1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright 2016 6WIND S.A. 3 * Copyright 2016 Mellanox Technologies, Ltd 4 */ 5 6 #include <netinet/in.h> 7 #include <sys/queue.h> 8 #include <stdalign.h> 9 #include <stdint.h> 10 #include <string.h> 11 12 /* Verbs header. */ 13 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */ 14 #ifdef PEDANTIC 15 #pragma GCC diagnostic ignored "-Wpedantic" 16 #endif 17 #include <infiniband/verbs.h> 18 #ifdef PEDANTIC 19 #pragma GCC diagnostic error "-Wpedantic" 20 #endif 21 22 #include <rte_common.h> 23 #include <rte_ether.h> 24 #include <rte_eth_ctrl.h> 25 #include <rte_ethdev_driver.h> 26 #include <rte_flow.h> 27 #include <rte_flow_driver.h> 28 #include <rte_malloc.h> 29 #include <rte_ip.h> 30 31 #include "mlx5.h" 32 #include "mlx5_defs.h" 33 #include "mlx5_prm.h" 34 #include "mlx5_glue.h" 35 #include "mlx5_flow.h" 36 37 /* Dev ops structure defined in mlx5.c */ 38 extern const struct eth_dev_ops mlx5_dev_ops; 39 extern const struct eth_dev_ops mlx5_dev_ops_isolate; 40 41 /** Device flow drivers. */ 42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT 43 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops; 44 #endif 45 extern const struct mlx5_flow_driver_ops mlx5_flow_tcf_drv_ops; 46 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops; 47 48 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops; 49 50 const struct mlx5_flow_driver_ops *flow_drv_ops[] = { 51 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops, 52 #ifdef HAVE_IBV_FLOW_DV_SUPPORT 53 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops, 54 #endif 55 [MLX5_FLOW_TYPE_TCF] = &mlx5_flow_tcf_drv_ops, 56 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops, 57 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops 58 }; 59 60 enum mlx5_expansion { 61 MLX5_EXPANSION_ROOT, 62 MLX5_EXPANSION_ROOT_OUTER, 63 MLX5_EXPANSION_ROOT_ETH_VLAN, 64 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN, 65 MLX5_EXPANSION_OUTER_ETH, 66 MLX5_EXPANSION_OUTER_ETH_VLAN, 67 MLX5_EXPANSION_OUTER_VLAN, 68 MLX5_EXPANSION_OUTER_IPV4, 69 MLX5_EXPANSION_OUTER_IPV4_UDP, 70 MLX5_EXPANSION_OUTER_IPV4_TCP, 71 MLX5_EXPANSION_OUTER_IPV6, 72 MLX5_EXPANSION_OUTER_IPV6_UDP, 73 MLX5_EXPANSION_OUTER_IPV6_TCP, 74 MLX5_EXPANSION_VXLAN, 75 MLX5_EXPANSION_VXLAN_GPE, 76 MLX5_EXPANSION_GRE, 77 MLX5_EXPANSION_MPLS, 78 MLX5_EXPANSION_ETH, 79 MLX5_EXPANSION_ETH_VLAN, 80 MLX5_EXPANSION_VLAN, 81 MLX5_EXPANSION_IPV4, 82 MLX5_EXPANSION_IPV4_UDP, 83 MLX5_EXPANSION_IPV4_TCP, 84 MLX5_EXPANSION_IPV6, 85 MLX5_EXPANSION_IPV6_UDP, 86 MLX5_EXPANSION_IPV6_TCP, 87 }; 88 89 /** Supported expansion of items. */ 90 static const struct rte_flow_expand_node mlx5_support_expansion[] = { 91 [MLX5_EXPANSION_ROOT] = { 92 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH, 93 MLX5_EXPANSION_IPV4, 94 MLX5_EXPANSION_IPV6), 95 .type = RTE_FLOW_ITEM_TYPE_END, 96 }, 97 [MLX5_EXPANSION_ROOT_OUTER] = { 98 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH, 99 MLX5_EXPANSION_OUTER_IPV4, 100 MLX5_EXPANSION_OUTER_IPV6), 101 .type = RTE_FLOW_ITEM_TYPE_END, 102 }, 103 [MLX5_EXPANSION_ROOT_ETH_VLAN] = { 104 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN), 105 .type = RTE_FLOW_ITEM_TYPE_END, 106 }, 107 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = { 108 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN), 109 .type = RTE_FLOW_ITEM_TYPE_END, 110 }, 111 [MLX5_EXPANSION_OUTER_ETH] = { 112 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4, 113 MLX5_EXPANSION_OUTER_IPV6, 114 MLX5_EXPANSION_MPLS), 115 .type = RTE_FLOW_ITEM_TYPE_ETH, 116 .rss_types = 0, 117 }, 118 [MLX5_EXPANSION_OUTER_ETH_VLAN] = { 119 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN), 120 .type = RTE_FLOW_ITEM_TYPE_ETH, 121 .rss_types = 0, 122 }, 123 [MLX5_EXPANSION_OUTER_VLAN] = { 124 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4, 125 MLX5_EXPANSION_OUTER_IPV6), 126 .type = RTE_FLOW_ITEM_TYPE_VLAN, 127 }, 128 [MLX5_EXPANSION_OUTER_IPV4] = { 129 .next = RTE_FLOW_EXPAND_RSS_NEXT 130 (MLX5_EXPANSION_OUTER_IPV4_UDP, 131 MLX5_EXPANSION_OUTER_IPV4_TCP, 132 MLX5_EXPANSION_GRE), 133 .type = RTE_FLOW_ITEM_TYPE_IPV4, 134 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 | 135 ETH_RSS_NONFRAG_IPV4_OTHER, 136 }, 137 [MLX5_EXPANSION_OUTER_IPV4_UDP] = { 138 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN, 139 MLX5_EXPANSION_VXLAN_GPE), 140 .type = RTE_FLOW_ITEM_TYPE_UDP, 141 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP, 142 }, 143 [MLX5_EXPANSION_OUTER_IPV4_TCP] = { 144 .type = RTE_FLOW_ITEM_TYPE_TCP, 145 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP, 146 }, 147 [MLX5_EXPANSION_OUTER_IPV6] = { 148 .next = RTE_FLOW_EXPAND_RSS_NEXT 149 (MLX5_EXPANSION_OUTER_IPV6_UDP, 150 MLX5_EXPANSION_OUTER_IPV6_TCP), 151 .type = RTE_FLOW_ITEM_TYPE_IPV6, 152 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 | 153 ETH_RSS_NONFRAG_IPV6_OTHER, 154 }, 155 [MLX5_EXPANSION_OUTER_IPV6_UDP] = { 156 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN, 157 MLX5_EXPANSION_VXLAN_GPE), 158 .type = RTE_FLOW_ITEM_TYPE_UDP, 159 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP, 160 }, 161 [MLX5_EXPANSION_OUTER_IPV6_TCP] = { 162 .type = RTE_FLOW_ITEM_TYPE_TCP, 163 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP, 164 }, 165 [MLX5_EXPANSION_VXLAN] = { 166 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH), 167 .type = RTE_FLOW_ITEM_TYPE_VXLAN, 168 }, 169 [MLX5_EXPANSION_VXLAN_GPE] = { 170 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH, 171 MLX5_EXPANSION_IPV4, 172 MLX5_EXPANSION_IPV6), 173 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE, 174 }, 175 [MLX5_EXPANSION_GRE] = { 176 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4), 177 .type = RTE_FLOW_ITEM_TYPE_GRE, 178 }, 179 [MLX5_EXPANSION_MPLS] = { 180 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4, 181 MLX5_EXPANSION_IPV6), 182 .type = RTE_FLOW_ITEM_TYPE_MPLS, 183 }, 184 [MLX5_EXPANSION_ETH] = { 185 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4, 186 MLX5_EXPANSION_IPV6), 187 .type = RTE_FLOW_ITEM_TYPE_ETH, 188 }, 189 [MLX5_EXPANSION_ETH_VLAN] = { 190 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN), 191 .type = RTE_FLOW_ITEM_TYPE_ETH, 192 }, 193 [MLX5_EXPANSION_VLAN] = { 194 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4, 195 MLX5_EXPANSION_IPV6), 196 .type = RTE_FLOW_ITEM_TYPE_VLAN, 197 }, 198 [MLX5_EXPANSION_IPV4] = { 199 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP, 200 MLX5_EXPANSION_IPV4_TCP), 201 .type = RTE_FLOW_ITEM_TYPE_IPV4, 202 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 | 203 ETH_RSS_NONFRAG_IPV4_OTHER, 204 }, 205 [MLX5_EXPANSION_IPV4_UDP] = { 206 .type = RTE_FLOW_ITEM_TYPE_UDP, 207 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP, 208 }, 209 [MLX5_EXPANSION_IPV4_TCP] = { 210 .type = RTE_FLOW_ITEM_TYPE_TCP, 211 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP, 212 }, 213 [MLX5_EXPANSION_IPV6] = { 214 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP, 215 MLX5_EXPANSION_IPV6_TCP), 216 .type = RTE_FLOW_ITEM_TYPE_IPV6, 217 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 | 218 ETH_RSS_NONFRAG_IPV6_OTHER, 219 }, 220 [MLX5_EXPANSION_IPV6_UDP] = { 221 .type = RTE_FLOW_ITEM_TYPE_UDP, 222 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP, 223 }, 224 [MLX5_EXPANSION_IPV6_TCP] = { 225 .type = RTE_FLOW_ITEM_TYPE_TCP, 226 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP, 227 }, 228 }; 229 230 static const struct rte_flow_ops mlx5_flow_ops = { 231 .validate = mlx5_flow_validate, 232 .create = mlx5_flow_create, 233 .destroy = mlx5_flow_destroy, 234 .flush = mlx5_flow_flush, 235 .isolate = mlx5_flow_isolate, 236 .query = mlx5_flow_query, 237 }; 238 239 /* Convert FDIR request to Generic flow. */ 240 struct mlx5_fdir { 241 struct rte_flow_attr attr; 242 struct rte_flow_item items[4]; 243 struct rte_flow_item_eth l2; 244 struct rte_flow_item_eth l2_mask; 245 union { 246 struct rte_flow_item_ipv4 ipv4; 247 struct rte_flow_item_ipv6 ipv6; 248 } l3; 249 union { 250 struct rte_flow_item_ipv4 ipv4; 251 struct rte_flow_item_ipv6 ipv6; 252 } l3_mask; 253 union { 254 struct rte_flow_item_udp udp; 255 struct rte_flow_item_tcp tcp; 256 } l4; 257 union { 258 struct rte_flow_item_udp udp; 259 struct rte_flow_item_tcp tcp; 260 } l4_mask; 261 struct rte_flow_action actions[2]; 262 struct rte_flow_action_queue queue; 263 }; 264 265 /* Map of Verbs to Flow priority with 8 Verbs priorities. */ 266 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = { 267 { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 }, 268 }; 269 270 /* Map of Verbs to Flow priority with 16 Verbs priorities. */ 271 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = { 272 { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 }, 273 { 9, 10, 11 }, { 12, 13, 14 }, 274 }; 275 276 /* Tunnel information. */ 277 struct mlx5_flow_tunnel_info { 278 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */ 279 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */ 280 }; 281 282 static struct mlx5_flow_tunnel_info tunnels_info[] = { 283 { 284 .tunnel = MLX5_FLOW_LAYER_VXLAN, 285 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP, 286 }, 287 { 288 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE, 289 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP, 290 }, 291 { 292 .tunnel = MLX5_FLOW_LAYER_GRE, 293 .ptype = RTE_PTYPE_TUNNEL_GRE, 294 }, 295 { 296 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP, 297 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP, 298 }, 299 { 300 .tunnel = MLX5_FLOW_LAYER_MPLS, 301 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE, 302 }, 303 }; 304 305 /** 306 * Discover the maximum number of priority available. 307 * 308 * @param[in] dev 309 * Pointer to the Ethernet device structure. 310 * 311 * @return 312 * number of supported flow priority on success, a negative errno 313 * value otherwise and rte_errno is set. 314 */ 315 int 316 mlx5_flow_discover_priorities(struct rte_eth_dev *dev) 317 { 318 struct { 319 struct ibv_flow_attr attr; 320 struct ibv_flow_spec_eth eth; 321 struct ibv_flow_spec_action_drop drop; 322 } flow_attr = { 323 .attr = { 324 .num_of_specs = 2, 325 }, 326 .eth = { 327 .type = IBV_FLOW_SPEC_ETH, 328 .size = sizeof(struct ibv_flow_spec_eth), 329 }, 330 .drop = { 331 .size = sizeof(struct ibv_flow_spec_action_drop), 332 .type = IBV_FLOW_SPEC_ACTION_DROP, 333 }, 334 }; 335 struct ibv_flow *flow; 336 struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev); 337 uint16_t vprio[] = { 8, 16 }; 338 int i; 339 int priority = 0; 340 341 if (!drop) { 342 rte_errno = ENOTSUP; 343 return -rte_errno; 344 } 345 for (i = 0; i != RTE_DIM(vprio); i++) { 346 flow_attr.attr.priority = vprio[i] - 1; 347 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr); 348 if (!flow) 349 break; 350 claim_zero(mlx5_glue->destroy_flow(flow)); 351 priority = vprio[i]; 352 } 353 switch (priority) { 354 case 8: 355 priority = RTE_DIM(priority_map_3); 356 break; 357 case 16: 358 priority = RTE_DIM(priority_map_5); 359 break; 360 default: 361 rte_errno = ENOTSUP; 362 DRV_LOG(ERR, 363 "port %u verbs maximum priority: %d expected 8/16", 364 dev->data->port_id, vprio[i]); 365 return -rte_errno; 366 } 367 mlx5_hrxq_drop_release(dev); 368 DRV_LOG(INFO, "port %u flow maximum priority: %d", 369 dev->data->port_id, priority); 370 return priority; 371 } 372 373 /** 374 * Adjust flow priority based on the highest layer and the request priority. 