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_action actions[2];
 243         struct rte_flow_item items[4];
 244         struct rte_flow_item_eth l2;
 245         struct rte_flow_item_eth l2_mask;
 246         union {
 247                 struct rte_flow_item_ipv4 ipv4;
 248                 struct rte_flow_item_ipv6 ipv6;
 249         } l3;
 250         union {
 251                 struct rte_flow_item_ipv4 ipv4;
 252                 struct rte_flow_item_ipv6 ipv6;
 253         } l3_mask;
 254         union {
 255                 struct rte_flow_item_udp udp;
 256                 struct rte_flow_item_tcp tcp;
 257         } l4;
 258         union {
 259                 struct rte_flow_item_udp udp;
 260                 struct rte_flow_item_tcp tcp;
 261         } l4_mask;
 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         uint32_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, uint32_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 mlx5_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 flow.
 529  *
 530  * @param[in] dev
 531  *   Pointer to the Ethernet device structure.
 532  * @param[in] flow
 533  *   Pointer to flow structure.
 534  */
 535 static void
 536 mlx5_flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
 537 {
 538         struct priv *priv = dev->data->dev_private;
 539         const int mark = !!(flow->actions &
 540                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
 541         const int tunnel = !!(flow->layers & MLX5_FLOW_LAYER_TUNNEL);
 542         unsigned int i;
 543
 544         for (i = 0; i != flow->rss.queue_num; ++i) {
 545                 int idx = (*flow->queue)[i];
 546                 struct mlx5_rxq_ctrl *rxq_ctrl =
 547                         container_of((*priv->rxqs)[idx],
 548                                      struct mlx5_rxq_ctrl, rxq);
 549
 550                 if (mark) {
 551                         rxq_ctrl->rxq.mark = 1;
 552                         rxq_ctrl->flow_mark_n++;
 553                 }
 554                 if (tunnel) {
 555                         unsigned int j;
 556
 557                         /* Increase the counter matching the flow. */
 558                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 559                                 if ((tunnels_info[j].tunnel & flow->layers) ==
 560                                     tunnels_info[j].tunnel) {
 561                                         rxq_ctrl->flow_tunnels_n[j]++;
 562                                         break;
 563                                 }
 564                         }
 565                         mlx5_flow_rxq_tunnel_ptype_update(rxq_ctrl);
 566                 }
 567         }
 568 }
 569
 570 /**
 571  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
 572  * @p flow if no other flow uses it with the same kind of request.
 573  *
 574  * @param dev
 575  *   Pointer to Ethernet device.
 576  * @param[in] flow
 577  *   Pointer to the flow.
 578  */
 579 static void
 580 mlx5_flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
 581 {
 582         struct priv *priv = dev->data->dev_private;
 583         const int mark = !!(flow->actions &
 584                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
 585         const int tunnel = !!(flow->layers & MLX5_FLOW_LAYER_TUNNEL);
 586         unsigned int i;
 587
 588         assert(dev->data->dev_started);
 589         for (i = 0; i != flow->rss.queue_num; ++i) {
 590                 int idx = (*flow->queue)[i];
 591                 struct mlx5_rxq_ctrl *rxq_ctrl =
 592                         container_of((*priv->rxqs)[idx],
 593                                      struct mlx5_rxq_ctrl, rxq);
 594
 595                 if (mark) {
 596                         rxq_ctrl->flow_mark_n--;
 597                         rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
 598                 }
 599                 if (tunnel) {
 600                         unsigned int j;
 601
 602                         /* Decrease the counter matching the flow. */
 603                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 604                                 if ((tunnels_info[j].tunnel & flow->layers) ==
 605                                     tunnels_info[j].tunnel) {
 606                                         rxq_ctrl->flow_tunnels_n[j]--;
 607                                         break;
 608                                 }
 609                         }
 610                         mlx5_flow_rxq_tunnel_ptype_update(rxq_ctrl);
 611                 }
 612         }
 613 }
 614
 615 /**
 616  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
 617  *
 618  * @param dev
 619  *   Pointer to Ethernet device.
