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 <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_GRE | RTE_PTYPE_L4_UDP,
 298         },
 299         {
 300                 .tunnel = MLX5_FLOW_LAYER_MPLS,
 301                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
 302         },
 303 };
 304
 305 /**
 306  * Discover the maximum number of priority available.
 307  *
 308  * @param[in] dev
 309  *   Pointer to the Ethernet device structure.
 310  *
 311  * @return
 312  *   number of supported flow priority on success, a negative errno
 313  *   value otherwise and rte_errno is set.
 314  */
 315 int
 316 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 317 {
 318         struct {
 319                 struct ibv_flow_attr attr;
 320                 struct ibv_flow_spec_eth eth;
 321                 struct ibv_flow_spec_action_drop drop;
 322         } flow_attr = {
 323                 .attr = {
 324                         .num_of_specs = 2,
 325                 },
 326                 .eth = {
 327                         .type = IBV_FLOW_SPEC_ETH,
 328                         .size = sizeof(struct ibv_flow_spec_eth),
 329                 },
 330                 .drop = {
 331                         .size = sizeof(struct ibv_flow_spec_action_drop),
 332                         .type = IBV_FLOW_SPEC_ACTION_DROP,
 333                 },
 334         };
 335         struct ibv_flow *flow;
 336         struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
 337         uint16_t vprio[] = { 8, 16 };
 338         int i;
 339         int priority = 0;
 340
 341         if (!drop) {
 342                 rte_errno = ENOTSUP;
 343                 return -rte_errno;
 344         }
 345         for (i = 0; i != RTE_DIM(vprio); i++) {
 346                 flow_attr.attr.priority = vprio[i] - 1;
 347                 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 348                 if (!flow)
 349                         break;
 350                 claim_zero(mlx5_glue->destroy_flow(flow));
 351                 priority = vprio[i];
 352         }
 353         switch (priority) {
 354         case 8:
 355                 priority = RTE_DIM(priority_map_3);
 356                 break;
 357         case 16:
 358                 priority = RTE_DIM(priority_map_5);
 359                 break;
 360         default:
 361                 rte_errno = ENOTSUP;
 362                 DRV_LOG(ERR,
 363                         "port %u verbs maximum priority: %d expected 8/16",
 364                         dev->data->port_id, vprio[i]);
 365                 return -rte_errno;
 366         }
 367         mlx5_hrxq_drop_release(dev);
 368         DRV_LOG(INFO, "port %u flow maximum priority: %d",
 369                 dev->data->port_id, priority);
 370         return priority;
 371 }
 372
 373 /**
 374  * Adjust flow priority based on the highest layer and the request priority.
 375  *
 376  * @param[in] dev
 377  *   Pointer to the Ethernet device structure.
 378  * @param[in] priority
 379  *   The rule base priority.
 380  * @param[in] subpriority
 381  *   The priority based on the items.
 382  *
 383  * @return
 384  *   The new priority.
 385  */
 386 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
 387                                    uint32_t subpriority)
 388 {
 389         uint32_t res = 0;
 390         struct priv *priv = dev->data->dev_private;
 391
 392         switch (priv->config.flow_prio) {
 393         case RTE_DIM(priority_map_3):
 394                 res = priority_map_3[priority][subpriority];
 395                 break;
 396         case RTE_DIM(priority_map_5):
 397                 res = priority_map_5[priority][subpriority];
 398                 break;
 399         }
 400         return  res;
 401 }
 402
 403 /**
 404  * Verify the @p item specifications (spec, last, mask) are compatible with the
 405  * NIC capabilities.
 406  *
 407  * @param[in] item
 408  *   Item specification.
 409  * @param[in] mask
 410  *   @p item->mask or flow default bit-masks.
 411  * @param[in] nic_mask
 412  *   Bit-masks covering supported fields by the NIC to compare with user mask.
 413  * @param[in] size
 414  *   Bit-masks size in bytes.
 415  * @param[out] error
 416  *   Pointer to error structure.
 417  *
 418  * @return
 419  *   0 on success, a negative errno value otherwise and rte_errno is set.
 420  */
 421 int
 422 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
 423                           const uint8_t *mask,
 424                           const uint8_t *nic_mask,
 425                           unsigned int size,
 426                           struct rte_flow_error *error)
 427 {
 428         unsigned int i;
 429
 430         assert(nic_mask);
 431         for (i = 0; i < size; ++i)
 432                 if ((nic_mask[i] | mask[i]) != nic_mask[i])
 433                         return rte_flow_error_set(error, ENOTSUP,
 434                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 435                                                   item,
 436                                                   "mask enables non supported"
 437                                                   " bits");
 438         if (!item->spec && (item->mask || item->last))
 439                 return rte_flow_error_set(error, EINVAL,
 440                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
 441                                           "mask/last without a spec is not"
 442                                           " supported");
 443         if (item->spec && item->last) {
 444                 uint8_t spec[size];
 445                 uint8_t last[size];
 446                 unsigned int i;
 447                 int ret;
 448
 449                 for (i = 0; i < size; ++i) {
 450                         spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
 451                         last[i] = ((const uint8_t *)item->last)[i] & mask[i];
 452                 }
 453                 ret = memcmp(spec, last, size);
 454                 if (ret != 0)
 455                         return rte_flow_error_set(error, EINVAL,
 456                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 457                                                   item,
 458                                                   "range is not valid");
 459         }
 460         return 0;
 461 }
 462
 463 /**
 464  * Adjust the hash fields according to the @p flow information.
 465  *
 466  * @param[in] dev_flow.
 467  *   Pointer to the mlx5_flow.
 468  * @param[in] tunnel
 469  *   1 when the hash field is for a tunnel item.
 470  * @param[in] layer_types
 471  *   ETH_RSS_* types.
 472  * @param[in] hash_fields
 473  *   Item hash fields.
 474  *
 475  * @return
 476  *   The hash fileds that should be used.
 477  */
 478 uint64_t
 479 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
 480                             int tunnel __rte_unused, uint64_t layer_types,
 481                             uint64_t hash_fields)
 482 {
 483         struct rte_flow *flow = dev_flow->flow;
 484 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
 485         int rss_request_inner = flow->rss.level >= 2;
 486
 487         /* Check RSS hash level for tunnel. */
 488         if (tunnel && rss_request_inner)
 489                 hash_fields |= IBV_RX_HASH_INNER;
 490         else if (tunnel || rss_request_inner)
 491                 return 0;
 492 #endif
 493         /* Check if requested layer matches RSS hash fields. */
 494         if (!(flow->rss.types & layer_types))
 495                 return 0;
 496         return hash_fields;
 497 }
 498
 499 /**
 500  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
 501  * if several tunnel rules are used on this queue, the tunnel ptype will be
 502  * cleared.
 503  *
 504  * @param rxq_ctrl
 505  *   Rx queue to update.
 506  */
 507 static void
 508 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
 509 {
 510         unsigned int i;
 511         uint32_t tunnel_ptype = 0;
 512
 513         /* Look up for the ptype to use. */
 514         for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
 515                 if (!rxq_ctrl->flow_tunnels_n[i])
 516                         continue;
 517                 if (!tunnel_ptype) {
 518                         tunnel_ptype = tunnels_info[i].ptype;
 519                 } else {
 520                         tunnel_ptype = 0;
 521                         break;
 522                 }
 523         }
 524         rxq_ctrl->rxq.tunnel = tunnel_ptype;
 525 }
 526
 527 /**
 528  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
 529  * flow.
 530  *
 531  * @param[in] dev
 532  *   Pointer to the Ethernet device structure.
 533  * @param[in] dev_flow
 534  *   Pointer to device flow structure.
 535  */
 536 static void
 537 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
 538 {
 539         struct priv *priv = dev->data->dev_private;
 540         struct rte_flow *flow = dev_flow->flow;
 541         const int mark = !!(flow->actions &
 542                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
 543         const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
 544         unsigned int i;
 545
 546         for (i = 0; i != flow->rss.queue_num; ++i) {
 547                 int idx = (*flow->queue)[i];
 548                 struct mlx5_rxq_ctrl *rxq_ctrl =
 549                         container_of((*priv->rxqs)[idx],
 550                                      struct mlx5_rxq_ctrl, rxq);
 551
 552                 if (mark) {
 553                         rxq_ctrl->rxq.mark = 1;
 554                         rxq_ctrl->flow_mark_n++;
 555                 }
 556                 if (tunnel) {
 557                         unsigned int j;
 558
 559                         /* Increase the counter matching the flow. */
 560                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 561                                 if ((tunnels_info[j].tunnel &
 562                                      dev_flow->layers) ==
 563                                     tunnels_info[j].tunnel) {
 564                                         rxq_ctrl->flow_tunnels_n[j]++;
 565                                         break;
 566                                 }
 567                         }
 568                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
 569                 }
 570         }
 571 }
 572
 573 /**
 574  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
 575  *
 576  * @param[in] dev
 577  *   Pointer to the Ethernet device structure.
 578  * @param[in] flow
 579  *   Pointer to flow structure.
 580  */
 581 static void
 582 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
 583 {
 584         struct mlx5_flow *dev_flow;
 585
 586         LIST_FOREACH(dev_flow, &flow->dev_flows, next)
 587                 flow_drv_rxq_flags_set(dev, dev_flow);
 588 }
 589
 590 /**
 591  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
 592  * device flow if no other flow uses it with the same kind of request.
 593  *
 594  * @param dev
 595  *   Pointer to Ethernet device.
 596  * @param[in] dev_flow
 597  *   Pointer to the device flow.
 598  */
 599 static void
 600 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
 601 {
 602         struct priv *priv = dev->data->dev_private;
 603         struct rte_flow *flow = dev_flow->flow;
 604         const int mark = !!(flow->actions &
 605                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
 606         const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
 607         unsigned int i;
 608
 609         assert(dev->data->dev_started);
 610         for (i = 0; i != flow->rss.queue_num; ++i) {
 611                 int idx = (*flow->queue)[i];
 612                 struct mlx5_rxq_ctrl *rxq_ctrl =
 613                         container_of((*priv->rxqs)[idx],
 614                                      struct mlx5_rxq_ctrl, rxq);
 615
 616                 if (mark) {
 617                         rxq_ctrl->flow_mark_n--;
 618                         rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
 619                 }
 620                 if (tunnel) {
 621                         unsigned int j;
 622
 623                         /* Decrease the counter matching the flow. */
 624                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 625                                 if ((tunnels_info[j].tunnel &
 626                                      dev_flow->layers) ==
 627                                     tunnels_info[j].tunnel) {
 628                                         rxq_ctrl->flow_tunnels_n[j]--;
 629                                         break;
 630                                 }
 631                         }
 632                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
 633                 }
 634         }
 635 }
 636
 637 /**
 638  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
 639  * @p flow if no other flow uses it with the same kind of request.
