(iter, prio) { 1116 if (iter->level > ft->level) 1117 break; 1118 prev = &iter->node.list; 1119 } 1120 list_add(&ft->node.list, prev); 1121 } 1122 1123 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns, 1124 struct mlx5_flow_table_attr *ft_attr, 1125 enum fs_flow_table_op_mod op_mod, 1126 u16 vport) 1127 { 1128 struct mlx5_flow_root_namespace *root = find_root(&ns->node); 1129 bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED; 1130 struct mlx5_flow_table *next_ft; 1131 struct fs_prio *fs_prio = NULL; 1132 struct mlx5_flow_table *ft; 1133 int err; 1134 1135 if (!root) { 1136 pr_err("mlx5: flow steering failed to find root of namespace\n"); 1137 return ERR_PTR(-ENODEV); 1138 } 1139 1140 mutex_lock(&root->chain_lock); 1141 fs_prio = find_prio(ns, ft_attr->prio); 1142 if (!fs_prio) { 1143 err = -EINVAL; 1144 goto unlock_root; 1145 } 1146 if (!unmanaged) { 1147 /* The level is related to the 1148 * priority level range. 1149 */ 1150 if (ft_attr->level >= fs_prio->num_levels) { 1151 err = -ENOSPC; 1152 goto unlock_root; 1153 } 1154 1155 ft_attr->level += fs_prio->start_level; 1156 } 1157 1158 /* The level is related to the 1159 * priority level range. 1160 */ 1161 ft = alloc_flow_table(ft_attr->level, 1162 vport, 1163 root->table_type, 1164 op_mod, ft_attr->flags); 1165 if (IS_ERR(ft)) { 1166 err = PTR_ERR(ft); 1167 goto unlock_root; 1168 } 1169 1170 tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table); 1171 next_ft = unmanaged ? ft_attr->next_ft : 1172 find_next_chained_ft(fs_prio); 1173 ft->def_miss_action = ns->def_miss_action; 1174 ft->ns = ns; 1175 err = root->cmds->create_flow_table(root, ft, ft_attr, next_ft); 1176 if (err) 1177 goto free_ft; 1178 1179 if (!unmanaged) { 1180 err = connect_flow_table(root->dev, ft, fs_prio); 1181 if (err) 1182 goto destroy_ft; 1183 } 1184 1185 ft->node.active = true; 1186 down_write_ref_node(&fs_prio->node, false); 1187 if (!unmanaged) { 1188 tree_add_node(&ft->node, &fs_prio->node); 1189 list_add_flow_table(ft, fs_prio); 1190 } else { 1191 ft->node.root = fs_prio->node.root; 1192 } 1193 fs_prio->num_ft++; 1194 up_write_ref_node(&fs_prio->node, false); 1195 mutex_unlock(&root->chain_lock); 1196 trace_mlx5_fs_add_ft(ft); 1197 return ft; 1198 destroy_ft: 1199 root->cmds->destroy_flow_table(root, ft); 1200 free_ft: 1201 rhltable_destroy(&ft->fgs_hash); 1202 kfree(ft); 1203 unlock_root: 1204 mutex_unlock(&root->chain_lock); 1205 return ERR_PTR(err); 1206 } 1207 1208 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns, 1209 struct mlx5_flow_table_attr *ft_attr) 1210 { 1211 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0); 1212 } 1213 EXPORT_SYMBOL(mlx5_create_flow_table); 1214 1215 u32 mlx5_flow_table_id(struct mlx5_flow_table *ft) 1216 { 1217 return ft->id; 1218 } 1219 EXPORT_SYMBOL(mlx5_flow_table_id); 1220 1221 struct mlx5_flow_table * 1222 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns, 1223 struct mlx5_flow_table_attr *ft_attr, u16 vport) 1224 { 1225 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport); 1226 } 1227 1228 struct mlx5_flow_table* 1229 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns, 1230 int prio, u32 level) 1231 { 1232 struct mlx5_flow_table_attr ft_attr = {}; 1233 1234 ft_attr.level = level; 1235 ft_attr.prio = prio; 1236 ft_attr.max_fte = 1; 1237 1238 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0); 1239 } 1240 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table); 1241 1242 #define MAX_FLOW_GROUP_SIZE BIT(24) 1243 struct mlx5_flow_table* 1244 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns, 1245 struct mlx5_flow_table_attr *ft_attr) 1246 { 1247 int num_reserved_entries = ft_attr->autogroup.num_reserved_entries; 1248 int max_num_groups = ft_attr->autogroup.max_num_groups; 1249 struct mlx5_flow_table *ft; 1250 int autogroups_max_fte; 1251 1252 ft = mlx5_create_flow_table(ns, ft_attr); 1253 if (IS_ERR(ft)) 1254 return ft; 1255 1256 autogroups_max_fte = ft->max_fte - num_reserved_entries; 1257 if (max_num_groups > autogroups_max_fte) 1258 goto err_validate; 1259 if (num_reserved_entries > ft->max_fte) 1260 goto err_validate; 1261 1262 /* Align the number of groups according to the largest group size */ 1263 if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE) 1264 max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1; 1265 1266 ft->autogroup.active = true; 1267 ft->autogroup.required_groups = max_num_groups; 1268 ft->autogroup.max_fte = autogroups_max_fte; 1269 /* We save place for flow groups in addition to max types */ 1270 ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1); 1271 1272 return ft; 1273 1274 err_validate: 1275 mlx5_destroy_flow_table(ft); 1276 return ERR_PTR(-ENOSPC); 1277 } 1278 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table); 1279 1280 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft, 1281 u32 *fg_in) 1282 { 1283 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 1284 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in, 1285 fg_in, match_criteria); 1286 u8 match_criteria_enable = MLX5_GET(create_flow_group_in, 1287 fg_in, 1288 match_criteria_enable); 1289 int start_index = MLX5_GET(create_flow_group_in, fg_in, 1290 start_flow_index); 1291 int end_index = MLX5_GET(create_flow_group_in, fg_in, 1292 end_flow_index); 1293 struct mlx5_flow_group *fg; 1294 int err; 1295 1296 if (ft->autogroup.active && start_index < ft->autogroup.max_fte) 1297 return ERR_PTR(-EPERM); 1298 1299 down_write_ref_node(&ft->node, false); 1300 fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria, 1301 start_index, end_index, 1302 ft->node.children.prev); 1303 up_write_ref_node(&ft->node, false); 1304 if (IS_ERR(fg)) 1305 return fg; 1306 1307 err = root->cmds->create_flow_group(root, ft, fg_in, fg); 1308 if (err) { 1309 tree_put_node(&fg->node, false); 1310 return ERR_PTR(err); 1311 } 1312 trace_mlx5_fs_add_fg(fg); 1313 fg->node.active = true; 1314 1315 return fg; 1316 } 1317 EXPORT_SYMBOL(mlx5_create_flow_group); 1318 1319 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest) 1320 { 1321 struct mlx5_flow_rule *rule; 1322 1323 rule = kzalloc(sizeof(*rule), GFP_KERNEL); 1324 if (!rule) 1325 return NULL; 1326 1327 INIT_LIST_HEAD(&rule->next_ft); 1328 rule->node.type = FS_TYPE_FLOW_DEST; 1329 if (dest) 1330 memcpy(&rule->dest_attr, dest, sizeof(*dest)); 1331 else 1332 rule->dest_attr.type = MLX5_FLOW_DESTINATION_TYPE_NONE; 1333 1334 return rule; 1335 } 1336 1337 static struct mlx5_flow_handle *alloc_handle(int num_rules) 1338 { 1339 struct mlx5_flow_handle *handle; 1340 1341 handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL); 1342 if (!handle) 1343 return NULL; 1344 1345 handle->num_rules = num_rules; 1346 1347 return handle; 1348 } 1349 1350 static void destroy_flow_handle(struct fs_fte *fte, 1351 struct mlx5_flow_handle *handle, 1352 struct mlx5_flow_destination *dest, 1353 int i) 1354 { 1355 for (; --i >= 0;) { 1356 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) { 1357 fte->dests_size--; 1358 list_del(&handle->rule[i]->node.list); 1359 kfree(handle->rule[i]); 1360 } 1361 } 1362 kfree(handle); 1363 } 1364 1365 static struct mlx5_flow_handle * 1366 create_flow_handle(struct fs_fte *fte, 1367 struct mlx5_flow_destination *dest, 1368 int dest_num, 1369 int *modify_mask, 1370 bool *new_rule) 1371 { 1372 struct mlx5_flow_handle *handle; 1373 struct mlx5_flow_rule *rule = NULL; 1374 static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS); 1375 static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST); 1376 int type; 1377 int i = 0; 1378 1379 handle = alloc_handle((dest_num) ? dest_num : 1); 1380 if (!handle) 1381 return ERR_PTR(-ENOMEM); 1382 1383 do { 1384 if (dest) { 1385 rule = find_flow_rule(fte, dest + i); 1386 if (rule) { 1387 refcount_inc(&rule->node.refcount); 1388 goto rule_found; 1389 } 1390 } 1391 1392 *new_rule = true; 1393 rule = alloc_rule(dest + i); 1394 if (!rule) 1395 goto free_rules; 1396 1397 /* Add dest to dests list- we need flow tables to be in the 1398 * end of the list for forward to next prio rules. 