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         if (rss->key_len < MLX5_RSS_HASH_KEY_LEN)
 916                 return rte_flow_error_set(error, ENOTSUP,
 917                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 918                                           &rss->key_len,
 919                                           "RSS hash key too small");
 920         if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
 921                 return rte_flow_error_set(error, ENOTSUP,
 922                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 923                                           &rss->key_len,
 924                                           "RSS hash key too large");
 925         if (rss->queue_num > priv->config.ind_table_max_size)
 926                 return rte_flow_error_set(error, ENOTSUP,
 927                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 928                                           &rss->queue_num,
 929                                           "number of queues too large");
 930         if (rss->types & MLX5_RSS_HF_MASK)
 931                 return rte_flow_error_set(error, ENOTSUP,
 932                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 933                                           &rss->types,
 934                                           "some RSS protocols are not"
 935                                           " supported");
 936         for (i = 0; i != rss->queue_num; ++i) {
 937                 if (!(*priv->rxqs)[rss->queue[i]])
 938                         return rte_flow_error_set
 939                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 940                                  &rss->queue[i], "queue is not configured");
 941         }
 942         if (attr->egress)
 943                 return rte_flow_error_set(error, ENOTSUP,
 944                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 945                                           "rss action not supported for "
 946                                           "egress");
 947         return 0;
 948 }
 949
 950 /*
 951  * Validate the count action.
 952  *
 953  * @param[in] dev
 954  *   Pointer to the Ethernet device structure.
 955  * @param[in] attr
 956  *   Attributes of flow that includes this action.
 957  * @param[out] error
 958  *   Pointer to error structure.
 959  *
 960  * @return
 961  *   0 on success, a negative errno value otherwise and rte_ernno is set.
 962  */
 963 int
 964 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
 965                                 const struct rte_flow_attr *attr,
 966                                 struct rte_flow_error *error)
 967 {
 968         if (attr->egress)
 969                 return rte_flow_error_set(error, ENOTSUP,
 970                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 971                                           "count action not supported for "
 972                                           "egress");
 973         return 0;
 974 }
 975
 976 /**
 977  * Verify the @p attributes will be correctly understood by the NIC and store
 978  * them in the @p flow if everything is correct.
 979  *
 980  * @param[in] dev
 981  *   Pointer to the Ethernet device structure.
 982  * @param[in] attributes
 983  *   Pointer to flow attributes
 984  * @param[out] error
 985  *   Pointer to error structure.
 986  *
 987  * @return
 988  *   0 on success, a negative errno value otherwise and rte_errno is set.
 989  */
 990 int
 991 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
 992                               const struct rte_flow_attr *attributes,
 993                               struct rte_flow_error *error)
 994 {
 995         struct priv *priv = dev->data->dev_private;
 996         uint32_t priority_max = priv->config.flow_prio - 1;
 997
 998         if (attributes->group)
 999                 return rte_flow_error_set(error, ENOTSUP,
1000                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1001                                           NULL, "groups is not supported");
1002         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1003             attributes->priority >= priority_max)
1004                 return rte_flow_error_set(error, ENOTSUP,
1005                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1006                                           NULL, "priority out of range");
1007         if (attributes->egress)
1008                 return rte_flow_error_set(error, ENOTSUP,
1009                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1010                                           "egress is not supported");
1011         if (attributes->transfer)
1012                 return rte_flow_error_set(error, ENOTSUP,
1013                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1014                                           NULL, "transfer is not supported");
1015         if (!attributes->ingress)
1016                 return rte_flow_error_set(error, EINVAL,
1017                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1018                                           NULL,
1019                                           "ingress attribute is mandatory");
1020         return 0;
1021 }
1022
1023 /**
1024  * Validate Ethernet item.
1025  *
1026  * @param[in] item
1027  *   Item specification.
1028  * @param[in] item_flags
1029  *   Bit-fields that holds the items detected until now.
1030  * @param[out] error
1031  *   Pointer to error structure.
1032  *
1033  * @return
1034  *   0 on success, a negative errno value otherwise and rte_errno is set.
1035  */
1036 int
1037 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1038                             uint64_t item_flags,
1039                             struct rte_flow_error *error)
1040 {
1041         const struct rte_flow_item_eth *mask = item->mask;
1042         const struct rte_flow_item_eth nic_mask = {
1043                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1044                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1045                 .type = RTE_BE16(0xffff),
1046         };
1047         int ret;
1048         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1049         const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1050                                        MLX5_FLOW_LAYER_OUTER_L2;
1051
1052         if (item_flags & ethm)
1053                 return rte_flow_error_set(error, ENOTSUP,
1054                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1055                                           "multiple L2 layers not supported");
1056         if (!mask)
1057                 mask = &rte_flow_item_eth_mask;
1058         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1059                                         (const uint8_t *)&nic_mask,
1060                                         sizeof(struct rte_flow_item_eth),
1061                                         error);
1062         return ret;
1063 }
1064
1065 /**
1066  * Validate VLAN item.
1067  *
1068  * @param[in] item
1069  *   Item specification.
1070  * @param[in] item_flags
1071  *   Bit-fields that holds the items detected until now.
1072  * @param[out] error
1073  *   Pointer to error structure.
1074  *
1075  * @return
1076  *   0 on success, a negative errno value otherwise and rte_errno is set.
1077  */
1078 int
1079 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1080                              uint64_t item_flags,
1081                              struct rte_flow_error *error)
1082 {
1083         const struct rte_flow_item_vlan *spec = item->spec;
1084         const struct rte_flow_item_vlan *mask = item->mask;
1085         const struct rte_flow_item_vlan nic_mask = {
1086                 .tci = RTE_BE16(0x0fff),
1087                 .inner_type = RTE_BE16(0xffff),
1088         };
1089         uint16_t vlan_tag = 0;
1090         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1091         int ret;
1092         const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1093                                         MLX5_FLOW_LAYER_INNER_L4) :
1094                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1095                                         MLX5_FLOW_LAYER_OUTER_L4);
1096         const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1097                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1098
1099         if (item_flags & vlanm)
1100                 return rte_flow_error_set(error, EINVAL,
1101                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1102                                           "multiple VLAN layers not supported");
1103         else if ((item_flags & l34m) != 0)
1104                 return rte_flow_error_set(error, EINVAL,
1105                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1106                                           "L2 layer cannot follow L3/L4 layer");
1107         if (!mask)
1108                 mask = &rte_flow_item_vlan_mask;
1109         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1110                                         (const uint8_t *)&nic_mask,
1111                                         sizeof(struct rte_flow_item_vlan),
1112                                         error);
1113         if (ret)
1114                 return ret;
1115         if (spec) {
1116                 vlan_tag = spec->tci;
1117                 vlan_tag &= mask->tci;
1118         }
1119         /*
1120          * From verbs perspective an empty VLAN is equivalent
1121          * to a packet without VLAN layer.
1122          */
1123         if (!vlan_tag)
1124                 return rte_flow_error_set(error, EINVAL,
1125                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1126                                           item->spec,
1127                                           "VLAN cannot be empty");
1128         return 0;
1129 }
1130
1131 /**
1132  * Validate IPV4 item.
1133  *
1134  * @param[in] item
1135  *   Item specification.
1136  * @param[in] item_flags
1137  *   Bit-fields that holds the items detected until now.
1138  * @param[out] error
1139  *   Pointer to error structure.
1140  *
1141  * @return
1142  *   0 on success, a negative errno value otherwise and rte_errno is set.
1143  */
1144 int
1145 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1146                              uint64_t item_flags,
1147                              struct rte_flow_error *error)
1148 {
1149         const struct rte_flow_item_ipv4 *mask = item->mask;
1150         const struct rte_flow_item_ipv4 nic_mask = {
1151                 .hdr = {
1152                         .src_addr = RTE_BE32(0xffffffff),
1153                         .dst_addr = RTE_BE32(0xffffffff),
1154                         .type_of_service = 0xff,
1155                         .next_proto_id = 0xff,
1156                 },
1157         };
1158         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1159         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1160                                       MLX5_FLOW_LAYER_OUTER_L3;
1161         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1162                                       MLX5_FLOW_LAYER_OUTER_L4;
1163         int ret;
1164
1165         if (item_flags & l3m)
1166                 return rte_flow_error_set(error, ENOTSUP,
1167                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1168                                           "multiple L3 layers not supported");
1169         else if (item_flags & l4m)
1170                 return rte_flow_error_set(error, EINVAL,
1171                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1172                                           "L3 cannot follow an L4 layer.");
1173         if (!mask)
1174                 mask = &rte_flow_item_ipv4_mask;
1175         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1176                                         (const uint8_t *)&nic_mask,
1177                                         sizeof(struct rte_flow_item_ipv4),
1178                                         error);
1179         if (ret < 0)
1180                 return ret;
1181         return 0;
1182 }
1183
1184 /**
1185  * Validate IPV6 item.