 640  *
 641  * @param dev
 642  *   Pointer to Ethernet device.
 643  * @param[in] flow
 644  *   Pointer to the flow.
 645  */
 646 static void
 647 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
 648 {
 649         struct mlx5_flow *dev_flow;
 650
 651         LIST_FOREACH(dev_flow, &flow->dev_flows, next)
 652                 flow_drv_rxq_flags_trim(dev, dev_flow);
 653 }
 654
 655 /**
 656  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
 657  *
 658  * @param dev
 659  *   Pointer to Ethernet device.
 660  */
 661 static void
 662 flow_rxq_flags_clear(struct rte_eth_dev *dev)
 663 {
 664         struct priv *priv = dev->data->dev_private;
 665         unsigned int i;
 666
 667         for (i = 0; i != priv->rxqs_n; ++i) {
 668                 struct mlx5_rxq_ctrl *rxq_ctrl;
 669                 unsigned int j;
 670
 671                 if (!(*priv->rxqs)[i])
 672                         continue;
 673                 rxq_ctrl = container_of((*priv->rxqs)[i],
 674                                         struct mlx5_rxq_ctrl, rxq);
 675                 rxq_ctrl->flow_mark_n = 0;
 676                 rxq_ctrl->rxq.mark = 0;
 677                 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
 678                         rxq_ctrl->flow_tunnels_n[j] = 0;
 679                 rxq_ctrl->rxq.tunnel = 0;
 680         }
 681 }
 682
 683 /*
 684  * Validate the flag action.
 685  *
 686  * @param[in] action_flags
 687  *   Bit-fields that holds the actions detected until now.
 688  * @param[in] attr
 689  *   Attributes of flow that includes this action.
 690  * @param[out] error
 691  *   Pointer to error structure.
 692  *
 693  * @return
 694  *   0 on success, a negative errno value otherwise and rte_errno is set.
 695  */
 696 int
 697 mlx5_flow_validate_action_flag(uint64_t action_flags,
 698                                const struct rte_flow_attr *attr,
 699                                struct rte_flow_error *error)
 700 {
 701
 702         if (action_flags & MLX5_FLOW_ACTION_DROP)
 703                 return rte_flow_error_set(error, EINVAL,
 704                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 705                                           "can't drop and flag in same flow");
 706         if (action_flags & MLX5_FLOW_ACTION_MARK)
 707                 return rte_flow_error_set(error, EINVAL,
 708                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 709                                           "can't mark and flag in same flow");
 710         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 711                 return rte_flow_error_set(error, EINVAL,
 712                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 713                                           "can't have 2 flag"
 714                                           " actions in same flow");
 715         if (attr->egress)
 716                 return rte_flow_error_set(error, ENOTSUP,
 717                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 718                                           "flag action not supported for "
 719                                           "egress");
 720         return 0;
 721 }
 722
 723 /*
 724  * Validate the mark action.
 725  *
 726  * @param[in] action
 727  *   Pointer to the queue action.
 728  * @param[in] action_flags
 729  *   Bit-fields that holds the actions detected until now.
 730  * @param[in] attr
 731  *   Attributes of flow that includes this action.
 732  * @param[out] error
 733  *   Pointer to error structure.
 734  *
 735  * @return
 736  *   0 on success, a negative errno value otherwise and rte_errno is set.
 737  */
 738 int
 739 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
 740                                uint64_t action_flags,
 741                                const struct rte_flow_attr *attr,
 742                                struct rte_flow_error *error)
 743 {
 744         const struct rte_flow_action_mark *mark = action->conf;
 745
 746         if (!mark)
 747                 return rte_flow_error_set(error, EINVAL,
 748                                           RTE_FLOW_ERROR_TYPE_ACTION,
 749                                           action,
 750                                           "configuration cannot be null");
 751         if (mark->id >= MLX5_FLOW_MARK_MAX)
 752                 return rte_flow_error_set(error, EINVAL,
 753                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 754                                           &mark->id,
 755                                           "mark id must in 0 <= id < "
 756                                           RTE_STR(MLX5_FLOW_MARK_MAX));
 757         if (action_flags & MLX5_FLOW_ACTION_DROP)
 758                 return rte_flow_error_set(error, EINVAL,
 759                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 760                                           "can't drop and mark in same flow");
 761         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 762                 return rte_flow_error_set(error, EINVAL,
 763                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 764                                           "can't flag and mark in same flow");
 765         if (action_flags & MLX5_FLOW_ACTION_MARK)
 766                 return rte_flow_error_set(error, EINVAL,
 767                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 768                                           "can't have 2 mark actions in same"
 769                                           " flow");
 770         if (attr->egress)
 771                 return rte_flow_error_set(error, ENOTSUP,
 772                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 773                                           "mark action not supported for "
 774                                           "egress");
 775         return 0;
 776 }
 777
 778 /*
 779  * Validate the drop action.
 780  *
 781  * @param[in] action_flags
 782  *   Bit-fields that holds the actions detected until now.
 783  * @param[in] attr
 784  *   Attributes of flow that includes this action.
 785  * @param[out] error
 786  *   Pointer to error structure.
 787  *
 788  * @return
 789  *   0 on success, a negative errno value otherwise and rte_ernno is set.
 790  */
 791 int
 792 mlx5_flow_validate_action_drop(uint64_t action_flags,
 793                                const struct rte_flow_attr *attr,
 794                                struct rte_flow_error *error)
 795 {
 796         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 797                 return rte_flow_error_set(error, EINVAL,
 798                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 799                                           "can't drop and flag in same flow");
 800         if (action_flags & MLX5_FLOW_ACTION_MARK)
 801                 return rte_flow_error_set(error, EINVAL,
 802                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 803                                           "can't drop and mark in same flow");
 804         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
 805                 return rte_flow_error_set(error, EINVAL,
 806                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 807                                           "can't have 2 fate actions in"
 808                                           " same flow");
 809         if (attr->egress)
 810                 return rte_flow_error_set(error, ENOTSUP,
 811                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 812                                           "drop action not supported for "
 813                                           "egress");
 814         return 0;
 815 }
 816
 817 /*
 818  * Validate the queue action.
 819  *
 820  * @param[in] action
 821  *   Pointer to the queue action.
 822  * @param[in] action_flags
 823  *   Bit-fields that holds the actions detected until now.
 824  * @param[in] dev
 825  *   Pointer to the Ethernet device structure.
 826  * @param[in] attr
 827  *   Attributes of flow that includes this action.
 828  * @param[out] error
 829  *   Pointer to error structure.
 830  *
 831  * @return
 832  *   0 on success, a negative errno value otherwise and rte_ernno is set.
 833  */
 834 int
 835 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 836                                 uint64_t action_flags,
 837                                 struct rte_eth_dev *dev,
 838                                 const struct rte_flow_attr *attr,
 839                                 struct rte_flow_error *error)
 840 {
 841         struct priv *priv = dev->data->dev_private;
 842         const struct rte_flow_action_queue *queue = action->conf;
 843
 844         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
 845                 return rte_flow_error_set(error, EINVAL,
 846                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 847                                           "can't have 2 fate actions in"
 848                                           " same flow");
 849         if (queue->index >= priv->rxqs_n)
 850                 return rte_flow_error_set(error, EINVAL,
 851                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 852                                           &queue->index,
 853                                           "queue index out of range");
 854         if (!(*priv->rxqs)[queue->index])
 855                 return rte_flow_error_set(error, EINVAL,
 856                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 857                                           &queue->index,
 858                                           "queue is not configured");
 859         if (attr->egress)
 860                 return rte_flow_error_set(error, ENOTSUP,
 861                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 862                                           "queue action not supported for "
 863                                           "egress");
 864         return 0;
 865 }
 866
 867 /*
 868  * Validate the rss action.
 869  *
 870  * @param[in] action
 871  *   Pointer to the queue action.
 872  * @param[in] action_flags
 873  *   Bit-fields that holds the actions detected until now.
 874  * @param[in] dev
 875  *   Pointer to the Ethernet device structure.
 876  * @param[in] attr
 877  *   Attributes of flow that includes this action.
 878  * @param[out] error
 879  *   Pointer to error structure.
 880  *
 881  * @return
 882  *   0 on success, a negative errno value otherwise and rte_ernno is set.
 883  */
 884 int
 885 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
 886                               uint64_t action_flags,
 887                               struct rte_eth_dev *dev,
 888                               const struct rte_flow_attr *attr,
 889                               struct rte_flow_error *error)
 890 {
 891         struct priv *priv = dev->data->dev_private;
 892         const struct rte_flow_action_rss *rss = action->conf;
 893         unsigned int i;
 894
 895         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
 896                 return rte_flow_error_set(error, EINVAL,
 897                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 898                                           "can't have 2 fate actions"
 899                                           " in same flow");
 900         if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
 901             rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
 902                 return rte_flow_error_set(error, ENOTSUP,
 903                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 904                                           &rss->func,
 905                                           "RSS hash function not supported");
 906 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
 907         if (rss->level > 2)
 908 #else
 909         if (rss->level > 1)
 910 #endif
 911                 return rte_flow_error_set(error, ENOTSUP,
 912                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 913                                           &rss->level,
 914                                           "tunnel RSS is not supported");
 915         /* allow RSS key_len 0 in case of NULL (default) RSS key. */
 916         if (rss->key_len == 0 && rss->key != NULL)
 917                 return rte_flow_error_set(error, ENOTSUP,
 918                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 919                                           &rss->key_len,
 920                                           "RSS hash key length 0");
 921         if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
 922                 return rte_flow_error_set(error, ENOTSUP,
 923                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 924                                           &rss->key_len,
 925                                           "RSS hash key too small");
 926         if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
 927                 return rte_flow_error_set(error, ENOTSUP,
 928                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 929                                           &rss->key_len,
 930                                           "RSS hash key too large");
 931         if (rss->queue_num > priv->config.ind_table_max_size)
 932                 return rte_flow_error_set(error, ENOTSUP,
 933                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 934                                           &rss->queue_num,
 935                                           "number of queues too large");
 936         if (rss->types & MLX5_RSS_HF_MASK)
 937                 return rte_flow_error_set(error, ENOTSUP,
 938                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 939                                           &rss->types,
 940                                           "some RSS protocols are not"
 941                                           " supported");
 942         for (i = 0; i != rss->queue_num; ++i) {
 943                 if (!(*priv->rxqs)[rss->queue[i]])
 944                         return rte_flow_error_set
 945                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 946                                  &rss->queue[i], "queue is not configured");
 947         }
 948         if (attr->egress)
 949                 return rte_flow_error_set(error, ENOTSUP,
 950                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 951                                           "rss action not supported for "
 952                                           "egress");
 953         return 0;
 954 }
 955
 956 /*
 957  * Validate the count action.