1399 */ 1400 tree_init_node(&rule->node, NULL, del_sw_hw_rule); 1401 if (dest && 1402 dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) 1403 list_add(&rule->node.list, &fte->node.children); 1404 else 1405 list_add_tail(&rule->node.list, &fte->node.children); 1406 if (dest) { 1407 fte->dests_size++; 1408 1409 if (is_fwd_dest_type(dest[i].type)) 1410 fte->fwd_dests++; 1411 1412 type = dest[i].type == 1413 MLX5_FLOW_DESTINATION_TYPE_COUNTER; 1414 *modify_mask |= type ? count : dst; 1415 } 1416 rule_found: 1417 handle->rule[i] = rule; 1418 } while (++i < dest_num); 1419 1420 return handle; 1421 1422 free_rules: 1423 destroy_flow_handle(fte, handle, dest, i); 1424 return ERR_PTR(-ENOMEM); 1425 } 1426 1427 /* fte should not be deleted while calling this function */ 1428 static struct mlx5_flow_handle * 1429 add_rule_fte(struct fs_fte *fte, 1430 struct mlx5_flow_group *fg, 1431 struct mlx5_flow_destination *dest, 1432 int dest_num, 1433 bool update_action) 1434 { 1435 struct mlx5_flow_root_namespace *root; 1436 struct mlx5_flow_handle *handle; 1437 struct mlx5_flow_table *ft; 1438 int modify_mask = 0; 1439 int err; 1440 bool new_rule = false; 1441 1442 handle = create_flow_handle(fte, dest, dest_num, &modify_mask, 1443 &new_rule); 1444 if (IS_ERR(handle) || !new_rule) 1445 goto out; 1446 1447 if (update_action) 1448 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION); 1449 1450 fs_get_obj(ft, fg->node.parent); 1451 root = find_root(&fg->node); 1452 if (!(fte->status & FS_FTE_STATUS_EXISTING)) 1453 err = root->cmds->create_fte(root, ft, fg, fte); 1454 else 1455 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte); 1456 if (err) 1457 goto free_handle; 1458 1459 fte->node.active = true; 1460 fte->status |= FS_FTE_STATUS_EXISTING; 1461 atomic_inc(&fg->node.version); 1462 1463 out: 1464 return handle; 1465 1466 free_handle: 1467 destroy_flow_handle(fte, handle, dest, handle->num_rules); 1468 return ERR_PTR(err); 1469 } 1470 1471 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table *ft, 1472 const struct mlx5_flow_spec *spec) 1473 { 1474 struct list_head *prev = &ft->node.children; 1475 u32 max_fte = ft->autogroup.max_fte; 1476 unsigned int candidate_index = 0; 1477 unsigned int group_size = 0; 1478 struct mlx5_flow_group *fg; 1479 1480 if (!ft->autogroup.active) 1481 return ERR_PTR(-ENOENT); 1482 1483 if (ft->autogroup.num_groups < ft->autogroup.required_groups) 1484 group_size = ft->autogroup.group_size; 1485 1486 /* max_fte == ft->autogroup.max_types */ 1487 if (group_size == 0) 1488 group_size = 1; 1489 1490 /* sorted by start_index */ 1491 fs_for_each_fg(fg, ft) { 1492 if (candidate_index + group_size > fg->start_index) 1493 candidate_index = fg->start_index + fg->max_ftes; 1494 else 1495 break; 1496 prev = &fg->node.list; 1497 } 1498 1499 if (candidate_index + group_size > max_fte) 1500 return ERR_PTR(-ENOSPC); 1501 1502 fg = alloc_insert_flow_group(ft, 1503 spec->match_criteria_enable, 1504 spec->match_criteria, 1505 candidate_index, 1506 candidate_index + group_size - 1, 1507 prev); 1508 if (IS_ERR(fg)) 1509 goto out; 1510 1511 if (group_size == ft->autogroup.group_size) 1512 ft->autogroup.num_groups++; 1513 1514 out: 1515 return fg; 1516 } 1517 1518 static int create_auto_flow_group(struct mlx5_flow_table *ft, 1519 struct mlx5_flow_group *fg) 1520 { 1521 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 1522 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in); 1523 void *match_criteria_addr; 1524 u8 src_esw_owner_mask_on; 1525 void *misc; 1526 int err; 1527 u32 *in; 1528 1529 in = kvzalloc(inlen, GFP_KERNEL); 1530 if (!in) 1531 return -ENOMEM; 1532 1533 MLX5_SET(create_flow_group_in, in, match_criteria_enable, 1534 fg->mask.match_criteria_enable); 1535 MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index); 1536 MLX5_SET(create_flow_group_in, in, end_flow_index, fg->start_index + 1537 fg->max_ftes - 1); 1538 1539 misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria, 1540 misc_parameters); 1541 src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc, 1542 source_eswitch_owner_vhca_id); 1543 MLX5_SET(create_flow_group_in, in, 1544 source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on); 1545 1546 match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in, 1547 in, match_criteria); 1548 memcpy(match_criteria_addr, fg->mask.match_criteria, 1549 sizeof(fg->mask.match_criteria)); 1550 1551 err = root->cmds->create_flow_group(root, ft, in, fg); 1552 if (!err) { 1553 fg->node.active = true; 1554 trace_mlx5_fs_add_fg(fg); 1555 } 1556 1557 kvfree(in); 1558 return err; 1559 } 1560 1561 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1, 1562 struct mlx5_flow_destination *d2) 1563 { 1564 if (d1->type == d2->type) { 1565 if (((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT || 1566 d1->type == MLX5_FLOW_DESTINATION_TYPE_UPLINK) && 1567 d1->vport.num == d2->vport.num && 1568 d1->vport.flags == d2->vport.flags && 1569 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ? 1570 (d1->vport.vhca_id == d2->vport.vhca_id) : true) && 1571 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ? 1572 (d1->vport.pkt_reformat->id == 1573 d2->vport.pkt_reformat->id) : true)) || 1574 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE && 1575 d1->ft == d2->ft) || 1576 (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR && 1577 d1->tir_num == d2->tir_num) || 1578 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM && 1579 d1->ft_num == d2->ft_num) || 1580 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER && 1581 d1->sampler_id == d2->sampler_id)) 1582 return true; 1583 } 1584 1585 return false; 1586 } 1587 1588 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte, 1589 struct mlx5_flow_destination *dest) 1590 { 1591 struct mlx5_flow_rule *rule; 1592 1593 list_for_each_entry(rule, &fte->node.children, node.list) { 1594 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest)) 1595 return rule; 1596 } 1597 return NULL; 1598 } 1599 1600 static bool check_conflicting_actions_vlan(const struct mlx5_fs_vlan *vlan0, 1601 const struct mlx5_fs_vlan *vlan1) 1602 { 1603 return vlan0->ethtype != vlan1->ethtype || 1604 vlan0->vid != vlan1->vid || 1605 vlan0->prio != vlan1->prio; 1606 } 1607 1608 static bool check_conflicting_actions(const struct mlx5_flow_act *act1, 1609 const struct mlx5_flow_act *act2) 1610 { 1611 u32 action1 = act1->action; 1612 u32 action2 = act2->action; 1613 u32 xored_actions; 1614 1615 xored_actions = action1 ^ action2; 1616 1617 /* if one rule only wants to count, it's ok */ 1618 if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT || 1619 action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT) 1620 return false; 1621 1622 if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP | 1623 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT | 1624 MLX5_FLOW_CONTEXT_ACTION_DECAP | 1625 MLX5_FLOW_CONTEXT_ACTION_MOD_HDR | 1626 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP | 1627 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH | 1628 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 | 1629 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2)) 1630 return true; 1631 1632 if (action1 & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT && 1633 act1->pkt_reformat != act2->pkt_reformat) 1634 return true; 1635 1636 if (action1 & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR && 1637 act1->modify_hdr != act2->modify_hdr) 1638 return true; 1639 1640 if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH && 1641 check_conflicting_actions_vlan(&act1->vlan[0], &act2->vlan[0])) 1642 return true; 1643 1644 if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2 && 1645 check_conflicting_actions_vlan(&act1->vlan[1], &act2->vlan[1])) 1646 return true; 1647 1648 return false; 1649 } 1650 1651 static int check_conflicting_ftes(struct fs_fte *fte, 1652 const struct mlx5_flow_context *flow_context, 1653 const struct mlx5_flow_act *flow_act) 1654 { 1655 if (check_conflicting_actions(flow_act, &fte->action)) { 1656 mlx5_core_warn(get_dev(&fte->node), 1657 "Found two FTEs with conflicting actions\n"); 1658 return -EEXIST; 1659 } 1660 1661 if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) && 1662 fte->flow_context.flow_tag != flow_context->flow_tag) { 1663 mlx5_core_warn(get_dev(&fte->node), 1664 "FTE flow tag %u already exists with different flow tag %u\n", 1665 fte->flow_context.flow_tag, 1666 flow_context->flow_tag); 1667 return -EEXIST; 1668 } 1669 1670 return 0; 1671 } 1672 1673 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg, 1674 const struct mlx5_flow_spec *spec, 1675 struct mlx5_flow_act *flow_act, 1676 struct mlx5_flow_destination *dest, 1677 int dest_num, 1678 struct fs_fte *fte) 1679 { 1680 struct mlx5_flow_handle *handle; 1681 int old_action; 1682 int i; 1683 int ret; 1684 1685 ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act); 1686 if (ret) 1687 return ERR_PTR(ret); 1688 1689 old_action = fte->action.action; 1690 fte->action.action |= flow_act->action; 1691 handle = add_rule_fte(fte, fg, dest, dest_num, 1692 old_action != flow_act->action); 1693 if (IS_ERR(handle)) { 1694 fte->action.action = old_action; 1695 return handle; 1696 } 1697 trace_mlx5_fs_set_fte(fte, false); 