1186  *
1187  * @param[in] item
1188  *   Item specification.
1189  * @param[in] item_flags
1190  *   Bit-fields that holds the items detected until now.
1191  * @param[out] error
1192  *   Pointer to error structure.
1193  *
1194  * @return
1195  *   0 on success, a negative errno value otherwise and rte_errno is set.
1196  */
1197 int
1198 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1199                              uint64_t item_flags,
1200                              struct rte_flow_error *error)
1201 {
1202         const struct rte_flow_item_ipv6 *mask = item->mask;
1203         const struct rte_flow_item_ipv6 nic_mask = {
1204                 .hdr = {
1205                         .src_addr =
1206                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1207                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1208                         .dst_addr =
1209                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1210                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1211                         .vtc_flow = RTE_BE32(0xffffffff),
1212                         .proto = 0xff,
1213                         .hop_limits = 0xff,
1214                 },
1215         };
1216         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1217         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1218                                       MLX5_FLOW_LAYER_OUTER_L3;
1219         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1220                                       MLX5_FLOW_LAYER_OUTER_L4;
1221         int ret;
1222
1223         if (item_flags & l3m)
1224                 return rte_flow_error_set(error, ENOTSUP,
1225                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1226                                           "multiple L3 layers not supported");
1227         else if (item_flags & l4m)
1228                 return rte_flow_error_set(error, EINVAL,
1229                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1230                                           "L3 cannot follow an L4 layer.");
1231         /*
1232          * IPv6 is not recognised by the NIC inside a GRE tunnel.
1233          * Such support has to be disabled as the rule will be
1234          * accepted.  Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1235          * Mellanox OFED 4.4-1.0.0.0.
1236          */
1237         if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1238                 return rte_flow_error_set(error, ENOTSUP,
1239                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1240                                           "IPv6 inside a GRE tunnel is"
1241                                           " not recognised.");
1242         if (!mask)
1243                 mask = &rte_flow_item_ipv6_mask;
1244         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1245                                         (const uint8_t *)&nic_mask,
1246                                         sizeof(struct rte_flow_item_ipv6),
1247                                         error);
1248         if (ret < 0)
1249                 return ret;
1250         return 0;
1251 }
1252
1253 /**
1254  * Validate UDP item.
1255  *
1256  * @param[in] item
1257  *   Item specification.
1258  * @param[in] item_flags
1259  *   Bit-fields that holds the items detected until now.
1260  * @param[in] target_protocol
1261  *   The next protocol in the previous item.
1262  * @param[in] flow_mask
1263  *   mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1264  * @param[out] error
1265  *   Pointer to error structure.
1266  *
1267  * @return
1268  *   0 on success, a negative errno value otherwise and rte_errno is set.
1269  */
1270 int
1271 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1272                             uint64_t item_flags,
1273                             uint8_t target_protocol,
1274                             struct rte_flow_error *error)
1275 {
1276         const struct rte_flow_item_udp *mask = item->mask;
1277         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1278         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1279                                       MLX5_FLOW_LAYER_OUTER_L3;
1280         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1281                                       MLX5_FLOW_LAYER_OUTER_L4;
1282         int ret;
1283
1284         if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1285                 return rte_flow_error_set(error, EINVAL,
1286                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1287                                           "protocol filtering not compatible"
1288                                           " with UDP layer");
1289         if (!(item_flags & l3m))
1290                 return rte_flow_error_set(error, EINVAL,
1291                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1292                                           "L3 is mandatory to filter on L4");
1293         if (item_flags & l4m)
1294                 return rte_flow_error_set(error, EINVAL,
1295                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1296                                           "multiple L4 layers not supported");
1297         if (!mask)
1298                 mask = &rte_flow_item_udp_mask;
1299         ret = mlx5_flow_item_acceptable
1300                 (item, (const uint8_t *)mask,
1301                  (const uint8_t *)&rte_flow_item_udp_mask,
1302                  sizeof(struct rte_flow_item_udp), error);
1303         if (ret < 0)
1304                 return ret;
1305         return 0;
1306 }
1307
1308 /**
1309  * Validate TCP item.
1310  *
1311  * @param[in] item
1312  *   Item specification.
1313  * @param[in] item_flags
1314  *   Bit-fields that holds the items detected until now.
1315  * @param[in] target_protocol
1316  *   The next protocol in the previous item.
1317  * @param[out] error
1318  *   Pointer to error structure.
1319  *
1320  * @return
1321  *   0 on success, a negative errno value otherwise and rte_errno is set.
1322  */
1323 int
1324 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1325                             uint64_t item_flags,
1326                             uint8_t target_protocol,
1327                             const struct rte_flow_item_tcp *flow_mask,
1328                             struct rte_flow_error *error)
1329 {
1330         const struct rte_flow_item_tcp *mask = item->mask;
1331         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1332         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1333                                       MLX5_FLOW_LAYER_OUTER_L3;
1334         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1335                                       MLX5_FLOW_LAYER_OUTER_L4;
1336         int ret;
1337
1338         assert(flow_mask);
1339         if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1340                 return rte_flow_error_set(error, EINVAL,
1341                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1342                                           "protocol filtering not compatible"
1343                                           " with TCP layer");
1344         if (!(item_flags & l3m))
1345                 return rte_flow_error_set(error, EINVAL,
1346                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1347                                           "L3 is mandatory to filter on L4");
1348         if (item_flags & l4m)
1349                 return rte_flow_error_set(error, EINVAL,
1350                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1351                                           "multiple L4 layers not supported");
1352         if (!mask)
1353                 mask = &rte_flow_item_tcp_mask;
1354         ret = mlx5_flow_item_acceptable
1355                 (item, (const uint8_t *)mask,
1356                  (const uint8_t *)flow_mask,
1357                  sizeof(struct rte_flow_item_tcp), error);
1358         if (ret < 0)
1359                 return ret;
1360         return 0;
1361 }
1362
1363 /**
1364  * Validate VXLAN item.
1365  *
1366  * @param[in] item
1367  *   Item specification.
1368  * @param[in] item_flags
1369  *   Bit-fields that holds the items detected until now.
1370  * @param[in] target_protocol
1371  *   The next protocol in the previous item.
1372  * @param[out] error
1373  *   Pointer to error structure.
1374  *
1375  * @return
1376  *   0 on success, a negative errno value otherwise and rte_errno is set.
1377  */
1378 int
1379 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1380                               uint64_t item_flags,
1381                               struct rte_flow_error *error)
1382 {
1383         const struct rte_flow_item_vxlan *spec = item->spec;
1384         const struct rte_flow_item_vxlan *mask = item->mask;
1385         int ret;
1386         union vni {
1387                 uint32_t vlan_id;
1388                 uint8_t vni[4];
1389         } id = { .vlan_id = 0, };
1390         uint32_t vlan_id = 0;
1391
1392
1393         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1394                 return rte_flow_error_set(error, ENOTSUP,
1395                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1396                                           "multiple tunnel layers not"
1397                                           " supported");
1398         /*
1399          * Verify only UDPv4 is present as defined in
1400          * https://tools.ietf.org/html/rfc7348
1401          */
1402         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1403                 return rte_flow_error_set(error, EINVAL,
1404                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1405                                           "no outer UDP layer found");
1406         if (!mask)
1407                 mask = &rte_flow_item_vxlan_mask;
1408         ret = mlx5_flow_item_acceptable
1409                 (item, (const uint8_t *)mask,
1410                  (const uint8_t *)&rte_flow_item_vxlan_mask,
1411                  sizeof(struct rte_flow_item_vxlan),
1412                  error);
1413         if (ret < 0)
1414                 return ret;
1415         if (spec) {
1416                 memcpy(&id.vni[1], spec->vni, 3);
1417                 vlan_id = id.vlan_id;
1418                 memcpy(&id.vni[1], mask->vni, 3);
1419                 vlan_id &= id.vlan_id;
1420         }
1421         /*
1422          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1423          * only this layer is defined in the Verbs specification it is
1424          * interpreted as wildcard and all packets will match this
1425          * rule, if it follows a full stack layer (ex: eth / ipv4 /
1426          * udp), all packets matching the layers before will also
1427          * match this rule.  To avoid such situation, VNI 0 is
1428          * currently refused.
