1 /* 2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 */ 32 33 #include <linux/mutex.h> 34 #include <linux/mlx5/driver.h> 35 #include <linux/mlx5/vport.h> 36 #include <linux/mlx5/eswitch.h> 37 38 #include "mlx5_core.h" 39 #include "fs_core.h" 40 #include "fs_cmd.h" 41 #include "fs_ft_pool.h" 42 #include "diag/fs_tracepoint.h" 43 44 #define INIT_TREE_NODE_ARRAY_SIZE(...) (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\ 45 sizeof(struct init_tree_node)) 46 47 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\ 48 ...) {.type = FS_TYPE_PRIO,\ 49 .min_ft_level = min_level_val,\ 50 .num_levels = num_levels_val,\ 51 .num_leaf_prios = num_prios_val,\ 52 .caps = caps_val,\ 53 .children = (struct init_tree_node[]) {__VA_ARGS__},\ 54 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \ 55 } 56 57 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\ 58 ADD_PRIO(num_prios_val, 0, num_levels_val, {},\ 59 __VA_ARGS__)\ 60 61 #define ADD_NS(def_miss_act, ...) {.type = FS_TYPE_NAMESPACE, \ 62 .def_miss_action = def_miss_act,\ 63 .children = (struct init_tree_node[]) {__VA_ARGS__},\ 64 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \ 65 } 66 67 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\ 68 sizeof(long)) 69 70 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap)) 71 72 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \ 73 .caps = (long[]) {__VA_ARGS__} } 74 75 #define FS_CHAINING_CAPS FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \ 76 FS_CAP(flow_table_properties_nic_receive.modify_root), \ 77 FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \ 78 FS_CAP(flow_table_properties_nic_receive.flow_table_modify)) 79 80 #define FS_CHAINING_CAPS_EGRESS \ 81 FS_REQUIRED_CAPS( \ 82 FS_CAP(flow_table_properties_nic_transmit.flow_modify_en), \ 83 FS_CAP(flow_table_properties_nic_transmit.modify_root), \ 84 FS_CAP(flow_table_properties_nic_transmit \ 85 .identified_miss_table_mode), \ 86 FS_CAP(flow_table_properties_nic_transmit.flow_table_modify)) 87 88 #define FS_CHAINING_CAPS_RDMA_TX \ 89 FS_REQUIRED_CAPS( \ 90 FS_CAP(flow_table_properties_nic_transmit_rdma.flow_modify_en), \ 91 FS_CAP(flow_table_properties_nic_transmit_rdma.modify_root), \ 92 FS_CAP(flow_table_properties_nic_transmit_rdma \ 93 .identified_miss_table_mode), \ 94 FS_CAP(flow_table_properties_nic_transmit_rdma \ 95 .flow_table_modify)) 96 97 #define LEFTOVERS_NUM_LEVELS 1 98 #define LEFTOVERS_NUM_PRIOS 1 99 100 #define RDMA_RX_COUNTERS_PRIO_NUM_LEVELS 1 101 #define RDMA_TX_COUNTERS_PRIO_NUM_LEVELS 1 102 103 #define BY_PASS_PRIO_NUM_LEVELS 1 104 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\ 105 LEFTOVERS_NUM_PRIOS) 106 107 #define KERNEL_RX_MACSEC_NUM_PRIOS 1 108 #define KERNEL_RX_MACSEC_NUM_LEVELS 2 109 #define KERNEL_RX_MACSEC_MIN_LEVEL (BY_PASS_MIN_LEVEL + KERNEL_RX_MACSEC_NUM_PRIOS) 110 111 #define ETHTOOL_PRIO_NUM_LEVELS 1 112 #define ETHTOOL_NUM_PRIOS 11 113 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS) 114 /* Promiscuous, Vlan, mac, ttc, inner ttc, {UDP/ANY/aRFS/accel/{esp, esp_err}}, IPsec policy */ 115 #define KERNEL_NIC_PRIO_NUM_LEVELS 8 116 #define KERNEL_NIC_NUM_PRIOS 1 117 /* One more level for tc */ 118 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1) 119 120 #define KERNEL_NIC_TC_NUM_PRIOS 1 121 #define KERNEL_NIC_TC_NUM_LEVELS 3 122 123 #define ANCHOR_NUM_LEVELS 1 124 #define ANCHOR_NUM_PRIOS 1 125 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1) 126 127 #define OFFLOADS_MAX_FT 2 128 #define OFFLOADS_NUM_PRIOS 2 129 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS) 130 131 #define LAG_PRIO_NUM_LEVELS 1 132 #define LAG_NUM_PRIOS 1 133 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + KERNEL_RX_MACSEC_MIN_LEVEL + 1) 134 135 #define KERNEL_TX_IPSEC_NUM_PRIOS 1 136 #define KERNEL_TX_IPSEC_NUM_LEVELS 2 137 #define KERNEL_TX_IPSEC_MIN_LEVEL (KERNEL_TX_IPSEC_NUM_LEVELS) 138 139 #define KERNEL_TX_MACSEC_NUM_PRIOS 1 140 #define KERNEL_TX_MACSEC_NUM_LEVELS 2 141 #define KERNEL_TX_MACSEC_MIN_LEVEL (KERNEL_TX_IPSEC_MIN_LEVEL + KERNEL_TX_MACSEC_NUM_PRIOS) 142 143 struct node_caps { 144 size_t arr_sz; 145 long *caps; 146 }; 147 148 static struct init_tree_node { 149 enum fs_node_type type; 150 struct init_tree_node *children; 151 int ar_size; 152 struct node_caps caps; 153 int min_ft_level; 154 int num_leaf_prios; 155 int prio; 156 int num_levels; 157 enum mlx5_flow_table_miss_action def_miss_action; 158 } root_fs = { 159 .type = FS_TYPE_NAMESPACE, 160 .ar_size = 8, 161 .children = (struct init_tree_node[]){ 162 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS, 163 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 164 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS, 165 BY_PASS_PRIO_NUM_LEVELS))), 166 ADD_PRIO(0, KERNEL_RX_MACSEC_MIN_LEVEL, 0, FS_CHAINING_CAPS, 167 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 168 ADD_MULTIPLE_PRIO(KERNEL_RX_MACSEC_NUM_PRIOS, 169 KERNEL_RX_MACSEC_NUM_LEVELS))), 170 ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS, 171 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 172 ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS, 173 LAG_PRIO_NUM_LEVELS))), 174 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS, 175 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 176 ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, 177 OFFLOADS_MAX_FT))), 178 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS, 179 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 180 ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS, 181 ETHTOOL_PRIO_NUM_LEVELS))), 182 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {}, 183 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 184 ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS, 185 KERNEL_NIC_TC_NUM_LEVELS), 186 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS, 187 KERNEL_NIC_PRIO_NUM_LEVELS))), 188 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS, 189 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 190 ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, 191 LEFTOVERS_NUM_LEVELS))), 192 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {}, 193 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 194 ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, 195 ANCHOR_NUM_LEVELS))), 196 } 197 }; 198 199 static struct init_tree_node egress_root_fs = { 200 .type = FS_TYPE_NAMESPACE, 201 .ar_size = 3, 202 .children = (struct init_tree_node[]) { 203 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0, 204 FS_CHAINING_CAPS_EGRESS, 205 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 206 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS, 207 BY_PASS_PRIO_NUM_LEVELS))), 208 ADD_PRIO(0, KERNEL_TX_IPSEC_MIN_LEVEL, 0, 209 FS_CHAINING_CAPS_EGRESS, 210 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 211 ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS, 212 KERNEL_TX_IPSEC_NUM_LEVELS))), 213 ADD_PRIO(0, KERNEL_TX_MACSEC_MIN_LEVEL, 0, 214 FS_CHAINING_CAPS_EGRESS, 215 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 216 ADD_MULTIPLE_PRIO(KERNEL_TX_MACSEC_NUM_PRIOS, 217 KERNEL_TX_MACSEC_NUM_LEVELS))), 218 } 219 }; 220 221 enum { 222 RDMA_RX_COUNTERS_PRIO, 223 RDMA_RX_BYPASS_PRIO, 224 RDMA_RX_KERNEL_PRIO, 225 }; 226 227 #define RDMA_RX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_REGULAR_PRIOS 228 #define RDMA_RX_KERNEL_MIN_LEVEL (RDMA_RX_BYPASS_MIN_LEVEL + 1) 229 #define RDMA_RX_COUNTERS_MIN_LEVEL (RDMA_RX_KERNEL_MIN_LEVEL + 2) 230 231 static struct init_tree_node rdma_rx_root_fs = { 232 .type = FS_TYPE_NAMESPACE, 233 .ar_size = 3, 234 .children = (struct init_tree_node[]) { 235 [RDMA_RX_COUNTERS_PRIO] = 236 ADD_PRIO(0, RDMA_RX_COUNTERS_MIN_LEVEL, 0, 237 FS_CHAINING_CAPS, 238 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 239 ADD_MULTIPLE_PRIO(MLX5_RDMA_RX_NUM_COUNTERS_PRIOS, 240 RDMA_RX_COUNTERS_PRIO_NUM_LEVELS))), 241 [RDMA_RX_BYPASS_PRIO] = 242 ADD_PRIO(0, RDMA_RX_BYPASS_MIN_LEVEL, 0, 243 FS_CHAINING_CAPS, 244 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 245 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS, 246 BY_PASS_PRIO_NUM_LEVELS))), 247 [RDMA_RX_KERNEL_PRIO] = 248 ADD_PRIO(0, RDMA_RX_KERNEL_MIN_LEVEL, 0, 249 FS_CHAINING_CAPS, 250 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN, 251 ADD_MULTIPLE_PRIO(1, 1))), 252 } 253 }; 254 255 enum { 256 RDMA_TX_COUNTERS_PRIO, 257 RDMA_TX_BYPASS_PRIO, 258 }; 259 260 #define RDMA_TX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_PRIOS 261 #define RDMA_TX_COUNTERS_MIN_LEVEL (RDMA_TX_BYPASS_MIN_LEVEL + 1) 262 263 static struct init_tree_node rdma_tx_root_fs = { 264 .type = FS_TYPE_NAMESPACE, 265 .ar_size = 2, 266 .children = (struct init_tree_node[]) { 267 [RDMA_TX_COUNTERS_PRIO] = 268 ADD_PRIO(0, RDMA_TX_COUNTERS_MIN_LEVEL, 0, 269 FS_CHAINING_CAPS, 270 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 271 ADD_MULTIPLE_PRIO(MLX5_RDMA_TX_NUM_COUNTERS_PRIOS, 272 RDMA_TX_COUNTERS_PRIO_NUM_LEVELS))), 273 [RDMA_TX_BYPASS_PRIO] = 274 ADD_PRIO(0, RDMA_TX_BYPASS_MIN_LEVEL, 0, 275 FS_CHAINING_CAPS_RDMA_TX, 276 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF, 277 ADD_MULTIPLE_PRIO(RDMA_TX_BYPASS_MIN_LEVEL, 278 BY_PASS_PRIO_NUM_LEVELS))), 279 } 280 }; 281 282 enum fs_i_lock_class { 283 FS_LOCK_GRANDPARENT, 284 FS_LOCK_PARENT, 285 FS_LOCK_CHILD 286 }; 287 288 static const struct rhashtable_params rhash_fte = { 289 .key_len = sizeof_field(struct fs_fte, val), 290 .key_offset = offsetof(struct fs_fte, val), 291 .head_offset = offsetof(struct fs_fte, hash), 292 .automatic_shrinking = true, 293 .min_size = 1, 294 }; 295 296 static const struct rhashtable_params rhash_fg = { 297 .key_len = sizeof_field(struct mlx5_flow_group, mask), 298 .key_offset = offsetof(struct mlx5_flow_group, mask), 299 .head_offset = offsetof(struct mlx5_flow_group, hash), 300 .automatic_shrinking = true, 301 .min_size = 1, 302 303 }; 304 305 static void del_hw_flow_table(struct fs_node *node); 306 static void del_hw_flow_group(struct fs_node *node); 307 static void del_hw_fte(struct fs_node *node); 308 static void del_sw_flow_table(struct fs_node *node); 309 static void del_sw_flow_group(struct fs_node *node); 310 static void del_sw_fte(struct fs_node *node); 311 static void del_sw_prio(struct fs_node *node); 312 static void del_sw_ns(struct fs_node *node); 313 /* Delete rule (destination) is special case that 314 * requires to lock the FTE for all the deletion process. 315 */ 316 static void del_sw_hw_rule(struct fs_node *node); 317 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1, 318 struct mlx5_flow_destination *d2); 319 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns); 320 static struct mlx5_flow_rule * 321 find_flow_rule(struct fs_fte *fte, 322 struct mlx5_flow_destination *dest); 323 324 static void tree_init_node(struct fs_node *node, 325 void (*del_hw_func)(struct fs_node *), 326 void (*del_sw_func)(struct fs_node *)) 327 { 328 refcount_set(&node->refcount, 1); 329 INIT_LIST_HEAD(&node->list); 330 INIT_LIST_HEAD(&node->children); 331 init_rwsem(&node->lock); 332 node->del_hw_func = del_hw_func; 333 node->del_sw_func = del_sw_func; 334 node->active = false; 335 } 336 337 static void tree_add_node(struct fs_node *node, struct fs_node *parent) 338 { 339 if (parent) 340 refcount_inc(&parent->refcount); 341 node->parent = parent; 342 343 /* Parent is the root */ 344 if (!parent) 345 node->root = node; 346 else 347 node->root = parent->root; 348 } 349 350 static int tree_get_node(struct fs_node *node) 351 { 352 return refcount_inc_not_zero(&node->refcount); 353 } 354 355 static void nested_down_read_ref_node(struct fs_node *node, 356 enum fs_i_lock_class class) 357 { 358 if (node) { 359 down_read_nested(&node->lock, class); 360 refcount_inc(&node->refcount); 361 } 362 } 363 364 static void nested_down_write_ref_node(struct fs_node *node, 365 enum fs_i_lock_class class) 366 { 367 if (node) { 368 down_write_nested(&node->lock, class); 369 refcount_inc(&node->refcount); 370 } 371 } 372 373 static void down_write_ref_node(struct fs_node *node, bool locked) 374 { 375 if (node) { 376 if (!locked) 377 down_write(&node->lock); 378 refcount_inc(&node->refcount); 379 } 380 } 381 382 static void up_read_ref_node(struct fs_node *node) 383 { 384 refcount_dec(&node->refcount); 385 up_read(&node->lock); 386 } 387 388 static void up_write_ref_node(struct fs_node *node, bool locked) 389 { 390 refcount_dec(&node->refcount); 391 if (!locked) 392 up_write(&node->lock); 393 } 394 395 static void tree_put_node(struct fs_node *node, bool locked) 396 { 397 struct fs_node *parent_node = node->parent; 398 399 if (refcount_dec_and_test(&node->refcount)) { 400 if (node->del_hw_func) 401 node->del_hw_func(node); 402 if (parent_node) { 403 