drivers/net/mlx5/mlx5_flow.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2016 6WIND S.A.
   3  * Copyright 2016 Mellanox Technologies, Ltd
   4  */
   5
   6 #include <sys/queue.h>
   7 #include <stdint.h>
   8 #include <string.h>
   9
  10 /* Verbs header. */
  11 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
  12 #ifdef PEDANTIC
  13 #pragma GCC diagnostic ignored "-Wpedantic"
  14 #endif
  15 #include <infiniband/verbs.h>
  16 #ifdef PEDANTIC
  17 #pragma GCC diagnostic error "-Wpedantic"
  18 #endif
  19
  20 #include <rte_common.h>
  21 #include <rte_ether.h>
  22 #include <rte_eth_ctrl.h>
  23 #include <rte_ethdev_driver.h>
  24 #include <rte_flow.h>
  25 #include <rte_flow_driver.h>
  26 #include <rte_malloc.h>
  27 #include <rte_ip.h>
  28
  29 #include "mlx5.h"
  30 #include "mlx5_defs.h"
  31 #include "mlx5_prm.h"
  32 #include "mlx5_glue.h"
  33
  34 /* Dev ops structure defined in mlx5.c */
  35 extern const struct eth_dev_ops mlx5_dev_ops;
  36 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
  37
  38 /* Pattern Layer bits. */
  39 #define MLX5_FLOW_LAYER_OUTER_L2 (1u << 0)
  40 #define MLX5_FLOW_LAYER_OUTER_L3_IPV4 (1u << 1)
  41 #define MLX5_FLOW_LAYER_OUTER_L3_IPV6 (1u << 2)
  42 #define MLX5_FLOW_LAYER_OUTER_L4_UDP (1u << 3)
  43 #define MLX5_FLOW_LAYER_OUTER_L4_TCP (1u << 4)
  44 #define MLX5_FLOW_LAYER_OUTER_VLAN (1u << 5)
  45 /* Masks. */
  46 #define MLX5_FLOW_LAYER_OUTER_L3 \
  47         (MLX5_FLOW_LAYER_OUTER_L3_IPV4 | MLX5_FLOW_LAYER_OUTER_L3_IPV6)
  48 #define MLX5_FLOW_LAYER_OUTER_L4 \
  49         (MLX5_FLOW_LAYER_OUTER_L4_UDP | MLX5_FLOW_LAYER_OUTER_L4_TCP)
  50
  51 /* Actions that modify the fate of matching traffic. */
  52 #define MLX5_FLOW_FATE_DROP (1u << 0)
  53 #define MLX5_FLOW_FATE_QUEUE (1u << 1)
  54
  55 /* possible L3 layers protocols filtering. */
  56 #define MLX5_IP_PROTOCOL_UDP 17
  57
  58 /** Handles information leading to a drop fate. */
  59 struct mlx5_flow_verbs {
  60         unsigned int size; /**< Size of the attribute. */
  61         struct {
  62                 struct ibv_flow_attr *attr;
  63                 /**< Pointer to the Specification buffer. */
  64                 uint8_t *specs; /**< Pointer to the specifications. */
  65         };
  66         struct ibv_flow *flow; /**< Verbs flow pointer. */
  67         struct mlx5_hrxq *hrxq; /**< Hash Rx queue object. */
  68 };
  69
  70 /* Flow structure. */
  71 struct rte_flow {
  72         TAILQ_ENTRY(rte_flow) next; /**< Pointer to the next flow structure. */
  73         struct rte_flow_attr attributes; /**< User flow attribute. */
  74         uint32_t l3_protocol_en:1; /**< Protocol filtering requested. */
  75         uint32_t layers;
  76         /**< Bit-fields of present layers see MLX5_FLOW_LAYER_*. */
  77         uint32_t fate;
  78         /**< Bit-fields of present fate see MLX5_FLOW_FATE_*. */
  79         uint8_t l3_protocol; /**< valid when l3_protocol_en is set. */
  80         struct mlx5_flow_verbs verbs; /* Verbs flow. */
  81         uint16_t queue; /**< Destination queue to redirect traffic to. */
  82 };
  83
  84 static const struct rte_flow_ops mlx5_flow_ops = {
  85         .validate = mlx5_flow_validate,
  86         .create = mlx5_flow_create,
  87         .destroy = mlx5_flow_destroy,
  88         .flush = mlx5_flow_flush,
  89         .isolate = mlx5_flow_isolate,
  90 };
  91
  92 /* Convert FDIR request to Generic flow. */
  93 struct mlx5_fdir {
  94         struct rte_flow_attr attr;
  95         struct rte_flow_action actions[2];
  96         struct rte_flow_item items[4];
  97         struct rte_flow_item_eth l2;
  98         struct rte_flow_item_eth l2_mask;
  99         union {
 100                 struct rte_flow_item_ipv4 ipv4;
 101                 struct rte_flow_item_ipv6 ipv6;
 102         } l3;
 103         union {
 104                 struct rte_flow_item_ipv4 ipv4;
 105                 struct rte_flow_item_ipv6 ipv6;
 106         } l3_mask;
 107         union {
 108                 struct rte_flow_item_udp udp;
 109                 struct rte_flow_item_tcp tcp;
 110         } l4;
 111         union {
 112                 struct rte_flow_item_udp udp;
 113                 struct rte_flow_item_tcp tcp;
 114         } l4_mask;
 115         struct rte_flow_action_queue queue;
 116 };
 117
 118 /* Verbs specification header. */
 119 struct ibv_spec_header {
 120         enum ibv_flow_spec_type type;
 121         uint16_t size;
 122 };
 123
 124  /**
 125   * Discover the maximum number of priority available.
 126   *
 127   * @param[in] dev
 128   *   Pointer to Ethernet device.
 129   *
 130   * @return
 131   *   number of supported flow priority on success, a negative errno value
 132   *   otherwise and rte_errno is set.
 133   */
 134 int
 135 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 136 {
 137         struct {
 138                 struct ibv_flow_attr attr;
 139                 struct ibv_flow_spec_eth eth;
 140                 struct ibv_flow_spec_action_drop drop;
 141         } flow_attr = {
 142                 .attr = {
 143                         .num_of_specs = 2,
 144                 },
 145                 .eth = {
 146                         .type = IBV_FLOW_SPEC_ETH,
 147                         .size = sizeof(struct ibv_flow_spec_eth),
 148                 },
 149                 .drop = {
 150                         .size = sizeof(struct ibv_flow_spec_action_drop),
 151                         .type = IBV_FLOW_SPEC_ACTION_DROP,
 152                 },
 153         };
 154         struct ibv_flow *flow;
 155         struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
 156         uint16_t vprio[] = { 8, 16 };
 157         int i;
 158
 159         if (!drop) {
 160                 rte_errno = ENOTSUP;
 161                 return -rte_errno;
 162         }
 163         for (i = 0; i != RTE_DIM(vprio); i++) {
 164                 flow_attr.attr.priority = vprio[i] - 1;
 165                 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 166                 if (!flow)
 167                         break;
 168                 claim_zero(mlx5_glue->destroy_flow(flow));
 169         }
 170         mlx5_hrxq_drop_release(dev);
 171         DRV_LOG(INFO, "port %u flow maximum priority: %d",
 172                 dev->data->port_id, vprio[i - 1]);
 173         return vprio[i - 1];
 174 }
 175
 176 /**
 177  * Verify the @p attributes will be correctly understood by the NIC and store
 178  * them in the @p flow if everything is correct.
 179  *
 180  * @param[in] dev
 181  *   Pointer to Ethernet device.
 182  * @param[in] attributes
 183  *   Pointer to flow attributes
 184  * @param[in, out] flow
 185  *   Pointer to the rte_flow structure.
 186  * @param[out] error
 187  *   Pointer to error structure.
 188  *
 189  * @return
 190  *   0 on success, a negative errno value otherwise and rte_errno is set.
 191  */
 192 static int
 193 mlx5_flow_attributes(struct rte_eth_dev *dev,
 194                      const struct rte_flow_attr *attributes,
 195                      struct rte_flow *flow,
 196                      struct rte_flow_error *error)
 197 {
 198         uint32_t priority_max =
 199                 ((struct priv *)dev->data->dev_private)->config.flow_prio;
 200
 201         if (attributes->group)
 202                 return rte_flow_error_set(error, ENOTSUP,
 203                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
 204                                           NULL,
 205                                           "groups is not supported");
 206         if (attributes->priority >= priority_max)
 207                 return rte_flow_error_set(error, ENOTSUP,
 208                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
 209                                           NULL,
 210                                           "priority out of range");
 211         if (attributes->egress)
 212                 return rte_flow_error_set(error, ENOTSUP,
 213                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
 214                                           NULL,
 215                                           "egress is not supported");
 216         if (attributes->transfer)
 217                 return rte_flow_error_set(error, ENOTSUP,
 218                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
 219                                           NULL,
 220                                           "transfer is not supported");
 221         if (!attributes->ingress)
 222                 return rte_flow_error_set(error, ENOTSUP,
 223                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
 224                                           NULL,
 225                                           "ingress attribute is mandatory");
 226         flow->attributes = *attributes;
 227         return 0;
 228 }
 229
 230 /**
 231  * Verify the @p item specifications (spec, last, mask) are compatible with the
 232  * NIC capabilities.
 233  *
 234  * @param[in] item
 235  *   Item specification.
 236  * @param[in] mask
 237  *   @p item->mask or flow default bit-masks.
 238  * @param[in] nic_mask
 239  *   Bit-masks covering supported fields by the NIC to compare with user mask.