1429          */
1430         if (!vlan_id)
1431                 return rte_flow_error_set(error, ENOTSUP,
1432                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1433                                           "VXLAN vni cannot be 0");
1434         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1435                 return rte_flow_error_set(error, ENOTSUP,
1436                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1437                                           "VXLAN tunnel must be fully defined");
1438         return 0;
1439 }
1440
1441 /**
1442  * Validate VXLAN_GPE item.
1443  *
1444  * @param[in] item
1445  *   Item specification.
1446  * @param[in] item_flags
1447  *   Bit-fields that holds the items detected until now.
1448  * @param[in] priv
1449  *   Pointer to the private data structure.
1450  * @param[in] target_protocol
1451  *   The next protocol in the previous item.
1452  * @param[out] error
1453  *   Pointer to error structure.
1454  *
1455  * @return
1456  *   0 on success, a negative errno value otherwise and rte_errno is set.
1457  */
1458 int
1459 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1460                                   uint64_t item_flags,
1461                                   struct rte_eth_dev *dev,
1462                                   struct rte_flow_error *error)
1463 {
1464         struct priv *priv = dev->data->dev_private;
1465         const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1466         const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1467         int ret;
1468         union vni {
1469                 uint32_t vlan_id;
1470                 uint8_t vni[4];
1471         } id = { .vlan_id = 0, };
1472         uint32_t vlan_id = 0;
1473
1474         if (!priv->config.l3_vxlan_en)
1475                 return rte_flow_error_set(error, ENOTSUP,
1476                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1477                                           "L3 VXLAN is not enabled by device"
1478                                           " parameter and/or not configured in"
1479                                           " firmware");
1480         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1481                 return rte_flow_error_set(error, ENOTSUP,
1482                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1483                                           "multiple tunnel layers not"
1484                                           " supported");
1485         /*
1486          * Verify only UDPv4 is present as defined in
1487          * https://tools.ietf.org/html/rfc7348
1488          */
1489         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1490                 return rte_flow_error_set(error, EINVAL,
1491                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1492                                           "no outer UDP layer found");
1493         if (!mask)
1494                 mask = &rte_flow_item_vxlan_gpe_mask;
1495         ret = mlx5_flow_item_acceptable
1496                 (item, (const uint8_t *)mask,
1497                  (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1498                  sizeof(struct rte_flow_item_vxlan_gpe),
1499                  error);
1500         if (ret < 0)
1501                 return ret;
1502         if (spec) {
1503                 if (spec->protocol)
1504                         return rte_flow_error_set(error, ENOTSUP,
1505                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1506                                                   item,
1507                                                   "VxLAN-GPE protocol"
1508                                                   " not supported");
1509                 memcpy(&id.vni[1], spec->vni, 3);
1510                 vlan_id = id.vlan_id;
1511                 memcpy(&id.vni[1], mask->vni, 3);
1512                 vlan_id &= id.vlan_id;
1513         }
1514         /*
1515          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1516          * layer is defined in the Verbs specification it is interpreted as
1517          * wildcard and all packets will match this rule, if it follows a full
1518          * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1519          * before will also match this rule.  To avoid such situation, VNI 0
1520          * is currently refused.
1521          */
1522         if (!vlan_id)
1523                 return rte_flow_error_set(error, ENOTSUP,
1524                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1525                                           "VXLAN-GPE vni cannot be 0");
1526         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1527                 return rte_flow_error_set(error, ENOTSUP,
1528                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1529                                           "VXLAN-GPE tunnel must be fully"
1530                                           " defined");
1531         return 0;
1532 }
1533
1534 /**
1535  * Validate GRE item.
1536  *
1537  * @param[in] item
1538  *   Item specification.
1539  * @param[in] item_flags
1540  *   Bit flags to mark detected items.
1541  * @param[in] target_protocol
1542  *   The next protocol in the previous item.
1543  * @param[out] error
1544  *   Pointer to error structure.
1545  *
1546  * @return
1547  *   0 on success, a negative errno value otherwise and rte_errno is set.
1548  */
1549 int
1550 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1551                             uint64_t item_flags,
1552                             uint8_t target_protocol,
1553                             struct rte_flow_error *error)
1554 {
1555         const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1556         const struct rte_flow_item_gre *mask = item->mask;
1557         int ret;
1558
1559         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1560                 return rte_flow_error_set(error, EINVAL,
1561                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1562                                           "protocol filtering not compatible"
1563                                           " with this GRE layer");
1564         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1565                 return rte_flow_error_set(error, ENOTSUP,
1566                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1567                                           "multiple tunnel layers not"
1568                                           " supported");
1569         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1570                 return rte_flow_error_set(error, ENOTSUP,
1571                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1572                                           "L3 Layer is missing");
1573         if (!mask)
1574                 mask = &rte_flow_item_gre_mask;
1575         ret = mlx5_flow_item_acceptable
1576                 (item, (const uint8_t *)mask,
1577                  (const uint8_t *)&rte_flow_item_gre_mask,
1578                  sizeof(struct rte_flow_item_gre), error);
1579         if (ret < 0)
1580                 return ret;
1581 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1582         if (spec && (spec->protocol & mask->protocol))
1583                 return rte_flow_error_set(error, ENOTSUP,
1584                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1585                                           "without MPLS support the"
1586                                           " specification cannot be used for"
1587                                           " filtering");
1588 #endif
1589         return 0;
1590 }
1591
1592 /**
1593  * Validate MPLS item.
1594  *
1595  * @param[in] item
1596  *   Item specification.
1597  * @param[in] item_flags
1598  *   Bit-fields that holds the items detected until now.
1599  * @param[in] target_protocol
1600  *   The next protocol in the previous item.
1601  * @param[out] error
1602  *   Pointer to error structure.
1603  *
1604  * @return
1605  *   0 on success, a negative errno value otherwise and rte_errno is set.
1606  */
1607 int
1608 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1609                              uint64_t item_flags __rte_unused,
1610                              uint8_t target_protocol __rte_unused,
1611                              struct rte_flow_error *error)
1612 {
1613 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1614         const struct rte_flow_item_mpls *mask = item->mask;
1615         int ret;
1616
1617         if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1618                 return rte_flow_error_set(error, EINVAL,
1619                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1620                                           "protocol filtering not compatible"
1621                                           " with MPLS layer");
1622         /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
1623         if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1624             !(item_flags & MLX5_FLOW_LAYER_GRE))
1625                 return rte_flow_error_set(error, ENOTSUP,
1626                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1627                                           "multiple tunnel layers not"
1628                                           " supported");
1629         if (!mask)
1630                 mask = &rte_flow_item_mpls_mask;
1631         ret = mlx5_flow_item_acceptable
1632                 (item, (const uint8_t *)mask,
1633                  (const uint8_t *)&rte_flow_item_mpls_mask,
1634                  sizeof(struct rte_flow_item_mpls), error);
1635         if (ret < 0)
1636                 return ret;
1637         return 0;
1638 #endif
1639         return rte_flow_error_set(error, ENOTSUP,
1640                                   RTE_FLOW_ERROR_TYPE_ITEM, item,
1641                                   "MPLS is not supported by Verbs, please"
1642                                   " update.");
1643 }
1644
1645 static int
1646 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1647                    const struct rte_flow_attr *attr __rte_unused,
1648                    const struct rte_flow_item items[] __rte_unused,
1649                    const struct rte_flow_action actions[] __rte_unused,
1650                    struct rte_flow_error *error __rte_unused)
1651 {
1652         rte_errno = ENOTSUP;
1653         return -rte_errno;
1654 }
1655
1656 static struct mlx5_flow *
1657 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1658                   const struct rte_flow_item items[] __rte_unused,
1659                   const struct rte_flow_action actions[] __rte_unused,
1660                   uint64_t *item_flags __rte_unused,
1661                   uint64_t *action_flags __rte_unused,
1662                   struct rte_flow_error *error __rte_unused)
1663 {
1664         rte_errno = ENOTSUP;
1665         return NULL;
1666 }
1667
1668 static int
1669 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1670                     struct mlx5_flow *dev_flow __rte_unused,
1671                     const struct rte_flow_attr *attr __rte_unused,
1672                     const struct rte_flow_item items[] __rte_unused,
1673                     const struct rte_flow_action actions[] __rte_unused,
1674                     struct rte_flow_error *error __rte_unused)
1675 {
1676         rte_errno = ENOTSUP;
1677         return -rte_errno;
1678 }
1679
1680 static int
1681 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1682                 struct rte_flow *flow __rte_unused,
1683                 struct rte_flow_error *error __rte_unused)
1684 {
1685         rte_errno = ENOTSUP;
1686         return -rte_errno;
1687 }
1688
1689 static void
1690 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1691                  struct rte_flow *flow __rte_unused)
1692 {
1693 }
1694
1695 static void
1696 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1697                   struct rte_flow *flow __rte_unused)
1698 {
1699 }
1700
1701 static int
1702 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1703                 struct rte_flow *flow __rte_unused,
1704                 const struct rte_flow_action *actions __rte_unused,
1705                 void *data __rte_unused,
1706                 struct rte_flow_error *error __rte_unused)
1707 {
1708         rte_errno = ENOTSUP;
1709         return -rte_errno;
1710 }
1711
1712 /* Void driver to protect from null pointer reference. */
1713 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1714         .validate = flow_null_validate,
1715         .prepare = flow_null_prepare,
1716         .translate = flow_null_translate,
1717         .apply = flow_null_apply,
1718         .remove = flow_null_remove,
1719         .destroy = flow_null_destroy,
1720         .query = flow_null_query,
1721 };
1722
1723 /**
1724  * Select flow driver type according to flow attributes and device
1725  * configuration.
