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 mlx5_hrxq_drop_release(dev); 354 switch (priority) { 355 case 8: 356 priority = RTE_DIM(priority_map_3); 357 break; 358 case 16: 359 priority = RTE_DIM(priority_map_5); 360 break; 361 default: 362 rte_errno = ENOTSUP; 363 DRV_LOG(ERR, 364 "port %u verbs maximum priority: %d expected 8/16", 365 dev->data->port_id, priority); 366 return -rte_errno; 367 } 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 mlx5_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 fields 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 mlx5_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 mlx5_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 mlx5_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_errno 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_errno 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 mlx5_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[in] item_flags 879 * Items that were detected. 880 * @param[out] error 881 * Pointer to error structure. 882 * 883 * @return 884 * 0 on success, a negative errno value otherwise and rte_errno is set. 885 */ 886 int 887 mlx5_flow_validate_action_rss(const struct rte_flow_action *action, 888 uint64_t action_flags, 889 struct rte_eth_dev *dev, 890 const struct rte_flow_attr *attr, 891 uint64_t item_flags, 892 struct rte_flow_error *error) 893 { 894 struct mlx5_priv *priv = dev->data->dev_private; 895 const struct rte_flow_action_rss *rss = action->conf; 896 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 897 unsigned int i; 898 899 if (action_flags & MLX5_FLOW_FATE_ACTIONS) 900 return rte_flow_error_set(error, EINVAL, 901 RTE_FLOW_ERROR_TYPE_ACTION, NULL, 902 "can't have 2 fate actions" 903 " in same flow"); 904 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT && 905 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ) 906 return rte_flow_error_set(error, ENOTSUP, 907 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 908 &rss->func, 909 "RSS hash function not supported"); 910 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT 911 if (rss->level > 2) 912 #else 913 if (rss->level > 1) 914 #endif 915 return rte_flow_error_set(error, ENOTSUP, 916 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 917 &rss->level, 918 "tunnel RSS is not supported"); 919 /* allow RSS key_len 0 in case of NULL (default) RSS key. */ 920 if (rss->key_len == 0 && rss->key != NULL) 921 return rte_flow_error_set(error, ENOTSUP, 922 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 923 &rss->key_len, 924 "RSS hash key length 0"); 925 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN) 926 return rte_flow_error_set(error, ENOTSUP, 927 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 928 &rss->key_len, 929 "RSS hash key too small"); 930 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN) 931 return rte_flow_error_set(error, ENOTSUP, 932 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 933 &rss->key_len, 934 "RSS hash key too large"); 935 if (rss->queue_num > priv->config.ind_table_max_size) 936 return rte_flow_error_set(error, ENOTSUP, 937 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 938 &rss->queue_num, 939 "number of queues too large"); 940 if (rss->types & MLX5_RSS_HF_MASK) 941 return rte_flow_error_set(error, ENOTSUP, 942 RTE_FLOW_ERROR_TYPE_ACTION_CONF, 943 &rss->types, 944 "some RSS protocols are not" 945 " supported"); 946 for (i = 0; i != rss->queue_num; ++i) { 947 if (!(*priv->rxqs)[rss->queue[i]]) 948 return rte_flow_error_set 949 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF, 950 &rss->queue[i], "queue is not configured"); 951 } 952 if (attr->egress) 953 return rte_flow_error_set(error, ENOTSUP, 954 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 955 "rss action not supported for " 956 "egress"); 957 if (rss->level > 1 && !tunnel) 958 return rte_flow_error_set(error, EINVAL, 959 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL, 960 "inner RSS is not supported for " 961 "non-tunnel flows"); 962 return 0; 963 } 964 965 /* 966 * Validate the count action. 967 * 968 * @param[in] dev 969 * Pointer to the Ethernet device structure. 970 * @param[in] attr 971 * Attributes of flow that includes this action. 972 * @param[out] error 973 * Pointer to error structure. 974 * 975 * @return 976 * 0 on success, a negative errno value otherwise and rte_errno is set. 977 */ 978 int 979 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused, 980 const struct rte_flow_attr *attr, 981 struct rte_flow_error *error) 982 { 983 if (attr->egress) 984 return rte_flow_error_set(error, ENOTSUP, 985 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 986 "count action not supported for " 987 "egress"); 988 return 0; 989 } 990 991 /** 992 * Verify the @p attributes will be correctly understood by the NIC and store 993 * them in the @p flow if everything is correct. 994 * 995 * @param[in] dev 996 * Pointer to the Ethernet device structure. 997 * @param[in] attributes 998 * Pointer to flow attributes 999 * @param[out] error 1000 * Pointer to error structure. 1001 * 1002 * @return 1003 * 0 on success, a negative errno value otherwise and rte_errno is set. 1004 */ 1005 int 1006 mlx5_flow_validate_attributes(struct rte_eth_dev *dev, 1007 const struct rte_flow_attr *attributes, 1008 struct rte_flow_error *error) 1009 { 1010 struct mlx5_priv *priv = dev->data->dev_private; 1011 uint32_t priority_max = priv->config.flow_prio - 1; 1012 1013 if (attributes->group) 1014 return rte_flow_error_set(error, ENOTSUP, 1015 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, 1016 NULL, "groups is not supported"); 1017 if (attributes->priority != MLX5_FLOW_PRIO_RSVD && 1018 attributes->priority >= priority_max) 1019 return rte_flow_error_set(error, ENOTSUP, 1020 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY, 1021 NULL, "priority out of range"); 1022 if (attributes->egress) 1023 return rte_flow_error_set(error, ENOTSUP, 1024 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL, 1025 "egress is not supported"); 1026 if (attributes->transfer) 1027 return rte_flow_error_set(error, ENOTSUP, 1028 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER, 1029 NULL, "transfer is not supported"); 1030 if (!attributes->ingress) 1031 return rte_flow_error_set(error, EINVAL, 1032 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS, 1033 NULL, 1034 "ingress attribute is mandatory"); 1035 return 0; 1036 } 1037 1038 /** 1039 * Validate Ethernet item. 1040 * 1041 * @param[in] item 1042 * Item specification. 1043 * @param[in] item_flags 1044 * Bit-fields that holds the items detected until now. 1045 * @param[out] error 1046 * Pointer to error structure. 1047 * 1048 * @return 1049 * 0 on success, a negative errno value otherwise and rte_errno is set. 1050 */ 1051 int 1052 mlx5_flow_validate_item_eth(const struct rte_flow_item *item, 1053 uint64_t item_flags, 1054 struct rte_flow_error *error) 1055 { 1056 const struct rte_flow_item_eth *mask = item->mask; 1057 const struct rte_flow_item_eth nic_mask = { 1058 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff", 1059 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff", 1060 .type = RTE_BE16(0xffff), 1061 }; 1062 int ret; 1063 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1064 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 : 1065 MLX5_FLOW_LAYER_OUTER_L2; 1066 1067 if (item_flags & ethm) 1068 return rte_flow_error_set(error, ENOTSUP, 1069 RTE_FLOW_ERROR_TYPE_ITEM, item, 1070 "multiple L2 layers not supported"); 1071 if (!mask) 1072 mask = &rte_flow_item_eth_mask; 1073 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, 1074 (const uint8_t *)&nic_mask, 1075 sizeof(struct rte_flow_item_eth), 1076 error); 1077 return ret; 1078 } 1079 1080 /** 1081 * Validate VLAN item. 1082 * 1083 * @param[in] item 1084 * Item specification. 1085 * @param[in] item_flags 1086 * Bit-fields that holds the items detected until now. 1087 * @param[out] error 1088 * Pointer to error structure. 