 240  * @param[in] size
 241  *   Bit-masks size in bytes.
 242  * @param[out] error
 243  *   Pointer to error structure.
 244  *
 245  * @return
 246  *   0 on success, a negative errno value otherwise and rte_errno is set.
 247  */
 248 static int
 249 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
 250                           const uint8_t *mask,
 251                           const uint8_t *nic_mask,
 252                           unsigned int size,
 253                           struct rte_flow_error *error)
 254 {
 255         unsigned int i;
 256
 257         assert(nic_mask);
 258         for (i = 0; i < size; ++i)
 259                 if ((nic_mask[i] | mask[i]) != nic_mask[i])
 260                         return rte_flow_error_set(error, ENOTSUP,
 261                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 262                                                   item,
 263                                                   "mask enables non supported"
 264                                                   " bits");
 265         if (!item->spec && (item->mask || item->last))
 266                 return rte_flow_error_set(error, EINVAL,
 267                                           RTE_FLOW_ERROR_TYPE_ITEM,
 268                                           item,
 269                                           "mask/last without a spec is not"
 270                                           " supported");
 271         if (item->spec && item->last) {
 272                 uint8_t spec[size];
 273                 uint8_t last[size];
 274                 unsigned int i;
 275                 int ret;
 276
 277                 for (i = 0; i < size; ++i) {
 278                         spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
 279                         last[i] = ((const uint8_t *)item->last)[i] & mask[i];
 280                 }
 281                 ret = memcmp(spec, last, size);
 282                 if (ret != 0)
 283                         return rte_flow_error_set(error, ENOTSUP,
 284                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 285                                                   item,
 286                                                   "range is not supported");
 287         }
 288         return 0;
 289 }
 290
 291 /**
 292  * Add a verbs specification into @p flow.
 293  *
 294  * @param[in, out] flow
 295  *   Pointer to flow structure.
 296  * @param[in] src
 297  *   Create specification.
 298  * @param[in] size
 299  *   Size in bytes of the specification to copy.
 300  */
 301 static void
 302 mlx5_flow_spec_verbs_add(struct rte_flow *flow, void *src, unsigned int size)
 303 {
 304         if (flow->verbs.specs) {
 305                 void *dst;
 306
 307                 dst = (void *)(flow->verbs.specs + flow->verbs.size);
 308                 memcpy(dst, src, size);
 309                 ++flow->verbs.attr->num_of_specs;
 310         }
 311         flow->verbs.size += size;
 312 }
 313
 314 /**
 315  * Convert the @p item into a Verbs specification after ensuring the NIC
 316  * will understand and process it correctly.
 317  * If the necessary size for the conversion is greater than the @p flow_size,
 318  * nothing is written in @p flow, the validation is still performed.
 319  *
 320  * @param[in] item
 321  *   Item specification.
 322  * @param[in, out] flow
 323  *   Pointer to flow structure.
 324  * @param[in] flow_size
 325  *   Size in bytes of the available space in @p flow, if too small, nothing is
 326  *   written.
 327  * @param[out] error
 328  *   Pointer to error structure.
 329  *
 330  * @return
 331  *   On success the number of bytes consumed/necessary, if the returned value
 332  *   is lesser or equal to @p flow_size, the @p item has fully been converted,
 333  *   otherwise another call with this returned memory size should be done.
 334  *   On error, a negative errno value is returned and rte_errno is set.
 335  */
 336 static int
 337 mlx5_flow_item_eth(const struct rte_flow_item *item, struct rte_flow *flow,
 338                    const size_t flow_size, struct rte_flow_error *error)
 339 {
 340         const struct rte_flow_item_eth *spec = item->spec;
 341         const struct rte_flow_item_eth *mask = item->mask;
 342         const struct rte_flow_item_eth nic_mask = {
 343                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
 344                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
 345                 .type = RTE_BE16(0xffff),
 346         };
 347         const unsigned int size = sizeof(struct ibv_flow_spec_eth);
 348         struct ibv_flow_spec_eth eth = {
 349                 .type = IBV_FLOW_SPEC_ETH,
 350                 .size = size,
 351         };
 352         int ret;
 353
 354         if (flow->layers & MLX5_FLOW_LAYER_OUTER_L2)
 355                 return rte_flow_error_set(error, ENOTSUP,
 356                                           RTE_FLOW_ERROR_TYPE_ITEM,
 357                                           item,
 358                                           "L2 layers already configured");
 359         if (!mask)
 360                 mask = &rte_flow_item_eth_mask;
 361         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
 362                                         (const uint8_t *)&nic_mask,
 363                                         sizeof(struct rte_flow_item_eth),
 364                                         error);
 365         if (ret)
 366                 return ret;
 367         flow->layers |= MLX5_FLOW_LAYER_OUTER_L2;
 368         if (size > flow_size)
 369                 return size;
 370         if (spec) {
 371                 unsigned int i;
 372
 373                 memcpy(&eth.val.dst_mac, spec->dst.addr_bytes, ETHER_ADDR_LEN);
 374                 memcpy(&eth.val.src_mac, spec->src.addr_bytes, ETHER_ADDR_LEN);
 375                 eth.val.ether_type = spec->type;
 376                 memcpy(&eth.mask.dst_mac, mask->dst.addr_bytes, ETHER_ADDR_LEN);
 377                 memcpy(&eth.mask.src_mac, mask->src.addr_bytes, ETHER_ADDR_LEN);
 378                 eth.mask.ether_type = mask->type;
 379                 /* Remove unwanted bits from values. */
 380                 for (i = 0; i < ETHER_ADDR_LEN; ++i) {
 381                         eth.val.dst_mac[i] &= eth.mask.dst_mac[i];
 382                         eth.val.src_mac[i] &= eth.mask.src_mac[i];
 383                 }
 384                 eth.val.ether_type &= eth.mask.ether_type;
 385         }
 386         mlx5_flow_spec_verbs_add(flow, &eth, size);
 387         return size;
 388 }
 389
 390 /**
 391  * Update the VLAN tag in the Verbs Ethernet specification.
 392  *
 393  * @param[in, out] attr
 394  *   Pointer to Verbs attributes structure.
 395  * @param[in] eth
 396  *   Verbs structure containing the VLAN information to copy.
 397  */
 398 static void
 399 mlx5_flow_item_vlan_update(struct ibv_flow_attr *attr,
 400                            struct ibv_flow_spec_eth *eth)
 401 {
 402         unsigned int i;
 403         enum ibv_flow_spec_type search = IBV_FLOW_SPEC_ETH;
 404         struct ibv_spec_header *hdr = (struct ibv_spec_header *)
 405                 ((uint8_t *)attr + sizeof(struct ibv_flow_attr));
 406
 407         for (i = 0; i != attr->num_of_specs; ++i) {
 408                 if (hdr->type == search) {
 409                         struct ibv_flow_spec_eth *e =
 410                                 (struct ibv_flow_spec_eth *)hdr;
 411
 412                         e->val.vlan_tag = eth->val.vlan_tag;
 413                         e->mask.vlan_tag = eth->mask.vlan_tag;
 414                         e->val.ether_type = eth->val.ether_type;
 415                         e->mask.ether_type = eth->mask.ether_type;
 416                         break;
 417                 }
 418                 hdr = (struct ibv_spec_header *)((uint8_t *)hdr + hdr->size);
 419         }
 420 }
 421
 422 /**
 423  * Convert the @p item into @p flow (or by updating the already present
 424  * Ethernet Verbs) specification after ensuring the NIC will understand and
 425  * process it correctly.
 426  * If the necessary size for the conversion is greater than the @p flow_size,
 427  * nothing is written in @p flow, the validation is still performed.
 428  *
 429  * @param[in] item
 430  *   Item specification.
 431  * @param[in, out] flow
 432  *   Pointer to flow structure.
 433  * @param[in] flow_size
 434  *   Size in bytes of the available space in @p flow, if too small, nothing is
 435  *   written.
 436  * @param[out] error
 437  *   Pointer to error structure.
 438  *
 439  * @return
 440  *   On success the number of bytes consumed/necessary, if the returned value
 441  *   is lesser or equal to @p flow_size, the @p item has fully been converted,
 442  *   otherwise another call with this returned memory size should be done.
 443  *   On error, a negative errno value is returned and rte_errno is set.
 444  */
 445 static int
 446 mlx5_flow_item_vlan(const struct rte_flow_item *item, struct rte_flow *flow,
 447                     const size_t flow_size, struct rte_flow_error *error)
 448 {
 449         const struct rte_flow_item_vlan *spec = item->spec;
 450         const struct rte_flow_item_vlan *mask = item->mask;
 451         const struct rte_flow_item_vlan nic_mask = {
 452                 .tci = RTE_BE16(0x0fff),
 453                 .inner_type = RTE_BE16(0xffff),
 454         };
 455         unsigned int size = sizeof(struct ibv_flow_spec_eth);
 456         struct ibv_flow_spec_eth eth = {
 457                 .type = IBV_FLOW_SPEC_ETH,
 458                 .size = size,
 459         };
 460         int ret;
 461         const uint32_t l34m = MLX5_FLOW_LAYER_OUTER_L3 |
 462                         MLX5_FLOW_LAYER_OUTER_L4;
 463         const uint32_t vlanm = MLX5_FLOW_LAYER_OUTER_VLAN;
 464         const uint32_t l2m = MLX5_FLOW_LAYER_OUTER_L2;
 465
 466         if (flow->layers & vlanm)
 467                 return rte_flow_error_set(error, ENOTSUP,
 468                                           RTE_FLOW_ERROR_TYPE_ITEM,
 469                                           item,
 470                                           "VLAN layer already configured");
 471         else if ((flow->layers & l34m) != 0)
 472                 return rte_flow_error_set(error, ENOTSUP,
 473                                           RTE_FLOW_ERROR_TYPE_ITEM,
 474                                           item,
 475                                           "L2 layer cannot follow L3/L4 layer");
 476         if (!mask)
 477                 mask = &rte_flow_item_vlan_mask;
 478         ret = mlx5_flow_item_acceptable
 479                 (item, (const uint8_t *)mask,
 480                  (const uint8_t *)&nic_mask,
 481                  sizeof(struct rte_flow_item_vlan), error);
 482         if (ret)
 483                 return ret;
 484         if (spec) {
 485                 eth.val.vlan_tag = spec->tci;
 486                 eth.mask.vlan_tag = mask->tci;
 487                 eth.val.vlan_tag &= eth.mask.vlan_tag;
 488                 eth.val.ether_type = spec->inner_type;
 489                 eth.mask.ether_type = mask->inner_type;
 490                 eth.val.ether_type &= eth.mask.ether_type;
 491         }
 492         /*
 493          * From verbs perspective an empty VLAN is equivalent
 494          * to a packet without VLAN layer.