1726  *
1727  * @param[in] dev
1728  *   Pointer to the dev structure.
1729  * @param[in] attr
1730  *   Pointer to the flow attributes.
1731  *
1732  * @return
1733  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1734  */
1735 static enum mlx5_flow_drv_type
1736 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1737 {
1738         struct priv *priv = dev->data->dev_private;
1739         enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1740
1741         if (attr->transfer)
1742                 type = MLX5_FLOW_TYPE_TCF;
1743         else
1744                 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1745                                                  MLX5_FLOW_TYPE_VERBS;
1746         return type;
1747 }
1748
1749 #define flow_get_drv_ops(type) flow_drv_ops[type]
1750
1751 /**
1752  * Flow driver validation API. This abstracts calling driver specific functions.
1753  * The type of flow driver is determined according to flow attributes.
1754  *
1755  * @param[in] dev
1756  *   Pointer to the dev structure.
1757  * @param[in] attr
1758  *   Pointer to the flow attributes.
1759  * @param[in] items
1760  *   Pointer to the list of items.
1761  * @param[in] actions
1762  *   Pointer to the list of actions.
1763  * @param[out] error
1764  *   Pointer to the error structure.
1765  *
1766  * @return
1767  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1768  */
1769 static inline int
1770 flow_drv_validate(struct rte_eth_dev *dev,
1771                   const struct rte_flow_attr *attr,
1772                   const struct rte_flow_item items[],
1773                   const struct rte_flow_action actions[],
1774                   struct rte_flow_error *error)
1775 {
1776         const struct mlx5_flow_driver_ops *fops;
1777         enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1778
1779         fops = flow_get_drv_ops(type);
1780         return fops->validate(dev, attr, items, actions, error);
1781 }
1782
1783 /**
1784  * Flow driver preparation API. This abstracts calling driver specific
1785  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1786  * calculates the size of memory required for device flow, allocates the memory,
1787  * initializes the device flow and returns the pointer.
1788  *
1789  * @param[in] attr
1790  *   Pointer to the flow attributes.
1791  * @param[in] items
1792  *   Pointer to the list of items.
1793  * @param[in] actions
1794  *   Pointer to the list of actions.
1795  * @param[out] item_flags
1796  *   Pointer to bit mask of all items detected.
1797  * @param[out] action_flags
1798  *   Pointer to bit mask of all actions detected.
1799  * @param[out] error
1800  *   Pointer to the error structure.
1801  *
1802  * @return
1803  *   Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1804  */
1805 static inline struct mlx5_flow *
1806 flow_drv_prepare(struct rte_flow *flow,
1807                  const struct rte_flow_attr *attr,
1808                  const struct rte_flow_item items[],
1809                  const struct rte_flow_action actions[],
1810                  uint64_t *item_flags,
1811                  uint64_t *action_flags,
1812                  struct rte_flow_error *error)
1813 {
1814         const struct mlx5_flow_driver_ops *fops;
1815         enum mlx5_flow_drv_type type = flow->drv_type;
1816
1817         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1818         fops = flow_get_drv_ops(type);
1819         return fops->prepare(attr, items, actions, item_flags, action_flags,
1820                              error);
1821 }
1822
1823 /**
1824  * Flow driver translation API. This abstracts calling driver specific
1825  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1826  * translates a generic flow into a driver flow. flow_drv_prepare() must
1827  * precede.
1828  *
1829  *
1830  * @param[in] dev
1831  *   Pointer to the rte dev structure.
1832  * @param[in, out] dev_flow
1833  *   Pointer to the mlx5 flow.
1834  * @param[in] attr
1835  *   Pointer to the flow attributes.
1836  * @param[in] items
1837  *   Pointer to the list of items.
1838  * @param[in] actions
1839  *   Pointer to the list of actions.
1840  * @param[out] error
1841  *   Pointer to the error structure.
1842  *
1843  * @return
1844  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1845  */
1846 static inline int
1847 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1848                    const struct rte_flow_attr *attr,
1849                    const struct rte_flow_item items[],
1850                    const struct rte_flow_action actions[],
1851                    struct rte_flow_error *error)
1852 {
1853         const struct mlx5_flow_driver_ops *fops;
1854         enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1855
1856         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1857         fops = flow_get_drv_ops(type);
1858         return fops->translate(dev, dev_flow, attr, items, actions, error);
1859 }
1860
1861 /**
1862  * Flow driver apply API. This abstracts calling driver specific functions.
1863  * Parent flow (rte_flow) should have driver type (drv_type). It applies
1864  * translated driver flows on to device. flow_drv_translate() must precede.
1865  *
1866  * @param[in] dev
1867  *   Pointer to Ethernet device structure.
1868  * @param[in, out] flow
1869  *   Pointer to flow structure.
1870  * @param[out] error
1871  *   Pointer to error structure.
1872  *
1873  * @return
1874  *   0 on success, a negative errno value otherwise and rte_errno is set.
1875  */
1876 static inline int
1877 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1878                struct rte_flow_error *error)
1879 {
1880         const struct mlx5_flow_driver_ops *fops;
1881         enum mlx5_flow_drv_type type = flow->drv_type;
1882
1883         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1884         fops = flow_get_drv_ops(type);
1885         return fops->apply(dev, flow, error);
1886 }
1887
1888 /**
1889  * Flow driver remove API. This abstracts calling driver specific functions.
1890  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1891  * on device. All the resources of the flow should be freed by calling
1892  * flow_dv_destroy().
1893  *
1894  * @param[in] dev
1895  *   Pointer to Ethernet device.
1896  * @param[in, out] flow
1897  *   Pointer to flow structure.
1898  */
1899 static inline void
1900 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1901 {
1902         const struct mlx5_flow_driver_ops *fops;
1903         enum mlx5_flow_drv_type type = flow->drv_type;
1904
1905         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1906         fops = flow_get_drv_ops(type);
1907         fops->remove(dev, flow);
1908 }
1909
1910 /**
1911  * Flow driver destroy API. This abstracts calling driver specific functions.
1912  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1913  * on device and releases resources of the flow.
1914  *
1915  * @param[in] dev
1916  *   Pointer to Ethernet device.
1917  * @param[in, out] flow
1918  *   Pointer to flow structure.
1919  */
1920 static inline void
1921 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1922 {
1923         const struct mlx5_flow_driver_ops *fops;
1924         enum mlx5_flow_drv_type type = flow->drv_type;
1925
1926         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1927         fops = flow_get_drv_ops(type);
1928         fops->destroy(dev, flow);
1929 }
1930
1931 /**
1932  * Validate a flow supported by the NIC.