1089 * 1090 * @return 1091 * 0 on success, a negative errno value otherwise and rte_errno is set. 1092 */ 1093 int 1094 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item, 1095 uint64_t item_flags, 1096 struct rte_flow_error *error) 1097 { 1098 const struct rte_flow_item_vlan *spec = item->spec; 1099 const struct rte_flow_item_vlan *mask = item->mask; 1100 const struct rte_flow_item_vlan nic_mask = { 1101 .tci = RTE_BE16(UINT16_MAX), 1102 .inner_type = RTE_BE16(UINT16_MAX), 1103 }; 1104 uint16_t vlan_tag = 0; 1105 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1106 int ret; 1107 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 | 1108 MLX5_FLOW_LAYER_INNER_L4) : 1109 (MLX5_FLOW_LAYER_OUTER_L3 | 1110 MLX5_FLOW_LAYER_OUTER_L4); 1111 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN : 1112 MLX5_FLOW_LAYER_OUTER_VLAN; 1113 1114 if (item_flags & vlanm) 1115 return rte_flow_error_set(error, EINVAL, 1116 RTE_FLOW_ERROR_TYPE_ITEM, item, 1117 "multiple VLAN layers not supported"); 1118 else if ((item_flags & l34m) != 0) 1119 return rte_flow_error_set(error, EINVAL, 1120 RTE_FLOW_ERROR_TYPE_ITEM, item, 1121 "L2 layer cannot follow L3/L4 layer"); 1122 if (!mask) 1123 mask = &rte_flow_item_vlan_mask; 1124 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, 1125 (const uint8_t *)&nic_mask, 1126 sizeof(struct rte_flow_item_vlan), 1127 error); 1128 if (ret) 1129 return ret; 1130 if (spec) { 1131 vlan_tag = spec->tci; 1132 vlan_tag &= mask->tci; 1133 } 1134 /* 1135 * From verbs perspective an empty VLAN is equivalent 1136 * to a packet without VLAN layer. 1137 */ 1138 if (!vlan_tag) 1139 return rte_flow_error_set(error, EINVAL, 1140 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, 1141 item->spec, 1142 "VLAN cannot be empty"); 1143 return 0; 1144 } 1145 1146 /** 1147 * Validate IPV4 item. 1148 * 1149 * @param[in] item 1150 * Item specification. 1151 * @param[in] item_flags 1152 * Bit-fields that holds the items detected until now. 1153 * @param[out] error 1154 * Pointer to error structure. 1155 * 1156 * @return 1157 * 0 on success, a negative errno value otherwise and rte_errno is set. 1158 */ 1159 int 1160 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item, 1161 uint64_t item_flags, 1162 struct rte_flow_error *error) 1163 { 1164 const struct rte_flow_item_ipv4 *mask = item->mask; 1165 const struct rte_flow_item_ipv4 nic_mask = { 1166 .hdr = { 1167 .src_addr = RTE_BE32(0xffffffff), 1168 .dst_addr = RTE_BE32(0xffffffff), 1169 .type_of_service = 0xff, 1170 .next_proto_id = 0xff, 1171 }, 1172 }; 1173 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1174 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 : 1175 MLX5_FLOW_LAYER_OUTER_L3; 1176 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 : 1177 MLX5_FLOW_LAYER_OUTER_L4; 1178 int ret; 1179 1180 if (item_flags & l3m) 1181 return rte_flow_error_set(error, ENOTSUP, 1182 RTE_FLOW_ERROR_TYPE_ITEM, item, 1183 "multiple L3 layers not supported"); 1184 else if (item_flags & l4m) 1185 return rte_flow_error_set(error, EINVAL, 1186 RTE_FLOW_ERROR_TYPE_ITEM, item, 1187 "L3 cannot follow an L4 layer."); 1188 if (!mask) 1189 mask = &rte_flow_item_ipv4_mask; 1190 else if (mask->hdr.next_proto_id != 0 && 1191 mask->hdr.next_proto_id != 0xff) 1192 return rte_flow_error_set(error, EINVAL, 1193 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask, 1194 "partial mask is not supported" 1195 " for protocol"); 1196 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, 1197 (const uint8_t *)&nic_mask, 1198 sizeof(struct rte_flow_item_ipv4), 1199 error); 1200 if (ret < 0) 1201 return ret; 1202 return 0; 1203 } 1204 1205 /** 1206 * Validate IPV6 item. 1207 * 1208 * @param[in] item 1209 * Item specification. 1210 * @param[in] item_flags 1211 * Bit-fields that holds the items detected until now. 1212 * @param[out] error 1213 * Pointer to error structure. 1214 * 1215 * @return 1216 * 0 on success, a negative errno value otherwise and rte_errno is set. 1217 */ 1218 int 1219 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item, 1220 uint64_t item_flags, 1221 struct rte_flow_error *error) 1222 { 1223 const struct rte_flow_item_ipv6 *mask = item->mask; 1224 const struct rte_flow_item_ipv6 nic_mask = { 1225 .hdr = { 1226 .src_addr = 1227 "\xff\xff\xff\xff\xff\xff\xff\xff" 1228 "\xff\xff\xff\xff\xff\xff\xff\xff", 1229 .dst_addr = 1230 "\xff\xff\xff\xff\xff\xff\xff\xff" 1231 "\xff\xff\xff\xff\xff\xff\xff\xff", 1232 .vtc_flow = RTE_BE32(0xffffffff), 1233 .proto = 0xff, 1234 .hop_limits = 0xff, 1235 }, 1236 }; 1237 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1238 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 : 1239 MLX5_FLOW_LAYER_OUTER_L3; 1240 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 : 1241 MLX5_FLOW_LAYER_OUTER_L4; 1242 int ret; 1243 1244 if (item_flags & l3m) 1245 return rte_flow_error_set(error, ENOTSUP, 1246 RTE_FLOW_ERROR_TYPE_ITEM, item, 1247 "multiple L3 layers not supported"); 1248 else if (item_flags & l4m) 1249 return rte_flow_error_set(error, EINVAL, 1250 RTE_FLOW_ERROR_TYPE_ITEM, item, 1251 "L3 cannot follow an L4 layer."); 1252 if (!mask) 1253 mask = &rte_flow_item_ipv6_mask; 1254 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, 1255 (const uint8_t *)&nic_mask, 1256 sizeof(struct rte_flow_item_ipv6), 1257 error); 1258 if (ret < 0) 1259 return ret; 1260 return 0; 1261 } 1262 1263 /** 1264 * Validate UDP item. 1265 * 1266 * @param[in] item 1267 * Item specification. 1268 * @param[in] item_flags 1269 * Bit-fields that holds the items detected until now. 1270 * @param[in] target_protocol 1271 * The next protocol in the previous item. 1272 * @param[in] flow_mask 1273 * mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask. 1274 * @param[out] error 1275 * Pointer to error structure. 1276 * 1277 * @return 1278 * 0 on success, a negative errno value otherwise and rte_errno is set. 1279 */ 1280 int 1281 mlx5_flow_validate_item_udp(const struct rte_flow_item *item, 1282 uint64_t item_flags, 1283 uint8_t target_protocol, 1284 struct rte_flow_error *error) 1285 { 1286 const struct rte_flow_item_udp *mask = item->mask; 1287 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1288 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 : 1289 MLX5_FLOW_LAYER_OUTER_L3; 1290 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 : 1291 MLX5_FLOW_LAYER_OUTER_L4; 1292 int ret; 1293 1294 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP) 1295 return rte_flow_error_set(error, EINVAL, 1296 RTE_FLOW_ERROR_TYPE_ITEM, item, 1297 "protocol filtering not compatible" 1298 " with UDP layer"); 1299 if (!(item_flags & l3m)) 1300 return rte_flow_error_set(error, EINVAL, 1301 RTE_FLOW_ERROR_TYPE_ITEM, item, 1302 "L3 is mandatory to filter on L4"); 1303 if (item_flags & l4m) 1304 return rte_flow_error_set(error, EINVAL, 1305 RTE_FLOW_ERROR_TYPE_ITEM, item, 1306 "multiple L4 layers not supported"); 1307 if (!mask) 1308 mask = &rte_flow_item_udp_mask; 1309 ret = mlx5_flow_item_acceptable 1310 (item, (const uint8_t *)mask, 1311 (const uint8_t *)&rte_flow_item_udp_mask, 1312 sizeof(struct rte_flow_item_udp), error); 1313 if (ret < 0) 1314 return ret; 1315 return 0; 1316 } 1317 1318 /** 1319 * Validate TCP item. 1320 * 1321 * @param[in] item 1322 * Item specification. 1323 * @param[in] item_flags 1324 * Bit-fields that holds the items detected until now. 1325 * @param[in] target_protocol 1326 * The next protocol in the previous item. 1327 * @param[out] error 1328 * Pointer to error structure. 1329 * 1330 * @return 1331 * 0 on success, a negative errno value otherwise and rte_errno is set. 1332 */ 1333 int 1334 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item, 1335 uint64_t item_flags, 1336 uint8_t target_protocol, 1337 const struct rte_flow_item_tcp *flow_mask, 1338 struct rte_flow_error *error) 1339 { 1340 const struct rte_flow_item_tcp *mask = item->mask; 1341 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL); 1342 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 : 1343 MLX5_FLOW_LAYER_OUTER_L3; 1344 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 : 1345 MLX5_FLOW_LAYER_OUTER_L4; 1346 int ret; 1347 1348 assert(flow_mask); 1349 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP) 1350 return rte_flow_error_set(error, EINVAL, 1351 RTE_FLOW_ERROR_TYPE_ITEM, item, 1352 "protocol filtering not compatible" 1353 " with TCP layer"); 1354 if (!