down_write_ref_node(parent_node, locked); 404 list_del_init(&node->list); 405 } 406 node->del_sw_func(node); 407 if (parent_node) 408 up_write_ref_node(parent_node, locked); 409 node = NULL; 410 } 411 if (!node && parent_node) 412 tree_put_node(parent_node, locked); 413 } 414 415 static int tree_remove_node(struct fs_node *node, bool locked) 416 { 417 if (refcount_read(&node->refcount) > 1) { 418 refcount_dec(&node->refcount); 419 return -EEXIST; 420 } 421 tree_put_node(node, locked); 422 return 0; 423 } 424 425 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns, 426 unsigned int prio) 427 { 428 struct fs_prio *iter_prio; 429 430 fs_for_each_prio(iter_prio, ns) { 431 if (iter_prio->prio == prio) 432 return iter_prio; 433 } 434 435 return NULL; 436 } 437 438 static bool is_fwd_next_action(u32 action) 439 { 440 return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO | 441 MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS); 442 } 443 444 static bool is_fwd_dest_type(enum mlx5_flow_destination_type type) 445 { 446 return type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM || 447 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE || 448 type == MLX5_FLOW_DESTINATION_TYPE_UPLINK || 449 type == MLX5_FLOW_DESTINATION_TYPE_VPORT || 450 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER || 451 type == MLX5_FLOW_DESTINATION_TYPE_TIR || 452 type == MLX5_FLOW_DESTINATION_TYPE_RANGE; 453 } 454 455 static bool check_valid_spec(const struct mlx5_flow_spec *spec) 456 { 457 int i; 458 459 for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++) 460 if (spec->match_value[i] & ~spec->match_criteria[i]) { 461 pr_warn("mlx5_core: match_value differs from match_criteria\n"); 462 return false; 463 } 464 465 return true; 466 } 467 468 struct mlx5_flow_root_namespace *find_root(struct fs_node *node) 469 { 470 struct fs_node *root; 471 struct mlx5_flow_namespace *ns; 472 473 root = node->root; 474 475 if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) { 476 pr_warn("mlx5: flow steering node is not in tree or garbaged\n"); 477 return NULL; 478 } 479 480 ns = container_of(root, struct mlx5_flow_namespace, node); 481 return container_of(ns, struct mlx5_flow_root_namespace, ns); 482 } 483 484 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node) 485 { 486 struct mlx5_flow_root_namespace *root = find_root(node); 487 488 if (root) 489 return root->dev->priv.steering; 490 return NULL; 491 } 492 493 static inline struct mlx5_core_dev *get_dev(struct fs_node *node) 494 { 495 struct mlx5_flow_root_namespace *root = find_root(node); 496 497 if (root) 498 return root->dev; 499 return NULL; 500 } 501 502 static void del_sw_ns(struct fs_node *node) 503 { 504 kfree(node); 505 } 506 507 static void del_sw_prio(struct fs_node *node) 508 { 509 kfree(node); 510 } 511 512 static void del_hw_flow_table(struct fs_node *node) 513 { 514 struct mlx5_flow_root_namespace *root; 515 struct mlx5_flow_table *ft; 516 struct mlx5_core_dev *dev; 517 int err; 518 519 fs_get_obj(ft, node); 520 dev = get_dev(&ft->node); 521 root = find_root(&ft->node); 522 trace_mlx5_fs_del_ft(ft); 523 524 if (node->active) { 525 err = root->cmds->destroy_flow_table(root, ft); 526 if (err) 527 mlx5_core_warn(dev, "flow steering can't destroy ft\n"); 528 } 529 } 530 531 static void del_sw_flow_table(struct fs_node *node) 532 { 533 struct mlx5_flow_table *ft; 534 struct fs_prio *prio; 535 536 fs_get_obj(ft, node); 537 538 rhltable_destroy(&ft->fgs_hash); 539 if (ft->node.parent) { 540 fs_get_obj(prio, ft->node.parent); 541 prio->num_ft--; 542 } 543 kfree(ft); 544 } 545 546 static void modify_fte(struct fs_fte *fte) 547 { 548 struct mlx5_flow_root_namespace *root; 549 struct mlx5_flow_table *ft; 550 struct mlx5_flow_group *fg; 551 struct mlx5_core_dev *dev; 552 int err; 553 554 fs_get_obj(fg, fte->node.parent); 555 fs_get_obj(ft, fg->node.parent); 556 dev = get_dev(&fte->node); 557 558 root = find_root(&ft->node); 559 err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte); 560 if (err) 561 mlx5_core_warn(dev, 562 "%s can't del rule fg id=%d fte_index=%d\n", 563 __func__, fg->id, fte->index); 564 fte->modify_mask = 0; 565 } 566 567 static void del_sw_hw_rule(struct fs_node *node) 568 { 569 struct mlx5_flow_rule *rule; 570 struct fs_fte *fte; 571 572 fs_get_obj(rule, node); 573 fs_get_obj(fte, rule->node.parent); 574 trace_mlx5_fs_del_rule(rule); 575 if (is_fwd_next_action(rule->sw_action)) { 576 mutex_lock(&rule->dest_attr.ft->lock); 577 list_del(&rule->next_ft); 578 mutex_unlock(&rule->dest_attr.ft->lock); 579 } 580 581 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER) { 582 --fte->dests_size; 583 fte->modify_mask |= 584 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) | 585 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS); 586 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT; 587 goto out; 588 } 589 590 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT) { 591 --fte->dests_size; 592 fte->modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION); 593 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW; 594 goto out; 595 } 596 597 if (is_fwd_dest_type(rule->dest_attr.type)) { 598 --fte->dests_size; 599 --fte->fwd_dests; 600 601 if (!fte->fwd_dests) 602 fte->action.action &= 603 ~MLX5_FLOW_CONTEXT_ACTION_FWD_DEST; 604 fte->modify_mask |= 605 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST); 606 goto out; 607 } 608 out: 609 kfree(rule); 610 } 611 612 static void del_hw_fte(struct fs_node *node) 613 { 614 struct mlx5_flow_root_namespace *root; 615 struct mlx5_flow_table *ft; 616 struct mlx5_flow_group *fg; 617 struct mlx5_core_dev *dev; 618 struct fs_fte *fte; 619 int err; 620 621 fs_get_obj(fte, node); 622 fs_get_obj(fg, fte->node.parent); 623 fs_get_obj(ft, fg->node.parent); 624 625 trace_mlx5_fs_del_fte(fte); 626 WARN_ON(fte->dests_size); 627 dev = get_dev(&ft->node); 628 root = find_root(&ft->node); 629 if (node->active) { 630 err = root->cmds->delete_fte(root, ft, fte); 631 if (err) 632 mlx5_core_warn(dev, 633 "flow steering can't delete fte in index %d of flow group id %d\n", 634 fte->index, fg->id); 635 node->active = false; 636 } 637 } 638 639 static void del_sw_fte(struct fs_node *node) 640 { 641 struct mlx5_flow_steering *steering = get_steering(node); 642 struct mlx5_flow_group *fg; 643 struct fs_fte *fte; 644 int err; 645 646 fs_get_obj(fte, node); 647 fs_get_obj(fg, fte->node.parent); 648 649 err = rhashtable_remove_fast(&fg->ftes_hash, 650 &fte->hash, 651 rhash_fte); 652 WARN_ON(err); 653 ida_free(&fg->fte_allocator, fte->index - fg->start_index); 654 kmem_cache_free(steering->ftes_cache, fte); 655 } 656 657 static void del_hw_flow_group(struct fs_node *node) 658 { 659 struct mlx5_flow_root_namespace *root; 660 struct mlx5_flow_group *fg; 661 struct mlx5_flow_table *ft; 662 struct mlx5_core_dev *dev; 663 664 fs_get_obj(fg, node); 665 fs_get_obj(ft, fg->node.parent); 666 dev = get_dev(&ft->node); 667 trace_mlx5_fs_del_fg(fg); 668 669 root = find_root(&ft->node); 670 if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg)) 671 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n", 672 fg->id, ft->id); 673 } 674 675 static void del_sw_flow_group(struct fs_node *node) 676 { 677 struct mlx5_flow_steering *steering = get_steering(node); 678 struct mlx5_flow_group *fg; 679 struct mlx5_flow_table *ft; 680 int err; 681 682 fs_get_obj(fg, node); 683 fs_get_obj(ft, fg->node.parent); 684 685 rhashtable_destroy(&fg->ftes_hash); 686 ida_destroy(&fg->fte_allocator); 687 if (ft->autogroup.active && 688 fg->max_ftes == ft->autogroup.group_size && 689 fg->start_index < ft->autogroup.max_fte) 690 ft->autogroup.num_groups--; 691 err = rhltable_remove(&ft->fgs_hash, 692 &fg->hash, 693 rhash_fg); 694 WARN_ON(err); 695 kmem_cache_free(steering->fgs_cache, fg); 696 } 697 698 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte) 699 { 700 int index; 701 int ret; 702 703 index = ida_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, GFP_KERNEL); 704 if (index < 0) 705 return index; 706 707 fte->index = index + fg->start_index; 708 ret = rhashtable_insert_fast(&fg->ftes_hash, 709 &fte->hash, 710 rhash_fte); 711 if (ret) 712 goto err_ida_remove; 713 714 tree_add_node(&fte->node, &fg->node); 715 list_add_tail(&fte->node.list, &fg->node.children); 716 return 0; 717 718 err_ida_remove: 719 ida_free(&fg->fte_allocator, index); 720 return ret; 721 } 722 723 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft, 724 const struct mlx5_flow_spec *spec, 725 struct mlx5_flow_act *flow_act) 726 { 727 struct mlx5_flow_steering *steering = get_steering(&ft->node); 728 struct fs_fte *fte; 729 730 fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL); 731 if (!fte) 732 return ERR_PTR(-ENOMEM); 733 734 memcpy(fte->val, &spec->match_value, sizeof(fte->val)); 735 fte->node.type = FS_TYPE_FLOW_ENTRY; 736 fte->action = *flow_act; 737 fte->flow_context = spec->flow_context; 738 739 tree_init_node(&fte->node, del_hw_fte, del_sw_fte); 740 741 return fte; 742 } 743 744 static void dealloc_flow_group(struct mlx5_flow_steering *steering, 745 struct mlx5_flow_group *fg) 746 { 747 rhashtable_destroy(&fg->ftes_hash); 748 kmem_cache_free(steering->fgs_cache, fg); 749 } 750 751 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering, 752 u8 match_criteria_enable, 753 const void *match_criteria, 754 int start_index, 755 int end_index) 756 { 757 struct mlx5_flow_group *fg; 758 int ret; 759 760 fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL); 761 if (!fg) 762 return ERR_PTR(-ENOMEM); 763 764 ret = rhashtable_init(&fg->ftes_hash, &rhash_fte); 765 if (ret) { 766 kmem_cache_free(steering->fgs_cache, fg); 767 return ERR_PTR(ret); 768 } 769 770 ida_init(&fg->fte_allocator); 771 fg->mask.match_criteria_enable = match_criteria_enable; 772 memcpy(&fg->mask.match_criteria, match_criteria, 773 sizeof(fg->mask.match_criteria)); 774 fg->node.type = FS_TYPE_FLOW_GROUP; 775 fg->start_index = start_index; 776 fg->max_ftes = end_index - start_index + 1; 777 778 return fg; 779 } 780 781 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft, 782 u8 match_criteria_enable, 783 const void *match_criteria, 784 int start_index, 785 int end_index, 786 struct list_head *prev) 787 { 788 struct mlx5_flow_steering *steering = get_steering(&ft->node); 789 struct mlx5_flow_group *fg; 790 int ret; 791 792 fg = alloc_flow_group(steering, match_criteria_enable, match_criteria, 793 start_index, end_index); 794 if (IS_ERR(fg)) 795 return fg; 796 797 /* initialize refcnt, add to parent list */ 798 ret = rhltable_insert(&ft->fgs_hash, 799 &fg->hash, 800 rhash_fg); 801 if (ret) { 802 dealloc_flow_group(steering, fg); 803 return ERR_PTR(ret); 804 } 805 806 tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group); 807 tree_add_node(&fg->node, &ft->node); 808 /* Add node to group list */ 809 list_add(&fg->node.list, prev); 810 atomic_inc(&ft->node.version); 811 812 return fg; 813 } 814 815 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, 816 enum fs_flow_table_type table_type, 817 enum fs_flow_table_op_mod op_mod, 818 u32 flags) 819 { 820 struct mlx5_flow_table *ft; 821 int ret; 822 823 ft = kzalloc(sizeof(*ft), GFP_KERNEL); 824 if (!ft) 825 return ERR_PTR(-ENOMEM); 826 827 ret = rhltable_init(&ft->fgs_hash, &rhash_fg); 828 if (ret) { 829 kfree(ft); 830 return ERR_PTR(ret); 831 } 832 833 ft->level = level; 834 ft->node.type = FS_TYPE_FLOW_TABLE; 835 ft->op_mod = op_mod; 836 ft->type = table_type; 837 ft->vport = vport; 838 ft->flags = flags; 839 INIT_LIST_HEAD(&ft->fwd_rules); 840 mutex_init(&ft->lock); 841 842 return ft; 843 } 844 845 /* If reverse is false, then we search for the first flow table in the 846 * root sub-tree from start(closest from right), else we search for the 847 * last flow table in the root sub-tree till start(closest from left). 