1698 1699 for (i = 0; i < handle->num_rules; i++) { 1700 if (refcount_read(&handle->rule[i]->node.refcount) == 1) { 1701 tree_add_node(&handle->rule[i]->node, &fte->node); 1702 trace_mlx5_fs_add_rule(handle->rule[i]); 1703 } 1704 } 1705 return handle; 1706 } 1707 1708 static bool counter_is_valid(u32 action) 1709 { 1710 return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP | 1711 MLX5_FLOW_CONTEXT_ACTION_ALLOW | 1712 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)); 1713 } 1714 1715 static bool dest_is_valid(struct mlx5_flow_destination *dest, 1716 struct mlx5_flow_act *flow_act, 1717 struct mlx5_flow_table *ft) 1718 { 1719 bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL; 1720 u32 action = flow_act->action; 1721 1722 if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)) 1723 return counter_is_valid(action); 1724 1725 if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)) 1726 return true; 1727 1728 if (ignore_level) { 1729 if (ft->type != FS_FT_FDB && 1730 ft->type != FS_FT_NIC_RX) 1731 return false; 1732 1733 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE && 1734 ft->type != dest->ft->type) 1735 return false; 1736 } 1737 1738 if (!dest || ((dest->type == 1739 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) && 1740 (dest->ft->level <= ft->level && !ignore_level))) 1741 return false; 1742 return true; 1743 } 1744 1745 struct match_list { 1746 struct list_head list; 1747 struct mlx5_flow_group *g; 1748 }; 1749 1750 static void free_match_list(struct match_list *head, bool ft_locked) 1751 { 1752 struct match_list *iter, *match_tmp; 1753 1754 list_for_each_entry_safe(iter, match_tmp, &head->list, 1755 list) { 1756 tree_put_node(&iter->g->node, ft_locked); 1757 list_del(&iter->list); 1758 kfree(iter); 1759 } 1760 } 1761 1762 static int build_match_list(struct match_list *match_head, 1763 struct mlx5_flow_table *ft, 1764 const struct mlx5_flow_spec *spec, 1765 struct mlx5_flow_group *fg, 1766 bool ft_locked) 1767 { 1768 struct rhlist_head *tmp, *list; 1769 struct mlx5_flow_group *g; 1770 int err = 0; 1771 1772 rcu_read_lock(); 1773 INIT_LIST_HEAD(&match_head->list); 1774 /* Collect all fgs which has a matching match_criteria */ 1775 list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg); 1776 /* RCU is atomic, we can't execute FW commands here */ 1777 rhl_for_each_entry_rcu(g, tmp, list, hash) { 1778 struct match_list *curr_match; 1779 1780 if (fg && fg != g) 1781 continue; 1782 1783 if (unlikely(!tree_get_node(&g->node))) 1784 continue; 1785 1786 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC); 1787 if (!curr_match) { 1788 rcu_read_unlock(); 1789 free_match_list(match_head, ft_locked); 1790 return -ENOMEM; 1791 } 1792 curr_match->g = g; 1793 list_add_tail(&curr_match->list, &match_head->list); 1794 } 1795 rcu_read_unlock(); 1796 return err; 1797 } 1798 1799 static u64 matched_fgs_get_version(struct list_head *match_head) 1800 { 1801 struct match_list *iter; 1802 u64 version = 0; 1803 1804 list_for_each_entry(iter, match_head, list) 1805 version += (u64)atomic_read(&iter->g->node.version); 1806 return version; 1807 } 1808 1809 static struct fs_fte * 1810 lookup_fte_locked(struct mlx5_flow_group *g, 1811 const u32 *match_value, 1812 bool take_write) 1813 { 1814 struct fs_fte *fte_tmp; 1815 1816 if (take_write) 1817 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT); 1818 else 1819 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT); 1820 fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value, 1821 rhash_fte); 1822 if (!fte_tmp || !tree_get_node(&fte_tmp->node)) { 1823 fte_tmp = NULL; 1824 goto out; 1825 } 1826 if (!fte_tmp->node.active) { 1827 tree_put_node(&fte_tmp->node, false); 1828 fte_tmp = NULL; 1829 goto out; 1830 } 1831 1832 nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD); 1833 out: 1834 if (take_write) 1835 up_write_ref_node(&g->node, false); 1836 else 1837 up_read_ref_node(&g->node); 1838 return fte_tmp; 1839 } 1840 1841 static struct mlx5_flow_handle * 1842 try_add_to_existing_fg(struct mlx5_flow_table *ft, 1843 struct list_head *match_head, 1844 const struct mlx5_flow_spec *spec, 1845 struct mlx5_flow_act *flow_act, 1846 struct mlx5_flow_destination *dest, 1847 int dest_num, 1848 int ft_version) 1849 { 1850 struct mlx5_flow_steering *steering = get_steering(&ft->node); 1851 struct mlx5_flow_group *g; 1852 struct mlx5_flow_handle *rule; 1853 struct match_list *iter; 1854 bool take_write = false; 1855 struct fs_fte *fte; 1856 u64 version = 0; 1857 int err; 1858 1859 fte = alloc_fte(ft, spec, flow_act); 1860 if (IS_ERR(fte)) 1861 return ERR_PTR(-ENOMEM); 1862 1863 search_again_locked: 1864 if (flow_act->flags & FLOW_ACT_NO_APPEND) 1865 goto skip_search; 1866 version = matched_fgs_get_version(match_head); 1867 /* Try to find an fte with identical match value and attempt update its 1868 * action. 1869 */ 1870 list_for_each_entry(iter, match_head, list) { 1871 struct fs_fte *fte_tmp; 1872 1873 g = iter->g; 1874 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write); 1875 if (!fte_tmp) 1876 continue; 1877 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp); 1878 /* No error check needed here, because insert_fte() is not called */ 1879 up_write_ref_node(&fte_tmp->node, false); 1880 tree_put_node(&fte_tmp->node, false); 1881 kmem_cache_free(steering->ftes_cache, fte); 1882 return rule; 1883 } 1884 1885 skip_search: 1886 /* No group with matching fte found, or we skipped the search. 1887 * Try to add a new fte to any matching fg. 1888 */ 1889 1890 /* Check the ft version, for case that new flow group 1891 * was added while the fgs weren't locked 1892 */ 1893 if (atomic_read(&ft->node.version) != ft_version) { 1894 rule = ERR_PTR(-EAGAIN); 1895 goto out; 1896 } 1897 1898 /* Check the fgs version. If version have changed it could be that an 1899 * FTE with the same match value was added while the fgs weren't 1900 * locked. 1901 */ 1902 if (!(flow_act->flags & FLOW_ACT_NO_APPEND) && 1903 version != matched_fgs_get_version(match_head)) { 1904 take_write = true; 1905 goto search_again_locked; 1906 } 1907 1908 list_for_each_entry(iter, match_head, list) { 1909 g = iter->g; 1910 1911 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT); 1912 1913 if (!g->node.active) { 1914 up_write_ref_node(&g->node, false); 1915 continue; 1916 } 1917 1918 err = insert_fte(g, fte); 1919 if (err) { 1920 up_write_ref_node(&g->node, false); 1921 if (err == -ENOSPC) 1922 continue; 1923 kmem_cache_free(steering->ftes_cache, fte); 1924 return ERR_PTR(err); 1925 } 1926 1927 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD); 1928 up_write_ref_node(&g->node, false); 1929 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte); 1930 up_write_ref_node(&fte->node, false); 1931 if (IS_ERR(rule)) 1932 tree_put_node(&fte->node, false); 1933 return rule; 1934 } 1935 rule = ERR_PTR(-ENOENT); 1936 out: 1937 kmem_cache_free(steering->ftes_cache, fte); 1938 return rule; 1939 } 1940 1941 static struct mlx5_flow_handle * 1942 _mlx5_add_flow_rules(struct mlx5_flow_table *ft, 1943 const struct mlx5_flow_spec *spec, 1944 struct mlx5_flow_act *flow_act, 1945 struct mlx5_flow_destination *dest, 1946 int dest_num) 1947 1948 { 1949 struct mlx5_flow_steering *steering = get_steering(&ft->node); 1950 struct mlx5_flow_handle *rule; 1951 struct match_list match_head; 1952 struct mlx5_flow_group *g; 1953 bool take_write = false; 1954 struct fs_fte *fte; 1955 int version; 1956 int err; 1957 int i; 1958 1959 if (!check_valid_spec(spec)) 1960 return ERR_PTR(-EINVAL); 1961 1962 if (flow_act->fg && ft->autogroup.active) 1963 return ERR_PTR(-EINVAL); 1964 1965 for (i = 0; i < dest_num; i++) { 1966 if (!dest_is_valid(&dest[i], flow_act, ft)) 1967 return ERR_PTR(-EINVAL); 1968 } 1969 nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT); 1970 search_again_locked: 1971 version = atomic_read(&ft->node.version); 1972 1973 /* Collect all fgs which has a matching match_criteria */ 1974 err = build_match_list(&match_head, ft, spec, flow_act->fg, take_write); 1975 if (err) { 1976 if (take_write) 1977 up_write_ref_node(&ft->node, false); 1978 else 1979 up_read_ref_node(&ft->node); 1980 return ERR_PTR(err); 1981 } 1982 1983 if (!take_write) 1984 up_read_ref_node(&ft->node); 1985 1986 rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest, 1987 dest_num, version); 1988 free_match_list(&match_head, take_write); 1989 if (!IS_ERR(rule) || 1990 (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) { 1991 if (take_write) 1992 up_write_ref_node(&ft->node, false); 1993 return rule; 1994 } 1995 1996 if (!take_write) { 1997 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT); 