 495          */
 496         if (!eth.mask.vlan_tag)
 497                 return rte_flow_error_set(error, EINVAL,
 498                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
 499                                           item->spec,
 500                                           "VLAN cannot be empty");
 501         if (!(flow->layers & l2m)) {
 502                 if (size <= flow_size)
 503                         mlx5_flow_spec_verbs_add(flow, &eth, size);
 504         } else {
 505                 if (flow->verbs.attr)
 506                         mlx5_flow_item_vlan_update(flow->verbs.attr, &eth);
 507                 size = 0; /* Only an update is done in eth specification. */
 508         }
 509         flow->layers |= MLX5_FLOW_LAYER_OUTER_L2 |
 510                 MLX5_FLOW_LAYER_OUTER_VLAN;
 511         return size;
 512 }
 513
 514 /**
 515  * Convert the @p item into a Verbs specification after ensuring the NIC
 516  * will understand and process it correctly.
 517  * If the necessary size for the conversion is greater than the @p flow_size,
 518  * nothing is written in @p flow, the validation is still performed.
 519  *
 520  * @param[in] item
 521  *   Item specification.
 522  * @param[in, out] flow
 523  *   Pointer to flow structure.
 524  * @param[in] flow_size
 525  *   Size in bytes of the available space in @p flow, if too small, nothing is
 526  *   written.
 527  * @param[out] error
 528  *   Pointer to error structure.
 529  *
 530  * @return
 531  *   On success the number of bytes consumed/necessary, if the returned value
 532  *   is lesser or equal to @p flow_size, the @p item has fully been converted,
 533  *   otherwise another call with this returned memory size should be done.
 534  *   On error, a negative errno value is returned and rte_errno is set.
 535  */
 536 static int
 537 mlx5_flow_item_ipv4(const struct rte_flow_item *item, struct rte_flow *flow,
 538                     const size_t flow_size, struct rte_flow_error *error)
 539 {
 540         const struct rte_flow_item_ipv4 *spec = item->spec;
 541         const struct rte_flow_item_ipv4 *mask = item->mask;
 542         const struct rte_flow_item_ipv4 nic_mask = {
 543                 .hdr = {
 544                         .src_addr = RTE_BE32(0xffffffff),
 545                         .dst_addr = RTE_BE32(0xffffffff),
 546                         .type_of_service = 0xff,
 547                         .next_proto_id = 0xff,
 548                 },
 549         };
 550         unsigned int size = sizeof(struct ibv_flow_spec_ipv4_ext);
 551         struct ibv_flow_spec_ipv4_ext ipv4 = {
 552                 .type = IBV_FLOW_SPEC_IPV4_EXT,
 553                 .size = size,
 554         };
 555         int ret;
 556
 557         if (flow->layers & MLX5_FLOW_LAYER_OUTER_L3)
 558                 return rte_flow_error_set(error, ENOTSUP,
 559                                           RTE_FLOW_ERROR_TYPE_ITEM,
 560                                           item,
 561                                           "multiple L3 layers not supported");
 562         else if (flow->layers & MLX5_FLOW_LAYER_OUTER_L4)
 563                 return rte_flow_error_set(error, ENOTSUP,
 564                                           RTE_FLOW_ERROR_TYPE_ITEM,
 565                                           item,
 566                                           "L3 cannot follow an L4 layer.");
 567         if (!mask)
 568                 mask = &rte_flow_item_ipv4_mask;
 569         ret = mlx5_flow_item_acceptable
 570                 (item, (const uint8_t *)mask,
 571                  (const uint8_t *)&nic_mask,
 572                  sizeof(struct rte_flow_item_ipv4), error);
 573         if (ret < 0)
 574                 return ret;
 575         flow->layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
 576         if (spec) {
 577                 ipv4.val = (struct ibv_flow_ipv4_ext_filter){
 578                         .src_ip = spec->hdr.src_addr,
 579                         .dst_ip = spec->hdr.dst_addr,
 580                         .proto = spec->hdr.next_proto_id,
 581                         .tos = spec->hdr.type_of_service,
 582                 };
 583                 ipv4.mask = (struct ibv_flow_ipv4_ext_filter){
 584                         .src_ip = mask->hdr.src_addr,
 585                         .dst_ip = mask->hdr.dst_addr,
 586                         .proto = mask->hdr.next_proto_id,
 587                         .tos = mask->hdr.type_of_service,
 588                 };
 589                 /* Remove unwanted bits from values. */
 590                 ipv4.val.src_ip &= ipv4.mask.src_ip;
 591                 ipv4.val.dst_ip &= ipv4.mask.dst_ip;
 592                 ipv4.val.proto &= ipv4.mask.proto;
 593                 ipv4.val.tos &= ipv4.mask.tos;
 594         }
 595         flow->l3_protocol_en = !!ipv4.mask.proto;
 596         flow->l3_protocol = ipv4.val.proto;
 597         if (size <= flow_size)
 598                 mlx5_flow_spec_verbs_add(flow, &ipv4, size);
 599         return size;
 600 }
 601
 602 /**
 603  * Convert the @p item into a Verbs specification after ensuring the NIC
 604  * will understand and process it correctly.
 605  * If the necessary size for the conversion is greater than the @p flow_size,
 606  * nothing is written in @p flow, the validation is still performed.
 607  *
 608  * @param[in] item
 609  *   Item specification.
 610  * @param[in, out] flow
 611  *   Pointer to flow structure.
 612  * @param[in] flow_size
 613  *   Size in bytes of the available space in @p flow, if too small, nothing is
 614  *   written.
 615  * @param[out] error
 616  *   Pointer to error structure.
 617  *
 618  * @return
 619  *   On success the number of bytes consumed/necessary, if the returned value
 620  *   is lesser or equal to @p flow_size, the @p item has fully been converted,
 621  *   otherwise another call with this returned memory size should be done.
 622  *   On error, a negative errno value is returned and rte_errno is set.