 958  *
 959  * @param[in] dev
 960  *   Pointer to the Ethernet device structure.
 961  * @param[in] attr
 962  *   Attributes of flow that includes this action.
 963  * @param[out] error
 964  *   Pointer to error structure.
 965  *
 966  * @return
 967  *   0 on success, a negative errno value otherwise and rte_ernno is set.
 968  */
 969 int
 970 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
 971                                 const struct rte_flow_attr *attr,
 972                                 struct rte_flow_error *error)
 973 {
 974         if (attr->egress)
 975                 return rte_flow_error_set(error, ENOTSUP,
 976                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 977                                           "count action not supported for "
 978                                           "egress");
 979         return 0;
 980 }
 981
 982 /**
 983  * Verify the @p attributes will be correctly understood by the NIC and store
 984  * them in the @p flow if everything is correct.
 985  *
 986  * @param[in] dev
 987  *   Pointer to the Ethernet device structure.
 988  * @param[in] attributes
 989  *   Pointer to flow attributes
 990  * @param[out] error
 991  *   Pointer to error structure.
 992  *
 993  * @return
 994  *   0 on success, a negative errno value otherwise and rte_errno is set.
 995  */
 996 int
 997 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
 998                               const struct rte_flow_attr *attributes,
 999                               struct rte_flow_error *error)
1000 {
1001         struct priv *priv = dev->data->dev_private;
1002         uint32_t priority_max = priv->config.flow_prio - 1;
1003
1004         if (attributes->group)
1005                 return rte_flow_error_set(error, ENOTSUP,
1006                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1007                                           NULL, "groups is not supported");
1008         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1009             attributes->priority >= priority_max)
1010                 return rte_flow_error_set(error, ENOTSUP,
1011                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1012                                           NULL, "priority out of range");
1013         if (attributes->egress)
1014                 return rte_flow_error_set(error, ENOTSUP,
1015                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1016                                           "egress is not supported");
1017         if (attributes->transfer)
1018                 return rte_flow_error_set(error, ENOTSUP,
1019                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1020                                           NULL, "transfer is not supported");
1021         if (!attributes->ingress)
1022                 return rte_flow_error_set(error, EINVAL,
1023                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1024                                           NULL,
1025                                           "ingress attribute is mandatory");
1026         return 0;
1027 }
1028
1029 /**
1030  * Validate Ethernet item.
1031  *
1032  * @param[in] item
1033  *   Item specification.
1034  * @param[in] item_flags
1035  *   Bit-fields that holds the items detected until now.
1036  * @param[out] error
1037  *   Pointer to error structure.
1038  *
1039  * @return
1040  *   0 on success, a negative errno value otherwise and rte_errno is set.
1041  */
1042 int
1043 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1044                             uint64_t item_flags,
1045                             struct rte_flow_error *error)
1046 {
1047         const struct rte_flow_item_eth *mask = item->mask;
1048         const struct rte_flow_item_eth nic_mask = {
1049                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1050                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1051                 .type = RTE_BE16(0xffff),
1052         };
1053         int ret;
1054         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1055         const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1056                                        MLX5_FLOW_LAYER_OUTER_L2;
1057
1058         if (item_flags & ethm)
1059                 return rte_flow_error_set(error, ENOTSUP,
1060                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1061                                           "multiple L2 layers not supported");
1062         if (!mask)
1063                 mask = &rte_flow_item_eth_mask;
1064         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1065                                         (const uint8_t *)&nic_mask,
1066                                         sizeof(struct rte_flow_item_eth),
1067                                         error);
1068         return ret;
1069 }
1070
1071 /**
1072  * Validate VLAN item.
1073  *
1074  * @param[in] item
1075  *   Item specification.
1076  * @param[in] item_flags
1077  *   Bit-fields that holds the items detected until now.
1078  * @param[out] error
1079  *   Pointer to error structure.
1080  *
1081  * @return
1082  *   0 on success, a negative errno value otherwise and rte_errno is set.
1083  */
1084 int
1085 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1086                              uint64_t item_flags,
1087                              struct rte_flow_error *error)
1088 {
1089         const struct rte_flow_item_vlan *spec = item->spec;
1090         const struct rte_flow_item_vlan *mask = item->mask;
1091         const struct rte_flow_item_vlan nic_mask = {
1092                 .tci = RTE_BE16(0x0fff),
1093                 .inner_type = RTE_BE16(0xffff),
1094         };
1095         uint16_t vlan_tag = 0;
1096         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1097         int ret;
1098         const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1099                                         MLX5_FLOW_LAYER_INNER_L4) :
1100                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1101                                         MLX5_FLOW_LAYER_OUTER_L4);
1102         const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1103                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1104
1105         if (item_flags & vlanm)
1106                 return rte_flow_error_set(error, EINVAL,
1107                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1108                                           "multiple VLAN layers not supported");
1109         else if ((item_flags & l34m) != 0)
1110                 return rte_flow_error_set(error, EINVAL,
1111                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1112                                           "L2 layer cannot follow L3/L4 layer");
1113         if (!mask)
1114                 mask = &rte_flow_item_vlan_mask;
1115         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1116                                         (const uint8_t *)&nic_mask,
1117                                         sizeof(struct rte_flow_item_vlan),
1118                                         error);
1119         if (ret)
1120                 return ret;
1121         if (spec) {
1122                 vlan_tag = spec->tci;
1123                 vlan_tag &= mask->tci;
1124         }
1125         /*
1126          * From verbs perspective an empty VLAN is equivalent
1127          * to a packet without VLAN layer.
1128          */
1129         if (!vlan_tag)
1130                 return rte_flow_error_set(error, EINVAL,
1131                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1132                                           item->spec,
1133                                           "VLAN cannot be empty");
1134         return 0;
1135 }
1136
1137 /**
1138  * Validate IPV4 item.
1139  *
1140  * @param[in] item
1141  *   Item specification.
1142  * @param[in] item_flags
1143  *   Bit-fields that holds the items detected until now.
1144  * @param[out] error
1145  *   Pointer to error structure.
1146  *
1147  * @return
1148  *   0 on success, a negative errno value otherwise and rte_errno is set.
1149  */
1150 int
1151 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1152                              uint64_t item_flags,
1153                              struct rte_flow_error *error)
1154 {
1155         const struct rte_flow_item_ipv4 *mask = item->mask;
1156         const struct rte_flow_item_ipv4 nic_mask = {
1157                 .hdr = {
1158                         .src_addr = RTE_BE32(0xffffffff),
1159                         .dst_addr = RTE_BE32(0xffffffff),
1160                         .type_of_service = 0xff,
1161                         .next_proto_id = 0xff,
1162                 },
1163         };
1164         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1165         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1166                                       MLX5_FLOW_LAYER_OUTER_L3;
1167         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1168                                       MLX5_FLOW_LAYER_OUTER_L4;
1169         int ret;
1170
1171         if (item_flags & l3m)
1172                 return rte_flow_error_set(error, ENOTSUP,
1173                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1174                                           "multiple L3 layers not supported");
1175         else if (item_flags & l4m)
1176                 return rte_flow_error_set(error, EINVAL,
1177                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1178                                           "L3 cannot follow an L4 layer.");
1179         if (!mask)
1180                 mask = &rte_flow_item_ipv4_mask;
1181         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1182                                         (const uint8_t *)&nic_mask,
1183                                         sizeof(struct rte_flow_item_ipv4),
1184                                         error);
1185         if (ret < 0)
1186                 return ret;
1187         return 0;
1188 }
1189
1190 /**
1191  * Validate IPV6 item.
1192  *
1193  * @param[in] item
1194  *   Item specification.
1195  * @param[in] item_flags
1196  *   Bit-fields that holds the items detected until now.
1197  * @param[out] error
1198  *   Pointer to error structure.
1199  *
1200  * @return
1201  *   0 on success, a negative errno value otherwise and rte_errno is set.
1202  */
1203 int
1204 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1205                              uint64_t item_flags,
1206                              struct rte_flow_error *error)
1207 {
1208         const struct rte_flow_item_ipv6 *mask = item->mask;
1209         const struct rte_flow_item_ipv6 nic_mask = {
1210                 .hdr = {
1211                         .src_addr =
1212                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1213                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1214                         .dst_addr =
1215                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1216                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1217                         .vtc_flow = RTE_BE32(0xffffffff),
1218                         .proto = 0xff,
1219                         .hop_limits = 0xff,
1220                 },
1221         };
1222         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1223         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1224                                       MLX5_FLOW_LAYER_OUTER_L3;
1225         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1226                                       MLX5_FLOW_LAYER_OUTER_L4;
1227         int ret;
1228
1229         if (item_flags & l3m)
1230                 return rte_flow_error_set(error, ENOTSUP,
1231                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1232                                           "multiple L3 layers not supported");
1233         else if (item_flags & l4m)
1234                 return rte_flow_error_set(error, EINVAL,
1235                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1236                                           "L3 cannot follow an L4 layer.");
1237         /*
1238          * IPv6 is not recognised by the NIC inside a GRE tunnel.
1239          * Such support has to be disabled as the rule will be
1240          * accepted.  Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1241          * Mellanox OFED 4.4-1.0.0.0.
1242          */
1243         if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1244                 return rte_flow_error_set(error, ENOTSUP,
1245                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1246                                           "IPv6 inside a GRE tunnel is"
1247                                           " not recognised.");
1248         if (!mask)
1249                 mask = &rte_flow_item_ipv6_mask;
1250         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1251                                         (const uint8_t *)&nic_mask,
1252                                         sizeof(struct rte_flow_item_ipv6),
1253                                         error);
1254         if (ret < 0)
1255                 return ret;
1256         return 0;
1257 }
1258
1259 /**
1260  * Validate UDP item.
1261  *
1262  * @param[in] item
1263  *   Item specification.
1264  * @param[in] item_flags
1265  *   Bit-fields that holds the items detected until now.
1266  * @param[in] target_protocol
1267  *   The next protocol in the previous item.
1268  * @param[in] flow_mask
1269  *   mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1270  * @param[out] error
1271  *   Pointer to error structure.
1272  *
1273  * @return
1274  *   0 on success, a negative errno value otherwise and rte_errno is set.