1933  *
1934  * @see rte_flow_validate()
1935  * @see rte_flow_ops
1936  */
1937 int
1938 mlx5_flow_validate(struct rte_eth_dev *dev,
1939                    const struct rte_flow_attr *attr,
1940                    const struct rte_flow_item items[],
1941                    const struct rte_flow_action actions[],
1942                    struct rte_flow_error *error)
1943 {
1944         int ret;
1945
1946         ret = flow_drv_validate(dev, attr, items, actions, error);
1947         if (ret < 0)
1948                 return ret;
1949         return 0;
1950 }
1951
1952 /**
1953  * Get RSS action from the action list.
1954  *
1955  * @param[in] actions
1956  *   Pointer to the list of actions.
1957  *
1958  * @return
1959  *   Pointer to the RSS action if exist, else return NULL.
1960  */
1961 static const struct rte_flow_action_rss*
1962 flow_get_rss_action(const struct rte_flow_action actions[])
1963 {
1964         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1965                 switch (actions->type) {
1966                 case RTE_FLOW_ACTION_TYPE_RSS:
1967                         return (const struct rte_flow_action_rss *)
1968                                actions->conf;
1969                 default:
1970                         break;
1971                 }
1972         }
1973         return NULL;
1974 }
1975
1976 static unsigned int
1977 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
1978 {
1979         const struct rte_flow_item *item;
1980         unsigned int has_vlan = 0;
1981
1982         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
1983                 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
1984                         has_vlan = 1;
1985                         break;
1986                 }
1987         }
1988         if (has_vlan)
1989                 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
1990                                        MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
1991         return rss_level < 2 ? MLX5_EXPANSION_ROOT :
1992                                MLX5_EXPANSION_ROOT_OUTER;
1993 }
1994
1995 /**
1996  * Create a flow and add it to @p list.
1997  *
1998  * @param dev
1999  *   Pointer to Ethernet device.
2000  * @param list
2001  *   Pointer to a TAILQ flow list.
2002  * @param[in] attr
2003  *   Flow rule attributes.
2004  * @param[in] items
2005  *   Pattern specification (list terminated by the END pattern item).
2006  * @param[in] actions
2007  *   Associated actions (list terminated by the END action).
2008  * @param[out] error
2009  *   Perform verbose error reporting if not NULL.
2010  *
2011  * @return
2012  *   A flow on success, NULL otherwise and rte_errno is set.
2013  */
2014 static struct rte_flow *
2015 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2016                  const struct rte_flow_attr *attr,
2017                  const struct rte_flow_item items[],
2018                  const struct rte_flow_action actions[],
2019                  struct rte_flow_error *error)
2020 {
2021         struct rte_flow *flow = NULL;
2022         struct mlx5_flow *dev_flow;
2023         uint64_t action_flags = 0;
2024         uint64_t item_flags = 0;
2025         const struct rte_flow_action_rss *rss;
2026         union {
2027                 struct rte_flow_expand_rss buf;
2028                 uint8_t buffer[2048];
2029         } expand_buffer;
2030         struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2031         int ret;
2032         uint32_t i;
2033         uint32_t flow_size;
2034
2035         ret = flow_drv_validate(dev, attr, items, actions, error);
2036         if (ret < 0)
2037                 return NULL;
2038         flow_size = sizeof(struct rte_flow);
2039         rss = flow_get_rss_action(actions);
2040         if (rss)
2041                 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2042                                             sizeof(void *));
2043         else
2044                 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2045         flow = rte_calloc(__func__, 1, flow_size, 0);
2046         flow->drv_type = flow_get_drv_type(dev, attr);
2047         assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2048                flow->drv_type < MLX5_FLOW_TYPE_MAX);
2049         flow->queue = (void *)(flow + 1);
2050         LIST_INIT(&flow->dev_flows);
2051         if (rss && rss->types) {
2052                 unsigned int graph_root;
2053
2054                 graph_root = find_graph_root(items, rss->level);
2055                 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2056                                           items, rss->types,
2057                                           mlx5_support_expansion,
2058                                           graph_root);
2059                 assert(ret > 0 &&
2060                        (unsigned int)ret < sizeof(expand_buffer.buffer));
2061         } else {
2062                 buf->entries = 1;
2063                 buf->entry[0].pattern = (void *)(uintptr_t)items;
2064         }
2065         for (i = 0; i < buf->entries; ++i) {
2066                 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2067                                             actions, &item_flags, &action_flags,
2068                                             error);
2069                 if (!dev_flow)
2070                         goto error;
2071                 dev_flow->flow = flow;
2072                 dev_flow->layers = item_flags;
2073                 /* Store actions once as expanded flows have same actions. */
2074                 if (i == 0)
2075                         flow->actions = action_flags;
2076                 assert(flow->actions == action_flags);
2077                 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2078                 ret = flow_drv_translate(dev, dev_flow, attr,
2079                                          buf->entry[i].pattern,
2080                                          actions, error);
2081                 if (ret < 0)
2082                         goto error;
2083         }
2084         if (dev->data->dev_started) {
2085                 ret = flow_drv_apply(dev, flow, error);
2086                 if (ret < 0)
2087                         goto error;
2088         }
2089         TAILQ_INSERT_TAIL(list, flow, next);
2090         flow_rxq_flags_set(dev, flow);
2091         return flow;
2092 error:
2093         ret = rte_errno; /* Save rte_errno before cleanup. */
2094         assert(flow);
2095         flow_drv_destroy(dev, flow);
2096         rte_free(flow);
2097         rte_errno = ret; /* Restore rte_errno. */
2098         return NULL;
2099 }
2100
2101 /**
2102  * Create a flow.
2103  *
2104  * @see rte_flow_create()
2105  * @see rte_flow_ops
2106  */
2107 struct rte_flow *
2108 mlx5_flow_create(struct rte_eth_dev *dev,
2109                  const struct rte_flow_attr *attr,
2110                  const struct rte_flow_item items[],
2111                  const struct rte_flow_action actions[],
2112                  struct rte_flow_error *error)
2113 {
2114         return flow_list_create(dev,
2115                                 &((struct priv *)dev->data->dev_private)->flows,
2116                                 attr, items, actions, error);
2117 }
2118
2119 /**
2120  * Destroy a flow in a list.
2121  *
2122  * @param dev
2123  *   Pointer to Ethernet device.
2124  * @param list
2125  *   Pointer to a TAILQ flow list.
2126  * @param[in] flow
2127  *   Flow to destroy.
2128  */
2129 static void
2130 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2131                   struct rte_flow *flow)
2132 {
2133         flow_drv_destroy(dev, flow);
2134         TAILQ_REMOVE(list, flow, next);
2135         /*
2136          * Update RX queue flags only if port is started, otherwise it is
2137          * already clean.
2138          */
2139         if (dev->data->dev_started)
2140                 flow_rxq_flags_trim(dev, flow);
2141         rte_free(flow->fdir);
2142         rte_free(flow);
2143 }
2144
2145 /**
2146  * Destroy all flows.
2147  *
2148  * @param dev
2149  *   Pointer to Ethernet device.
2150  * @param list
2151  *   Pointer to a TAILQ flow list.
2152  */
2153 void
2154 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2155 {
2156         while (!TAILQ_EMPTY(list)) {
2157                 struct rte_flow *flow;
2158
2159                 flow = TAILQ_FIRST(list);
2160                 flow_list_destroy(dev, list, flow);
2161         }
2162 }
2163
2164 /**
2165  * Remove all flows.
2166  *
2167  * @param dev
2168  *   Pointer to Ethernet device.
2169  * @param list
2170  *   Pointer to a TAILQ flow list.
2171  */
2172 void
2173 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2174 {
2175         struct rte_flow *flow;
2176
2177         TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2178                 flow_drv_remove(dev, flow);
2179         flow_rxq_flags_clear(dev);
2180 }
2181
2182 /**
2183  * Add all flows.
2184  *
2185  * @param dev
2186  *   Pointer to Ethernet device.
2187  * @param list
2188  *   Pointer to a TAILQ flow list.
2189  *
2190  * @return
2191  *   0 on success, a negative errno value otherwise and rte_errno is set.