 623  */
 624 static int
 625 mlx5_flow_item_ipv6(const struct rte_flow_item *item, struct rte_flow *flow,
 626                     const size_t flow_size, struct rte_flow_error *error)
 627 {
 628         const struct rte_flow_item_ipv6 *spec = item->spec;
 629         const struct rte_flow_item_ipv6 *mask = item->mask;
 630         const struct rte_flow_item_ipv6 nic_mask = {
 631                 .hdr = {
 632                         .src_addr =
 633                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
 634                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
 635                         .dst_addr =
 636                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
 637                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
 638                         .vtc_flow = RTE_BE32(0xffffffff),
 639                         .proto = 0xff,
 640                         .hop_limits = 0xff,
 641                 },
 642         };
 643         unsigned int size = sizeof(struct ibv_flow_spec_ipv6);
 644         struct ibv_flow_spec_ipv6 ipv6 = {
 645                 .type = IBV_FLOW_SPEC_IPV6,
 646                 .size = size,
 647         };
 648         int ret;
 649
 650         if (flow->layers & MLX5_FLOW_LAYER_OUTER_L3)
 651                 return rte_flow_error_set(error, ENOTSUP,
 652                                           RTE_FLOW_ERROR_TYPE_ITEM,
 653                                           item,
 654                                           "multiple L3 layers not supported");
 655         else if (flow->layers & MLX5_FLOW_LAYER_OUTER_L4)
 656                 return rte_flow_error_set(error, ENOTSUP,
 657                                           RTE_FLOW_ERROR_TYPE_ITEM,
 658                                           item,
 659                                           "L3 cannot follow an L4 layer.");
 660         if (!mask)
 661                 mask = &rte_flow_item_ipv6_mask;
 662         ret = mlx5_flow_item_acceptable
 663                 (item, (const uint8_t *)mask,
 664                  (const uint8_t *)&nic_mask,
 665                  sizeof(struct rte_flow_item_ipv6), error);
 666         if (ret < 0)
 667                 return ret;
 668         flow->layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
 669         if (spec) {
 670                 unsigned int i;
 671                 uint32_t vtc_flow_val;
 672                 uint32_t vtc_flow_mask;
 673
 674                 memcpy(&ipv6.val.src_ip, spec->hdr.src_addr,
 675                        RTE_DIM(ipv6.val.src_ip));
 676                 memcpy(&ipv6.val.dst_ip, spec->hdr.dst_addr,
 677                        RTE_DIM(ipv6.val.dst_ip));
 678                 memcpy(&ipv6.mask.src_ip, mask->hdr.src_addr,
 679                        RTE_DIM(ipv6.mask.src_ip));
 680                 memcpy(&ipv6.mask.dst_ip, mask->hdr.dst_addr,
 681                        RTE_DIM(ipv6.mask.dst_ip));
 682                 vtc_flow_val = rte_be_to_cpu_32(spec->hdr.vtc_flow);
 683                 vtc_flow_mask = rte_be_to_cpu_32(mask->hdr.vtc_flow);
 684                 ipv6.val.flow_label =
 685                         rte_cpu_to_be_32((vtc_flow_val & IPV6_HDR_FL_MASK) >>
 686                                          IPV6_HDR_FL_SHIFT);
 687                 ipv6.val.traffic_class = (vtc_flow_val & IPV6_HDR_TC_MASK) >>
 688                                          IPV6_HDR_TC_SHIFT;
 689                 ipv6.val.next_hdr = spec->hdr.proto;
 690                 ipv6.val.hop_limit = spec->hdr.hop_limits;
 691                 ipv6.mask.flow_label =
 692                         rte_cpu_to_be_32((vtc_flow_mask & IPV6_HDR_FL_MASK) >>
 693                                          IPV6_HDR_FL_SHIFT);
 694                 ipv6.mask.traffic_class = (vtc_flow_mask & IPV6_HDR_TC_MASK) >>
 695                                           IPV6_HDR_TC_SHIFT;
 696                 ipv6.mask.next_hdr = mask->hdr.proto;
 697                 ipv6.mask.hop_limit = mask->hdr.hop_limits;
 698                 /* Remove unwanted bits from values. */
 699                 for (i = 0; i < RTE_DIM(ipv6.val.src_ip); ++i) {
 700                         ipv6.val.src_ip[i] &= ipv6.mask.src_ip[i];
 701                         ipv6.val.dst_ip[i] &= ipv6.mask.dst_ip[i];
 702                 }
 703                 ipv6.val.flow_label &= ipv6.mask.flow_label;
 704                 ipv6.val.traffic_class &= ipv6.mask.traffic_class;
 705                 ipv6.val.next_hdr &= ipv6.mask.next_hdr;
 706                 ipv6.val.hop_limit &= ipv6.mask.hop_limit;
 707         }
 708         flow->l3_protocol_en = !!ipv6.mask.next_hdr;
 709         flow->l3_protocol = ipv6.val.next_hdr;
 710         if (size <= flow_size)
 711                 mlx5_flow_spec_verbs_add(flow, &ipv6, size);
 712         return size;
 713 }
 714
 715 /**
 716  * Convert the @p item into a Verbs specification after ensuring the NIC
 717  * will understand and process it correctly.
 718  * If the necessary size for the conversion is greater than the @p flow_size,
 719  * nothing is written in @p flow, the validation is still performed.
 720  *
 721  * @param[in] item
 722  *   Item specification.
 723  * @param[in, out] flow
 724  *   Pointer to flow structure.
 725  * @param[in] flow_size
 726  *   Size in bytes of the available space in @p flow, if too small, nothing is
 727  *   written.
 728  * @param[out] error
 729  *   Pointer to error structure.
 730  *
 731  * @return
 732  *   On success the number of bytes consumed/necessary, if the returned value
 733  *   is lesser or equal to @p flow_size, the @p item has fully been converted,
 734  *   otherwise another call with this returned memory size should be done.
 735  *   On error, a negative errno value is returned and rte_errno is set.
 736  */
 737 static int
 738 mlx5_flow_item_udp(const struct rte_flow_item *item, struct rte_flow *flow,
 739                    const size_t flow_size, struct rte_flow_error *error)
 740 {
 741         const struct rte_flow_item_udp *spec = item->spec;
 742         const struct rte_flow_item_udp *mask = item->mask;
 743         unsigned int size = sizeof(struct ibv_flow_spec_tcp_udp);
 744         struct ibv_flow_spec_tcp_udp udp = {
 745                 .type = IBV_FLOW_SPEC_UDP,
 746                 .size = size,
 747         };
 748         int ret;
 749
 750         if (!(flow->layers & MLX5_FLOW_LAYER_OUTER_L3))
 751                 return rte_flow_error_set(error, ENOTSUP,
 752                                           RTE_FLOW_ERROR_TYPE_ITEM,
 753                                           item,
 754                                           "L3 is mandatory to filter on L4");
 755         if (flow->layers & MLX5_FLOW_LAYER_OUTER_L4)
 756                 return rte_flow_error_set(error, ENOTSUP,
 757                                           RTE_FLOW_ERROR_TYPE_ITEM,
 758                                           item,
 759                                           "L4 layer is already present");
 760         if (flow->l3_protocol_en && flow->l3_protocol != MLX5_IP_PROTOCOL_UDP)
 761                 return rte_flow_error_set(error, ENOTSUP,
 762                                           RTE_FLOW_ERROR_TYPE_ITEM,
 763                                           item,
 764                                           "protocol filtering not compatible"
 765                                           " with UDP layer");
 766         if (!mask)
 767                 mask = &rte_flow_item_udp_mask;
 768         ret = mlx5_flow_item_acceptable
 769                 (item, (const uint8_t *)mask,
 770                  (const uint8_t *)&rte_flow_item_udp_mask,
 771                  sizeof(struct rte_flow_item_udp), error);
 772         if (ret < 0)
 773                 return ret;
 774         flow->layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
 775         if (size > flow_size)
 776                 return size;
 777         if (spec) {
 778                 udp.val.dst_port = spec->hdr.dst_port;
 779                 udp.val.src_port = spec->hdr.src_port;
 780                 udp.mask.dst_port = mask->hdr.dst_port;
 781                 udp.mask.src_port = mask->hdr.src_port;
 782                 /* Remove unwanted bits from values. */
 783                 udp.val.src_port &= udp.mask.src_port;
 784                 udp.val.dst_port &= udp.mask.dst_port;
 785         }
 786         mlx5_flow_spec_verbs_add(flow, &udp, size);
 787         return size;
 788 }
 789
 790 /**
 791  * Convert the @p pattern into a Verbs specifications after ensuring the NIC
 792  * will understand and process it correctly.
 793  * The conversion is performed item per item, each of them is written into
 794  * the @p flow if its size is lesser or equal to @p flow_size.
 795  * Validation and memory consumption computation are still performed until the
 796  * end of @p pattern, unless an error is encountered.
 797  *
 798  * @param[in] pattern
 799  *   Flow pattern.
 800  * @param[in, out] flow
 801  *   Pointer to the rte_flow structure.
 802  * @param[in] flow_size
 803  *   Size in bytes of the available space in @p flow, if too small some
 804  *   garbage may be present.
 805  * @param[out] error
 806  *   Pointer to error structure.
 807  *
 808  * @return
 809  *   On success the number of bytes consumed/necessary, if the returned value
 810  *   is lesser or equal to @p flow_size, the @pattern  has fully been
 811  *   converted, otherwise another call with this returned memory size should
 812  *   be done.
 813  *   On error, a negative errno value is returned and rte_errno is set.
 814  */
 815 static int
 816 mlx5_flow_items(const struct rte_flow_item pattern[],
 817                 struct rte_flow *flow, const size_t flow_size,
 818                 struct rte_flow_error *error)
 819 {
 820         int remain = flow_size;
 821         size_t size = 0;
 822
 823         for (; pattern->type != RTE_FLOW_ITEM_TYPE_END; pattern++) {
 824                 int ret = 0;
 825
 826                 switch (pattern->type) {
 827                 case RTE_FLOW_ITEM_TYPE_VOID:
 828                         break;
 829                 case RTE_FLOW_ITEM_TYPE_ETH:
 830                         ret = mlx5_flow_item_eth(pattern, flow, remain, error);
 831                         break;
 832                 case RTE_FLOW_ITEM_TYPE_VLAN:
 833                         ret = mlx5_flow_item_vlan(pattern, flow, remain, error);
 834                         break;
 835                 case RTE_FLOW_ITEM_TYPE_IPV4:
 836                         ret = mlx5_flow_item_ipv4(pattern, flow, remain, error);
 837                         break;
 838                 case RTE_FLOW_ITEM_TYPE_IPV6:
 839                         ret = mlx5_flow_item_ipv6(pattern, flow, remain, error);
 840                         break;
 841                 case RTE_FLOW_ITEM_TYPE_UDP:
 842                         ret = mlx5_flow_item_udp(pattern, flow, remain, error);
 843                         break;
 844                 default:
 845                         return rte_flow_error_set(error, ENOTSUP,
 846                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 847                                                   pattern,
 848                                                   "item not supported");
 849                 }
 850                 if (ret < 0)
 851                         return ret;
 852                 if (remain > ret)
 853                         remain -= ret;
 854                 else
 855                         remain = 0;
 856                 size += ret;
 857         }
 858         if (!flow->layers) {
 859                 const struct rte_flow_item item = {
 860                         .type = RTE_FLOW_ITEM_TYPE_ETH,
 861                 };
 862
 863                 return mlx5_flow_item_eth(&item, flow, flow_size, error);
 864         }
 865         return size;
 866 }
 867
 868 /**
 869  * Convert the @p action into a Verbs specification after ensuring the NIC
 870  * will understand and process it correctly.
 871  * If the necessary size for the conversion is greater than the @p flow_size,
 872  * nothing is written in @p flow, the validation is still performed.
 873  *
 874  * @param[in] action
 875  *   Action configuration.
 876  * @param[in, out] flow
 877  *   Pointer to flow structure.
 878  * @param[in] flow_size
 879  *   Size in bytes of the available space in @p flow, if too small, nothing is
 880  *   written.