(item_flags & l3m)) 1355 return rte_flow_error_set(error, EINVAL, 1356 RTE_FLOW_ERROR_TYPE_ITEM, item, 1357 "L3 is mandatory to filter on L4"); 1358 if (item_flags & l4m) 1359 return rte_flow_error_set(error, EINVAL, 1360 RTE_FLOW_ERROR_TYPE_ITEM, item, 1361 "multiple L4 layers not supported"); 1362 if (!mask) 1363 mask = &rte_flow_item_tcp_mask; 1364 ret = mlx5_flow_item_acceptable 1365 (item, (const uint8_t *)mask, 1366 (const uint8_t *)flow_mask, 1367 sizeof(struct rte_flow_item_tcp), error); 1368 if (ret < 0) 1369 return ret; 1370 return 0; 1371 } 1372 1373 /** 1374 * Validate VXLAN item. 1375 * 1376 * @param[in] item 1377 * Item specification. 1378 * @param[in] item_flags 1379 * Bit-fields that holds the items detected until now. 1380 * @param[in] target_protocol 1381 * The next protocol in the previous item. 1382 * @param[out] error 1383 * Pointer to error structure. 1384 * 1385 * @return 1386 * 0 on success, a negative errno value otherwise and rte_errno is set. 1387 */ 1388 int 1389 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item, 1390 uint64_t item_flags, 1391 struct rte_flow_error *error) 1392 { 1393 const struct rte_flow_item_vxlan *spec = item->spec; 1394 const struct rte_flow_item_vxlan *mask = item->mask; 1395 int ret; 1396 union vni { 1397 uint32_t vlan_id; 1398 uint8_t vni[4]; 1399 } id = { .vlan_id = 0, }; 1400 uint32_t vlan_id = 0; 1401 1402 1403 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) 1404 return rte_flow_error_set(error, ENOTSUP, 1405 RTE_FLOW_ERROR_TYPE_ITEM, item, 1406 "multiple tunnel layers not" 1407 " supported"); 1408 /* 1409 * Verify only UDPv4 is present as defined in 1410 * https://tools.ietf.org/html/rfc7348 1411 */ 1412 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)) 1413 return rte_flow_error_set(error, EINVAL, 1414 RTE_FLOW_ERROR_TYPE_ITEM, item, 1415 "no outer UDP layer found"); 1416 if (!mask) 1417 mask = &rte_flow_item_vxlan_mask; 1418 ret = mlx5_flow_item_acceptable 1419 (item, (const uint8_t *)mask, 1420 (const uint8_t *)&rte_flow_item_vxlan_mask, 1421 sizeof(struct rte_flow_item_vxlan), 1422 error); 1423 if (ret < 0) 1424 return ret; 1425 if (spec) { 1426 memcpy(&id.vni[1], spec->vni, 3); 1427 vlan_id = id.vlan_id; 1428 memcpy(&id.vni[1], mask->vni, 3); 1429 vlan_id &= id.vlan_id; 1430 } 1431 /* 1432 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if 1433 * only this layer is defined in the Verbs specification it is 1434 * interpreted as wildcard and all packets will match this 1435 * rule, if it follows a full stack layer (ex: eth / ipv4 / 1436 * udp), all packets matching the layers before will also 1437 * match this rule. To avoid such situation, VNI 0 is 1438 * currently refused. 1439 */ 1440 if (!vlan_id) 1441 return rte_flow_error_set(error, ENOTSUP, 1442 RTE_FLOW_ERROR_TYPE_ITEM, item, 1443 "VXLAN vni cannot be 0"); 1444 if (!(item_flags & MLX5_FLOW_LAYER_OUTER)) 1445 return rte_flow_error_set(error, ENOTSUP, 1446 RTE_FLOW_ERROR_TYPE_ITEM, item, 1447 "VXLAN tunnel must be fully defined"); 1448 return 0; 1449 } 1450 1451 /** 1452 * Validate VXLAN_GPE item. 1453 * 1454 * @param[in] item 1455 * Item specification. 1456 * @param[in] item_flags 1457 * Bit-fields that holds the items detected until now. 1458 * @param[in] priv 1459 * Pointer to the private data structure. 1460 * @param[in] target_protocol 1461 * The next protocol in the previous item. 1462 * @param[out] error 1463 * Pointer to error structure. 1464 * 1465 * @return 1466 * 0 on success, a negative errno value otherwise and rte_errno is set. 1467 */ 1468 int 1469 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item, 1470 uint64_t item_flags, 1471 struct rte_eth_dev *dev, 1472 struct rte_flow_error *error) 1473 { 1474 struct mlx5_priv *priv = dev->data->dev_private; 1475 const struct rte_flow_item_vxlan_gpe *spec = item->spec; 1476 const struct rte_flow_item_vxlan_gpe *mask = item->mask; 1477 int ret; 1478 union vni { 1479 uint32_t vlan_id; 1480 uint8_t vni[4]; 1481 } id = { .vlan_id = 0, }; 1482 uint32_t vlan_id = 0; 1483 1484 if (!priv->config.l3_vxlan_en) 1485 return rte_flow_error_set(error, ENOTSUP, 1486 RTE_FLOW_ERROR_TYPE_ITEM, item, 1487 "L3 VXLAN is not enabled by device" 1488 " parameter and/or not configured in" 1489 " firmware"); 1490 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) 1491 return rte_flow_error_set(error, ENOTSUP, 1492 RTE_FLOW_ERROR_TYPE_ITEM, item, 1493 "multiple tunnel layers not" 1494 " supported"); 1495 /* 1496 * Verify only UDPv4 is present as defined in 1497 * https://tools.ietf.org/html/rfc7348 1498 */ 1499 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)) 1500 return rte_flow_error_set(error, EINVAL, 1501 RTE_FLOW_ERROR_TYPE_ITEM, item, 1502 "no outer UDP layer found"); 1503 if (!mask) 1504 mask = &rte_flow_item_vxlan_gpe_mask; 1505 ret = mlx5_flow_item_acceptable 1506 (item, (const uint8_t *)mask, 1507 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask, 1508 sizeof(struct rte_flow_item_vxlan_gpe), 1509 error); 1510 if (ret < 0) 1511 return ret; 1512 if (spec) { 1513 if (spec->protocol) 1514 return rte_flow_error_set(error, ENOTSUP, 1515 RTE_FLOW_ERROR_TYPE_ITEM, 1516 item, 1517 "VxLAN-GPE protocol" 1518 " not supported"); 1519 memcpy(&id.vni[1], spec->vni, 3); 1520 vlan_id = id.vlan_id; 1521 memcpy(&id.vni[1], mask->vni, 3); 1522 vlan_id &= id.vlan_id; 1523 } 1524 /* 1525 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this 1526 * layer is defined in the Verbs specification it is interpreted as 1527 * wildcard and all packets will match this rule, if it follows a full 1528 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers 1529 * before will also match this rule. To avoid such situation, VNI 0 1530 * is currently refused. 1531 */ 1532 if (!vlan_id) 1533 return rte_flow_error_set(error, ENOTSUP, 1534 RTE_FLOW_ERROR_TYPE_ITEM, item, 1535 "VXLAN-GPE vni cannot be 0"); 1536 if (!(item_flags & MLX5_FLOW_LAYER_OUTER)) 1537 return rte_flow_error_set(error, ENOTSUP, 1538 RTE_FLOW_ERROR_TYPE_ITEM, item, 1539 "VXLAN-GPE tunnel must be fully" 1540 " defined"); 1541 return 0; 1542 } 1543 1544 /** 1545 * Validate GRE item. 1546 * 1547 * @param[in] item 1548 * Item specification. 1549 * @param[in] item_flags 1550 * Bit flags to mark detected items. 1551 * @param[in] target_protocol 1552 * The next protocol in the previous item. 1553 * @param[out] error 1554 * Pointer to error structure. 1555 * 1556 * @return 1557 * 0 on success, a negative errno value otherwise and rte_errno is set. 1558 */ 1559 int 1560 mlx5_flow_validate_item_gre(const struct rte_flow_item *item, 1561 uint64_t item_flags, 1562 uint8_t target_protocol, 1563 struct rte_flow_error *error) 1564 { 1565 const struct rte_flow_item_gre *spec __rte_unused = item->spec; 1566 const struct rte_flow_item_gre *mask = item->mask; 1567 int ret; 1568 1569 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE) 1570 return rte_flow_error_set(error, EINVAL, 1571 RTE_FLOW_ERROR_TYPE_ITEM, item, 1572 "protocol filtering not compatible" 1573 " with this GRE layer"); 1574 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) 1575 return rte_flow_error_set(error, ENOTSUP, 1576 RTE_FLOW_ERROR_TYPE_ITEM, item, 1577 "multiple tunnel layers not" 1578 " supported"); 1579 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3)) 1580 return rte_flow_error_set(error, ENOTSUP, 1581 RTE_FLOW_ERROR_TYPE_ITEM, item, 1582 "L3 Layer is missing"); 1583 if (!mask) 1584 mask = &rte_flow_item_gre_mask; 1585 ret = mlx5_flow_item_acceptable 1586 (item, (const uint8_t *)mask, 1587 (const uint8_t *)&rte_flow_item_gre_mask, 1588 sizeof(struct rte_flow_item_gre), error); 1589 if (ret < 0) 1590 return ret; 1591 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT 1592 if (spec && (spec->protocol & mask->protocol)) 1593 return rte_flow_error_set(error, ENOTSUP, 1594 RTE_FLOW_ERROR_TYPE_ITEM, item, 1595 "without MPLS support the" 1596 " specification cannot be used for" 1597 " filtering"); 1598 #endif 1599 return 0; 1600 } 1601 1602 /** 1603 * Validate MPLS item. 1604 * 1605 * @param[in] dev 1606 * Pointer to the rte_eth_dev structure. 