1275  */
1276 int
1277 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1278                             uint64_t item_flags,
1279                             uint8_t target_protocol,
1280                             struct rte_flow_error *error)
1281 {
1282         const struct rte_flow_item_udp *mask = item->mask;
1283         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1284         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1285                                       MLX5_FLOW_LAYER_OUTER_L3;
1286         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1287                                       MLX5_FLOW_LAYER_OUTER_L4;
1288         int ret;
1289
1290         if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1291                 return rte_flow_error_set(error, EINVAL,
1292                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1293                                           "protocol filtering not compatible"
1294                                           " with UDP layer");
1295         if (!(item_flags & l3m))
1296                 return rte_flow_error_set(error, EINVAL,
1297                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1298                                           "L3 is mandatory to filter on L4");
1299         if (item_flags & l4m)
1300                 return rte_flow_error_set(error, EINVAL,
1301                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1302                                           "multiple L4 layers not supported");
1303         if (!mask)
1304                 mask = &rte_flow_item_udp_mask;
1305         ret = mlx5_flow_item_acceptable
1306                 (item, (const uint8_t *)mask,
1307                  (const uint8_t *)&rte_flow_item_udp_mask,
1308                  sizeof(struct rte_flow_item_udp), error);
1309         if (ret < 0)
1310                 return ret;
1311         return 0;
1312 }
1313
1314 /**
1315  * Validate TCP item.
1316  *
1317  * @param[in] item
1318  *   Item specification.
1319  * @param[in] item_flags
1320  *   Bit-fields that holds the items detected until now.
1321  * @param[in] target_protocol
1322  *   The next protocol in the previous item.
1323  * @param[out] error
1324  *   Pointer to error structure.
1325  *
1326  * @return
1327  *   0 on success, a negative errno value otherwise and rte_errno is set.
1328  */
1329 int
1330 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1331                             uint64_t item_flags,
1332                             uint8_t target_protocol,
1333                             const struct rte_flow_item_tcp *flow_mask,
1334                             struct rte_flow_error *error)
1335 {
1336         const struct rte_flow_item_tcp *mask = item->mask;
1337         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1338         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1339                                       MLX5_FLOW_LAYER_OUTER_L3;
1340         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1341                                       MLX5_FLOW_LAYER_OUTER_L4;
1342         int ret;
1343
1344         assert(flow_mask);
1345         if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1346                 return rte_flow_error_set(error, EINVAL,
1347                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1348                                           "protocol filtering not compatible"
1349                                           " with TCP layer");
1350         if (!(item_flags & l3m))
1351                 return rte_flow_error_set(error, EINVAL,
1352                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1353                                           "L3 is mandatory to filter on L4");
1354         if (item_flags & l4m)
1355                 return rte_flow_error_set(error, EINVAL,
1356                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1357                                           "multiple L4 layers not supported");
1358         if (!mask)
1359                 mask = &rte_flow_item_tcp_mask;
1360         ret = mlx5_flow_item_acceptable
1361                 (item, (const uint8_t *)mask,
1362                  (const uint8_t *)flow_mask,
1363                  sizeof(struct rte_flow_item_tcp), error);
1364         if (ret < 0)
1365                 return ret;
1366         return 0;
1367 }
1368
1369 /**
1370  * Validate VXLAN item.
1371  *
1372  * @param[in] item
1373  *   Item specification.
1374  * @param[in] item_flags
1375  *   Bit-fields that holds the items detected until now.
1376  * @param[in] target_protocol
1377  *   The next protocol in the previous item.
1378  * @param[out] error
1379  *   Pointer to error structure.
1380  *
1381  * @return
1382  *   0 on success, a negative errno value otherwise and rte_errno is set.
1383  */
1384 int
1385 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1386                               uint64_t item_flags,
1387                               struct rte_flow_error *error)
1388 {
1389         const struct rte_flow_item_vxlan *spec = item->spec;
1390         const struct rte_flow_item_vxlan *mask = item->mask;
1391         int ret;
1392         union vni {
1393                 uint32_t vlan_id;
1394                 uint8_t vni[4];
1395         } id = { .vlan_id = 0, };
1396         uint32_t vlan_id = 0;
1397
1398
1399         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1400                 return rte_flow_error_set(error, ENOTSUP,
1401                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1402                                           "multiple tunnel layers not"
1403                                           " supported");
1404         /*
1405          * Verify only UDPv4 is present as defined in
1406          * https://tools.ietf.org/html/rfc7348
1407          */
1408         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1409                 return rte_flow_error_set(error, EINVAL,
1410                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1411                                           "no outer UDP layer found");
1412         if (!mask)
1413                 mask = &rte_flow_item_vxlan_mask;
1414         ret = mlx5_flow_item_acceptable
1415                 (item, (const uint8_t *)mask,
1416                  (const uint8_t *)&rte_flow_item_vxlan_mask,
1417                  sizeof(struct rte_flow_item_vxlan),
1418                  error);
1419         if (ret < 0)
1420                 return ret;
1421         if (spec) {
1422                 memcpy(&id.vni[1], spec->vni, 3);
1423                 vlan_id = id.vlan_id;
1424                 memcpy(&id.vni[1], mask->vni, 3);
1425                 vlan_id &= id.vlan_id;
1426         }
1427         /*
1428          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1429          * only this layer is defined in the Verbs specification it is
1430          * interpreted as wildcard and all packets will match this
1431          * rule, if it follows a full stack layer (ex: eth / ipv4 /
1432          * udp), all packets matching the layers before will also
1433          * match this rule.  To avoid such situation, VNI 0 is
1434          * currently refused.
1435          */
1436         if (!vlan_id)
1437                 return rte_flow_error_set(error, ENOTSUP,
1438                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1439                                           "VXLAN vni cannot be 0");
1440         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1441                 return rte_flow_error_set(error, ENOTSUP,
1442                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1443                                           "VXLAN tunnel must be fully defined");
1444         return 0;
1445 }
1446
1447 /**
1448  * Validate VXLAN_GPE item.
1449  *
1450  * @param[in] item
1451  *   Item specification.
1452  * @param[in] item_flags
1453  *   Bit-fields that holds the items detected until now.
1454  * @param[in] priv
1455  *   Pointer to the private data structure.
1456  * @param[in] target_protocol
1457  *   The next protocol in the previous item.
1458  * @param[out] error
1459  *   Pointer to error structure.
1460  *
1461  * @return
1462  *   0 on success, a negative errno value otherwise and rte_errno is set.
1463  */
1464 int
1465 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1466                                   uint64_t item_flags,
1467                                   struct rte_eth_dev *dev,
1468                                   struct rte_flow_error *error)
1469 {
1470         struct priv *priv = dev->data->dev_private;
1471         const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1472         const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1473         int ret;
1474         union vni {
1475                 uint32_t vlan_id;
1476                 uint8_t vni[4];
1477         } id = { .vlan_id = 0, };
1478         uint32_t vlan_id = 0;
1479
1480         if (!priv->config.l3_vxlan_en)
1481                 return rte_flow_error_set(error, ENOTSUP,
1482                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1483                                           "L3 VXLAN is not enabled by device"
1484                                           " parameter and/or not configured in"
1485                                           " firmware");
1486         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1487                 return rte_flow_error_set(error, ENOTSUP,
1488                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1489                                           "multiple tunnel layers not"
1490                                           " supported");
1491         /*
1492          * Verify only UDPv4 is present as defined in
1493          * https://tools.ietf.org/html/rfc7348
1494          */
1495         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1496                 return rte_flow_error_set(error, EINVAL,
1497                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1498                                           "no outer UDP layer found");
1499         if (!mask)
1500                 mask = &rte_flow_item_vxlan_gpe_mask;
1501         ret = mlx5_flow_item_acceptable
1502                 (item, (const uint8_t *)mask,
1503                  (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1504                  sizeof(struct rte_flow_item_vxlan_gpe),
1505                  error);
1506         if (ret < 0)
1507                 return ret;
1508         if (spec) {
1509                 if (spec->protocol)
1510                         return rte_flow_error_set(error, ENOTSUP,
1511                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1512                                                   item,
1513                                                   "VxLAN-GPE protocol"
1514                                                   " not supported");
1515                 memcpy(&id.vni[1], spec->vni, 3);
1516                 vlan_id = id.vlan_id;
1517                 memcpy(&id.vni[1], mask->vni, 3);
1518                 vlan_id &= id.vlan_id;
1519         }
1520         /*
1521          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1522          * layer is defined in the Verbs specification it is interpreted as
1523          * wildcard and all packets will match this rule, if it follows a full
1524          * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1525          * before will also match this rule.  To avoid such situation, VNI 0
1526          * is currently refused.
1527          */
1528         if (!vlan_id)
1529                 return rte_flow_error_set(error, ENOTSUP,
1530                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1531                                           "VXLAN-GPE vni cannot be 0");
1532         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1533                 return rte_flow_error_set(error, ENOTSUP,
1534                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1535                                           "VXLAN-GPE tunnel must be fully"
1536                                           " defined");
1537         return 0;
1538 }
1539
1540 /**
1541  * Validate GRE item.
1542  *
1543  * @param[in] item
1544  *   Item specification.
1545  * @param[in] item_flags
1546  *   Bit flags to mark detected items.
1547  * @param[in] target_protocol
1548  *   The next protocol in the previous item.
1549  * @param[out] error
1550  *   Pointer to error structure.
1551  *
1552  * @return
1553  *   0 on success, a negative errno value otherwise and rte_errno is set.
1554  */
1555 int
1556 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1557                             uint64_t item_flags,
1558                             uint8_t target_protocol,
1559                             struct rte_flow_error *error)
1560 {
1561         const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1562         const struct rte_flow_item_gre *mask = item->mask;
1563         int ret;
1564
1565         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1566                 return rte_flow_error_set(error, EINVAL,
1567                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1568                                           "protocol filtering not compatible"
1569                                           " with this GRE layer");
1570         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1571                 return rte_flow_error_set(error, ENOTSUP,
1572                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1573                                           "multiple tunnel layers not"
1574                                           " supported");
1575         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1576                 return rte_flow_error_set(error, ENOTSUP,
1577                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1578                                           "L3 Layer is missing");
1579         if (!mask)
1580                 mask = &rte_flow_item_gre_mask;
1581         ret = mlx5_flow_item_acceptable
1582                 (item, (const uint8_t *)mask,
1583                  (const uint8_t *)&rte_flow_item_gre_mask,
1584                  sizeof(struct rte_flow_item_gre), error);
1585         if (ret < 0)
1586                 return ret;
1587 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1588         if (spec && (spec->protocol & mask->protocol))
1589                 return rte_flow_error_set(error, ENOTSUP,
1590                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1591                                           "without MPLS support the"
1592                                           " specification cannot be used for"
1593                                           " filtering");
1594 #endif
1595         return 0;
1596 }
1597
1598 /**
1599  * Validate MPLS item.
1600  *
1601  * @param[in] item
1602  *   Item specification.
1603  * @param[in] item_flags
1604  *   Bit-fields that holds the items detected until now.
1605  * @param[in] target_protocol
1606  *   The next protocol in the previous item.
1607  * @param[out] error
1608  *   Pointer to error structure.
1609  *
1610  * @return
1611  *   0 on success, a negative errno value otherwise and rte_errno is set.