 881  * @param[out] error
 882  *   Pointer to error structure.
 883  *
 884  * @return
 885  *   On success the number of bytes consumed/necessary, if the returned value
 886  *   is lesser or equal to @p flow_size, the @p action has fully been
 887  *   converted, otherwise another call with this returned memory size should
 888  *   be done.
 889  *   On error, a negative errno value is returned and rte_errno is set.
 890  */
 891 static int
 892 mlx5_flow_action_drop(const struct rte_flow_action *action,
 893                       struct rte_flow *flow, const size_t flow_size,
 894                       struct rte_flow_error *error)
 895 {
 896         unsigned int size = sizeof(struct ibv_flow_spec_action_drop);
 897         struct ibv_flow_spec_action_drop drop = {
 898                         .type = IBV_FLOW_SPEC_ACTION_DROP,
 899                         .size = size,
 900         };
 901
 902         if (flow->fate)
 903                 return rte_flow_error_set(error, ENOTSUP,
 904                                           RTE_FLOW_ERROR_TYPE_ACTION,
 905                                           action,
 906                                           "multiple fate actions are not"
 907                                           " supported");
 908         if (size < flow_size)
 909                 mlx5_flow_spec_verbs_add(flow, &drop, size);
 910         flow->fate |= MLX5_FLOW_FATE_DROP;
 911         return size;
 912 }
 913
 914 /**
 915  * Convert the @p action into @p flow after ensuring the NIC will understand
 916  * and process it correctly.
 917  *
 918  * @param[in] dev
 919  *   Pointer to Ethernet device structure.
 920  * @param[in] action
 921  *   Action configuration.
 922  * @param[in, out] flow
 923  *   Pointer to flow structure.
 924  * @param[out] error
 925  *   Pointer to error structure.
 926  *
 927  * @return
 928  *   0 on success, a negative errno value otherwise and rte_errno is set.
 929  */
 930 static int
 931 mlx5_flow_action_queue(struct rte_eth_dev *dev,
 932                        const struct rte_flow_action *action,
 933                        struct rte_flow *flow,
 934                        struct rte_flow_error *error)
 935 {
 936         struct priv *priv = dev->data->dev_private;
 937         const struct rte_flow_action_queue *queue = action->conf;
 938
 939         if (flow->fate)
 940                 return rte_flow_error_set(error, ENOTSUP,
 941                                           RTE_FLOW_ERROR_TYPE_ACTION,
 942                                           action,
 943                                           "multiple fate actions are not"
 944                                           " supported");
 945         if (queue->index >= priv->rxqs_n)
 946                 return rte_flow_error_set(error, EINVAL,
 947                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 948                                           &queue->index,
 949                                           "queue index out of range");
 950         if (!(*priv->rxqs)[queue->index])
 951                 return rte_flow_error_set(error, EINVAL,
 952                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 953                                           &queue->index,
 954                                           "queue is not configured");
 955         flow->queue = queue->index;
 956         flow->fate |= MLX5_FLOW_FATE_QUEUE;
 957         return 0;
 958 }
 959
 960 /**
 961  * Convert the @p action into @p flow after ensuring the NIC will understand
 962  * and process it correctly.
 963  * The conversion is performed action per action, each of them is written into
 964  * the @p flow if its size is lesser or equal to @p flow_size.
 965  * Validation and memory consumption computation are still performed until the
 966  * end of @p action, unless an error is encountered.
 967  *
 968  * @param[in] dev
 969  *   Pointer to Ethernet device structure.
 970  * @param[in] actions
 971  *   Pointer to flow actions array.
 972  * @param[in, out] flow
 973  *   Pointer to the rte_flow structure.
 974  * @param[in] flow_size
 975  *   Size in bytes of the available space in @p flow, if too small some
 976  *   garbage may be present.
 977  * @param[out] error
 978  *   Pointer to error structure.
 979  *
 980  * @return
 981  *   On success the number of bytes consumed/necessary, if the returned value
 982  *   is lesser or equal to @p flow_size, the @p actions has fully been
 983  *   converted, otherwise another call with this returned memory size should
 984  *   be done.
 985  *   On error, a negative errno value is returned and rte_errno is set.
 986  */
 987 static int
 988 mlx5_flow_actions(struct rte_eth_dev *dev,
 989                   const struct rte_flow_action actions[],
 990                   struct rte_flow *flow, const size_t flow_size,
 991                   struct rte_flow_error *error)
 992 {
 993         size_t size = 0;
 994         int remain = flow_size;
 995         int ret = 0;
 996
 997         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
 998                 switch (actions->type) {
 999                 case RTE_FLOW_ACTION_TYPE_VOID:
1000                         break;
1001                 case RTE_FLOW_ACTION_TYPE_DROP:
1002                         ret = mlx5_flow_action_drop(actions, flow, remain,
1003                                                     error);
1004                         break;
1005                 case RTE_FLOW_ACTION_TYPE_QUEUE:
1006                         ret = mlx5_flow_action_queue(dev, actions, flow, error);
1007                         break;
1008                 default:
1009                         return rte_flow_error_set(error, ENOTSUP,
1010                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1011                                                   actions,
1012                                                   "action not supported");
1013                 }
1014                 if (ret < 0)
1015                         return ret;
1016                 if (remain > ret)
1017                         remain -= ret;
1018                 else
1019                         remain = 0;
1020                 size += ret;
1021         }
1022         if (!flow->fate)
1023                 return rte_flow_error_set(error, ENOTSUP,
1024                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1025                                           NULL,
1026                                           "no fate action found");
1027         return size;
1028 }
1029
1030 /**
1031  * Convert the @p attributes, @p pattern, @p action, into an flow for the NIC
1032  * after ensuring the NIC will understand and process it correctly.
1033  * The conversion is only performed item/action per item/action, each of
1034  * them is written into the @p flow if its size is lesser or equal to @p
1035  * flow_size.
1036  * Validation and memory consumption computation are still performed until the
1037  * end, unless an error is encountered.
1038  *
1039  * @param[in] dev
1040  *   Pointer to Ethernet device.
1041  * @param[in, out] flow
1042  *   Pointer to flow structure.
1043  * @param[in] flow_size
1044  *   Size in bytes of the available space in @p flow, if too small some
1045  *   garbage may be present.
1046  * @param[in] attributes
1047  *   Flow rule attributes.
1048  * @param[in] pattern
1049  *   Pattern specification (list terminated by the END pattern item).
1050  * @param[in] actions
1051  *   Associated actions (list terminated by the END action).
1052  * @param[out] error
1053  *   Perform verbose error reporting if not NULL.
1054  *
1055  * @return
1056  *   On success the number of bytes consumed/necessary, if the returned value
1057  *   is lesser or equal to @p flow_size, the flow has fully been converted and
1058  *   can be applied, otherwise another call with this returned memory size
1059  *   should be done.
1060  *   On error, a negative errno value is returned and rte_errno is set.
1061  */
1062 static int
1063 mlx5_flow_merge(struct rte_eth_dev *dev, struct rte_flow *flow,
1064                 const size_t flow_size,
1065                 const struct rte_flow_attr *attributes,
1066                 const struct rte_flow_item pattern[],
1067                 const struct rte_flow_action actions[],
1068                 struct rte_flow_error *error)
1069 {
1070         struct rte_flow local_flow = { .layers = 0, };
1071         size_t size = sizeof(*flow) + sizeof(struct ibv_flow_attr);
1072         int remain = (flow_size > size) ? flow_size - size : 0;
1073         int ret;
1074
1075         if (!remain)
1076                 flow = &local_flow;
1077         ret = mlx5_flow_attributes(dev, attributes, flow, error);
1078         if (ret < 0)
1079                 return ret;
1080         ret = mlx5_flow_items(pattern, flow, remain, error);
1081         if (ret < 0)
1082                 return ret;
1083         size += ret;
1084         remain = (flow_size > size) ? flow_size - size : 0;
1085         ret = mlx5_flow_actions(dev, actions, flow, remain, error);
1086         if (ret < 0)
1087                 return ret;
1088         size += ret;
1089         if (size <= flow_size)
1090                 flow->verbs.attr->priority = flow->attributes.priority;
1091         return size;
1092 }
1093
1094 /**
1095  * Validate a flow supported by the NIC.
1096  *
1097  * @see rte_flow_validate()
1098  * @see rte_flow_ops
1099  */
1100 int
1101 mlx5_flow_validate(struct rte_eth_dev *dev,
1102                    const struct rte_flow_attr *attr,
1103                    const struct rte_flow_item items[],
1104                    const struct rte_flow_action actions[],
1105                    struct rte_flow_error *error)
1106 {
1107         int ret = mlx5_flow_merge(dev, NULL, 0, attr, items, actions, error);
1108
1109         if (ret < 0)
1110                 return ret;
1111         return 0;
1112 }
1113
1114 /**
1115  * Remove the flow.
1116  *
1117  * @param[in] dev
1118  *   Pointer to Ethernet device.
1119  * @param[in, out] flow
1120  *   Pointer to flow structure.
1121  */
1122 static void
1123 mlx5_flow_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1124 {
1125         if (flow->fate & MLX5_FLOW_FATE_DROP) {
1126                 if (flow->verbs.flow) {
1127                         claim_zero(mlx5_glue->destroy_flow(flow->verbs.flow));
1128                         flow->verbs.flow = NULL;
1129                 }
1130         }
1131         if (flow->verbs.hrxq) {
1132                 if (flow->fate & MLX5_FLOW_FATE_DROP)
1133                         mlx5_hrxq_drop_release(dev);
1134                 else if (flow->fate & MLX5_FLOW_FATE_QUEUE)
1135                         mlx5_hrxq_release(dev, flow->verbs.hrxq);
1136                 flow->verbs.hrxq = NULL;
1137         }
1138 }
1139
1140 /**
1141  * Apply the flow.