 620  */
 621 static void
 622 mlx5_flow_rxq_flags_clear(struct rte_eth_dev *dev)
 623 {
 624         struct priv *priv = dev->data->dev_private;
 625         unsigned int i;
 626
 627         for (i = 0; i != priv->rxqs_n; ++i) {
 628                 struct mlx5_rxq_ctrl *rxq_ctrl;
 629                 unsigned int j;
 630
 631                 if (!(*priv->rxqs)[i])
 632                         continue;
 633                 rxq_ctrl = container_of((*priv->rxqs)[i],
 634                                         struct mlx5_rxq_ctrl, rxq);
 635                 rxq_ctrl->flow_mark_n = 0;
 636                 rxq_ctrl->rxq.mark = 0;
 637                 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
 638                         rxq_ctrl->flow_tunnels_n[j] = 0;
 639                 rxq_ctrl->rxq.tunnel = 0;
 640         }
 641 }
 642
 643 /*
 644  * Validate the flag action.
 645  *
 646  * @param[in] action_flags
 647  *   Bit-fields that holds the actions detected until now.
 648  * @param[in] attr
 649  *   Attributes of flow that includes this action.
 650  * @param[out] error
 651  *   Pointer to error structure.
 652  *
 653  * @return
 654  *   0 on success, a negative errno value otherwise and rte_errno is set.
 655  */
 656 int
 657 mlx5_flow_validate_action_flag(uint64_t action_flags,
 658                                const struct rte_flow_attr *attr,
 659                                struct rte_flow_error *error)
 660 {
 661
 662         if (action_flags & MLX5_FLOW_ACTION_DROP)
 663                 return rte_flow_error_set(error, EINVAL,
 664                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 665                                           "can't drop and flag in same flow");
 666         if (action_flags & MLX5_FLOW_ACTION_MARK)
 667                 return rte_flow_error_set(error, EINVAL,
 668                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 669                                           "can't mark and flag in same flow");
 670         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 671                 return rte_flow_error_set(error, EINVAL,
 672                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 673                                           "can't have 2 flag"
 674                                           " actions in same flow");
 675         if (attr->egress)
 676                 return rte_flow_error_set(error, ENOTSUP,
 677                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 678                                           "flag action not supported for "
 679                                           "egress");
 680         return 0;
 681 }
 682
 683 /*
 684  * Validate the mark action.
 685  *
 686  * @param[in] action
 687  *   Pointer to the queue action.
 688  * @param[in] action_flags
 689  *   Bit-fields that holds the actions detected until now.
 690  * @param[in] attr
 691  *   Attributes of flow that includes this action.
 692  * @param[out] error
 693  *   Pointer to error structure.
 694  *
 695  * @return
 696  *   0 on success, a negative errno value otherwise and rte_errno is set.
 697  */
 698 int
 699 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
 700                                uint64_t action_flags,
 701                                const struct rte_flow_attr *attr,
 702                                struct rte_flow_error *error)
 703 {
 704         const struct rte_flow_action_mark *mark = action->conf;
 705
 706         if (!mark)
 707                 return rte_flow_error_set(error, EINVAL,
 708                                           RTE_FLOW_ERROR_TYPE_ACTION,
 709                                           action,
 710                                           "configuration cannot be null");
 711         if (mark->id >= MLX5_FLOW_MARK_MAX)
 712                 return rte_flow_error_set(error, EINVAL,
 713                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 714                                           &mark->id,
 715                                           "mark id must in 0 <= id < "
 716                                           RTE_STR(MLX5_FLOW_MARK_MAX));
 717         if (action_flags & MLX5_FLOW_ACTION_DROP)
 718                 return rte_flow_error_set(error, EINVAL,
 719                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 720                                           "can't drop and mark in same flow");
 721         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 722                 return rte_flow_error_set(error, EINVAL,
 723                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 724                                           "can't flag and mark in same flow");
 725         if (action_flags & MLX5_FLOW_ACTION_MARK)
 726                 return rte_flow_error_set(error, EINVAL,
 727                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 728                                           "can't have 2 mark actions in same"
 729                                           " flow");
 730         if (attr->egress)
 731                 return rte_flow_error_set(error, ENOTSUP,
 732                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 733                                           "mark action not supported for "
 734                                           "egress");
 735         return 0;
 736 }
 737
 738 /*
 739  * Validate the drop action.
 740  *
 741  * @param[in] action_flags
 742  *   Bit-fields that holds the actions detected until now.
 743  * @param[in] attr
 744  *   Attributes of flow that includes this action.