1612  */
1613 int
1614 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1615                              uint64_t item_flags __rte_unused,
1616                              uint8_t target_protocol __rte_unused,
1617                              struct rte_flow_error *error)
1618 {
1619 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1620         const struct rte_flow_item_mpls *mask = item->mask;
1621         int ret;
1622
1623         if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1624                 return rte_flow_error_set(error, EINVAL,
1625                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1626                                           "protocol filtering not compatible"
1627                                           " with MPLS layer");
1628         /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
1629         if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1630             !(item_flags & MLX5_FLOW_LAYER_GRE))
1631                 return rte_flow_error_set(error, ENOTSUP,
1632                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1633                                           "multiple tunnel layers not"
1634                                           " supported");
1635         if (!mask)
1636                 mask = &rte_flow_item_mpls_mask;
1637         ret = mlx5_flow_item_acceptable
1638                 (item, (const uint8_t *)mask,
1639                  (const uint8_t *)&rte_flow_item_mpls_mask,
1640                  sizeof(struct rte_flow_item_mpls), error);
1641         if (ret < 0)
1642                 return ret;
1643         return 0;
1644 #endif
1645         return rte_flow_error_set(error, ENOTSUP,
1646                                   RTE_FLOW_ERROR_TYPE_ITEM, item,
1647                                   "MPLS is not supported by Verbs, please"
1648                                   " update.");
1649 }
1650
1651 static int
1652 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1653                    const struct rte_flow_attr *attr __rte_unused,
1654                    const struct rte_flow_item items[] __rte_unused,
1655                    const struct rte_flow_action actions[] __rte_unused,
1656                    struct rte_flow_error *error __rte_unused)
1657 {
1658         rte_errno = ENOTSUP;
1659         return -rte_errno;
1660 }
1661
1662 static struct mlx5_flow *
1663 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1664                   const struct rte_flow_item items[] __rte_unused,
1665                   const struct rte_flow_action actions[] __rte_unused,
1666                   uint64_t *item_flags __rte_unused,
1667                   uint64_t *action_flags __rte_unused,
1668                   struct rte_flow_error *error __rte_unused)
1669 {
1670         rte_errno = ENOTSUP;
1671         return NULL;
1672 }
1673
1674 static int
1675 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1676                     struct mlx5_flow *dev_flow __rte_unused,
1677                     const struct rte_flow_attr *attr __rte_unused,
1678                     const struct rte_flow_item items[] __rte_unused,
1679                     const struct rte_flow_action actions[] __rte_unused,
1680                     struct rte_flow_error *error __rte_unused)
1681 {
1682         rte_errno = ENOTSUP;
1683         return -rte_errno;
1684 }
1685
1686 static int
1687 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1688                 struct rte_flow *flow __rte_unused,
1689                 struct rte_flow_error *error __rte_unused)
1690 {
1691         rte_errno = ENOTSUP;
1692         return -rte_errno;
1693 }
1694
1695 static void
1696 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1697                  struct rte_flow *flow __rte_unused)
1698 {
1699 }
1700
1701 static void
1702 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1703                   struct rte_flow *flow __rte_unused)
1704 {
1705 }
1706
1707 static int
1708 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1709                 struct rte_flow *flow __rte_unused,
1710                 const struct rte_flow_action *actions __rte_unused,
1711                 void *data __rte_unused,
1712                 struct rte_flow_error *error __rte_unused)
1713 {
1714         rte_errno = ENOTSUP;
1715         return -rte_errno;
1716 }
1717
1718 /* Void driver to protect from null pointer reference. */
1719 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1720         .validate = flow_null_validate,
1721         .prepare = flow_null_prepare,
1722         .translate = flow_null_translate,
1723         .apply = flow_null_apply,
1724         .remove = flow_null_remove,
1725         .destroy = flow_null_destroy,
1726         .query = flow_null_query,
1727 };
1728
1729 /**
1730  * Select flow driver type according to flow attributes and device
1731  * configuration.
1732  *
1733  * @param[in] dev
1734  *   Pointer to the dev structure.
1735  * @param[in] attr
1736  *   Pointer to the flow attributes.
1737  *
1738  * @return
1739  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1740  */
1741 static enum mlx5_flow_drv_type
1742 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1743 {
1744         struct priv *priv = dev->data->dev_private;
1745         enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1746
1747         if (attr->transfer)
1748                 type = MLX5_FLOW_TYPE_TCF;
1749         else
1750                 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1751                                                  MLX5_FLOW_TYPE_VERBS;
1752         return type;
1753 }
1754
1755 #define flow_get_drv_ops(type) flow_drv_ops[type]
1756
1757 /**
1758  * Flow driver validation API. This abstracts calling driver specific functions.
1759  * The type of flow driver is determined according to flow attributes.
1760  *
1761  * @param[in] dev
1762  *   Pointer to the dev structure.
1763  * @param[in] attr
1764  *   Pointer to the flow attributes.
1765  * @param[in] items
1766  *   Pointer to the list of items.
1767  * @param[in] actions
1768  *   Pointer to the list of actions.
1769  * @param[out] error
1770  *   Pointer to the error structure.
1771  *
1772  * @return
1773  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1774  */
1775 static inline int
1776 flow_drv_validate(struct rte_eth_dev *dev,
1777                   const struct rte_flow_attr *attr,
1778                   const struct rte_flow_item items[],
1779                   const struct rte_flow_action actions[],
1780                   struct rte_flow_error *error)
1781 {
1782         const struct mlx5_flow_driver_ops *fops;
1783         enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1784
1785         fops = flow_get_drv_ops(type);
1786         return fops->validate(dev, attr, items, actions, error);
1787 }
1788
1789 /**
1790  * Flow driver preparation API. This abstracts calling driver specific
1791  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1792  * calculates the size of memory required for device flow, allocates the memory,
1793  * initializes the device flow and returns the pointer.
1794  *
1795  * @param[in] attr
1796  *   Pointer to the flow attributes.
1797  * @param[in] items
1798  *   Pointer to the list of items.
1799  * @param[in] actions
1800  *   Pointer to the list of actions.
1801  * @param[out] item_flags
1802  *   Pointer to bit mask of all items detected.
1803  * @param[out] action_flags
1804  *   Pointer to bit mask of all actions detected.
1805  * @param[out] error
1806  *   Pointer to the error structure.
1807  *
1808  * @return
1809  *   Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1810  */
1811 static inline struct mlx5_flow *
1812 flow_drv_prepare(struct rte_flow *flow,
1813                  const struct rte_flow_attr *attr,
1814                  const struct rte_flow_item items[],
1815                  const struct rte_flow_action actions[],
1816                  uint64_t *item_flags,
1817                  uint64_t *action_flags,
1818                  struct rte_flow_error *error)
1819 {
1820         const struct mlx5_flow_driver_ops *fops;
1821         enum mlx5_flow_drv_type type = flow->drv_type;
1822
1823         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1824         fops = flow_get_drv_ops(type);
1825         return fops->prepare(attr, items, actions, item_flags, action_flags,
1826                              error);
1827 }
1828
1829 /**
1830  * Flow driver translation API. This abstracts calling driver specific
1831  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1832  * translates a generic flow into a driver flow. flow_drv_prepare() must
1833  * precede.
1834  *
1835  *
1836  * @param[in] dev
1837  *   Pointer to the rte dev structure.
1838  * @param[in, out] dev_flow
1839  *   Pointer to the mlx5 flow.
1840  * @param[in] attr
1841  *   Pointer to the flow attributes.
1842  * @param[in] items
1843  *   Pointer to the list of items.
1844  * @param[in] actions
1845  *   Pointer to the list of actions.
1846  * @param[out] error
1847  *   Pointer to the error structure.
1848  *
1849  * @return
1850  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1851  */
1852 static inline int
1853 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1854                    const struct rte_flow_attr *attr,
1855                    const struct rte_flow_item items[],
1856                    const struct rte_flow_action actions[],
1857                    struct rte_flow_error *error)
1858 {
1859         const struct mlx5_flow_driver_ops *fops;
1860         enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1861
1862         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1863         fops = flow_get_drv_ops(type);
1864         return fops->translate(dev, dev_flow, attr, items, actions, error);
1865 }
1866
1867 /**
1868  * Flow driver apply API. This abstracts calling driver specific functions.
1869  * Parent flow (rte_flow) should have driver type (drv_type). It applies
1870  * translated driver flows on to device. flow_drv_translate() must precede.
1871  *
1872  * @param[in] dev
1873  *   Pointer to Ethernet device structure.
1874  * @param[in, out] flow
1875  *   Pointer to flow structure.
1876  * @param[out] error
1877  *   Pointer to error structure.
1878  *
1879  * @return
1880  *   0 on success, a negative errno value otherwise and rte_errno is set.
1881  */
1882 static inline int
1883 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1884                struct rte_flow_error *error)
1885 {
1886         const struct mlx5_flow_driver_ops *fops;
1887         enum mlx5_flow_drv_type type = flow->drv_type;
1888
1889         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1890         fops = flow_get_drv_ops(type);
1891         return fops->apply(dev, flow, error);
1892 }
1893
1894 /**
1895  * Flow driver remove API. This abstracts calling driver specific functions.
1896  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1897  * on device. All the resources of the flow should be freed by calling
1898  * flow_dv_destroy().
1899  *
1900  * @param[in] dev
1901  *   Pointer to Ethernet device.
1902  * @param[in, out] flow
1903  *   Pointer to flow structure.
1904  */
1905 static inline void
1906 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
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         fops->remove(dev, flow);
1914 }
1915
1916 /**
1917  * Flow driver destroy 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 and releases resources of the flow.
1920  *
1921  * @param[in] dev
1922  *   Pointer to Ethernet device.
1923  * @param[in, out] flow
1924  *   Pointer to flow structure.
1925  */
1926 static inline void
1927 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1928 {
1929         const struct mlx5_flow_driver_ops *fops;
1930         enum mlx5_flow_drv_type type = flow->drv_type;
1931
1932         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1933         fops = flow_get_drv_ops(type);
1934         fops->destroy(dev, flow);
1935 }
1936
1937 /**
1938  * Validate a flow supported by the NIC.
1939  *
1940  * @see rte_flow_validate()
1941  * @see rte_flow_ops
1942  */
1943 int
1944 mlx5_flow_validate(struct rte_eth_dev *dev,
1945                    const struct rte_flow_attr *attr,
1946                    const struct rte_flow_item items[],
1947                    const struct rte_flow_action actions[],
1948                    struct rte_flow_error *error)
1949 {
1950         int ret;
1951
1952         ret = flow_drv_validate(dev, attr, items, actions, error);
1953         if (ret < 0)
1954                 return ret;
1955         return 0;
1956 }
1957
1958 /**
1959  * Get RSS action from the action list.
