1 /* 2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 */ 32 33 #include <linux/mutex.h> 34 #include <linux/mlx5/driver.h> 35 #include <linux/mlx5/vport.h> 36 #include <linux/mlx5/eswitch.h> 37 38 #include "mlx5_core.h" 39 #include "fs_core.h" 40 #include "fs_cmd.h" 41 #include "fs_ft_pool.h" 42 #include "diag/fs_tracepoint.h" 43 #include "accel/ipsec.h" 44 #include "fpga/ipsec.h" 45 46 #define INIT_TREE_NODE_ARRAY_SIZE(...) (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\ 47 sizeof(struct init_tree_node)) 48 49 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\ 50 ...) {.type = FS_TYPE_PRIO,\ 51 .min_ft_level = min_level_val,\ 52 .num_levels = num_levels_val,\ 53 .num_leaf_prios = num_prios_val,\ 54 .caps = caps_val,\ 55 .children = (struct init_tree_node[]) {__VA_ARGS__},\ 56 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \ 57 } 58 59 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\ 60 ADD_PRIO(num_prios_val, 0, num_levels_val, {},\ 61 __VA_ARGS__)\ 62 63 #define ADD_NS(def_miss_act, ...) {.type = FS_TYPE_NAMESPACE, \ 64 .def_miss_action = def_miss_act,\ 65 .children = (struct init_tree_node[]) {__VA_ARGS__},\ 66 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \ 67 } 68 69 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\ 70 sizeof(long)) 71 72 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap)) 73 74 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \ 75 .caps = (long[]) {__VA_ARGS__} } 76 77 #define FS_CHAINING_CAPS FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \ 78 FS_CAP(flow_table_properties_nic_receive.modify_root), \ 79 FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \ 80 FS_CAP(flow_table_properties_nic_receive.flow_table_modify)) 81 82 #define FS_CHAINING_CAPS_EGRESS \ 83 FS_REQUIRED_CAPS( \ 84 FS_CAP(flow_table_properties_nic_transmit.flow_modify_en), \ 85 FS_CAP(flow_table_properties_nic_transmit.modify_root), \ 86 FS_CAP(flow_table_properties_nic_transmit \ 87 .identified_miss_table_mode), \ 88 FS_CAP(flow_table_properties_nic_transmit.flow_table_modify)) 89 90 #define FS_CHAINING_CAPS_RDMA_TX \ 91 FS_REQUIRED_CAPS( \ 92 FS_CAP(flow_table_properties_nic_transmit_rdma.flow_modify_en), \ 93 FS_CAP(flow_table_properties_nic_transmit_rdma.modify_root), \ 94 FS_CAP(flow_table_properties_nic_transmit_rdma \ 95 .identified_miss_table_mode), \ 96 FS_CAP(flow_table_properties_nic_transmit_rdma \ 97 .flow_table_modify)) 98 99 #define LEFTOVERS_NUM_LEVELS 1 100 #define LEFTOVERS_NUM_PRIOS 1 101 102 #define RDMA_RX_COUNTERS_PRIO_NUM_LEVELS 1 103 #define RDMA_TX_COUNTERS_PRIO_NUM_LEVELS 1 104 105 #define BY_PASS_PRIO_NUM_LEVELS 1 106 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\ 107 LEFTOVERS_NUM_PRIOS) 108 109 #define ETHTOOL_PRIO_NUM_LEVELS 1 110 #define ETHTOOL_NUM_PRIOS 11 111 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS) 112 /* Promiscuous, Vlan, mac, ttc, inner ttc, {UDP/ANY/aRFS/accel/{esp, esp_err}} */ 113 #define KERNEL_NIC_PRIO_NUM_LEVELS 7 114 #define KERNEL_NIC_NUM_PRIOS 1 115 /* One more level for tc */ 116 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1) 117 118 #define KERNEL_NIC_TC_NUM_PRIOS 1 119 #define KERNEL_NIC_TC_NUM_LEVELS 2 120 121 #define ANCHOR_NUM_LEVELS 1 122 #define ANCHOR_NUM_PRIOS 1 123 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1) 124 125 #define OFFLOADS_MAX_FT 2 126 #define OFFLOADS_NUM_PRIOS 2 127 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS) 128 129 #define LAG_PRIO_NUM_LEVELS 1 130 #define LAG_NUM_PRIOS 1 131 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1) 132 133 #define KERNEL_TX_IPSEC_NUM_PRIOS 1 134 #define KERNEL_TX_IPSEC_NUM_LEVELS 1 135 #define KERNEL_TX_MIN_LEVEL (KERNEL_TX_IPSEC_NUM_LEVELS) 136 137 struct node_caps { 138 size_t arr_sz; 139 long *caps; 140 }; 141 142 static struct init_tree_node { 143 enum fs_node_type type; 144 struct init_tree_node *children; 145 int ar_size; 146 struct node_caps caps; 147 int min_ft_level; 148 int num_leaf_prios; 149 int prio; 150 int num_levels; 151 enum mlx5_flow_table_miss_action def_miss_action; 152 } root_fs = { 153 .type = FS_TYPE_NAMESPACE, 154 .ar_size = 7, 155 .children = (struct init_tree_node[]){ 156 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS, 157 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 158 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS, 159 BY_PASS_PRIO_NUM_LEVELS))), 160 ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS, 161 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 162 ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS, 163 LAG_PRIO_NUM_LEVELS))), 164 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS, 165 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 166 ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, 167 OFFLOADS_MAX_FT))), 168 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS, 169 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 170 ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS, 171 ETHTOOL_PRIO_NUM_LEVELS))), 172 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {}, 173 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 174 ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS, 175 KERNEL_NIC_TC_NUM_LEVELS), 176 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS, 177 KERNEL_NIC_PRIO_NUM_LEVELS))), 178 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS, 179 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 180 ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, 181 LEFTOVERS_NUM_LEVELS))), 182 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {}, 183 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 184 ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, 185 ANCHOR_NUM_LEVELS))), 186 } 187 }; 188 189 static struct init_tree_node egress_root_fs = { 190 .type = FS_TYPE_NAMESPACE, 191 #ifdef CONFIG_MLX5_IPSEC 192 .ar_size = 2, 193 #else 194 .ar_size = 1, 195 #endif 196 .children = (struct init_tree_node[]) { 197 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0, 198 FS_CHAINING_CAPS_EGRESS, 199 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 200 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS, 201 BY_PASS_PRIO_NUM_LEVELS))), 202 #ifdef CONFIG_MLX5_IPSEC 203 ADD_PRIO(0, KERNEL_TX_MIN_LEVEL, 0, 204 FS_CHAINING_CAPS_EGRESS, 205 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 206 ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS, 207 KERNEL_TX_IPSEC_NUM_LEVELS))), 208 #endif 209 } 210 }; 211 212 enum { 213 RDMA_RX_COUNTERS_PRIO, 214 RDMA_RX_BYPASS_PRIO, 215 RDMA_RX_KERNEL_PRIO, 216 }; 217 218 #define RDMA_RX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_REGULAR_PRIOS 219 #define RDMA_RX_KERNEL_MIN_LEVEL (RDMA_RX_BYPASS_MIN_LEVEL + 1) 220 #define RDMA_RX_COUNTERS_MIN_LEVEL (RDMA_RX_KERNEL_MIN_LEVEL + 2) 221 222 static struct init_tree_node rdma_rx_root_fs = { 223 .type = FS_TYPE_NAMESPACE, 224 .ar_size = 3, 225 .children = (struct init_tree_node[]) { 226 [RDMA_RX_COUNTERS_PRIO] = 227 ADD_PRIO(0, RDMA_RX_COUNTERS_MIN_LEVEL, 0, 228 FS_CHAINING_CAPS, 229 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 230 ADD_MULTIPLE_PRIO(MLX5_RDMA_RX_NUM_COUNTERS_PRIOS, 231 RDMA_RX_COUNTERS_PRIO_NUM_LEVELS))), 232 [RDMA_RX_BYPASS_PRIO] = 233 ADD_PRIO(0, RDMA_RX_BYPASS_MIN_LEVEL, 0, 234 FS_CHAINING_CAPS, 235 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 236 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS, 237 BY_PASS_PRIO_NUM_LEVELS))), 238 [RDMA_RX_KERNEL_PRIO] = 239 ADD_PRIO(0, RDMA_RX_KERNEL_MIN_LEVEL, 0, 240 FS_CHAINING_CAPS, 241 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN, 242 ADD_MULTIPLE_PRIO(1, 1))), 243 } 244 }; 245 246 enum { 247 RDMA_TX_COUNTERS_PRIO, 248 RDMA_TX_BYPASS_PRIO, 249 }; 250 251 #define RDMA_TX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_PRIOS 252 #define RDMA_TX_COUNTERS_MIN_LEVEL (RDMA_TX_BYPASS_MIN_LEVEL + 1) 253 254 static struct init_tree_node rdma_tx_root_fs = { 255 .type = FS_TYPE_NAMESPACE, 256 .ar_size = 2, 257 .children = (struct init_tree_node[]) { 258 [RDMA_TX_COUNTERS_PRIO] = 259 ADD_PRIO(0, RDMA_TX_COUNTERS_MIN_LEVEL, 0, 260 FS_CHAINING_CAPS, 261 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 262 ADD_MULTIPLE_PRIO(MLX5_RDMA_TX_NUM_COUNTERS_PRIOS, 263 RDMA_TX_COUNTERS_PRIO_NUM_LEVELS))), 264 [RDMA_TX_BYPASS_PRIO] = 265 ADD_PRIO(0, RDMA_TX_BYPASS_MIN_LEVEL, 0, 266 FS_CHAINING_CAPS_RDMA_TX, 267 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 268 ADD_MULTIPLE_PRIO(RDMA_TX_BYPASS_MIN_LEVEL, 269 BY_PASS_PRIO_NUM_LEVELS))), 270 } 271 }; 272 273 enum fs_i_lock_class { 274 FS_LOCK_GRANDPARENT, 275 FS_LOCK_PARENT, 276 FS_LOCK_CHILD 277 }; 278 279 static const struct rhashtable_params rhash_fte = { 280 .key_len = sizeof_field(struct fs_fte, val), 281 .key_offset = offsetof(struct fs_fte, val), 282 .head_offset = offsetof(struct fs_fte, hash), 283 .automatic_shrinking = true, 284 .min_size = 1, 285 }; 286 287 static const struct rhashtable_params rhash_fg = { 288 .key_len = sizeof_field(struct mlx5_flow_group, mask), 289 .key_offset = offsetof(struct mlx5_flow_group, mask), 290 .head_offset = offsetof(struct mlx5_flow_group, hash), 291 .automatic_shrinking = true, 292 .min_size = 1, 293 294 }; 295 296 static void del_hw_flow_table(struct fs_node *node); 297 static void del_hw_flow_group(struct fs_node *node); 298 static void del_hw_fte(struct fs_node *node); 299 static void del_sw_flow_table(struct fs_node *node); 300 static void del_sw_flow_group(struct fs_node *node); 301 static void del_sw_fte(struct fs_node *node); 302 static void del_sw_prio(struct fs_node *node); 303 static void del_sw_ns(struct fs_node *node); 304 /* Delete rule (destination) is special case that 305 * requires to lock the FTE for all the deletion process. 306 */ 307 static void del_sw_hw_rule(struct fs_node *node); 308 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1, 309 struct mlx5_flow_destination *d2); 310 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns); 311 static struct mlx5_flow_rule * 312 find_flow_rule(struct fs_fte *fte, 313 struct mlx5_flow_destination *dest); 314 315 static void tree_init_node(struct fs_node *node, 316 void (*del_hw_func)(struct fs_node *), 317 void (*del_sw_func)(struct fs_node *)) 318 { 319 refcount_set(&node->refcount, 1); 320 INIT_LIST_HEAD(&node->list); 321 INIT_LIST_HEAD(&node->children); 322 init_rwsem(&node->lock); 323 node->del_hw_func = del_hw_func; 324 node->del_sw_func = del_sw_func; 325 node->active = false; 326 } 327 328 static void tree_add_node(struct fs_node *node, struct fs_node *parent) 329 { 330 if (parent) 331 refcount_inc(&parent->refcount); 332 node->parent = parent; 333 334 /* Parent is the root */ 335 if (!parent) 336 node->root = node; 337 else 338 node->root = parent->root; 339 } 340 341 static int tree_get_node(struct fs_node *node) 342 { 343 return refcount_inc_not_zero(&node->refcount); 344 } 345 346 static void nested_down_read_ref_node(struct fs_node *node, 347 enum fs_i_lock_class class) 348 { 349 if (node) { 350 down_read_nested(&node->lock, class); 351 refcount_inc(&node->refcount); 352 } 353 } 354 355 static void nested_down_write_ref_node(struct fs_node *node, 356 enum fs_i_lock_class class) 357 { 358 if (node) { 359 down_write_nested(&node->lock, class); 360 refcount_inc(&node->refcount); 361 } 362 } 363 364 static void down_write_ref_node(struct fs_node *node, bool locked) 365 { 366 if (node) { 367 if (!locked) 368 down_write(&node->lock); 369 refcount_inc(&node->refcount); 370 } 371 } 372 373 static void up_read_ref_node(struct fs_node *node) 374 { 375 refcount_dec(&node->refcount); 376 up_read(&node->lock); 377 } 378 379 static void up_write_ref_node(struct fs_node *node, bool locked) 380 { 381 refcount_dec(&node->refcount); 382 if (!locked) 383 up_write(&node->lock); 384 } 385 386 static void tree_put_node(struct fs_node *node, bool locked) 387 { 388 struct fs_node *parent_node = node->parent; 389 390 if (refcount_dec_and_test(&node->refcount)) { 391 if (node->del_hw_func) 392 node->del_hw_func(node); 393 if (parent_node) { 394 down_write_ref_node(parent_node, locked); 395 list_del_init(&node->list); 396 } 397 node->del_sw_func(node); 398 if (parent_node) 399 up_write_ref_node(parent_node, locked); 400 node = NULL; 401 } 402 if (!node && parent_node) 403 tree_put_node(parent_node, locked); 404 } 405 406 static