1998 take_write = true; 1999 } 2000 2001 if (PTR_ERR(rule) == -EAGAIN || 2002 version != atomic_read(&ft->node.version)) 2003 goto search_again_locked; 2004 2005 g = alloc_auto_flow_group(ft, spec); 2006 if (IS_ERR(g)) { 2007 rule = ERR_CAST(g); 2008 up_write_ref_node(&ft->node, false); 2009 return rule; 2010 } 2011 2012 fte = alloc_fte(ft, spec, flow_act); 2013 if (IS_ERR(fte)) { 2014 up_write_ref_node(&ft->node, false); 2015 err = PTR_ERR(fte); 2016 goto err_alloc_fte; 2017 } 2018 2019 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT); 2020 up_write_ref_node(&ft->node, false); 2021 2022 err = create_auto_flow_group(ft, g); 2023 if (err) 2024 goto err_release_fg; 2025 2026 err = insert_fte(g, fte); 2027 if (err) 2028 goto err_release_fg; 2029 2030 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD); 2031 up_write_ref_node(&g->node, false); 2032 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte); 2033 up_write_ref_node(&fte->node, false); 2034 if (IS_ERR(rule)) 2035 tree_put_node(&fte->node, false); 2036 tree_put_node(&g->node, false); 2037 return rule; 2038 2039 err_release_fg: 2040 up_write_ref_node(&g->node, false); 2041 kmem_cache_free(steering->ftes_cache, fte); 2042 err_alloc_fte: 2043 tree_put_node(&g->node, false); 2044 return ERR_PTR(err); 2045 } 2046 2047 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft) 2048 { 2049 return ((ft->type == FS_FT_NIC_RX) && 2050 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs))); 2051 } 2052 2053 struct mlx5_flow_handle * 2054 mlx5_add_flow_rules(struct mlx5_flow_table *ft, 2055 const struct mlx5_flow_spec *spec, 2056 struct mlx5_flow_act *flow_act, 2057 struct mlx5_flow_destination *dest, 2058 int num_dest) 2059 { 2060 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 2061 static const struct mlx5_flow_spec zero_spec = {}; 2062 struct mlx5_flow_destination *gen_dest = NULL; 2063 struct mlx5_flow_table *next_ft = NULL; 2064 struct mlx5_flow_handle *handle = NULL; 2065 u32 sw_action = flow_act->action; 2066 int i; 2067 2068 if (!spec) 2069 spec = &zero_spec; 2070 2071 if (!is_fwd_next_action(sw_action)) 2072 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest); 2073 2074 if (!fwd_next_prio_supported(ft)) 2075 return ERR_PTR(-EOPNOTSUPP); 2076 2077 mutex_lock(&root->chain_lock); 2078 next_ft = find_next_fwd_ft(ft, flow_act); 2079 if (!next_ft) { 2080 handle = ERR_PTR(-EOPNOTSUPP); 2081 goto unlock; 2082 } 2083 2084 gen_dest = kcalloc(num_dest + 1, sizeof(*dest), 2085 GFP_KERNEL); 2086 if (!gen_dest) { 2087 handle = ERR_PTR(-ENOMEM); 2088 goto unlock; 2089 } 2090 for (i = 0; i < num_dest; i++) 2091 gen_dest[i] = dest[i]; 2092 gen_dest[i].type = 2093 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE; 2094 gen_dest[i].ft = next_ft; 2095 dest = gen_dest; 2096 num_dest++; 2097 flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO | 2098 MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS); 2099 flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST; 2100 handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest); 2101 if (IS_ERR(handle)) 2102 goto unlock; 2103 2104 if (list_empty(&handle->rule[num_dest - 1]->next_ft)) { 2105 mutex_lock(&next_ft->lock); 2106 list_add(&handle->rule[num_dest - 1]->next_ft, 2107 &next_ft->fwd_rules); 2108 mutex_unlock(&next_ft->lock); 2109 handle->rule[num_dest - 1]->sw_action = sw_action; 2110 handle->rule[num_dest - 1]->ft = ft; 2111 } 2112 unlock: 2113 mutex_unlock(&root->chain_lock); 2114 kfree(gen_dest); 2115 return handle; 2116 } 2117 EXPORT_SYMBOL(mlx5_add_flow_rules); 2118 2119 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle) 2120 { 2121 struct fs_fte *fte; 2122 int i; 2123 2124 /* In order to consolidate the HW changes we lock the FTE for other 2125 * changes, and increase its refcount, in order not to perform the 2126 * "del" functions of the FTE. Will handle them here. 2127 * The removal of the rules is done under locked FTE. 2128 * After removing all the handle's rules, if there are remaining 2129 * rules, it means we just need to modify the FTE in FW, and 2130 * unlock/decrease the refcount we increased before. 2131 * Otherwise, it means the FTE should be deleted. First delete the 2132 * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of 2133 * the FTE, which will handle the last decrease of the refcount, as 2134 * well as required handling of its parent. 2135 */ 2136 fs_get_obj(fte, handle->rule[0]->node.parent); 2137 down_write_ref_node(&fte->node, false); 2138 for (i = handle->num_rules - 1; i >= 0; i--) 2139 tree_remove_node(&handle->rule[i]->node, true); 2140 if (list_empty(&fte->node.children)) { 2141 fte->node.del_hw_func(&fte->node); 2142 /* Avoid double call to del_hw_fte */ 2143 fte->node.del_hw_func = NULL; 2144 up_write_ref_node(&fte->node, false); 2145 tree_put_node(&fte->node, false); 2146 } else if (fte->dests_size) { 2147 if (fte->modify_mask) 2148 modify_fte(fte); 2149 up_write_ref_node(&fte->node, false); 2150 } else { 2151 up_write_ref_node(&fte->node, false); 2152 } 2153 kfree(handle); 2154 } 2155 EXPORT_SYMBOL(mlx5_del_flow_rules); 2156 2157 /* Assuming prio->node.children(flow tables) is sorted by level */ 2158 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft) 2159 { 2160 struct fs_prio *prio; 2161 2162 fs_get_obj(prio, ft->node.parent); 2163 2164 if (!list_is_last(&ft->node.list, &prio->node.children)) 2165 return list_next_entry(ft, node.list); 2166 return find_next_chained_ft(prio); 2167 } 2168 2169 static int update_root_ft_destroy(struct mlx5_flow_table *ft) 2170 { 2171 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 2172 struct mlx5_ft_underlay_qp *uqp; 2173 struct mlx5_flow_table *new_root_ft = NULL; 2174 int err = 0; 2175 u32 qpn; 2176 2177 if (root->root_ft != ft) 2178 return 0; 2179 2180 new_root_ft = find_next_ft(ft); 2181 if (!new_root_ft) { 2182 root->root_ft = NULL; 2183 return 0; 2184 } 2185 2186 if (list_empty(&root->underlay_qpns)) { 2187 /* Don't set any QPN (zero) in case QPN list is empty */ 2188 qpn = 0; 2189 err = root->cmds->update_root_ft(root, new_root_ft, 2190 qpn, false); 2191 } else { 2192 list_for_each_entry(uqp, &root->underlay_qpns, list) { 2193 qpn = uqp->qpn; 2194 err = root->cmds->update_root_ft(root, 2195 new_root_ft, qpn, 2196 false); 2197 if (err) 2198 break; 2199 } 2200 } 2201 2202 if (err) 2203 mlx5_core_warn(root->dev, 2204 "Update root flow table of id(%u) qpn(%d) failed\n", 2205 ft->id, qpn); 2206 else 2207 root->root_ft = new_root_ft; 2208 2209 return 0; 2210 } 2211 2212 /* Connect flow table from previous priority to 2213 * the next flow table. 2214 */ 2215 static int disconnect_flow_table(struct mlx5_flow_table *ft) 2216 { 2217 struct mlx5_core_dev *dev = get_dev(&ft->node); 2218 struct mlx5_flow_table *next_ft; 2219 struct fs_prio *prio; 2220 int err = 0; 2221 2222 err = update_root_ft_destroy(ft); 2223 if (err) 2224 return err; 2225 2226 fs_get_obj(prio, ft->node.parent); 2227 if (!(list_first_entry(&prio->node.children, 2228 struct mlx5_flow_table, 2229 node.list) == ft)) 2230 return 0; 2231 2232 next_ft = find_next_ft(ft); 2233 err = connect_fwd_rules(dev, next_ft, ft); 2234 if (err) 2235 return err; 2236 2237 err = connect_prev_fts(dev, next_ft, prio); 2238 if (err) 2239 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n", 2240 ft->id); 2241 return err; 2242 } 2243 2244 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft) 2245 { 2246 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 2247 int err = 0; 2248 2249 mutex_lock(&root->chain_lock); 2250 if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED)) 2251 err = disconnect_flow_table(ft); 2252 if (err) { 2253 mutex_unlock(&root->chain_lock); 2254 return err; 2255 } 2256 if (tree_remove_node(&ft->node, false)) 2257 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n", 2258 ft->id); 2259 mutex_unlock(&root->chain_lock); 2260 2261 return err; 2262 } 2263 EXPORT_SYMBOL(mlx5_destroy_flow_table); 2264 2265 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg) 2266 { 2267 if (tree_remove_node(&fg->node, false)) 2268 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n", 2269 fg->id); 2270 } 2271 EXPORT_SYMBOL(mlx5_destroy_flow_group); 2272 2273 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev, 2274 int n) 2275 { 2276 struct mlx5_flow_steering *steering = dev->priv.steering; 2277 2278 if (!steering || !steering->fdb_sub_ns) 2279 return NULL; 2280 2281 return steering->fdb_sub_ns[n]; 2282 } 2283 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns); 2284 2285 static bool is_nic_rx_ns(enum mlx5_flow_namespace_type type) 2286 { 2287 switch (type) { 2288 case MLX5_FLOW_NAMESPACE_BYPASS: 