2192  */
2193 int
2194 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2195 {
2196         struct rte_flow *flow;
2197         struct rte_flow_error error;
2198         int ret = 0;
2199
2200         TAILQ_FOREACH(flow, list, next) {
2201                 ret = flow_drv_apply(dev, flow, &error);
2202                 if (ret < 0)
2203                         goto error;
2204                 flow_rxq_flags_set(dev, flow);
2205         }
2206         return 0;
2207 error:
2208         ret = rte_errno; /* Save rte_errno before cleanup. */
2209         mlx5_flow_stop(dev, list);
2210         rte_errno = ret; /* Restore rte_errno. */
2211         return -rte_errno;
2212 }
2213
2214 /**
2215  * Verify the flow list is empty
2216  *
2217  * @param dev
2218  *  Pointer to Ethernet device.
2219  *
2220  * @return the number of flows not released.
2221  */
2222 int
2223 mlx5_flow_verify(struct rte_eth_dev *dev)
2224 {
2225         struct priv *priv = dev->data->dev_private;
2226         struct rte_flow *flow;
2227         int ret = 0;
2228
2229         TAILQ_FOREACH(flow, &priv->flows, next) {
2230                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
2231                         dev->data->port_id, (void *)flow);
2232                 ++ret;
2233         }
2234         return ret;
2235 }
2236
2237 /**
2238  * Enable a control flow configured from the control plane.
2239  *
2240  * @param dev
2241  *   Pointer to Ethernet device.
2242  * @param eth_spec
2243  *   An Ethernet flow spec to apply.
2244  * @param eth_mask
2245  *   An Ethernet flow mask to apply.
2246  * @param vlan_spec
2247  *   A VLAN flow spec to apply.
2248  * @param vlan_mask
2249  *   A VLAN flow mask to apply.
2250  *
2251  * @return
2252  *   0 on success, a negative errno value otherwise and rte_errno is set.
2253  */
2254 int
2255 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2256                     struct rte_flow_item_eth *eth_spec,
2257                     struct rte_flow_item_eth *eth_mask,
2258                     struct rte_flow_item_vlan *vlan_spec,
2259                     struct rte_flow_item_vlan *vlan_mask)
2260 {
2261         struct priv *priv = dev->data->dev_private;
2262         const struct rte_flow_attr attr = {
2263                 .ingress = 1,
2264                 .priority = MLX5_FLOW_PRIO_RSVD,
2265         };
2266         struct rte_flow_item items[] = {
2267                 {
2268                         .type = RTE_FLOW_ITEM_TYPE_ETH,
2269                         .spec = eth_spec,
2270                         .last = NULL,
2271                         .mask = eth_mask,
2272                 },
2273                 {
2274                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2275                                               RTE_FLOW_ITEM_TYPE_END,
2276                         .spec = vlan_spec,
2277                         .last = NULL,
2278                         .mask = vlan_mask,
2279                 },
2280                 {
2281                         .type = RTE_FLOW_ITEM_TYPE_END,
2282                 },
2283         };
2284         uint16_t queue[priv->reta_idx_n];
2285         struct rte_flow_action_rss action_rss = {
2286                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2287                 .level = 0,
2288                 .types = priv->rss_conf.rss_hf,
2289                 .key_len = priv->rss_conf.rss_key_len,
2290                 .queue_num = priv->reta_idx_n,
2291                 .key = priv->rss_conf.rss_key,
2292                 .queue = queue,
2293         };
2294         struct rte_flow_action actions[] = {
2295                 {
2296                         .type = RTE_FLOW_ACTION_TYPE_RSS,
2297                         .conf = &action_rss,
2298                 },
2299                 {
2300                         .type = RTE_FLOW_ACTION_TYPE_END,
2301                 },
2302         };
2303         struct rte_flow *flow;
2304         struct rte_flow_error error;
2305         unsigned int i;
2306
2307         if (!priv->reta_idx_n) {
2308                 rte_errno = EINVAL;
2309                 return -rte_errno;
2310         }
2311         for (i = 0; i != priv->reta_idx_n; ++i)
2312                 queue[i] = (*priv->reta_idx)[i];
2313         flow = flow_list_create(dev, &priv->ctrl_flows,
2314                                 &attr, items, actions, &error);
2315         if (!flow)
2316                 return -rte_errno;
2317         return 0;
2318 }
2319
2320 /**
2321  * Enable a flow control configured from the control plane.
2322  *
2323  * @param dev
2324  *   Pointer to Ethernet device.
2325  * @param eth_spec
2326  *   An Ethernet flow spec to apply.
2327  * @param eth_mask
2328  *   An Ethernet flow mask to apply.
2329  *
2330  * @return
2331  *   0 on success, a negative errno value otherwise and rte_errno is set.
2332  */
2333 int
2334 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2335                struct rte_flow_item_eth *eth_spec,
2336                struct rte_flow_item_eth *eth_mask)
2337 {
2338         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2339 }
2340
2341 /**
2342  * Destroy a flow.
2343  *
2344  * @see rte_flow_destroy()
2345  * @see rte_flow_ops
2346  */
2347 int
2348 mlx5_flow_destroy(struct rte_eth_dev *dev,
2349                   struct rte_flow *flow,
2350                   struct rte_flow_error *error __rte_unused)
2351 {
2352         struct priv *priv = dev->data->dev_private;
2353
2354         flow_list_destroy(dev, &priv->flows, flow);
2355         return 0;
2356 }
2357
2358 /**
2359  * Destroy all flows.
2360  *
2361  * @see rte_flow_flush()
2362  * @see rte_flow_ops
2363  */
2364 int
2365 mlx5_flow_flush(struct rte_eth_dev *dev,
2366                 struct rte_flow_error *error __rte_unused)
2367 {
2368         struct priv *priv = dev->data->dev_private;
2369
2370         mlx5_flow_list_flush(dev, &priv->flows);
2371         return 0;
2372 }
2373
2374 /**
2375  * Isolated mode.
2376  *
2377  * @see rte_flow_isolate()
2378  * @see rte_flow_ops
2379  */
2380 int
2381 mlx5_flow_isolate(struct rte_eth_dev *dev,
2382                   int enable,
2383                   struct rte_flow_error *error)
2384 {
2385         struct priv *priv = dev->data->dev_private;
2386
2387         if (dev->data->dev_started) {
2388                 rte_flow_error_set(error, EBUSY,
2389                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2390                                    NULL,
2391                                    "port must be stopped first");
2392                 return -rte_errno;
2393         }
2394         priv->isolated = !!enable;
2395         if (enable)
2396                 dev->dev_ops = &mlx5_dev_ops_isolate;
2397         else
2398                 dev->dev_ops = &mlx5_dev_ops;
2399         return 0;
2400 }
2401
2402 /**
2403  * Query a flow.
2404  *
2405  * @see rte_flow_query()
2406  * @see rte_flow_ops
2407  */
2408 static int
2409 flow_drv_query(struct rte_eth_dev *dev,
2410                struct rte_flow *flow,
2411                const struct rte_flow_action *actions,
2412                void *data,
2413                struct rte_flow_error *error)
2414 {
2415         const struct mlx5_flow_driver_ops *fops;
2416         enum mlx5_flow_drv_type ftype = flow->drv_type;
2417
2418         assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2419         fops = flow_get_drv_ops(ftype);
2420
2421         return fops->query(dev, flow, actions, data, error);
2422 }
2423
2424 /**
2425  * Query a flow.
2426  *
2427  * @see rte_flow_query()
2428  * @see rte_flow_ops
2429  */
2430 int
2431 mlx5_flow_query(struct rte_eth_dev *dev,
2432                 struct rte_flow *flow,
2433                 const struct rte_flow_action *actions,
2434                 void *data,
2435                 struct rte_flow_error *error)
2436 {
2437         int ret;
2438
2439         ret = flow_drv_query(dev, flow, actions, data, error);
2440         if (ret < 0)
2441                 return ret;
2442         return 0;
2443 }
2444
2445 /**
2446  * Convert a flow director filter to a generic flow.
2447  *
2448  * @param dev
2449  *   Pointer to Ethernet device.
2450  * @param fdir_filter
2451  *   Flow director filter to add.
2452  * @param attributes
2453  *   Generic flow parameters structure.
2454  *
2455  * @return
2456  *   0 on success, a negative errno value otherwise and rte_errno is set.