1607 * @param[in] item 1608 * Item specification. 1609 * @param[in] item_flags 1610 * Bit-fields that holds the items detected until now. 1611 * @param[in] prev_layer 1612 * The protocol layer indicated in previous item. 1613 * @param[out] error 1614 * Pointer to error structure. 1615 * 1616 * @return 1617 * 0 on success, a negative errno value otherwise and rte_errno is set. 1618 */ 1619 int 1620 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused, 1621 const struct rte_flow_item *item __rte_unused, 1622 uint64_t item_flags __rte_unused, 1623 uint64_t prev_layer __rte_unused, 1624 struct rte_flow_error *error) 1625 { 1626 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT 1627 const struct rte_flow_item_mpls *mask = item->mask; 1628 struct mlx5_priv *priv = dev->data->dev_private; 1629 int ret; 1630 1631 if (!priv->config.mpls_en) 1632 return rte_flow_error_set(error, ENOTSUP, 1633 RTE_FLOW_ERROR_TYPE_ITEM, item, 1634 "MPLS not supported or" 1635 " disabled in firmware" 1636 " configuration."); 1637 /* MPLS over IP, UDP, GRE is allowed */ 1638 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 | 1639 MLX5_FLOW_LAYER_OUTER_L4_UDP | 1640 MLX5_FLOW_LAYER_GRE))) 1641 return rte_flow_error_set(error, EINVAL, 1642 RTE_FLOW_ERROR_TYPE_ITEM, item, 1643 "protocol filtering not compatible" 1644 " with MPLS layer"); 1645 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */ 1646 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) && 1647 !(item_flags & MLX5_FLOW_LAYER_GRE)) 1648 return rte_flow_error_set(error, ENOTSUP, 1649 RTE_FLOW_ERROR_TYPE_ITEM, item, 1650 "multiple tunnel layers not" 1651 " supported"); 1652 if (!mask) 1653 mask = &rte_flow_item_mpls_mask; 1654 ret = mlx5_flow_item_acceptable 1655 (item, (const uint8_t *)mask, 1656 (const uint8_t *)&rte_flow_item_mpls_mask, 1657 sizeof(struct rte_flow_item_mpls), error); 1658 if (ret < 0) 1659 return ret; 1660 return 0; 1661 #endif 1662 return rte_flow_error_set(error, ENOTSUP, 1663 RTE_FLOW_ERROR_TYPE_ITEM, item, 1664 "MPLS is not supported by Verbs, please" 1665 " update."); 1666 } 1667 1668 static int 1669 flow_null_validate(struct rte_eth_dev *dev __rte_unused, 1670 const struct rte_flow_attr *attr __rte_unused, 1671 const struct rte_flow_item items[] __rte_unused, 1672 const struct rte_flow_action actions[] __rte_unused, 1673 struct rte_flow_error *error) 1674 { 1675 return rte_flow_error_set(error, ENOTSUP, 1676 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL); 1677 } 1678 1679 static struct mlx5_flow * 1680 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused, 1681 const struct rte_flow_item items[] __rte_unused, 1682 const struct rte_flow_action actions[] __rte_unused, 1683 struct rte_flow_error *error) 1684 { 1685 rte_flow_error_set(error, ENOTSUP, 1686 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL); 1687 return NULL; 1688 } 1689 1690 static int 1691 flow_null_translate(struct rte_eth_dev *dev __rte_unused, 1692 struct mlx5_flow *dev_flow __rte_unused, 1693 const struct rte_flow_attr *attr __rte_unused, 1694 const struct rte_flow_item items[] __rte_unused, 1695 const struct rte_flow_action actions[] __rte_unused, 1696 struct rte_flow_error *error) 1697 { 1698 return rte_flow_error_set(error, ENOTSUP, 1699 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL); 1700 } 1701 1702 static int 1703 flow_null_apply(struct rte_eth_dev *dev __rte_unused, 1704 struct rte_flow *flow __rte_unused, 1705 struct rte_flow_error *error) 1706 { 1707 return rte_flow_error_set(error, ENOTSUP, 1708 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL); 1709 } 1710 1711 static void 1712 flow_null_remove(struct rte_eth_dev *dev __rte_unused, 1713 struct rte_flow *flow __rte_unused) 1714 { 1715 } 1716 1717 static void 1718 flow_null_destroy(struct rte_eth_dev *dev __rte_unused, 1719 struct rte_flow *flow __rte_unused) 1720 { 1721 } 1722 1723 static int 1724 flow_null_query(struct rte_eth_dev *dev __rte_unused, 1725 struct rte_flow *flow __rte_unused, 1726 const struct rte_flow_action *actions __rte_unused, 1727 void *data __rte_unused, 1728 struct rte_flow_error *error) 1729 { 1730 return rte_flow_error_set(error, ENOTSUP, 1731 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL); 1732 } 1733 1734 /* Void driver to protect from null pointer reference. */ 1735 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = { 1736 .validate = flow_null_validate, 1737 .prepare = flow_null_prepare, 1738 .translate = flow_null_translate, 1739 .apply = flow_null_apply, 1740 .remove = flow_null_remove, 1741 .destroy = flow_null_destroy, 1742 .query = flow_null_query, 1743 }; 1744 1745 /** 1746 * Select flow driver type according to flow attributes and device 1747 * configuration. 1748 * 1749 * @param[in] dev 1750 * Pointer to the dev structure. 1751 * @param[in] attr 1752 * Pointer to the flow attributes. 1753 * 1754 * @return 1755 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise. 1756 */ 1757 static enum mlx5_flow_drv_type 1758 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr) 1759 { 1760 struct mlx5_priv *priv = dev->data->dev_private; 1761 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX; 1762 1763 if (attr->transfer) 1764 type = MLX5_FLOW_TYPE_TCF; 1765 else 1766 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV : 1767 MLX5_FLOW_TYPE_VERBS; 1768 return type; 1769 } 1770 1771 #define flow_get_drv_ops(type) flow_drv_ops[type] 1772 1773 /** 1774 * Flow driver validation API. This abstracts calling driver specific functions. 1775 * The type of flow driver is determined according to flow attributes. 1776 * 1777 * @param[in] dev 1778 * Pointer to the dev structure. 1779 * @param[in] attr 1780 * Pointer to the flow attributes. 1781 * @param[in] items 1782 * Pointer to the list of items. 1783 * @param[in] actions 1784 * Pointer to the list of actions. 1785 * @param[out] error 1786 * Pointer to the error structure. 1787 * 1788 * @return 1789 * 0 on success, a negative errno value otherwise and rte_errno is set. 1790 */ 1791 static inline int 1792 flow_drv_validate(struct rte_eth_dev *dev, 1793 const struct rte_flow_attr *attr, 1794 const struct rte_flow_item items[], 1795 const struct rte_flow_action actions[], 1796 struct rte_flow_error *error) 1797 { 1798 const struct mlx5_flow_driver_ops *fops; 1799 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr); 1800 1801 fops = flow_get_drv_ops(type); 1802 return fops->validate(dev, attr, items, actions, error); 1803 } 1804 1805 /** 1806 * Flow driver preparation API. This abstracts calling driver specific 1807 * functions. Parent flow (rte_flow) should have driver type (drv_type). It 1808 * calculates the size of memory required for device flow, allocates the memory, 1809 * initializes the device flow and returns the pointer. 1810 * 1811 * @note 1812 * This function initializes device flow structure such as dv, tcf or verbs in 1813 * struct mlx5_flow. However, it is caller's responsibility to initialize the 1814 * rest. For example, adding returning device flow to flow->dev_flow list and 1815 * setting backward reference to the flow should be done out of this function. 1816 * layers field is not filled either. 1817 * 1818 * @param[in] attr 1819 * Pointer to the flow attributes. 1820 * @param[in] items 1821 * Pointer to the list of items. 1822 * @param[in] actions 1823 * Pointer to the list of actions. 1824 * @param[out] error 1825 * Pointer to the error structure. 