375 * 376 * @param[in] dev 377 * Pointer to the Ethernet device structure. 378 * @param[in] priority 379 * The rule base priority. 380 * @param[in] subpriority 381 * The priority based on the items. 382 * 383 * @return 384 * The new priority. 385 */ 386 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority, 387 uint32_t subpriority) 388 { 389 uint32_t res = 0; 390 struct priv *priv = dev->data->dev_private; 391 392 switch (priv->config.flow_prio) { 393 case RTE_DIM(priority_map_3): 394 res = priority_map_3[priority][subpriority]; 395 break; 396 case RTE_DIM(priority_map_5): 397 res = priority_map_5[priority][subpriority]; 398 break; 399 } 400 return res; 401 } 402 403 /** 404 * Verify the @p item specifications (spec, last, mask) are compatible with the 405 * NIC capabilities. 406 * 407 * @param[in] item 408 * Item specification. 409 * @param[in] mask 410 * @p item->mask or flow default bit-masks. 411 * @param[in] nic_mask 412 * Bit-masks covering supported fields by the NIC to compare with user mask. 413 * @param[in] size 414 * Bit-masks size in bytes. 415 * @param[out] error 416 * Pointer to error structure. 417 * 418 * @return 419 * 0 on success, a negative errno value otherwise and rte_errno is set. 420 */ 421 int 422 mlx5_flow_item_acceptable(const struct rte_flow_item *item, 423 const uint8_t *mask, 424 const uint8_t *nic_mask, 425 unsigned int size, 426 struct rte_flow_error *error) 427 { 428 unsigned int i; 429 430 assert(nic_mask); 431 for (i = 0; i < size; ++i) 432 if ((nic_mask[i] | mask[i]) != nic_mask[i]) 433 return rte_flow_error_set(error, ENOTSUP, 434 RTE_FLOW_ERROR_TYPE_ITEM, 435 item, 436 "mask enables non supported" 437 " bits"); 438 if (!item->spec && (item->mask || item->last)) 439 return rte_flow_error_set(error, EINVAL, 440 RTE_FLOW_ERROR_TYPE_ITEM, item, 441 "mask/last without a spec is not" 442 " supported"); 443 if (item->spec && item->last) { 444 uint8_t spec[size]; 445 uint8_t last[size]; 446 unsigned int i; 447 int ret; 448 449 for (i = 0; i < size; ++i) { 450 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i]; 451 last[i] = ((const uint8_t *)item->last)[i] & mask[i]; 452 } 453 ret = memcmp(spec, last, size); 454 if (ret != 0) 455 return rte_flow_error_set(error, EINVAL, 456 RTE_FLOW_ERROR_TYPE_ITEM, 457 item, 458 "range is not valid"); 459 } 460 return 0; 461 } 462 463 /** 464 * Adjust the hash fields according to the @p flow information. 465 * 466 * @param[in] dev_flow. 467 * Pointer to the mlx5_flow. 468 * @param[in] tunnel 469 * 1 when the hash field is for a tunnel item. 470 * @param[in] layer_types 471 * ETH_RSS_* types. 472 * @param[in] hash_fields 473 * Item hash fields. 474 * 475 * @return 476 * The hash fileds that should be used. 477 */ 478 uint64_t 479 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow, 480 int tunnel __rte_unused, uint64_t layer_types, 481 uint64_t hash_fields) 482 { 483 struct rte_flow *flow = dev_flow->flow; 484 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT 485 int rss_request_inner = flow->rss.level >= 2; 486 487 /* Check RSS hash level for tunnel. */ 488 if (tunnel && rss_request_inner) 489 hash_fields |= IBV_RX_HASH_INNER; 490 else if (tunnel || rss_request_inner) 491 return 0; 492 #endif 493 /* Check if requested layer matches RSS hash fields. */ 494 if (!(flow->rss.types & layer_types)) 495 return 0; 496 return hash_fields; 497 } 498 499 /** 500 * Lookup and set the ptype in the data Rx part. A single Ptype can be used, 501 * if several tunnel rules are used on this queue, the tunnel ptype will be 502 * cleared. 503 * 504 * @param rxq_ctrl 505 * Rx queue to update. 506 */ 507 static void 508 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl) 509 { 510 unsigned int i; 511 uint32_t tunnel_ptype = 0; 512 513 /* Look up for the ptype to use. */ 514 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) { 515 if (!rxq_ctrl->flow_tunnels_n[i]) 516 continue; 517 if (!tunnel_ptype) { 518 tunnel_ptype = tunnels_info[i].ptype; 519 } else { 520 tunnel_ptype = 0; 521 break; 522 } 523 } 524 rxq_ctrl->rxq.tunnel = tunnel_ptype; 525 } 526 527 /** 528 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive 529 * flow. 530 * 531 * @param[in] dev 532 * Pointer to the Ethernet device structure. 533 * @param[in] dev_flow 534 * Pointer to device flow structure. 535 */ 536 static void 537 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow) 538 { 539 struct priv *priv = dev->data->dev_private; 540 struct rte_flow *flow = dev_flow->flow; 541 const int mark = !!(flow->actions & 542 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK)); 543 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL); 544 unsigned int i; 545 546 for (i = 0; i != flow->rss.queue_num; ++i) { 547 int idx = (*flow->queue)[i]; 548 struct mlx5_rxq_ctrl *rxq_ctrl = 549 container_of((*priv->rxqs)[idx], 550 struct mlx5_rxq_ctrl, rxq); 551 552 if (mark) { 553 rxq_ctrl->rxq.mark = 1; 554 rxq_ctrl->flow_mark_n++; 555 } 556 if (tunnel) { 557 unsigned int j; 558 559 /* Increase the counter matching the flow. */ 560 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) { 561 if ((tunnels_info[j].tunnel & 562 dev_flow->layers) == 563 tunnels_info[j].tunnel) { 564 rxq_ctrl->flow_tunnels_n[j]++; 565 break; 566 } 567 } 568 flow_rxq_tunnel_ptype_update(rxq_ctrl); 569 } 570 } 571 } 572 573 /** 574 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow 575 * 576 * @param[in] dev 577 * Pointer to the Ethernet device structure. 578 * @param[in] flow 579 * Pointer to flow structure. 580 */ 581 static void 582 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow) 583 { 584 struct mlx5_flow *dev_flow; 585 586 LIST_FOREACH(dev_flow, &flow->dev_flows, next) 587 flow_drv_rxq_flags_set(dev, dev_flow); 588 } 589 590 /** 591 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the 592 * device flow if no other flow uses it with the same kind of request. 593 * 594 * @param dev 595 * Pointer to Ethernet device. 596 * @param[in] dev_flow 597 * Pointer to the device flow. 598 */ 599 static void 600 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow) 601 { 602 struct priv *priv = dev->data->dev_private; 603 struct rte_flow *flow = dev_flow->flow; 604 const int mark = !!(flow->actions & 605 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK)); 606 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL); 607 unsigned int i; 608 609 assert(dev->data->dev_started); 610 for (i = 0; i != flow->rss.queue_num; ++i) { 611 int idx = (*flow->queue)[i]; 612 struct mlx5_rxq_ctrl *rxq_ctrl = 613 container_of((*priv->rxqs)[idx], 614 struct mlx5_rxq_ctrl, rxq); 615 616 if (mark) { 617 rxq_ctrl->flow_mark_n--; 618 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n; 619 } 620 if (tunnel) { 621 unsigned int j; 622 623 /* Decrease the counter matching the flow. */ 624 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) { 625 if ((tunnels_info[j].tunnel & 626 dev_flow->layers) == 627 tunnels_info[j].tunnel) { 628 rxq_ctrl->flow_tunnels_n[j]--; 629 break; 630 } 631 } 632 flow_rxq_tunnel_ptype_update(rxq_ctrl); 633 } 634 } 635 } 636 637 /** 638 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the 639 * @p flow if no other flow uses it with the same kind of request. 640 * 641 * @param dev 642 * Pointer to Ethernet device. 643 * @param[in] flow 644 * Pointer to the flow. 645 */ 646 static void 647 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow) 648 { 649 struct mlx5_flow *dev_flow; 650 651 LIST_FOREACH(dev_flow, &flow->dev_flows, next) 652 flow_drv_rxq_flags_trim(dev, dev_flow); 653 } 654 655 /** 656 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues. 657 * 658 * @param dev 659 * Pointer to Ethernet device. 660 */ 661 static void 662 flow_rxq_flags_clear(struct rte_eth_dev *dev) 663 { 664 struct priv *priv = dev->data->dev_private; 665 unsigned int i; 666 667 for (i = 0; i != priv->rxqs_n; ++i) { 668 struct mlx5_rxq_ctrl *rxq_ctrl; 669 unsigned int j; 670 671 if (!(*priv->rxqs)[i]) 672 continue; 673 rxq_ctrl = container_of((*priv->rxqs)[i], 674 struct mlx5_rxq_ctrl, rxq); 675 rxq_ctrl->flow_mark_n = 0; 676 rxq_ctrl->rxq.mark = 0; 677 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) 678 rxq_ctrl->flow_tunnels_n[j] = 0; 679 rxq_ctrl->rxq.tunnel = 0; 680 } 681 } 682 683 /* 684 * Validate the flag action. 685 * 686 * @param[in] action_flags 687 * Bit-fields that holds the actions detected until now. 688 * @param[in] attr 689 * Attributes of flow that includes this action. 690 * @param[out] error 691 * Pointer to error structure. 692 * 693 * @return 694 * 0 on success, a negative errno value otherwise and rte_errno is set. 695 */ 696 int 697 mlx5_flow_validate_action_flag(uint64_t action_flags, 698 const struct rte_flow_attr *attr, 699 struct rte_flow_error *error) 700 { 701 702 if (action_flags & MLX5_FLOW_ACTION_DROP) 703 return rte_flow_error_set(error, EINVAL, 704 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 705 "can't drop and flag in same flow"); 706 if (action_flags & MLX5_FLOW_ACTION_MARK) 707 return rte_flow_error_set(error, EINVAL, 708 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 709 "can't mark and flag in same flow"); 710 if (action_flags & MLX5_FLOW_ACTION_FLAG) 711 return rte_flow_error_set(error, EINVAL, 712 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 713 "can't have 2 flag" 714 " actions in same flow"); 715 if (attr->egress) 716 return rte_flow_error_set(error, ENOTSUP, 717 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 718 "flag action not supported for " 719 "egress"); 720 return 0; 721 } 722 723 /* 724 * Validate the mark action. 725 * 726 * @param[in] action 727 * Pointer to the queue action. 728 * @param[in] action_flags 729 * Bit-fields that holds the actions detected until now. 730 * @param[in] attr 731 * Attributes of flow that includes this action. 732 * @param[out] error 733 * Pointer to error structure. 734 * 735 * @return 736 * 0 on success, a negative errno value otherwise and rte_errno is set. 737 */ 738 int 739 mlx5_flow_validate_action_mark(const struct rte_flow_action *action, 740 uint64_t action_flags, 741 const struct rte_flow_attr *attr, 742 struct rte_flow_error *error) 743 { 744 const struct rte_flow_action_mark *mark = action->conf; 745 746 if (!mark) 747 return rte_flow_error_set(error, EINVAL, 748 RTE_FLOW_ERROR_TYPE_ACTION, 749 action, 750 "configuration cannot be null"); 751 if (mark->id >= MLX5_FLOW_MARK_MAX) 752 return rte_flow_error_set(error, EINVAL, 753 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 754 &mark->id, 755 "mark id must in 0 <= id < " 756 RTE_STR(MLX5_FLOW_MARK_MAX)); 757 if (action_flags & MLX5_FLOW_ACTION_DROP) 758 return rte_flow_error_set(error, EINVAL, 759 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 760 "can't drop and mark in same flow"); 761 if (action_flags & MLX5_FLOW_ACTION_FLAG) 762 return rte_flow_error_set(error, EINVAL, 763 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 764 "can't flag and mark in same flow"); 765 if (action_flags & MLX5_FLOW_ACTION_MARK) 766 return rte_flow_error_set(error, EINVAL, 767 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 768 "can't have 2 mark actions in same" 769 " flow"); 770 if (attr->egress) 771 return rte_flow_error_set(error, ENOTSUP, 772 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 773 "mark action not supported for " 774 "egress"); 775 return 0; 776 } 777 778 /* 779 * Validate the drop action. 