1960  *
1961  * @param[in] actions
1962  *   Pointer to the list of actions.
1963  *
1964  * @return
1965  *   Pointer to the RSS action if exist, else return NULL.
1966  */
1967 static const struct rte_flow_action_rss*
1968 flow_get_rss_action(const struct rte_flow_action actions[])
1969 {
1970         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1971                 switch (actions->type) {
1972                 case RTE_FLOW_ACTION_TYPE_RSS:
1973                         return (const struct rte_flow_action_rss *)
1974                                actions->conf;
1975                 default:
1976                         break;
1977                 }
1978         }
1979         return NULL;
1980 }
1981
1982 static unsigned int
1983 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
1984 {
1985         const struct rte_flow_item *item;
1986         unsigned int has_vlan = 0;
1987
1988         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
1989                 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
1990                         has_vlan = 1;
1991                         break;
1992                 }
1993         }
1994         if (has_vlan)
1995                 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
1996                                        MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
1997         return rss_level < 2 ? MLX5_EXPANSION_ROOT :
1998                                MLX5_EXPANSION_ROOT_OUTER;
1999 }
2000
2001 /**
2002  * Create a flow and add it to @p list.
2003  *
2004  * @param dev
2005  *   Pointer to Ethernet device.
2006  * @param list
2007  *   Pointer to a TAILQ flow list.
2008  * @param[in] attr
2009  *   Flow rule attributes.
2010  * @param[in] items
2011  *   Pattern specification (list terminated by the END pattern item).
2012  * @param[in] actions
2013  *   Associated actions (list terminated by the END action).
2014  * @param[out] error
2015  *   Perform verbose error reporting if not NULL.
2016  *
2017  * @return
2018  *   A flow on success, NULL otherwise and rte_errno is set.
2019  */
2020 static struct rte_flow *
2021 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2022                  const struct rte_flow_attr *attr,
2023                  const struct rte_flow_item items[],
2024                  const struct rte_flow_action actions[],
2025                  struct rte_flow_error *error)
2026 {
2027         struct rte_flow *flow = NULL;
2028         struct mlx5_flow *dev_flow;
2029         uint64_t action_flags = 0;
2030         uint64_t item_flags = 0;
2031         const struct rte_flow_action_rss *rss;
2032         union {
2033                 struct rte_flow_expand_rss buf;
2034                 uint8_t buffer[2048];
2035         } expand_buffer;
2036         struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2037         int ret;
2038         uint32_t i;
2039         uint32_t flow_size;
2040
2041         ret = flow_drv_validate(dev, attr, items, actions, error);
2042         if (ret < 0)
2043                 return NULL;
2044         flow_size = sizeof(struct rte_flow);
2045         rss = flow_get_rss_action(actions);
2046         if (rss)
2047                 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2048                                             sizeof(void *));
2049         else
2050                 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2051         flow = rte_calloc(__func__, 1, flow_size, 0);
2052         flow->drv_type = flow_get_drv_type(dev, attr);
2053         assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2054                flow->drv_type < MLX5_FLOW_TYPE_MAX);
2055         flow->queue = (void *)(flow + 1);
2056         LIST_INIT(&flow->dev_flows);
2057         if (rss && rss->types) {
2058                 unsigned int graph_root;
2059
2060                 graph_root = find_graph_root(items, rss->level);
2061                 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2062                                           items, rss->types,
2063                                           mlx5_support_expansion,
2064                                           graph_root);
2065                 assert(ret > 0 &&
2066                        (unsigned int)ret < sizeof(expand_buffer.buffer));
2067         } else {
2068                 buf->entries = 1;
2069                 buf->entry[0].pattern = (void *)(uintptr_t)items;
2070         }
2071         for (i = 0; i < buf->entries; ++i) {
2072                 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2073                                             actions, &item_flags, &action_flags,
2074                                             error);
2075                 if (!dev_flow)
2076                         goto error;
2077                 dev_flow->flow = flow;
2078                 dev_flow->layers = item_flags;
2079                 /* Store actions once as expanded flows have same actions. */
2080                 if (i == 0)
2081                         flow->actions = action_flags;
2082                 assert(flow->actions == action_flags);
2083                 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2084                 ret = flow_drv_translate(dev, dev_flow, attr,
2085                                          buf->entry[i].pattern,
2086                                          actions, error);
2087                 if (ret < 0)
2088                         goto error;
2089         }
2090         if (dev->data->dev_started) {
2091                 ret = flow_drv_apply(dev, flow, error);
2092                 if (ret < 0)
2093                         goto error;
2094         }
2095         TAILQ_INSERT_TAIL(list, flow, next);
2096         flow_rxq_flags_set(dev, flow);
2097         return flow;
2098 error:
2099         ret = rte_errno; /* Save rte_errno before cleanup. */
2100         assert(flow);
2101         flow_drv_destroy(dev, flow);
2102         rte_free(flow);
2103         rte_errno = ret; /* Restore rte_errno. */
2104         return NULL;
2105 }
2106
2107 /**
2108  * Create a flow.
2109  *
2110  * @see rte_flow_create()
2111  * @see rte_flow_ops
2112  */
2113 struct rte_flow *
2114 mlx5_flow_create(struct rte_eth_dev *dev,
2115                  const struct rte_flow_attr *attr,
2116                  const struct rte_flow_item items[],
2117                  const struct rte_flow_action actions[],
2118                  struct rte_flow_error *error)
2119 {
2120         return flow_list_create(dev,
2121                                 &((struct priv *)dev->data->dev_private)->flows,
2122                                 attr, items, actions, error);
2123 }
2124
2125 /**
2126  * Destroy a flow in a list.
2127  *
2128  * @param dev
2129  *   Pointer to Ethernet device.
2130  * @param list
2131  *   Pointer to a TAILQ flow list.
2132  * @param[in] flow
2133  *   Flow to destroy.
2134  */
2135 static void
2136 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2137                   struct rte_flow *flow)
2138 {
2139         flow_drv_destroy(dev, flow);
2140         TAILQ_REMOVE(list, flow, next);
2141         /*
2142          * Update RX queue flags only if port is started, otherwise it is
2143          * already clean.
2144          */
2145         if (dev->data->dev_started)
2146                 flow_rxq_flags_trim(dev, flow);
2147         rte_free(flow->fdir);
2148         rte_free(flow);
2149 }
2150
2151 /**
2152  * Destroy all flows.
2153  *
2154  * @param dev
2155  *   Pointer to Ethernet device.
2156  * @param list
2157  *   Pointer to a TAILQ flow list.
2158  */
2159 void
2160 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2161 {
2162         while (!TAILQ_EMPTY(list)) {
2163                 struct rte_flow *flow;
2164
2165                 flow = TAILQ_FIRST(list);
2166                 flow_list_destroy(dev, list, flow);
2167         }
2168 }
2169
2170 /**
2171  * Remove 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_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2180 {
2181         struct rte_flow *flow;
2182
2183         TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2184                 flow_drv_remove(dev, flow);
2185         flow_rxq_flags_clear(dev);
2186 }
2187
2188 /**
2189  * Add all flows.
2190  *
2191  * @param dev
2192  *   Pointer to Ethernet device.
2193  * @param list
2194  *   Pointer to a TAILQ flow list.
2195  *
2196  * @return
2197  *   0 on success, a negative errno value otherwise and rte_errno is set.
2198  */
2199 int
2200 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2201 {
2202         struct rte_flow *flow;
2203         struct rte_flow_error error;
2204         int ret = 0;
2205
2206         TAILQ_FOREACH(flow, list, next) {
2207                 ret = flow_drv_apply(dev, flow, &error);
2208                 if (ret < 0)
2209                         goto error;
2210                 flow_rxq_flags_set(dev, flow);
2211         }
2212         return 0;
2213 error:
2214         ret = rte_errno; /* Save rte_errno before cleanup. */
2215         mlx5_flow_stop(dev, list);
2216         rte_errno = ret; /* Restore rte_errno. */
2217         return -rte_errno;
2218 }
2219
2220 /**
2221  * Verify the flow list is empty
2222  *
2223  * @param dev
2224  *  Pointer to Ethernet device.
2225  *
2226  * @return the number of flows not released.
2227  */
2228 int
2229 mlx5_flow_verify(struct rte_eth_dev *dev)
2230 {
2231         struct priv *priv = dev->data->dev_private;
2232         struct rte_flow *flow;
2233         int ret = 0;
2234
2235         TAILQ_FOREACH(flow, &priv->flows, next) {
2236                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
2237                         dev->data->port_id, (void *)flow);
2238                 ++ret;
2239         }
2240         return ret;
2241 }
2242
2243 /**
2244  * Enable a control flow configured from the control plane.
2245  *
2246  * @param dev
2247  *   Pointer to Ethernet device.
2248  * @param eth_spec
2249  *   An Ethernet flow spec to apply.
2250  * @param eth_mask
2251  *   An Ethernet flow mask to apply.
2252  * @param vlan_spec
2253  *   A VLAN flow spec to apply.
2254  * @param vlan_mask
2255  *   A VLAN flow mask to apply.
2256  *
2257  * @return
2258  *   0 on success, a negative errno value otherwise and rte_errno is set.
2259  */
2260 int
2261 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2262                     struct rte_flow_item_eth *eth_spec,
2263                     struct rte_flow_item_eth *eth_mask,
2264                     struct rte_flow_item_vlan *vlan_spec,
2265                     struct rte_flow_item_vlan *vlan_mask)
2266 {
2267         struct priv *priv = dev->data->dev_private;
2268         const struct rte_flow_attr attr = {
2269                 .ingress = 1,
2270                 .priority = MLX5_FLOW_PRIO_RSVD,
2271         };
2272         struct rte_flow_item items[] = {
2273                 {
2274                         .type = RTE_FLOW_ITEM_TYPE_ETH,
2275                         .spec = eth_spec,
2276                         .last = NULL,
2277                         .mask = eth_mask,
2278                 },
2279                 {
2280                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2281                                               RTE_FLOW_ITEM_TYPE_END,
2282                         .spec = vlan_spec,
2283                         .last = NULL,
2284                         .mask = vlan_mask,
2285                 },
2286                 {
2287                         .type = RTE_FLOW_ITEM_TYPE_END,
2288                 },
2289         };
2290         uint16_t queue[priv->reta_idx_n];
2291         struct rte_flow_action_rss action_rss = {
2292                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2293                 .level = 0,
2294                 .types = priv->rss_conf.rss_hf,
2295                 .key_len = priv->rss_conf.rss_key_len,
2296                 .queue_num = priv->reta_idx_n,
2297                 .key = priv->rss_conf.rss_key,
2298                 .queue = queue,
2299         };
2300         struct rte_flow_action actions[] = {
2301                 {
2302                         .type = RTE_FLOW_ACTION_TYPE_RSS,
2303                         .conf = &action_rss,
2304                 },
2305                 {
2306                         .type = RTE_FLOW_ACTION_TYPE_END,
2307                 },
2308         };
2309         struct rte_flow *flow;
2310         struct rte_flow_error error;
2311         unsigned int i;
2312
2313         if (!priv->reta_idx_n) {
2314                 rte_errno = EINVAL;
2315                 return -rte_errno;
2316         }
2317         for (i = 0; i != priv->reta_idx_n; ++i)
2318                 queue[i] = (*priv->reta_idx)[i];
2319         flow = flow_list_create(dev, &priv->ctrl_flows,
2320                                 &attr, items, actions, &error);
2321         if (!flow)
2322                 return -rte_errno;
2323         return 0;
2324 }
2325
2326 /**
2327  * Enable a flow control configured from the control plane.