 745  * @param[out] error
 746  *   Pointer to error structure.
 747  *
 748  * @return
 749  *   0 on success, a negative errno value otherwise and rte_ernno is set.
 750  */
 751 int
 752 mlx5_flow_validate_action_drop(uint64_t action_flags,
 753                                const struct rte_flow_attr *attr,
 754                                struct rte_flow_error *error)
 755 {
 756         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 757                 return rte_flow_error_set(error, EINVAL,
 758                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 759                                           "can't drop and flag in same flow");
 760         if (action_flags & MLX5_FLOW_ACTION_MARK)
 761                 return rte_flow_error_set(error, EINVAL,
 762                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 763                                           "can't drop and mark in same flow");
 764         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
 765                 return rte_flow_error_set(error, EINVAL,
 766                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 767                                           "can't have 2 fate actions in"
 768                                           " same flow");
 769         if (attr->egress)
 770                 return rte_flow_error_set(error, ENOTSUP,
 771                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 772                                           "drop action not supported for "
 773                                           "egress");
 774         return 0;
 775 }
 776
 777 /*
 778  * Validate the queue action.
 779  *
 780  * @param[in] action
 781  *   Pointer to the queue action.
 782  * @param[in] action_flags
 783  *   Bit-fields that holds the actions detected until now.
 784  * @param[in] dev
 785  *   Pointer to the Ethernet device structure.
 786  * @param[in] attr
 787  *   Attributes of flow that includes this action.
 788  * @param[out] error
 789  *   Pointer to error structure.
 790  *
 791  * @return
 792  *   0 on success, a negative errno value otherwise and rte_ernno is set.
 793  */
 794 int
 795 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 796                                 uint64_t action_flags,
 797                                 struct rte_eth_dev *dev,
 798                                 const struct rte_flow_attr *attr,
 799                                 struct rte_flow_error *error)
 800 {
 801         struct priv *priv = dev->data->dev_private;
 802         const struct rte_flow_action_queue *queue = action->conf;
 803
 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 (queue->index >= priv->rxqs_n)
 810                 return rte_flow_error_set(error, EINVAL,
 811                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 812                                           &queue->index,
 813                                           "queue index out of range");
 814         if (!(*priv->rxqs)[queue->index])
 815                 return rte_flow_error_set(error, EINVAL,
 816                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 817                                           &queue->index,
 818                                           "queue is not configured");
 819         if (attr->egress)
 820                 return rte_flow_error_set(error, ENOTSUP,
 821                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 822                                           "queue action not supported for "
 823                                           "egress");
 824         return 0;
 825 }
 826
 827 /*
 828  * Validate the rss action.
 829  *
 830  * @param[in] action
 831  *   Pointer to the queue action.
 832  * @param[in] action_flags
 833  *   Bit-fields that holds the actions detected until now.
 834  * @param[in] dev
 835  *   Pointer to the Ethernet device structure.
 836  * @param[in] attr
 837  *   Attributes of flow that includes this action.
 838  * @param[out] error
 839  *   Pointer to error structure.
 840  *
 841  * @return
 842  *   0 on success, a negative errno value otherwise and rte_ernno is set.