780 * 781 * @param[in] action_flags 782 * Bit-fields that holds the actions detected until now. 783 * @param[in] attr 784 * Attributes of flow that includes this action. 785 * @param[out] error 786 * Pointer to error structure. 787 * 788 * @return 789 * 0 on success, a negative errno value otherwise and rte_ernno is set. 790 */ 791 int 792 mlx5_flow_validate_action_drop(uint64_t action_flags, 793 const struct rte_flow_attr *attr, 794 struct rte_flow_error *error) 795 { 796 if (action_flags & MLX5_FLOW_ACTION_FLAG) 797 return rte_flow_error_set(error, EINVAL, 798 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 799 "can't drop and flag in same flow"); 800 if (action_flags & MLX5_FLOW_ACTION_MARK) 801 return rte_flow_error_set(error, EINVAL, 802 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 803 "can't drop and mark in same flow"); 804 if (action_flags & MLX5_FLOW_FATE_ACTIONS) 805 return rte_flow_error_set(error, EINVAL, 806 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 807 "can't have 2 fate actions in" 808 " same flow"); 809 if (attr->egress) 810 return rte_flow_error_set(error, ENOTSUP, 811 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 812 "drop action not supported for " 813 "egress"); 814 return 0; 815 } 816 817 /* 818 * Validate the queue action. 819 * 820 * @param[in] action 821 * Pointer to the queue action. 822 * @param[in] action_flags 823 * Bit-fields that holds the actions detected until now. 824 * @param[in] dev 825 * Pointer to the Ethernet device structure. 826 * @param[in] attr 827 * Attributes of flow that includes this action. 828 * @param[out] error 829 * Pointer to error structure. 830 * 831 * @return 832 * 0 on success, a negative errno value otherwise and rte_ernno is set. 833 */ 834 int 835 mlx5_flow_validate_action_queue(const struct rte_flow_action *action, 836 uint64_t action_flags, 837 struct rte_eth_dev *dev, 838 const struct rte_flow_attr *attr, 839 struct rte_flow_error *error) 840 { 841 struct priv *priv = dev->data->dev_private; 842 const struct rte_flow_action_queue *queue = action->conf; 843 844 if (action_flags & MLX5_FLOW_FATE_ACTIONS) 845 return rte_flow_error_set(error, EINVAL, 846 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 847 "can't have 2 fate actions in" 848 " same flow"); 849 if (queue->index >= priv->rxqs_n) 850 return rte_flow_error_set(error, EINVAL, 851 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 852 &queue->index, 853 "queue index out of range"); 854 if (!(*priv->rxqs)[queue->index]) 855 return rte_flow_error_set(error, EINVAL, 856 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 857 &queue->index, 858 "queue is not configured"); 859 if (attr->egress) 860 return rte_flow_error_set(error, ENOTSUP, 861 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 862 "queue action not supported for " 863 "egress"); 864 return 0; 865 } 866 867 /* 868 * Validate the rss action. 869 * 870 * @param[in] action 871 * Pointer to the queue action. 872 * @param[in] action_flags 873 * Bit-fields that holds the actions detected until now. 874 * @param[in] dev 875 * Pointer to the Ethernet device structure. 876 * @param[in] attr 877 * Attributes of flow that includes this action. 878 * @param[out] error 879 * Pointer to error structure. 880 * 881 * @return 882 * 0 on success, a negative errno value otherwise and rte_ernno is set. 883 */ 884 int 885 mlx5_flow_validate_action_rss(const struct rte_flow_action *action, 886 uint64_t action_flags, 887 struct rte_eth_dev *dev, 888 const struct rte_flow_attr *attr, 889 struct rte_flow_error *error) 890 { 891 struct priv *priv = dev->data->dev_private; 892 const struct rte_flow_action_rss *rss = action->conf; 893 unsigned int i; 894 895 if (action_flags & MLX5_FLOW_FATE_ACTIONS) 896 return rte_flow_error_set(error, EINVAL, 897 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 898 "can't have 2 fate actions" 899 " in same flow"); 900 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT && 901 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ) 902 return rte_flow_error_set(error, ENOTSUP, 903 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 904 &rss->func, 905 "RSS hash function not supported"); 906 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT 907 if (rss->level > 2) 908 #else 909 if (rss->level > 1) 910 #endif 911 return rte_flow_error_set(error, ENOTSUP, 912 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 913 &rss->level, 914 "tunnel RSS is not supported"); 915 /* allow RSS key_len 0 in case of NULL (default) RSS key. */ 916 if (rss->key_len == 0 && rss->key != NULL) 917 return rte_flow_error_set(error, ENOTSUP, 918 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 919 &rss->key_len, 920 "RSS hash key length 0"); 921 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN) 922 return rte_flow_error_set(error, ENOTSUP, 923 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 924 &rss->key_len, 925 "RSS hash key too small"); 926 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN) 927 return rte_flow_error_set(error, ENOTSUP, 928 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 929 &rss->key_len, 930 "RSS hash key too large"); 931 if (rss->queue_num > priv->config.ind_table_max_size) 932 return rte_flow_error_set(error, ENOTSUP, 933 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 934 &rss->queue_num, 935 "number of queues too large"); 936 if (rss->types & MLX5_RSS_HF_MASK) 937 return rte_flow_error_set(error, ENOTSUP, 938 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 939 &rss->types, 940 "some RSS protocols are not" 941 " supported"); 942 for (i = 0; i != rss->queue_num; ++i) { 943 if (!(*priv->rxqs)[rss->queue[i]]) 944 return rte_flow_error_set 945 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF, 946 &rss->queue[i], "queue is not configured"); 947 } 948 if (attr->egress) 949 return rte_flow_error_set(error, ENOTSUP, 950 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 951 "rss action not supported for " 952 "egress"); 953 return 0; 954 } 955 956 /* 957 * Validate the count action. 958 * 959 * @param[in] dev 960 * Pointer to the Ethernet device structure. 961 * @param[in] attr 962 * Attributes of flow that includes this action. 963 * @param[out] error 964 * Pointer to error structure. 965 * 966 * @return 967 * 0 on success, a negative errno value otherwise and rte_ernno is set. 968 */ 969 int 970 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused, 971 const struct rte_flow_attr *attr, 972 struct rte_flow_error *error) 973 { 974 if (attr->egress) 975 return rte_flow_error_set(error, ENOTSUP, 976 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 977 "count action not supported for " 978 "egress"); 979 return 0; 980 } 981 982 /** 983 * Verify the @p attributes will be correctly understood by the NIC and store 984 * them in the @p flow if everything is correct. 985 * 986 * @param[in] dev 987 * Pointer to the Ethernet device structure. 988 * @param[in] attributes 989 * Pointer to flow attributes 990 * @param[out] error 991 * Pointer to error structure. 992 * 993 * @return 994 * 0 on success, a negative errno value otherwise and rte_errno is set. 995 */ 996 int 997 mlx5_flow_validate_attributes(struct rte_eth_dev *dev, 998 const struct rte_flow_attr *attributes, 999 struct rte_flow_error *error) 1000 { 1001 struct priv *priv = dev->data->dev_private; 1002 uint32_t priority_max = priv->config.flow_prio - 1; 1003 1004 if (attributes->group) 1005 return rte_flow_error_set(error, ENOTSUP, 1006 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, 1007 NULL, "groups is not supported"); 1008 if (attributes->priority != MLX5_FLOW_PRIO_RSVD && 1009 attributes->priority >= priority_max) 1010 return rte_flow_error_set(error, ENOTSUP, 1011 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY, 1012 NULL, "priority out of range"); 1013 if (attributes->egress) 1014 return rte_flow_error_set(error, ENOTSUP, 1015 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 1016 "egress is not supported"); 1017 if (attributes->transfer) 1018 return rte_flow_error_set(error, ENOTSUP, 1019 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER, 1020 NULL, "transfer is not supported"); 1021 if (!attributes->ingress) 1022 return rte_flow_error_set(error, EINVAL, 1023 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS, 1024 NULL, 1025 "ingress attribute is mandatory"); 1026 return 0; 1027 } 1028 1029 /** 1030 * Validate Ethernet item. 1031 * 1032 * @param[in] item 1033 * Item specification. 1034 * @param[in] item_flags 1035 * Bit-fields that holds the items detected until now. 1036 * @param[out] error 1037 * Pointer to error structure. 1038 * 1039 * @return 1040 * 0 on success, a negative errno value otherwise and rte_errno is set. 1041 */ 1042 int 1043 mlx5_flow_validate_item_eth(const struct rte_flow_item *item, 1044 uint64_t item_flags, 1045 struct rte_flow_error *error) 1046 { 1047 const struct rte_flow_item_eth *mask = item->mask; 1048 const struct rte_flow_item_eth nic_mask = { 1049 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff", 1050 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff", 1051 .type = RTE_BE16(0xffff), 1052 }; 1053 int ret; 1054 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1055 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 : 1056 MLX5_FLOW_LAYER_OUTER_L2; 1057 1058 if (item_flags & ethm) 1059 return rte_flow_error_set(error, ENOTSUP, 1060 RTE_FLOW_ERROR_TYPE_ITEM, item, 1061 "multiple L2 layers not supported"); 1062 if (!mask) 1063 mask = &rte_flow_item_eth_mask; 1064 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, 1065 (const uint8_t *)&nic_mask, 1066 sizeof(struct rte_flow_item_eth), 1067 error); 1068 return ret; 1069 } 1070 1071 /** 1072 * Validate VLAN item. 1073 * 1074 * @param[in] item 1075 * Item specification. 1076 * @param[in] item_flags 1077 * Bit-fields that holds the items detected until now. 1078 * @param[out] error 1079 * Pointer to error structure. 