2328  *
2329  * @param dev
2330  *   Pointer to Ethernet device.
2331  * @param eth_spec
2332  *   An Ethernet flow spec to apply.
2333  * @param eth_mask
2334  *   An Ethernet flow mask to apply.
2335  *
2336  * @return
2337  *   0 on success, a negative errno value otherwise and rte_errno is set.
2338  */
2339 int
2340 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2341                struct rte_flow_item_eth *eth_spec,
2342                struct rte_flow_item_eth *eth_mask)
2343 {
2344         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2345 }
2346
2347 /**
2348  * Destroy a flow.
2349  *
2350  * @see rte_flow_destroy()
2351  * @see rte_flow_ops
2352  */
2353 int
2354 mlx5_flow_destroy(struct rte_eth_dev *dev,
2355                   struct rte_flow *flow,
2356                   struct rte_flow_error *error __rte_unused)
2357 {
2358         struct priv *priv = dev->data->dev_private;
2359
2360         flow_list_destroy(dev, &priv->flows, flow);
2361         return 0;
2362 }
2363
2364 /**
2365  * Destroy all flows.
2366  *
2367  * @see rte_flow_flush()
2368  * @see rte_flow_ops
2369  */
2370 int
2371 mlx5_flow_flush(struct rte_eth_dev *dev,
2372                 struct rte_flow_error *error __rte_unused)
2373 {
2374         struct priv *priv = dev->data->dev_private;
2375
2376         mlx5_flow_list_flush(dev, &priv->flows);
2377         return 0;
2378 }
2379
2380 /**
2381  * Isolated mode.
2382  *
2383  * @see rte_flow_isolate()
2384  * @see rte_flow_ops
2385  */
2386 int
2387 mlx5_flow_isolate(struct rte_eth_dev *dev,
2388                   int enable,
2389                   struct rte_flow_error *error)
2390 {
2391         struct priv *priv = dev->data->dev_private;
2392
2393         if (dev->data->dev_started) {
2394                 rte_flow_error_set(error, EBUSY,
2395                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2396                                    NULL,
2397                                    "port must be stopped first");
2398                 return -rte_errno;
2399         }
2400         priv->isolated = !!enable;
2401         if (enable)
2402                 dev->dev_ops = &mlx5_dev_ops_isolate;
2403         else
2404                 dev->dev_ops = &mlx5_dev_ops;
2405         return 0;
2406 }
2407
2408 /**
2409  * Query a flow.
2410  *
2411  * @see rte_flow_query()
2412  * @see rte_flow_ops
2413  */
2414 static int
2415 flow_drv_query(struct rte_eth_dev *dev,
2416                struct rte_flow *flow,
2417                const struct rte_flow_action *actions,
2418                void *data,
2419                struct rte_flow_error *error)
2420 {
2421         const struct mlx5_flow_driver_ops *fops;
2422         enum mlx5_flow_drv_type ftype = flow->drv_type;
2423
2424         assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2425         fops = flow_get_drv_ops(ftype);
2426
2427         return fops->query(dev, flow, actions, data, error);
2428 }
2429
2430 /**
2431  * Query a flow.
2432  *
2433  * @see rte_flow_query()
2434  * @see rte_flow_ops
2435  */
2436 int
2437 mlx5_flow_query(struct rte_eth_dev *dev,
2438                 struct rte_flow *flow,
2439                 const struct rte_flow_action *actions,
2440                 void *data,
2441                 struct rte_flow_error *error)
2442 {
2443         int ret;
2444
2445         ret = flow_drv_query(dev, flow, actions, data, error);
2446         if (ret < 0)
2447                 return ret;
2448         return 0;
2449 }
2450
2451 /**
2452  * Convert a flow director filter to a generic flow.
2453  *
2454  * @param dev
2455  *   Pointer to Ethernet device.
2456  * @param fdir_filter
2457  *   Flow director filter to add.
2458  * @param attributes
2459  *   Generic flow parameters structure.
2460  *
2461  * @return
2462  *   0 on success, a negative errno value otherwise and rte_errno is set.
2463  */
2464 static int
2465 flow_fdir_filter_convert(struct rte_eth_dev *dev,
2466                          const struct rte_eth_fdir_filter *fdir_filter,
2467                          struct mlx5_fdir *attributes)
2468 {
2469         struct priv *priv = dev->data->dev_private;
2470         const struct rte_eth_fdir_input *input = &fdir_filter->input;
2471         const struct rte_eth_fdir_masks *mask =
2472                 &dev->data->dev_conf.fdir_conf.mask;
2473
2474         /* Validate queue number. */
2475         if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2476                 DRV_LOG(ERR, "port %u invalid queue number %d",
2477                         dev->data->port_id, fdir_filter->action.rx_queue);
2478                 rte_errno = EINVAL;
2479                 return -rte_errno;
2480         }
2481         attributes->attr.ingress = 1;
2482         attributes->items[0] = (struct rte_flow_item) {
2483                 .type = RTE_FLOW_ITEM_TYPE_ETH,
2484                 .spec = &attributes->l2,
2485                 .mask = &attributes->l2_mask,
2486         };
2487         switch (fdir_filter->action.behavior) {
2488         case RTE_ETH_FDIR_ACCEPT:
2489                 attributes->actions[0] = (struct rte_flow_action){
2490                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
2491                         .conf = &attributes->queue,
2492                 };
2493                 break;
2494         case RTE_ETH_FDIR_REJECT:
2495                 attributes->actions[0] = (struct rte_flow_action){
2496                         .type = RTE_FLOW_ACTION_TYPE_DROP,
2497                 };
2498                 break;
2499         default:
2500                 DRV_LOG(ERR, "port %u invalid behavior %d",
2501                         dev->data->port_id,
2502                         fdir_filter->action.behavior);
2503                 rte_errno = ENOTSUP;
2504                 return -rte_errno;
2505         }
2506         attributes->queue.index = fdir_filter->action.rx_queue;
2507         /* Handle L3. */
2508         switch (fdir_filter->input.flow_type) {
2509         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2510         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2511         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2512                 attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2513                         .src_addr = input->flow.ip4_flow.src_ip,
2514                         .dst_addr = input->flow.ip4_flow.dst_ip,
2515                         .time_to_live = input->flow.ip4_flow.ttl,
2516                         .type_of_service = input->flow.ip4_flow.tos,
2517                 };
2518                 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2519                         .src_addr = mask->ipv4_mask.src_ip,
2520                         .dst_addr = mask->ipv4_mask.dst_ip,
2521                         .time_to_live = mask->ipv4_mask.ttl,
2522                         .type_of_service = mask->ipv4_mask.tos,
2523                         .next_proto_id = mask->ipv4_mask.proto,
2524                 };
2525                 attributes->items[1] = (struct rte_flow_item){
2526                         .type = RTE_FLOW_ITEM_TYPE_IPV4,
2527                         .spec = &attributes->l3,
2528                         .mask = &attributes->l3_mask,
2529                 };
2530                 break;
2531         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2532         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2533         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2534                 attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2535                         .hop_limits = input->flow.ipv6_flow.hop_limits,
2536                         .proto = input->flow.ipv6_flow.proto,
2537                 };
2538
2539                 memcpy(attributes->l3.ipv6.hdr.src_addr,
2540                        input->flow.ipv6_flow.src_ip,
2541                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2542                 memcpy(attributes->l3.ipv6.hdr.dst_addr,
2543                        input->flow.ipv6_flow.dst_ip,
2544                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2545                 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2546                        mask->ipv6_mask.src_ip,
2547                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2548                 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2549                        mask->ipv6_mask.dst_ip,
2550                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2551                 attributes->items[1] = (struct rte_flow_item){
2552                         .type = RTE_FLOW_ITEM_TYPE_IPV6,
2553                         .spec = &attributes->l3,
2554                         .mask = &attributes->l3_mask,
2555                 };
2556                 break;
2557         default:
2558                 DRV_LOG(ERR, "port %u invalid flow type%d",
2559                         dev->data->port_id, fdir_filter->input.flow_type);
2560                 rte_errno = ENOTSUP;
2561                 return -rte_errno;
2562         }
2563         /* Handle L4. */
2564         switch (fdir_filter->input.flow_type) {
2565         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2566                 attributes->l4.udp.hdr = (struct udp_hdr){
2567                         .src_port = input->flow.udp4_flow.src_port,
2568                         .dst_port = input->flow.udp4_flow.dst_port,
2569                 };
2570                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2571                         .src_port = mask->src_port_mask,
2572                         .dst_port = mask->dst_port_mask,
2573                 };
2574                 attributes->items[2] = (struct rte_flow_item){
2575                         .type = RTE_FLOW_ITEM_TYPE_UDP,
2576                         .spec = &attributes->l4,
2577                         .mask = &attributes->l4_mask,
2578                 };
2579                 break;
2580         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2581                 attributes->l4.tcp.hdr = (struct tcp_hdr){
2582                         .src_port = input->flow.tcp4_flow.src_port,
2583                         .dst_port = input->flow.tcp4_flow.dst_port,
2584                 };
2585                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2586                         .src_port = mask->src_port_mask,
2587                         .dst_port = mask->dst_port_mask,
2588                 };
2589                 attributes->items[2] = (struct rte_flow_item){
2590                         .type = RTE_FLOW_ITEM_TYPE_TCP,
2591                         .spec = &attributes->l4,
2592                         .mask = &attributes->l4_mask,
2593                 };
2594                 break;
2595         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2596                 attributes->l4.udp.hdr = (struct udp_hdr){
2597                         .src_port = input->flow.udp6_flow.src_port,
2598                         .dst_port = input->flow.udp6_flow.dst_port,
2599                 };
2600                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2601                         .src_port = mask->src_port_mask,
2602                         .dst_port = mask->dst_port_mask,
2603                 };
2604                 attributes->items[2] = (struct rte_flow_item){
2605                         .type = RTE_FLOW_ITEM_TYPE_UDP,
2606                         .spec = &attributes->l4,
2607                         .mask = &attributes->l4_mask,
2608                 };
2609                 break;
2610         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2611                 attributes->l4.tcp.hdr = (struct tcp_hdr){
2612                         .src_port = input->flow.tcp6_flow.src_port,
2613                         .dst_port = input->flow.tcp6_flow.dst_port,
2614                 };
2615                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2616                         .src_port = mask->src_port_mask,
2617                         .dst_port = mask->dst_port_mask,
2618                 };
2619                 attributes->items[2] = (struct rte_flow_item){
2620                         .type = RTE_FLOW_ITEM_TYPE_TCP,
2621                         .spec = &attributes->l4,
2622                         .mask = &attributes->l4_mask,
2623                 };
2624                 break;
2625         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2626         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2627                 break;
2628         default:
2629                 DRV_LOG(ERR, "port %u invalid flow type%d",
2630                         dev->data->port_id, fdir_filter->input.flow_type);
2631                 rte_errno = ENOTSUP;
2632                 return -rte_errno;
2633         }
2634         return 0;
2635 }
2636
2637 #define FLOW_FDIR_CMP(f1, f2, fld) \
2638         memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
2639
2640 /**
2641  * Compare two FDIR flows. If items and actions are identical, the two flows are
2642  * regarded as same.