1826 * 1827 * @return 1828 * Pointer to device flow on success, otherwise NULL and rte_errno is set. 1829 */ 1830 static inline struct mlx5_flow * 1831 flow_drv_prepare(const struct rte_flow *flow, 1832 const struct rte_flow_attr *attr, 1833 const struct rte_flow_item items[], 1834 const struct rte_flow_action actions[], 1835 struct rte_flow_error *error) 1836 { 1837 const struct mlx5_flow_driver_ops *fops; 1838 enum mlx5_flow_drv_type type = flow->drv_type; 1839 1840 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1841 fops = flow_get_drv_ops(type); 1842 return fops->prepare(attr, items, actions, error); 1843 } 1844 1845 /** 1846 * Flow driver translation API. This abstracts calling driver specific 1847 * functions. Parent flow (rte_flow) should have driver type (drv_type). It 1848 * translates a generic flow into a driver flow. flow_drv_prepare() must 1849 * precede. 1850 * 1851 * @note 1852 * dev_flow->layers could be filled as a result of parsing during translation 1853 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled 1854 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion, 1855 * flow->actions could be overwritten even though all the expanded dev_flows 1856 * have the same actions. 1857 * 1858 * @param[in] dev 1859 * Pointer to the rte dev structure. 1860 * @param[in, out] dev_flow 1861 * Pointer to the mlx5 flow. 1862 * @param[in] attr 1863 * Pointer to the flow attributes. 1864 * @param[in] items 1865 * Pointer to the list of items. 1866 * @param[in] actions 1867 * Pointer to the list of actions. 1868 * @param[out] error 1869 * Pointer to the error structure. 1870 * 1871 * @return 1872 * 0 on success, a negative errno value otherwise and rte_errno is set. 1873 */ 1874 static inline int 1875 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow, 1876 const struct rte_flow_attr *attr, 1877 const struct rte_flow_item items[], 1878 const struct rte_flow_action actions[], 1879 struct rte_flow_error *error) 1880 { 1881 const struct mlx5_flow_driver_ops *fops; 1882 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type; 1883 1884 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1885 fops = flow_get_drv_ops(type); 1886 return fops->translate(dev, dev_flow, attr, items, actions, error); 1887 } 1888 1889 /** 1890 * Flow driver apply API. This abstracts calling driver specific functions. 1891 * Parent flow (rte_flow) should have driver type (drv_type). It applies 1892 * translated driver flows on to device. flow_drv_translate() must precede. 1893 * 1894 * @param[in] dev 1895 * Pointer to Ethernet device structure. 1896 * @param[in, out] flow 1897 * Pointer to flow structure. 1898 * @param[out] error 1899 * Pointer to error structure. 1900 * 1901 * @return 1902 * 0 on success, a negative errno value otherwise and rte_errno is set. 1903 */ 1904 static inline int 1905 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow, 1906 struct rte_flow_error *error) 1907 { 1908 const struct mlx5_flow_driver_ops *fops; 1909 enum mlx5_flow_drv_type type = flow->drv_type; 1910 1911 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1912 fops = flow_get_drv_ops(type); 1913 return fops->apply(dev, flow, error); 1914 } 1915 1916 /** 1917 * Flow driver remove API. This abstracts calling driver specific functions. 1918 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow 1919 * on device. All the resources of the flow should be freed by calling 1920 * flow_drv_destroy(). 1921 * 1922 * @param[in] dev 1923 * Pointer to Ethernet device. 1924 * @param[in, out] flow 1925 * Pointer to flow structure. 1926 */ 1927 static inline void 1928 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow) 1929 { 1930 const struct mlx5_flow_driver_ops *fops; 1931 enum mlx5_flow_drv_type type = flow->drv_type; 1932 1933 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1934 fops = flow_get_drv_ops(type); 1935 fops->remove(dev, flow); 1936 } 1937 1938 /** 1939 * Flow driver destroy API. This abstracts calling driver specific functions. 1940 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow 1941 * on device and releases resources of the flow. 1942 * 1943 * @param[in] dev 1944 * Pointer to Ethernet device. 1945 * @param[in, out] flow 1946 * Pointer to flow structure. 1947 */ 1948 static inline void 1949 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow) 1950 { 1951 const struct mlx5_flow_driver_ops *fops; 1952 enum mlx5_flow_drv_type type = flow->drv_type; 1953 1954 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX); 1955 fops = flow_get_drv_ops(type); 1956 fops->destroy(dev, flow); 1957 } 1958 1959 /** 1960 * Validate a flow supported by the NIC. 1961 * 1962 * @see rte_flow_validate() 1963 * @see rte_flow_ops 1964 */ 1965 int 1966 mlx5_flow_validate(struct rte_eth_dev *dev, 1967 const struct rte_flow_attr *attr, 1968 const struct rte_flow_item items[], 1969 const struct rte_flow_action actions[], 1970 struct rte_flow_error *error) 1971 { 1972 int ret; 1973 1974 ret = flow_drv_validate(dev, attr, items, actions, error); 1975 if (ret < 0) 1976 return ret; 1977 return 0; 1978 } 1979 1980 /** 1981 * Get RSS action from the action list. 1982 * 1983 * @param[in] actions 1984 * Pointer to the list of actions. 1985 * 1986 * @return 1987 * Pointer to the RSS action if exist, else return NULL. 1988 */ 1989 static const struct rte_flow_action_rss* 1990 flow_get_rss_action(const struct rte_flow_action actions[]) 1991 { 1992 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) { 1993 switch (actions->type) { 1994 case RTE_FLOW_ACTION_TYPE_RSS: 1995 return (const struct rte_flow_action_rss *) 1996 actions->conf; 1997 default: 1998 break; 1999 } 2000 } 2001 return NULL; 2002 } 2003 2004 static unsigned int 2005 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level) 2006 { 2007 const struct rte_flow_item *item; 2008 unsigned int has_vlan = 0; 2009 2010 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) { 2011 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) { 2012 has_vlan = 1; 2013 break; 2014 } 2015 } 2016 if (has_vlan) 2017 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN : 2018 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN; 2019 return rss_level < 2 ? MLX5_EXPANSION_ROOT : 2020 MLX5_EXPANSION_ROOT_OUTER; 2021 } 2022 2023 /** 2024 * Create a flow and add it to @p list. 2025 * 2026 * @param dev 2027 * Pointer to Ethernet device. 2028 * @param list 2029 * Pointer to a TAILQ flow list. 2030 * @param[in] attr 2031 * Flow rule attributes. 2032 * @param[in] items 2033 * Pattern specification (list terminated by the END pattern item). 2034 * @param[in] actions 2035 * Associated actions (list terminated by the END action). 2036 * @param[out] error 2037 * Perform verbose error reporting if not NULL. 2038 * 2039 * @return 2040 * A flow on success, NULL otherwise and rte_errno is set. 2041 */ 2042 static struct rte_flow * 2043 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list, 2044 const struct rte_flow_attr *attr, 2045 const struct rte_flow_item items[], 2046 const struct rte_flow_action actions[], 2047 struct rte_flow_error *error) 2048 { 2049 struct rte_flow *flow = NULL; 2050 struct mlx5_flow *dev_flow; 2051 const struct rte_flow_action_rss *rss; 2052 union { 2053 struct rte_flow_expand_rss buf; 2054 uint8_t buffer[2048]; 2055 } expand_buffer; 2056 struct rte_flow_expand_rss *buf = &expand_buffer.buf; 2057 int ret; 2058 uint32_t i; 2059 uint32_t flow_size; 2060 2061 ret = flow_drv_validate(dev, attr, items, actions, error); 2062 if (ret < 0) 2063 return NULL; 2064 flow_size = sizeof(struct rte_flow); 2065 rss = flow_get_rss_action(actions); 2066 if (rss) 2067 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t), 2068 sizeof(void *)); 2069 else 2070 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *)); 2071 flow = rte_calloc(__func__, 1, flow_size, 0); 2072 if (!flow) { 2073 rte_errno = ENOMEM; 2074 return NULL; 2075 } 2076 flow->drv_type = flow_get_drv_type(dev, attr); 2077 assert(flow->drv_type > MLX5_FLOW_TYPE_MIN && 2078 flow->drv_type < MLX5_FLOW_TYPE_MAX); 2079 flow->queue = (void *)(flow + 1); 2080 