848 */ 849 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node *root, 850 struct list_head *start, 851 bool reverse) 852 { 853 #define list_advance_entry(pos, reverse) \ 854 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list)) 855 856 #define list_for_each_advance_continue(pos, head, reverse) \ 857 for (pos = list_advance_entry(pos, reverse); \ 858 &pos->list != (head); \ 859 pos = list_advance_entry(pos, reverse)) 860 861 struct fs_node *iter = list_entry(start, struct fs_node, list); 862 struct mlx5_flow_table *ft = NULL; 863 864 if (!root || root->type == FS_TYPE_PRIO_CHAINS) 865 return NULL; 866 867 list_for_each_advance_continue(iter, &root->children, reverse) { 868 if (iter->type == FS_TYPE_FLOW_TABLE) { 869 fs_get_obj(ft, iter); 870 return ft; 871 } 872 ft = find_closest_ft_recursive(iter, &iter->children, reverse); 873 if (ft) 874 return ft; 875 } 876 877 return ft; 878 } 879 880 /* If reverse is false then return the first flow table in next priority of 881 * prio in the tree, else return the last flow table in the previous priority 882 * of prio in the tree. 883 */ 884 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse) 885 { 886 struct mlx5_flow_table *ft = NULL; 887 struct fs_node *curr_node; 888 struct fs_node *parent; 889 890 parent = prio->node.parent; 891 curr_node = &prio->node; 892 while (!ft && parent) { 893 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse); 894 curr_node = parent; 895 parent = curr_node->parent; 896 } 897 return ft; 898 } 899 900 /* Assuming all the tree is locked by mutex chain lock */ 901 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio) 902 { 903 return find_closest_ft(prio, false); 904 } 905 906 /* Assuming all the tree is locked by mutex chain lock */ 907 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio) 908 { 909 return find_closest_ft(prio, true); 910 } 911 912 static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft, 913 struct mlx5_flow_act *flow_act) 914 { 915 struct fs_prio *prio; 916 bool next_ns; 917 918 next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS; 919 fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent); 920 921 return find_next_chained_ft(prio); 922 } 923 924 static int connect_fts_in_prio(struct mlx5_core_dev *dev, 925 struct fs_prio *prio, 926 struct mlx5_flow_table *ft) 927 { 928 struct mlx5_flow_root_namespace *root = find_root(&prio->node); 929 struct mlx5_flow_table *iter; 930 int err; 931 932 fs_for_each_ft(iter, prio) { 933 err = root->cmds->modify_flow_table(root, iter, ft); 934 if (err) { 935 mlx5_core_err(dev, 936 "Failed to modify flow table id %d, type %d, err %d\n", 937 iter->id, iter->type, err); 938 /* The driver is out of sync with the FW */ 939 return err; 940 } 941 } 942 return 0; 943 } 944 945 /* Connect flow tables from previous priority of prio to ft */ 946 static int connect_prev_fts(struct mlx5_core_dev *dev, 947 struct mlx5_flow_table *ft, 948 struct fs_prio *prio) 949 { 950 struct mlx5_flow_table *prev_ft; 951 952 prev_ft = find_prev_chained_ft(prio); 953 if (prev_ft) { 954 struct fs_prio *prev_prio; 955 956 fs_get_obj(prev_prio, prev_ft->node.parent); 957 return connect_fts_in_prio(dev, prev_prio, ft); 958 } 959 return 0; 960 } 961 962 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio 963 *prio) 964 { 965 struct mlx5_flow_root_namespace *root = find_root(&prio->node); 966 struct mlx5_ft_underlay_qp *uqp; 967 int min_level = INT_MAX; 968 int err = 0; 969 u32 qpn; 970 971 if (root->root_ft) 972 min_level = root->root_ft->level; 973 974 if (ft->level >= min_level) 975 return 0; 976 977 if (list_empty(&root->underlay_qpns)) { 978 /* Don't set any QPN (zero) in case QPN list is empty */ 979 qpn = 0; 980 err = root->cmds->update_root_ft(root, ft, qpn, false); 981 } else { 982 list_for_each_entry(uqp, &root->underlay_qpns, list) { 983 qpn = uqp->qpn; 984 err = root->cmds->update_root_ft(root, ft, 985 qpn, false); 986 if (err) 987 break; 988 } 989 } 990 991 if (err) 992 mlx5_core_warn(root->dev, 993 "Update root flow table of id(%u) qpn(%d) failed\n", 994 ft->id, qpn); 995 else 996 root->root_ft = ft; 997 998 return err; 999 } 1000 1001 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule, 1002 struct mlx5_flow_destination *dest) 1003 { 1004 struct mlx5_flow_root_namespace *root; 1005 struct mlx5_flow_table *ft; 1006 struct mlx5_flow_group *fg; 1007 struct fs_fte *fte; 1008 int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST); 1009 int err = 0; 1010 1011 fs_get_obj(fte, rule->node.parent); 1012 if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)) 1013 return -EINVAL; 1014 down_write_ref_node(&fte->node, false); 1015 fs_get_obj(fg, fte->node.parent); 1016 fs_get_obj(ft, fg->node.parent); 1017 1018 memcpy(&rule->dest_attr, dest, sizeof(*dest)); 1019 root = find_root(&ft->node); 1020 err = root->cmds->update_fte(root, ft, fg, 1021 modify_mask, fte); 1022 up_write_ref_node(&fte->node, false); 1023 1024 return err; 1025 } 1026 1027 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle, 1028 struct mlx5_flow_destination *new_dest, 1029 struct mlx5_flow_destination *old_dest) 1030 { 1031 int i; 1032 1033 if (!old_dest) { 1034 if (handle->num_rules != 1) 1035 return -EINVAL; 1036 return _mlx5_modify_rule_destination(handle->rule[0], 1037 new_dest); 1038 } 1039 1040 for (i = 0; i < handle->num_rules; i++) { 1041 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr)) 1042 return _mlx5_modify_rule_destination(handle->rule[i], 1043 new_dest); 1044 } 1045 1046 return -EINVAL; 1047 } 1048 1049 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */ 1050 static int connect_fwd_rules(struct mlx5_core_dev *dev, 1051 struct mlx5_flow_table *new_next_ft, 1052 struct mlx5_flow_table *old_next_ft) 1053 { 1054 struct mlx5_flow_destination dest = {}; 1055 struct mlx5_flow_rule *iter; 1056 int err = 0; 1057 1058 /* new_next_ft and old_next_ft could be NULL only 1059 * when we create/destroy the anchor flow table. 1060 */ 1061 if (!new_next_ft || !old_next_ft) 1062 return 0; 1063 1064 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE; 1065 dest.ft = new_next_ft; 1066 1067 mutex_lock(&old_next_ft->lock); 1068 list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules); 1069 mutex_unlock(&old_next_ft->lock); 1070 list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) { 1071 if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) && 1072 iter->ft->ns == new_next_ft->ns) 1073 continue; 1074 1075 err = _mlx5_modify_rule_destination(iter, &dest); 1076 if (err) 1077 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n", 1078 new_next_ft->id); 1079 } 1080 return 0; 1081 } 1082 1083 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft, 1084 struct fs_prio *prio) 1085 { 1086 struct mlx5_flow_table *next_ft, *first_ft; 1087 int err = 0; 1088 1089 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */ 1090 1091 first_ft = list_first_entry_or_null(&prio->node.children, 1092 struct mlx5_flow_table, node.list); 1093 if (!first_ft || first_ft->level > ft->level) { 1094 err = connect_prev_fts(dev, ft, prio); 1095 if (err) 1096 return err; 1097 1098 next_ft = first_ft ? first_ft : find_next_chained_ft(prio); 1099 err = connect_fwd_rules(dev, ft, next_ft); 1100 if (err) 1101 return err; 1102 } 1103 1104 if (MLX5_CAP_FLOWTABLE(dev, 1105 flow_table_properties_nic_receive.modify_root)) 1106 err = update_root_ft_create(ft, prio); 1107 return err; 1108 } 1109 1110 static void list_add_flow_table(struct mlx5_flow_table *ft, 1111 struct fs_prio *prio) 1112 { 1113 struct list_head *prev = &prio->node.children; 1114 struct mlx5_flow_table *iter; 1115 1116 fs_for_each_ft(iter, prio) { 1117 if (iter->level > ft->level) 1118 break; 1119 prev = &iter->node.list; 1120 } 1121 list_add(&ft->node.list, prev); 1122 } 1123 1124 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns, 1125 struct mlx5_flow_table_attr *ft_attr, 1126 enum fs_flow_table_op_mod op_mod, 1127 u16 vport) 1128 { 1129 struct mlx5_flow_root_namespace *root = find_root(&ns->node); 1130 bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED; 1131 struct mlx5_flow_table *next_ft; 1132 struct fs_prio *fs_prio = NULL; 1133 struct mlx5_flow_table *ft; 1134 int err; 1135 1136 if (!root) { 1137 pr_err("mlx5: flow steering failed to find root of namespace\n"); 1138 return ERR_PTR(-ENODEV); 1139 } 1140 1141 mutex_lock(&root->chain_lock); 1142 fs_prio = find_prio(ns, ft_attr->prio); 1143 if (!fs_prio) { 1144 err = -EINVAL; 1145 goto unlock_root; 1146 } 1147 if (!unmanaged) { 1148 /* The level is related to the 1149 * priority level range. 1150 */ 1151 if (ft_attr->level >= fs_prio->num_levels) { 1152 err = -ENOSPC; 1153 goto unlock_root; 1154 } 1155 1156 ft_attr->level += fs_prio->start_level; 1157 } 1158 1159 /* The level is related to the 1160 * priority level range. 1161 */ 1162 ft = alloc_flow_table(ft_attr->level, 1163 vport, 1164 root->table_type, 1165 op_mod, ft_attr->flags); 1166 if (IS_ERR(ft)) { 1167 err = PTR_ERR(ft); 1168 goto unlock_root; 1169 } 1170 1171 tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table); 1172 next_ft = unmanaged ? ft_attr->next_ft : 1173 find_next_chained_ft(fs_prio); 1174 ft->def_miss_action = ns->def_miss_action; 1175 ft->ns = ns; 1176 err = root->cmds->create_flow_table(root, ft, ft_attr, next_ft); 1177 if (err) 1178 goto free_ft; 1179 1180 if (!unmanaged) { 1181 err = connect_flow_table(root->dev, ft, fs_prio); 1182 if (err) 1183 goto destroy_ft; 1184 } 1185 1186 ft->node.active = true; 1187 down_write_ref_node(&fs_prio->node, false); 1188 if (!unmanaged) { 1189 tree_add_node(&ft->node, &fs_prio->node); 1190 list_add_flow_table(ft, fs_prio); 1191 } else { 1192 ft->node.root = fs_prio->node.root; 1193 } 1194 fs_prio->num_ft++; 1195 up_write_ref_node(&fs_prio->node, false); 1196 mutex_unlock(&root->chain_lock); 1197 trace_mlx5_fs_add_ft(ft); 1198 return ft; 1199 destroy_ft: 1200 root->cmds->destroy_flow_table(root, ft); 1201 free_ft: 1202 rhltable_destroy(&ft->fgs_hash); 1203 kfree(ft); 1204 unlock_root: 1205 mutex_unlock(&root->chain_lock); 1206 return ERR_PTR(err); 1207 } 1208 1209 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns, 1210 struct mlx5_flow_table_attr *ft_attr) 1211 { 1212 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0); 1213 } 1214 EXPORT_SYMBOL(mlx5_create_flow_table); 1215 1216 u32 mlx5_flow_table_id(struct mlx5_flow_table *ft) 1217 { 1218 return ft->id; 1219 } 1220 EXPORT_SYMBOL(mlx5_flow_table_id); 1221 1222 struct mlx5_flow_table * 1223 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns, 1224 struct mlx5_flow_table_attr *ft_attr, u16 vport) 1225 { 1226 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport); 1227 } 1228 1229 struct mlx5_flow_table* 1230 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns, 1231 int prio, u32 level) 1232 { 1233 struct mlx5_flow_table_attr ft_attr = {}; 1234 1235 ft_attr.level = level; 1236 ft_attr.prio = prio; 1237 ft_attr.max_fte = 1; 1238 1239 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0); 1240 } 1241 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table); 1242 1243 #define MAX_FLOW_GROUP_SIZE BIT(24) 1244 struct mlx5_flow_table* 1245 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns, 1246 struct mlx5_flow_table_attr *ft_attr) 1247 { 1248 int num_reserved_entries = ft_attr->autogroup.num_reserved_entries; 1249 int max_num_groups = ft_attr->autogroup.max_num_groups; 1250 struct mlx5_flow_table *ft; 1251 int autogroups_max_fte; 1252 1253 ft = mlx5_create_flow_table(ns, ft_attr); 1254 if (IS_ERR(ft)) 1255 return ft; 1256 1257 autogroups_max_fte = ft->max_fte - num_reserved_entries; 1258 if (max_num_groups > autogroups_max_fte) 1259 goto err_validate; 1260 if (num_reserved_entries > ft->max_fte) 1261 goto err_validate; 1262 1263 /* Align the number of groups according to the largest group size */ 1264 if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE) 1265 max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1; 1266 1267 ft->autogroup.active = true; 1268 ft->autogroup.required_groups = max_num_groups; 1269 ft->autogroup.max_fte = autogroups_max_fte; 1270 /* We save place for flow groups in addition to max types */ 1271 ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1); 1272 1273 return ft; 1274 1275 err_validate: 1276 mlx5_destroy_flow_table(ft); 1277 return ERR_PTR(-ENOSPC); 1278 } 1279 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table); 1280 1281 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft, 1282 u32 *fg_in) 1283 { 1284 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 1285 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in, 1286 fg_in, match_criteria); 1287 u8 match_criteria_enable = MLX5_GET(create_flow_group_in, 1288 fg_in, 1289 match_criteria_enable); 1290 int start_index = MLX5_GET(create_flow_group_in, fg_in, 1291 start_flow_index); 1292 int end_index = MLX5_GET(create_flow_group_in, fg_in, 1293 end_flow_index); 1294 struct mlx5_flow_group *fg; 1295 int err; 1296 1297 if (ft->autogroup.active && start_index < ft->autogroup.max_fte) 1298 return ERR_PTR(-EPERM); 1299 1300 down_write_ref_node(&ft->node, false); 1301 fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria, 1302 start_index, end_index, 1303 ft->node.children.prev); 1304 up_write_ref_node(&ft->node, false); 1305 if (IS_ERR(fg)) 1306 return fg; 1307 1308 err = root->cmds->create_flow_group(root, ft, fg_in, fg); 1309 if (err) { 1310 tree_put_node(&fg->node, false); 1311 return ERR_PTR(err); 1312 } 1313 trace_mlx5_fs_add_fg(fg); 1314 fg->node.active = true; 1315 1316 return fg; 1317 } 1318 EXPORT_SYMBOL(mlx5_create_flow_group); 