int tree_remove_node(struct fs_node *node, bool locked) 407 { 408 if (refcount_read(&node->refcount) > 1) { 409 refcount_dec(&node->refcount); 410 return -EEXIST; 411 } 412 tree_put_node(node, locked); 413 return 0; 414 } 415 416 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns, 417 unsigned int prio) 418 { 419 struct fs_prio *iter_prio; 420 421 fs_for_each_prio(iter_prio, ns) { 422 if (iter_prio->prio == prio) 423 return iter_prio; 424 } 425 426 return NULL; 427 } 428 429 static bool is_fwd_next_action(u32 action) 430 { 431 return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO | 432 MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS); 433 } 434 435 static bool check_valid_spec(const struct mlx5_flow_spec *spec) 436 { 437 int i; 438 439 for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++) 440 if (spec->match_value[i] & ~spec->match_criteria[i]) { 441 pr_warn("mlx5_core: match_value differs from match_criteria\n"); 442 return false; 443 } 444 445 return true; 446 } 447 448 struct mlx5_flow_root_namespace *find_root(struct fs_node *node) 449 { 450 struct fs_node *root; 451 struct mlx5_flow_namespace *ns; 452 453 root = node->root; 454 455 if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) { 456 pr_warn("mlx5: flow steering node is not in tree or garbaged\n"); 457 return NULL; 458 } 459 460 ns = container_of(root, struct mlx5_flow_namespace, node); 461 return container_of(ns, struct mlx5_flow_root_namespace, ns); 462 } 463 464 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node) 465 { 466 struct mlx5_flow_root_namespace *root = find_root(node); 467 468 if (root) 469 return root->dev->priv.steering; 470 return NULL; 471 } 472 473 static inline struct mlx5_core_dev *get_dev(struct fs_node *node) 474 { 475 struct mlx5_flow_root_namespace *root = find_root(node); 476 477 if (root) 478 return root->dev; 479 return NULL; 480 } 481 482 static void del_sw_ns(struct fs_node *node) 483 { 484 kfree(node); 485 } 486 487 static void del_sw_prio(struct fs_node *node) 488 { 489 kfree(node); 490 } 491 492 static void del_hw_flow_table(struct fs_node *node) 493 { 494 struct mlx5_flow_root_namespace *root; 495 struct mlx5_flow_table *ft; 496 struct mlx5_core_dev *dev; 497 int err; 498 499 fs_get_obj(ft, node); 500 dev = get_dev(&ft->node); 501 root = find_root(&ft->node); 502 trace_mlx5_fs_del_ft(ft); 503 504 if (node->active) { 505 err = root->cmds->destroy_flow_table(root, ft); 506 if (err) 507 mlx5_core_warn(dev, "flow steering can't destroy ft\n"); 508 } 509 } 510 511 static void del_sw_flow_table(struct fs_node *node) 512 { 513 struct mlx5_flow_table *ft; 514 struct fs_prio *prio; 515 516 fs_get_obj(ft, node); 517 518 rhltable_destroy(&ft->fgs_hash); 519 if (ft->node.parent) { 520 fs_get_obj(prio, ft->node.parent); 521 prio->num_ft--; 522 } 523 kfree(ft); 524 } 525 526 static void modify_fte(struct fs_fte *fte) 527 { 528 struct mlx5_flow_root_namespace *root; 529 struct mlx5_flow_table *ft; 530 struct mlx5_flow_group *fg; 531 struct mlx5_core_dev *dev; 532 int err; 533 534 fs_get_obj(fg, fte->node.parent); 535 fs_get_obj(ft, fg->node.parent); 536 dev = get_dev(&fte->node); 537 538 root = find_root(&ft->node); 539 err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte); 540 if (err) 541 mlx5_core_warn(dev, 542 "%s can't del rule fg id=%d fte_index=%d\n", 543 __func__, fg->id, fte->index); 544 fte->modify_mask = 0; 545 } 546 547 static void del_sw_hw_rule(struct fs_node *node) 548 { 549 struct mlx5_flow_rule *rule; 550 struct fs_fte *fte; 551 552 fs_get_obj(rule, node); 553 fs_get_obj(fte, rule->node.parent); 554 trace_mlx5_fs_del_rule(rule); 555 if (is_fwd_next_action(rule->sw_action)) { 556 mutex_lock(&rule->dest_attr.ft->lock); 557 list_del(&rule->next_ft); 558 mutex_unlock(&rule->dest_attr.ft->lock); 559 } 560 561 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER && 562 --fte->dests_size) { 563 fte->modify_mask |= 564 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) | 565 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS); 566 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT; 567 goto out; 568 } 569 570 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT && 571 --fte->dests_size) { 572 fte->modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION); 573 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW; 574 goto out; 575 } 576 577 if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) && 578 --fte->dests_size) { 579 fte->modify_mask |= 580 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST); 581 } 582 out: 583 kfree(rule); 584 } 585 586 static void del_hw_fte(struct fs_node *node) 587 { 588 struct mlx5_flow_root_namespace *root; 589 struct mlx5_flow_table *ft; 590 struct mlx5_flow_group *fg; 591 struct mlx5_core_dev *dev; 592 struct fs_fte *fte; 593 int err; 594 595 fs_get_obj(fte, node); 596 fs_get_obj(fg, fte->node.parent); 597 fs_get_obj(ft, fg->node.parent); 598 599 trace_mlx5_fs_del_fte(fte); 600 dev = get_dev(&ft->node); 601 root = find_root(&ft->node); 602 if (node->active) { 603 err = root->cmds->delete_fte(root, ft, fte); 604 if (err) 605 mlx5_core_warn(dev, 606 "flow steering can't delete fte in index %d of flow group id %d\n", 607 fte->index, fg->id); 608 node->active = false; 609 } 610 } 611 612 static void del_sw_fte(struct fs_node *node) 613 { 614 struct mlx5_flow_steering *steering = get_steering(node); 615 struct mlx5_flow_group *fg; 616 struct fs_fte *fte; 617 int err; 618 619 fs_get_obj(fte, node); 620 fs_get_obj(fg, fte->node.parent); 621 622 err = rhashtable_remove_fast(&fg->ftes_hash, 623 &fte->hash, 624 rhash_fte); 625 WARN_ON(err); 626 ida_free(&fg->fte_allocator, fte->index - fg->start_index); 627 kmem_cache_free(steering->ftes_cache, fte); 628 } 629 630 static void del_hw_flow_group(struct fs_node *node) 631 { 632 struct mlx5_flow_root_namespace *root; 633 struct mlx5_flow_group *fg; 634 struct mlx5_flow_table *ft; 635 struct mlx5_core_dev *dev; 636 637 fs_get_obj(fg, node); 638 fs_get_obj(ft, fg->node.parent); 639 dev = get_dev(&ft->node); 640 trace_mlx5_fs_del_fg(fg); 641 642 root = find_root(&ft->node); 643 if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg)) 644 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n", 645 fg->id, ft->id); 646 } 647 648 static void del_sw_flow_group(struct fs_node *node) 649 { 650 struct mlx5_flow_steering *steering = get_steering(node); 651 struct mlx5_flow_group *fg; 652 struct mlx5_flow_table *ft; 653 int err; 654 655 fs_get_obj(fg, node); 656 fs_get_obj(ft, fg->node.parent); 657 658 rhashtable_destroy(&fg->ftes_hash); 659 ida_destroy(&fg->fte_allocator); 660 if (ft->autogroup.active && 661 fg->max_ftes == ft->autogroup.group_size && 662 fg->start_index < ft->autogroup.max_fte) 663 ft->autogroup.num_groups--; 664 err = rhltable_remove(&ft->fgs_hash, 665 &fg->hash, 666 rhash_fg); 667 WARN_ON(err); 668 kmem_cache_free(steering->fgs_cache, fg); 669 } 670 671 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte) 672 { 673 int index; 674 int ret; 675 676 index = ida_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, GFP_KERNEL); 677 if (index < 0) 678 return index; 679 680 fte->index = index + fg->start_index; 681 ret = rhashtable_insert_fast(&fg->ftes_hash, 682 &fte->hash, 683 rhash_fte); 684 if (ret) 685 goto err_ida_remove; 686 687 tree_add_node(&fte->node, &fg->node); 688 list_add_tail(&fte->node.list, &fg->node.children); 689 return 0; 690 691 err_ida_remove: 692 ida_free(&fg->fte_allocator, index); 693 return ret; 694 } 695 696 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft, 697 const struct mlx5_flow_spec *spec, 698 struct mlx5_flow_act *flow_act) 699 { 700 struct mlx5_flow_steering *steering = get_steering(&ft->node); 701 struct fs_fte *fte; 702 703 fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL); 704 if (!fte) 705 return ERR_PTR(-ENOMEM); 706 707 memcpy(fte->val, &spec->match_value, sizeof(fte->val)); 708 fte->node.type = FS_TYPE_FLOW_ENTRY; 709 fte->action = *flow_act; 710 fte->flow_context = spec->flow_context; 711 712 tree_init_node(&fte->node, del_hw_fte, del_sw_fte); 713 714 return fte; 715 } 716 717 static void dealloc_flow_group(struct mlx5_flow_steering *steering, 718 struct mlx5_flow_group *fg) 719 { 720 rhashtable_destroy(&fg->ftes_hash); 721 kmem_cache_free(steering->fgs_cache, fg); 722 } 723 724 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering, 725 u8 match_criteria_enable, 726 const void *match_criteria, 727 int start_index, 728 int end_index) 729 { 730 struct mlx5_flow_group *fg; 731 int ret; 732 733 fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL); 734 if (!fg) 735 return ERR_PTR(-ENOMEM); 736 737 ret = rhashtable_init(&fg->ftes_hash, &rhash_fte); 738 if (ret) { 739 kmem_cache_free(steering->fgs_cache, fg); 740 return ERR_PTR(ret); 741 } 742 743 ida_init(&fg->fte_allocator); 744 fg->mask.match_criteria_enable = match_criteria_enable; 745 memcpy(&fg->mask.match_criteria, match_criteria, 746 sizeof(fg->mask.match_criteria)); 747 fg->node.type = FS_TYPE_FLOW_GROUP; 748 fg->start_index = start_index; 749 fg->max_ftes = end_index - start_index + 1; 750 751 return fg; 752 } 753 754 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft, 755 u8 match_criteria_enable, 756 const void *match_criteria, 757 int start_index, 758 int end_index, 759 struct list_head *prev) 760 { 761 struct mlx5_flow_steering *steering = get_steering(&ft->node); 762 struct mlx5_flow_group *fg; 763 int ret; 764 765 fg = alloc_flow_group(steering, match_criteria_enable, match_criteria, 766 start_index, end_index); 767 if (IS_ERR(fg)) 768 return fg; 769 770 /* initialize refcnt, add to parent list */ 771 ret = rhltable_insert(&ft->fgs_hash, 772 &fg->hash, 773 rhash_fg); 774 if (ret) { 775 dealloc_flow_group(steering, fg); 776 return ERR_PTR(ret); 777 } 778 779 tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group); 780 tree_add_node(&fg->node, &ft->node); 781 /* Add node to group list */ 782 list_add(&fg->node.list, prev); 783 atomic_inc(&ft->node.version); 784 785 return fg; 786 } 787 788 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, 789 enum fs_flow_table_type table_type, 790 enum fs_flow_table_op_mod op_mod, 791 u32 flags) 792 { 793 struct mlx5_flow_table *ft; 794 int ret; 795 796 ft = kzalloc(sizeof(*ft), GFP_KERNEL); 797 if (!ft) 798 return ERR_PTR(-ENOMEM); 799 800 ret = rhltable_init(&ft->fgs_hash, &rhash_fg); 801 if (ret) { 802 kfree(ft); 803 return ERR_PTR(ret); 804 } 805 806 ft->level = level; 807 ft->node.type = FS_TYPE_FLOW_TABLE; 808 ft->op_mod = op_mod; 809 ft->type = table_type; 810 ft->vport = vport; 811 ft->flags = flags; 812 INIT_LIST_HEAD(&ft->fwd_rules); 813 mutex_init(&ft->lock); 814 815 return ft; 816 } 817 818 /* If reverse is false, then we search for the first flow table in the 819 * root sub-tree from start(closest from right), else we search for the 820 * last flow table in the root sub-tree till start(closest from left). 