LIST_INIT(&flow->dev_flows); 2081 if (rss && rss->types) { 2082 unsigned int graph_root; 2083 2084 graph_root = find_graph_root(items, rss->level); 2085 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer), 2086 items, rss->types, 2087 mlx5_support_expansion, 2088 graph_root); 2089 assert(ret > 0 && 2090 (unsigned int)ret < sizeof(expand_buffer.buffer)); 2091 } else { 2092 buf->entries = 1; 2093 buf->entry[0].pattern = (void *)(uintptr_t)items; 2094 } 2095 for (i = 0; i < buf->entries; ++i) { 2096 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern, 2097 actions, error); 2098 if (!dev_flow) 2099 goto error; 2100 dev_flow->flow = flow; 2101 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next); 2102 ret = flow_drv_translate(dev, dev_flow, attr, 2103 buf->entry[i].pattern, 2104 actions, error); 2105 if (ret < 0) 2106 goto error; 2107 } 2108 if (dev->data->dev_started) { 2109 ret = flow_drv_apply(dev, flow, error); 2110 if (ret < 0) 2111 goto error; 2112 } 2113 TAILQ_INSERT_TAIL(list, flow, next); 2114 flow_rxq_flags_set(dev, flow); 2115 return flow; 2116 error: 2117 ret = rte_errno; /* Save rte_errno before cleanup. */ 2118 assert(flow); 2119 flow_drv_destroy(dev, flow); 2120 rte_free(flow); 2121 rte_errno = ret; /* Restore rte_errno. */ 2122 return NULL; 2123 } 2124 2125 /** 2126 * Create a flow. 2127 * 2128 * @see rte_flow_create() 2129 * @see rte_flow_ops 2130 */ 2131 struct rte_flow * 2132 mlx5_flow_create(struct rte_eth_dev *dev, 2133 const struct rte_flow_attr *attr, 2134 const struct rte_flow_item items[], 2135 const struct rte_flow_action actions[], 2136 struct rte_flow_error *error) 2137 { 2138 struct mlx5_priv *priv = dev->data->dev_private; 2139 2140 return flow_list_create(dev, &priv->flows, 2141 attr, items, actions, error); 2142 } 2143 2144 /** 2145 * Destroy a flow in a list. 2146 * 2147 * @param dev 2148 * Pointer to Ethernet device. 2149 * @param list 2150 * Pointer to a TAILQ flow list. 2151 * @param[in] flow 2152 * Flow to destroy. 2153 */ 2154 static void 2155 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list, 2156 struct rte_flow *flow) 2157 { 2158 /* 2159 * Update RX queue flags only if port is started, otherwise it is 2160 * already clean. 2161 */ 2162 if (dev->data->dev_started) 2163 flow_rxq_flags_trim(dev, flow); 2164 flow_drv_destroy(dev, flow); 2165 TAILQ_REMOVE(list, flow, next); 2166 rte_free(flow->fdir); 2167 rte_free(flow); 2168 } 2169 2170 /** 2171 * Destroy all flows. 2172 * 2173 * @param dev 2174 * Pointer to Ethernet device. 2175 * @param list 2176 * Pointer to a TAILQ flow list. 2177 */ 2178 void 2179 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list) 2180 { 2181 while (!TAILQ_EMPTY(list)) { 2182 struct rte_flow *flow; 2183 2184 flow = TAILQ_FIRST(list); 2185 flow_list_destroy(dev, list, flow); 2186 } 2187 } 2188 2189 /** 2190 * Remove all flows. 2191 * 2192 * @param dev 2193 * Pointer to Ethernet device. 2194 * @param list 2195 * Pointer to a TAILQ flow list. 2196 */ 2197 void 2198 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list) 2199 { 2200 struct rte_flow *flow; 2201 2202 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) 2203 flow_drv_remove(dev, flow); 2204 flow_rxq_flags_clear(dev); 2205 } 2206 2207 /** 2208 * Add all flows. 2209 * 2210 * @param dev 2211 * Pointer to Ethernet device. 2212 * @param list 2213 * Pointer to a TAILQ flow list. 2214 * 2215 * @return 2216 * 0 on success, a negative errno value otherwise and rte_errno is set. 2217 */ 2218 int 2219 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list) 2220 { 2221 struct rte_flow *flow; 2222 struct rte_flow_error error; 2223 int ret = 0; 2224 2225 TAILQ_FOREACH(flow, list, next) { 2226 ret = flow_drv_apply(dev, flow, &error); 2227 if (ret < 0) 2228 goto error; 2229 flow_rxq_flags_set(dev, flow); 2230 } 2231 return 0; 2232 error: 2233 ret = rte_errno; /* Save rte_errno before cleanup. */ 2234 mlx5_flow_stop(dev, list); 2235 rte_errno = ret; /* Restore rte_errno. */ 2236 return -rte_errno; 2237 } 2238 2239 /** 2240 * Verify the flow list is empty 2241 * 2242 * @param dev 2243 * Pointer to Ethernet device. 2244 * 2245 * @return the number of flows not released. 2246 */ 2247 int 2248 mlx5_flow_verify(struct rte_eth_dev *dev) 2249 { 2250 struct mlx5_priv *priv = dev->data->dev_private; 2251 struct rte_flow *flow; 2252 int ret = 0; 2253 2254 TAILQ_FOREACH(flow, &priv->flows, next) { 2255 DRV_LOG(DEBUG, "port %u flow %p still referenced", 2256 dev->data->port_id, (void *)flow); 2257 ++ret; 2258 } 2259 return ret; 2260 } 2261 2262 /** 2263 * Enable a control flow configured from the control plane. 2264 * 2265 * @param dev 2266 * Pointer to Ethernet device. 2267 * @param eth_spec 2268 * An Ethernet flow spec to apply. 2269 * @param eth_mask 2270 * An Ethernet flow mask to apply. 2271 * @param vlan_spec 2272 * A VLAN flow spec to apply. 2273 * @param vlan_mask 2274 * A VLAN flow mask to apply. 2275 * 2276 * @return 2277 * 0 on success, a negative errno value otherwise and rte_errno is set. 2278 */ 2279 int 2280 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev, 2281 struct rte_flow_item_eth *eth_spec, 2282 struct rte_flow_item_eth *eth_mask, 2283 struct rte_flow_item_vlan *vlan_spec, 2284 struct rte_flow_item_vlan *vlan_mask) 2285 { 2286 struct mlx5_priv *priv = dev->data->dev_private; 2287 const struct rte_flow_attr attr = { 2288 .ingress = 1, 2289 .priority = MLX5_FLOW_PRIO_RSVD, 2290 }; 2291 struct rte_flow_item items[] = { 2292 { 2293 .type = RTE_FLOW_ITEM_TYPE_ETH, 2294 .spec = eth_spec, 2295 .last = NULL, 2296 .mask = eth_mask, 2297 }, 2298 { 2299 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN : 2300 RTE_FLOW_ITEM_TYPE_END, 2301 .spec = vlan_spec, 2302 .last = NULL, 2303 .mask = vlan_mask, 2304 }, 2305 { 2306 .type = RTE_FLOW_ITEM_TYPE_END, 2307 }, 2308 }; 2309 uint16_t queue[priv->reta_idx_n]; 2310 struct rte_flow_action_rss action_rss = { 2311 .func = RTE_ETH_HASH_FUNCTION_DEFAULT, 2312 .level = 0, 2313 .types = priv->rss_conf.rss_hf, 2314 .key_len = priv->rss_conf.rss_key_len, 2315 .queue_num = priv->reta_idx_n, 2316 .key = priv->rss_conf.rss_key, 2317 .queue = queue, 2318 }; 2319 struct rte_flow_action actions[] = { 2320 { 2321 .type = RTE_FLOW_ACTION_TYPE_RSS, 2322 .conf = &action_rss, 2323 }, 2324 { 2325 .type = RTE_FLOW_ACTION_TYPE_END, 2326 }, 2327 }; 2328 struct rte_flow *flow; 2329 struct rte_flow_error error; 2330 unsigned int i; 2331 2332 if (!priv->reta_idx_n || !priv->rxqs_n) { 2333 rte_errno = EINVAL; 2334 return -rte_errno; 2335 } 2336 for (i = 0; i != priv->reta_idx_n; ++i) 2337 queue[i] = (*priv->reta_idx)[i]; 2338 flow = flow_list_create(dev, &priv->ctrl_flows, 2339 &attr, items, actions, &error); 2340 if (!flow) 2341 return -rte_errno; 2342 return 0; 2343 } 2344 2345 /** 2346 * Enable a flow control configured from the control plane. 2347 * 2348 * @param dev 2349 * Pointer to Ethernet device. 2350 * @param eth_spec 2351 * An Ethernet flow spec to apply. 2352 * @param eth_mask 2353 * An Ethernet flow mask to apply. 2354 * 2355 * @return 2356 * 0 on success, a negative errno value otherwise and rte_errno is set. 2357 */ 2358 int 2359 mlx5_ctrl_flow(struct rte_eth_dev *dev, 2360 struct rte_flow_item_eth *eth_spec, 2361 struct rte_flow_item_eth *eth_mask) 2362 { 2363 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL); 2364 } 2365 2366 /** 2367 * Destroy a flow. 2368 * 2369 * @see rte_flow_destroy() 2370 * @see rte_flow_ops 2371 */ 2372 int 2373 mlx5_flow_destroy(struct rte_eth_dev *dev, 2374 struct rte_flow *flow, 2375 struct rte_flow_error *error __rte_unused) 2376 { 2377 struct mlx5_priv *priv = dev->data->dev_private; 2378 2379 flow_list_destroy(dev, &priv->flows, flow); 2380 return 0; 2381 } 2382 2383 /** 2384 * Destroy all flows. 