1142  *
1143  * @param[in] dev
1144  *   Pointer to Ethernet device structure.
1145  * @param[in, out] flow
1146  *   Pointer to flow structure.
1147  * @param[out] error
1148  *   Pointer to error structure.
1149  *
1150  * @return
1151  *   0 on success, a negative errno value otherwise and rte_errno is set.
1152  */
1153 static int
1154 mlx5_flow_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1155                 struct rte_flow_error *error)
1156 {
1157         if (flow->fate & MLX5_FLOW_FATE_DROP) {
1158                 flow->verbs.hrxq = mlx5_hrxq_drop_new(dev);
1159                 if (!flow->verbs.hrxq)
1160                         return rte_flow_error_set
1161                                 (error, errno,
1162                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1163                                  NULL,
1164                                  "cannot allocate Drop queue");
1165         } else if (flow->fate & MLX5_FLOW_FATE_QUEUE) {
1166                 struct mlx5_hrxq *hrxq;
1167
1168                 hrxq = mlx5_hrxq_get(dev, rss_hash_default_key,
1169                                      rss_hash_default_key_len, 0,
1170                                      &flow->queue, 1, 0, 0);
1171                 if (!hrxq)
1172                         hrxq = mlx5_hrxq_new(dev, rss_hash_default_key,
1173                                              rss_hash_default_key_len, 0,
1174                                              &flow->queue, 1, 0, 0);
1175                 if (!hrxq)
1176                         return rte_flow_error_set(error, rte_errno,
1177                                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1178                                         NULL,
1179                                         "cannot create flow");
1180                 flow->verbs.hrxq = hrxq;
1181         }
1182         flow->verbs.flow =
1183                 mlx5_glue->create_flow(flow->verbs.hrxq->qp, flow->verbs.attr);
1184         if (!flow->verbs.flow) {
1185                 if (flow->fate & MLX5_FLOW_FATE_DROP)
1186                         mlx5_hrxq_drop_release(dev);
1187                 else
1188                         mlx5_hrxq_release(dev, flow->verbs.hrxq);
1189                 flow->verbs.hrxq = NULL;
1190                 return rte_flow_error_set(error, errno,
1191                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1192                                           NULL,
1193                                           "kernel module refuses to create"
1194                                           " flow");
1195         }
1196         return 0;
1197 }
1198
1199 /**
1200  * Create a flow and add it to @p list.
1201  *
1202  * @param dev
1203  *   Pointer to Ethernet device.
1204  * @param list
1205  *   Pointer to a TAILQ flow list.
1206  * @param[in] attr
1207  *   Flow rule attributes.
1208  * @param[in] items
1209  *   Pattern specification (list terminated by the END pattern item).
1210  * @param[in] actions
1211  *   Associated actions (list terminated by the END action).
1212  * @param[out] error
1213  *   Perform verbose error reporting if not NULL.
1214  *
1215  * @return
1216  *   A flow on success, NULL otherwise and rte_errno is set.
1217  */
1218 static struct rte_flow *
1219 mlx5_flow_list_create(struct rte_eth_dev *dev,
1220                       struct mlx5_flows *list,
1221                       const struct rte_flow_attr *attr,
1222                       const struct rte_flow_item items[],
1223                       const struct rte_flow_action actions[],
1224                       struct rte_flow_error *error)
1225 {
1226         struct rte_flow *flow;
1227         size_t size;
1228         int ret;
1229
1230         ret = mlx5_flow_merge(dev, NULL, 0, attr, items, actions, error);
1231         if (ret < 0)
1232                 return NULL;
1233         size = ret;
1234         flow = rte_zmalloc(__func__, size, 0);
1235         if (!flow) {
1236                 rte_flow_error_set(error, ENOMEM,
1237                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1238                                    NULL,
1239                                    "cannot allocate memory");
1240                 return NULL;
1241         }
1242         flow->verbs.attr = (struct ibv_flow_attr *)(flow + 1);
1243         flow->verbs.specs = (uint8_t *)(flow->verbs.attr + 1);
1244         ret = mlx5_flow_merge(dev, flow, size, attr, items, actions, error);
1245         if (ret < 0)
1246                 goto error;
1247         assert((size_t)ret == size);
1248         if (dev->data->dev_started) {
1249                 ret = mlx5_flow_apply(dev, flow, error);
1250                 if (ret < 0)
1251                         goto error;
1252         }
1253         TAILQ_INSERT_TAIL(list, flow, next);
1254         return flow;
1255 error:
1256         ret = rte_errno; /* Save rte_errno before cleanup. */
1257         mlx5_flow_remove(dev, flow);
1258         rte_free(flow);
1259         rte_errno = ret; /* Restore rte_errno. */
1260         return NULL;
1261 }
1262
1263 /**
1264  * Create a flow.
1265  *
1266  * @see rte_flow_create()
1267  * @see rte_flow_ops
1268  */
1269 struct rte_flow *
1270 mlx5_flow_create(struct rte_eth_dev *dev,
1271                  const struct rte_flow_attr *attr,
1272                  const struct rte_flow_item items[],
1273                  const struct rte_flow_action actions[],
1274                  struct rte_flow_error *error)
1275 {
1276         return mlx5_flow_list_create
1277                 (dev, &((struct priv *)dev->data->dev_private)->flows,
1278                  attr, items, actions, error);
1279 }
1280
1281 /**
1282  * Destroy a flow in a list.
1283  *
1284  * @param dev
1285  *   Pointer to Ethernet device.
1286  * @param list
1287  *   Pointer to a TAILQ flow list.
1288  * @param[in] flow
1289  *   Flow to destroy.
1290  */
1291 static void
1292 mlx5_flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
1293                        struct rte_flow *flow)
1294 {
1295         mlx5_flow_remove(dev, flow);
1296         TAILQ_REMOVE(list, flow, next);
1297         rte_free(flow);
1298 }
1299
1300 /**
1301  * Destroy all flows.
1302  *
1303  * @param dev
1304  *   Pointer to Ethernet device.
1305  * @param list
1306  *   Pointer to a TAILQ flow list.
1307  */
1308 void
1309 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
1310 {
1311         while (!TAILQ_EMPTY(list)) {
1312                 struct rte_flow *flow;
1313
1314                 flow = TAILQ_FIRST(list);
1315                 mlx5_flow_list_destroy(dev, list, flow);
1316         }
1317 }
1318
1319 /**
1320  * Remove all flows.
1321  *
1322  * @param dev
1323  *   Pointer to Ethernet device.
1324  * @param list
1325  *   Pointer to a TAILQ flow list.
1326  */
1327 void
1328 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
1329 {
1330         struct rte_flow *flow;
1331
1332         TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
1333                 mlx5_flow_remove(dev, flow);
1334 }
1335
1336 /**
1337  * Add all flows.
1338  *
1339  * @param dev
1340  *   Pointer to Ethernet device.
1341  * @param list
1342  *   Pointer to a TAILQ flow list.
1343  *
1344  * @return
1345  *   0 on success, a negative errno value otherwise and rte_errno is set.
1346  */
1347 int
1348 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
1349 {
1350         struct rte_flow *flow;
1351         struct rte_flow_error error;
1352         int ret = 0;
1353
1354         TAILQ_FOREACH(flow, list, next) {
1355                 ret = mlx5_flow_apply(dev, flow, &error);
1356                 if (ret < 0)
1357                         goto error;
1358         }
1359         return 0;
1360 error:
1361         ret = rte_errno; /* Save rte_errno before cleanup. */
1362         mlx5_flow_stop(dev, list);
1363         rte_errno = ret; /* Restore rte_errno. */
1364         return -rte_errno;
1365 }
1366
1367 /**
1368  * Verify the flow list is empty
1369  *
1370  * @param dev
1371  *  Pointer to Ethernet device.
1372  *
1373  * @return the number of flows not released.
1374  */
1375 int
1376 mlx5_flow_verify(struct rte_eth_dev *dev)
1377 {
1378         struct priv *priv = dev->data->dev_private;
1379         struct rte_flow *flow;
1380         int ret = 0;
1381
1382         TAILQ_FOREACH(flow, &priv->flows, next) {
1383                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
1384                         dev->data->port_id, (void *)flow);
1385                 ++ret;
1386         }
1387         return ret;
1388 }
1389
1390 /**
1391  * Enable a control flow configured from the control plane.
1392  *
1393  * @param dev
1394  *   Pointer to Ethernet device.
1395  * @param eth_spec
1396  *   An Ethernet flow spec to apply.
1397  * @param eth_mask
1398  *   An Ethernet flow mask to apply.
1399  * @param vlan_spec
1400  *   A VLAN flow spec to apply.
1401  * @param vlan_mask
1402  *   A VLAN flow mask to apply.
1403  *
1404  * @return
1405  *   0 on success, a negative errno value otherwise and rte_errno is set.