821 */ 822 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node *root, 823 struct list_head *start, 824 bool reverse) 825 { 826 #define list_advance_entry(pos, reverse) \ 827 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list)) 828 829 #define list_for_each_advance_continue(pos, head, reverse) \ 830 for (pos = list_advance_entry(pos, reverse); \ 831 &pos->list != (head); \ 832 pos = list_advance_entry(pos, reverse)) 833 834 struct fs_node *iter = list_entry(start, struct fs_node, list); 835 struct mlx5_flow_table *ft = NULL; 836 837 if (!root || root->type == FS_TYPE_PRIO_CHAINS) 838 return NULL; 839 840 list_for_each_advance_continue(iter, &root->children, reverse) { 841 if (iter->type == FS_TYPE_FLOW_TABLE) { 842 fs_get_obj(ft, iter); 843 return ft; 844 } 845 ft = find_closest_ft_recursive(iter, &iter->children, reverse); 846 if (ft) 847 return ft; 848 } 849 850 return ft; 851 } 852 853 /* If reverse is false then return the first flow table in next priority of 854 * prio in the tree, else return the last flow table in the previous priority 855 * of prio in the tree. 856 */ 857 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse) 858 { 859 struct mlx5_flow_table *ft = NULL; 860 struct fs_node *curr_node; 861 struct fs_node *parent; 862 863 parent = prio->node.parent; 864 curr_node = &prio->node; 865 while (!ft && parent) { 866 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse); 867 curr_node = parent; 868 parent = curr_node->parent; 869 } 870 return ft; 871 } 872 873 /* Assuming all the tree is locked by mutex chain lock */ 874 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio) 875 { 876 return find_closest_ft(prio, false); 877 } 878 879 /* Assuming all the tree is locked by mutex chain lock */ 880 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio) 881 { 882 return find_closest_ft(prio, true); 883 } 884 885 static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft, 886 struct mlx5_flow_act *flow_act) 887 { 888 struct fs_prio *prio; 889 bool next_ns; 890 891 next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS; 892 fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent); 893 894 return find_next_chained_ft(prio); 895 } 896 897 static int connect_fts_in_prio(struct mlx5_core_dev *dev, 898 struct fs_prio *prio, 899 struct mlx5_flow_table *ft) 900 { 901 struct mlx5_flow_root_namespace *root = find_root(&prio->node); 902 struct mlx5_flow_table *iter; 903 int err; 904 905 fs_for_each_ft(iter, prio) { 906 err = root->cmds->modify_flow_table(root, iter, ft); 907 if (err) { 908 mlx5_core_err(dev, 909 "Failed to modify flow table id %d, type %d, err %d\n", 910 iter->id, iter->type, err); 911 /* The driver is out of sync with the FW */ 912 return err; 913 } 914 } 915 return 0; 916 } 917 918 /* Connect flow tables from previous priority of prio to ft */ 919 static int connect_prev_fts(struct mlx5_core_dev *dev, 920 struct mlx5_flow_table *ft, 921 struct fs_prio *prio) 922 { 923 struct mlx5_flow_table *prev_ft; 924 925 prev_ft = find_prev_chained_ft(prio); 926 if (prev_ft) { 927 struct fs_prio *prev_prio; 928 929 fs_get_obj(prev_prio, prev_ft->node.parent); 930 return connect_fts_in_prio(dev, prev_prio, ft); 931 } 932 return 0; 933 } 934 935 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio 936 *prio) 937 { 938 struct mlx5_flow_root_namespace *root = find_root(&prio->node); 939 struct mlx5_ft_underlay_qp *uqp; 940 int min_level = INT_MAX; 941 int err = 0; 942 u32 qpn; 943 944 if (root->root_ft) 945 min_level = root->root_ft->level; 946 947 if (ft->level >= min_level) 948 return 0; 949 950 if (list_empty(&root->underlay_qpns)) { 951 /* Don't set any QPN (zero) in case QPN list is empty */ 952 qpn = 0; 953 err = root->cmds->update_root_ft(root, ft, qpn, false); 954 } else { 955 list_for_each_entry(uqp, &root->underlay_qpns, list) { 956 qpn = uqp->qpn; 957 err = root->cmds->update_root_ft(root, ft, 958 qpn, false); 959 if (err) 960 break; 961 } 962 } 963 964 if (err) 965 mlx5_core_warn(root->dev, 966 "Update root flow table of id(%u) qpn(%d) failed\n", 967 ft->id, qpn); 968 else 969 root->root_ft = ft; 970 971 return err; 972 } 973 974 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule, 975 struct mlx5_flow_destination *dest) 976 { 977 struct mlx5_flow_root_namespace *root; 978 struct mlx5_flow_table *ft; 979 struct mlx5_flow_group *fg; 980 struct fs_fte *fte; 981 int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST); 982 int err = 0; 983 984 fs_get_obj(fte, rule->node.parent); 985 if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)) 986 return -EINVAL; 987 down_write_ref_node(&fte->node, false); 988 fs_get_obj(fg, fte->node.parent); 989 fs_get_obj(ft, fg->node.parent); 990 991 memcpy(&rule->dest_attr, dest, sizeof(*dest)); 992 root = find_root(&ft->node); 993 err = root->cmds->update_fte(root, ft, fg, 994 modify_mask, fte); 995 up_write_ref_node(&fte->node, false); 996 997 return err; 998 } 999 1000 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle, 1001 struct mlx5_flow_destination *new_dest, 1002 struct mlx5_flow_destination *old_dest) 1003 { 1004 int i; 1005 1006 if (!old_dest) { 1007 if (handle->num_rules != 1) 1008 return -EINVAL; 1009 return _mlx5_modify_rule_destination(handle->rule[0], 1010 new_dest); 1011 } 1012 1013 for (i = 0; i < handle->num_rules; i++) { 1014 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr)) 1015 return _mlx5_modify_rule_destination(handle->rule[i], 1016 new_dest); 1017 } 1018 1019 return -EINVAL; 1020 } 1021 1022 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */ 1023 static int connect_fwd_rules(struct mlx5_core_dev *dev, 1024 struct mlx5_flow_table *new_next_ft, 1025 struct mlx5_flow_table *old_next_ft) 1026 { 1027 struct mlx5_flow_destination dest = {}; 1028 struct mlx5_flow_rule *iter; 1029 int err = 0; 1030 1031 /* new_next_ft and old_next_ft could be NULL only 1032 * when we create/destroy the anchor flow table. 1033 */ 1034 if (!new_next_ft || !old_next_ft) 1035 return 0; 1036 1037 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE; 1038 dest.ft = new_next_ft; 1039 1040 mutex_lock(&old_next_ft->lock); 1041 list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules); 1042 mutex_unlock(&old_next_ft->lock); 1043 list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) { 1044 if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) && 1045 iter->ft->ns == new_next_ft->ns) 1046 continue; 1047 1048 err = _mlx5_modify_rule_destination(iter, &dest); 1049 if (err) 1050 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n", 1051 new_next_ft->id); 1052 } 1053 return 0; 1054 } 1055 1056 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft, 1057 struct fs_prio *prio) 1058 { 1059 struct mlx5_flow_table *next_ft, *first_ft; 1060 int err = 0; 1061 1062 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */ 1063 1064 first_ft = list_first_entry_or_null(&prio->node.children, 1065 struct mlx5_flow_table, node.list); 1066 if (!first_ft || first_ft->level > ft->level) { 1067 err = connect_prev_fts(dev, ft, prio); 1068 if (err) 1069 return err; 1070 1071 next_ft = first_ft ? first_ft : find_next_chained_ft(prio); 1072 err = connect_fwd_rules(dev, ft, next_ft); 1073 if (err) 1074 return err; 1075 } 1076 1077 if (MLX5_CAP_FLOWTABLE(dev, 1078 flow_table_properties_nic_receive.modify_root)) 1079 err = update_root_ft_create(ft, prio); 1080 return err; 1081 } 1082 1083 static void list_add_flow_table(struct mlx5_flow_table *ft, 1084 struct fs_prio *prio) 1085 { 1086 struct list_head *prev = &prio->node.children; 1087 struct mlx5_flow_table *iter; 1088 1089 fs_for_each_ft(iter, prio) { 1090 if (iter->level > ft->level) 1091 break; 1092 prev = &iter->node.list; 1093 } 1094 list_add(&ft->node.list, prev); 1095 } 1096 1097 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns, 1098 struct mlx5_flow_table_attr *ft_attr, 1099 enum fs_flow_table_op_mod op_mod, 1100 u16 vport) 1101 { 1102 struct mlx5_flow_root_namespace *root = find_root(&ns->node); 1103 bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED; 1104 struct mlx5_flow_table *next_ft; 1105 struct fs_prio *fs_prio = NULL; 1106 struct mlx5_flow_table *ft; 1107 int err; 1108 1109 if (!root) { 1110 pr_err("mlx5: flow steering failed to find root of namespace\n"); 1111 return ERR_PTR(-ENODEV); 1112 } 1113 1114 mutex_lock(&root->chain_lock); 1115 fs_prio = find_prio(ns, ft_attr->prio); 1116 if (!fs_prio) { 1117 err = -EINVAL; 1118 goto unlock_root; 1119 } 1120 if (!unmanaged) { 1121 /* The level is related to the 1122 * priority level range. 1123 */ 1124 if (ft_attr->level >= fs_prio->num_levels) { 1125 err = -ENOSPC; 1126 goto unlock_root; 1127 } 1128 1129 ft_attr->level += fs_prio->start_level; 1130 } 1131 1132 /* The level is related to the 1133 * priority level range. 1134 */ 1135 ft = alloc_flow_table(ft_attr->level, 1136 vport, 1137 root->table_type, 1138 op_mod, ft_attr->flags); 1139 if (IS_ERR(ft)) { 1140 err = PTR_ERR(ft); 1141 goto unlock_root; 1142 } 1143 1144 tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table); 1145 next_ft = unmanaged ? ft_attr->next_ft : 1146 find_next_chained_ft(fs_prio); 1147 ft->def_miss_action = ns->def_miss_action; 1148 ft->ns = ns; 1149 err = root->cmds->create_flow_table(root, ft, ft_attr->max_fte, next_ft); 1150 if (err) 1151 goto free_ft; 1152 1153 if (!unmanaged) { 1154 err = connect_flow_table(root->dev, ft, fs_prio); 1155 if (err) 1156 goto destroy_ft; 1157 } 1158 1159 ft->node.active = true; 1160 down_write_ref_node(&fs_prio->node, false); 1161 if (!unmanaged) { 1162 tree_add_node(&ft->node, &fs_prio->node); 1163 list_add_flow_table(ft, fs_prio); 1164 } else { 1165 ft->node.root = fs_prio->node.root; 1166 } 1167 fs_prio->num_ft++; 1168 up_write_ref_node(&fs_prio->node, false); 1169 mutex_unlock(&root->chain_lock); 1170 trace_mlx5_fs_add_ft(ft); 1171 return ft; 1172 destroy_ft: 1173 root->cmds->destroy_flow_table(root, ft); 1174 free_ft: 1175 rhltable_destroy(&ft->fgs_hash); 1176 kfree(ft); 1177 unlock_root: 1178 mutex_unlock(&root->chain_lock); 1179 return ERR_PTR(err); 1180 } 1181 1182 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns, 1183 struct mlx5_flow_table_attr *ft_attr) 1184 { 1185 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0); 1186 } 1187 EXPORT_SYMBOL(mlx5_create_flow_table); 1188 1189 struct mlx5_flow_table * 1190 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns, 1191 struct mlx5_flow_table_attr *ft_attr, u16 vport) 1192 { 1193 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport); 1194 } 1195 1196 struct mlx5_flow_table* 1197 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns, 1198 int prio, u32 level) 1199 { 1200 struct mlx5_flow_table_attr ft_attr = {}; 1201 1202 ft_attr.level = level; 1203 ft_attr.prio = prio; 1204 ft_attr.max_fte = 1; 1205 1206 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0); 1207 } 1208 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table); 1209 1210 #define MAX_FLOW_GROUP_SIZE BIT(24) 1211 struct mlx5_flow_table* 1212 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns, 1213 struct mlx5_flow_table_attr *ft_attr) 1214 { 1215 int num_reserved_entries = ft_attr->autogroup.num_reserved_entries; 1216 int max_num_groups = ft_attr->autogroup.max_num_groups; 1217 struct mlx5_flow_table *ft; 1218 int autogroups_max_fte; 1219 1220 ft = mlx5_create_flow_table(ns, ft_attr); 1221 if (IS_ERR(ft)) 1222 return ft; 1223 1224 autogroups_max_fte = ft->max_fte - num_reserved_entries; 1225 if (max_num_groups > autogroups_max_fte) 1226 goto err_validate; 1227 if (num_reserved_entries > ft->max_fte) 1228 goto err_validate; 1229 1230 /* Align the number of groups according to the largest group size */ 1231 if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE) 1232 max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1; 1233 1234 ft->autogroup.active = true; 1235 ft->autogroup.required_groups = max_num_groups; 1236 ft->autogroup.max_fte = autogroups_max_fte; 1237 /* We save place for flow groups in addition to max types */ 1238 ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1); 1239 1240 return ft; 1241 1242 err_validate: 1243 mlx5_destroy_flow_table(ft); 1244 return ERR_PTR(-ENOSPC); 1245 } 1246 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table); 1247 1248 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft, 1249 u32 *fg_in) 1250 { 1251 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 1252 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in, 1253 fg_in, match_criteria); 1254 u8 match_criteria_enable = MLX5_GET(create_flow_group_in, 1255 fg_in, 1256 match_criteria_enable); 1257 int start_index = MLX5_GET(create_flow_group_in, fg_in, 1258 start_flow_index); 1259 int end_index = MLX5_GET(create_flow_group_in, fg_in, 1260 end_flow_index); 1261 struct mlx5_flow_group *fg; 1262 int err; 1263 1264 if (ft->autogroup.active && start_index < ft->autogroup.max_fte) 1265 return ERR_PTR(-EPERM); 1266 1267 down_write_ref_node(&ft->node, false); 1268 fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria, 1269 start_index, end_index, 1270 ft->node.children.prev); 1271 up_write_ref_node(&ft->node, false); 1272 if (IS_ERR(fg)) 1273 return fg; 1274 1275 err = root->cmds->create_flow_group(root, ft, fg_in, fg); 1276 if (err) { 1277 tree_put_node(&fg->node, false); 1278 return ERR_PTR(err); 1279 } 1280 trace_mlx5_fs_add_fg(fg); 1281 fg->node.active = true; 1282 1283 return fg; 1284 } 1285 EXPORT_SYMBOL(mlx5_create_flow_group); 1286 1287 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest) 1288 { 1289 struct mlx5_flow_rule *rule; 1290 1291 rule = kzalloc(sizeof(*rule), GFP_KERNEL); 1292 if (!rule) 1293 return NULL; 1294 1295 INIT_LIST_HEAD(&rule->next_ft); 1296 rule->node.type = FS_TYPE_FLOW_DEST; 1297 if (dest) 1298 memcpy(&rule->dest_attr, dest, sizeof(*dest)); 1299 1300 return rule; 1301 } 1302 1303 static struct mlx5_flow_handle *alloc_handle(int num_rules) 1304 { 1305 struct mlx5_flow_handle *handle; 1306 1307 handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL); 1308 if (!handle) 1309 return NULL; 1310 1311 handle->num_rules = num_rules; 1312 1313 return handle; 1314 } 1315 1316 static void destroy_flow_handle(struct fs_fte *fte, 1317 struct mlx5_flow_handle *handle, 1318 struct mlx5_flow_destination *dest, 1319 int i) 1320 { 1321 for (; --i >= 0;) { 1322 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) { 1323 fte->dests_size--; 1324 list_del(&handle->rule[i]->node.list); 1325 kfree(handle->rule[i]); 1326 } 1327 } 1328 kfree(handle); 1329 } 1330 1331 static struct mlx5_flow_handle * 1332 create_flow_handle(struct fs_fte *fte, 1333 struct mlx5_flow_destination *dest, 1334 int dest_num, 1335 int *modify_mask, 1336 bool *new_rule) 1337 { 1338 struct mlx5_flow_handle *handle; 1339 struct mlx5_flow_rule *rule = NULL; 1340 static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS); 1341 static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST); 1342 int type; 1343 int i = 0; 1344 1345 handle = alloc_handle((dest_num) ? dest_num : 1); 1346 if (!handle) 1347 return ERR_PTR(-ENOMEM); 1348 1349 do { 1350 if (dest) { 1351 rule = find_flow_rule(fte, dest + i); 1352 if (rule) { 1353 refcount_inc(&rule->node.refcount); 1354 goto rule_found; 1355 } 1356 } 1357 1358 *new_rule = true; 1359 rule = alloc_rule(dest + i); 1360 if (!rule) 1361 goto free_rules; 1362 1363 /* Add dest to dests list- we need flow tables to be in the 1364 * end of the list for forward to next prio rules. 