2457  */
2458 static int
2459 flow_fdir_filter_convert(struct rte_eth_dev *dev,
2460                          const struct rte_eth_fdir_filter *fdir_filter,
2461                          struct mlx5_fdir *attributes)
2462 {
2463         struct priv *priv = dev->data->dev_private;
2464         const struct rte_eth_fdir_input *input = &fdir_filter->input;
2465         const struct rte_eth_fdir_masks *mask =
2466                 &dev->data->dev_conf.fdir_conf.mask;
2467
2468         /* Validate queue number. */
2469         if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2470                 DRV_LOG(ERR, "port %u invalid queue number %d",
2471                         dev->data->port_id, fdir_filter->action.rx_queue);
2472                 rte_errno = EINVAL;
2473                 return -rte_errno;
2474         }
2475         attributes->attr.ingress = 1;
2476         attributes->items[0] = (struct rte_flow_item) {
2477                 .type = RTE_FLOW_ITEM_TYPE_ETH,
2478                 .spec = &attributes->l2,
2479                 .mask = &attributes->l2_mask,
2480         };
2481         switch (fdir_filter->action.behavior) {
2482         case RTE_ETH_FDIR_ACCEPT:
2483                 attributes->actions[0] = (struct rte_flow_action){
2484                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
2485                         .conf = &attributes->queue,
2486                 };
2487                 break;
2488         case RTE_ETH_FDIR_REJECT:
2489                 attributes->actions[0] = (struct rte_flow_action){
2490                         .type = RTE_FLOW_ACTION_TYPE_DROP,
2491                 };
2492                 break;
2493         default:
2494                 DRV_LOG(ERR, "port %u invalid behavior %d",
2495                         dev->data->port_id,
2496                         fdir_filter->action.behavior);
2497                 rte_errno = ENOTSUP;
2498                 return -rte_errno;
2499         }
2500         attributes->queue.index = fdir_filter->action.rx_queue;
2501         /* Handle L3. */
2502         switch (fdir_filter->input.flow_type) {
2503         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2504         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2505         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2506                 attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2507                         .src_addr = input->flow.ip4_flow.src_ip,
2508                         .dst_addr = input->flow.ip4_flow.dst_ip,
2509                         .time_to_live = input->flow.ip4_flow.ttl,
2510                         .type_of_service = input->flow.ip4_flow.tos,
2511                 };
2512                 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2513                         .src_addr = mask->ipv4_mask.src_ip,
2514                         .dst_addr = mask->ipv4_mask.dst_ip,
2515                         .time_to_live = mask->ipv4_mask.ttl,
2516                         .type_of_service = mask->ipv4_mask.tos,
2517                         .next_proto_id = mask->ipv4_mask.proto,
2518                 };
2519                 attributes->items[1] = (struct rte_flow_item){
2520                         .type = RTE_FLOW_ITEM_TYPE_IPV4,
2521                         .spec = &attributes->l3,
2522                         .mask = &attributes->l3_mask,
2523                 };
2524                 break;
2525         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2526         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2527         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2528                 attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2529                         .hop_limits = input->flow.ipv6_flow.hop_limits,
2530                         .proto = input->flow.ipv6_flow.proto,
2531                 };
2532
2533                 memcpy(attributes->l3.ipv6.hdr.src_addr,
2534                        input->flow.ipv6_flow.src_ip,
2535                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2536                 memcpy(attributes->l3.ipv6.hdr.dst_addr,
2537                        input->flow.ipv6_flow.dst_ip,
2538                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2539                 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2540                        mask->ipv6_mask.src_ip,
2541                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2542                 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2543                        mask->ipv6_mask.dst_ip,
2544                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2545                 attributes->items[1] = (struct rte_flow_item){
2546                         .type = RTE_FLOW_ITEM_TYPE_IPV6,
2547                         .spec = &attributes->l3,
2548                         .mask = &attributes->l3_mask,
2549                 };
2550                 break;
2551         default:
2552                 DRV_LOG(ERR, "port %u invalid flow type%d",
2553                         dev->data->port_id, fdir_filter->input.flow_type);
2554                 rte_errno = ENOTSUP;
2555                 return -rte_errno;
2556         }
2557         /* Handle L4. */
2558         switch (fdir_filter->input.flow_type) {
2559         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2560                 attributes->l4.udp.hdr = (struct udp_hdr){
2561                         .src_port = input->flow.udp4_flow.src_port,
2562                         .dst_port = input->flow.udp4_flow.dst_port,
2563                 };
2564                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2565                         .src_port = mask->src_port_mask,
2566                         .dst_port = mask->dst_port_mask,
2567                 };
2568                 attributes->items[2] = (struct rte_flow_item){
2569                         .type = RTE_FLOW_ITEM_TYPE_UDP,
2570                         .spec = &attributes->l4,
2571                         .mask = &attributes->l4_mask,
2572                 };
2573                 break;
2574         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2575                 attributes->l4.tcp.hdr = (struct tcp_hdr){
2576                         .src_port = input->flow.tcp4_flow.src_port,
2577                         .dst_port = input->flow.tcp4_flow.dst_port,
2578                 };
2579                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2580                         .src_port = mask->src_port_mask,
2581                         .dst_port = mask->dst_port_mask,
2582                 };
2583                 attributes->items[2] = (struct rte_flow_item){
2584                         .type = RTE_FLOW_ITEM_TYPE_TCP,
2585                         .spec = &attributes->l4,
2586                         .mask = &attributes->l4_mask,
2587                 };
2588                 break;
2589         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2590                 attributes->l4.udp.hdr = (struct udp_hdr){
2591                         .src_port = input->flow.udp6_flow.src_port,
2592                         .dst_port = input->flow.udp6_flow.dst_port,
2593                 };
2594                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2595                         .src_port = mask->src_port_mask,
2596                         .dst_port = mask->dst_port_mask,
2597                 };
2598                 attributes->items[2] = (struct rte_flow_item){
2599                         .type = RTE_FLOW_ITEM_TYPE_UDP,
2600                         .spec = &attributes->l4,
2601                         .mask = &attributes->l4_mask,
2602                 };
2603                 break;
2604         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2605                 attributes->l4.tcp.hdr = (struct tcp_hdr){
2606                         .src_port = input->flow.tcp6_flow.src_port,
2607                         .dst_port = input->flow.tcp6_flow.dst_port,
2608                 };
2609                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2610                         .src_port = mask->src_port_mask,
2611                         .dst_port = mask->dst_port_mask,
2612                 };
2613                 attributes->items[2] = (struct rte_flow_item){
2614                         .type = RTE_FLOW_ITEM_TYPE_TCP,
2615                         .spec = &attributes->l4,
2616                         .mask = &attributes->l4_mask,
2617                 };
2618                 break;
2619         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2620         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2621                 break;
2622         default:
2623                 DRV_LOG(ERR, "port %u invalid flow type%d",
2624                         dev->data->port_id, fdir_filter->input.flow_type);
2625                 rte_errno = ENOTSUP;
2626                 return -rte_errno;
2627         }
2628         return 0;
2629 }
2630
2631 #define FLOW_FDIR_CMP(f1, f2, fld) \
2632         memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
2633
2634 /**
2635  * Compare two FDIR flows. If items and actions are identical, the two flows are
2636  * regarded as same.
2637  *
2638  * @param dev
2639  *   Pointer to Ethernet device.
2640  * @param f1
2641  *   FDIR flow to compare.
2642  * @param f2
2643  *   FDIR flow to compare.
2644  *
2645  * @return
2646  *   Zero on match, 1 otherwise.
2647  */
2648 static int
2649 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
2650 {
2651         if (FLOW_FDIR_CMP(f1, f2, attr) ||
2652             FLOW_FDIR_CMP(f1, f2, l2) ||
2653             FLOW_FDIR_CMP(f1, f2, l2_mask) ||
2654             FLOW_FDIR_CMP(f1, f2, l3) ||
2655             FLOW_FDIR_CMP(f1, f2, l3_mask) ||
2656             FLOW_FDIR_CMP(f1, f2, l4) ||
2657             FLOW_FDIR_CMP(f1, f2, l4_mask) ||
2658             FLOW_FDIR_CMP(f1, f2, actions[0]))
2659                 return 1;
2660         if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
2661             FLOW_FDIR_CMP(f1, f2, queue))
2662                 return 1;
2663         return 0;
2664 }
2665
2666 /**
2667  * Search device flow list to find out a matched FDIR flow.