 843  */
 844 int
 845 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
 846                               uint64_t action_flags,
 847                               struct rte_eth_dev *dev,
 848                               const struct rte_flow_attr *attr,
 849                               struct rte_flow_error *error)
 850 {
 851         struct priv *priv = dev->data->dev_private;
 852         const struct rte_flow_action_rss *rss = action->conf;
 853         unsigned int i;
 854
 855         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
 856                 return rte_flow_error_set(error, EINVAL,
 857                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 858                                           "can't have 2 fate actions"
 859                                           " in same flow");
 860         if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
 861             rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
 862                 return rte_flow_error_set(error, ENOTSUP,
 863                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 864                                           &rss->func,
 865                                           "RSS hash function not supported");
 866 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
 867         if (rss->level > 2)
 868 #else
 869         if (rss->level > 1)
 870 #endif
 871                 return rte_flow_error_set(error, ENOTSUP,
 872                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 873                                           &rss->level,
 874                                           "tunnel RSS is not supported");
 875         if (rss->key_len < MLX5_RSS_HASH_KEY_LEN)
 876                 return rte_flow_error_set(error, ENOTSUP,
 877                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 878                                           &rss->key_len,
 879                                           "RSS hash key too small");
 880         if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
 881                 return rte_flow_error_set(error, ENOTSUP,
 882                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 883                                           &rss->key_len,
 884                                           "RSS hash key too large");
 885         if (rss->queue_num > priv->config.ind_table_max_size)
 886                 return rte_flow_error_set(error, ENOTSUP,
 887                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 888                                           &rss->queue_num,
 889                                           "number of queues too large");
 890         if (rss->types & MLX5_RSS_HF_MASK)
 891                 return rte_flow_error_set(error, ENOTSUP,
 892                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 893                                           &rss->types,
 894                                           "some RSS protocols are not"
 895                                           " supported");
 896         for (i = 0; i != rss->queue_num; ++i) {
 897                 if (!(*priv->rxqs)[rss->queue[i]])
 898                         return rte_flow_error_set
 899                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 900                                  &rss->queue[i], "queue is not configured");
 901         }
 902         if (attr->egress)
 903                 return rte_flow_error_set(error, ENOTSUP,
 904                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 905                                           "rss action not supported for "
 906                                           "egress");
 907         return 0;
 908 }
 909
 910 /*
 911  * Validate the count action.
 912  *
 913  * @param[in] dev
 914  *   Pointer to the Ethernet device structure.
 915  * @param[in] attr
 916  *   Attributes of flow that includes this action.
 917  * @param[out] error
 918  *   Pointer to error structure.
 919  *
 920  * @return
 921  *   0 on success, a negative errno value otherwise and rte_ernno is set.
 922  */
 923 int
 924 mlx5_flow_validate_action_count(struct rte_eth_dev *dev,
 925                                 const struct rte_flow_attr *attr,
 926                                 struct rte_flow_error *error)
 927 {
 928         struct priv *priv = dev->data->dev_private;
 929
 930         if (!priv->config.flow_counter_en)
 931                 return rte_flow_error_set(error, ENOTSUP,
 932                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 933                                           "flow counters are not supported.");
 934         if (attr->egress)
 935                 return rte_flow_error_set(error, ENOTSUP,
 936                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 937                                           "count action not supported for "
 938                                           "egress");
 939         return 0;
 940 }
 941
 942 /**
 943  * Verify the @p attributes will be correctly understood by the NIC and store
 944  * them in the @p flow if everything is correct.
 945  *
 946  * @param[in] dev
 947  *   Pointer to the Ethernet device structure.
 948  * @param[in] attributes
 949  *   Pointer to flow attributes
 950  * @param[out] error
 951  *   Pointer to error structure.
 952  *
 953  * @return
 954  *   0 on success, a negative errno value otherwise and rte_errno is set.
 955  */
 956 int
 957 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
 958                               const struct rte_flow_attr *attributes,
 959                               struct rte_flow_error *error)
 960 {
 961         struct priv *priv = dev->data->dev_private;
 962         uint32_t priority_max = priv->config.flow_prio - 1;
 963
 964         if (attributes->group)
 965                 return rte_flow_error_set(error, ENOTSUP,
 966                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
 967                                           NULL, "groups is not supported");
 968         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
 969             attributes->priority >= priority_max)
 970                 return rte_flow_error_set(error, ENOTSUP,
 971                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
 972                                           NULL, "priority out of range");
 973         if (attributes->egress)
 974                 return rte_flow_error_set(error, ENOTSUP,
 975                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 976                                           "egress is not supported");
 977         if (attributes->transfer)
 978                 return rte_flow_error_set(error, ENOTSUP,
 979                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
 980                                           NULL, "transfer is not supported");
 981         if (!attributes->ingress)
 982                 return rte_flow_error_set(error, EINVAL,
 983                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
 984                                           NULL,
 985                                           "ingress attribute is mandatory");
 986         return 0;
 987 }
 988
 989 /**
 990  * Validate Ethernet item.
 991  *
 992  * @param[in] item
 993  *   Item specification.
 994  * @param[in] item_flags
 995  *   Bit-fields that holds the items detected until now.
 996  * @param[out] error
 997  *   Pointer to error structure.
 998  *
 999  * @return
1000  *   0 on success, a negative errno value otherwise and rte_errno is set.