1365 */ 1366 tree_init_node(&rule->node, NULL, del_sw_hw_rule); 1367 if (dest && 1368 dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) 1369 list_add(&rule->node.list, &fte->node.children); 1370 else 1371 list_add_tail(&rule->node.list, &fte->node.children); 1372 if (dest) { 1373 fte->dests_size++; 1374 1375 type = dest[i].type == 1376 MLX5_FLOW_DESTINATION_TYPE_COUNTER; 1377 *modify_mask |= type ? count : dst; 1378 } 1379 rule_found: 1380 handle->rule[i] = rule; 1381 } while (++i < dest_num); 1382 1383 return handle; 1384 1385 free_rules: 1386 destroy_flow_handle(fte, handle, dest, i); 1387 return ERR_PTR(-ENOMEM); 1388 } 1389 1390 /* fte should not be deleted while calling this function */ 1391 static struct mlx5_flow_handle * 1392 add_rule_fte(struct fs_fte *fte, 1393 struct mlx5_flow_group *fg, 1394 struct mlx5_flow_destination *dest, 1395 int dest_num, 1396 bool update_action) 1397 { 1398 struct mlx5_flow_root_namespace *root; 1399 struct mlx5_flow_handle *handle; 1400 struct mlx5_flow_table *ft; 1401 int modify_mask = 0; 1402 int err; 1403 bool new_rule = false; 1404 1405 handle = create_flow_handle(fte, dest, dest_num, &modify_mask, 1406 &new_rule); 1407 if (IS_ERR(handle) || !new_rule) 1408 goto out; 1409 1410 if (update_action) 1411 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION); 1412 1413 fs_get_obj(ft, fg->node.parent); 1414 root = find_root(&fg->node); 1415 if (!(fte->status & FS_FTE_STATUS_EXISTING)) 1416 err = root->cmds->create_fte(root, ft, fg, fte); 1417 else 1418 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte); 1419 if (err) 1420 goto free_handle; 1421 1422 fte->node.active = true; 1423 fte->status |= FS_FTE_STATUS_EXISTING; 1424 atomic_inc(&fg->node.version); 1425 1426 out: 1427 return handle; 1428 1429 free_handle: 1430 destroy_flow_handle(fte, handle, dest, handle->num_rules); 1431 return ERR_PTR(err); 1432 } 1433 1434 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table *ft, 1435 const struct mlx5_flow_spec *spec) 1436 { 1437 struct list_head *prev = &ft->node.children; 1438 u32 max_fte = ft->autogroup.max_fte; 1439 unsigned int candidate_index = 0; 1440 unsigned int group_size = 0; 1441 struct mlx5_flow_group *fg; 1442 1443 if (!ft->autogroup.active) 1444 return ERR_PTR(-ENOENT); 1445 1446 if (ft->autogroup.num_groups < ft->autogroup.required_groups) 1447 group_size = ft->autogroup.group_size; 1448 1449 /* max_fte == ft->autogroup.max_types */ 1450 if (group_size == 0) 1451 group_size = 1; 1452 1453 /* sorted by start_index */ 1454 fs_for_each_fg(fg, ft) { 1455 if (candidate_index + group_size > fg->start_index) 1456 candidate_index = fg->start_index + fg->max_ftes; 1457 else 1458 break; 1459 prev = &fg->node.list; 1460 } 1461 1462 if (candidate_index + group_size > max_fte) 1463 return ERR_PTR(-ENOSPC); 1464 1465 fg = alloc_insert_flow_group(ft, 1466 spec->match_criteria_enable, 1467 spec->match_criteria, 1468 candidate_index, 1469 candidate_index + group_size - 1, 1470 prev); 1471 if (IS_ERR(fg)) 1472 goto out; 1473 1474 if (group_size == ft->autogroup.group_size) 1475 ft->autogroup.num_groups++; 1476 1477 out: 1478 return fg; 1479 } 1480 1481 static int create_auto_flow_group(struct mlx5_flow_table *ft, 1482 struct mlx5_flow_group *fg) 1483 { 1484 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 1485 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in); 1486 void *match_criteria_addr; 1487 u8 src_esw_owner_mask_on; 1488 void *misc; 1489 int err; 1490 u32 *in; 1491 1492 in = kvzalloc(inlen, GFP_KERNEL); 1493 if (!in) 1494 return -ENOMEM; 1495 1496 MLX5_SET(create_flow_group_in, in, match_criteria_enable, 1497 fg->mask.match_criteria_enable); 1498 MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index); 1499 MLX5_SET(create_flow_group_in, in, end_flow_index, fg->start_index + 1500 fg->max_ftes - 1); 1501 1502 misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria, 1503 misc_parameters); 1504 src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc, 1505 source_eswitch_owner_vhca_id); 1506 MLX5_SET(create_flow_group_in, in, 1507 source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on); 1508 1509 match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in, 1510 in, match_criteria); 1511 memcpy(match_criteria_addr, fg->mask.match_criteria, 1512 sizeof(fg->mask.match_criteria)); 1513 1514 err = root->cmds->create_flow_group(root, ft, in, fg); 1515 if (!err) { 1516 fg->node.active = true; 1517 trace_mlx5_fs_add_fg(fg); 1518 } 1519 1520 kvfree(in); 1521 return err; 1522 } 1523 1524 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1, 1525 struct mlx5_flow_destination *d2) 1526 { 1527 if (d1->type == d2->type) { 1528 if (((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT || 1529 d1->type == MLX5_FLOW_DESTINATION_TYPE_UPLINK) && 1530 d1->vport.num == d2->vport.num && 1531 d1->vport.flags == d2->vport.flags && 1532 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ? 1533 (d1->vport.vhca_id == d2->vport.vhca_id) : true) && 1534 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ? 1535 (d1->vport.pkt_reformat->id == 1536 d2->vport.pkt_reformat->id) : true)) || 1537 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE && 1538 d1->ft == d2->ft) || 1539 (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR && 1540 d1->tir_num == d2->tir_num) || 1541 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM && 1542 d1->ft_num == d2->ft_num) || 1543 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER && 1544 d1->sampler_id == d2->sampler_id)) 1545 return true; 1546 } 1547 1548 return false; 1549 } 1550 1551 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte, 1552 struct mlx5_flow_destination *dest) 1553 { 1554 struct mlx5_flow_rule *rule; 1555 1556 list_for_each_entry(rule, &fte->node.children, node.list) { 1557 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest)) 1558 return rule; 1559 } 1560 return NULL; 1561 } 1562 1563 static bool check_conflicting_actions(u32 action1, u32 action2) 1564 { 1565 u32 xored_actions = action1 ^ action2; 1566 1567 /* if one rule only wants to count, it's ok */ 1568 if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT || 1569 action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT) 1570 return false; 1571 1572 if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP | 1573 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT | 1574 MLX5_FLOW_CONTEXT_ACTION_DECAP | 1575 MLX5_FLOW_CONTEXT_ACTION_MOD_HDR | 1576 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP | 1577 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH | 1578 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 | 1579 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2)) 1580 return true; 1581 1582 return false; 1583 } 1584 1585 static int check_conflicting_ftes(struct fs_fte *fte, 1586 const struct mlx5_flow_context *flow_context, 1587 const struct mlx5_flow_act *flow_act) 1588 { 1589 if (check_conflicting_actions(flow_act->action, fte->action.action)) { 1590 mlx5_core_warn(get_dev(&fte->node), 1591 "Found two FTEs with conflicting actions\n"); 1592 return -EEXIST; 1593 } 1594 1595 if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) && 1596 fte->flow_context.flow_tag != flow_context->flow_tag) { 1597 mlx5_core_warn(get_dev(&fte->node), 1598 "FTE flow tag %u already exists with different flow tag %u\n", 1599 fte->flow_context.flow_tag, 1600 flow_context->flow_tag); 1601 return -EEXIST; 1602 } 1603 1604 return 0; 1605 } 1606 1607 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg, 1608 const struct mlx5_flow_spec *spec, 1609 struct mlx5_flow_act *flow_act, 1610 struct mlx5_flow_destination *dest, 1611 int dest_num, 1612 struct fs_fte *fte) 1613 { 1614 struct mlx5_flow_handle *handle; 1615 int old_action; 1616 int i; 1617 int ret; 1618 1619 ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act); 1620 if (ret) 1621 return ERR_PTR(ret); 1622 1623 old_action = fte->action.action; 1624 fte->action.action |= flow_act->action; 1625 handle = add_rule_fte(fte, fg, dest, dest_num, 1626 old_action != flow_act->action); 1627 if (IS_ERR(handle)) { 1628 fte->action.action = old_action; 1629 return handle; 1630 } 1631 trace_mlx5_fs_set_fte(fte, false); 1632 1633 for (i = 0; i < handle->num_rules; i++) { 1634 if (refcount_read(&handle->rule[i]->node.refcount) == 1) { 1635 tree_add_node(&handle->rule[i]->node, &fte->node); 1636 trace_mlx5_fs_add_rule(handle->rule[i]); 1637 } 1638 } 1639 return handle; 1640 } 1641 1642 static bool counter_is_valid(u32 action) 1643 { 1644 return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP | 1645 MLX5_FLOW_CONTEXT_ACTION_ALLOW | 1646 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)); 1647 } 1648 1649 static bool dest_is_valid(struct mlx5_flow_destination *dest, 1650 struct mlx5_flow_act *flow_act, 1651 struct mlx5_flow_table *ft) 1652 { 1653 bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL; 1654 u32 action = flow_act->action; 1655 1656 if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)) 1657 return counter_is_valid(action); 1658 1659 if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)) 1660 return true; 1661 1662 if (ignore_level) { 1663 if (ft->type != FS_FT_FDB && 1664 ft->type != FS_FT_NIC_RX) 1665 return false; 1666 1667 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE && 1668 ft->type != dest->ft->type) 1669 return false; 1670 } 1671 1672 if (!dest || ((dest->type == 1673 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) && 1674 (dest->ft->level <= ft->level && !ignore_level))) 1675 return false; 1676 return true; 1677 } 1678 1679 struct match_list { 1680 struct list_head list; 1681 struct mlx5_flow_group *g; 1682 }; 1683 1684 static void free_match_list(struct match_list *head, bool ft_locked) 1685 { 1686 struct match_list *iter, *match_tmp; 1687 1688 list_for_each_entry_safe(iter, match_tmp, &head->list, 1689 list) { 1690 tree_put_node(&iter->g->node, ft_locked); 1691 list_del(&iter->list); 1692 kfree(iter); 1693 } 1694 } 1695 1696 static int build_match_list(struct match_list *match_head, 1697 struct mlx5_flow_table *ft, 1698 const struct mlx5_flow_spec *spec, 1699 struct mlx5_flow_group *fg, 1700 bool ft_locked) 1701 { 1702 struct rhlist_head *tmp, *list; 1703 struct mlx5_flow_group *g; 1704 int err = 0; 1705 1706 rcu_read_lock(); 1707 INIT_LIST_HEAD(&match_head->list); 1708 /* Collect all fgs which has a matching match_criteria */ 1709 list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg); 1710 /* RCU is atomic, we can't execute FW commands here */ 1711 rhl_for_each_entry_rcu(g, tmp, list, hash) { 