1406  */
1407 int
1408 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
1409                     struct rte_flow_item_eth *eth_spec,
1410                     struct rte_flow_item_eth *eth_mask,
1411                     struct rte_flow_item_vlan *vlan_spec,
1412                     struct rte_flow_item_vlan *vlan_mask)
1413 {
1414         struct priv *priv = dev->data->dev_private;
1415         const struct rte_flow_attr attr = {
1416                 .ingress = 1,
1417                 .priority = priv->config.flow_prio - 1,
1418         };
1419         struct rte_flow_item items[] = {
1420                 {
1421                         .type = RTE_FLOW_ITEM_TYPE_ETH,
1422                         .spec = eth_spec,
1423                         .last = NULL,
1424                         .mask = eth_mask,
1425                 },
1426                 {
1427                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
1428                                 RTE_FLOW_ITEM_TYPE_END,
1429                         .spec = vlan_spec,
1430                         .last = NULL,
1431                         .mask = vlan_mask,
1432                 },
1433                 {
1434                         .type = RTE_FLOW_ITEM_TYPE_END,
1435                 },
1436         };
1437         uint16_t queue[priv->reta_idx_n];
1438         struct rte_flow_action_rss action_rss = {
1439                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
1440                 .level = 0,
1441                 .types = priv->rss_conf.rss_hf,
1442                 .key_len = priv->rss_conf.rss_key_len,
1443                 .queue_num = priv->reta_idx_n,
1444                 .key = priv->rss_conf.rss_key,
1445                 .queue = queue,
1446         };
1447         struct rte_flow_action actions[] = {
1448                 {
1449                         .type = RTE_FLOW_ACTION_TYPE_RSS,
1450                         .conf = &action_rss,
1451                 },
1452                 {
1453                         .type = RTE_FLOW_ACTION_TYPE_END,
1454                 },
1455         };
1456         struct rte_flow *flow;
1457         struct rte_flow_error error;
1458         unsigned int i;
1459
1460         if (!priv->reta_idx_n) {
1461                 rte_errno = EINVAL;
1462                 return -rte_errno;
1463         }
1464         for (i = 0; i != priv->reta_idx_n; ++i)
1465                 queue[i] = (*priv->reta_idx)[i];
1466         flow = mlx5_flow_list_create(dev, &priv->ctrl_flows, &attr, items,
1467                                      actions, &error);
1468         if (!flow)
1469                 return -rte_errno;
1470         return 0;
1471 }
1472
1473 /**
1474  * Enable a flow control configured from the control plane.
1475  *
1476  * @param dev
1477  *   Pointer to Ethernet device.
1478  * @param eth_spec
1479  *   An Ethernet flow spec to apply.
1480  * @param eth_mask
1481  *   An Ethernet flow mask to apply.
1482  *
1483  * @return
1484  *   0 on success, a negative errno value otherwise and rte_errno is set.
1485  */
1486 int
1487 mlx5_ctrl_flow(struct rte_eth_dev *dev,
1488                struct rte_flow_item_eth *eth_spec,
1489                struct rte_flow_item_eth *eth_mask)
1490 {
1491         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
1492 }
1493
1494 /**
1495  * Destroy a flow.
1496  *
1497  * @see rte_flow_destroy()
1498  * @see rte_flow_ops
1499  */
1500 int
1501 mlx5_flow_destroy(struct rte_eth_dev *dev,
1502                   struct rte_flow *flow,
1503                   struct rte_flow_error *error __rte_unused)
1504 {
1505         struct priv *priv = dev->data->dev_private;
1506
1507         mlx5_flow_list_destroy(dev, &priv->flows, flow);
1508         return 0;
1509 }
1510
1511 /**
1512  * Destroy all flows.
1513  *
1514  * @see rte_flow_flush()
1515  * @see rte_flow_ops
1516  */
1517 int
1518 mlx5_flow_flush(struct rte_eth_dev *dev,
1519                 struct rte_flow_error *error __rte_unused)
1520 {
1521         struct priv *priv = dev->data->dev_private;
1522
1523         mlx5_flow_list_flush(dev, &priv->flows);
1524         return 0;
1525 }
1526
1527 /**
1528  * Isolated mode.
1529  *
1530  * @see rte_flow_isolate()
1531  * @see rte_flow_ops
1532  */
1533 int
1534 mlx5_flow_isolate(struct rte_eth_dev *dev,
1535                   int enable,
1536                   struct rte_flow_error *error)
1537 {
1538         struct priv *priv = dev->data->dev_private;
1539
1540         if (dev->data->dev_started) {
1541                 rte_flow_error_set(error, EBUSY,
1542                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1543                                    NULL,
1544                                    "port must be stopped first");
1545                 return -rte_errno;
1546         }
1547         priv->isolated = !!enable;
1548         if (enable)
1549                 dev->dev_ops = &mlx5_dev_ops_isolate;
1550         else
1551                 dev->dev_ops = &mlx5_dev_ops;
1552         return 0;
1553 }
1554
1555 /**
1556  * Convert a flow director filter to a generic flow.
1557  *
1558  * @param dev
1559  *   Pointer to Ethernet device.
1560  * @param fdir_filter
1561  *   Flow director filter to add.
1562  * @param attributes
1563  *   Generic flow parameters structure.
1564  *
1565  * @return
1566  *   0 on success, a negative errno value otherwise and rte_errno is set.
1567  */
1568 static int
1569 mlx5_fdir_filter_convert(struct rte_eth_dev *dev,
1570                          const struct rte_eth_fdir_filter *fdir_filter,
1571                          struct mlx5_fdir *attributes)
1572 {
1573         struct priv *priv = dev->data->dev_private;
1574         const struct rte_eth_fdir_input *input = &fdir_filter->input;
1575         const struct rte_eth_fdir_masks *mask =
1576                 &dev->data->dev_conf.fdir_conf.mask;
1577
1578         /* Validate queue number. */
1579         if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
1580                 DRV_LOG(ERR, "port %u invalid queue number %d",
1581                         dev->data->port_id, fdir_filter->action.rx_queue);
1582                 rte_errno = EINVAL;
1583                 return -rte_errno;
1584         }
1585         attributes->attr.ingress = 1;
1586         attributes->items[0] = (struct rte_flow_item) {
1587                 .type = RTE_FLOW_ITEM_TYPE_ETH,
1588                 .spec = &attributes->l2,
1589                 .mask = &attributes->l2_mask,
1590         };
1591         switch (fdir_filter->action.behavior) {
1592         case RTE_ETH_FDIR_ACCEPT:
1593                 attributes->actions[0] = (struct rte_flow_action){
1594                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
1595                         .conf = &attributes->queue,
1596                 };
1597                 break;
1598         case RTE_ETH_FDIR_REJECT:
1599                 attributes->actions[0] = (struct rte_flow_action){
1600                         .type = RTE_FLOW_ACTION_TYPE_DROP,
1601                 };
1602                 break;
1603         default:
1604                 DRV_LOG(ERR, "port %u invalid behavior %d",
1605                         dev->data->port_id,
1606                         fdir_filter->action.behavior);
1607                 rte_errno = ENOTSUP;
1608                 return -rte_errno;
1609         }
1610         attributes->queue.index = fdir_filter->action.rx_queue;
1611         /* Handle L3. */
1612         switch (fdir_filter->input.flow_type) {
1613         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
1614         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
1615         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
1616                 attributes->l3.ipv4.hdr = (struct ipv4_hdr){
1617                         .src_addr = input->flow.ip4_flow.src_ip,
1618                         .dst_addr = input->flow.ip4_flow.dst_ip,
1619                         .time_to_live = input->flow.ip4_flow.ttl,
1620                         .type_of_service = input->flow.ip4_flow.tos,
1621                         .next_proto_id = input->flow.ip4_flow.proto,
1622                 };
1623                 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
1624                         .src_addr = mask->ipv4_mask.src_ip,
1625                         .dst_addr = mask->ipv4_mask.dst_ip,
1626                         .time_to_live = mask->ipv4_mask.ttl,
1627                         .type_of_service = mask->ipv4_mask.tos,
1628                         .next_proto_id = mask->ipv4_mask.proto,
1629                 };
1630                 attributes->items[1] = (struct rte_flow_item){
1631                         .type = RTE_FLOW_ITEM_TYPE_IPV4,
1632                         .spec = &attributes->l3,
1633                         .mask = &attributes->l3_mask,
1634                 };
1635                 break;
1636         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
1637         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
1638         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
1639                 attributes->l3.ipv6.hdr = (struct ipv6_hdr){
1640                         .hop_limits = input->flow.ipv6_flow.hop_limits,
1641                         .proto = input->flow.ipv6_flow.proto,
1642                 };
1643
1644                 memcpy(attributes->l3.ipv6.hdr.src_addr,
1645                        input->flow.ipv6_flow.src_ip,
1646                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
1647                 memcpy(attributes->l3.ipv6.hdr.dst_addr,
1648                        input->flow.ipv6_flow.dst_ip,
1649                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
1650                 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
1651                        mask->ipv6_mask.src_ip,
1652                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
1653                 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
1654                        mask->ipv6_mask.dst_ip,
1655                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
1656                 attributes->items[1] = (struct rte_flow_item){
1657                         .type = RTE_FLOW_ITEM_TYPE_IPV6,
1658                         .spec = &attributes->l3,
1659                         .mask = &attributes->l3_mask,
1660                 };
1661                 break;
1662         default:
1663                 DRV_LOG(ERR, "port %u invalid flow type%d",
1664                         dev->data->port_id, fdir_filter->input.flow_type);
1665                 rte_errno = ENOTSUP;
1666                 return -rte_errno;
1667         }
1668         /* Handle L4. */
1669         switch (fdir_filter->input.flow_type) {
1670         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
1671                 attributes->l4.udp.hdr = (struct udp_hdr){
1672                         .src_port = input->flow.udp4_flow.src_port,
1673                         .dst_port = input->flow.udp4_flow.dst_port,
1674                 };
1675                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
1676                         .src_port = mask->src_port_mask,
1677                         .dst_port = mask->dst_port_mask,
1678                 };
1679                 attributes->items[2] = (struct rte_flow_item){
1680                         .type = RTE_FLOW_ITEM_TYPE_UDP,
1681                         .spec = &attributes->l4,
1682                         .mask = &attributes->l4_mask,
1683                 };
1684                 break;
1685         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
1686                 attributes->l4.tcp.hdr = (struct tcp_hdr){
1687                         .src_port = input->flow.tcp4_flow.src_port,
1688                         .dst_port = input->flow.tcp4_flow.dst_port,
1689                 };
1690                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
1691                         .src_port = mask->src_port_mask,
1692                         .dst_port = mask->dst_port_mask,
1693                 };
1694                 attributes->items[2] = (struct rte_flow_item){
1695                         .type = RTE_FLOW_ITEM_TYPE_TCP,
1696                         .spec = &attributes->l4,
1697                         .mask = &attributes->l4_mask,
1698                 };
1699                 break;
1700         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
1701                 attributes->l4.udp.hdr = (struct udp_hdr){
1702                         .src_port = input->flow.udp6_flow.src_port,
1703                         .dst_port = input->flow.udp6_flow.dst_port,
1704                 };
1705                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
1706                         .src_port = mask->src_port_mask,
1707                         .dst_port = mask->dst_port_mask,
1708                 };
1709                 attributes->items[2] = (struct rte_flow_item){
1710                         .type = RTE_FLOW_ITEM_TYPE_UDP,
1711                         .spec = &attributes->l4,
1712                         .mask = &attributes->l4_mask,
1713                 };
1714                 break;
1715         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
1716                 attributes->l4.tcp.hdr = (struct tcp_hdr){
1717                         .src_port = input->flow.tcp6_flow.src_port,
1718                         .dst_port = input->flow.tcp6_flow.dst_port,
1719                 };
1720                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
1721                         .src_port = mask->src_port_mask,
1722                         .dst_port = mask->dst_port_mask,
1723                 };
1724                 attributes->items[2] = (struct rte_flow_item){
1725                         .type = RTE_FLOW_ITEM_TYPE_TCP,
1726                         .spec = &attributes->l4,
1727                         .mask = &attributes->l4_mask,
1728                 };
1729                 break;
1730         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
1731         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
1732                 break;
1733         default:
1734                 DRV_LOG(ERR, "port %u invalid flow type%d",
1735                         dev->data->port_id, fdir_filter->input.flow_type);
1736                 rte_errno = ENOTSUP;
1737                 return -rte_errno;
1738         }
1739         return 0;
1740 }
1741
1742 /**
1743  * Add new flow director filter and store it in list.