2289 case MLX5_FLOW_NAMESPACE_KERNEL_RX_MACSEC: 2290 case MLX5_FLOW_NAMESPACE_LAG: 2291 case MLX5_FLOW_NAMESPACE_OFFLOADS: 2292 case MLX5_FLOW_NAMESPACE_ETHTOOL: 2293 case MLX5_FLOW_NAMESPACE_KERNEL: 2294 case MLX5_FLOW_NAMESPACE_LEFTOVERS: 2295 case MLX5_FLOW_NAMESPACE_ANCHOR: 2296 return true; 2297 default: 2298 return false; 2299 } 2300 } 2301 2302 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev, 2303 enum mlx5_flow_namespace_type type) 2304 { 2305 struct mlx5_flow_steering *steering = dev->priv.steering; 2306 struct mlx5_flow_root_namespace *root_ns; 2307 int prio = 0; 2308 struct fs_prio *fs_prio; 2309 struct mlx5_flow_namespace *ns; 2310 2311 if (!steering) 2312 return NULL; 2313 2314 switch (type) { 2315 case MLX5_FLOW_NAMESPACE_FDB: 2316 if (steering->fdb_root_ns) 2317 return &steering->fdb_root_ns->ns; 2318 return NULL; 2319 case MLX5_FLOW_NAMESPACE_PORT_SEL: 2320 if (steering->port_sel_root_ns) 2321 return &steering->port_sel_root_ns->ns; 2322 return NULL; 2323 case MLX5_FLOW_NAMESPACE_SNIFFER_RX: 2324 if (steering->sniffer_rx_root_ns) 2325 return &steering->sniffer_rx_root_ns->ns; 2326 return NULL; 2327 case MLX5_FLOW_NAMESPACE_SNIFFER_TX: 2328 if (steering->sniffer_tx_root_ns) 2329 return &steering->sniffer_tx_root_ns->ns; 2330 return NULL; 2331 case MLX5_FLOW_NAMESPACE_FDB_BYPASS: 2332 root_ns = steering->fdb_root_ns; 2333 prio = FDB_BYPASS_PATH; 2334 break; 2335 case MLX5_FLOW_NAMESPACE_EGRESS: 2336 case MLX5_FLOW_NAMESPACE_EGRESS_IPSEC: 2337 case MLX5_FLOW_NAMESPACE_EGRESS_MACSEC: 2338 root_ns = steering->egress_root_ns; 2339 prio = type - MLX5_FLOW_NAMESPACE_EGRESS; 2340 break; 2341 case MLX5_FLOW_NAMESPACE_RDMA_RX: 2342 root_ns = steering->rdma_rx_root_ns; 2343 prio = RDMA_RX_BYPASS_PRIO; 2344 break; 2345 case MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL: 2346 root_ns = steering->rdma_rx_root_ns; 2347 prio = RDMA_RX_KERNEL_PRIO; 2348 break; 2349 case MLX5_FLOW_NAMESPACE_RDMA_TX: 2350 root_ns = steering->rdma_tx_root_ns; 2351 break; 2352 case MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS: 2353 root_ns = steering->rdma_rx_root_ns; 2354 prio = RDMA_RX_COUNTERS_PRIO; 2355 break; 2356 case MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS: 2357 root_ns = steering->rdma_tx_root_ns; 2358 prio = RDMA_TX_COUNTERS_PRIO; 2359 break; 2360 default: /* Must be NIC RX */ 2361 WARN_ON(!is_nic_rx_ns(type)); 2362 root_ns = steering->root_ns; 2363 prio = type; 2364 break; 2365 } 2366 2367 if (!root_ns) 2368 return NULL; 2369 2370 fs_prio = find_prio(&root_ns->ns, prio); 2371 if (!fs_prio) 2372 return NULL; 2373 2374 ns = list_first_entry(&fs_prio->node.children, 2375 typeof(*ns), 2376 node.list); 2377 2378 return ns; 2379 } 2380 EXPORT_SYMBOL(mlx5_get_flow_namespace); 2381 2382 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev, 2383 enum mlx5_flow_namespace_type type, 2384 int vport) 2385 { 2386 struct mlx5_flow_steering *steering = dev->priv.steering; 2387 2388 if (!steering) 2389 return NULL; 2390 2391 switch (type) { 2392 case MLX5_FLOW_NAMESPACE_ESW_EGRESS: 2393 if (vport >= steering->esw_egress_acl_vports) 2394 return NULL; 2395 if (steering->esw_egress_root_ns && 2396 steering->esw_egress_root_ns[vport]) 2397 return &steering->esw_egress_root_ns[vport]->ns; 2398 else 2399 return NULL; 2400 case MLX5_FLOW_NAMESPACE_ESW_INGRESS: 2401 if (vport >= steering->esw_ingress_acl_vports) 2402 return NULL; 2403 if (steering->esw_ingress_root_ns && 2404 steering->esw_ingress_root_ns[vport]) 2405 return &steering->esw_ingress_root_ns[vport]->ns; 2406 else 2407 return NULL; 2408 default: 2409 return NULL; 2410 } 2411 } 2412 2413 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns, 2414 unsigned int prio, 2415 int num_levels, 2416 enum fs_node_type type) 2417 { 2418 struct fs_prio *fs_prio; 2419 2420 fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL); 2421 if (!fs_prio) 2422 return ERR_PTR(-ENOMEM); 2423 2424 fs_prio->node.type = type; 2425 tree_init_node(&fs_prio->node, NULL, del_sw_prio); 2426 tree_add_node(&fs_prio->node, &ns->node); 2427 fs_prio->num_levels = num_levels; 2428 fs_prio->prio = prio; 2429 list_add_tail(&fs_prio->node.list, &ns->node.children); 2430 2431 return fs_prio; 2432 } 2433 2434 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns, 2435 unsigned int prio, 2436 int num_levels) 2437 { 2438 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS); 2439 } 2440 2441 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns, 2442 unsigned int prio, int num_levels) 2443 { 2444 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO); 2445 } 2446 2447 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace 2448 *ns) 2449 { 2450 ns->node.type = FS_TYPE_NAMESPACE; 2451 2452 return ns; 2453 } 2454 2455 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio, 2456 int def_miss_act) 2457 { 2458 struct mlx5_flow_namespace *ns; 2459 2460 ns = kzalloc(sizeof(*ns), GFP_KERNEL); 2461 if (!ns) 2462 return ERR_PTR(-ENOMEM); 2463 2464 fs_init_namespace(ns); 2465 ns->def_miss_action = def_miss_act; 2466 tree_init_node(&ns->node, NULL, del_sw_ns); 2467 tree_add_node(&ns->node, &prio->node); 2468 list_add_tail(&ns->node.list, &prio->node.children); 2469 2470 return ns; 2471 } 2472 2473 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio, 2474 struct init_tree_node *prio_metadata) 2475 { 2476 struct fs_prio *fs_prio; 2477 int i; 2478 2479 for (i = 0; i < prio_metadata->num_leaf_prios; i++) { 2480 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels); 2481 if (IS_ERR(fs_prio)) 2482 return PTR_ERR(fs_prio); 2483 } 2484 return 0; 2485 } 2486 2487 #define FLOW_TABLE_BIT_SZ 1 2488 #define GET_FLOW_TABLE_CAP(dev, offset) \ 2489 ((be32_to_cpu(*((__be32 *)(dev->caps.hca[MLX5_CAP_FLOW_TABLE]->cur) + \ 2490 offset / 32)) >> \ 2491 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ) 2492 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps) 2493 { 2494 int i; 2495 2496 for (i = 0; i < caps->arr_sz; i++) { 2497 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i])) 2498 return false; 2499 } 2500 return true; 2501 } 2502 2503 static int init_root_tree_recursive(struct mlx5_flow_steering *steering, 2504 struct init_tree_node *init_node, 2505 struct fs_node *fs_parent_node, 2506 struct init_tree_node *init_parent_node, 2507 int prio) 2508 { 2509 int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev, 2510 flow_table_properties_nic_receive. 2511 max_ft_level); 2512 struct mlx5_flow_namespace *fs_ns; 2513 struct fs_prio *fs_prio; 2514 struct fs_node *base; 2515 int i; 2516 int err; 2517 2518 if (init_node->type == FS_TYPE_PRIO) { 2519 if ((init_node->min_ft_level > max_ft_level) || 2520 !has_required_caps(steering->dev, &init_node->caps)) 2521 return 0; 2522 2523 fs_get_obj(fs_ns, fs_parent_node); 2524 if (init_node->num_leaf_prios) 2525 return create_leaf_prios(fs_ns, prio, init_node); 2526 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels); 2527 if (IS_ERR(fs_prio)) 2528 return PTR_ERR(fs_prio); 2529 base = &fs_prio->node; 2530 } else if (init_node->type == FS_TYPE_NAMESPACE) { 2531 fs_get_obj(fs_prio, fs_parent_node); 2532 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action); 2533 if (IS_ERR(fs_ns)) 2534 return PTR_ERR(fs_ns); 2535 base = &fs_ns->node; 2536 } else { 2537 return -EINVAL; 2538 } 2539 prio = 0; 2540 for (i = 0; i < init_node->ar_size; i++) { 2541 err = init_root_tree_recursive(steering, &init_node->children[i], 2542 base, init_node, prio); 2543 if (err) 2544 return err; 2545 if (init_node->children[i].type == FS_TYPE_PRIO && 2546 init_node->children[i].num_leaf_prios) { 2547 prio += init_node->children[i].num_leaf_prios; 2548 } 2549 } 2550 2551 return 0; 2552 } 2553 2554 static int init_root_tree(struct mlx5_flow_steering *steering, 2555 struct init_tree_node *init_node, 2556 struct fs_node *fs_parent_node) 2557 { 2558 int err; 2559 int i; 2560 2561 for (i = 0; i < init_node->ar_size; i++) { 2562 err = init_root_tree_recursive(steering, &init_node->children[i], 2563 fs_parent_node, 2564 init_node, i); 2565 if (err) 2566 return err; 2567 } 2568 return 0; 2569 } 2570 2571 static void del_sw_root_ns(struct fs_node *node) 2572 { 2573 struct mlx5_flow_root_namespace *root_ns; 2574 struct mlx5_flow_namespace *ns; 2575 2576 fs_get_obj(ns, node); 2577 root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns); 2578 mutex_destroy(&root_ns->chain_lock); 2579 kfree(node); 2580 } 2581 2582 static struct mlx5_flow_root_namespace 2583 *create_root_ns(struct mlx5_flow_steering *steering, 2584 enum fs_flow_table_type table_type) 2585 { 2586 const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type); 2587 struct