1080 * 1081 * @return 1082 * 0 on success, a negative errno value otherwise and rte_errno is set. 1083 */ 1084 int 1085 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item, 1086 uint64_t item_flags, 1087 struct rte_flow_error *error) 1088 { 1089 const struct rte_flow_item_vlan *spec = item->spec; 1090 const struct rte_flow_item_vlan *mask = item->mask; 1091 const struct rte_flow_item_vlan nic_mask = { 1092 .tci = RTE_BE16(0x0fff), 1093 .inner_type = RTE_BE16(0xffff), 1094 }; 1095 uint16_t vlan_tag = 0; 1096 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1097 int ret; 1098 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 | 1099 MLX5_FLOW_LAYER_INNER_L4) : 1100 (MLX5_FLOW_LAYER_OUTER_L3 | 1101 MLX5_FLOW_LAYER_OUTER_L4); 1102 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN : 1103 MLX5_FLOW_LAYER_OUTER_VLAN; 1104 1105 if (item_flags & vlanm) 1106 return rte_flow_error_set(error, EINVAL, 1107 RTE_FLOW_ERROR_TYPE_ITEM, item, 1108 "multiple VLAN layers not supported"); 1109 else if ((item_flags & l34m) != 0) 1110 return rte_flow_error_set(error, EINVAL, 1111 RTE_FLOW_ERROR_TYPE_ITEM, item, 1112 "L2 layer cannot follow L3/L4 layer"); 1113 if (!mask) 1114 mask = &rte_flow_item_vlan_mask; 1115 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, 1116 (const uint8_t *)&nic_mask, 1117 sizeof(struct rte_flow_item_vlan), 1118 error); 1119 if (ret) 1120 return ret; 1121 if (spec) { 1122 vlan_tag = spec->tci; 1123 vlan_tag &= mask->tci; 1124 } 1125 /* 1126 * From verbs perspective an empty VLAN is equivalent 1127 * to a packet without VLAN layer. 1128 */ 1129 if (!vlan_tag) 1130 return rte_flow_error_set(error, EINVAL, 1131 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, 1132 item->spec, 1133 "VLAN cannot be empty"); 1134 return 0; 1135 } 1136 1137 /** 1138 * Validate IPV4 item. 1139 * 1140 * @param[in] item 1141 * Item specification. 1142 * @param[in] item_flags 1143 * Bit-fields that holds the items detected until now. 1144 * @param[out] error 1145 * Pointer to error structure. 1146 * 1147 * @return 1148 * 0 on success, a negative errno value otherwise and rte_errno is set. 1149 */ 1150 int 1151 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item, 1152 uint64_t item_flags, 1153 struct rte_flow_error *error) 1154 { 1155 const struct rte_flow_item_ipv4 *mask = item->mask; 1156 const struct rte_flow_item_ipv4 nic_mask = { 1157 .hdr = { 1158 .src_addr = RTE_BE32(0xffffffff), 1159 .dst_addr = RTE_BE32(0xffffffff), 1160 .type_of_service = 0xff, 1161 .next_proto_id = 0xff, 1162 }, 1163 }; 1164 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1165 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 : 1166 MLX5_FLOW_LAYER_OUTER_L3; 1167 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 : 1168 MLX5_FLOW_LAYER_OUTER_L4; 1169 int ret; 1170 1171 if (item_flags & l3m) 1172 return rte_flow_error_set(error, ENOTSUP, 1173 RTE_FLOW_ERROR_TYPE_ITEM, item, 1174 "multiple L3 layers not supported"); 1175 else if (item_flags & l4m) 1176 return rte_flow_error_set(error, EINVAL, 1177 RTE_FLOW_ERROR_TYPE_ITEM, item, 1178 "L3 cannot follow an L4 layer."); 1179 if (!mask) 1180 mask = &rte_flow_item_ipv4_mask; 1181 else if (mask->hdr.next_proto_id != 0 && 1182 mask->hdr.next_proto_id != 0xff) 1183 return rte_flow_error_set(error, EINVAL, 1184 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask, 1185 "partial mask is not supported" 1186 " for protocol"); 1187 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, 1188 (const uint8_t *)&nic_mask, 1189 sizeof(struct rte_flow_item_ipv4), 1190 error); 1191 if (ret < 0) 1192 return ret; 1193 return 0; 1194 } 1195 1196 /** 1197 * Validate IPV6 item. 1198 * 1199 * @param[in] item 1200 * Item specification. 1201 * @param[in] item_flags 1202 * Bit-fields that holds the items detected until now. 1203 * @param[out] error 1204 * Pointer to error structure. 1205 * 1206 * @return 1207 * 0 on success, a negative errno value otherwise and rte_errno is set. 1208 */ 1209 int 1210 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item, 1211 uint64_t item_flags, 1212 struct rte_flow_error *error) 1213 { 1214 const struct rte_flow_item_ipv6 *mask = item->mask; 1215 const struct rte_flow_item_ipv6 nic_mask = { 1216 .hdr = { 1217 .src_addr = 1218 "\xff\xff\xff\xff\xff\xff\xff\xff" 1219 "\xff\xff\xff\xff\xff\xff\xff\xff", 1220 .dst_addr = 1221 "\xff\xff\xff\xff\xff\xff\xff\xff" 1222 "\xff\xff\xff\xff\xff\xff\xff\xff", 1223 .vtc_flow = RTE_BE32(0xffffffff), 1224 .proto = 0xff, 1225 .hop_limits = 0xff, 1226 }, 1227 }; 1228 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1229 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 : 1230 MLX5_FLOW_LAYER_OUTER_L3; 1231 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 : 1232 MLX5_FLOW_LAYER_OUTER_L4; 1233 int ret; 1234 1235 if (item_flags & l3m) 1236 return rte_flow_error_set(error, ENOTSUP, 1237 RTE_FLOW_ERROR_TYPE_ITEM, item, 1238 "multiple L3 layers not supported"); 1239 else if (item_flags & l4m) 1240 return rte_flow_error_set(error, EINVAL, 1241 RTE_FLOW_ERROR_TYPE_ITEM, item, 1242 "L3 cannot follow an L4 layer."); 1243 if (!mask) 1244 mask = &rte_flow_item_ipv6_mask; 1245 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, 1246 (const uint8_t *)&nic_mask, 1247 sizeof(struct rte_flow_item_ipv6), 1248 error); 1249 if (ret < 0) 1250 return ret; 1251 return 0; 1252 } 1253 1254 /** 1255 * Validate UDP item. 1256 * 1257 * @param[in] item 1258 * Item specification. 1259 * @param[in] item_flags 1260 * Bit-fields that holds the items detected until now. 1261 * @param[in] target_protocol 1262 * The next protocol in the previous item. 1263 * @param[in] flow_mask 1264 * mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask. 1265 * @param[out] error 1266 * Pointer to error structure. 1267 * 1268 * @return 1269 * 0 on success, a negative errno value otherwise and rte_errno is set. 1270 */ 1271 int 1272 mlx5_flow_validate_item_udp(const struct rte_flow_item *item, 1273 uint64_t item_flags, 1274 uint8_t target_protocol, 1275 struct rte_flow_error *error) 1276 { 1277 const struct rte_flow_item_udp *mask = item->mask; 1278 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1279 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 : 1280 MLX5_FLOW_LAYER_OUTER_L3; 1281 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 : 1282 MLX5_FLOW_LAYER_OUTER_L4; 1283 int ret; 1284 1285 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP) 1286 return rte_flow_error_set(error, EINVAL, 1287 RTE_FLOW_ERROR_TYPE_ITEM, item, 1288 "protocol filtering not compatible" 1289 " with UDP layer"); 1290 if (!(item_flags & l3m)) 1291 return rte_flow_error_set(error, EINVAL, 1292 RTE_FLOW_ERROR_TYPE_ITEM, item, 1293 "L3 is mandatory to filter on L4"); 1294 if (item_flags & l4m) 1295 return rte_flow_error_set(error, EINVAL, 1296 RTE_FLOW_ERROR_TYPE_ITEM, item, 1297 "multiple L4 layers not supported"); 1298 if (!mask) 1299 mask = &rte_flow_item_udp_mask; 1300 ret = mlx5_flow_item_acceptable 1301 (item, (const uint8_t *)mask, 1302 (const uint8_t *)&rte_flow_item_udp_mask, 1303 sizeof(struct rte_flow_item_udp), error); 1304 if (ret < 0) 1305 return ret; 1306 return 0; 1307 } 1308 1309 /** 1310 * Validate TCP item. 1311 * 1312 * @param[in] item 1313 * Item specification. 1314 * @param[in] item_flags 1315 * Bit-fields that holds the items detected until now. 1316 * @param[in] target_protocol 1317 * The next protocol in the previous item. 1318 * @param[out] error 1319 * Pointer to error structure. 1320 * 1321 * @return 1322 * 0 on success, a negative errno value otherwise and rte_errno is set. 1323 */ 1324 int 1325 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item, 1326 uint64_t item_flags, 1327 uint8_t target_protocol, 1328 const struct rte_flow_item_tcp *flow_mask, 1329 struct rte_flow_error *error) 1330 { 1331 const struct rte_flow_item_tcp *mask = item->mask; 1332 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1333 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 : 1334 MLX5_FLOW_LAYER_OUTER_L3; 1335 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 : 1336 MLX5_FLOW_LAYER_OUTER_L4; 1337 int ret; 1338 1339 assert(flow_mask); 1340 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP) 1341 return rte_flow_error_set(error, EINVAL, 1342 RTE_FLOW_ERROR_TYPE_ITEM, item, 1343 "protocol filtering not compatible" 1344 " with TCP layer"); 1345 if (!(item_flags & l3m)) 1346 return rte_flow_error_set(error, EINVAL, 1347 RTE_FLOW_ERROR_TYPE_ITEM, item, 1348 "L3 is mandatory to filter on L4"); 1349 if (item_flags & l4m) 1350 return rte_flow_error_set(error, EINVAL, 1351 RTE_FLOW_ERROR_TYPE_ITEM, item, 1352 "multiple L4 layers not supported"); 1353 if (!mask) 1354 mask = &rte_flow_item_tcp_mask; 1355 ret = mlx5_flow_item_acceptable 1356 (item, (const uint8_t *)mask, 1357 (const uint8_t *)flow_mask, 1358 sizeof(struct rte_flow_item_tcp), error); 1359 if (ret < 0) 1360 return ret; 1361 return 0; 1362 } 1363 1364 /** 1365 * Validate VXLAN item. 1366 * 1367 * @param[in] item 1368 * Item specification. 1369 * @param[in] item_flags 1370 * Bit-fields that holds the items detected until now. 1371 * @param[in] target_protocol 1372 * The next protocol in the previous item. 1373 * @param[out] error 1374 * Pointer to error structure. 1375 * 1376 * @return 1377 * 0 on success, a negative errno value otherwise and rte_errno is set. 1378 */ 1379 int 1380 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item, 1381 uint64_t item_flags, 1382 struct rte_flow_error *error) 1383 { 1384 const struct rte_flow_item_vxlan *spec = item->spec; 1385 const struct rte_flow_item_vxlan *mask = item->mask; 1386 int ret; 1387 union vni { 1388 uint32_t vlan_id; 1389 uint8_t vni[4]; 1390 } id = { .vlan_id = 0, }; 1391 uint32_t vlan_id = 0; 1392 1393 1394 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) 1395 return rte_flow_error_set(error, ENOTSUP, 1396 RTE_FLOW_ERROR_TYPE_ITEM, item, 1397 "multiple tunnel layers not" 1398 " supported"); 1399 /* 1400 * Verify only UDPv4 is present as defined in 1401 * https://tools.ietf.org/html/rfc7348 1402 */ 1403 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)) 1404 return rte_flow_error_set(error, EINVAL, 1405 RTE_FLOW_ERROR_TYPE_ITEM, item, 1406 "no outer UDP layer found"); 1407 if (!mask) 1408 mask = &rte_flow_item_vxlan_mask; 1409 ret = mlx5_flow_item_acceptable 1410 (item, (const uint8_t *)mask, 1411 (const uint8_t *)&rte_flow_item_vxlan_mask, 1412 sizeof(struct rte_flow_item_vxlan), 1413 error); 1414 if (ret < 0) 1415 return ret; 1416 if (spec) { 1417 memcpy(&id.vni[1], spec->vni, 3); 1418 vlan_id = id.vlan_id; 1419 memcpy(&id.vni[1], mask->vni, 3); 1420 vlan_id &= id.vlan_id; 1421 } 1422 /* 1423 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if 1424 * only this layer is defined in the Verbs specification it is 1425 * interpreted as wildcard and all packets will match this 1426 * rule, if it follows a full stack layer (ex: eth / ipv4 / 1427 * udp), all packets matching the layers before will also 1428 * match this rule. To avoid such situation, VNI 0 is 1429 * currently refused. 1430 */ 1431 if (!vlan_id) 1432 return rte_flow_error_set(error, ENOTSUP, 1433 RTE_FLOW_ERROR_TYPE_ITEM, item, 1434 "VXLAN vni cannot be 0"); 1435 if (!(item_flags & MLX5_FLOW_LAYER_OUTER)) 1436 return rte_flow_error_set(error, ENOTSUP, 1437 RTE_FLOW_ERROR_TYPE_ITEM, item, 1438 "VXLAN tunnel must be fully defined"); 1439 return 0; 1440 } 1441 1442 /** 1443 * Validate VXLAN_GPE item. 1444 * 1445 * @param[in] item 1446 * Item specification. 