1712 struct match_list *curr_match; 1713 1714 if (fg && fg != g) 1715 continue; 1716 1717 if (unlikely(!tree_get_node(&g->node))) 1718 continue; 1719 1720 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC); 1721 if (!curr_match) { 1722 rcu_read_unlock(); 1723 free_match_list(match_head, ft_locked); 1724 return -ENOMEM; 1725 } 1726 curr_match->g = g; 1727 list_add_tail(&curr_match->list, &match_head->list); 1728 } 1729 rcu_read_unlock(); 1730 return err; 1731 } 1732 1733 static u64 matched_fgs_get_version(struct list_head *match_head) 1734 { 1735 struct match_list *iter; 1736 u64 version = 0; 1737 1738 list_for_each_entry(iter, match_head, list) 1739 version += (u64)atomic_read(&iter->g->node.version); 1740 return version; 1741 } 1742 1743 static struct fs_fte * 1744 lookup_fte_locked(struct mlx5_flow_group *g, 1745 const u32 *match_value, 1746 bool take_write) 1747 { 1748 struct fs_fte *fte_tmp; 1749 1750 if (take_write) 1751 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT); 1752 else 1753 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT); 1754 fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value, 1755 rhash_fte); 1756 if (!fte_tmp || !tree_get_node(&fte_tmp->node)) { 1757 fte_tmp = NULL; 1758 goto out; 1759 } 1760 if (!fte_tmp->node.active) { 1761 tree_put_node(&fte_tmp->node, false); 1762 fte_tmp = NULL; 1763 goto out; 1764 } 1765 1766 nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD); 1767 out: 1768 if (take_write) 1769 up_write_ref_node(&g->node, false); 1770 else 1771 up_read_ref_node(&g->node); 1772 return fte_tmp; 1773 } 1774 1775 static struct mlx5_flow_handle * 1776 try_add_to_existing_fg(struct mlx5_flow_table *ft, 1777 struct list_head *match_head, 1778 const struct mlx5_flow_spec *spec, 1779 struct mlx5_flow_act *flow_act, 1780 struct mlx5_flow_destination *dest, 1781 int dest_num, 1782 int ft_version) 1783 { 1784 struct mlx5_flow_steering *steering = get_steering(&ft->node); 1785 struct mlx5_flow_group *g; 1786 struct mlx5_flow_handle *rule; 1787 struct match_list *iter; 1788 bool take_write = false; 1789 struct fs_fte *fte; 1790 u64 version = 0; 1791 int err; 1792 1793 fte = alloc_fte(ft, spec, flow_act); 1794 if (IS_ERR(fte)) 1795 return ERR_PTR(-ENOMEM); 1796 1797 search_again_locked: 1798 if (flow_act->flags & FLOW_ACT_NO_APPEND) 1799 goto skip_search; 1800 version = matched_fgs_get_version(match_head); 1801 /* Try to find an fte with identical match value and attempt update its 1802 * action. 1803 */ 1804 list_for_each_entry(iter, match_head, list) { 1805 struct fs_fte *fte_tmp; 1806 1807 g = iter->g; 1808 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write); 1809 if (!fte_tmp) 1810 continue; 1811 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp); 1812 /* No error check needed here, because insert_fte() is not called */ 1813 up_write_ref_node(&fte_tmp->node, false); 1814 tree_put_node(&fte_tmp->node, false); 1815 kmem_cache_free(steering->ftes_cache, fte); 1816 return rule; 1817 } 1818 1819 skip_search: 1820 /* No group with matching fte found, or we skipped the search. 1821 * Try to add a new fte to any matching fg. 1822 */ 1823 1824 /* Check the ft version, for case that new flow group 1825 * was added while the fgs weren't locked 1826 */ 1827 if (atomic_read(&ft->node.version) != ft_version) { 1828 rule = ERR_PTR(-EAGAIN); 1829 goto out; 1830 } 1831 1832 /* Check the fgs version. If version have changed it could be that an 1833 * FTE with the same match value was added while the fgs weren't 1834 * locked. 1835 */ 1836 if (!(flow_act->flags & FLOW_ACT_NO_APPEND) && 1837 version != matched_fgs_get_version(match_head)) { 1838 take_write = true; 1839 goto search_again_locked; 1840 } 1841 1842 list_for_each_entry(iter, match_head, list) { 1843 g = iter->g; 1844 1845 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT); 1846 1847 if (!g->node.active) { 1848 up_write_ref_node(&g->node, false); 1849 continue; 1850 } 1851 1852 err = insert_fte(g, fte); 1853 if (err) { 1854 up_write_ref_node(&g->node, false); 1855 if (err == -ENOSPC) 1856 continue; 1857 kmem_cache_free(steering->ftes_cache, fte); 1858 return ERR_PTR(err); 1859 } 1860 1861 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD); 1862 up_write_ref_node(&g->node, false); 1863 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte); 1864 up_write_ref_node(&fte->node, false); 1865 if (IS_ERR(rule)) 1866 tree_put_node(&fte->node, false); 1867 return rule; 1868 } 1869 rule = ERR_PTR(-ENOENT); 1870 out: 1871 kmem_cache_free(steering->ftes_cache, fte); 1872 return rule; 1873 } 1874 1875 static struct mlx5_flow_handle * 1876 _mlx5_add_flow_rules(struct mlx5_flow_table *ft, 1877 const struct mlx5_flow_spec *spec, 1878 struct mlx5_flow_act *flow_act, 1879 struct mlx5_flow_destination *dest, 1880 int dest_num) 1881 1882 { 1883 struct mlx5_flow_steering *steering = get_steering(&ft->node); 1884 struct mlx5_flow_handle *rule; 1885 struct match_list match_head; 1886 struct mlx5_flow_group *g; 1887 bool take_write = false; 1888 struct fs_fte *fte; 1889 int version; 1890 int err; 1891 int i; 1892 1893 if (!check_valid_spec(spec)) 1894 return ERR_PTR(-EINVAL); 1895 1896 if (flow_act->fg && ft->autogroup.active) 1897 return ERR_PTR(-EINVAL); 1898 1899 for (i = 0; i < dest_num; i++) { 1900 if (!dest_is_valid(&dest[i], flow_act, ft)) 1901 return ERR_PTR(-EINVAL); 1902 } 1903 nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT); 1904 search_again_locked: 1905 version = atomic_read(&ft->node.version); 1906 1907 /* Collect all fgs which has a matching match_criteria */ 1908 err = build_match_list(&match_head, ft, spec, flow_act->fg, take_write); 1909 if (err) { 1910 if (take_write) 1911 up_write_ref_node(&ft->node, false); 1912 else 1913 up_read_ref_node(&ft->node); 1914 return ERR_PTR(err); 1915 } 1916 1917 if (!take_write) 1918 up_read_ref_node(&ft->node); 1919 1920 rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest, 1921 dest_num, version); 1922 free_match_list(&match_head, take_write); 1923 if (!IS_ERR(rule) || 1924 (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) { 1925 if (take_write) 1926 up_write_ref_node(&ft->node, false); 1927 return rule; 1928 } 1929 1930 if (!take_write) { 1931 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT); 1932 take_write = true; 1933 } 1934 1935 if (PTR_ERR(rule) == -EAGAIN || 1936 version != atomic_read(&ft->node.version)) 1937 goto search_again_locked; 1938 1939 g = alloc_auto_flow_group(ft, spec); 1940 if (IS_ERR(g)) { 1941 rule = ERR_CAST(g); 1942 up_write_ref_node(&ft->node, false); 1943 return rule; 1944 } 1945 1946 fte = alloc_fte(ft, spec, flow_act); 1947 if (IS_ERR(fte)) { 1948 up_write_ref_node(&ft->node, false); 1949 err = PTR_ERR(fte); 1950 goto err_alloc_fte; 1951 } 1952 1953 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT); 1954 up_write_ref_node(&ft->node, false); 1955 1956 err = create_auto_flow_group(ft, g); 1957 if (err) 1958 goto err_release_fg; 1959 1960 err = insert_fte(g, fte); 1961 if (err) 1962 goto err_release_fg; 1963 1964 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD); 1965 up_write_ref_node(&g->node, false); 1966 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte); 1967 up_write_ref_node(&fte->node, false); 1968 if (IS_ERR(rule)) 1969 tree_put_node(&fte->node, false); 1970 tree_put_node(&g->node, false); 1971 return rule; 1972 1973 err_release_fg: 1974 up_write_ref_node(&g->node, false); 1975 kmem_cache_free(steering->ftes_cache, fte); 1976 err_alloc_fte: 1977 tree_put_node(&g->node, false); 1978 return ERR_PTR(err); 1979 } 1980 1981 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft) 1982 { 1983 return ((ft->type == FS_FT_NIC_RX) && 1984 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs))); 1985 } 1986 1987 struct mlx5_flow_handle * 1988 mlx5_add_flow_rules(struct mlx5_flow_table *ft, 1989 const struct mlx5_flow_spec *spec, 1990 struct mlx5_flow_act *flow_act, 1991 struct mlx5_flow_destination *dest, 1992 int num_dest) 1993 { 1994 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 1995 static const struct mlx5_flow_spec zero_spec = {}; 1996 struct mlx5_flow_destination *gen_dest = NULL; 1997 struct mlx5_flow_table *next_ft = NULL; 1998 struct mlx5_flow_handle *handle = NULL; 1999 u32 sw_action = flow_act->action; 2000 int i; 2001 2002 if (!spec) 2003 spec = &zero_spec; 2004 2005 if (!is_fwd_next_action(sw_action)) 2006 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest); 2007 2008 if (!fwd_next_prio_supported(ft)) 2009 return ERR_PTR(-EOPNOTSUPP); 2010 2011 mutex_lock(&root->chain_lock); 2012 next_ft = find_next_fwd_ft(ft, flow_act); 2013 if (!next_ft) { 2014 handle = ERR_PTR(-EOPNOTSUPP); 2015 goto unlock; 2016 } 2017 2018 gen_dest = kcalloc(num_dest + 1, sizeof(*dest), 2019 GFP_KERNEL); 2020 if (!gen_dest) { 2021 handle = ERR_PTR(-ENOMEM); 2022 goto unlock; 2023 } 2024 for (i = 0; i < num_dest; i++) 2025 gen_dest[i] = dest[i]; 2026 gen_dest[i].type = 2027 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE; 2028 gen_dest[i].ft = next_ft; 2029 dest = gen_dest; 2030 num_dest++; 2031 flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO | 2032 MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS); 2033 flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST; 2034 handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest); 2035 if (IS_ERR(handle)) 2036 goto unlock; 2037 2038 if (list_empty(&handle->rule[num_dest - 1]->next_ft)) { 2039 mutex_lock(&next_ft->lock); 2040 list_add(&handle->rule[num_dest - 1]->next_ft, 2041 &next_ft->fwd_rules); 2042 mutex_unlock(&next_ft->lock); 2043 handle->rule[num_dest - 1]->sw_action = sw_action; 2044 handle->rule[num_dest - 1]->ft = ft; 2045 } 2046 unlock: 2047 mutex_unlock(&root->chain_lock); 2048 kfree(gen_dest); 2049 return handle; 2050 } 2051 EXPORT_SYMBOL(mlx5_add_flow_rules); 2052 2053 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle) 2054 { 2055 struct fs_fte *fte; 2056 int i; 2057 2058 /* In order to consolidate the HW changes we lock the FTE for other 2059 * changes, and increase its refcount, in order not to perform the 2060 * "del" functions of the FTE. Will handle them here. 2061 * The removal of the rules is done under locked FTE. 2062 * After removing all the handle's rules, if there are remaining 2063 * rules, it means we just need to modify the FTE in FW, and 2064 * unlock/decrease the refcount we increased before. 2065 * Otherwise, it means the FTE should be deleted. First delete the 2066 * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of 2067 * the FTE, which will handle the last decrease of the refcount, as 2068 * well as required handling of its parent. 2069 */ 2070 fs_get_obj(fte, handle->rule[0]->node.parent); 2071 down_write_ref_node(&fte->node, false); 2072 for (i = handle->num_rules - 1; i >= 0; i--) 2073 tree_remove_node(&handle->rule[i]->node, true); 2074 if (fte->dests_size) { 2075 if (fte->modify_mask) 2076 modify_fte(fte); 2077 up_write_ref_node(&fte->node, false); 2078 } else if (list_empty(&fte->node.children)) { 2079 del_hw_fte(&fte->node); 2080 /* Avoid double call to del_hw_fte */ 2081 fte->node.del_hw_func = NULL; 2082 up_write_ref_node(&fte->node, false); 2083 tree_put_node(&fte->node, false); 2084 } else { 2085 up_write_ref_node(&fte->node, false); 2086 } 2087 kfree(handle); 2088 } 2089 EXPORT_SYMBOL(mlx5_del_flow_rules); 2090 2091 /* Assuming prio->node.children(flow tables) is sorted by level */ 2092 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft) 2093 { 2094 struct fs_prio *prio; 2095 2096 fs_get_obj(prio, ft->node.parent); 2097 2098 if (!list_is_last(&ft->node.list, &prio->node.children)) 2099 return list_next_entry(ft, node.list); 2100 return find_next_chained_ft(prio); 2101 } 2102 2103 static int update_root_ft_destroy(struct mlx5_flow_table *ft) 2104 { 2105 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 2106 struct mlx5_ft_underlay_qp *uqp; 2107 struct mlx5_flow_table *new_root_ft = NULL; 2108 int err = 0; 2109 u32 qpn; 2110 2111 if (root->root_ft != ft) 2112 return 0; 2113 2114 new_root_ft = find_next_ft(ft); 2115 if (!new_root_ft) { 2116 root->root_ft = NULL; 2117 return 0; 2118 } 2119 2120 if (list_empty(&root->underlay_qpns)) { 2121 /* Don't set any QPN (zero) in case QPN list is empty */ 2122 qpn = 0; 2123 err = root->cmds->update_root_ft(root, new_root_ft, 2124 qpn, false); 2125 } else { 2126 list_for_each_entry(uqp, &root->underlay_qpns, list) { 2127 qpn = uqp->qpn; 2128 err = root->cmds->update_root_ft(root, 2129 new_root_ft, qpn, 2130 false); 2131 if (err) 2132 break; 2133 } 2134 } 2135 2136 if (err) 2137 mlx5_core_warn(root->dev, 2138 "Update root flow table of id(%u) qpn(%d) failed\n", 2139 ft->id, qpn); 2140 else 2141 root->root_ft = new_root_ft; 2142 2143 return 0; 2144 } 2145 2146 /* Connect flow table from previous priority to 2147 * the next flow table. 