1001  */
1002 int
1003 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1004                             uint64_t item_flags,
1005                             struct rte_flow_error *error)
1006 {
1007         const struct rte_flow_item_eth *mask = item->mask;
1008         const struct rte_flow_item_eth nic_mask = {
1009                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1010                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1011                 .type = RTE_BE16(0xffff),
1012         };
1013         int ret;
1014         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1015
1016         if (item_flags & MLX5_FLOW_LAYER_OUTER_L2)
1017                 return rte_flow_error_set(error, ENOTSUP,
1018                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1019                                           "3 levels of l2 are not supported");
1020         if ((item_flags & MLX5_FLOW_LAYER_INNER_L2) && !tunnel)
1021                 return rte_flow_error_set(error, ENOTSUP,
1022                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1023                                           "2 L2 without tunnel are not supported");
1024         if (!mask)
1025                 mask = &rte_flow_item_eth_mask;
1026         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1027                                         (const uint8_t *)&nic_mask,
1028                                         sizeof(struct rte_flow_item_eth),
1029                                         error);
1030         return ret;
1031 }
1032
1033 /**
1034  * Validate VLAN item.
1035  *
1036  * @param[in] item
1037  *   Item specification.
1038  * @param[in] item_flags
1039  *   Bit-fields that holds the items detected until now.
1040  * @param[out] error
1041  *   Pointer to error structure.
1042  *
1043  * @return
1044  *   0 on success, a negative errno value otherwise and rte_errno is set.
1045  */
1046 int
1047 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1048                              int64_t item_flags,
1049                              struct rte_flow_error *error)
1050 {
1051         const struct rte_flow_item_vlan *spec = item->spec;
1052         const struct rte_flow_item_vlan *mask = item->mask;
1053         const struct rte_flow_item_vlan nic_mask = {
1054                 .tci = RTE_BE16(0x0fff),
1055                 .inner_type = RTE_BE16(0xffff),
1056         };
1057         uint16_t vlan_tag = 0;
1058         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1059         int ret;
1060         const uint32_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1061                                         MLX5_FLOW_LAYER_INNER_L4) :
1062                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1063                                         MLX5_FLOW_LAYER_OUTER_L4);
1064         const uint32_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1065                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1066
1067         if (item_flags & vlanm)
1068                 return rte_flow_error_set(error, EINVAL,
1069                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1070                                           "VLAN layer already configured");
1071         else if ((item_flags & l34m) != 0)
1072                 return rte_flow_error_set(error, EINVAL,
1073                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1074                                           "L2 layer cannot follow L3/L4 layer");
1075         if (!mask)
1076                 mask = &rte_flow_item_vlan_mask;
1077         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1078                                         (const uint8_t *)&nic_mask,
1079                                         sizeof(struct rte_flow_item_vlan),
1080                                         error);
1081         if (ret)
1082                 return ret;
1083         if (spec) {
1084                 vlan_tag = spec->tci;
1085                 vlan_tag &= mask->tci;
1086         }
1087         /*
1088          * From verbs perspective an empty VLAN is equivalent
1089          * to a packet without VLAN layer.
1090          */
1091         if (!vlan_tag)
1092                 return rte_flow_error_set(error, EINVAL,
1093                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1094                                           item->spec,
1095                                           "VLAN cannot be empty");
1096         return 0;
1097 }
1098
1099 /**
1100  * Validate IPV4 item.
1101  *
1102  * @param[in] item
1103  *   Item specification.
1104  * @param[in] item_flags
1105  *   Bit-fields that holds the items detected until now.
1106  * @param[out] error
1107  *   Pointer to error structure.
1108  *
1109  * @return
1110  *   0 on success, a negative errno value otherwise and rte_errno is set.
1111  */
1112 int
1113 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1114                              int64_t item_flags,
1115                              struct rte_flow_error *error)
1116 {
1117         const struct rte_flow_item_ipv4 *mask = item->mask;
1118         const struct rte_flow_item_ipv4 nic_mask = {
1119                 .hdr = {
1120                         .src_addr = RTE_BE32(0xffffffff),
1121                         .dst_addr = RTE_BE32(0xffffffff),
1122                         .type_of_service = 0xff,
1123                         .next_proto_id = 0xff,
1124                 },
1125         };
1126         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1127         int ret;
1128
1129         if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1130                                    MLX5_FLOW_LAYER_OUTER_L3))
1131                 return rte_flow_error_set(error, ENOTSUP,
1132                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1133                                           "multiple L3 layers not supported");
1134         else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1135                                         MLX5_FLOW_LAYER_OUTER_L4))
1136                 return rte_flow_error_set(error, EINVAL,
1137                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1138                                           "L3 cannot follow an L4 layer.");
1139         if (!mask)
1140                 mask = &rte_flow_item_ipv4_mask;
1141         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1142                                         (const uint8_t *)&nic_mask,
1143                                         sizeof(struct rte_flow_item_ipv4),
1144                                         error);
1145         if (ret < 0)
1146                 return ret;
1147         return 0;
1148 }
1149
1150 /**
1151  * Validate IPV6 item.