mlx5_flow_root_namespace *root_ns; 2588 struct mlx5_flow_namespace *ns; 2589 2590 /* Create the root namespace */ 2591 root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL); 2592 if (!root_ns) 2593 return NULL; 2594 2595 root_ns->dev = steering->dev; 2596 root_ns->table_type = table_type; 2597 root_ns->cmds = cmds; 2598 2599 INIT_LIST_HEAD(&root_ns->underlay_qpns); 2600 2601 ns = &root_ns->ns; 2602 fs_init_namespace(ns); 2603 mutex_init(&root_ns->chain_lock); 2604 tree_init_node(&ns->node, NULL, del_sw_root_ns); 2605 tree_add_node(&ns->node, NULL); 2606 2607 return root_ns; 2608 } 2609 2610 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level); 2611 2612 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level) 2613 { 2614 struct fs_prio *prio; 2615 2616 fs_for_each_prio(prio, ns) { 2617 /* This updates prio start_level and num_levels */ 2618 set_prio_attrs_in_prio(prio, acc_level); 2619 acc_level += prio->num_levels; 2620 } 2621 return acc_level; 2622 } 2623 2624 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level) 2625 { 2626 struct mlx5_flow_namespace *ns; 2627 int acc_level_ns = acc_level; 2628 2629 prio->start_level = acc_level; 2630 fs_for_each_ns(ns, prio) { 2631 /* This updates start_level and num_levels of ns's priority descendants */ 2632 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level); 2633 2634 /* If this a prio with chains, and we can jump from one chain 2635 * (namespace) to another, so we accumulate the levels 2636 */ 2637 if (prio->node.type == FS_TYPE_PRIO_CHAINS) 2638 acc_level = acc_level_ns; 2639 } 2640 2641 if (!prio->num_levels) 2642 prio->num_levels = acc_level_ns - prio->start_level; 2643 WARN_ON(prio->num_levels < acc_level_ns - prio->start_level); 2644 } 2645 2646 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns) 2647 { 2648 struct mlx5_flow_namespace *ns = &root_ns->ns; 2649 struct fs_prio *prio; 2650 int start_level = 0; 2651 2652 fs_for_each_prio(prio, ns) { 2653 set_prio_attrs_in_prio(prio, start_level); 2654 start_level += prio->num_levels; 2655 } 2656 } 2657 2658 #define ANCHOR_PRIO 0 2659 #define ANCHOR_SIZE 1 2660 #define ANCHOR_LEVEL 0 2661 static int create_anchor_flow_table(struct mlx5_flow_steering *steering) 2662 { 2663 struct mlx5_flow_namespace *ns = NULL; 2664 struct mlx5_flow_table_attr ft_attr = {}; 2665 struct mlx5_flow_table *ft; 2666 2667 ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR); 2668 if (WARN_ON(!ns)) 2669 return -EINVAL; 2670 2671 ft_attr.max_fte = ANCHOR_SIZE; 2672 ft_attr.level = ANCHOR_LEVEL; 2673 ft_attr.prio = ANCHOR_PRIO; 2674 2675 ft = mlx5_create_flow_table(ns, &ft_attr); 2676 if (IS_ERR(ft)) { 2677 mlx5_core_err(steering->dev, "Failed to create last anchor flow table"); 2678 return PTR_ERR(ft); 2679 } 2680 return 0; 2681 } 2682 2683 static int init_root_ns(struct mlx5_flow_steering *steering) 2684 { 2685 int err; 2686 2687 steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX); 2688 if (!steering->root_ns) 2689 return -ENOMEM; 2690 2691 err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node); 2692 if (err) 2693 goto out_err; 2694 2695 set_prio_attrs(steering->root_ns); 2696 err = create_anchor_flow_table(steering); 2697 if (err) 2698 goto out_err; 2699 2700 return 0; 2701 2702 out_err: 2703 cleanup_root_ns(steering->root_ns); 2704 steering->root_ns = NULL; 2705 return err; 2706 } 2707 2708 static void clean_tree(struct fs_node *node) 2709 { 2710 if (node) { 2711 struct fs_node *iter; 2712 struct fs_node *temp; 2713 2714 tree_get_node(node); 2715 list_for_each_entry_safe(iter, temp, &node->children, list) 2716 clean_tree(iter); 2717 tree_put_node(node, false); 2718 tree_remove_node(node, false); 2719 } 2720 } 2721 2722 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns) 2723 { 2724 if (!root_ns) 2725 return; 2726 2727 clean_tree(&root_ns->ns.node); 2728 } 2729 2730 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering) 2731 { 2732 struct fs_prio *prio; 2733 2734 steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX); 2735 if (!steering->sniffer_tx_root_ns) 2736 return -ENOMEM; 2737 2738 /* Create single prio */ 2739 prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1); 2740 return PTR_ERR_OR_ZERO(prio); 2741 } 2742 2743 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering) 2744 { 2745 struct fs_prio *prio; 2746 2747 steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX); 2748 if (!steering->sniffer_rx_root_ns) 2749 return -ENOMEM; 2750 2751 /* Create single prio */ 2752 prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1); 2753 return PTR_ERR_OR_ZERO(prio); 2754 } 2755 2756 #define PORT_SEL_NUM_LEVELS 3 2757 static int init_port_sel_root_ns(struct mlx5_flow_steering *steering) 2758 { 2759 struct fs_prio *prio; 2760 2761 steering->port_sel_root_ns = create_root_ns(steering, FS_FT_PORT_SEL); 2762 if (!steering->port_sel_root_ns) 2763 return -ENOMEM; 2764 2765 /* Create single prio */ 2766 prio = fs_create_prio(&steering->port_sel_root_ns->ns, 0, 2767 PORT_SEL_NUM_LEVELS); 2768 return PTR_ERR_OR_ZERO(prio); 2769 } 2770 2771 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering) 2772 { 2773 int err; 2774 2775 steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX); 2776 if (!steering->rdma_rx_root_ns) 2777 return -ENOMEM; 2778 2779 err = init_root_tree(steering, &rdma_rx_root_fs, 2780 &steering->rdma_rx_root_ns->ns.node); 2781 if (err) 2782 goto out_err; 2783 2784 set_prio_attrs(steering->rdma_rx_root_ns); 2785 2786 return 0; 2787 2788 out_err: 2789 cleanup_root_ns(steering->rdma_rx_root_ns); 2790 steering->rdma_rx_root_ns = NULL; 2791 return err; 2792 } 2793 2794 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering) 2795 { 2796 int err; 2797 2798 steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX); 2799 if (!steering->rdma_tx_root_ns) 2800 return -ENOMEM; 2801 2802 err = init_root_tree(steering, &rdma_tx_root_fs, 2803 &steering->rdma_tx_root_ns->ns.node); 2804 if (err) 2805 goto out_err; 2806 2807 set_prio_attrs(steering->rdma_tx_root_ns); 2808 2809 return 0; 2810 2811 out_err: 2812 cleanup_root_ns(steering->rdma_tx_root_ns); 2813 steering->rdma_tx_root_ns = NULL; 2814 return err; 2815 } 2816 2817 /* FT and tc chains are stored in the same array so we can re-use the 2818 * mlx5_get_fdb_sub_ns() and tc api for FT chains. 2819 * When creating a new ns for each chain store it in the first available slot. 2820 * Assume tc chains are created and stored first and only then the FT chain. 2821 */ 2822 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering, 2823 struct mlx5_flow_namespace *ns) 2824 { 2825 int chain = 0; 2826 2827 while (steering->fdb_sub_ns[chain]) 2828 ++chain; 2829 2830 steering->fdb_sub_ns[chain] = ns; 2831 } 2832 2833 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering, 2834 struct fs_prio *maj_prio) 2835 { 2836 struct mlx5_flow_namespace *ns; 2837 struct fs_prio *min_prio; 2838 int prio; 2839 2840 ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF); 2841 if (IS_ERR(ns)) 2842 return PTR_ERR(ns); 2843 2844 for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) { 2845 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO); 2846 if (IS_ERR(min_prio)) 2847 return PTR_ERR(min_prio); 2848 } 2849 2850 store_fdb_sub_ns_prio_chain(steering, ns); 2851 2852 return 0; 2853 } 2854 2855 static int create_fdb_chains(struct mlx5_flow_steering *steering, 2856 int fs_prio, 2857 int chains) 2858 { 2859 struct fs_prio *maj_prio; 2860 int levels; 2861 int chain; 2862 int err; 2863 2864 levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains; 2865 maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns, 2866 fs_prio, 2867 levels); 2868 if (IS_ERR(maj_prio)) 2869 return PTR_ERR(maj_prio); 2870 2871 for (chain = 0; chain < chains; chain++) { 2872 err = create_fdb_sub_ns_prio_chain(steering, maj_prio); 2873 if (err) 2874 return err; 2875 } 2876 2877 return 0; 2878 } 2879 2880 static int create_fdb_fast_path(struct mlx5_flow_steering *steering) 2881 { 2882 int err; 2883 2884 steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS, 2885 sizeof(*steering->fdb_sub_ns), 2886 GFP_KERNEL); 2887 if (!steering->fdb_sub_ns) 2888 return -ENOMEM; 2889 2890 err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1); 2891 if (err) 2892 return err; 2893 2894 err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1); 2895 if (err) 2896 return err; 2897 2898 return 0; 2899 } 2900 2901 static int create_fdb_bypass(struct mlx5_flow_steering *steering) 2902 { 2903 struct mlx5_flow_namespace *ns; 2904 struct fs_prio *prio; 2905 int i; 2906 2907 prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH, 0); 2908 if (IS_ERR(prio)) 2909 