2385 * 2386 * @see rte_flow_flush() 2387 * @see rte_flow_ops 2388 */ 2389 int 2390 mlx5_flow_flush(struct rte_eth_dev *dev, 2391 struct rte_flow_error *error __rte_unused) 2392 { 2393 struct mlx5_priv *priv = dev->data->dev_private; 2394 2395 mlx5_flow_list_flush(dev, &priv->flows); 2396 return 0; 2397 } 2398 2399 /** 2400 * Isolated mode. 2401 * 2402 * @see rte_flow_isolate() 2403 * @see rte_flow_ops 2404 */ 2405 int 2406 mlx5_flow_isolate(struct rte_eth_dev *dev, 2407 int enable, 2408 struct rte_flow_error *error) 2409 { 2410 struct mlx5_priv *priv = dev->data->dev_private; 2411 2412 if (dev->data->dev_started) { 2413 rte_flow_error_set(error, EBUSY, 2414 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, 2415 NULL, 2416 "port must be stopped first"); 2417 return -rte_errno; 2418 } 2419 priv->isolated = !!enable; 2420 if (enable) 2421 dev->dev_ops = &mlx5_dev_ops_isolate; 2422 else 2423 dev->dev_ops = &mlx5_dev_ops; 2424 return 0; 2425 } 2426 2427 /** 2428 * Query a flow. 2429 * 2430 * @see rte_flow_query() 2431 * @see rte_flow_ops 2432 */ 2433 static int 2434 flow_drv_query(struct rte_eth_dev *dev, 2435 struct rte_flow *flow, 2436 const struct rte_flow_action *actions, 2437 void *data, 2438 struct rte_flow_error *error) 2439 { 2440 const struct mlx5_flow_driver_ops *fops; 2441 enum mlx5_flow_drv_type ftype = flow->drv_type; 2442 2443 assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX); 2444 fops = flow_get_drv_ops(ftype); 2445 2446 return fops->query(dev, flow, actions, data, error); 2447 } 2448 2449 /** 2450 * Query a flow. 2451 * 2452 * @see rte_flow_query() 2453 * @see rte_flow_ops 2454 */ 2455 int 2456 mlx5_flow_query(struct rte_eth_dev *dev, 2457 struct rte_flow *flow, 2458 const struct rte_flow_action *actions, 2459 void *data, 2460 struct rte_flow_error *error) 2461 { 2462 int ret; 2463 2464 ret = flow_drv_query(dev, flow, actions, data, error); 2465 if (ret < 0) 2466 return ret; 2467 return 0; 2468 } 2469 2470 /** 2471 * Convert a flow director filter to a generic flow. 2472 * 2473 * @param dev 2474 * Pointer to Ethernet device. 2475 * @param fdir_filter 2476 * Flow director filter to add. 2477 * @param attributes 2478 * Generic flow parameters structure. 2479 * 2480 * @return 2481 * 0 on success, a negative errno value otherwise and rte_errno is set. 2482 */ 2483 static int 2484 flow_fdir_filter_convert(struct rte_eth_dev *dev, 2485 const struct rte_eth_fdir_filter *fdir_filter, 2486 struct mlx5_fdir *attributes) 2487 { 2488 struct mlx5_priv *priv = dev->data->dev_private; 2489 const struct rte_eth_fdir_input *input = &fdir_filter->input; 2490 const struct rte_eth_fdir_masks *mask = 2491 &dev->data->dev_conf.fdir_conf.mask; 2492 2493 /* Validate queue number. */ 2494 if (fdir_filter->action.rx_queue >= priv->rxqs_n) { 2495 DRV_LOG(ERR, "port %u invalid queue number %d", 2496 dev->data->port_id, fdir_filter->action.rx_queue); 2497 rte_errno = EINVAL; 2498 return -rte_errno; 2499 } 2500 attributes->attr.ingress = 1; 2501 attributes->items[0] = (struct rte_flow_item) { 2502 .type = RTE_FLOW_ITEM_TYPE_ETH, 2503 .spec = &attributes->l2, 2504 .mask = &attributes->l2_mask, 2505 }; 2506 switch (fdir_filter->action.behavior) { 2507 case RTE_ETH_FDIR_ACCEPT: 2508 attributes->actions[0] = (struct rte_flow_action){ 2509 .type = RTE_FLOW_ACTION_TYPE_QUEUE, 2510 .conf = &attributes->queue, 2511 }; 2512 break; 2513 case RTE_ETH_FDIR_REJECT: 2514 attributes->actions[0] = (struct rte_flow_action){ 2515 .type = RTE_FLOW_ACTION_TYPE_DROP, 2516 }; 2517 break; 2518 default: 2519 DRV_LOG(ERR, "port %u invalid behavior %d", 2520 dev->data->port_id, 2521 fdir_filter->action.behavior); 2522 rte_errno = ENOTSUP; 2523 return -rte_errno; 2524 } 2525 attributes->queue.index = fdir_filter->action.rx_queue; 2526 /* Handle L3. */ 2527 switch (fdir_filter->input.flow_type) { 2528 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP: 2529 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP: 2530 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER: 2531 attributes->l3.ipv4.hdr = (struct ipv4_hdr){ 2532 .src_addr = input->flow.ip4_flow.src_ip, 2533 .dst_addr = input->flow.ip4_flow.dst_ip, 2534 .time_to_live = input->flow.ip4_flow.ttl, 2535 .type_of_service = input->flow.ip4_flow.tos, 2536 }; 2537 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){ 2538 .src_addr = mask->ipv4_mask.src_ip, 2539 .dst_addr = mask->ipv4_mask.dst_ip, 2540 .time_to_live = mask->ipv4_mask.ttl, 2541 .type_of_service = mask->ipv4_mask.tos, 2542 .next_proto_id = mask->ipv4_mask.proto, 2543 }; 2544 attributes->items[1] = (struct rte_flow_item){ 2545 .type = RTE_FLOW_ITEM_TYPE_IPV4, 2546 .spec = &attributes->l3, 2547 .mask = &attributes->l3_mask, 2548 }; 2549 break; 2550 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP: 2551 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP: 2552 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER: 2553 attributes->l3.ipv6.hdr = (struct ipv6_hdr){ 2554 .hop_limits = input->flow.ipv6_flow.hop_limits, 2555 .proto = input->flow.ipv6_flow.proto, 2556 }; 2557 2558 memcpy(attributes->l3.ipv6.hdr.src_addr, 2559 input->flow.ipv6_flow.src_ip, 2560 RTE_DIM(attributes->l3.ipv6.hdr.src_addr)); 2561 memcpy(attributes->l3.ipv6.hdr.dst_addr, 2562 input->flow.ipv6_flow.dst_ip, 2563 RTE_DIM(attributes->l3.ipv6.hdr.src_addr)); 2564 memcpy(attributes->l3_mask.ipv6.hdr.src_addr, 2565 mask->ipv6_mask.src_ip, 2566 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr)); 2567 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr, 2568 mask->ipv6_mask.dst_ip, 2569 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr)); 2570 attributes->items[1] = (struct rte_flow_item){ 2571 .type = RTE_FLOW_ITEM_TYPE_IPV6, 2572 .spec = &attributes->l3, 2573 .mask = &attributes->l3_mask, 2574 }; 2575 break; 2576 default: 2577 DRV_LOG(ERR, "port %u invalid flow type%d", 2578 dev->data->port_id, fdir_filter->input.flow_type); 2579 rte_errno = ENOTSUP; 2580 return -rte_errno; 2581 } 2582 /* Handle L4. */ 2583 switch (fdir_filter->input.flow_type) { 2584 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP: 2585 attributes->l4.udp.hdr = (struct udp_hdr){ 2586 .src_port = input->flow.udp4_flow.src_port, 2587 .dst_port = input->flow.udp4_flow.dst_port, 2588 }; 2589 attributes->l4_mask.udp.hdr = (struct udp_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_UDP, 2595 .spec = &attributes->l4, 2596 .mask = &attributes->l4_mask, 2597 }; 2598 break; 2599 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP: 2600 attributes->l4.tcp.hdr = (struct tcp_hdr){ 2601 .src_port = input->flow.tcp4_flow.src_port, 2602 .dst_port = input->flow.tcp4_flow.dst_port, 2603 }; 2604 attributes->l4_mask.tcp.hdr = (struct tcp_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_TCP, 2610 .spec = &attributes->l4, 2611 .mask = &attributes->l4_mask, 2612 }; 2613 break; 2614 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP: 2615 attributes->l4.udp.hdr = (struct udp_hdr){ 2616 .src_port = input->flow.udp6_flow.src_port, 2617 .dst_port = input->flow.udp6_flow.dst_port, 2618 }; 2619 attributes->l4_mask.udp.hdr = (struct udp_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_UDP, 2625 .spec = &attributes->l4, 2626 .mask = &attributes->l4_mask, 2627 }; 2628 break; 2629 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP: 2630 attributes->l4.tcp.hdr = (struct tcp_hdr){ 2631 .src_port = input->flow.tcp6_flow.src_port, 2632 .dst_port = input->flow.tcp6_flow.dst_port, 2633 }; 2634 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){ 2635 .src_port = mask->src_port_mask, 2636 .dst_port = mask->dst_port_mask, 2637 }; 2638 attributes->items[2] = (struct rte_flow_item){ 2639 .type = RTE_FLOW_ITEM_TYPE_TCP, 2640 .spec = &attributes->l4, 2641 .mask = &attributes->l4_mask, 2642 }; 2643 break; 2644 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER: 2645 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER: 2646 break; 2647 default: 2648 DRV_LOG(ERR, "port %u invalid flow type%d", 2649 dev->data->port_id, fdir_filter->input.flow_type); 2650 rte_errno = ENOTSUP; 2651 return -rte_errno; 2652 } 2653 return 0; 2654 } 2655 2656 #define FLOW_FDIR_CMP(f1, f2, fld) \ 2657 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld)) 2658 2659 /** 2660 * Compare two FDIR flows. If items and actions are identical, the two flows are 2661 * regarded as same. 2662 * 2663 * @param dev 2664 * Pointer to Ethernet device. 