1447 * @param[in] item_flags 1448 * Bit-fields that holds the items detected until now. 1449 * @param[in] priv 1450 * Pointer to the private data structure. 1451 * @param[in] target_protocol 1452 * The next protocol in the previous item. 1453 * @param[out] error 1454 * Pointer to error structure. 1455 * 1456 * @return 1457 * 0 on success, a negative errno value otherwise and rte_errno is set. 1458 */ 1459 int 1460 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item, 1461 uint64_t item_flags, 1462 struct rte_eth_dev *dev, 1463 struct rte_flow_error *error) 1464 { 1465 struct priv *priv = dev->data->dev_private; 1466 const struct rte_flow_item_vxlan_gpe *spec = item->spec; 1467 const struct rte_flow_item_vxlan_gpe *mask = item->mask; 1468 int ret; 1469 union vni { 1470 uint32_t vlan_id; 1471 uint8_t vni[4]; 1472 } id = { .vlan_id = 0, }; 1473 uint32_t vlan_id = 0; 1474 1475 if (!priv->config.l3_vxlan_en) 1476 return rte_flow_error_set(error, ENOTSUP, 1477 RTE_FLOW_ERROR_TYPE_ITEM, item, 1478 "L3 VXLAN is not enabled by device" 1479 " parameter and/or not configured in" 1480 " firmware"); 1481 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) 1482 return rte_flow_error_set(error, ENOTSUP, 1483 RTE_FLOW_ERROR_TYPE_ITEM, item, 1484 "multiple tunnel layers not" 1485 " supported"); 1486 /* 1487 * Verify only UDPv4 is present as defined in 1488 * https://tools.ietf.org/html/rfc7348 1489 */ 1490 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)) 1491 return rte_flow_error_set(error, EINVAL, 1492 RTE_FLOW_ERROR_TYPE_ITEM, item, 1493 "no outer UDP layer found"); 1494 if (!mask) 1495 mask = &rte_flow_item_vxlan_gpe_mask; 1496 ret = mlx5_flow_item_acceptable 1497 (item, (const uint8_t *)mask, 1498 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask, 1499 sizeof(struct rte_flow_item_vxlan_gpe), 1500 error); 1501 if (ret < 0) 1502 return ret; 1503 if (spec) { 1504 if (spec->protocol) 1505 return rte_flow_error_set(error, ENOTSUP, 1506 RTE_FLOW_ERROR_TYPE_ITEM, 1507 item, 1508 "VxLAN-GPE protocol" 1509 " not supported"); 1510 memcpy(&id.vni[1], spec->vni, 3); 1511 vlan_id = id.vlan_id; 1512 memcpy(&id.vni[1], mask->vni, 3); 1513 vlan_id &= id.vlan_id; 1514 } 1515 /* 1516 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this 1517 * layer is defined in the Verbs specification it is interpreted as 1518 * wildcard and all packets will match this rule, if it follows a full 1519 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers 1520 * before will also match this rule. To avoid such situation, VNI 0 1521 * is currently refused. 1522 */ 1523 if (!vlan_id) 1524 return rte_flow_error_set(error, ENOTSUP, 1525 RTE_FLOW_ERROR_TYPE_ITEM, item, 1526 "VXLAN-GPE vni cannot be 0"); 1527 if (!(item_flags & MLX5_FLOW_LAYER_OUTER)) 1528 return rte_flow_error_set(error, ENOTSUP, 1529 RTE_FLOW_ERROR_TYPE_ITEM, item, 1530 "VXLAN-GPE tunnel must be fully" 1531 " defined"); 1532 return 0; 1533 } 1534 1535 /** 1536 * Validate GRE item. 1537 * 1538 * @param[in] item 1539 * Item specification. 1540 * @param[in] item_flags 1541 * Bit flags to mark detected items. 1542 * @param[in] target_protocol 1543 * The next protocol in the previous item. 1544 * @param[out] error 1545 * Pointer to error structure. 1546 * 1547 * @return 1548 * 0 on success, a negative errno value otherwise and rte_errno is set. 1549 */ 1550 int 1551 mlx5_flow_validate_item_gre(const struct rte_flow_item *item, 1552 uint64_t item_flags, 1553 uint8_t target_protocol, 1554 struct rte_flow_error *error) 1555 { 1556 const struct rte_flow_item_gre *spec __rte_unused = item->spec; 1557 const struct rte_flow_item_gre *mask = item->mask; 1558 int ret; 1559 1560 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE) 1561 return rte_flow_error_set(error, EINVAL, 1562 RTE_FLOW_ERROR_TYPE_ITEM, item, 1563 "protocol filtering not compatible" 1564 " with this GRE layer"); 1565 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) 1566 return rte_flow_error_set(error, ENOTSUP, 1567 RTE_FLOW_ERROR_TYPE_ITEM, item, 1568 "multiple tunnel layers not" 1569 " supported"); 1570 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3)) 1571 return rte_flow_error_set(error, ENOTSUP, 1572 RTE_FLOW_ERROR_TYPE_ITEM, item, 1573 "L3 Layer is missing"); 1574 if (!mask) 1575 mask = &rte_flow_item_gre_mask; 1576 ret = mlx5_flow_item_acceptable 1577 (item, (const uint8_t *)mask, 1578 (const uint8_t *)&rte_flow_item_gre_mask, 1579 sizeof(struct rte_flow_item_gre), error); 1580 if (ret < 0) 1581 return ret; 1582 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT 1583 if (spec && (spec->protocol & mask->protocol)) 1584 return rte_flow_error_set(error, ENOTSUP, 1585 RTE_FLOW_ERROR_TYPE_ITEM, item, 1586 "without MPLS support the" 1587 " specification cannot be used for" 1588 " filtering"); 1589 #endif 1590 return 0; 1591 } 1592 1593 /** 1594 * Validate MPLS item. 1595 * 1596 * @param[in] dev 1597 * Pointer to the rte_eth_dev structure. 1598 * @param[in] item 1599 * Item specification. 1600 * @param[in] item_flags 1601 * Bit-fields that holds the items detected until now. 1602 * @param[in] prev_layer 1603 * The protocol layer indicated in previous item. 1604 * @param[out] error 1605 * Pointer to error structure. 1606 * 1607 * @return 1608 * 0 on success, a negative errno value otherwise and rte_errno is set. 1609 */ 1610 int 1611 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused, 1612 const struct rte_flow_item *item __rte_unused, 1613 uint64_t item_flags __rte_unused, 1614 uint64_t prev_layer __rte_unused, 1615 struct rte_flow_error *error) 1616 { 1617 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT 1618 const struct rte_flow_item_mpls *mask = item->mask; 1619 struct priv *priv = dev->data->dev_private; 1620 int ret; 1621 1622 if (!priv->config.mpls_en) 1623 return rte_flow_error_set(error, ENOTSUP, 1624 RTE_FLOW_ERROR_TYPE_ITEM, item, 1625 "MPLS not supported or" 1626 " disabled in firmware" 1627 " configuration."); 1628 /* MPLS over IP, UDP, GRE is allowed */ 1629 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 | 1630 MLX5_FLOW_LAYER_OUTER_L4_UDP | 1631 MLX5_FLOW_LAYER_GRE))) 1632 return rte_flow_error_set(error, EINVAL, 1633 RTE_FLOW_ERROR_TYPE_ITEM, item, 1634 "protocol filtering not compatible" 1635 " with MPLS layer"); 1636 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */ 1637 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) && 1638 !(item_flags & MLX5_FLOW_LAYER_GRE)) 1639 return rte_flow_error_set(error, ENOTSUP, 1640 RTE_FLOW_ERROR_TYPE_ITEM, item, 1641 "multiple tunnel layers not" 1642 " supported"); 1643 if (!mask) 1644 mask = &rte_flow_item_mpls_mask; 1645 ret = mlx5_flow_item_acceptable 1646 (item, (const uint8_t *)mask, 1647 (const uint8_t *)&rte_flow_item_mpls_mask, 1648 sizeof(struct rte_flow_item_mpls), error); 1649 if (ret < 0) 1650 return ret; 1651 return 0; 1652 #endif 1653 return rte_flow_error_set(error, ENOTSUP, 1654 RTE_FLOW_ERROR_TYPE_ITEM, item, 1655 "MPLS is not supported by Verbs, please" 1656 " update."); 1657 } 1658 1659 static int 1660 flow_null_validate(struct rte_eth_dev *dev __rte_unused, 1661 const struct rte_flow_attr *attr __rte_unused, 1662 const struct rte_flow_item items[] __rte_unused, 1663 const struct rte_flow_action actions[] __rte_unused, 1664 struct rte_flow_error *error __rte_unused) 1665 { 1666 rte_errno = ENOTSUP; 1667 return -rte_errno; 1668 } 1669 1670 static struct mlx5_flow * 1671 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused, 1672 const struct rte_flow_item items[] __rte_unused, 1673 const struct rte_flow_action actions[] __rte_unused, 1674 struct rte_flow_error *error __rte_unused) 1675 { 1676 rte_errno = ENOTSUP; 1677 return NULL; 1678 } 1679 1680 static int 1681 flow_null_translate(struct rte_eth_dev *dev __rte_unused, 1682 struct mlx5_flow *dev_flow __rte_unused, 1683 const struct rte_flow_attr *attr __rte_unused, 1684 const struct rte_flow_item items[] __rte_unused, 1685 const struct rte_flow_action actions[] __rte_unused, 1686 struct rte_flow_error *error __rte_unused) 1687 { 1688 rte_errno = ENOTSUP; 1689 return -rte_errno; 1690 } 1691 1692 static int 1693 flow_null_apply(struct rte_eth_dev *dev __rte_unused, 1694 struct rte_flow *flow __rte_unused, 1695 struct rte_flow_error *error __rte_unused) 1696 { 1697 rte_errno = ENOTSUP; 1698 return -rte_errno; 1699 } 1700 1701 static void 1702 flow_null_remove(struct rte_eth_dev *dev __rte_unused, 1703 struct rte_flow *flow __rte_unused) 1704 { 1705 } 1706 1707 static void 1708 flow_null_destroy(struct rte_eth_dev *dev __rte_unused, 1709 struct rte_flow *flow __rte_unused) 1710 { 1711 } 1712 1713 static int 1714 flow_null_query(struct rte_eth_dev *dev __rte_unused, 1715 struct rte_flow *flow __rte_unused, 1716 const struct rte_flow_action *actions __rte_unused, 1717 void *data __rte_unused, 1718 struct rte_flow_error *error __rte_unused) 1719 { 1720 rte_errno = ENOTSUP; 1721 return -rte_errno; 1722 } 1723 1724 /* Void driver to protect from null pointer reference. */ 1725 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = { 1726 .validate = flow_null_validate, 1727 .prepare = flow_null_prepare, 1728 .translate = flow_null_translate, 1729 .apply = flow_null_apply, 1730 .remove = flow_null_remove, 1731 .destroy = flow_null_destroy, 1732 .query = flow_null_query, 1733 }; 1734 1735 /** 1736 * Select flow driver type according to flow attributes and device 1737 * configuration. 1738 * 1739 * @param[in] dev 1740 * Pointer to the dev structure. 1741 * @param[in] attr 1742 * Pointer to the flow attributes. 1743 * 1744 * @return 1745 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise. 1746 */ 1747 static enum mlx5_flow_drv_type 1748 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr) 1749 { 1750 struct priv *priv = dev->data->dev_private; 1751 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX; 1752 1753 if (attr->transfer) 1754 type = MLX5_FLOW_TYPE_TCF; 1755 else 1756 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV : 1757 MLX5_FLOW_TYPE_VERBS; 1758 return type; 1759 } 1760 1761 #define flow_get_drv_ops(type) flow_drv_ops[type] 1762 1763 /** 1764 * Flow driver validation API. This abstracts calling driver specific functions. 1765 * The type of flow driver is determined according to flow attributes. 1766 * 1767 * @param[in] dev 1768 * Pointer to the dev structure. 1769 * @param[in] attr 1770 * Pointer to the flow attributes. 1771 * @param[in] items 1772 * Pointer to the list of items. 1773 * @param[in] actions 1774 * Pointer to the list of actions. 1775 * @param[out] error 1776 * Pointer to the error structure. 1777 * 1778 * @return 1779 * 0 on success, a negative errno value otherwise and rte_ernno is set. 1780 */ 1781 static inline int 1782 flow_drv_validate(struct rte_eth_dev *dev, 1783 const struct rte_flow_attr *attr, 1784 const struct rte_flow_item items[], 1785 const struct rte_flow_action actions[], 1786 struct rte_flow_error *error) 1787 { 1788 const struct mlx5_flow_driver_ops *fops; 1789 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr); 1790 1791 fops = flow_get_drv_ops(type); 1792 return fops->validate(dev, attr, items, actions, error); 1793 } 1794 1795 /** 1796 * Flow driver preparation API. This abstracts calling driver specific 1797 * functions. Parent flow (rte_flow) should have driver type (drv_type). It 1798 * calculates the size of memory required for device flow, allocates the memory, 1799 * initializes the device flow and returns the pointer. 