return PTR_ERR(prio); 2910 2911 ns = fs_create_namespace(prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF); 2912 if (IS_ERR(ns)) 2913 return PTR_ERR(ns); 2914 2915 for (i = 0; i < MLX5_BY_PASS_NUM_REGULAR_PRIOS; i++) { 2916 prio = fs_create_prio(ns, i, 1); 2917 if (IS_ERR(prio)) 2918 return PTR_ERR(prio); 2919 } 2920 return 0; 2921 } 2922 2923 static void cleanup_fdb_root_ns(struct mlx5_flow_steering *steering) 2924 { 2925 cleanup_root_ns(steering->fdb_root_ns); 2926 steering->fdb_root_ns = NULL; 2927 kfree(steering->fdb_sub_ns); 2928 steering->fdb_sub_ns = NULL; 2929 } 2930 2931 static int init_fdb_root_ns(struct mlx5_flow_steering *steering) 2932 { 2933 struct fs_prio *maj_prio; 2934 int err; 2935 2936 steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB); 2937 if (!steering->fdb_root_ns) 2938 return -ENOMEM; 2939 2940 err = create_fdb_bypass(steering); 2941 if (err) 2942 goto out_err; 2943 2944 err = create_fdb_fast_path(steering); 2945 if (err) 2946 goto out_err; 2947 2948 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1); 2949 if (IS_ERR(maj_prio)) { 2950 err = PTR_ERR(maj_prio); 2951 goto out_err; 2952 } 2953 2954 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 3); 2955 if (IS_ERR(maj_prio)) { 2956 err = PTR_ERR(maj_prio); 2957 goto out_err; 2958 } 2959 2960 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1); 2961 if (IS_ERR(maj_prio)) { 2962 err = PTR_ERR(maj_prio); 2963 goto out_err; 2964 } 2965 2966 /* We put this priority last, knowing that nothing will get here 2967 * unless explicitly forwarded to. This is possible because the 2968 * slow path tables have catch all rules and nothing gets passed 2969 * those tables. 2970 */ 2971 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1); 2972 if (IS_ERR(maj_prio)) { 2973 err = PTR_ERR(maj_prio); 2974 goto out_err; 2975 } 2976 2977 set_prio_attrs(steering->fdb_root_ns); 2978 return 0; 2979 2980 out_err: 2981 cleanup_fdb_root_ns(steering); 2982 return err; 2983 } 2984 2985 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport) 2986 { 2987 struct fs_prio *prio; 2988 2989 steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL); 2990 if (!steering->esw_egress_root_ns[vport]) 2991 return -ENOMEM; 2992 2993 /* create 1 prio*/ 2994 prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1); 2995 return PTR_ERR_OR_ZERO(prio); 2996 } 2997 2998 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport) 2999 { 3000 struct fs_prio *prio; 3001 3002 steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL); 3003 if (!steering->esw_ingress_root_ns[vport]) 3004 return -ENOMEM; 3005 3006 /* create 1 prio*/ 3007 prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1); 3008 return PTR_ERR_OR_ZERO(prio); 3009 } 3010 3011 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports) 3012 { 3013 struct mlx5_flow_steering *steering = dev->priv.steering; 3014 int err; 3015 int i; 3016 3017 steering->esw_egress_root_ns = 3018 kcalloc(total_vports, 3019 sizeof(*steering->esw_egress_root_ns), 3020 GFP_KERNEL); 3021 if (!steering->esw_egress_root_ns) 3022 return -ENOMEM; 3023 3024 for (i = 0; i < total_vports; i++) { 3025 err = init_egress_acl_root_ns(steering, i); 3026 if (err) 3027 goto cleanup_root_ns; 3028 } 3029 steering->esw_egress_acl_vports = total_vports; 3030 return 0; 3031 3032 cleanup_root_ns: 3033 for (i--; i >= 0; i--) 3034 cleanup_root_ns(steering->esw_egress_root_ns[i]); 3035 kfree(steering->esw_egress_root_ns); 3036 steering->esw_egress_root_ns = NULL; 3037 return err; 3038 } 3039 3040 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev) 3041 { 3042 struct mlx5_flow_steering *steering = dev->priv.steering; 3043 int i; 3044 3045 if (!steering->esw_egress_root_ns) 3046 return; 3047 3048 for (i = 0; i < steering->esw_egress_acl_vports; i++) 3049 cleanup_root_ns(steering->esw_egress_root_ns[i]); 3050 3051 kfree(steering->esw_egress_root_ns); 3052 steering->esw_egress_root_ns = NULL; 3053 } 3054 3055 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports) 3056 { 3057 struct mlx5_flow_steering *steering = dev->priv.steering; 3058 int err; 3059 int i; 3060 3061 steering->esw_ingress_root_ns = 3062 kcalloc(total_vports, 3063 sizeof(*steering->esw_ingress_root_ns), 3064 GFP_KERNEL); 3065 if (!steering->esw_ingress_root_ns) 3066 return -ENOMEM; 3067 3068 for (i = 0; i < total_vports; i++) { 3069 err = init_ingress_acl_root_ns(steering, i); 3070 if (err) 3071 goto cleanup_root_ns; 3072 } 3073 steering->esw_ingress_acl_vports = total_vports; 3074 return 0; 3075 3076 cleanup_root_ns: 3077 for (i--; i >= 0; i--) 3078 cleanup_root_ns(steering->esw_ingress_root_ns[i]); 3079 kfree(steering->esw_ingress_root_ns); 3080 steering->esw_ingress_root_ns = NULL; 3081 return err; 3082 } 3083 3084 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev) 3085 { 3086 struct mlx5_flow_steering *steering = dev->priv.steering; 3087 int i; 3088 3089 if (!steering->esw_ingress_root_ns) 3090 return; 3091 3092 for (i = 0; i < steering->esw_ingress_acl_vports; i++) 3093 cleanup_root_ns(steering->esw_ingress_root_ns[i]); 3094 3095 kfree(steering->esw_ingress_root_ns); 3096 steering->esw_ingress_root_ns = NULL; 3097 } 3098 3099 u32 mlx5_fs_get_capabilities(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type type) 3100 { 3101 struct mlx5_flow_root_namespace *root; 3102 struct mlx5_flow_namespace *ns; 3103 3104 ns = mlx5_get_flow_namespace(dev, type); 3105 if (!ns) 3106 return 0; 3107 3108 root = find_root(&ns->node); 3109 if (!root) 3110 return 0; 3111 3112 return root->cmds->get_capabilities(root, root->table_type); 3113 } 3114 3115 static int init_egress_root_ns(struct mlx5_flow_steering *steering) 3116 { 3117 int err; 3118 3119 steering->egress_root_ns = create_root_ns(steering, 3120 FS_FT_NIC_TX); 3121 if (!steering->egress_root_ns) 3122 return -ENOMEM; 3123 3124 err = init_root_tree(steering, &egress_root_fs, 3125 &steering->egress_root_ns->ns.node); 3126 if (err) 3127 goto cleanup; 3128 set_prio_attrs(steering->egress_root_ns); 3129 return 0; 3130 cleanup: 3131 cleanup_root_ns(steering->egress_root_ns); 3132 steering->egress_root_ns = NULL; 3133 return err; 3134 } 3135 3136 void mlx5_fs_core_cleanup(struct mlx5_core_dev *dev) 3137 { 3138 struct mlx5_flow_steering *steering = dev->priv.steering; 3139 3140 cleanup_root_ns(steering->root_ns); 3141 cleanup_fdb_root_ns(steering); 3142 cleanup_root_ns(steering->port_sel_root_ns); 3143 cleanup_root_ns(steering->sniffer_rx_root_ns); 3144 cleanup_root_ns(steering->sniffer_tx_root_ns); 3145 cleanup_root_ns(steering->rdma_rx_root_ns); 3146 cleanup_root_ns(steering->rdma_tx_root_ns); 3147 cleanup_root_ns(steering->egress_root_ns); 3148 } 3149 3150 int mlx5_fs_core_init(struct mlx5_core_dev *dev) 3151 { 3152 struct mlx5_flow_steering *steering = dev->priv.steering; 3153 int err = 0; 3154 3155 if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) && 3156 (MLX5_CAP_GEN(dev, nic_flow_table))) || 3157 ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) && 3158 MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) && 3159 MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) { 3160 err = init_root_ns(steering); 3161 if (err) 3162 goto err; 3163 } 3164 3165 if (MLX5_ESWITCH_MANAGER(dev)) { 3166 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) { 3167 err = init_fdb_root_ns(steering); 3168 if (err) 3169 goto err; 3170 } 3171 } 3172 3173 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) { 3174 err = init_sniffer_rx_root_ns(steering); 3175 if (err) 3176 goto err; 3177 } 3178 3179 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) { 3180 err = init_sniffer_tx_root_ns(steering); 3181 if (err) 3182 goto err; 3183 } 3184 3185 if (MLX5_CAP_FLOWTABLE_PORT_SELECTION(dev, ft_support)) { 3186 err = init_port_sel_root_ns(steering); 3187 if (err) 3188 goto err; 3189 } 3190 3191 if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) && 3192 MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) { 3193 err = init_rdma_rx_root_ns(steering); 3194 if (err) 3195 goto err; 3196 } 3197 3198 if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) { 3199 err = init_rdma_tx_root_ns(steering); 3200 if (err) 3201 goto err; 3202 } 3203 3204 if (MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) { 3205 err = init_egress_root_ns(steering); 3206 if (err) 3207 goto err; 3208 } 3209 3210 return 0; 3211 3212 err: 3213 mlx5_fs_core_cleanup(dev); 3214 return err; 3215 } 3216 3217 void mlx5_fs_core_free(struct mlx5_core_dev *dev) 3218 { 3219 struct mlx5_flow_steering *steering = dev->priv.steering; 3220 3221 kmem_cache_destroy(steering->ftes_cache); 3222 kmem_cache_destroy(steering->fgs_cache); 3223 