2148 */ 2149 static int disconnect_flow_table(struct mlx5_flow_table *ft) 2150 { 2151 struct mlx5_core_dev *dev = get_dev(&ft->node); 2152 struct mlx5_flow_table *next_ft; 2153 struct fs_prio *prio; 2154 int err = 0; 2155 2156 err = update_root_ft_destroy(ft); 2157 if (err) 2158 return err; 2159 2160 fs_get_obj(prio, ft->node.parent); 2161 if (!(list_first_entry(&prio->node.children, 2162 struct mlx5_flow_table, 2163 node.list) == ft)) 2164 return 0; 2165 2166 next_ft = find_next_ft(ft); 2167 err = connect_fwd_rules(dev, next_ft, ft); 2168 if (err) 2169 return err; 2170 2171 err = connect_prev_fts(dev, next_ft, prio); 2172 if (err) 2173 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n", 2174 ft->id); 2175 return err; 2176 } 2177 2178 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft) 2179 { 2180 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 2181 int err = 0; 2182 2183 mutex_lock(&root->chain_lock); 2184 if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED)) 2185 err = disconnect_flow_table(ft); 2186 if (err) { 2187 mutex_unlock(&root->chain_lock); 2188 return err; 2189 } 2190 if (tree_remove_node(&ft->node, false)) 2191 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n", 2192 ft->id); 2193 mutex_unlock(&root->chain_lock); 2194 2195 return err; 2196 } 2197 EXPORT_SYMBOL(mlx5_destroy_flow_table); 2198 2199 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg) 2200 { 2201 if (tree_remove_node(&fg->node, false)) 2202 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n", 2203 fg->id); 2204 } 2205 EXPORT_SYMBOL(mlx5_destroy_flow_group); 2206 2207 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev, 2208 int n) 2209 { 2210 struct mlx5_flow_steering *steering = dev->priv.steering; 2211 2212 if (!steering || !steering->fdb_sub_ns) 2213 return NULL; 2214 2215 return steering->fdb_sub_ns[n]; 2216 } 2217 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns); 2218 2219 static bool is_nic_rx_ns(enum mlx5_flow_namespace_type type) 2220 { 2221 switch (type) { 2222 case MLX5_FLOW_NAMESPACE_BYPASS: 2223 case MLX5_FLOW_NAMESPACE_LAG: 2224 case MLX5_FLOW_NAMESPACE_OFFLOADS: 2225 case MLX5_FLOW_NAMESPACE_ETHTOOL: 2226 case MLX5_FLOW_NAMESPACE_KERNEL: 2227 case MLX5_FLOW_NAMESPACE_LEFTOVERS: 2228 case MLX5_FLOW_NAMESPACE_ANCHOR: 2229 return true; 2230 default: 2231 return false; 2232 } 2233 } 2234 2235 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev, 2236 enum mlx5_flow_namespace_type type) 2237 { 2238 struct mlx5_flow_steering *steering = dev->priv.steering; 2239 struct mlx5_flow_root_namespace *root_ns; 2240 int prio = 0; 2241 struct fs_prio *fs_prio; 2242 struct mlx5_flow_namespace *ns; 2243 2244 if (!steering) 2245 return NULL; 2246 2247 switch (type) { 2248 case MLX5_FLOW_NAMESPACE_FDB: 2249 if (steering->fdb_root_ns) 2250 return &steering->fdb_root_ns->ns; 2251 return NULL; 2252 case MLX5_FLOW_NAMESPACE_PORT_SEL: 2253 if (steering->port_sel_root_ns) 2254 return &steering->port_sel_root_ns->ns; 2255 return NULL; 2256 case MLX5_FLOW_NAMESPACE_SNIFFER_RX: 2257 if (steering->sniffer_rx_root_ns) 2258 return &steering->sniffer_rx_root_ns->ns; 2259 return NULL; 2260 case MLX5_FLOW_NAMESPACE_SNIFFER_TX: 2261 if (steering->sniffer_tx_root_ns) 2262 return &steering->sniffer_tx_root_ns->ns; 2263 return NULL; 2264 case MLX5_FLOW_NAMESPACE_FDB_BYPASS: 2265 root_ns = steering->fdb_root_ns; 2266 prio = FDB_BYPASS_PATH; 2267 break; 2268 case MLX5_FLOW_NAMESPACE_EGRESS: 2269 case MLX5_FLOW_NAMESPACE_EGRESS_KERNEL: 2270 root_ns = steering->egress_root_ns; 2271 prio = type - MLX5_FLOW_NAMESPACE_EGRESS; 2272 break; 2273 case MLX5_FLOW_NAMESPACE_RDMA_RX: 2274 root_ns = steering->rdma_rx_root_ns; 2275 prio = RDMA_RX_BYPASS_PRIO; 2276 break; 2277 case MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL: 2278 root_ns = steering->rdma_rx_root_ns; 2279 prio = RDMA_RX_KERNEL_PRIO; 2280 break; 2281 case MLX5_FLOW_NAMESPACE_RDMA_TX: 2282 root_ns = steering->rdma_tx_root_ns; 2283 break; 2284 case MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS: 2285 root_ns = steering->rdma_rx_root_ns; 2286 prio = RDMA_RX_COUNTERS_PRIO; 2287 break; 2288 case MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS: 2289 root_ns = steering->rdma_tx_root_ns; 2290 prio = RDMA_TX_COUNTERS_PRIO; 2291 break; 2292 default: /* Must be NIC RX */ 2293 WARN_ON(!is_nic_rx_ns(type)); 2294 root_ns = steering->root_ns; 2295 prio = type; 2296 break; 2297 } 2298 2299 if (!root_ns) 2300 return NULL; 2301 2302 fs_prio = find_prio(&root_ns->ns, prio); 2303 if (!fs_prio) 2304 return NULL; 2305 2306 ns = list_first_entry(&fs_prio->node.children, 2307 typeof(*ns), 2308 node.list); 2309 2310 return ns; 2311 } 2312 EXPORT_SYMBOL(mlx5_get_flow_namespace); 2313 2314 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev, 2315 enum mlx5_flow_namespace_type type, 2316 int vport) 2317 { 2318 struct mlx5_flow_steering *steering = dev->priv.steering; 2319 2320 if (!steering) 2321 return NULL; 2322 2323 switch (type) { 2324 case MLX5_FLOW_NAMESPACE_ESW_EGRESS: 2325 if (vport >= steering->esw_egress_acl_vports) 2326 return NULL; 2327 if (steering->esw_egress_root_ns && 2328 steering->esw_egress_root_ns[vport]) 2329 return &steering->esw_egress_root_ns[vport]->ns; 2330 else 2331 return NULL; 2332 case MLX5_FLOW_NAMESPACE_ESW_INGRESS: 2333 if (vport >= steering->esw_ingress_acl_vports) 2334 return NULL; 2335 if (steering->esw_ingress_root_ns && 2336 steering->esw_ingress_root_ns[vport]) 2337 return &steering->esw_ingress_root_ns[vport]->ns; 2338 else 2339 return NULL; 2340 default: 2341 return NULL; 2342 } 2343 } 2344 2345 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns, 2346 unsigned int prio, 2347 int num_levels, 2348 enum fs_node_type type) 2349 { 2350 struct fs_prio *fs_prio; 2351 2352 fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL); 2353 if (!fs_prio) 2354 return ERR_PTR(-ENOMEM); 2355 2356 fs_prio->node.type = type; 2357 tree_init_node(&fs_prio->node, NULL, del_sw_prio); 2358 tree_add_node(&fs_prio->node, &ns->node); 2359 fs_prio->num_levels = num_levels; 2360 fs_prio->prio = prio; 2361 list_add_tail(&fs_prio->node.list, &ns->node.children); 2362 2363 return fs_prio; 2364 } 2365 2366 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns, 2367 unsigned int prio, 2368 int num_levels) 2369 { 2370 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS); 2371 } 2372 2373 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns, 2374 unsigned int prio, int num_levels) 2375 { 2376 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO); 2377 } 2378 2379 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace 2380 *ns) 2381 { 2382 ns->node.type = FS_TYPE_NAMESPACE; 2383 2384 return ns; 2385 } 2386 2387 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio, 2388 int def_miss_act) 2389 { 2390 struct mlx5_flow_namespace *ns; 2391 2392 ns = kzalloc(sizeof(*ns), GFP_KERNEL); 2393 if (!ns) 2394 return ERR_PTR(-ENOMEM); 2395 2396 fs_init_namespace(ns); 2397 ns->def_miss_action = def_miss_act; 2398 tree_init_node(&ns->node, NULL, del_sw_ns); 2399 tree_add_node(&ns->node, &prio->node); 2400 list_add_tail(&ns->node.list, &prio->node.children); 2401 2402 return ns; 2403 } 2404 2405 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio, 2406 struct init_tree_node *prio_metadata) 2407 { 2408 struct fs_prio *fs_prio; 2409 int i; 2410 2411 for (i = 0; i < prio_metadata->num_leaf_prios; i++) { 2412 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels); 2413 if (IS_ERR(fs_prio)) 2414 return PTR_ERR(fs_prio); 2415 } 2416 return 0; 2417 } 2418 2419 #define FLOW_TABLE_BIT_SZ 1 2420 #define GET_FLOW_TABLE_CAP(dev, offset) \ 2421 ((be32_to_cpu(*((__be32 *)(dev->caps.hca[MLX5_CAP_FLOW_TABLE]->cur) + \ 2422 offset / 32)) >> \ 2423 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ) 2424 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps) 2425 { 2426 int i; 2427 2428 for (i = 0; i < caps->arr_sz; i++) { 2429 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i])) 2430 return false; 2431 } 2432 return true; 2433 } 2434 2435 static int init_root_tree_recursive(struct mlx5_flow_steering *steering, 2436 struct init_tree_node *init_node, 2437 struct fs_node *fs_parent_node, 2438 struct init_tree_node *init_parent_node, 2439 int prio) 2440 { 2441 int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev, 2442 flow_table_properties_nic_receive. 2443 max_ft_level); 2444 struct mlx5_flow_namespace *fs_ns; 2445 struct fs_prio *fs_prio; 2446 struct fs_node *base; 2447 int i; 2448 int err; 2449 2450 if (init_node->type == FS_TYPE_PRIO) { 2451 if ((init_node->min_ft_level > max_ft_level) || 2452 !has_required_caps(steering->dev, &init_node->caps)) 2453 return 0; 2454 2455 fs_get_obj(fs_ns, fs_parent_node); 2456 if (init_node->num_leaf_prios) 2457 return create_leaf_prios(fs_ns, prio, init_node); 2458 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels); 2459 if (IS_ERR(fs_prio)) 2460 return PTR_ERR(fs_prio); 2461 base = &fs_prio->node; 2462 } else if (init_node->type == FS_TYPE_NAMESPACE) { 2463 fs_get_obj(fs_prio, fs_parent_node); 2464 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action); 2465 if (IS_ERR(fs_ns)) 2466 return PTR_ERR(fs_ns); 2467 base = &fs_ns->node; 2468 } else { 2469 return -EINVAL; 2470 } 2471 prio = 0; 2472 for (i = 0; i < init_node->ar_size; i++) { 2473 err = init_root_tree_recursive(steering, &init_node->children[i], 2474 base, init_node, prio); 2475 if (err) 2476 return err; 2477 if (init_node->children[i].type == FS_TYPE_PRIO && 2478 init_node->children[i].num_leaf_prios) { 2479 prio += init_node->children[i].num_leaf_prios; 2480 } 2481 } 2482 2483 return 0; 2484 } 2485 2486 static int init_root_tree(struct mlx5_flow_steering *steering, 2487 struct init_tree_node *init_node, 2488 struct fs_node *fs_parent_node) 2489 { 2490 int err; 2491 int i; 2492 2493 for (i = 0; i < init_node->ar_size; i++) { 2494 err = init_root_tree_recursive(steering, &init_node->children[i], 2495 fs_parent_node, 2496 init_node, i); 2497 if (err) 2498 return err; 2499 } 2500 return 0; 2501 } 2502 2503 static void del_sw_root_ns(struct fs_node *node) 2504 { 2505 struct mlx5_flow_root_namespace *root_ns; 2506 struct mlx5_flow_namespace *ns; 2507 2508 fs_get_obj(ns, node); 2509 root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns); 2510 mutex_destroy(&root_ns->chain_lock); 2511 kfree(node); 2512 } 2513 2514 static struct mlx5_flow_root_namespace 2515 *create_root_ns(struct mlx5_flow_steering *steering, 2516 enum fs_flow_table_type table_type) 2517 { 2518 const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type); 2519 struct mlx5_flow_root_namespace *root_ns; 2520 struct mlx5_flow_namespace *ns; 2521 2522 if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE && 2523 (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX)) 2524 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type); 2525 2526 /* Create the root namespace */ 2527 root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL); 2528 if (!root_ns) 2529 return NULL; 2530 2531 root_ns->dev = steering->dev; 2532 root_ns->table_type = table_type; 2533 root_ns->cmds = cmds; 2534 2535 INIT_LIST_HEAD(&root_ns->underlay_qpns); 2536 2537 ns = &root_ns->ns; 2538 fs_init_namespace(ns); 2539 mutex_init(&root_ns->chain_lock); 2540 tree_init_node(&ns->node, NULL, del_sw_root_ns); 2541 tree_add_node(&ns->node, NULL); 2542 2543 return root_ns; 2544 } 2545 2546 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level); 2547 2548 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level) 2549 { 2550 struct fs_prio *prio; 2551 2552 fs_for_each_prio(prio, ns) { 2553 /* This updates prio start_level and num_levels */ 2554 set_prio_attrs_in_prio(prio, acc_level); 2555 acc_level += prio->num_levels; 2556 } 2557 return