1152  *
1153  * @param[in] item
1154  *   Item specification.
1155  * @param[in] item_flags
1156  *   Bit-fields that holds the items detected until now.
1157  * @param[out] error
1158  *   Pointer to error structure.
1159  *
1160  * @return
1161  *   0 on success, a negative errno value otherwise and rte_errno is set.
1162  */
1163 int
1164 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1165                              uint64_t item_flags,
1166                              struct rte_flow_error *error)
1167 {
1168         const struct rte_flow_item_ipv6 *mask = item->mask;
1169         const struct rte_flow_item_ipv6 nic_mask = {
1170                 .hdr = {
1171                         .src_addr =
1172                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1173                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1174                         .dst_addr =
1175                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1176                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1177                         .vtc_flow = RTE_BE32(0xffffffff),
1178                         .proto = 0xff,
1179                         .hop_limits = 0xff,
1180                 },
1181         };
1182         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1183         int ret;
1184
1185         if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1186                                    MLX5_FLOW_LAYER_OUTER_L3))
1187                 return rte_flow_error_set(error, ENOTSUP,
1188                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1189                                           "multiple L3 layers not supported");
1190         else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1191                                         MLX5_FLOW_LAYER_OUTER_L4))
1192                 return rte_flow_error_set(error, EINVAL,
1193                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1194                                           "L3 cannot follow an L4 layer.");
1195         /*
1196          * IPv6 is not recognised by the NIC inside a GRE tunnel.
1197          * Such support has to be disabled as the rule will be
1198          * accepted.  Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1199          * Mellanox OFED 4.4-1.0.0.0.
1200          */
1201         if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1202                 return rte_flow_error_set(error, ENOTSUP,
1203                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1204                                           "IPv6 inside a GRE tunnel is"
1205                                           " not recognised.");
1206         if (!mask)
1207                 mask = &rte_flow_item_ipv6_mask;
1208         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1209                                         (const uint8_t *)&nic_mask,
1210                                         sizeof(struct rte_flow_item_ipv6),
1211                                         error);
1212         if (ret < 0)
1213                 return ret;
1214         return 0;
1215 }
1216
1217 /**
1218  * Validate UDP item.
1219  *
1220  * @param[in] item
1221  *   Item specification.
1222  * @param[in] item_flags
1223  *   Bit-fields that holds the items detected until now.
1224  * @param[in] target_protocol
1225  *   The next protocol in the previous item.
1226  * @param[in] flow_mask
1227  *   mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1228  * @param[out] error
1229  *   Pointer to error structure.
1230  *
1231  * @return
1232  *   0 on success, a negative errno value otherwise and rte_errno is set.
1233  */
1234 int
1235 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1236                             uint64_t item_flags,
1237                             uint8_t target_protocol,
1238                             struct rte_flow_error *error)
1239 {
1240         const struct rte_flow_item_udp *mask = item->mask;
1241         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1242         int ret;
1243
1244         if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1245                 return rte_flow_error_set(error, EINVAL,
1246                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1247                                           "protocol filtering not compatible"
1248                                           " with UDP layer");
1249         if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1250                                      MLX5_FLOW_LAYER_OUTER_L3)))
1251                 return rte_flow_error_set(error, EINVAL,
1252                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1253                                           "L3 is mandatory to filter on L4");
1254         if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1255                                    MLX5_FLOW_LAYER_OUTER_L4))
1256                 return rte_flow_error_set(error, EINVAL,
1257                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1258                                           "L4 layer is already present");
1259         if (!mask)
1260                 mask = &rte_flow_item_udp_mask;
1261         ret = mlx5_flow_item_acceptable
1262                 (item, (const uint8_t *)mask,
1263                  (const uint8_t *)&rte_flow_item_udp_mask,
1264                  sizeof(struct rte_flow_item_udp), error);
1265         if (ret < 0)
1266                 return ret;
1267         return 0;
1268 }
1269
1270 /**
1271  * Validate TCP item.