1319 1320 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest) 1321 { 1322 struct mlx5_flow_rule *rule; 1323 1324 rule = kzalloc(sizeof(*rule), GFP_KERNEL); 1325 if (!rule) 1326 return NULL; 1327 1328 INIT_LIST_HEAD(&rule->next_ft); 1329 rule->node.type = FS_TYPE_FLOW_DEST; 1330 if (dest) 1331 memcpy(&rule->dest_attr, dest, sizeof(*dest)); 1332 else 1333 rule->dest_attr.type = MLX5_FLOW_DESTINATION_TYPE_NONE; 1334 1335 return rule; 1336 } 1337 1338 static struct mlx5_flow_handle *alloc_handle(int num_rules) 1339 { 1340 struct mlx5_flow_handle *handle; 1341 1342 handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL); 1343 if (!handle) 1344 return NULL; 1345 1346 handle->num_rules = num_rules; 1347 1348 return handle; 1349 } 1350 1351 static void destroy_flow_handle(struct fs_fte *fte, 1352 struct mlx5_flow_handle *handle, 1353 struct mlx5_flow_destination *dest, 1354 int i) 1355 { 1356 for (; --i >= 0;) { 1357 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) { 1358 fte->dests_size--; 1359 list_del(&handle->rule[i]->node.list); 1360 kfree(handle->rule[i]); 1361 } 1362 } 1363 kfree(handle); 1364 } 1365 1366 static struct mlx5_flow_handle * 1367 create_flow_handle(struct fs_fte *fte, 1368 struct mlx5_flow_destination *dest, 1369 int dest_num, 1370 int *modify_mask, 1371 bool *new_rule) 1372 { 1373 struct mlx5_flow_handle *handle; 1374 struct mlx5_flow_rule *rule = NULL; 1375 static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS); 1376 static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST); 1377 int type; 1378 int i = 0; 1379 1380 handle = alloc_handle((dest_num) ? dest_num : 1); 1381 if (!handle) 1382 return ERR_PTR(-ENOMEM); 1383 1384 do { 1385 if (dest) { 1386 rule = find_flow_rule(fte, dest + i); 1387 if (rule) { 1388 refcount_inc(&rule->node.refcount); 1389 goto rule_found; 1390 } 1391 } 1392 1393 *new_rule = true; 1394 rule = alloc_rule(dest + i); 1395 if (!rule) 1396 goto free_rules; 1397 1398 /* Add dest to dests list- we need flow tables to be in the 1399 * end of the list for forward to next prio rules. 1400 */ 1401 tree_init_node(&rule->node, NULL, del_sw_hw_rule); 1402 if (dest && 1403 dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) 1404 list_add(&rule->node.list, &fte->node.children); 1405 else 1406 list_add_tail(&rule->node.list, &fte->node.children); 1407 if (dest) { 1408 fte->dests_size++; 1409 1410 if (is_fwd_dest_type(dest[i].type)) 1411 fte->fwd_dests++; 1412 1413 type = dest[i].type == 1414 MLX5_FLOW_DESTINATION_TYPE_COUNTER; 1415 *modify_mask |= type ? count : dst; 1416 } 1417 rule_found: 1418 handle->rule[i] = rule; 1419 } while (++i < dest_num); 1420 1421 return handle; 1422 1423 free_rules: 1424 destroy_flow_handle(fte, handle, dest, i); 1425 return ERR_PTR(-ENOMEM); 1426 } 1427 1428 /* fte should not be deleted while calling this function */ 1429 static struct mlx5_flow_handle * 1430 add_rule_fte(struct fs_fte *fte, 1431 struct mlx5_flow_group *fg, 1432 struct mlx5_flow_destination *dest, 1433 int dest_num, 1434 bool update_action) 1435 { 1436 struct mlx5_flow_root_namespace *root; 1437 struct mlx5_flow_handle *handle; 1438 struct mlx5_flow_table *ft; 1439 int modify_mask = 0; 1440 int err; 1441 bool new_rule = false; 1442 1443 handle = create_flow_handle(fte, dest, dest_num, &modify_mask, 1444 &new_rule); 1445 if (IS_ERR(handle) || !new_rule) 1446 goto out; 1447 1448 if (update_action) 1449 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION); 1450 1451 fs_get_obj(ft, fg->node.parent); 1452 root = find_root(&fg->node); 1453 if (!(fte->status & FS_FTE_STATUS_EXISTING)) 1454 err = root->cmds->create_fte(root, ft, fg, fte); 1455 else 1456 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte); 1457 if (err) 1458 goto free_handle; 1459 1460 fte->node.active = true; 1461 fte->status |= FS_FTE_STATUS_EXISTING; 1462 atomic_inc(&fg->node.version); 1463 1464 out: 1465 return handle; 1466 1467 free_handle: 1468 destroy_flow_handle(fte, handle, dest, handle->num_rules); 1469 return ERR_PTR(err); 1470 } 1471 1472 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table *ft, 1473 const struct mlx5_flow_spec *spec) 1474 { 1475 struct list_head *prev = &ft->node.children; 1476 u32 max_fte = ft->autogroup.max_fte; 1477 unsigned int candidate_index = 0; 1478 unsigned int group_size = 0; 1479 struct mlx5_flow_group *fg; 1480 1481 if (!ft->autogroup.active) 1482 return ERR_PTR(-ENOENT); 1483 1484 if (ft->autogroup.num_groups < ft->autogroup.required_groups) 1485 group_size = ft->autogroup.group_size; 1486 1487 /* max_fte == ft->autogroup.max_types */ 1488 if (group_size == 0) 1489 group_size = 1; 1490 1491 /* sorted by start_index */ 1492 fs_for_each_fg(fg, ft) { 1493 if (candidate_index + group_size > fg->start_index) 1494 candidate_index = fg->start_index + fg->max_ftes; 1495 else 1496 break; 1497 prev = &fg->node.list; 1498 } 1499 1500 if (candidate_index + group_size > max_fte) 1501 return ERR_PTR(-ENOSPC); 1502 1503 fg = alloc_insert_flow_group(ft, 1504 spec->match_criteria_enable, 1505 spec->match_criteria, 1506 candidate_index, 1507 candidate_index + group_size - 1, 1508 prev); 1509 if (IS_ERR(fg)) 1510 goto out; 1511 1512 if (group_size == ft->autogroup.group_size) 1513 ft->autogroup.num_groups++; 1514 1515 out: 1516 return fg; 1517 } 1518 1519 static int create_auto_flow_group(struct mlx5_flow_table *ft, 1520 struct mlx5_flow_group *fg) 1521 { 1522 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 1523 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in); 1524 void *match_criteria_addr; 1525 u8 src_esw_owner_mask_on; 1526 void *misc; 1527 int err; 1528 u32 *in; 1529 1530 in = kvzalloc(inlen, GFP_KERNEL); 1531 if (!in) 1532 return -ENOMEM; 1533 1534 MLX5_SET(create_flow_group_in, in, match_criteria_enable, 1535 fg->mask.match_criteria_enable); 1536 MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index); 1537 MLX5_SET(create_flow_group_in, in, end_flow_index, fg->start_index + 1538 fg->max_ftes - 1); 1539 1540 misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria, 1541 misc_parameters); 1542 src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc, 1543 source_eswitch_owner_vhca_id); 1544 MLX5_SET(create_flow_group_in, in, 1545 source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on); 1546 1547 match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in, 1548 in, match_criteria); 1549 memcpy(match_criteria_addr, fg->mask.match_criteria, 1550 sizeof(fg->mask.match_criteria)); 1551 1552 err = root->cmds->create_flow_group(root, ft, in, fg); 1553 if (!err) { 1554 fg->node.active = true; 1555 trace_mlx5_fs_add_fg(fg); 1556 } 1557 1558 kvfree(in); 1559 return err; 1560 } 1561 1562 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1, 1563 struct mlx5_flow_destination *d2) 1564 { 1565 if (d1->type == d2->type) { 1566 if (((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT || 1567 d1->type == MLX5_FLOW_DESTINATION_TYPE_UPLINK) && 1568 d1->vport.num == d2->vport.num && 1569 d1->vport.flags == d2->vport.flags && 1570 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ? 1571 (d1->vport.vhca_id == d2->vport.vhca_id) : true) && 1572 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ? 1573 (d1->vport.pkt_reformat->id == 1574 d2->vport.pkt_reformat->id) : true)) || 1575 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE && 1576 d1->ft == d2->ft) || 1577 (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR && 1578 d1->tir_num == d2->tir_num) || 1579 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM && 1580 d1->ft_num == d2->ft_num) || 1581 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER && 1582 d1->sampler_id == d2->sampler_id) || 1583 (d1->type == MLX5_FLOW_DESTINATION_TYPE_RANGE && 1584 d1->range.field == d2->range.field && 1585 d1->range.hit_ft == d2->range.hit_ft && 1586 d1->range.miss_ft == d2->range.miss_ft && 1587 d1->range.min == d2->range.min && 1588 d1->range.max == d2->range.max)) 1589 return true; 1590 } 1591 1592 return false; 1593 } 1594 1595 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte, 1596 struct mlx5_flow_destination *dest) 1597 { 1598 struct mlx5_flow_rule *rule; 1599 1600 list_for_each_entry(rule, &fte->node.children, node.list) { 1601 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest)) 1602 return rule; 1603 } 1604 return NULL; 1605 } 1606 1607 static bool check_conflicting_actions_vlan(const struct mlx5_fs_vlan *vlan0, 1608 const struct mlx5_fs_vlan *vlan1) 1609 { 1610 return vlan0->ethtype != vlan1->ethtype || 1611 vlan0->vid != vlan1->vid || 1612 vlan0->prio != vlan1->prio; 1613 } 1614 1615 static bool check_conflicting_actions(const struct mlx5_flow_act *act1, 1616 const struct mlx5_flow_act *act2) 1617 { 1618 u32 action1 = act1->action; 1619 u32 action2 = act2->action; 1620 u32 xored_actions; 1621 1622 xored_actions = action1 ^ action2; 1623 1624 /* if one rule only wants to count, it's ok */ 1625 if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT || 1626 action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT) 1627 return false; 1628 1629 if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP | 1630 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT | 1631 MLX5_FLOW_CONTEXT_ACTION_DECAP | 1632 MLX5_FLOW_CONTEXT_ACTION_MOD_HDR | 1633 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP | 1634 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH | 1635 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 | 1636 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2)) 1637 return true; 1638 1639 if (action1 & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT && 1640 act1->pkt_reformat != act2->pkt_reformat) 1641 return true; 1642 1643 if (action1 & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR && 1644 act1->modify_hdr != act2->modify_hdr) 1645 return true; 1646 1647 if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH && 1648 check_conflicting_actions_vlan(&act1->vlan[0], &act2->vlan[0])) 1649 return true; 1650 1651 if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2 && 1652 check_conflicting_actions_vlan(&act1->vlan[1], &act2->vlan[1])) 1653 return true; 1654 1655 return false; 1656 } 1657 1658 static int check_conflicting_ftes(struct fs_fte *fte, 1659 const struct mlx5_flow_context *flow_context, 1660 const struct mlx5_flow_act *flow_act) 1661 { 1662 if (check_conflicting_actions(flow_act, &fte->action)) { 1663 mlx5_core_warn(get_dev(&fte->node), 1664 "Found two FTEs with conflicting actions\n"); 1665 return -EEXIST; 1666 } 1667 1668 if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) && 1669 fte->flow_context.flow_tag != flow_context->flow_tag) { 1670 mlx5_core_warn(get_dev(&fte->node), 1671 "FTE flow tag %u already exists with different flow tag %u\n", 1672 fte->flow_context.flow_tag, 1673 flow_context->flow_tag); 1674 return -EEXIST; 1675 } 1676 1677 return 0; 1678 } 1679 1680 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg, 1681 const struct mlx5_flow_spec *spec, 1682 struct mlx5_flow_act *flow_act, 1683 struct mlx5_flow_destination *dest, 1684 int dest_num, 1685 struct fs_fte *fte) 1686 { 1687 struct mlx5_flow_handle *handle; 1688 int old_action; 1689 int i; 1690 int ret; 1691 1692 ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act); 1693 if (ret) 1694 return ERR_PTR(ret); 1695 1696 old_action = fte->action.action; 1697 fte->action.action |= flow_act->action; 1698 handle = add_rule_fte(fte, fg, dest, dest_num, 1699 old_action != flow_act->action); 1700 if (IS_ERR(handle)) { 1701 fte->action.action = old_action; 1702 return handle; 1703 } 1704 trace_mlx5_fs_set_fte(fte, false); 1705 1706 for (i = 0; i < handle->num_rules; i++) { 1707 if (refcount_read(&handle->rule[i]->node.refcount) == 1) { 1708 tree_add_node(&handle->rule[i]->node, &fte->node); 1709 trace_mlx5_fs_add_rule(handle->rule[i]); 1710 } 1711 } 1712 return handle; 1713 } 1714 1715 static bool counter_is_valid(u32 action) 1716 { 1717 return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP | 1718 MLX5_FLOW_CONTEXT_ACTION_ALLOW | 1719 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)); 1720 } 1721 1722 static bool dest_is_valid(struct mlx5_flow_destination *dest, 1723 struct mlx5_flow_act *flow_act, 1724 struct mlx5_flow_table *ft) 1725 { 1726 bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL; 1727 u32 action = flow_act->action; 1728 1729 if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)) 1730 return counter_is_valid(action); 1731 1732 if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)) 1733 return true; 1734 1735 if (ignore_level) { 1736 if (ft->type != FS_FT_FDB && 1737 ft->type != FS_FT_NIC_RX) 1738 return false; 1739 1740 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE && 1741 ft->type != dest->ft->type) 1742 return false; 1743 } 1744 1745 if (!dest || ((dest->type == 1746 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) && 1747 (dest->ft->level <= ft->level && !ignore_level))) 1748 return false; 1749 return true; 1750 } 1751 1752 struct match_list { 1753 struct list_head list; 1754 struct mlx5_flow_group *g; 1755 }; 1756 1757 static void free_match_list(struct match_list *head, bool ft_locked) 1758 { 1759 struct match_list *iter, *match_tmp; 1760 1761 list_for_each_entry_safe(iter, match_tmp, &head->list, 1762 list) { 1763 tree_put_node(&iter->g->node, ft_locked); 1764 list_del(&iter->list); 1765 kfree(iter); 1766 } 1767 } 1768 1769 static int build_match_list(struct match_list *match_head, 1770 struct mlx5_flow_table *ft, 1771 const struct mlx5_flow_spec *spec, 1772 struct mlx5_flow_group *fg, 1773 bool ft_locked) 1774 { 1775 struct rhlist_head *tmp, *list; 1776 struct mlx5_flow_group *g; 1777 int err = 0; 1778 1779 rcu_read_lock(); 1780 INIT_LIST_HEAD(&match_head->list); 1781 /* Collect all fgs which has a matching match_criteria */ 1782 list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg); 1783 /* RCU is atomic, we can't execute FW commands here */ 1784 rhl_for_each_entry_rcu(g, tmp, list, hash) { 1785 struct match_list *curr_match; 1786 1787 if (fg && fg != g) 1788 continue; 1789 1790 if (unlikely(!tree_get_node(&g->node))) 1791 continue; 1792 1793 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC); 1794 if (!curr_match) { 1795 rcu_read_unlock(); 1796 free_match_list(match_head, ft_locked); 1797 return -ENOMEM; 1798 } 1799 curr_match->g = g; 1800 list_add_tail(&curr_match->list, &match_head->list); 1801 } 1802 rcu_read_unlock(); 1803 return err; 1804 } 1805 1806 static u64 matched_fgs_get_version(struct list_head *match_head) 1807 { 1808 struct match_list *iter; 1809 u64 version = 0; 1810 1811 list_for_each_entry(iter, match_head, list) 1812 version += (u64)atomic_read(&iter->g->node.version); 1813 return version; 1814 } 1815 1816 static struct fs_fte * 1817 lookup_fte_locked(struct mlx5_flow_group *g, 1818 const u32 *match_value, 1819 bool take_write) 1820 { 1821 struct fs_fte *fte_tmp; 1822 1823 if (take_write) 1824 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT); 1825 else 1826 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT); 1827 fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value, 1828 rhash_fte); 1829 if (!fte_tmp || !tree_get_node(&fte_tmp->node)) { 1830 fte_tmp = NULL; 1831 goto out; 1832 } 1833 if (!fte_tmp->node.active) { 1834 tree_put_node(&fte_tmp->node, false); 1835 fte_tmp = NULL; 1836 goto out; 1837 } 1838 1839 nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD); 1840 out: 1841 if (take_write) 1842 up_write_ref_node(&g->node, false); 1843 else 1844 up_read_ref_node(&g->node); 1845 return fte_tmp; 1846 } 1847 1848 static struct mlx5_flow_handle * 1849 try_add_to_existing_fg(struct mlx5_flow_table *ft, 1850 struct list_head *match_head, 1851 const struct mlx5_flow_spec *spec, 1852 struct mlx5_flow_act *flow_act, 1853 struct mlx5_flow_destination *dest, 1854 int dest_num, 1855 int ft_version) 1856 { 1857 struct mlx5_flow_steering *steering = get_steering(&ft->node); 1858 struct mlx5_flow_group *g; 1859 struct mlx5_flow_handle *rule; 1860 struct match_list *iter; 1861 bool take_write = false; 1862 struct fs_fte *fte; 1863 u64 version = 0; 1864 int err; 1865 1866 fte = alloc_fte(ft, spec, flow_act); 1867 if (IS_ERR(fte)) 1868 return ERR_PTR(-ENOMEM); 1869 1870 search_again_locked: 1871 if (flow_act->flags & FLOW_ACT_NO_APPEND) 1872 goto skip_search; 1873 version = matched_fgs_get_version(match_head); 1874 /* Try to find an fte with identical match value and attempt update its 1875 * action. 1876 */ 1877 list_for_each_entry(iter, match_head, list) { 1878 struct fs_fte *fte_tmp; 1879 1880 g = iter->g; 1881 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write); 1882 if (!fte_tmp) 1883 continue; 1884 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp); 1885 /* No error check needed here, because insert_fte() is not called */ 1886 up_write_ref_node(&fte_tmp->node, false); 1887 tree_put_node(&fte_tmp->node, false); 1888 kmem_cache_free(steering->ftes_cache, fte); 1889 return rule; 1890 } 1891 1892 skip_search: 1893 /* No group with matching fte found, or we skipped the search. 1894 * Try to add a new fte to any matching fg. 1895 */ 1896 1897 /* Check the ft version, for case that new flow group 1898 * was added while the fgs weren't locked 1899 */ 1900 if (atomic_read(&ft->node.version) != ft_version) { 1901 rule = ERR_PTR(-EAGAIN); 1902 goto out; 1903 } 1904 1905 /* Check the fgs version. If version have changed it could be that an 1906 * FTE with the same match value was added while the fgs weren't 1907 * locked. 1908 */ 1909 if (!(flow_act->flags & FLOW_ACT_NO_APPEND) && 1910 version != matched_fgs_get_version(match_head)) { 1911 take_write = true; 1912 goto search_again_locked; 1913 } 1914 1915 list_for_each_entry(iter, match_head, list) { 1916 g = iter->g; 1917 1918 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT); 1919 1920 if (!g->node.active) { 1921 up_write_ref_node(&g->node, false); 1922 continue; 1923 } 1924 1925 err = insert_fte(g, fte); 1926 if (err) { 1927 up_write_ref_node(&g->node, false); 1928 if (err == -ENOSPC) 1929 continue; 1930 kmem_cache_free(steering->ftes_cache, fte); 1931 return ERR_PTR(err); 1932 } 1933 1934 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD); 1935 up_write_ref_node(&g->node, false); 1936 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte); 1937 up_write_ref_node(&fte->node, false); 1938 if (IS_ERR(rule)) 1939 tree_put_node(&fte->node, false); 1940 return rule; 1941 } 1942 rule = ERR_PTR(-ENOENT); 1943 out: 1944 kmem_cache_free(steering->ftes_cache, fte); 1945 return rule; 1946 } 1947 1948 static struct mlx5_flow_handle * 1949 _mlx5_add_flow_rules(struct mlx5_flow_table *ft, 1950 const struct mlx5_flow_spec *spec, 1951 struct mlx5_flow_act *flow_act, 1952 struct mlx5_flow_destination *dest, 1953 int dest_num) 1954 1955 { 1956 struct mlx5_flow_steering *steering = get_steering(&ft->node); 1957 struct mlx5_flow_handle *rule; 1958 struct match_list match_head; 1959 struct mlx5_flow_group *g; 1960 bool take_write = false; 1961 struct fs_fte *fte; 1962 int version; 1963 int err; 1964 int i; 1965 1966 if (!check_valid_spec(spec)) 1967 return ERR_PTR(-EINVAL); 1968 1969 if (flow_act->fg && ft->autogroup.active) 1970 return ERR_PTR(-EINVAL); 1971 1972 if (dest && dest_num <= 0) 1973 return ERR_PTR(-EINVAL); 1974 1975 for (i = 0; i < dest_num; i++) { 1976 if (!dest_is_valid(&dest[i], flow_act, ft)) 1977 return ERR_PTR(-EINVAL); 1978 } 1979 nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT); 1980 search_again_locked: 1981 version = atomic_read(&ft->node.version); 1982 1983 /* Collect all fgs which has a matching match_criteria */ 1984 err = build_match_list(&match_head, ft, spec, flow_act->fg, take_write); 1985 if (err) { 1986 if (take_write) 1987 up_write_ref_node(&ft->node, false); 1988 else 1989 up_read_ref_node(&ft->node); 1990 return ERR_PTR(err); 1991 } 1992 1993 if (!take_write) 1994 up_read_ref_node(&ft->node); 1995 1996 rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest, 1997 dest_num, version); 1998 free_match_list(&match_head, take_write); 1999 if (!IS_ERR(rule) || 2000 (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) { 2001 if (take_write) 2002 up_write_ref_node(&ft->node, false); 2003 return rule; 2004 } 2005 2006 if (!take_write) { 2007 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT); 2008 take_write = true; 2009 } 2010 2011 if (PTR_ERR(rule) == -EAGAIN || 2012 version != atomic_read(&ft->node.version)) 2013 goto search_again_locked; 2014 2015 g = alloc_auto_flow_group(ft, spec); 2016 if (IS_ERR(g)) { 2017 rule = ERR_CAST(g); 2018 up_write_ref_node(&ft->node, false); 2019 return rule; 2020 } 2021 2022 fte = alloc_fte(ft, spec, flow_act); 2023 if (IS_ERR(fte)) { 2024 up_write_ref_node(&ft->node, false); 2025 err = PTR_ERR(fte); 2026 goto err_alloc_fte; 2027 } 2028 2029 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT); 2030 up_write_ref_node(&ft->node, false); 2031 2032 err = create_auto_flow_group(ft, g); 2033 if (err) 2034 goto err_release_fg; 2035 2036 err = insert_fte(g, fte); 2037 if (err) 2038 goto err_release_fg; 2039 2040 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD); 2041 up_write_ref_node(&g->node, false); 2042 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte); 2043 up_write_ref_node(&fte->node, false); 2044 if (IS_ERR(rule)) 2045 tree_put_node(&fte->node, false); 2046 tree_put_node(&g->node, false); 2047 return rule; 2048 2049 err_release_fg: 2050 up_write_ref_node(&g->node, false); 2051 kmem_cache_free(steering->ftes_cache, fte); 2052 err_alloc_fte: 2053 tree_put_node(&g->node, false); 2054 return ERR_PTR(err); 2055 } 2056 2057 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft) 2058 { 2059 return ((ft->type == FS_FT_NIC_RX) && 2060 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs))); 2061 } 2062 2063 struct mlx5_flow_handle * 2064 mlx5_add_flow_rules(struct mlx5_flow_table *ft, 2065 const struct mlx5_flow_spec *spec, 2066 struct mlx5_flow_act *flow_act, 2067 struct mlx5_flow_destination *dest, 2068 int num_dest) 2069 { 2070 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 2071 static const struct mlx5_flow_spec zero_spec = {}; 2072 struct mlx5_flow_destination *gen_dest = NULL; 2073 struct mlx5_flow_table *next_ft = NULL; 2074 struct mlx5_flow_handle *handle = NULL; 2075 u32 sw_action = flow_act->action; 2076 int i; 2077 2078 if (!spec) 2079 spec = &zero_spec; 2080 2081 if (!is_fwd_next_action(sw_action)) 2082 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest); 2083 2084 if (!fwd_next_prio_supported(ft)) 2085 return ERR_PTR(-EOPNOTSUPP); 2086 2087 mutex_lock(&root->chain_lock); 2088 next_ft = find_next_fwd_ft(ft, flow_act); 2089 if (!next_ft) { 2090 handle = ERR_PTR(-EOPNOTSUPP); 2091 goto unlock; 2092 } 2093 2094 gen_dest = kcalloc(num_dest + 1, sizeof(*dest), 2095 GFP_KERNEL); 2096 if (!gen_dest) { 2097 handle = ERR_PTR(-ENOMEM); 2098 goto unlock; 2099 } 2100 for (i = 0; i < num_dest; i++) 2101 gen_dest[i] = dest[i]; 2102 gen_dest[i].type = 2103 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE; 2104 gen_dest[i].ft = next_ft; 2105 dest = gen_dest; 2106 num_dest++; 2107 flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO | 2108 MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS); 2109 flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST; 2110 handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest); 2111 if (IS_ERR(handle)) 2112 goto unlock; 2113 2114 if (list_empty(&handle->rule[num_dest - 1]->next_ft)) { 2115 mutex_lock(&next_ft->lock); 2116 list_add(&handle->rule[num_dest - 1]->next_ft, 2117 &next_ft->fwd_rules); 2118 mutex_unlock(&next_ft->lock); 2119 handle->rule[num_dest - 1]->sw_action = sw_action; 2120 handle->rule[num_dest - 1]->ft = ft; 2121 } 2122 unlock: 2123 mutex_unlock(&root->chain_lock); 2124 kfree(gen_dest); 2125 return handle; 2126 } 2127 EXPORT_SYMBOL(mlx5_add_flow_rules); 2128 2129 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle) 2130 { 2131 struct fs_fte *fte; 2132 int i; 2133 2134 /* In order to consolidate the HW changes we lock the FTE for other 2135 * changes, and increase its refcount, in order not to perform the 2136 * "del" functions of the FTE. Will handle them here. 2137 * The removal of the rules is done under locked FTE. 2138 * After removing all the handle's rules, if there are remaining 2139 * rules, it means we just need to modify the FTE in FW, and 2140 * unlock/decrease the refcount we increased before. 2141 * Otherwise, it means the FTE should be deleted. First delete the 2142 * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of 2143 * the FTE, which will handle the last decrease of the refcount, as 2144 * well as required handling of its parent. 