kfree(steering); 3224 mlx5_ft_pool_destroy(dev); 3225 mlx5_cleanup_fc_stats(dev); 3226 } 3227 3228 int mlx5_fs_core_alloc(struct mlx5_core_dev *dev) 3229 { 3230 struct mlx5_flow_steering *steering; 3231 int err = 0; 3232 3233 err = mlx5_init_fc_stats(dev); 3234 if (err) 3235 return err; 3236 3237 err = mlx5_ft_pool_init(dev); 3238 if (err) 3239 goto err; 3240 3241 steering = kzalloc(sizeof(*steering), GFP_KERNEL); 3242 if (!steering) { 3243 err = -ENOMEM; 3244 goto err; 3245 } 3246 3247 steering->dev = dev; 3248 dev->priv.steering = steering; 3249 3250 if (mlx5_fs_dr_is_supported(dev)) 3251 steering->mode = MLX5_FLOW_STEERING_MODE_SMFS; 3252 else 3253 steering->mode = MLX5_FLOW_STEERING_MODE_DMFS; 3254 3255 steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs", 3256 sizeof(struct mlx5_flow_group), 0, 3257 0, NULL); 3258 steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0, 3259 0, NULL); 3260 if (!steering->ftes_cache || !steering->fgs_cache) { 3261 err = -ENOMEM; 3262 goto err; 3263 } 3264 3265 return 0; 3266 3267 err: 3268 mlx5_fs_core_free(dev); 3269 return err; 3270 } 3271 3272 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn) 3273 { 3274 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns; 3275 struct mlx5_ft_underlay_qp *new_uqp; 3276 int err = 0; 3277 3278 new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL); 3279 if (!new_uqp) 3280 return -ENOMEM; 3281 3282 mutex_lock(&root->chain_lock); 3283 3284 if (!root->root_ft) { 3285 err = -EINVAL; 3286 goto update_ft_fail; 3287 } 3288 3289 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn, 3290 false); 3291 if (err) { 3292 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n", 3293 underlay_qpn, err); 3294 goto update_ft_fail; 3295 } 3296 3297 new_uqp->qpn = underlay_qpn; 3298 list_add_tail(&new_uqp->list, &root->underlay_qpns); 3299 3300 mutex_unlock(&root->chain_lock); 3301 3302 return 0; 3303 3304 update_ft_fail: 3305 mutex_unlock(&root->chain_lock); 3306 kfree(new_uqp); 3307 return err; 3308 } 3309 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn); 3310 3311 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn) 3312 { 3313 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns; 3314 struct mlx5_ft_underlay_qp *uqp; 3315 bool found = false; 3316 int err = 0; 3317 3318 mutex_lock(&root->chain_lock); 3319 list_for_each_entry(uqp, &root->underlay_qpns, list) { 3320 if (uqp->qpn == underlay_qpn) { 3321 found = true; 3322 break; 3323 } 3324 } 3325 3326 if (!found) { 3327 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n", 3328 underlay_qpn); 3329 err = -EINVAL; 3330 goto out; 3331 } 3332 3333 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn, 3334 true); 3335 if (err) 3336 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n", 3337 underlay_qpn, err); 3338 3339 list_del(&uqp->list); 3340 mutex_unlock(&root->chain_lock); 3341 kfree(uqp); 3342 3343 return 0; 3344 3345 out: 3346 mutex_unlock(&root->chain_lock); 3347 return err; 3348 } 3349 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn); 3350 3351 static struct mlx5_flow_root_namespace 3352 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type) 3353 { 3354 struct mlx5_flow_namespace *ns; 3355 3356 if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS || 3357 ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS) 3358 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0); 3359 else 3360 ns = mlx5_get_flow_namespace(dev, ns_type); 3361 if (!ns) 3362 return NULL; 3363 3364 return find_root(&ns->node); 3365 } 3366 3367 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev, 3368 u8 ns_type, u8 num_actions, 3369 void *modify_actions) 3370 { 3371 struct mlx5_flow_root_namespace *root; 3372 struct mlx5_modify_hdr *modify_hdr; 3373 int err; 3374 3375 root = get_root_namespace(dev, ns_type); 3376 if (!root) 3377 return ERR_PTR(-EOPNOTSUPP); 3378 3379 modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL); 3380 if (!modify_hdr) 3381 return ERR_PTR(-ENOMEM); 3382 3383 modify_hdr->ns_type = ns_type; 3384 err = root->cmds->modify_header_alloc(root, ns_type, num_actions, 3385 modify_actions, modify_hdr); 3386 if (err) { 3387 kfree(modify_hdr); 3388 return ERR_PTR(err); 3389 } 3390 3391 return modify_hdr; 3392 } 3393 EXPORT_SYMBOL(mlx5_modify_header_alloc); 3394 3395 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev, 3396 struct mlx5_modify_hdr *modify_hdr) 3397 { 3398 struct mlx5_flow_root_namespace *root; 3399 3400 root = get_root_namespace(dev, modify_hdr->ns_type); 3401 if (WARN_ON(!root)) 3402 return; 3403 root->cmds->modify_header_dealloc(root, modify_hdr); 3404 kfree(modify_hdr); 3405 } 3406 EXPORT_SYMBOL(mlx5_modify_header_dealloc); 3407 3408 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev, 3409 struct mlx5_pkt_reformat_params *params, 3410 enum mlx5_flow_namespace_type ns_type) 3411 { 3412 struct mlx5_pkt_reformat *pkt_reformat; 3413 struct mlx5_flow_root_namespace *root; 3414 int err; 3415 3416 root = get_root_namespace(dev, ns_type); 3417 if (!root) 3418 return ERR_PTR(-EOPNOTSUPP); 3419 3420 pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL); 3421 if (!pkt_reformat) 3422 return ERR_PTR(-ENOMEM); 3423 3424 pkt_reformat->ns_type = ns_type; 3425 pkt_reformat->reformat_type = params->type; 3426 err = root->cmds->packet_reformat_alloc(root, params, ns_type, 3427 pkt_reformat); 3428 if (err) { 3429 kfree(pkt_reformat); 3430 return ERR_PTR(err); 3431 } 3432 3433 return pkt_reformat; 3434 } 3435 EXPORT_SYMBOL(mlx5_packet_reformat_alloc); 3436 3437 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev, 3438 struct mlx5_pkt_reformat *pkt_reformat) 3439 { 3440 struct mlx5_flow_root_namespace *root; 3441 3442 root = get_root_namespace(dev, pkt_reformat->ns_type); 3443 if (WARN_ON(!root)) 3444 return; 3445 root->cmds->packet_reformat_dealloc(root, pkt_reformat); 3446 kfree(pkt_reformat); 3447 } 3448 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc); 3449 3450 int mlx5_get_match_definer_id(struct mlx5_flow_definer *definer) 3451 { 3452 return definer->id; 3453 } 3454 3455 struct mlx5_flow_definer * 3456 mlx5_create_match_definer(struct mlx5_core_dev *dev, 3457 enum mlx5_flow_namespace_type ns_type, u16 format_id, 3458 u32 *match_mask) 3459 { 3460 struct mlx5_flow_root_namespace *root; 3461 struct mlx5_flow_definer *definer; 3462 int id; 3463 3464 root = get_root_namespace(dev, ns_type); 3465 if (!root) 3466 return ERR_PTR(-EOPNOTSUPP); 3467 3468 definer = kzalloc(sizeof(*definer), GFP_KERNEL); 3469 if (!definer) 3470 return ERR_PTR(-ENOMEM); 3471 3472 definer->ns_type = ns_type; 3473 id = root->cmds->create_match_definer(root, format_id, match_mask); 3474 if (id < 0) { 3475 mlx5_core_warn(root->dev, "Failed to create match definer (%d)\n", id); 3476 kfree(definer); 3477 return ERR_PTR(id); 3478 } 3479 definer->id = id; 3480 return definer; 3481 } 3482 3483 void mlx5_destroy_match_definer(struct mlx5_core_dev *dev, 3484 struct mlx5_flow_definer *definer) 3485 { 3486 struct mlx5_flow_root_namespace *root; 3487 3488 root = get_root_namespace(dev, definer->ns_type); 3489 if (WARN_ON(!root)) 3490 return; 3491 3492 root->cmds->destroy_match_definer(root, definer->id); 3493 kfree(definer); 3494 } 3495 3496 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns, 3497 struct mlx5_flow_root_namespace *peer_ns) 3498 { 3499 if (peer_ns && ns->mode != peer_ns->mode) { 3500 mlx5_core_err(ns->dev, 3501 "Can't peer namespace of different steering mode\n"); 3502 return -EINVAL; 3503 } 3504 3505 return ns->cmds->set_peer(ns, peer_ns); 3506 } 3507 3508 /* This function should be called only at init stage of the namespace. 3509 * It is not safe to call this function while steering operations 3510 * are executed in the namespace. 3511 */ 3512 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns, 3513 enum mlx5_flow_steering_mode mode) 3514 { 3515 struct mlx5_flow_root_namespace *root; 3516 const struct mlx5_flow_cmds *cmds; 3517 int err; 3518 3519 root = find_root(&ns->node); 3520 if (&root->ns != ns) 3521 /* Can't set cmds to non root namespace */ 3522 return -EINVAL; 3523 3524 if (root->table_type != FS_FT_FDB) 3525 return -EOPNOTSUPP; 3526 3527 if (root->mode == mode) 3528 return 0; 3529 3530 if (mode == MLX5_FLOW_STEERING_MODE_SMFS) 3531 cmds = mlx5_fs_cmd_get_dr_cmds(); 3532 else 3533 cmds = mlx5_fs_cmd_get_fw_cmds(); 3534 if (!cmds) 3535 return -EOPNOTSUPP; 3536 3537 err = cmds->create_ns(root); 3538 if (err) { 3539 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n", 3540 err); 3541 return err; 3542 } 3543 3544 root->cmds->destroy_ns(root); 3545 root->cmds = cmds; 3546 root->mode = mode; 3547 3548 return 0; 3549 } 3550