1800 * 1801 * @note 1802 * This function initializes device flow structure such as dv, tcf or verbs in 1803 * struct mlx5_flow. However, it is caller's responsibility to initialize the 1804 * rest. For example, adding returning device flow to flow->dev_flow list and 1805 * setting backward reference to the flow should be done out of this function. 1806 * layers field is not filled either. 1807 * 1808 * @param[in] attr 1809 * Pointer to the flow attributes. 1810 * @param[in] items 1811 * Pointer to the list of items. 1812 * @param[in] actions 1813 * Pointer to the list of actions. 1814 * @param[out] error 1815 * Pointer to the error structure. 1816 * 1817 * @return 1818 * Pointer to device flow on success, otherwise NULL and rte_ernno is set. 1819 */ 1820 static inline struct mlx5_flow * 1821 flow_drv_prepare(const struct rte_flow *flow, 1822 const struct rte_flow_attr *attr, 1823 const struct rte_flow_item items[], 1824 const struct rte_flow_action actions[], 1825 struct rte_flow_error *error) 1826 { 1827 const struct mlx5_flow_driver_ops *fops; 1828 enum mlx5_flow_drv_type type = flow->drv_type; 1829 1830 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1831 fops = flow_get_drv_ops(type); 1832 return fops->prepare(attr, items, actions, error); 1833 } 1834 1835 /** 1836 * Flow driver translation API. This abstracts calling driver specific 1837 * functions. Parent flow (rte_flow) should have driver type (drv_type). It 1838 * translates a generic flow into a driver flow. flow_drv_prepare() must 1839 * precede. 1840 * 1841 * @note 1842 * dev_flow->layers could be filled as a result of parsing during translation 1843 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled 1844 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion, 1845 * flow->actions could be overwritten even though all the expanded dev_flows 1846 * have the same actions. 1847 * 1848 * @param[in] dev 1849 * Pointer to the rte dev structure. 1850 * @param[in, out] dev_flow 1851 * Pointer to the mlx5 flow. 1852 * @param[in] attr 1853 * Pointer to the flow attributes. 1854 * @param[in] items 1855 * Pointer to the list of items. 1856 * @param[in] actions 1857 * Pointer to the list of actions. 1858 * @param[out] error 1859 * Pointer to the error structure. 1860 * 1861 * @return 1862 * 0 on success, a negative errno value otherwise and rte_ernno is set. 1863 */ 1864 static inline int 1865 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow, 1866 const struct rte_flow_attr *attr, 1867 const struct rte_flow_item items[], 1868 const struct rte_flow_action actions[], 1869 struct rte_flow_error *error) 1870 { 1871 const struct mlx5_flow_driver_ops *fops; 1872 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type; 1873 1874 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1875 fops = flow_get_drv_ops(type); 1876 return fops->translate(dev, dev_flow, attr, items, actions, error); 1877 } 1878 1879 /** 1880 * Flow driver apply API. This abstracts calling driver specific functions. 1881 * Parent flow (rte_flow) should have driver type (drv_type). It applies 1882 * translated driver flows on to device. flow_drv_translate() must precede. 1883 * 1884 * @param[in] dev 1885 * Pointer to Ethernet device structure. 1886 * @param[in, out] flow 1887 * Pointer to flow structure. 1888 * @param[out] error 1889 * Pointer to error structure. 1890 * 1891 * @return 1892 * 0 on success, a negative errno value otherwise and rte_errno is set. 1893 */ 1894 static inline int 1895 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow, 1896 struct rte_flow_error *error) 1897 { 1898 const struct mlx5_flow_driver_ops *fops; 1899 enum mlx5_flow_drv_type type = flow->drv_type; 1900 1901 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1902 fops = flow_get_drv_ops(type); 1903 return fops->apply(dev, flow, error); 1904 } 1905 1906 /** 1907 * Flow driver remove API. This abstracts calling driver specific functions. 1908 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow 1909 * on device. All the resources of the flow should be freed by calling 1910 * flow_drv_destroy(). 1911 * 1912 * @param[in] dev 1913 * Pointer to Ethernet device. 1914 * @param[in, out] flow 1915 * Pointer to flow structure. 1916 */ 1917 static inline void 1918 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow) 1919 { 1920 const struct mlx5_flow_driver_ops *fops; 1921 enum mlx5_flow_drv_type type = flow->drv_type; 1922 1923 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1924 fops = flow_get_drv_ops(type); 1925 fops->remove(dev, flow); 1926 } 1927 1928 /** 1929 * Flow driver destroy API. This abstracts calling driver specific functions. 1930 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow 1931 * on device and releases resources of the flow. 1932 * 1933 * @param[in] dev 1934 * Pointer to Ethernet device. 1935 * @param[in, out] flow 1936 * Pointer to flow structure. 1937 */ 1938 static inline void 1939 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow) 1940 { 1941 const struct mlx5_flow_driver_ops *fops; 1942 enum mlx5_flow_drv_type type = flow->drv_type; 1943 1944 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1945 fops = flow_get_drv_ops(type); 1946 fops->destroy(dev, flow); 1947 } 1948 1949 /** 1950 * Validate a flow supported by the NIC. 1951 * 1952 * @see rte_flow_validate() 1953 * @see rte_flow_ops 1954 */ 1955 int 1956 mlx5_flow_validate(struct rte_eth_dev *dev, 1957 const struct rte_flow_attr *attr, 1958 const struct rte_flow_item items[], 1959 const struct rte_flow_action actions[], 1960 struct rte_flow_error *error) 1961 { 1962 int ret; 1963 1964 ret = flow_drv_validate(dev, attr, items, actions, error); 1965 if (ret < 0) 1966 return ret; 1967 return 0; 1968 } 1969 1970 /** 1971 * Get RSS action from the action list. 1972 * 1973 * @param[in] actions 1974 * Pointer to the list of actions. 1975 * 1976 * @return 1977 * Pointer to the RSS action if exist, else return NULL. 1978 */ 1979 static const struct rte_flow_action_rss* 1980 flow_get_rss_action(const struct rte_flow_action actions[]) 1981 { 1982 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) { 1983 switch (actions->type) { 1984 case RTE_FLOW_ACTION_TYPE_RSS: 1985 return (const struct rte_flow_action_rss *) 1986 actions->conf; 1987 default: 1988 break; 1989 } 1990 } 1991 return NULL; 1992 } 1993 1994 static unsigned int 1995 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level) 1996 { 1997 const struct rte_flow_item *item; 1998 unsigned int has_vlan = 0; 1999 2000 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) { 2001 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) { 2002 has_vlan = 1; 2003 break; 2004 } 2005 } 2006 if (has_vlan) 2007 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN : 2008 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN; 2009 return rss_level < 2 ? MLX5_EXPANSION_ROOT : 2010 MLX5_EXPANSION_ROOT_OUTER; 2011 } 2012 2013 /** 2014 * Create a flow and add it to @p list. 2015 * 2016 * @param dev 2017 * Pointer to Ethernet device. 2018 * @param list 2019 * Pointer to a TAILQ flow list. 2020 * @param[in] attr 2021 * Flow rule attributes. 2022 * @param[in] items 2023 * Pattern specification (list terminated by the END pattern item). 2024 * @param[in] actions 2025 * Associated actions (list terminated by the END action). 2026 * @param[out] error 2027 * Perform verbose error reporting if not NULL. 2028 * 2029 * @return 2030 * A flow on success, NULL otherwise and rte_errno is set. 2031 */ 2032 static struct rte_flow * 2033 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list, 2034 const struct rte_flow_attr *attr, 2035 const struct rte_flow_item items[], 2036 const struct rte_flow_action actions[], 2037 struct rte_flow_error *error) 2038 { 2039 struct rte_flow *flow = NULL; 2040 struct mlx5_flow *dev_flow; 2041 const struct rte_flow_action_rss *rss; 2042 union { 2043 struct rte_flow_expand_rss buf; 2044 uint8_t buffer[2048]; 2045 } expand_buffer; 2046 struct rte_flow_expand_rss *buf = &expand_buffer.buf; 2047 int ret; 2048 uint32_t i; 2049 uint32_t flow_size; 2050 2051 ret = flow_drv_validate(dev, attr, items, actions, error); 2052 if (ret < 0) 2053 return NULL; 2054 flow_size = sizeof(struct rte_flow); 2055 rss = flow_get_rss_action(actions); 2056 if (rss) 2057 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t), 2058 sizeof(void *)); 2059 else 2060 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *)); 2061 flow = rte_calloc(__func__, 1, flow_size, 0); 2062 flow->drv_type = flow_get_drv_type(dev, attr); 2063 assert(flow->drv_type > MLX5_FLOW_TYPE_MIN && 2064 flow->drv_type < MLX5_FLOW_TYPE_MAX); 2065 flow->queue = (void *)(flow + 1); 2066 LIST_INIT(&flow->dev_flows); 2067 if (rss && rss->types) { 2068 unsigned int graph_root; 2069 2070 graph_root = find_graph_root(items, rss->level); 2071 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer), 2072 items, rss->types, 2073 mlx5_support_expansion, 2074 graph_root); 2075 assert(ret > 0 && 2076 (unsigned int)ret < sizeof(expand_buffer.buffer)); 2077 } else { 2078 buf->entries = 1; 2079 buf->entry[0].pattern = (void *)(uintptr_t)items; 2080 } 2081 for (i = 0; i < buf->entries; ++i) { 2082 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern, 2083 actions, error); 2084 if (!dev_flow) 2085 goto error; 2086 dev_flow->flow = flow; 2087 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next); 2088 ret = flow_drv_translate(dev, dev_flow, attr, 2089 buf->entry[i].pattern, 2090 actions, error); 2091 if (ret < 0) 2092 goto error; 2093 } 2094 if (dev->data->dev_started) { 2095 ret = flow_drv_apply(dev, flow, error); 2096 if (ret < 0) 2097 goto error; 2098 } 2099 TAILQ_INSERT_TAIL(list, flow, next); 2100 flow_rxq_flags_set(dev, flow); 2101 return flow; 2102 error: 2103 ret = rte_errno; /* Save rte_errno before cleanup. */ 2104 assert(flow); 2105 flow_drv_destroy(dev, flow); 2106 rte_free(flow); 2107 rte_errno = ret; /* Restore rte_errno. */ 2108 return NULL; 2109 } 2110 2111 /** 2112 * Create a flow. 2113 * 2114 * @see rte_flow_create() 2115 * @see rte_flow_ops 2116 */ 2117 struct rte_flow * 2118 mlx5_flow_create(struct rte_eth_dev *dev, 2119 const struct rte_flow_attr *attr, 2120 const struct rte_flow_item items[], 2121 const struct rte_flow_action actions[], 2122 struct rte_flow_error *error) 2123 { 2124 return flow_list_create(dev, 2125 &((struct priv *)dev->data->dev_private)->flows, 2126 attr, items, actions, error); 2127 } 2128 2129 /** 2130 * Destroy a flow in a list. 2131 * 2132 * @param dev 2133 * Pointer to Ethernet device. 2134 * @param list 2135 * Pointer to a TAILQ flow list. 2136 * @param[in] flow 2137 * Flow to destroy. 2138 */ 2139 static void 2140 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list, 2141 struct rte_flow *flow) 2142 { 2143 /* 2144 * Update RX queue flags only if port is started, otherwise it is 2145 * already clean. 