2145 */ 2146 fs_get_obj(fte, handle->rule[0]->node.parent); 2147 down_write_ref_node(&fte->node, false); 2148 for (i = handle->num_rules - 1; i >= 0; i--) 2149 tree_remove_node(&handle->rule[i]->node, true); 2150 if (list_empty(&fte->node.children)) { 2151 fte->node.del_hw_func(&fte->node); 2152 /* Avoid double call to del_hw_fte */ 2153 fte->node.del_hw_func = NULL; 2154 up_write_ref_node(&fte->node, false); 2155 tree_put_node(&fte->node, false); 2156 } else if (fte->dests_size) { 2157 if (fte->modify_mask) 2158 modify_fte(fte); 2159 up_write_ref_node(&fte->node, false); 2160 } else { 2161 up_write_ref_node(&fte->node, false); 2162 } 2163 kfree(handle); 2164 } 2165 EXPORT_SYMBOL(mlx5_del_flow_rules); 2166 2167 /* Assuming prio->node.children(flow tables) is sorted by level */ 2168 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft) 2169 { 2170 struct fs_prio *prio; 2171 2172 fs_get_obj(prio, ft->node.parent); 2173 2174 if (!list_is_last(&ft->node.list, &prio->node.children)) 2175 return list_next_entry(ft, node.list); 2176 return find_next_chained_ft(prio); 2177 } 2178 2179 static int update_root_ft_destroy(struct mlx5_flow_table *ft) 2180 { 2181 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 2182 struct mlx5_ft_underlay_qp *uqp; 2183 struct mlx5_flow_table *new_root_ft = NULL; 2184 int err = 0; 2185 u32 qpn; 2186 2187 if (root->root_ft != ft) 2188 return 0; 2189 2190 new_root_ft = find_next_ft(ft); 2191 if (!new_root_ft) { 2192 root->root_ft = NULL; 2193 return 0; 2194 } 2195 2196 if (list_empty(&root->underlay_qpns)) { 2197 /* Don't set any QPN (zero) in case QPN list is empty */ 2198 qpn = 0; 2199 err = root->cmds->update_root_ft(root, new_root_ft, 2200 qpn, false); 2201 } else { 2202 list_for_each_entry(uqp, &root->underlay_qpns, list) { 2203 qpn = uqp->qpn; 2204 err = root->cmds->update_root_ft(root, 2205 new_root_ft, qpn, 2206 false); 2207 if (err) 2208 break; 2209 } 2210 } 2211 2212 if (err) 2213 mlx5_core_warn(root->dev, 2214 "Update root flow table of id(%u) qpn(%d) failed\n", 2215 ft->id, qpn); 2216 else 2217 root->root_ft = new_root_ft; 2218 2219 return 0; 2220 } 2221 2222 /* Connect flow table from previous priority to 2223 * the next flow table. 2224 */ 2225 static int disconnect_flow_table(struct mlx5_flow_table *ft) 2226 { 2227 struct mlx5_core_dev *dev = get_dev(&ft->node); 2228 struct mlx5_flow_table *next_ft; 2229 struct fs_prio *prio; 2230 int err = 0; 2231 2232 err = update_root_ft_destroy(ft); 2233 if (err) 2234 return err; 2235 2236 fs_get_obj(prio, ft->node.parent); 2237 if (!(list_first_entry(&prio->node.children, 2238 struct mlx5_flow_table, 2239 node.list) == ft)) 2240 return 0; 2241 2242 next_ft = find_next_ft(ft); 2243 err = connect_fwd_rules(dev, next_ft, ft); 2244 if (err) 2245 return err; 2246 2247 err = connect_prev_fts(dev, next_ft, prio); 2248 if (err) 2249 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n", 2250 ft->id); 2251 return err; 2252 } 2253 2254 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft) 2255 { 2256 struct mlx5_flow_root_namespace *root = find_root(&ft->node); 2257 int err = 0; 2258 2259 mutex_lock(&root->chain_lock); 2260 if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED)) 2261 err = disconnect_flow_table(ft); 2262 if (err) { 2263 mutex_unlock(&root->chain_lock); 2264 return err; 2265 } 2266 if (tree_remove_node(&ft->node, false)) 2267 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n", 2268 ft->id); 2269 mutex_unlock(&root->chain_lock); 2270 2271 return err; 2272 } 2273 EXPORT_SYMBOL(mlx5_destroy_flow_table); 2274 2275 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg) 2276 { 2277 if (tree_remove_node(&fg->node, false)) 2278 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n", 2279 fg->id); 2280 } 2281 EXPORT_SYMBOL(mlx5_destroy_flow_group); 2282 2283 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev, 2284 int n) 2285 { 2286 struct mlx5_flow_steering *steering = dev->priv.steering; 2287 2288 if (!steering || !steering->fdb_sub_ns) 2289 return NULL; 2290 2291 return steering->fdb_sub_ns[n]; 2292 } 2293 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns); 2294 2295 static bool is_nic_rx_ns(enum mlx5_flow_namespace_type type) 2296 { 2297 switch (type) { 2298 case MLX5_FLOW_NAMESPACE_BYPASS: 2299 case MLX5_FLOW_NAMESPACE_KERNEL_RX_MACSEC: 2300 case MLX5_FLOW_NAMESPACE_LAG: 2301 case MLX5_FLOW_NAMESPACE_OFFLOADS: 2302 case MLX5_FLOW_NAMESPACE_ETHTOOL: 2303 case MLX5_FLOW_NAMESPACE_KERNEL: 2304 case MLX5_FLOW_NAMESPACE_LEFTOVERS: 2305 case MLX5_FLOW_NAMESPACE_ANCHOR: 2306 return true; 2307 default: 2308 return false; 2309 } 2310 } 2311 2312 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev, 2313 enum mlx5_flow_namespace_type type) 2314 { 2315 struct mlx5_flow_steering *steering = dev->priv.steering; 2316 struct mlx5_flow_root_namespace *root_ns; 2317 int prio = 0; 2318 struct fs_prio *fs_prio; 2319 struct mlx5_flow_namespace *ns; 2320 2321 if (!steering) 2322 return NULL; 2323 2324 switch (type) { 2325 case MLX5_FLOW_NAMESPACE_FDB: 2326 if (steering->fdb_root_ns) 2327 return &steering->fdb_root_ns->ns; 2328 return NULL; 2329 case MLX5_FLOW_NAMESPACE_PORT_SEL: 2330 if (steering->port_sel_root_ns) 2331 return &steering->port_sel_root_ns->ns; 2332 return NULL; 2333 case MLX5_FLOW_NAMESPACE_SNIFFER_RX: 2334 if (steering->sniffer_rx_root_ns) 2335 return &steering->sniffer_rx_root_ns->ns; 2336 return NULL; 2337 case MLX5_FLOW_NAMESPACE_SNIFFER_TX: 2338 if (steering->sniffer_tx_root_ns) 2339 return &steering->sniffer_tx_root_ns->ns; 2340 return NULL; 2341 case MLX5_FLOW_NAMESPACE_FDB_BYPASS: 2342 root_ns = steering->fdb_root_ns; 2343 prio = FDB_BYPASS_PATH; 2344 break; 2345 case MLX5_FLOW_NAMESPACE_EGRESS: 2346 case MLX5_FLOW_NAMESPACE_EGRESS_IPSEC: 2347 case MLX5_FLOW_NAMESPACE_EGRESS_MACSEC: 2348 root_ns = steering->egress_root_ns; 2349 prio = type - MLX5_FLOW_NAMESPACE_EGRESS; 2350 break; 2351 case MLX5_FLOW_NAMESPACE_RDMA_RX: 2352 root_ns = steering->rdma_rx_root_ns; 2353 prio = RDMA_RX_BYPASS_PRIO; 2354 break; 2355 case MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL: 2356 root_ns = steering->rdma_rx_root_ns; 2357 prio = RDMA_RX_KERNEL_PRIO; 2358 break; 2359 case MLX5_FLOW_NAMESPACE_RDMA_TX: 2360 root_ns = steering->rdma_tx_root_ns; 2361 break; 2362 case MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS: 2363 root_ns = steering->rdma_rx_root_ns; 2364 prio = RDMA_RX_COUNTERS_PRIO; 2365 break; 2366 case MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS: 2367 root_ns = steering->rdma_tx_root_ns; 2368 prio = RDMA_TX_COUNTERS_PRIO; 2369 break; 2370 default: /* Must be NIC RX */ 2371 WARN_ON(!is_nic_rx_ns(type)); 2372 root_ns = steering->root_ns; 2373 prio = type; 2374 break; 2375 } 2376 2377 if (!root_ns) 2378 return NULL; 2379 2380 fs_prio = find_prio(&root_ns->ns, prio); 2381 if (!fs_prio) 2382 return NULL; 2383 2384 ns = list_first_entry(&fs_prio->node.children, 2385 typeof(*ns), 2386 node.list); 2387 2388 return ns; 2389 } 2390 EXPORT_SYMBOL(mlx5_get_flow_namespace); 2391 2392 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev, 2393 enum mlx5_flow_namespace_type type, 2394 int vport) 2395 { 2396 struct mlx5_flow_steering *steering = dev->priv.steering; 2397 2398 if (!steering) 2399 return NULL; 2400 2401 switch (type) { 2402 case MLX5_FLOW_NAMESPACE_ESW_EGRESS: 2403 if (vport >= steering->esw_egress_acl_vports) 2404 return NULL; 2405 if (steering->esw_egress_root_ns && 2406 steering->esw_egress_root_ns[vport]) 2407 return &steering->esw_egress_root_ns[vport]->ns; 2408 else 2409 return NULL; 2410 case MLX5_FLOW_NAMESPACE_ESW_INGRESS: 2411 if (vport >= steering->esw_ingress_acl_vports) 2412 return NULL; 2413 if (steering->esw_ingress_root_ns && 2414 steering->esw_ingress_root_ns[vport]) 2415 return &steering->esw_ingress_root_ns[vport]->ns; 2416 else 2417 return NULL; 2418 default: 2419 return NULL; 2420 } 2421 } 2422 2423 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns, 2424 unsigned int prio, 2425 int num_levels, 2426 enum fs_node_type type) 2427 { 2428 struct fs_prio *fs_prio; 2429 2430 fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL); 2431 if (!fs_prio) 2432 return ERR_PTR(-ENOMEM); 2433 2434 fs_prio->node.type = type; 2435 tree_init_node(&fs_prio->node, NULL, del_sw_prio); 2436 tree_add_node(&fs_prio->node, &ns->node); 2437 fs_prio->num_levels = num_levels; 2438 fs_prio->prio = prio; 2439 list_add_tail(&fs_prio->node.list, &ns->node.children); 2440 2441 return fs_prio; 2442 } 2443 2444 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns, 2445 unsigned int prio, 2446 int num_levels) 2447 { 2448 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS); 2449 } 2450 2451 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns, 2452 unsigned int prio, int num_levels) 2453 { 2454 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO); 2455 } 2456 2457 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace 2458 *ns) 2459 { 2460 ns->node.type = FS_TYPE_NAMESPACE; 2461 2462 return ns; 2463 } 2464 2465 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio, 2466 int def_miss_act) 2467 { 2468 struct mlx5_flow_namespace *ns; 2469 2470 ns = kzalloc(sizeof(*ns), GFP_KERNEL); 2471 if (!ns) 2472 return ERR_PTR(-ENOMEM); 2473 2474 fs_init_namespace(ns); 2475 ns->def_miss_action = def_miss_act; 2476 tree_init_node(&ns->node, NULL, del_sw_ns); 2477 tree_add_node(&ns->node, &prio->node); 2478 list_add_tail(&ns->node.list, &prio->node.children); 2479 2480 return ns; 2481 } 2482 2483 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio, 2484 struct init_tree_node *prio_metadata) 2485 { 2486 struct fs_prio *fs_prio; 2487 int i; 2488 2489 for (i = 0; i < prio_metadata->num_leaf_prios; i++) { 2490 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels); 2491 if (IS_ERR(fs_prio)) 2492 return PTR_ERR(fs_prio); 2493 } 2494 return 0; 2495 } 2496 2497 #define FLOW_TABLE_BIT_SZ 1 2498 #define GET_FLOW_TABLE_CAP(dev, offset) \ 2499 ((be32_to_cpu(*((__be32 *)(dev->caps.hca[MLX5_CAP_FLOW_TABLE]->cur) + \ 2500 offset / 32)) >> \ 2501 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ) 2502 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps) 2503 { 2504 int i; 2505 2506 for (i = 0; i < caps->arr_sz; i++) { 2507 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i])) 2508 return false; 2509 } 2510 return true; 2511 } 2512 2513 static int init_root_tree_recursive(struct mlx5_flow_steering *steering, 2514 struct init_tree_node *init_node, 2515 struct fs_node *fs_parent_node, 2516 struct init_tree_node *init_parent_node, 2517 int prio) 2518 { 2519 int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev, 2520 flow_table_properties_nic_receive. 2521 max_ft_level); 2522 struct mlx5_flow_namespace *fs_ns; 2523 struct fs_prio *fs_prio; 2524 struct fs_node *base; 2525 int i; 2526 int err; 2527 2528 if (init_node->type == FS_TYPE_PRIO) { 2529 if ((init_node->min_ft_level > max_ft_level) || 2530 !has_required_caps(steering->dev, &init_node->caps)) 2531 return 0; 2532 2533 fs_get_obj(fs_ns, fs_parent_node); 2534 if (init_node->num_leaf_prios) 2535 return create_leaf_prios(fs_ns, prio, init_node); 2536 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels); 2537 if (IS_ERR(fs_prio)) 2538 return PTR_ERR(fs_prio); 2539 base = &fs_prio->node; 2540 } else if (init_node->type == FS_TYPE_NAMESPACE) { 2541 fs_get_obj(fs_prio, fs_parent_node); 2542 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action); 2543 if (IS_ERR(fs_ns)) 2544 return PTR_ERR(fs_ns); 2545 base = &fs_ns->node; 2546 } else { 2547 return -EINVAL; 2548 } 2549 prio = 0; 2550 for (i = 0; i < init_node->ar_size; i++) { 2551 err = init_root_tree_recursive(steering, &init_node->children[i], 2552 base, init_node, prio); 2553 if (err) 2554 return err; 2555 if (init_node->children[i].type == FS_TYPE_PRIO && 2556 init_node->children[i].num_leaf_prios) { 2557 prio += init_node->children[i].num_leaf_prios; 2558 } 2559 } 2560 2561 return 0; 2562 } 2563 2564 static int init_root_tree(struct mlx5_flow_steering *steering, 2565 struct init_tree_node *init_node, 2566 struct fs_node *fs_parent_node) 2567 { 2568 int err; 2569 int i; 2570 2571 for (i = 0; i < init_node->ar_size; i++) { 2572 err = init_root_tree_recursive(steering, &init_node->children[i], 2573 fs_parent_node, 2574 init_node, i); 2575 if (err) 2576 return err; 2577 } 2578 return 0; 2579 } 2580 2581 static void del_sw_root_ns(struct fs_node *node) 2582 { 2583 struct mlx5_flow_root_namespace *root_ns; 2584 struct mlx5_flow_namespace *ns; 2585 2586 fs_get_obj(ns, node); 2587 root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns); 2588 mutex_destroy(&root_ns->chain_lock); 2589 kfree(node); 2590 } 2591 2592 static struct mlx5_flow_root_namespace 2593 *create_root_ns(struct mlx5_flow_steering *steering, 2594 enum fs_flow_table_type table_type) 2595 { 2596 const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type); 2597 struct mlx5_flow_root_namespace *root_ns; 2598 struct mlx5_flow_namespace *ns; 2599 2600 /* Create the root namespace */ 2601 root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL); 2602 if (!root_ns) 2603 return NULL; 2604 2605 root_ns->dev = steering->dev; 2606 root_ns->table_type = table_type; 2607 root_ns->cmds = cmds; 2608 2609 INIT_LIST_HEAD(&root_ns->underlay_qpns); 2610 2611 ns = &root_ns->ns; 2612 fs_init_namespace(ns); 2613 mutex_init(&root_ns->chain_lock); 2614 tree_init_node(&ns->node, NULL, del_sw_root_ns); 2615 