acc_level; 2558 } 2559 2560 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level) 2561 { 2562 struct mlx5_flow_namespace *ns; 2563 int acc_level_ns = acc_level; 2564 2565 prio->start_level = acc_level; 2566 fs_for_each_ns(ns, prio) { 2567 /* This updates start_level and num_levels of ns's priority descendants */ 2568 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level); 2569 2570 /* If this a prio with chains, and we can jump from one chain 2571 * (namespace) to another, so we accumulate the levels 2572 */ 2573 if (prio->node.type == FS_TYPE_PRIO_CHAINS) 2574 acc_level = acc_level_ns; 2575 } 2576 2577 if (!prio->num_levels) 2578 prio->num_levels = acc_level_ns - prio->start_level; 2579 WARN_ON(prio->num_levels < acc_level_ns - prio->start_level); 2580 } 2581 2582 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns) 2583 { 2584 struct mlx5_flow_namespace *ns = &root_ns->ns; 2585 struct fs_prio *prio; 2586 int start_level = 0; 2587 2588 fs_for_each_prio(prio, ns) { 2589 set_prio_attrs_in_prio(prio, start_level); 2590 start_level += prio->num_levels; 2591 } 2592 } 2593 2594 #define ANCHOR_PRIO 0 2595 #define ANCHOR_SIZE 1 2596 #define ANCHOR_LEVEL 0 2597 static int create_anchor_flow_table(struct mlx5_flow_steering *steering) 2598 { 2599 struct mlx5_flow_namespace *ns = NULL; 2600 struct mlx5_flow_table_attr ft_attr = {}; 2601 struct mlx5_flow_table *ft; 2602 2603 ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR); 2604 if (WARN_ON(!ns)) 2605 return -EINVAL; 2606 2607 ft_attr.max_fte = ANCHOR_SIZE; 2608 ft_attr.level = ANCHOR_LEVEL; 2609 ft_attr.prio = ANCHOR_PRIO; 2610 2611 ft = mlx5_create_flow_table(ns, &ft_attr); 2612 if (IS_ERR(ft)) { 2613 mlx5_core_err(steering->dev, "Failed to create last anchor flow table"); 2614 return PTR_ERR(ft); 2615 } 2616 return 0; 2617 } 2618 2619 static int init_root_ns(struct mlx5_flow_steering *steering) 2620 { 2621 int err; 2622 2623 steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX); 2624 if (!steering->root_ns) 2625 return -ENOMEM; 2626 2627 err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node); 2628 if (err) 2629 goto out_err; 2630 2631 set_prio_attrs(steering->root_ns); 2632 err = create_anchor_flow_table(steering); 2633 if (err) 2634 goto out_err; 2635 2636 return 0; 2637 2638 out_err: 2639 cleanup_root_ns(steering->root_ns); 2640 steering->root_ns = NULL; 2641 return err; 2642 } 2643 2644 static void clean_tree(struct fs_node *node) 2645 { 2646 if (node) { 2647 struct fs_node *iter; 2648 struct fs_node *temp; 2649 2650 tree_get_node(node); 2651 list_for_each_entry_safe(iter, temp, &node->children, list) 2652 clean_tree(iter); 2653 tree_put_node(node, false); 2654 tree_remove_node(node, false); 2655 } 2656 } 2657 2658 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns) 2659 { 2660 if (!root_ns) 2661 return; 2662 2663 clean_tree(&root_ns->ns.node); 2664 } 2665 2666 void mlx5_cleanup_fs(struct mlx5_core_dev *dev) 2667 { 2668 struct mlx5_flow_steering *steering = dev->priv.steering; 2669 2670 cleanup_root_ns(steering->root_ns); 2671 cleanup_root_ns(steering->fdb_root_ns); 2672 steering->fdb_root_ns = NULL; 2673 kfree(steering->fdb_sub_ns); 2674 steering->fdb_sub_ns = NULL; 2675 cleanup_root_ns(steering->port_sel_root_ns); 2676 cleanup_root_ns(steering->sniffer_rx_root_ns); 2677 cleanup_root_ns(steering->sniffer_tx_root_ns); 2678 cleanup_root_ns(steering->rdma_rx_root_ns); 2679 cleanup_root_ns(steering->rdma_tx_root_ns); 2680 cleanup_root_ns(steering->egress_root_ns); 2681 mlx5_cleanup_fc_stats(dev); 2682 kmem_cache_destroy(steering->ftes_cache); 2683 kmem_cache_destroy(steering->fgs_cache); 2684 mlx5_ft_pool_destroy(dev); 2685 kfree(steering); 2686 } 2687 2688 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering) 2689 { 2690 struct fs_prio *prio; 2691 2692 steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX); 2693 if (!steering->sniffer_tx_root_ns) 2694 return -ENOMEM; 2695 2696 /* Create single prio */ 2697 prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1); 2698 return PTR_ERR_OR_ZERO(prio); 2699 } 2700 2701 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering) 2702 { 2703 struct fs_prio *prio; 2704 2705 steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX); 2706 if (!steering->sniffer_rx_root_ns) 2707 return -ENOMEM; 2708 2709 /* Create single prio */ 2710 prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1); 2711 return PTR_ERR_OR_ZERO(prio); 2712 } 2713 2714 #define PORT_SEL_NUM_LEVELS 3 2715 static int init_port_sel_root_ns(struct mlx5_flow_steering *steering) 2716 { 2717 struct fs_prio *prio; 2718 2719 steering->port_sel_root_ns = create_root_ns(steering, FS_FT_PORT_SEL); 2720 if (!steering->port_sel_root_ns) 2721 return -ENOMEM; 2722 2723 /* Create single prio */ 2724 prio = fs_create_prio(&steering->port_sel_root_ns->ns, 0, 2725 PORT_SEL_NUM_LEVELS); 2726 return PTR_ERR_OR_ZERO(prio); 2727 } 2728 2729 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering) 2730 { 2731 int err; 2732 2733 steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX); 2734 if (!steering->rdma_rx_root_ns) 2735 return -ENOMEM; 2736 2737 err = init_root_tree(steering, &rdma_rx_root_fs, 2738 &steering->rdma_rx_root_ns->ns.node); 2739 if (err) 2740 goto out_err; 2741 2742 set_prio_attrs(steering->rdma_rx_root_ns); 2743 2744 return 0; 2745 2746 out_err: 2747 cleanup_root_ns(steering->rdma_rx_root_ns); 2748 steering->rdma_rx_root_ns = NULL; 2749 return err; 2750 } 2751 2752 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering) 2753 { 2754 int err; 2755 2756 steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX); 2757 if (!steering->rdma_tx_root_ns) 2758 return -ENOMEM; 2759 2760 err = init_root_tree(steering, &rdma_tx_root_fs, 2761 &steering->rdma_tx_root_ns->ns.node); 2762 if (err) 2763 goto out_err; 2764 2765 set_prio_attrs(steering->rdma_tx_root_ns); 2766 2767 return 0; 2768 2769 out_err: 2770 cleanup_root_ns(steering->rdma_tx_root_ns); 2771 steering->rdma_tx_root_ns = NULL; 2772 return err; 2773 } 2774 2775 /* FT and tc chains are stored in the same array so we can re-use the 2776 * mlx5_get_fdb_sub_ns() and tc api for FT chains. 2777 * When creating a new ns for each chain store it in the first available slot. 2778 * Assume tc chains are created and stored first and only then the FT chain. 2779 */ 2780 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering, 2781 struct mlx5_flow_namespace *ns) 2782 { 2783 int chain = 0; 2784 2785 while (steering->fdb_sub_ns[chain]) 2786 ++chain; 2787 2788 steering->fdb_sub_ns[chain] = ns; 2789 } 2790 2791 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering, 2792 struct fs_prio *maj_prio) 2793 { 2794 struct mlx5_flow_namespace *ns; 2795 struct fs_prio *min_prio; 2796 int prio; 2797 2798 ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF); 2799 if (IS_ERR(ns)) 2800 return PTR_ERR(ns); 2801 2802 for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) { 2803 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO); 2804 if (IS_ERR(min_prio)) 2805 return PTR_ERR(min_prio); 2806 } 2807 2808 store_fdb_sub_ns_prio_chain(steering, ns); 2809 2810 return 0; 2811 } 2812 2813 static int create_fdb_chains(struct mlx5_flow_steering *steering, 2814 int fs_prio, 2815 int chains) 2816 { 2817 struct fs_prio *maj_prio; 2818 int levels; 2819 int chain; 2820 int err; 2821 2822 levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains; 2823 maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns, 2824 fs_prio, 2825 levels); 2826 if (IS_ERR(maj_prio)) 2827 return PTR_ERR(maj_prio); 2828 2829 for (chain = 0; chain < chains; chain++) { 2830 err = create_fdb_sub_ns_prio_chain(steering, maj_prio); 2831 if (err) 2832 return err; 2833 } 2834 2835 return 0; 2836 } 2837 2838 static int create_fdb_fast_path(struct mlx5_flow_steering *steering) 2839 { 2840 int err; 2841 2842 steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS, 2843 sizeof(*steering->fdb_sub_ns), 2844 GFP_KERNEL); 2845 if (!steering->fdb_sub_ns) 2846 return -ENOMEM; 2847 2848 err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1); 2849 if (err) 2850 return err; 2851 2852 err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1); 2853 if (err) 2854 return err; 2855 2856 return 0; 2857 } 2858 2859 static int create_fdb_bypass(struct mlx5_flow_steering *steering) 2860 { 2861 struct mlx5_flow_namespace *ns; 2862 struct fs_prio *prio; 2863 int i; 2864 2865 prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH, 0); 2866 if (IS_ERR(prio)) 2867 return PTR_ERR(prio); 2868 2869 ns = fs_create_namespace(prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF); 2870 if (IS_ERR(ns)) 2871 return PTR_ERR(ns); 2872 2873 for (i = 0; i < MLX5_BY_PASS_NUM_REGULAR_PRIOS; i++) { 2874 prio = fs_create_prio(ns, i, 1); 2875 if (IS_ERR(prio)) 2876 return PTR_ERR(prio); 2877 } 2878 return 0; 2879 } 2880 2881 static int init_fdb_root_ns(struct mlx5_flow_steering *steering) 2882 { 2883 struct fs_prio *maj_prio; 2884 int err; 2885 2886 steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB); 2887 if (!steering->fdb_root_ns) 2888 return -ENOMEM; 2889 2890 err = create_fdb_bypass(steering); 2891 if (err) 2892 goto out_err; 2893 2894 err = create_fdb_fast_path(steering); 2895 if (err) 2896 goto out_err; 2897 2898 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1); 2899 if (IS_ERR(maj_prio)) { 2900 err = PTR_ERR(maj_prio); 2901 goto out_err; 2902 } 2903 2904 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 3); 2905 if (IS_ERR(maj_prio)) { 2906 err = PTR_ERR(maj_prio); 2907 goto out_err; 2908 } 2909 2910 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1); 2911 if (IS_ERR(maj_prio)) { 2912 err = PTR_ERR(maj_prio); 2913 goto out_err; 2914 } 2915 2916 /* We put this priority last, knowing that nothing will get here 2917 * unless explicitly forwarded to. This is possible because the 2918 * slow path tables have catch all rules and nothing gets passed 2919 * those tables. 2920 */ 2921 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1); 2922 if (IS_ERR(maj_prio)) { 2923 err = PTR_ERR(maj_prio); 2924 goto out_err; 2925 } 2926 2927 set_prio_attrs(steering->fdb_root_ns); 2928 return 0; 2929 2930 out_err: 2931 cleanup_root_ns(steering->fdb_root_ns); 2932 kfree(steering->fdb_sub_ns); 2933 steering->fdb_sub_ns = NULL; 2934 steering->fdb_root_ns = NULL; 2935 return err; 2936 } 2937 2938 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport) 2939 { 2940 struct fs_prio *prio; 2941 2942 steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL); 2943 if (!steering->esw_egress_root_ns[vport]) 2944 return -ENOMEM; 2945 2946 /* create 1 prio*/ 2947 prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1); 2948 return PTR_ERR_OR_ZERO(prio); 2949 } 2950 2951 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport) 2952 { 2953 struct fs_prio *prio; 2954 2955 steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL); 2956 if (!steering->esw_ingress_root_ns[vport]) 2957 return -ENOMEM; 2958 2959 /* create 1 prio*/ 2960 prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1); 2961 return PTR_ERR_OR_ZERO(prio); 2962 } 2963 2964 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports) 2965 { 2966 struct mlx5_flow_steering *steering = dev->priv.steering; 2967 int err; 2968 int i; 2969 2970 steering->esw_egress_root_ns = 2971 kcalloc(total_vports, 2972 sizeof(*steering->esw_egress_root_ns), 2973 GFP_KERNEL); 2974 if (!steering->esw_egress_root_ns) 2975 return -ENOMEM; 2976 2977 for (i = 0; i < total_vports; i++) { 2978 err = init_egress_acl_root_ns(steering, i); 2979 if (err) 2980 goto cleanup_root_ns; 2981 } 2982 steering->esw_egress_acl_vports = total_vports; 2983 return 0; 2984 2985 cleanup_root_ns: 2986 for (i--; i >= 0; i--) 2987 cleanup_root_ns(steering->esw_egress_root_ns[i]); 2988 kfree(steering->esw_egress_root_ns); 2989 steering->esw_egress_root_ns = NULL; 2990 return err; 2991 } 2992 2993 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev) 2994 { 2995 struct mlx5_flow_steering *steering = dev->priv.steering; 2996 int i; 2997 2998 if (!steering->esw_egress_root_ns) 2999 return; 3000 3001 for (i = 0; i < steering->esw_egress_acl_vports; i++) 3002 cleanup_root_ns(steering->esw_egress_root_ns[i]); 3003 3004 kfree(steering->esw_egress_root_ns); 3005 steering->esw_egress_root_ns = NULL; 3006 } 3007 3008 