2665 * @param f1 2666 * FDIR flow to compare. 2667 * @param f2 2668 * FDIR flow to compare. 2669 * 2670 * @return 2671 * Zero on match, 1 otherwise. 2672 */ 2673 static int 2674 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2) 2675 { 2676 if (FLOW_FDIR_CMP(f1, f2, attr) || 2677 FLOW_FDIR_CMP(f1, f2, l2) || 2678 FLOW_FDIR_CMP(f1, f2, l2_mask) || 2679 FLOW_FDIR_CMP(f1, f2, l3) || 2680 FLOW_FDIR_CMP(f1, f2, l3_mask) || 2681 FLOW_FDIR_CMP(f1, f2, l4) || 2682 FLOW_FDIR_CMP(f1, f2, l4_mask) || 2683 FLOW_FDIR_CMP(f1, f2, actions[0].type)) 2684 return 1; 2685 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE && 2686 FLOW_FDIR_CMP(f1, f2, queue)) 2687 return 1; 2688 return 0; 2689 } 2690 2691 /** 2692 * Search device flow list to find out a matched FDIR flow. 2693 * 2694 * @param dev 2695 * Pointer to Ethernet device. 2696 * @param fdir_flow 2697 * FDIR flow to lookup. 2698 * 2699 * @return 2700 * Pointer of flow if found, NULL otherwise. 2701 */ 2702 static struct rte_flow * 2703 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow) 2704 { 2705 struct mlx5_priv *priv = dev->data->dev_private; 2706 struct rte_flow *flow = NULL; 2707 2708 assert(fdir_flow); 2709 TAILQ_FOREACH(flow, &priv->flows, next) { 2710 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) { 2711 DRV_LOG(DEBUG, "port %u found FDIR flow %p", 2712 dev->data->port_id, (void *)flow); 2713 break; 2714 } 2715 } 2716 return flow; 2717 } 2718 2719 /** 2720 * Add new flow director filter and store it in list. 2721 * 2722 * @param dev 2723 * Pointer to Ethernet device. 2724 * @param fdir_filter 2725 * Flow director filter to add. 2726 * 2727 * @return 2728 * 0 on success, a negative errno value otherwise and rte_errno is set. 2729 */ 2730 static int 2731 flow_fdir_filter_add(struct rte_eth_dev *dev, 2732 const struct rte_eth_fdir_filter *fdir_filter) 2733 { 2734 struct mlx5_priv *priv = dev->data->dev_private; 2735 struct mlx5_fdir *fdir_flow; 2736 struct rte_flow *flow; 2737 int ret; 2738 2739 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0); 2740 if (!fdir_flow) { 2741 rte_errno = ENOMEM; 2742 return -rte_errno; 2743 } 2744 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow); 2745 if (ret) 2746 goto error; 2747 flow = flow_fdir_filter_lookup(dev, fdir_flow); 2748 if (flow) { 2749 rte_errno = EEXIST; 2750 goto error; 2751 } 2752 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr, 2753 fdir_flow->items, fdir_flow->actions, NULL); 2754 if (!flow) 2755 goto error; 2756 assert(!flow->fdir); 2757 flow->fdir = fdir_flow; 2758 DRV_LOG(DEBUG, "port %u created FDIR flow %p", 2759 dev->data->port_id, (void *)flow); 2760 return 0; 2761 error: 2762 rte_free(fdir_flow); 2763 return -rte_errno; 2764 } 2765 2766 /** 2767 * Delete specific filter. 2768 * 2769 * @param dev 2770 * Pointer to Ethernet device. 2771 * @param fdir_filter 2772 * Filter to be deleted. 2773 * 2774 * @return 2775 * 0 on success, a negative errno value otherwise and rte_errno is set. 2776 */ 2777 static int 2778 flow_fdir_filter_delete(struct rte_eth_dev *dev, 2779 const struct rte_eth_fdir_filter *fdir_filter) 2780 { 2781 struct mlx5_priv *priv = dev->data->dev_private; 2782 struct rte_flow *flow; 2783 struct mlx5_fdir fdir_flow = { 2784 .attr.group = 0, 2785 }; 2786 int ret; 2787 2788 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow); 2789 if (ret) 2790 return -rte_errno; 2791 flow = flow_fdir_filter_lookup(dev, &fdir_flow); 2792 if (!flow) { 2793 rte_errno = ENOENT; 2794 return -rte_errno; 2795 } 2796 flow_list_destroy(dev, &priv->flows, flow); 2797 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p", 2798 dev->data->port_id, (void *)flow); 2799 return 0; 2800 } 2801 2802 /** 2803 * Update queue for specific filter. 2804 * 2805 * @param dev 2806 * Pointer to Ethernet device. 2807 * @param fdir_filter 2808 * Filter to be updated. 2809 * 2810 * @return 2811 * 0 on success, a negative errno value otherwise and rte_errno is set. 2812 */ 2813 static int 2814 flow_fdir_filter_update(struct rte_eth_dev *dev, 2815 const struct rte_eth_fdir_filter *fdir_filter) 2816 { 2817 int ret; 2818 2819 ret = flow_fdir_filter_delete(dev, fdir_filter); 2820 if (ret) 2821 return ret; 2822 return flow_fdir_filter_add(dev, fdir_filter); 2823 } 2824 2825 /** 2826 * Flush all filters. 2827 * 2828 * @param dev 2829 * Pointer to Ethernet device. 2830 */ 2831 static void 2832 flow_fdir_filter_flush(struct rte_eth_dev *dev) 2833 { 2834 struct mlx5_priv *priv = dev->data->dev_private; 2835 2836 mlx5_flow_list_flush(dev, &priv->flows); 2837 } 2838 2839 /** 2840 * Get flow director information. 2841 * 2842 * @param dev 2843 * Pointer to Ethernet device. 2844 * @param[out] fdir_info 2845 * Resulting flow director information. 2846 */ 2847 static void 2848 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info) 2849 { 2850 struct rte_eth_fdir_masks *mask = 2851 &dev->data->dev_conf.fdir_conf.mask; 2852 2853 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode; 2854 fdir_info->guarant_spc = 0; 2855 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask)); 2856 fdir_info->max_flexpayload = 0; 2857 fdir_info->flow_types_mask[0] = 0; 2858 fdir_info->flex_payload_unit = 0; 2859 fdir_info->max_flex_payload_segment_num = 0; 2860 fdir_info->flex_payload_limit = 0; 2861 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf)); 2862 } 2863 2864 /** 2865 * Deal with flow director operations. 2866 * 2867 * @param dev 2868 * Pointer to Ethernet device. 2869 * @param filter_op 2870 * Operation to perform. 2871 * @param arg 2872 * Pointer to operation-specific structure. 2873 * 2874 * @return 2875 * 0 on success, a negative errno value otherwise and rte_errno is set. 2876 */ 2877 static int 2878 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op, 2879 void *arg) 2880 { 2881 enum rte_fdir_mode fdir_mode = 2882 dev->data->dev_conf.fdir_conf.mode; 2883 2884 if (filter_op == RTE_ETH_FILTER_NOP) 2885 return 0; 2886 if (fdir_mode != RTE_FDIR_MODE_PERFECT && 2887 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) { 2888 DRV_LOG(ERR, "port %u flow director mode %d not supported", 2889 dev->data->port_id, fdir_mode); 2890 rte_errno = EINVAL; 2891 return -rte_errno; 2892 } 2893 switch (filter_op) { 2894 case RTE_ETH_FILTER_ADD: 2895 return flow_fdir_filter_add(dev, arg); 2896 case RTE_ETH_FILTER_UPDATE: 2897 return flow_fdir_filter_update(dev, arg); 2898 case RTE_ETH_FILTER_DELETE: 2899 return flow_fdir_filter_delete(dev, arg); 2900 case RTE_ETH_FILTER_FLUSH: 2901 flow_fdir_filter_flush(dev); 2902 break; 2903 case RTE_ETH_FILTER_INFO: 2904 flow_fdir_info_get(dev, arg); 2905 break; 2906 default: 2907 DRV_LOG(DEBUG, "port %u unknown operation %u", 2908 dev->data->port_id, filter_op); 2909 rte_errno = EINVAL; 2910 return -rte_errno; 2911 } 2912 return 0; 2913 } 2914 2915 /** 2916 * Manage filter operations. 2917 * 2918 * @param dev 2919 * Pointer to Ethernet device structure. 2920 * @param filter_type 2921 * Filter type. 2922 * @param filter_op 2923 * Operation to perform. 2924 * @param arg 2925 * Pointer to operation-specific structure. 2926 * 2927 * @return 2928 * 0 on success, a negative errno value otherwise and rte_errno is set. 2929 */ 2930 int 2931 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev, 2932 enum rte_filter_type filter_type, 2933 enum rte_filter_op filter_op, 2934 void *arg) 2935 { 2936 switch (filter_type) { 2937 case RTE_ETH_FILTER_GENERIC: 2938 if (filter_op != RTE_ETH_FILTER_GET) { 2939 rte_errno = EINVAL; 2940 return -rte_errno; 2941 } 2942 *(const void **)arg = &mlx5_flow_ops; 2943 return 0; 2944 case RTE_ETH_FILTER_FDIR: 2945 return flow_fdir_ctrl_func(dev, filter_op, arg); 2946 default: 2947 DRV_LOG(ERR, "port %u filter type (%d) not supported", 2948 dev->data->port_id, filter_type); 2949 rte_errno = ENOTSUP; 2950 return -rte_errno; 2951 } 2952 return 0; 2953 } 2954