2668  *
2669  * @param dev
2670  *   Pointer to Ethernet device.
2671  * @param fdir_flow
2672  *   FDIR flow to lookup.
2673  *
2674  * @return
2675  *   Pointer of flow if found, NULL otherwise.
2676  */
2677 static struct rte_flow *
2678 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
2679 {
2680         struct priv *priv = dev->data->dev_private;
2681         struct rte_flow *flow = NULL;
2682
2683         assert(fdir_flow);
2684         TAILQ_FOREACH(flow, &priv->flows, next) {
2685                 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
2686                         DRV_LOG(DEBUG, "port %u found FDIR flow %p",
2687                                 dev->data->port_id, (void *)flow);
2688                         break;
2689                 }
2690         }
2691         return flow;
2692 }
2693
2694 /**
2695  * Add new flow director filter and store it in list.
2696  *
2697  * @param dev
2698  *   Pointer to Ethernet device.
2699  * @param fdir_filter
2700  *   Flow director filter to add.
2701  *
2702  * @return
2703  *   0 on success, a negative errno value otherwise and rte_errno is set.
2704  */
2705 static int
2706 flow_fdir_filter_add(struct rte_eth_dev *dev,
2707                      const struct rte_eth_fdir_filter *fdir_filter)
2708 {
2709         struct priv *priv = dev->data->dev_private;
2710         struct mlx5_fdir *fdir_flow;
2711         struct rte_flow *flow;
2712         int ret;
2713
2714         fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
2715         if (!fdir_flow) {
2716                 rte_errno = ENOMEM;
2717                 return -rte_errno;
2718         }
2719         ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
2720         if (ret)
2721                 goto error;
2722         flow = flow_fdir_filter_lookup(dev, fdir_flow);
2723         if (flow) {
2724                 rte_errno = EEXIST;
2725                 goto error;
2726         }
2727         flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
2728                                 fdir_flow->items, fdir_flow->actions, NULL);
2729         if (!flow)
2730                 goto error;
2731         assert(!flow->fdir);
2732         flow->fdir = fdir_flow;
2733         DRV_LOG(DEBUG, "port %u created FDIR flow %p",
2734                 dev->data->port_id, (void *)flow);
2735         return 0;
2736 error:
2737         rte_free(fdir_flow);
2738         return -rte_errno;
2739 }
2740
2741 /**
2742  * Delete specific filter.
2743  *
2744  * @param dev
2745  *   Pointer to Ethernet device.
2746  * @param fdir_filter
2747  *   Filter to be deleted.
2748  *
2749  * @return
2750  *   0 on success, a negative errno value otherwise and rte_errno is set.
2751  */
2752 static int
2753 flow_fdir_filter_delete(struct rte_eth_dev *dev,
2754                         const struct rte_eth_fdir_filter *fdir_filter)
2755 {
2756         struct priv *priv = dev->data->dev_private;
2757         struct rte_flow *flow;
2758         struct mlx5_fdir fdir_flow = {
2759                 .attr.group = 0,
2760         };
2761         int ret;
2762
2763         ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
2764         if (ret)
2765                 return -rte_errno;
2766         flow = flow_fdir_filter_lookup(dev, &fdir_flow);
2767         if (!flow) {
2768                 rte_errno = ENOENT;
2769                 return -rte_errno;
2770         }
2771         flow_list_destroy(dev, &priv->flows, flow);
2772         DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
2773                 dev->data->port_id, (void *)flow);
2774         return 0;
2775 }
2776
2777 /**
2778  * Update queue for specific filter.
2779  *
2780  * @param dev
2781  *   Pointer to Ethernet device.
2782  * @param fdir_filter
2783  *   Filter to be updated.
2784  *
2785  * @return
2786  *   0 on success, a negative errno value otherwise and rte_errno is set.
2787  */
2788 static int
2789 flow_fdir_filter_update(struct rte_eth_dev *dev,
2790                         const struct rte_eth_fdir_filter *fdir_filter)
2791 {
2792         int ret;
2793
2794         ret = flow_fdir_filter_delete(dev, fdir_filter);
2795         if (ret)
2796                 return ret;
2797         return flow_fdir_filter_add(dev, fdir_filter);
2798 }
2799
2800 /**
2801  * Flush all filters.
2802  *
2803  * @param dev
2804  *   Pointer to Ethernet device.
2805  */
2806 static void
2807 flow_fdir_filter_flush(struct rte_eth_dev *dev)
2808 {
2809         struct priv *priv = dev->data->dev_private;
2810
2811         mlx5_flow_list_flush(dev, &priv->flows);
2812 }
2813
2814 /**
2815  * Get flow director information.
2816  *
2817  * @param dev
2818  *   Pointer to Ethernet device.
2819  * @param[out] fdir_info
2820  *   Resulting flow director information.
2821  */
2822 static void
2823 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2824 {
2825         struct rte_eth_fdir_masks *mask =
2826                 &dev->data->dev_conf.fdir_conf.mask;
2827
2828         fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2829         fdir_info->guarant_spc = 0;
2830         rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2831         fdir_info->max_flexpayload = 0;
2832         fdir_info->flow_types_mask[0] = 0;
2833         fdir_info->flex_payload_unit = 0;
2834         fdir_info->max_flex_payload_segment_num = 0;
2835         fdir_info->flex_payload_limit = 0;
2836         memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2837 }
2838
2839 /**
2840  * Deal with flow director operations.
2841  *
2842  * @param dev
2843  *   Pointer to Ethernet device.
2844  * @param filter_op
2845  *   Operation to perform.
2846  * @param arg
2847  *   Pointer to operation-specific structure.
2848  *
2849  * @return
2850  *   0 on success, a negative errno value otherwise and rte_errno is set.
2851  */
2852 static int
2853 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2854                     void *arg)
2855 {
2856         enum rte_fdir_mode fdir_mode =
2857                 dev->data->dev_conf.fdir_conf.mode;
2858
2859         if (filter_op == RTE_ETH_FILTER_NOP)
2860                 return 0;
2861         if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2862             fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2863                 DRV_LOG(ERR, "port %u flow director mode %d not supported",
2864                         dev->data->port_id, fdir_mode);
2865                 rte_errno = EINVAL;
2866                 return -rte_errno;
2867         }
2868         switch (filter_op) {
2869         case RTE_ETH_FILTER_ADD:
2870                 return flow_fdir_filter_add(dev, arg);
2871         case RTE_ETH_FILTER_UPDATE:
2872                 return flow_fdir_filter_update(dev, arg);
2873         case RTE_ETH_FILTER_DELETE:
2874                 return flow_fdir_filter_delete(dev, arg);
2875         case RTE_ETH_FILTER_FLUSH:
2876                 flow_fdir_filter_flush(dev);
2877                 break;
2878         case RTE_ETH_FILTER_INFO:
2879                 flow_fdir_info_get(dev, arg);
2880                 break;
2881         default:
2882                 DRV_LOG(DEBUG, "port %u unknown operation %u",
2883                         dev->data->port_id, filter_op);
2884                 rte_errno = EINVAL;
2885                 return -rte_errno;
2886         }
2887         return 0;
2888 }
2889
2890 /**
2891  * Manage filter operations.
2892  *
2893  * @param dev
2894  *   Pointer to Ethernet device structure.
2895  * @param filter_type
2896  *   Filter type.
2897  * @param filter_op
2898  *   Operation to perform.
2899  * @param arg
2900  *   Pointer to operation-specific structure.
2901  *
2902  * @return
2903  *   0 on success, a negative errno value otherwise and rte_errno is set.
2904  */
2905 int
2906 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2907                      enum rte_filter_type filter_type,
2908                      enum rte_filter_op filter_op,
2909                      void *arg)
2910 {
2911         switch (filter_type) {
2912         case RTE_ETH_FILTER_GENERIC:
2913                 if (filter_op != RTE_ETH_FILTER_GET) {
2914                         rte_errno = EINVAL;
2915                         return -rte_errno;
2916                 }
2917                 *(const void **)arg = &mlx5_flow_ops;
2918                 return 0;
2919         case RTE_ETH_FILTER_FDIR:
2920                 return flow_fdir_ctrl_func(dev, filter_op, arg);
2921         default:
2922                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
2923                         dev->data->port_id, filter_type);
2924                 rte_errno = ENOTSUP;
2925                 return -rte_errno;
2926         }
2927         return 0;
2928 }