1272  *
1273  * @param[in] item
1274  *   Item specification.
1275  * @param[in] item_flags
1276  *   Bit-fields that holds the items detected until now.
1277  * @param[in] target_protocol
1278  *   The next protocol in the previous item.
1279  * @param[out] error
1280  *   Pointer to error structure.
1281  *
1282  * @return
1283  *   0 on success, a negative errno value otherwise and rte_errno is set.
1284  */
1285 int
1286 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1287                             uint64_t item_flags,
1288                             uint8_t target_protocol,
1289                             const struct rte_flow_item_tcp *flow_mask,
1290                             struct rte_flow_error *error)
1291 {
1292         const struct rte_flow_item_tcp *mask = item->mask;
1293         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1294         int ret;
1295
1296         assert(flow_mask);
1297         if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1298                 return rte_flow_error_set(error, EINVAL,
1299                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1300                                           "protocol filtering not compatible"
1301                                           " with TCP layer");
1302         if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1303                                      MLX5_FLOW_LAYER_OUTER_L3)))
1304                 return rte_flow_error_set(error, EINVAL,
1305                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1306                                           "L3 is mandatory to filter on L4");
1307         if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1308                                    MLX5_FLOW_LAYER_OUTER_L4))
1309                 return rte_flow_error_set(error, EINVAL,
1310                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1311                                           "L4 layer is already present");
1312         if (!mask)
1313                 mask = &rte_flow_item_tcp_mask;
1314         ret = mlx5_flow_item_acceptable
1315                 (item, (const uint8_t *)mask,
1316                  (const uint8_t *)flow_mask,
1317                  sizeof(struct rte_flow_item_tcp), error);
1318         if (ret < 0)
1319                 return ret;
1320         return 0;
1321 }
1322
1323 /**
1324  * Validate VXLAN item.
1325  *
1326  * @param[in] item
1327  *   Item specification.
1328  * @param[in] item_flags
1329  *   Bit-fields that holds the items detected until now.
1330  * @param[in] target_protocol
1331  *   The next protocol in the previous item.
1332  * @param[out] error
1333  *   Pointer to error structure.
1334  *
1335  * @return
1336  *   0 on success, a negative errno value otherwise and rte_errno is set.
1337  */
1338 int
1339 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1340                               uint64_t item_flags,
1341                               struct rte_flow_error *error)
1342 {
1343         const struct rte_flow_item_vxlan *spec = item->spec;
1344         const struct rte_flow_item_vxlan *mask = item->mask;
1345         int ret;
1346         union vni {
1347                 uint32_t vlan_id;
1348                 uint8_t vni[4];
1349         } id = { .vlan_id = 0, };
1350         uint32_t vlan_id = 0;
1351
1352
1353         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1354                 return rte_flow_error_set(error, ENOTSUP,
1355                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1356                                           "a tunnel is already present");
1357         /*
1358          * Verify only UDPv4 is present as defined in
1359          * https://tools.ietf.org/html/rfc7348
1360          */
1361         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1362                 return rte_flow_error_set(error, EINVAL,
1363                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1364                                           "no outer UDP layer found");
1365         if (!mask)
1366                 mask = &rte_flow_item_vxlan_mask;
1367         ret = mlx5_flow_item_acceptable
1368                 (item, (const uint8_t *)mask,
1369                  (const uint8_t *)&rte_flow_item_vxlan_mask,
1370                  sizeof(struct rte_flow_item_vxlan),
1371                  error);
1372         if (ret < 0)
1373                 return ret;
1374         if (spec) {
1375                 memcpy(&id.vni[1], spec->vni, 3);
1376                 vlan_id = id.vlan_id;
1377                 memcpy(&id.vni[1], mask->vni, 3);
1378                 vlan_id &= id.vlan_id;
1379         }
1380         /*
1381          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1382          * only this layer is defined in the Verbs specification it is
1383          * interpreted as wildcard and all packets will match this
1384          * rule, if it follows a full stack layer (ex: eth / ipv4 /
1385          * udp), all packets matching the layers before will also
1386          * match this rule.  To avoid such situation, VNI 0 is
1387          * currently refused.