2643  *
2644  * @param dev
2645  *   Pointer to Ethernet device.
2646  * @param f1
2647  *   FDIR flow to compare.
2648  * @param f2
2649  *   FDIR flow to compare.
2650  *
2651  * @return
2652  *   Zero on match, 1 otherwise.
2653  */
2654 static int
2655 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
2656 {
2657         if (FLOW_FDIR_CMP(f1, f2, attr) ||
2658             FLOW_FDIR_CMP(f1, f2, l2) ||
2659             FLOW_FDIR_CMP(f1, f2, l2_mask) ||
2660             FLOW_FDIR_CMP(f1, f2, l3) ||
2661             FLOW_FDIR_CMP(f1, f2, l3_mask) ||
2662             FLOW_FDIR_CMP(f1, f2, l4) ||
2663             FLOW_FDIR_CMP(f1, f2, l4_mask) ||
2664             FLOW_FDIR_CMP(f1, f2, actions[0]))
2665                 return 1;
2666         if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
2667             FLOW_FDIR_CMP(f1, f2, queue))
2668                 return 1;
2669         return 0;
2670 }
2671
2672 /**
2673  * Search device flow list to find out a matched FDIR flow.
2674  *
2675  * @param dev
2676  *   Pointer to Ethernet device.
2677  * @param fdir_flow
2678  *   FDIR flow to lookup.
2679  *
2680  * @return
2681  *   Pointer of flow if found, NULL otherwise.
2682  */
2683 static struct rte_flow *
2684 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
2685 {
2686         struct priv *priv = dev->data->dev_private;
2687         struct rte_flow *flow = NULL;
2688
2689         assert(fdir_flow);
2690         TAILQ_FOREACH(flow, &priv->flows, next) {
2691                 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
2692                         DRV_LOG(DEBUG, "port %u found FDIR flow %p",
2693                                 dev->data->port_id, (void *)flow);
2694                         break;
2695                 }
2696         }
2697         return flow;
2698 }
2699
2700 /**
2701  * Add new flow director filter and store it in list.
2702  *
2703  * @param dev
2704  *   Pointer to Ethernet device.
2705  * @param fdir_filter
2706  *   Flow director filter to add.
2707  *
2708  * @return
2709  *   0 on success, a negative errno value otherwise and rte_errno is set.
2710  */
2711 static int
2712 flow_fdir_filter_add(struct rte_eth_dev *dev,
2713                      const struct rte_eth_fdir_filter *fdir_filter)
2714 {
2715         struct priv *priv = dev->data->dev_private;
2716         struct mlx5_fdir *fdir_flow;
2717         struct rte_flow *flow;
2718         int ret;
2719
2720         fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
2721         if (!fdir_flow) {
2722                 rte_errno = ENOMEM;
2723                 return -rte_errno;
2724         }
2725         ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
2726         if (ret)
2727                 goto error;
2728         flow = flow_fdir_filter_lookup(dev, fdir_flow);
2729         if (flow) {
2730                 rte_errno = EEXIST;
2731                 goto error;
2732         }
2733         flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
2734                                 fdir_flow->items, fdir_flow->actions, NULL);
2735         if (!flow)
2736                 goto error;
2737         assert(!flow->fdir);
2738         flow->fdir = fdir_flow;
2739         DRV_LOG(DEBUG, "port %u created FDIR flow %p",
2740                 dev->data->port_id, (void *)flow);
2741         return 0;
2742 error:
2743         rte_free(fdir_flow);
2744         return -rte_errno;
2745 }
2746
2747 /**
2748  * Delete specific filter.
2749  *
2750  * @param dev
2751  *   Pointer to Ethernet device.
2752  * @param fdir_filter
2753  *   Filter to be deleted.
2754  *
2755  * @return
2756  *   0 on success, a negative errno value otherwise and rte_errno is set.
2757  */
2758 static int
2759 flow_fdir_filter_delete(struct rte_eth_dev *dev,
2760                         const struct rte_eth_fdir_filter *fdir_filter)
2761 {
2762         struct priv *priv = dev->data->dev_private;
2763         struct rte_flow *flow;
2764         struct mlx5_fdir fdir_flow = {
2765                 .attr.group = 0,
2766         };
2767         int ret;
2768
2769         ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
2770         if (ret)
2771                 return -rte_errno;
2772         flow = flow_fdir_filter_lookup(dev, &fdir_flow);
2773         if (!flow) {
2774                 rte_errno = ENOENT;
2775                 return -rte_errno;
2776         }
2777         flow_list_destroy(dev, &priv->flows, flow);
2778         DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
2779                 dev->data->port_id, (void *)flow);
2780         return 0;
2781 }
2782
2783 /**
2784  * Update queue for specific filter.
2785  *
2786  * @param dev
2787  *   Pointer to Ethernet device.
2788  * @param fdir_filter
2789  *   Filter to be updated.
2790  *
2791  * @return
2792  *   0 on success, a negative errno value otherwise and rte_errno is set.
2793  */
2794 static int
2795 flow_fdir_filter_update(struct rte_eth_dev *dev,
2796                         const struct rte_eth_fdir_filter *fdir_filter)
2797 {
2798         int ret;
2799
2800         ret = flow_fdir_filter_delete(dev, fdir_filter);
2801         if (ret)
2802                 return ret;
2803         return flow_fdir_filter_add(dev, fdir_filter);
2804 }
2805
2806 /**
2807  * Flush all filters.
2808  *
2809  * @param dev
2810  *   Pointer to Ethernet device.
2811  */
2812 static void
2813 flow_fdir_filter_flush(struct rte_eth_dev *dev)
2814 {
2815         struct priv *priv = dev->data->dev_private;
2816
2817         mlx5_flow_list_flush(dev, &priv->flows);
2818 }
2819
2820 /**
2821  * Get flow director information.
2822  *
2823  * @param dev
2824  *   Pointer to Ethernet device.
2825  * @param[out] fdir_info
2826  *   Resulting flow director information.
2827  */
2828 static void
2829 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2830 {
2831         struct rte_eth_fdir_masks *mask =
2832                 &dev->data->dev_conf.fdir_conf.mask;
2833
2834         fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2835         fdir_info->guarant_spc = 0;
2836         rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2837         fdir_info->max_flexpayload = 0;
2838         fdir_info->flow_types_mask[0] = 0;
2839         fdir_info->flex_payload_unit = 0;
2840         fdir_info->max_flex_payload_segment_num = 0;
2841         fdir_info->flex_payload_limit = 0;
2842         memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2843 }
2844
2845 /**
2846  * Deal with flow director operations.
2847  *
2848  * @param dev
2849  *   Pointer to Ethernet device.
2850  * @param filter_op
2851  *   Operation to perform.
2852  * @param arg
2853  *   Pointer to operation-specific structure.
2854  *
2855  * @return
2856  *   0 on success, a negative errno value otherwise and rte_errno is set.
2857  */
2858 static int
2859 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2860                     void *arg)
2861 {
2862         enum rte_fdir_mode fdir_mode =
2863                 dev->data->dev_conf.fdir_conf.mode;
2864
2865         if (filter_op == RTE_ETH_FILTER_NOP)
2866                 return 0;
2867         if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2868             fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2869                 DRV_LOG(ERR, "port %u flow director mode %d not supported",
2870                         dev->data->port_id, fdir_mode);
2871                 rte_errno = EINVAL;
2872                 return -rte_errno;
2873         }
2874         switch (filter_op) {
2875         case RTE_ETH_FILTER_ADD:
2876                 return flow_fdir_filter_add(dev, arg);
2877         case RTE_ETH_FILTER_UPDATE:
2878                 return flow_fdir_filter_update(dev, arg);
2879         case RTE_ETH_FILTER_DELETE:
2880                 return flow_fdir_filter_delete(dev, arg);
2881         case RTE_ETH_FILTER_FLUSH:
2882                 flow_fdir_filter_flush(dev);
2883                 break;
2884         case RTE_ETH_FILTER_INFO:
2885                 flow_fdir_info_get(dev, arg);
2886                 break;
2887         default:
2888                 DRV_LOG(DEBUG, "port %u unknown operation %u",
2889                         dev->data->port_id, filter_op);
2890                 rte_errno = EINVAL;
2891                 return -rte_errno;
2892         }
2893         return 0;
2894 }
2895
2896 /**
2897  * Manage filter operations.
2898  *
2899  * @param dev
2900  *   Pointer to Ethernet device structure.
2901  * @param filter_type
2902  *   Filter type.
2903  * @param filter_op
2904  *   Operation to perform.
2905  * @param arg
2906  *   Pointer to operation-specific structure.
2907  *
2908  * @return
2909  *   0 on success, a negative errno value otherwise and rte_errno is set.
2910  */
2911 int
2912 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2913                      enum rte_filter_type filter_type,
2914                      enum rte_filter_op filter_op,
2915                      void *arg)
2916 {
2917         switch (filter_type) {
2918         case RTE_ETH_FILTER_GENERIC:
2919                 if (filter_op != RTE_ETH_FILTER_GET) {
2920                         rte_errno = EINVAL;
2921                         return -rte_errno;
2922                 }
2923                 *(const void **)arg = &mlx5_flow_ops;
2924                 return 0;
2925         case RTE_ETH_FILTER_FDIR:
2926                 return flow_fdir_ctrl_func(dev, filter_op, arg);
2927         default:
2928                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
2929                         dev->data->port_id, filter_type);
2930                 rte_errno = ENOTSUP;
2931                 return -rte_errno;
2932         }
2933         return 0;
2934 }