1744  *
1745  * @param dev
1746  *   Pointer to Ethernet device.
1747  * @param fdir_filter
1748  *   Flow director filter to add.
1749  *
1750  * @return
1751  *   0 on success, a negative errno value otherwise and rte_errno is set.
1752  */
1753 static int
1754 mlx5_fdir_filter_add(struct rte_eth_dev *dev,
1755                      const struct rte_eth_fdir_filter *fdir_filter)
1756 {
1757         struct priv *priv = dev->data->dev_private;
1758         struct mlx5_fdir attributes = {
1759                 .attr.group = 0,
1760                 .l2_mask = {
1761                         .dst.addr_bytes = "\x00\x00\x00\x00\x00\x00",
1762                         .src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
1763                         .type = 0,
1764                 },
1765         };
1766         struct rte_flow_error error;
1767         struct rte_flow *flow;
1768         int ret;
1769
1770         ret = mlx5_fdir_filter_convert(dev, fdir_filter, &attributes);
1771         if (ret)
1772                 return ret;
1773         flow = mlx5_flow_list_create(dev, &priv->flows, &attributes.attr,
1774                                      attributes.items, attributes.actions,
1775                                      &error);
1776         if (flow) {
1777                 DRV_LOG(DEBUG, "port %u FDIR created %p", dev->data->port_id,
1778                         (void *)flow);
1779                 return 0;
1780         }
1781         return -rte_errno;
1782 }
1783
1784 /**
1785  * Delete specific filter.
1786  *
1787  * @param dev
1788  *   Pointer to Ethernet device.
1789  * @param fdir_filter
1790  *   Filter to be deleted.
1791  *
1792  * @return
1793  *   0 on success, a negative errno value otherwise and rte_errno is set.
1794  */
1795 static int
1796 mlx5_fdir_filter_delete(struct rte_eth_dev *dev __rte_unused,
1797                         const struct rte_eth_fdir_filter *fdir_filter
1798                         __rte_unused)
1799 {
1800         rte_errno = ENOTSUP;
1801         return -rte_errno;
1802 }
1803
1804 /**
1805  * Update queue for specific filter.
1806  *
1807  * @param dev
1808  *   Pointer to Ethernet device.
1809  * @param fdir_filter
1810  *   Filter to be updated.
1811  *
1812  * @return
1813  *   0 on success, a negative errno value otherwise and rte_errno is set.
1814  */
1815 static int
1816 mlx5_fdir_filter_update(struct rte_eth_dev *dev,
1817                         const struct rte_eth_fdir_filter *fdir_filter)
1818 {
1819         int ret;
1820
1821         ret = mlx5_fdir_filter_delete(dev, fdir_filter);
1822         if (ret)
1823                 return ret;
1824         return mlx5_fdir_filter_add(dev, fdir_filter);
1825 }
1826
1827 /**
1828  * Flush all filters.
1829  *
1830  * @param dev
1831  *   Pointer to Ethernet device.
1832  */
1833 static void
1834 mlx5_fdir_filter_flush(struct rte_eth_dev *dev)
1835 {
1836         struct priv *priv = dev->data->dev_private;
1837
1838         mlx5_flow_list_flush(dev, &priv->flows);
1839 }
1840
1841 /**
1842  * Get flow director information.
1843  *
1844  * @param dev
1845  *   Pointer to Ethernet device.
1846  * @param[out] fdir_info
1847  *   Resulting flow director information.
1848  */
1849 static void
1850 mlx5_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
1851 {
1852         struct rte_eth_fdir_masks *mask =
1853                 &dev->data->dev_conf.fdir_conf.mask;
1854
1855         fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
1856         fdir_info->guarant_spc = 0;
1857         rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
1858         fdir_info->max_flexpayload = 0;
1859         fdir_info->flow_types_mask[0] = 0;
1860         fdir_info->flex_payload_unit = 0;
1861         fdir_info->max_flex_payload_segment_num = 0;
1862         fdir_info->flex_payload_limit = 0;
1863         memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
1864 }
1865
1866 /**
1867  * Deal with flow director operations.
1868  *
1869  * @param dev
1870  *   Pointer to Ethernet device.
1871  * @param filter_op
1872  *   Operation to perform.
1873  * @param arg
1874  *   Pointer to operation-specific structure.
1875  *
1876  * @return
1877  *   0 on success, a negative errno value otherwise and rte_errno is set.
1878  */
1879 static int
1880 mlx5_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
1881                     void *arg)
1882 {
1883         enum rte_fdir_mode fdir_mode =
1884                 dev->data->dev_conf.fdir_conf.mode;
1885
1886         if (filter_op == RTE_ETH_FILTER_NOP)
1887                 return 0;
1888         if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
1889             fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
1890                 DRV_LOG(ERR, "port %u flow director mode %d not supported",
1891                         dev->data->port_id, fdir_mode);
1892                 rte_errno = EINVAL;
1893                 return -rte_errno;
1894         }
1895         switch (filter_op) {
1896         case RTE_ETH_FILTER_ADD:
1897                 return mlx5_fdir_filter_add(dev, arg);
1898         case RTE_ETH_FILTER_UPDATE:
1899                 return mlx5_fdir_filter_update(dev, arg);
1900         case RTE_ETH_FILTER_DELETE:
1901                 return mlx5_fdir_filter_delete(dev, arg);
1902         case RTE_ETH_FILTER_FLUSH:
1903                 mlx5_fdir_filter_flush(dev);
1904                 break;
1905         case RTE_ETH_FILTER_INFO:
1906                 mlx5_fdir_info_get(dev, arg);
1907                 break;
1908         default:
1909                 DRV_LOG(DEBUG, "port %u unknown operation %u",
1910                         dev->data->port_id, filter_op);
1911                 rte_errno = EINVAL;
1912                 return -rte_errno;
1913         }
1914         return 0;
1915 }
1916
1917 /**
1918  * Manage filter operations.
1919  *
1920  * @param dev
1921  *   Pointer to Ethernet device structure.
1922  * @param filter_type
1923  *   Filter type.
1924  * @param filter_op
1925  *   Operation to perform.
1926  * @param arg
1927  *   Pointer to operation-specific structure.
1928  *
1929  * @return
1930  *   0 on success, a negative errno value otherwise and rte_errno is set.
1931  */
1932 int
1933 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
1934                      enum rte_filter_type filter_type,
1935                      enum rte_filter_op filter_op,
1936                      void *arg)
1937 {
1938         switch (filter_type) {
1939         case RTE_ETH_FILTER_GENERIC:
1940                 if (filter_op != RTE_ETH_FILTER_GET) {
1941                         rte_errno = EINVAL;
1942                         return -rte_errno;
1943                 }
1944                 *(const void **)arg = &mlx5_flow_ops;
1945                 return 0;
1946         case RTE_ETH_FILTER_FDIR:
1947                 return mlx5_fdir_ctrl_func(dev, filter_op, arg);
1948         default:
1949                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
1950                         dev->data->port_id, filter_type);
1951                 rte_errno = ENOTSUP;
1952                 return -rte_errno;
1953         }
1954         return 0;
1955 }