tree_add_node(&ns->node, NULL); 2616 2617 return root_ns; 2618 } 2619 2620 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level); 2621 2622 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level) 2623 { 2624 struct fs_prio *prio; 2625 2626 fs_for_each_prio(prio, ns) { 2627 /* This updates prio start_level and num_levels */ 2628 set_prio_attrs_in_prio(prio, acc_level); 2629 acc_level += prio->num_levels; 2630 } 2631 return acc_level; 2632 } 2633 2634 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level) 2635 { 2636 struct mlx5_flow_namespace *ns; 2637 int acc_level_ns = acc_level; 2638 2639 prio->start_level = acc_level; 2640 fs_for_each_ns(ns, prio) { 2641 /* This updates start_level and num_levels of ns's priority descendants */ 2642 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level); 2643 2644 /* If this a prio with chains, and we can jump from one chain 2645 * (namespace) to another, so we accumulate the levels 2646 */ 2647 if (prio->node.type == FS_TYPE_PRIO_CHAINS) 2648 acc_level = acc_level_ns; 2649 } 2650 2651 if (!prio->num_levels) 2652 prio->num_levels = acc_level_ns - prio->start_level; 2653 WARN_ON(prio->num_levels < acc_level_ns - prio->start_level); 2654 } 2655 2656 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns) 2657 { 2658 struct mlx5_flow_namespace *ns = &root_ns->ns; 2659 struct fs_prio *prio; 2660 int start_level = 0; 2661 2662 fs_for_each_prio(prio, ns) { 2663 set_prio_attrs_in_prio(prio, start_level); 2664 start_level += prio->num_levels; 2665 } 2666 } 2667 2668 #define ANCHOR_PRIO 0 2669 #define ANCHOR_SIZE 1 2670 #define ANCHOR_LEVEL 0 2671 static int create_anchor_flow_table(struct mlx5_flow_steering *steering) 2672 { 2673 struct mlx5_flow_namespace *ns = NULL; 2674 struct mlx5_flow_table_attr ft_attr = {}; 2675 struct mlx5_flow_table *ft; 2676 2677 ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR); 2678 if (WARN_ON(!ns)) 2679 return -EINVAL; 2680 2681 ft_attr.max_fte = ANCHOR_SIZE; 2682 ft_attr.level = ANCHOR_LEVEL; 2683 ft_attr.prio = ANCHOR_PRIO; 2684 2685 ft = mlx5_create_flow_table(ns, &ft_attr); 2686 if (IS_ERR(ft)) { 2687 mlx5_core_err(steering->dev, "Failed to create last anchor flow table"); 2688 return PTR_ERR(ft); 2689 } 2690 return 0; 2691 } 2692 2693 static int init_root_ns(struct mlx5_flow_steering *steering) 2694 { 2695 int err; 2696 2697 steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX); 2698 if (!steering->root_ns) 2699 return -ENOMEM; 2700 2701 err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node); 2702 if (err) 2703 goto out_err; 2704 2705 set_prio_attrs(steering->root_ns); 2706 err = create_anchor_flow_table(steering); 2707 if (err) 2708 goto out_err; 2709 2710 return 0; 2711 2712 out_err: 2713 cleanup_root_ns(steering->root_ns); 2714 steering->root_ns = NULL; 2715 return err; 2716 } 2717 2718 static void clean_tree(struct fs_node *node) 2719 { 2720 if (node) { 2721 struct fs_node *iter; 2722 struct fs_node *temp; 2723 2724 tree_get_node(node); 2725 list_for_each_entry_safe(iter, temp, &node->children, list) 2726 clean_tree(iter); 2727 tree_put_node(node, false); 2728 tree_remove_node(node, false); 2729 } 2730 } 2731 2732 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns) 2733 { 2734 if (!root_ns) 2735 return; 2736 2737 clean_tree(&root_ns->ns.node); 2738 } 2739 2740 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering) 2741 { 2742 struct fs_prio *prio; 2743 2744 steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX); 2745 if (!steering->sniffer_tx_root_ns) 2746 return -ENOMEM; 2747 2748 /* Create single prio */ 2749 prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1); 2750 return PTR_ERR_OR_ZERO(prio); 2751 } 2752 2753 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering) 2754 { 2755 struct fs_prio *prio; 2756 2757 steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX); 2758 if (!steering->sniffer_rx_root_ns) 2759 return -ENOMEM; 2760 2761 /* Create single prio */ 2762 prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1); 2763 return PTR_ERR_OR_ZERO(prio); 2764 } 2765 2766 #define PORT_SEL_NUM_LEVELS 3 2767 static int init_port_sel_root_ns(struct mlx5_flow_steering *steering) 2768 { 2769 struct fs_prio *prio; 2770 2771 steering->port_sel_root_ns = create_root_ns(steering, FS_FT_PORT_SEL); 2772 if (!steering->port_sel_root_ns) 2773 return -ENOMEM; 2774 2775 /* Create single prio */ 2776 prio = fs_create_prio(&steering->port_sel_root_ns->ns, 0, 2777 PORT_SEL_NUM_LEVELS); 2778 return PTR_ERR_OR_ZERO(prio); 2779 } 2780 2781 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering) 2782 { 2783 int err; 2784 2785 steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX); 2786 if (!steering->rdma_rx_root_ns) 2787 return -ENOMEM; 2788 2789 err = init_root_tree(steering, &rdma_rx_root_fs, 2790 &steering->rdma_rx_root_ns->ns.node); 2791 if (err) 2792 goto out_err; 2793 2794 set_prio_attrs(steering->rdma_rx_root_ns); 2795 2796 return 0; 2797 2798 out_err: 2799 cleanup_root_ns(steering->rdma_rx_root_ns); 2800 steering->rdma_rx_root_ns = NULL; 2801 return err; 2802 } 2803 2804 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering) 2805 { 2806 int err; 2807 2808 steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX); 2809 if (!steering->rdma_tx_root_ns) 2810 return -ENOMEM; 2811 2812 err = init_root_tree(steering, &rdma_tx_root_fs, 2813 &steering->rdma_tx_root_ns->ns.node); 2814 if (err) 2815 goto out_err; 2816 2817 set_prio_attrs(steering->rdma_tx_root_ns); 2818 2819 return 0; 2820 2821 out_err: 2822 cleanup_root_ns(steering->rdma_tx_root_ns); 2823 steering->rdma_tx_root_ns = NULL; 2824 return err; 2825 } 2826 2827 /* FT and tc chains are stored in the same array so we can re-use the 2828 * mlx5_get_fdb_sub_ns() and tc api for FT chains. 2829 * When creating a new ns for each chain store it in the first available slot. 2830 * Assume tc chains are created and stored first and only then the FT chain. 2831 */ 2832 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering, 2833 struct mlx5_flow_namespace *ns) 2834 { 2835 int chain = 0; 2836 2837 while (steering->fdb_sub_ns[chain]) 2838 ++chain; 2839 2840 steering->fdb_sub_ns[chain] = ns; 2841 } 2842 2843 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering, 2844 struct fs_prio *maj_prio) 2845 { 2846 struct mlx5_flow_namespace *ns; 2847 struct fs_prio *min_prio; 2848 int prio; 2849 2850 ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF); 2851 if (IS_ERR(ns)) 2852 return PTR_ERR(ns); 2853 2854 for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) { 2855 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO); 2856 if (IS_ERR(min_prio)) 2857 return PTR_ERR(min_prio); 2858 } 2859 2860 store_fdb_sub_ns_prio_chain(steering, ns); 2861 2862 return 0; 2863 } 2864 2865 static int create_fdb_chains(struct mlx5_flow_steering *steering, 2866 int fs_prio, 2867 int chains) 2868 { 2869 struct fs_prio *maj_prio; 2870 int levels; 2871 int chain; 2872 int err; 2873 2874 levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains; 2875 maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns, 2876 fs_prio, 2877 levels); 2878 if (IS_ERR(maj_prio)) 2879 return PTR_ERR(maj_prio); 2880 2881 for (chain = 0; chain < chains; chain++) { 2882 err = create_fdb_sub_ns_prio_chain(steering, maj_prio); 2883 if (err) 2884 return err; 2885 } 2886 2887 return 0; 2888 } 2889 2890 static int create_fdb_fast_path(struct mlx5_flow_steering *steering) 2891 { 2892 int err; 2893 2894 steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS, 2895 sizeof(*steering->fdb_sub_ns), 2896 GFP_KERNEL); 2897 if (!steering->fdb_sub_ns) 2898 return -ENOMEM; 2899 2900 err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1); 2901 if (err) 2902 return err; 2903 2904 err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1); 2905 if (err) 2906 return err; 2907 2908 return 0; 2909 } 2910 2911 static int create_fdb_bypass(struct mlx5_flow_steering *steering) 2912 { 2913 struct mlx5_flow_namespace *ns; 2914 struct fs_prio *prio; 2915 int i; 2916 2917 prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH, 0); 2918 if (IS_ERR(prio)) 2919 return PTR_ERR(prio); 2920 2921 ns = fs_create_namespace(prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF); 2922 if (IS_ERR(ns)) 2923 return PTR_ERR(ns); 2924 2925 for (i = 0; i < MLX5_BY_PASS_NUM_REGULAR_PRIOS; i++) { 2926 prio = fs_create_prio(ns, i, 1); 2927 if (IS_ERR(prio)) 2928 return PTR_ERR(prio); 2929 } 2930 return 0; 2931 } 2932 2933 static void cleanup_fdb_root_ns(struct mlx5_flow_steering *steering) 2934 { 2935 cleanup_root_ns(steering->fdb_root_ns); 2936 steering->fdb_root_ns = NULL; 2937 kfree(steering->fdb_sub_ns); 2938 steering->fdb_sub_ns = NULL; 2939 } 2940 2941 static int init_fdb_root_ns(struct mlx5_flow_steering *steering) 2942 { 2943 struct fs_prio *maj_prio; 2944 int err; 2945 2946 steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB); 2947 if (!steering->fdb_root_ns) 2948 return -ENOMEM; 2949 2950 err = create_fdb_bypass(steering); 2951 if (err) 2952 goto out_err; 2953 2954 err = create_fdb_fast_path(steering); 2955 if (err) 2956 goto out_err; 2957 2958 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1); 2959 if (IS_ERR(maj_prio)) { 2960 err = PTR_ERR(maj_prio); 2961 goto out_err; 2962 } 2963 2964 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 3); 2965 if (IS_ERR(maj_prio)) { 2966 err = PTR_ERR(maj_prio); 2967 goto out_err; 2968 } 2969 2970 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1); 2971 if (IS_ERR(maj_prio)) { 2972 err = PTR_ERR(maj_prio); 2973 goto out_err; 2974 } 2975 2976 /* We put this priority last, knowing that nothing will get here 2977 * unless explicitly forwarded to. This is possible because the 2978 * slow path tables have catch all rules and nothing gets passed 2979 * those tables. 2980 */ 2981 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1); 2982 if (IS_ERR(maj_prio)) { 2983 err = PTR_ERR(maj_prio); 2984 goto out_err; 2985 } 2986 2987 set_prio_attrs(steering->fdb_root_ns); 2988 return 0; 2989 2990 out_err: 2991 cleanup_fdb_root_ns(steering); 2992 return err; 2993 } 2994 2995 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport) 2996 { 2997 struct fs_prio *prio; 2998 2999 steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL); 3000 if (!steering->esw_egress_root_ns[vport]) 3001 return -ENOMEM; 3002 3003 /* create 1 prio*/ 3004 prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1); 3005 return PTR_ERR_OR_ZERO(prio); 3006 } 3007 3008 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport) 3009 { 3010 struct fs_prio *prio; 3011 3012 steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL); 3013 if (!steering->esw_ingress_root_ns[vport]) 3014 return -ENOMEM; 3015 3016 /* create 1 prio*/ 3017 prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1); 3018 return PTR_ERR_OR_ZERO(prio); 3019 } 3020 3021 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports) 3022 { 3023 struct mlx5_flow_steering *steering = dev->priv.steering; 3024 int err; 3025 int i; 3026 3027 steering->esw_egress_root_ns = 3028 kcalloc(total_vports, 3029 sizeof(*steering->esw_egress_root_ns), 3030 GFP_KERNEL); 3031 if (!steering->esw_egress_root_ns) 3032 return -ENOMEM; 3033 3034 for (i = 0; i < total_vports; i++) { 3035 err = init_egress_acl_root_ns(steering, i); 3036 if (err) 3037 goto cleanup_root_ns; 3038 } 3039 steering->esw_egress_acl_vports = total_vports; 3040 return 0; 3041 3042 cleanup_root_ns: 3043 for (i--; i >= 0; i--) 3044 cleanup_root_ns(steering->esw_egress_root_ns[i]); 3045 kfree(steering->esw_egress_root_ns); 3046 steering->esw_egress_root_ns = NULL; 3047 return err; 3048 } 3049 3050 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev) 3051 { 3052 struct mlx5_flow_steering *steering = dev->priv.steering; 3053 int i; 3054 3055 if (!steering->esw_egress_root_ns) 3056 return; 3057 3058 for (i = 0; i < steering->esw_egress_acl_vports; i++) 3059 cleanup_root_ns(steering->esw_egress_root_ns[i]); 3060 3061 kfree(steering->esw_egress_root_ns); 3062 steering->esw_egress_root_ns = NULL; 3063 } 3064 3065 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports) 3066 { 3067 struct mlx5_flow_steering *steering = dev->priv.steering; 3068 int err; 3069 int i; 3070 3071 steering->esw_ingress_root_ns = 3072 kcalloc(total_vports, 3073 sizeof(*steering->esw_ingress_root_ns), 3074 GFP_KERNEL); 3075 if (!steering->esw_ingress_root_ns) 3076 return -ENOMEM; 3077 3078 for (i = 0; i < total_vports; i++) { 3079 err = init_ingress_acl_root_ns(steering, i); 3080 if (err) 3081 goto cleanup_root_ns; 3082 } 3083 steering->esw_ingress_acl_vports = total_vports; 3084 return 0; 3085 3086 cleanup_root_ns: 3087 for (i--; i >= 0; i--) 3088 