2146 */ 2147 if (dev->data->dev_started) 2148 flow_rxq_flags_trim(dev, flow); 2149 flow_drv_destroy(dev, flow); 2150 TAILQ_REMOVE(list, flow, next); 2151 rte_free(flow->fdir); 2152 rte_free(flow); 2153 } 2154 2155 /** 2156 * Destroy all flows. 2157 * 2158 * @param dev 2159 * Pointer to Ethernet device. 2160 * @param list 2161 * Pointer to a TAILQ flow list. 2162 */ 2163 void 2164 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list) 2165 { 2166 while (!TAILQ_EMPTY(list)) { 2167 struct rte_flow *flow; 2168 2169 flow = TAILQ_FIRST(list); 2170 flow_list_destroy(dev, list, flow); 2171 } 2172 } 2173 2174 /** 2175 * Remove all flows. 2176 * 2177 * @param dev 2178 * Pointer to Ethernet device. 2179 * @param list 2180 * Pointer to a TAILQ flow list. 2181 */ 2182 void 2183 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list) 2184 { 2185 struct rte_flow *flow; 2186 2187 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) 2188 flow_drv_remove(dev, flow); 2189 flow_rxq_flags_clear(dev); 2190 } 2191 2192 /** 2193 * Add all flows. 2194 * 2195 * @param dev 2196 * Pointer to Ethernet device. 2197 * @param list 2198 * Pointer to a TAILQ flow list. 2199 * 2200 * @return 2201 * 0 on success, a negative errno value otherwise and rte_errno is set. 2202 */ 2203 int 2204 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list) 2205 { 2206 struct rte_flow *flow; 2207 struct rte_flow_error error; 2208 int ret = 0; 2209 2210 TAILQ_FOREACH(flow, list, next) { 2211 ret = flow_drv_apply(dev, flow, &error); 2212 if (ret < 0) 2213 goto error; 2214 flow_rxq_flags_set(dev, flow); 2215 } 2216 return 0; 2217 error: 2218 ret = rte_errno; /* Save rte_errno before cleanup. */ 2219 mlx5_flow_stop(dev, list); 2220 rte_errno = ret; /* Restore rte_errno. */ 2221 return -rte_errno; 2222 } 2223 2224 /** 2225 * Verify the flow list is empty 2226 * 2227 * @param dev 2228 * Pointer to Ethernet device. 2229 * 2230 * @return the number of flows not released. 2231 */ 2232 int 2233 mlx5_flow_verify(struct rte_eth_dev *dev) 2234 { 2235 struct priv *priv = dev->data->dev_private; 2236 struct rte_flow *flow; 2237 int ret = 0; 2238 2239 TAILQ_FOREACH(flow, &priv->flows, next) { 2240 DRV_LOG(DEBUG, "port %u flow %p still referenced", 2241 dev->data->port_id, (void *)flow); 2242 ++ret; 2243 } 2244 return ret; 2245 } 2246 2247 /** 2248 * Enable a control flow configured from the control plane. 2249 * 2250 * @param dev 2251 * Pointer to Ethernet device. 2252 * @param eth_spec 2253 * An Ethernet flow spec to apply. 2254 * @param eth_mask 2255 * An Ethernet flow mask to apply. 2256 * @param vlan_spec 2257 * A VLAN flow spec to apply. 2258 * @param vlan_mask 2259 * A VLAN flow mask to apply. 2260 * 2261 * @return 2262 * 0 on success, a negative errno value otherwise and rte_errno is set. 2263 */ 2264 int 2265 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev, 2266 struct rte_flow_item_eth *eth_spec, 2267 struct rte_flow_item_eth *eth_mask, 2268 struct rte_flow_item_vlan *vlan_spec, 2269 struct rte_flow_item_vlan *vlan_mask) 2270 { 2271 struct priv *priv = dev->data->dev_private; 2272 const struct rte_flow_attr attr = { 2273 .ingress = 1, 2274 .priority = MLX5_FLOW_PRIO_RSVD, 2275 }; 2276 struct rte_flow_item items[] = { 2277 { 2278 .type = RTE_FLOW_ITEM_TYPE_ETH, 2279 .spec = eth_spec, 2280 .last = NULL, 2281 .mask = eth_mask, 2282 }, 2283 { 2284 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN : 2285 RTE_FLOW_ITEM_TYPE_END, 2286 .spec = vlan_spec, 2287 .last = NULL, 2288 .mask = vlan_mask, 2289 }, 2290 { 2291 .type = RTE_FLOW_ITEM_TYPE_END, 2292 }, 2293 }; 2294 uint16_t queue[priv->reta_idx_n]; 2295 struct rte_flow_action_rss action_rss = { 2296 .func = RTE_ETH_HASH_FUNCTION_DEFAULT, 2297 .level = 0, 2298 .types = priv->rss_conf.rss_hf, 2299 .key_len = priv->rss_conf.rss_key_len, 2300 .queue_num = priv->reta_idx_n, 2301 .key = priv->rss_conf.rss_key, 2302 .queue = queue, 2303 }; 2304 struct rte_flow_action actions[] = { 2305 { 2306 .type = RTE_FLOW_ACTION_TYPE_RSS, 2307 .conf = &action_rss, 2308 }, 2309 { 2310 .type = RTE_FLOW_ACTION_TYPE_END, 2311 }, 2312 }; 2313 struct rte_flow *flow; 2314 struct rte_flow_error error; 2315 unsigned int i; 2316 2317 if (!priv->reta_idx_n || !priv->rxqs_n) { 2318 rte_errno = EINVAL; 2319 return -rte_errno; 2320 } 2321 for (i = 0; i != priv->reta_idx_n; ++i) 2322 queue[i] = (*priv->reta_idx)[i]; 2323 flow = flow_list_create(dev, &priv->ctrl_flows, 2324 &attr, items, actions, &error); 2325 if (!flow) 2326 return -rte_errno; 2327 return 0; 2328 } 2329 2330 /** 2331 * Enable a flow control configured from the control plane. 2332 * 2333 * @param dev 2334 * Pointer to Ethernet device. 2335 * @param eth_spec 2336 * An Ethernet flow spec to apply. 2337 * @param eth_mask 2338 * An Ethernet flow mask to apply. 2339 * 2340 * @return 2341 * 0 on success, a negative errno value otherwise and rte_errno is set. 2342 */ 2343 int 2344 mlx5_ctrl_flow(struct rte_eth_dev *dev, 2345 struct rte_flow_item_eth *eth_spec, 2346 struct rte_flow_item_eth *eth_mask) 2347 { 2348 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL); 2349 } 2350 2351 /** 2352 * Destroy a flow. 2353 * 2354 * @see rte_flow_destroy() 2355 * @see rte_flow_ops 2356 */ 2357 int 2358 mlx5_flow_destroy(struct rte_eth_dev *dev, 2359 struct rte_flow *flow, 2360 struct rte_flow_error *error __rte_unused) 2361 { 2362 struct priv *priv = dev->data->dev_private; 2363 2364 flow_list_destroy(dev, &priv->flows, flow); 2365 return 0; 2366 } 2367 2368 /** 2369 * Destroy all flows. 2370 * 2371 * @see rte_flow_flush() 2372 * @see rte_flow_ops 2373 */ 2374 int 2375 mlx5_flow_flush(struct rte_eth_dev *dev, 2376 struct rte_flow_error *error __rte_unused) 2377 { 2378 struct priv *priv = dev->data->dev_private; 2379 2380 mlx5_flow_list_flush(dev, &priv->flows); 2381 return 0; 2382 } 2383 2384 /** 2385 * Isolated mode. 2386 * 2387 * @see rte_flow_isolate() 2388 * @see rte_flow_ops 2389 */ 2390 int 2391 mlx5_flow_isolate(struct rte_eth_dev *dev, 2392 int enable, 2393 struct rte_flow_error *error) 2394 { 2395 struct priv *priv = dev->data->dev_private; 2396 2397 if (dev->data->dev_started) { 2398 rte_flow_error_set(error, EBUSY, 2399 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, 2400 NULL, 2401 "port must be stopped first"); 2402 return -rte_errno; 2403 } 2404 priv->isolated = !!enable; 2405 if (enable) 2406 dev->dev_ops = &mlx5_dev_ops_isolate; 2407 else 2408 dev->dev_ops = &mlx5_dev_ops; 2409 return 0; 2410 } 2411 2412 /** 2413 * Query a flow. 2414 * 2415 * @see rte_flow_query() 2416 * @see rte_flow_ops 2417 */ 2418 static int 2419 flow_drv_query(struct rte_eth_dev *dev, 2420 struct rte_flow *flow, 2421 const struct rte_flow_action *actions, 2422 void *data, 2423 struct rte_flow_error *error) 2424 { 2425 const struct mlx5_flow_driver_ops *fops; 2426 enum mlx5_flow_drv_type ftype = flow->drv_type; 2427 2428 assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX); 2429 fops = flow_get_drv_ops(ftype); 2430 2431 return fops->query(dev, flow, actions, data, error); 2432 } 2433 2434 /** 2435 * Query a flow. 2436 * 2437 * @see rte_flow_query() 2438 * @see rte_flow_ops 2439 */ 2440 int 2441 mlx5_flow_query(struct rte_eth_dev *dev, 2442 struct rte_flow *flow, 2443 const struct rte_flow_action *actions, 2444 void *data, 2445 struct rte_flow_error *error) 2446 { 2447 int ret; 2448 2449 ret = flow_drv_query(dev, flow, actions, data, error); 2450 if (ret < 0) 2451 return ret; 2452 return 0; 2453 } 2454 2455 /** 2456 * Convert a flow director filter to a generic flow. 2457 * 2458 * @param dev 2459 * Pointer to Ethernet device. 2460 * @param fdir_filter 2461 * Flow director filter to add. 2462 * @param attributes 2463 * Generic flow parameters structure. 2464 * 2465 * @return 2466 * 0 on success, a negative errno value otherwise and rte_errno is set. 2467 */ 2468 static int 2469 flow_fdir_filter_convert(struct rte_eth_dev *dev, 2470 const struct rte_eth_fdir_filter *fdir_filter, 2471 struct mlx5_fdir *attributes) 2472 { 2473 struct priv *priv = dev->data->dev_private; 2474 const struct rte_eth_fdir_input *input = &fdir_filter->input; 2475 const struct rte_eth_fdir_masks *mask = 2476 &dev->data->dev_conf.fdir_conf.mask; 2477 2478 /* Validate queue number. */ 2479 if (fdir_filter->action.rx_queue >= priv->rxqs_n) { 2480 DRV_LOG(ERR, "port %u invalid queue number %d", 2481 dev->data->port_id, fdir_filter->action.rx_queue); 2482 rte_errno = EINVAL; 2483 return -rte_errno; 2484 } 2485 attributes->attr.ingress = 1; 2486 attributes->items[0] = (struct rte_flow_item) { 2487 .type = RTE_FLOW_ITEM_TYPE_ETH, 2488 .spec = &attributes->l2, 2489 .mask = &attributes->l2_mask, 2490 }; 2491 switch (fdir_filter->action.behavior) { 2492 case RTE_ETH_FDIR_ACCEPT: 2493 attributes->actions[0] = (struct rte_flow_action){ 2494 .type = RTE_FLOW_ACTION_TYPE_QUEUE, 2495 .conf = &attributes->queue, 2496 }; 2497 break; 2498 case RTE_ETH_FDIR_REJECT: 2499 attributes->actions[0] = (struct rte_flow_action){ 2500 .type = RTE_FLOW_ACTION_TYPE_DROP, 2501 }; 2502 break; 2503 default: 2504 DRV_LOG(ERR, "port %u invalid behavior %d", 2505 dev->data->port_id, 2506 fdir_filter->action.behavior); 2507 rte_errno = ENOTSUP; 2508 return -rte_errno; 2509 } 2510 attributes->queue.index = fdir_filter->action.rx_queue; 2511 /* Handle L3. */ 2512 switch (fdir_filter->input.flow_type) { 2513 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP: 2514 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP: 2515 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER: 2516 attributes->l3.ipv4.hdr = (struct ipv4_hdr){ 2517 .src_addr = input->flow.ip4_flow.src_ip, 2518 .dst_addr = input->flow.ip4_flow.dst_ip, 2519 .time_to_live = input->flow.ip4_flow.ttl, 2520 .type_of_service = input->flow.ip4_flow.tos, 2521 }; 2522 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){ 2523 .src_addr = mask->ipv4_mask.src_ip, 2524 .dst_addr = mask->ipv4_mask.dst_ip, 2525 .time_to_live = mask->ipv4_mask.ttl, 2526 .type_of_service = mask->ipv4_mask.tos, 2527 .next_proto_id = mask->ipv4_mask.proto, 2528 }; 2529 attributes->items[1] = (struct rte_flow_item){ 2530 .type = RTE_FLOW_ITEM_TYPE_IPV4, 2531 .spec = &attributes->l3, 2532 .mask = &attributes->l3_mask, 2533 }; 2534 break; 2535 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP: 2536 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP: 2537 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER: 2538 attributes->l3.ipv6.hdr = (struct ipv6_hdr){ 2539 .hop_limits = input->flow.ipv6_flow.hop_limits, 2540 .proto = input->flow.ipv6_flow.proto, 2541 }; 2542 2543 memcpy(attributes->l3.ipv6.hdr.src_addr, 2544 input->flow.ipv6_flow.src_ip, 2545 RTE_DIM(attributes->l3.ipv6.hdr.src_addr)); 2546 memcpy(attributes->l3.ipv6.hdr.dst_addr, 2547 input->flow.ipv6_flow.dst_ip, 2548 RTE_DIM(attributes->l3.ipv6.hdr.src_addr)); 2549 memcpy(attributes->l3_mask.ipv6.hdr.src_addr, 2550 mask->ipv6_mask.src_ip, 2551 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr)); 2552 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr, 