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports) 3009 { 3010 struct mlx5_flow_steering *steering = dev->priv.steering; 3011 int err; 3012 int i; 3013 3014 steering->esw_ingress_root_ns = 3015 kcalloc(total_vports, 3016 sizeof(*steering->esw_ingress_root_ns), 3017 GFP_KERNEL); 3018 if (!steering->esw_ingress_root_ns) 3019 return -ENOMEM; 3020 3021 for (i = 0; i < total_vports; i++) { 3022 err = init_ingress_acl_root_ns(steering, i); 3023 if (err) 3024 goto cleanup_root_ns; 3025 } 3026 steering->esw_ingress_acl_vports = total_vports; 3027 return 0; 3028 3029 cleanup_root_ns: 3030 for (i--; i >= 0; i--) 3031 cleanup_root_ns(steering->esw_ingress_root_ns[i]); 3032 kfree(steering->esw_ingress_root_ns); 3033 steering->esw_ingress_root_ns = NULL; 3034 return err; 3035 } 3036 3037 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev) 3038 { 3039 struct mlx5_flow_steering *steering = dev->priv.steering; 3040 int i; 3041 3042 if (!steering->esw_ingress_root_ns) 3043 return; 3044 3045 for (i = 0; i < steering->esw_ingress_acl_vports; i++) 3046 cleanup_root_ns(steering->esw_ingress_root_ns[i]); 3047 3048 kfree(steering->esw_ingress_root_ns); 3049 steering->esw_ingress_root_ns = NULL; 3050 } 3051 3052 u32 mlx5_fs_get_capabilities(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type type) 3053 { 3054 struct mlx5_flow_root_namespace *root; 3055 struct mlx5_flow_namespace *ns; 3056 3057 ns = mlx5_get_flow_namespace(dev, type); 3058 if (!ns) 3059 return 0; 3060 3061 root = find_root(&ns->node); 3062 if (!root) 3063 return 0; 3064 3065 return root->cmds->get_capabilities(root, root->table_type); 3066 } 3067 3068 static int init_egress_root_ns(struct mlx5_flow_steering *steering) 3069 { 3070 int err; 3071 3072 steering->egress_root_ns = create_root_ns(steering, 3073 FS_FT_NIC_TX); 3074 if (!steering->egress_root_ns) 3075 return -ENOMEM; 3076 3077 err = init_root_tree(steering, &egress_root_fs, 3078 &steering->egress_root_ns->ns.node); 3079 if (err) 3080 goto cleanup; 3081 set_prio_attrs(steering->egress_root_ns); 3082 return 0; 3083 cleanup: 3084 cleanup_root_ns(steering->egress_root_ns); 3085 steering->egress_root_ns = NULL; 3086 return err; 3087 } 3088 3089 int mlx5_init_fs(struct mlx5_core_dev *dev) 3090 { 3091 struct mlx5_flow_steering *steering; 3092 int err = 0; 3093 3094 err = mlx5_init_fc_stats(dev); 3095 if (err) 3096 return err; 3097 3098 err = mlx5_ft_pool_init(dev); 3099 if (err) 3100 return err; 3101 3102 steering = kzalloc(sizeof(*steering), GFP_KERNEL); 3103 if (!steering) { 3104 err = -ENOMEM; 3105 goto err; 3106 } 3107 3108 steering->dev = dev; 3109 dev->priv.steering = steering; 3110 3111 if (mlx5_fs_dr_is_supported(dev)) 3112 steering->mode = MLX5_FLOW_STEERING_MODE_SMFS; 3113 else 3114 steering->mode = MLX5_FLOW_STEERING_MODE_DMFS; 3115 3116 steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs", 3117 sizeof(struct mlx5_flow_group), 0, 3118 0, NULL); 3119 steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0, 3120 0, NULL); 3121 if (!steering->ftes_cache || !steering->fgs_cache) { 3122 err = -ENOMEM; 3123 goto err; 3124 } 3125 3126 if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) && 3127 (MLX5_CAP_GEN(dev, nic_flow_table))) || 3128 ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) && 3129 MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) && 3130 MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) { 3131 err = init_root_ns(steering); 3132 if (err) 3133 goto err; 3134 } 3135 3136 if (MLX5_ESWITCH_MANAGER(dev)) { 3137 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) { 3138 err = init_fdb_root_ns(steering); 3139 if (err) 3140 goto err; 3141 } 3142 } 3143 3144 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) { 3145 err = init_sniffer_rx_root_ns(steering); 3146 if (err) 3147 goto err; 3148 } 3149 3150 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) { 3151 err = init_sniffer_tx_root_ns(steering); 3152 if (err) 3153 goto err; 3154 } 3155 3156 if (MLX5_CAP_FLOWTABLE_PORT_SELECTION(dev, ft_support)) { 3157 err = init_port_sel_root_ns(steering); 3158 if (err) 3159 goto err; 3160 } 3161 3162 if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) && 3163 MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) { 3164 err = init_rdma_rx_root_ns(steering); 3165 if (err) 3166 goto err; 3167 } 3168 3169 if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) { 3170 err = init_rdma_tx_root_ns(steering); 3171 if (err) 3172 goto err; 3173 } 3174 3175 if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE || 3176 MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) { 3177 err = init_egress_root_ns(steering); 3178 if (err) 3179 goto err; 3180 } 3181 3182 return 0; 3183 err: 3184 mlx5_cleanup_fs(dev); 3185 return err; 3186 } 3187 3188 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn) 3189 { 3190 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns; 3191 struct mlx5_ft_underlay_qp *new_uqp; 3192 int err = 0; 3193 3194 new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL); 3195 if (!new_uqp) 3196 return -ENOMEM; 3197 3198 mutex_lock(&root->chain_lock); 3199 3200 if (!root->root_ft) { 3201 err = -EINVAL; 3202 goto update_ft_fail; 3203 } 3204 3205 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn, 3206 false); 3207 if (err) { 3208 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n", 3209 underlay_qpn, err); 3210 goto update_ft_fail; 3211 } 3212 3213 new_uqp->qpn = underlay_qpn; 3214 list_add_tail(&new_uqp->list, &root->underlay_qpns); 3215 3216 mutex_unlock(&root->chain_lock); 3217 3218 return 0; 3219 3220 update_ft_fail: 3221 mutex_unlock(&root->chain_lock); 3222 kfree(new_uqp); 3223 return err; 3224 } 3225 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn); 3226 3227 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn) 3228 { 3229 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns; 3230 struct mlx5_ft_underlay_qp *uqp; 3231 bool found = false; 3232 int err = 0; 3233 3234 mutex_lock(&root->chain_lock); 3235 list_for_each_entry(uqp, &root->underlay_qpns, list) { 3236 if (uqp->qpn == underlay_qpn) { 3237 found = true; 3238 break; 3239 } 3240 } 3241 3242 if (!found) { 3243 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n", 3244 underlay_qpn); 3245 err = -EINVAL; 3246 goto out; 3247 } 3248 3249 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn, 3250 true); 3251 if (err) 3252 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n", 3253 underlay_qpn, err); 3254 3255 list_del(&uqp->list); 3256 mutex_unlock(&root->chain_lock); 3257 kfree(uqp); 3258 3259 return 0; 3260 3261 out: 3262 mutex_unlock(&root->chain_lock); 3263 return err; 3264 } 3265 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn); 3266 3267 static struct mlx5_flow_root_namespace 3268 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type) 3269 { 3270 struct mlx5_flow_namespace *ns; 3271 3272 if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS || 3273 ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS) 3274 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0); 3275 else 3276 ns = mlx5_get_flow_namespace(dev, ns_type); 3277 if (!ns) 3278 return NULL; 3279 3280 return find_root(&ns->node); 3281 } 3282 3283 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev, 3284 u8 ns_type, u8 num_actions, 3285 void *modify_actions) 3286 { 3287 struct mlx5_flow_root_namespace *root; 3288 struct mlx5_modify_hdr *modify_hdr; 3289 int err; 3290 3291 root = get_root_namespace(dev, ns_type); 3292 if (!root) 3293 return ERR_PTR(-EOPNOTSUPP); 3294 3295 modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL); 3296 if (!modify_hdr) 3297 return ERR_PTR(-ENOMEM); 3298 3299 modify_hdr->ns_type = ns_type; 3300 err = root->cmds->modify_header_alloc(root, ns_type, num_actions, 3301 modify_actions, modify_hdr); 3302 if (err) { 3303 kfree(modify_hdr); 3304 return ERR_PTR(err); 3305 } 3306 3307 return modify_hdr; 3308 } 3309 EXPORT_SYMBOL(mlx5_modify_header_alloc); 3310 3311 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev, 3312 struct mlx5_modify_hdr *modify_hdr) 3313 { 3314 struct mlx5_flow_root_namespace *root; 3315 3316 root = get_root_namespace(dev, modify_hdr->ns_type); 3317 if (WARN_ON(!root)) 3318 return; 3319 root->cmds->modify_header_dealloc(root, modify_hdr); 3320 kfree(modify_hdr); 3321 } 3322 EXPORT_SYMBOL(mlx5_modify_header_dealloc); 3323 3324 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev, 3325 struct mlx5_pkt_reformat_params *params, 3326 enum mlx5_flow_namespace_type ns_type) 3327 { 3328 struct mlx5_pkt_reformat *pkt_reformat; 3329 struct mlx5_flow_root_namespace *root; 3330 int err; 3331 3332 root = get_root_namespace(dev, ns_type); 3333 if (!root) 3334 return ERR_PTR(-EOPNOTSUPP); 3335 3336 pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL); 3337 if (!pkt_reformat) 3338 return ERR_PTR(-ENOMEM); 3339 3340 pkt_reformat->ns_type = ns_type; 3341 pkt_reformat->reformat_type = params->type; 3342 err = root->cmds->packet_reformat_alloc(root, params, ns_type, 3343 pkt_reformat); 3344 if (err) { 3345 kfree(pkt_reformat); 3346 return ERR_PTR(err); 3347 } 3348 3349 return pkt_reformat; 3350 } 3351 EXPORT_SYMBOL(mlx5_packet_reformat_alloc); 3352 3353 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev, 3354 struct mlx5_pkt_reformat *pkt_reformat) 3355 { 3356 struct mlx5_flow_root_namespace *root; 3357 3358 root = get_root_namespace(dev, pkt_reformat->ns_type); 3359 if (WARN_ON(!root)) 3360 return; 3361 root->cmds->packet_reformat_dealloc(root, pkt_reformat); 3362 kfree(pkt_reformat); 3363 } 3364 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc); 3365 3366 int mlx5_get_match_definer_id(struct mlx5_flow_definer *definer) 3367 { 3368 return definer->id; 3369 } 3370 3371 struct mlx5_flow_definer * 3372 mlx5_create_match_definer(struct mlx5_core_dev *dev, 3373 enum mlx5_flow_namespace_type ns_type, u16 format_id, 3374 u32 *match_mask) 3375 { 3376 struct mlx5_flow_root_namespace *root; 3377 struct mlx5_flow_definer *definer; 3378 int id; 3379 3380 root = get_root_namespace(dev, ns_type); 3381 if (!root) 3382 return ERR_PTR(-EOPNOTSUPP); 3383 3384 definer = kzalloc(sizeof(*definer), GFP_KERNEL); 3385 if (!definer) 3386 return ERR_PTR(-ENOMEM); 3387 3388 definer->ns_type = ns_type; 3389 id = root->cmds->create_match_definer(root, format_id, match_mask); 3390 if (id < 0) { 3391 mlx5_core_warn(root->dev, "Failed to create match definer (%d)\n", id); 3392 kfree(definer); 3393 return ERR_PTR(id); 3394 } 3395 definer->id = id; 3396 return definer; 3397 } 3398 3399 void mlx5_destroy_match_definer(struct mlx5_core_dev *dev, 3400 struct mlx5_flow_definer *definer) 3401 { 3402 struct mlx5_flow_root_namespace *root; 3403 3404 root = get_root_namespace(dev, definer->ns_type); 3405 if (WARN_ON(!root)) 3406 return; 3407 3408 root->cmds->destroy_match_definer(root, definer->id); 3409 kfree(definer); 3410 } 3411 3412 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns, 3413 struct mlx5_flow_root_namespace *peer_ns) 3414 { 3415 if (peer_ns && ns->mode != peer_ns->mode) { 3416 mlx5_core_err(ns->dev, 3417 "Can't peer namespace of different steering mode\n"); 3418 return -EINVAL; 3419 } 3420 3421 return ns->cmds->set_peer(ns, peer_ns); 3422 } 3423 3424 /* This function should be called only at init stage of the namespace. 3425 * It is not safe to call this function while steering operations 3426 * are executed in the namespace. 3427 */ 3428 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns, 3429 enum mlx5_flow_steering_mode mode) 3430 { 3431 struct mlx5_flow_root_namespace *root; 3432 const struct mlx5_flow_cmds *cmds; 3433 int err; 3434 3435 root = find_root(&ns->node); 3436 if (&root->ns != ns) 3437 /* Can't set cmds to non root namespace */ 3438 return -EINVAL; 3439 3440 if (root->table_type != FS_FT_FDB) 3441 return -EOPNOTSUPP; 3442 3443 if (root->mode == mode) 3444 return 0; 3445 3446 if (mode == MLX5_FLOW_STEERING_MODE_SMFS) 3447 cmds = mlx5_fs_cmd_get_dr_cmds(); 3448 else 3449 cmds = mlx5_fs_cmd_get_fw_cmds(); 3450 if (!cmds) 3451 return -EOPNOTSUPP; 3452 3453 err = cmds->create_ns(root); 3454 if (err) { 3455 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n", 3456 err); 3457 return err; 3458 } 3459 3460 root->cmds->destroy_ns(root); 3461 root->cmds = cmds; 3462 root->mode = mode; 3463 3464 return 0; 3465 } 3466