cleanup_root_ns(steering->esw_ingress_root_ns[i]); 3089 kfree(steering->esw_ingress_root_ns); 3090 steering->esw_ingress_root_ns = NULL; 3091 return err; 3092 } 3093 3094 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev) 3095 { 3096 struct mlx5_flow_steering *steering = dev->priv.steering; 3097 int i; 3098 3099 if (!steering->esw_ingress_root_ns) 3100 return; 3101 3102 for (i = 0; i < steering->esw_ingress_acl_vports; i++) 3103 cleanup_root_ns(steering->esw_ingress_root_ns[i]); 3104 3105 kfree(steering->esw_ingress_root_ns); 3106 steering->esw_ingress_root_ns = NULL; 3107 } 3108 3109 u32 mlx5_fs_get_capabilities(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type type) 3110 { 3111 struct mlx5_flow_root_namespace *root; 3112 struct mlx5_flow_namespace *ns; 3113 3114 ns = mlx5_get_flow_namespace(dev, type); 3115 if (!ns) 3116 return 0; 3117 3118 root = find_root(&ns->node); 3119 if (!root) 3120 return 0; 3121 3122 return root->cmds->get_capabilities(root, root->table_type); 3123 } 3124 3125 static int init_egress_root_ns(struct mlx5_flow_steering *steering) 3126 { 3127 int err; 3128 3129 steering->egress_root_ns = create_root_ns(steering, 3130 FS_FT_NIC_TX); 3131 if (!steering->egress_root_ns) 3132 return -ENOMEM; 3133 3134 err = init_root_tree(steering, &egress_root_fs, 3135 &steering->egress_root_ns->ns.node); 3136 if (err) 3137 goto cleanup; 3138 set_prio_attrs(steering->egress_root_ns); 3139 return 0; 3140 cleanup: 3141 cleanup_root_ns(steering->egress_root_ns); 3142 steering->egress_root_ns = NULL; 3143 return err; 3144 } 3145 3146 void mlx5_fs_core_cleanup(struct mlx5_core_dev *dev) 3147 { 3148 struct mlx5_flow_steering *steering = dev->priv.steering; 3149 3150 cleanup_root_ns(steering->root_ns); 3151 cleanup_fdb_root_ns(steering); 3152 cleanup_root_ns(steering->port_sel_root_ns); 3153 cleanup_root_ns(steering->sniffer_rx_root_ns); 3154 cleanup_root_ns(steering->sniffer_tx_root_ns); 3155 cleanup_root_ns(steering->rdma_rx_root_ns); 3156 cleanup_root_ns(steering->rdma_tx_root_ns); 3157 cleanup_root_ns(steering->egress_root_ns); 3158 } 3159 3160 int mlx5_fs_core_init(struct mlx5_core_dev *dev) 3161 { 3162 struct mlx5_flow_steering *steering = dev->priv.steering; 3163 int err = 0; 3164 3165 if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) && 3166 (MLX5_CAP_GEN(dev, nic_flow_table))) || 3167 ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) && 3168 MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) && 3169 MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) { 3170 err = init_root_ns(steering); 3171 if (err) 3172 goto err; 3173 } 3174 3175 if (MLX5_ESWITCH_MANAGER(dev)) { 3176 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) { 3177 err = init_fdb_root_ns(steering); 3178 if (err) 3179 goto err; 3180 } 3181 } 3182 3183 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) { 3184 err = init_sniffer_rx_root_ns(steering); 3185 if (err) 3186 goto err; 3187 } 3188 3189 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) { 3190 err = init_sniffer_tx_root_ns(steering); 3191 if (err) 3192 goto err; 3193 } 3194 3195 if (MLX5_CAP_FLOWTABLE_PORT_SELECTION(dev, ft_support)) { 3196 err = init_port_sel_root_ns(steering); 3197 if (err) 3198 goto err; 3199 } 3200 3201 if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) && 3202 MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) { 3203 err = init_rdma_rx_root_ns(steering); 3204 if (err) 3205 goto err; 3206 } 3207 3208 if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) { 3209 err = init_rdma_tx_root_ns(steering); 3210 if (err) 3211 goto err; 3212 } 3213 3214 if (MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) { 3215 err = init_egress_root_ns(steering); 3216 if (err) 3217 goto err; 3218 } 3219 3220 return 0; 3221 3222 err: 3223 mlx5_fs_core_cleanup(dev); 3224 return err; 3225 } 3226 3227 void mlx5_fs_core_free(struct mlx5_core_dev *dev) 3228 { 3229 struct mlx5_flow_steering *steering = dev->priv.steering; 3230 3231 kmem_cache_destroy(steering->ftes_cache); 3232 kmem_cache_destroy(steering->fgs_cache); 3233 kfree(steering); 3234 mlx5_ft_pool_destroy(dev); 3235 mlx5_cleanup_fc_stats(dev); 3236 } 3237 3238 int mlx5_fs_core_alloc(struct mlx5_core_dev *dev) 3239 { 3240 struct mlx5_flow_steering *steering; 3241 int err = 0; 3242 3243 err = mlx5_init_fc_stats(dev); 3244 if (err) 3245 return err; 3246 3247 err = mlx5_ft_pool_init(dev); 3248 if (err) 3249 goto err; 3250 3251 steering = kzalloc(sizeof(*steering), GFP_KERNEL); 3252 if (!steering) { 3253 err = -ENOMEM; 3254 goto err; 3255 } 3256 3257 steering->dev = dev; 3258 dev->priv.steering = steering; 3259 3260 if (mlx5_fs_dr_is_supported(dev)) 3261 steering->mode = MLX5_FLOW_STEERING_MODE_SMFS; 3262 else 3263 steering->mode = MLX5_FLOW_STEERING_MODE_DMFS; 3264 3265 steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs", 3266 sizeof(struct mlx5_flow_group), 0, 3267 0, NULL); 3268 steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0, 3269 0, NULL); 3270 if (!steering->ftes_cache || !steering->fgs_cache) { 3271 err = -ENOMEM; 3272 goto err; 3273 } 3274 3275 return 0; 3276 3277 err: 3278 mlx5_fs_core_free(dev); 3279 return err; 3280 } 3281 3282 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn) 3283 { 3284 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns; 3285 struct mlx5_ft_underlay_qp *new_uqp; 3286 int err = 0; 3287 3288 new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL); 3289 if (!new_uqp) 3290 return -ENOMEM; 3291 3292 mutex_lock(&root->chain_lock); 3293 3294 if (!root->root_ft) { 3295 err = -EINVAL; 3296 goto update_ft_fail; 3297 } 3298 3299 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn, 3300 false); 3301 if (err) { 3302 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n", 3303 underlay_qpn, err); 3304 goto update_ft_fail; 3305 } 3306 3307 new_uqp->qpn = underlay_qpn; 3308 list_add_tail(&new_uqp->list, &root->underlay_qpns); 3309 3310 mutex_unlock(&root->chain_lock); 3311 3312 return 0; 3313 3314 update_ft_fail: 3315 mutex_unlock(&root->chain_lock); 3316 kfree(new_uqp); 3317 return err; 3318 } 3319 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn); 3320 3321 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn) 3322 { 3323 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns; 3324 struct mlx5_ft_underlay_qp *uqp; 3325 bool found = false; 3326 int err = 0; 3327 3328 mutex_lock(&root->chain_lock); 3329 list_for_each_entry(uqp, &root->underlay_qpns, list) { 3330 if (uqp->qpn == underlay_qpn) { 3331 found = true; 3332 break; 3333 } 3334 } 3335 3336 if (!found) { 3337 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n", 3338 underlay_qpn); 3339 err = -EINVAL; 3340 goto out; 3341 } 3342 3343 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn, 3344 true); 3345 if (err) 3346 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n", 3347 underlay_qpn, err); 3348 3349 list_del(&uqp->list); 3350 mutex_unlock(&root->chain_lock); 3351 kfree(uqp); 3352 3353 return 0; 3354 3355 out: 3356 mutex_unlock(&root->chain_lock); 3357 return err; 3358 } 3359 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn); 3360 3361 static struct mlx5_flow_root_namespace 3362 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type) 3363 { 3364 struct mlx5_flow_namespace *ns; 3365 3366 if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS || 3367 ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS) 3368 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0); 3369 else 3370 ns = mlx5_get_flow_namespace(dev, ns_type); 3371 if (!ns) 3372 return NULL; 3373 3374 return find_root(&ns->node); 3375 } 3376 3377 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev, 3378 u8 ns_type, u8 num_actions, 3379 void *modify_actions) 3380 { 3381 struct mlx5_flow_root_namespace *root; 3382 struct mlx5_modify_hdr *modify_hdr; 3383 int err; 3384 3385 root = get_root_namespace(dev, ns_type); 3386 if (!root) 3387 return ERR_PTR(-EOPNOTSUPP); 3388 3389 modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL); 3390 if (!modify_hdr) 3391 return ERR_PTR(-ENOMEM); 3392 3393 modify_hdr->ns_type = ns_type; 3394 err = root->cmds->modify_header_alloc(root, ns_type, num_actions, 3395 modify_actions, modify_hdr); 3396 if (err) { 3397 kfree(modify_hdr); 3398 return ERR_PTR(err); 3399 } 3400 3401 return modify_hdr; 3402 } 3403 EXPORT_SYMBOL(mlx5_modify_header_alloc); 3404 3405 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev, 3406 struct mlx5_modify_hdr *modify_hdr) 3407 { 3408 struct mlx5_flow_root_namespace *root; 3409 3410 root = get_root_namespace(dev, modify_hdr->ns_type); 3411 if (WARN_ON(!root)) 3412 return; 3413 root->cmds->modify_header_dealloc(root, modify_hdr); 3414 kfree(modify_hdr); 3415 } 3416 EXPORT_SYMBOL(mlx5_modify_header_dealloc); 3417 3418 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev, 3419 struct mlx5_pkt_reformat_params *params, 3420 enum mlx5_flow_namespace_type ns_type) 3421 { 3422 struct mlx5_pkt_reformat *pkt_reformat; 3423 struct mlx5_flow_root_namespace *root; 3424 int err; 3425 3426 root = get_root_namespace(dev, ns_type); 3427 if (!root) 3428 return ERR_PTR(-EOPNOTSUPP); 3429 3430 pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL); 3431 if (!pkt_reformat) 3432 return ERR_PTR(-ENOMEM); 3433 3434 pkt_reformat->ns_type = ns_type; 3435 pkt_reformat->reformat_type = params->type; 3436 err = root->cmds->packet_reformat_alloc(root, params, ns_type, 3437 pkt_reformat); 3438 if (err) { 3439 kfree(pkt_reformat); 3440 return ERR_PTR(err); 3441 } 3442 3443 return pkt_reformat; 3444 } 3445 EXPORT_SYMBOL(mlx5_packet_reformat_alloc); 3446 3447 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev, 3448 struct mlx5_pkt_reformat *pkt_reformat) 3449 { 3450 struct mlx5_flow_root_namespace *root; 3451 3452 root = get_root_namespace(dev, pkt_reformat->ns_type); 3453 if (WARN_ON(!root)) 3454 return; 3455 root->cmds->packet_reformat_dealloc(root, pkt_reformat); 3456 kfree(pkt_reformat); 3457 } 3458 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc); 3459 3460 int mlx5_get_match_definer_id(struct mlx5_flow_definer *definer) 3461 { 3462 return definer->id; 3463 } 3464 3465 struct mlx5_flow_definer * 3466 mlx5_create_match_definer(struct mlx5_core_dev *dev, 3467 enum mlx5_flow_namespace_type ns_type, u16 format_id, 3468 u32 *match_mask) 3469 { 3470 struct mlx5_flow_root_namespace *root; 3471 struct mlx5_flow_definer *definer; 3472 int id; 3473 3474 root = get_root_namespace(dev, ns_type); 3475 if (!root) 3476 return ERR_PTR(-EOPNOTSUPP); 3477 3478 definer = kzalloc(sizeof(*definer), GFP_KERNEL); 3479 if (!definer) 3480 return ERR_PTR(-ENOMEM); 3481 3482 definer->ns_type = ns_type; 3483 id = root->cmds->create_match_definer(root, format_id, match_mask); 3484 if (id < 0) { 3485 mlx5_core_warn(root->dev, "Failed to create match definer (%d)\n", id); 3486 kfree(definer); 3487 return ERR_PTR(id); 3488 } 3489 definer->id = id; 3490 return definer; 3491 } 3492 3493 void mlx5_destroy_match_definer(struct mlx5_core_dev *dev, 3494 struct mlx5_flow_definer *definer) 3495 { 3496 struct mlx5_flow_root_namespace *root; 3497 3498 root = get_root_namespace(dev, definer->ns_type); 3499 if (WARN_ON(!root)) 3500 return; 3501 3502 root->cmds->destroy_match_definer(root, definer->id); 3503 kfree(definer); 3504 } 3505 3506 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns, 3507 struct mlx5_flow_root_namespace *peer_ns) 3508 { 3509 if (peer_ns && ns->mode != peer_ns->mode) { 3510 mlx5_core_err(ns->dev, 3511 "Can't peer namespace of different steering mode\n"); 3512 return -EINVAL; 3513 } 3514 3515 return ns->cmds->set_peer(ns, peer_ns); 3516 } 3517 3518 /* This function should be called only at init stage of the namespace. 3519 * It is not safe to call this function while steering operations 3520 * are executed in the namespace. 3521 */ 3522 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns, 3523 enum mlx5_flow_steering_mode mode) 3524 { 3525 struct mlx5_flow_root_namespace *root; 3526 const struct mlx5_flow_cmds *cmds; 3527 int err; 3528 3529 root = find_root(&ns->node); 3530 if (&root->ns != ns) 3531 /* Can't set cmds to non root namespace */ 3532 return -EINVAL; 3533 3534 if (root->table_type != FS_FT_FDB) 3535 return -EOPNOTSUPP; 3536 3537 if (root->mode == mode) 3538 return 0; 3539 3540 if (mode == MLX5_FLOW_STEERING_MODE_SMFS) 3541 cmds = mlx5_fs_cmd_get_dr_cmds(); 3542 else 3543 cmds = mlx5_fs_cmd_get_fw_cmds(); 3544 if (!cmds) 3545 return -EOPNOTSUPP; 3546 3547 err = cmds->create_ns(root); 3548 if (err) { 3549 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n", 3550 err); 3551 return err; 3552 } 3553 3554 root->cmds->destroy_ns(root); 3555 root->cmds = cmds; 3556 root->mode = mode; 3557 3558 return 0; 3559 } 3560