2553 mask->ipv6_mask.dst_ip, 2554 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr)); 2555 attributes->items[1] = (struct rte_flow_item){ 2556 .type = RTE_FLOW_ITEM_TYPE_IPV6, 2557 .spec = &attributes->l3, 2558 .mask = &attributes->l3_mask, 2559 }; 2560 break; 2561 default: 2562 DRV_LOG(ERR, "port %u invalid flow type%d", 2563 dev->data->port_id, fdir_filter->input.flow_type); 2564 rte_errno = ENOTSUP; 2565 return -rte_errno; 2566 } 2567 /* Handle L4. */ 2568 switch (fdir_filter->input.flow_type) { 2569 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP: 2570 attributes->l4.udp.hdr = (struct udp_hdr){ 2571 .src_port = input->flow.udp4_flow.src_port, 2572 .dst_port = input->flow.udp4_flow.dst_port, 2573 }; 2574 attributes->l4_mask.udp.hdr = (struct udp_hdr){ 2575 .src_port = mask->src_port_mask, 2576 .dst_port = mask->dst_port_mask, 2577 }; 2578 attributes->items[2] = (struct rte_flow_item){ 2579 .type = RTE_FLOW_ITEM_TYPE_UDP, 2580 .spec = &attributes->l4, 2581 .mask = &attributes->l4_mask, 2582 }; 2583 break; 2584 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP: 2585 attributes->l4.tcp.hdr = (struct tcp_hdr){ 2586 .src_port = input->flow.tcp4_flow.src_port, 2587 .dst_port = input->flow.tcp4_flow.dst_port, 2588 }; 2589 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){ 2590 .src_port = mask->src_port_mask, 2591 .dst_port = mask->dst_port_mask, 2592 }; 2593 attributes->items[2] = (struct rte_flow_item){ 2594 .type = RTE_FLOW_ITEM_TYPE_TCP, 2595 .spec = &attributes->l4, 2596 .mask = &attributes->l4_mask, 2597 }; 2598 break; 2599 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP: 2600 attributes->l4.udp.hdr = (struct udp_hdr){ 2601 .src_port = input->flow.udp6_flow.src_port, 2602 .dst_port = input->flow.udp6_flow.dst_port, 2603 }; 2604 attributes->l4_mask.udp.hdr = (struct udp_hdr){ 2605 .src_port = mask->src_port_mask, 2606 .dst_port = mask->dst_port_mask, 2607 }; 2608 attributes->items[2] = (struct rte_flow_item){ 2609 .type = RTE_FLOW_ITEM_TYPE_UDP, 2610 .spec = &attributes->l4, 2611 .mask = &attributes->l4_mask, 2612 }; 2613 break; 2614 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP: 2615 attributes->l4.tcp.hdr = (struct tcp_hdr){ 2616 .src_port = input->flow.tcp6_flow.src_port, 2617 .dst_port = input->flow.tcp6_flow.dst_port, 2618 }; 2619 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){ 2620 .src_port = mask->src_port_mask, 2621 .dst_port = mask->dst_port_mask, 2622 }; 2623 attributes->items[2] = (struct rte_flow_item){ 2624 .type = RTE_FLOW_ITEM_TYPE_TCP, 2625 .spec = &attributes->l4, 2626 .mask = &attributes->l4_mask, 2627 }; 2628 break; 2629 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER: 2630 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER: 2631 break; 2632 default: 2633 DRV_LOG(ERR, "port %u invalid flow type%d", 2634 dev->data->port_id, fdir_filter->input.flow_type); 2635 rte_errno = ENOTSUP; 2636 return -rte_errno; 2637 } 2638 return 0; 2639 } 2640 2641 #define FLOW_FDIR_CMP(f1, f2, fld) \ 2642 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld)) 2643 2644 /** 2645 * Compare two FDIR flows. If items and actions are identical, the two flows are 2646 * regarded as same. 2647 * 2648 * @param dev 2649 * Pointer to Ethernet device. 2650 * @param f1 2651 * FDIR flow to compare. 2652 * @param f2 2653 * FDIR flow to compare. 2654 * 2655 * @return 2656 * Zero on match, 1 otherwise. 2657 */ 2658 static int 2659 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2) 2660 { 2661 if (FLOW_FDIR_CMP(f1, f2, attr) || 2662 FLOW_FDIR_CMP(f1, f2, l2) || 2663 FLOW_FDIR_CMP(f1, f2, l2_mask) || 2664 FLOW_FDIR_CMP(f1, f2, l3) || 2665 FLOW_FDIR_CMP(f1, f2, l3_mask) || 2666 FLOW_FDIR_CMP(f1, f2, l4) || 2667 FLOW_FDIR_CMP(f1, f2, l4_mask) || 2668 FLOW_FDIR_CMP(f1, f2, actions[0].type)) 2669 return 1; 2670 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE && 2671 FLOW_FDIR_CMP(f1, f2, queue)) 2672 return 1; 2673 return 0; 2674 } 2675 2676 /** 2677 * Search device flow list to find out a matched FDIR flow. 2678 * 2679 * @param dev 2680 * Pointer to Ethernet device. 2681 * @param fdir_flow 2682 * FDIR flow to lookup. 2683 * 2684 * @return 2685 * Pointer of flow if found, NULL otherwise. 2686 */ 2687 static struct rte_flow * 2688 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow) 2689 { 2690 struct priv *priv = dev->data->dev_private; 2691 struct rte_flow *flow = NULL; 2692 2693 assert(fdir_flow); 2694 TAILQ_FOREACH(flow, &priv->flows, next) { 2695 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) { 2696 DRV_LOG(DEBUG, "port %u found FDIR flow %p", 2697 dev->data->port_id, (void *)flow); 2698 break; 2699 } 2700 } 2701 return flow; 2702 } 2703 2704 /** 2705 * Add new flow director filter and store it in list. 2706 * 2707 * @param dev 2708 * Pointer to Ethernet device. 2709 * @param fdir_filter 2710 * Flow director filter to add. 2711 * 2712 * @return 2713 * 0 on success, a negative errno value otherwise and rte_errno is set. 2714 */ 2715 static int 2716 flow_fdir_filter_add(struct rte_eth_dev *dev, 2717 const struct rte_eth_fdir_filter *fdir_filter) 2718 { 2719 struct priv *priv = dev->data->dev_private; 2720 struct mlx5_fdir *fdir_flow; 2721 struct rte_flow *flow; 2722 int ret; 2723 2724 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0); 2725 if (!fdir_flow) { 2726 rte_errno = ENOMEM; 2727 return -rte_errno; 2728 } 2729 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow); 2730 if (ret) 2731 goto error; 2732 flow = flow_fdir_filter_lookup(dev, fdir_flow); 2733 if (flow) { 2734 rte_errno = EEXIST; 2735 goto error; 2736 } 2737 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr, 2738 fdir_flow->items, fdir_flow->actions, NULL); 2739 if (!flow) 2740 goto error; 2741 assert(!flow->fdir); 2742 flow->fdir = fdir_flow; 2743 DRV_LOG(DEBUG, "port %u created FDIR flow %p", 2744 dev->data->port_id, (void *)flow); 2745 return 0; 2746 error: 2747 rte_free(fdir_flow); 2748 return -rte_errno; 2749 } 2750 2751 /** 2752 * Delete specific filter. 2753 * 2754 * @param dev 2755 * Pointer to Ethernet device. 2756 * @param fdir_filter 2757 * Filter to be deleted. 2758 * 2759 * @return 2760 * 0 on success, a negative errno value otherwise and rte_errno is set. 2761 */ 2762 static int 2763 flow_fdir_filter_delete(struct rte_eth_dev *dev, 2764 const struct rte_eth_fdir_filter *fdir_filter) 2765 { 2766 struct priv *priv = dev->data->dev_private; 2767 struct rte_flow *flow; 2768 struct mlx5_fdir fdir_flow = { 2769 .attr.group = 0, 2770 }; 2771 int ret; 2772 2773 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow); 2774 if (ret) 2775 return -rte_errno; 2776 flow = flow_fdir_filter_lookup(dev, &fdir_flow); 2777 if (!flow) { 2778 rte_errno = ENOENT; 2779 return -rte_errno; 2780 } 2781 flow_list_destroy(dev, &priv->flows, flow); 2782 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p", 2783 dev->data->port_id, (void *)flow); 2784 return 0; 2785 } 2786 2787 /** 2788 * Update queue for specific filter. 2789 * 2790 * @param dev 2791 * Pointer to Ethernet device. 2792 * @param fdir_filter 2793 * Filter to be updated. 2794 * 2795 * @return 2796 * 0 on success, a negative errno value otherwise and rte_errno is set. 2797 */ 2798 static int 2799 flow_fdir_filter_update(struct rte_eth_dev *dev, 2800 const struct rte_eth_fdir_filter *fdir_filter) 2801 { 2802 int ret; 2803 2804 ret = flow_fdir_filter_delete(dev, fdir_filter); 2805 if (ret) 2806 return ret; 2807 return flow_fdir_filter_add(dev, fdir_filter); 2808 } 2809 2810 /** 2811 * Flush all filters. 2812 * 2813 * @param dev 2814 * Pointer to Ethernet device. 2815 */ 2816 static void 2817 flow_fdir_filter_flush(struct rte_eth_dev *dev) 2818 { 2819 struct priv *priv = dev->data->dev_private; 2820 2821 mlx5_flow_list_flush(dev, &priv->flows); 2822 } 2823 2824 /** 2825 * Get flow director information. 2826 * 2827 * @param dev 2828 * Pointer to Ethernet device. 2829 * @param[out] fdir_info 2830 * Resulting flow director information. 2831 */ 2832 static void 2833 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info) 2834 { 2835 struct rte_eth_fdir_masks *mask = 2836 &dev->data->dev_conf.fdir_conf.mask; 2837 2838 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode; 2839 fdir_info->guarant_spc = 0; 2840 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask)); 2841 fdir_info->max_flexpayload = 0; 2842 fdir_info->flow_types_mask[0] = 0; 2843 fdir_info->flex_payload_unit = 0; 2844 fdir_info->max_flex_payload_segment_num = 0; 2845 fdir_info->flex_payload_limit = 0; 2846 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf)); 2847 } 2848 2849 /** 2850 * Deal with flow director operations. 2851 * 2852 * @param dev 2853 * Pointer to Ethernet device. 2854 * @param filter_op 2855 * Operation to perform. 2856 * @param arg 2857 * Pointer to operation-specific structure. 2858 * 2859 * @return 2860 * 0 on success, a negative errno value otherwise and rte_errno is set. 2861 */ 2862 static int 2863 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op, 2864 void *arg) 2865 { 2866 enum rte_fdir_mode fdir_mode = 2867 dev->data->dev_conf.fdir_conf.mode; 2868 2869 if (filter_op == RTE_ETH_FILTER_NOP) 2870 return 0; 2871 if (fdir_mode != RTE_FDIR_MODE_PERFECT && 2872 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) { 2873 DRV_LOG(ERR, "port %u flow director mode %d not supported", 2874 dev->data->port_id, fdir_mode); 2875 rte_errno = EINVAL; 2876 return -rte_errno; 2877 } 2878 switch (filter_op) { 2879 case RTE_ETH_FILTER_ADD: 2880 return flow_fdir_filter_add(dev, arg); 2881 case RTE_ETH_FILTER_UPDATE: 2882 return flow_fdir_filter_update(dev, arg); 2883 case RTE_ETH_FILTER_DELETE: 2884 return flow_fdir_filter_delete(dev, arg); 2885 case RTE_ETH_FILTER_FLUSH: 2886 flow_fdir_filter_flush(dev); 2887 break; 2888 case RTE_ETH_FILTER_INFO: 2889 flow_fdir_info_get(dev, arg); 2890 break; 2891 default: 2892 DRV_LOG(DEBUG, "port %u unknown operation %u", 2893 dev->data->port_id, filter_op); 2894 rte_errno = EINVAL; 2895 return -rte_errno; 2896 } 2897 return 0; 2898 } 2899 2900 /** 2901 * Manage filter operations. 2902 * 2903 * @param dev 2904 * Pointer to Ethernet device structure. 2905 * @param filter_type 2906 * Filter type. 2907 * @param filter_op 2908 * Operation to perform. 2909 * @param arg 2910 * Pointer to operation-specific structure. 2911 * 2912 * @return 2913 * 0 on success, a negative errno value otherwise and rte_errno is set. 2914 */ 2915 int 2916 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev, 2917 enum rte_filter_type filter_type, 2918 enum rte_filter_op filter_op, 2919 void *arg) 2920 { 2921 switch (filter_type) { 2922 case RTE_ETH_FILTER_GENERIC: 2923 if (filter_op != RTE_ETH_FILTER_GET) { 2924 rte_errno = EINVAL; 2925 return -rte_errno; 2926 } 2927 *(const void **)arg = &mlx5_flow_ops; 2928 return 0; 2929 case RTE_ETH_FILTER_FDIR: 2930 return flow_fdir_ctrl_func(dev, filter_op, arg); 2931 default: 2932 DRV_LOG(ERR, "port %u filter type (%d) not supported", 2933 dev->data->port_id, filter_type); 2934 rte_errno = ENOTSUP; 2935 return -rte_errno; 2936 } 2937 return 0; 2938 } 2939