1388          */
1389         if (!vlan_id)
1390                 return rte_flow_error_set(error, ENOTSUP,
1391                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1392                                           "VXLAN vni cannot be 0");
1393         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1394                 return rte_flow_error_set(error, ENOTSUP,
1395                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1396                                           "VXLAN tunnel must be fully defined");
1397         return 0;
1398 }
1399
1400 /**
1401  * Validate VXLAN_GPE item.
1402  *
1403  * @param[in] item
1404  *   Item specification.
1405  * @param[in] item_flags
1406  *   Bit-fields that holds the items detected until now.
1407  * @param[in] priv
1408  *   Pointer to the private data structure.
1409  * @param[in] target_protocol
1410  *   The next protocol in the previous item.
1411  * @param[out] error
1412  *   Pointer to error structure.
1413  *
1414  * @return
1415  *   0 on success, a negative errno value otherwise and rte_errno is set.
1416  */
1417 int
1418 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1419                                   uint64_t item_flags,
1420                                   struct rte_eth_dev *dev,
1421                                   struct rte_flow_error *error)
1422 {
1423         struct priv *priv = dev->data->dev_private;
1424         const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1425         const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1426         int ret;
1427         union vni {
1428                 uint32_t vlan_id;
1429                 uint8_t vni[4];
1430         } id = { .vlan_id = 0, };
1431         uint32_t vlan_id = 0;
1432
1433         if (!priv->config.l3_vxlan_en)
1434                 return rte_flow_error_set(error, ENOTSUP,
1435                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1436                                           "L3 VXLAN is not enabled by device"
1437                                           " parameter and/or not configured in"
1438                                           " firmware");
1439         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1440                 return rte_flow_error_set(error, ENOTSUP,
1441                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1442                                           "a tunnel is already present");
1443         /*
1444          * Verify only UDPv4 is present as defined in
1445          * https://tools.ietf.org/html/rfc7348
1446          */
1447         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1448                 return rte_flow_error_set(error, EINVAL,
1449                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1450                                           "no outer UDP layer found");
1451         if (!mask)
1452                 mask = &rte_flow_item_vxlan_gpe_mask;
1453         ret = mlx5_flow_item_acceptable
1454                 (item, (const uint8_t *)mask,
1455                  (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1456                  sizeof(struct rte_flow_item_vxlan_gpe),
1457                  error);
1458         if (ret < 0)
1459                 return ret;
1460         if (spec) {
1461                 if (spec->protocol)
1462                         return rte_flow_error_set(error, ENOTSUP,
1463                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1464                                                   item,
1465                                                   "VxLAN-GPE protocol"
1466                                                   " not supported");
1467                 memcpy(&id.vni[1], spec->vni, 3);
1468                 vlan_id = id.vlan_id;
1469                 memcpy(&id.vni[1], mask->vni, 3);
1470                 vlan_id &= id.vlan_id;
1471         }
1472         /*
1473          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1474          * layer is defined in the Verbs specification it is interpreted as
1475          * wildcard and all packets will match this rule, if it follows a full
1476          * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1477          * before will also match this rule.  To avoid such situation, VNI 0
1478          * is currently refused.
1479          */
1480         if (!vlan_id)
1481                 return rte_flow_error_set(error, ENOTSUP,
1482                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1483                                           "VXLAN-GPE vni cannot be 0");
1484         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1485                 return rte_flow_error_set(error, ENOTSUP,
1486                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1487                                           "VXLAN-GPE tunnel must be fully"
1488                                           " defined");
1489         return 0;
1490 }
1491
1492 /**
1493  * Validate GRE item.
1494  *
1495  * @param[in] item
1496  *   Item specification.
1497  * @param[in] item_flags
1498  *   Bit flags to mark detected items.
1499  * @param[in] target_protocol
1500  *   The next protocol in the previous item.
1501  * @param[out] error
1502  *   Pointer to error structure.
1503  *
1504  * @return
1505  *   0 on success, a negative errno value otherwise and rte_errno is set.
1506  */
1507 int
1508 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1509                             uint64_t item_flags,
1510                             uint8_t target_protocol,
1511                             struct rte_flow_error *error)
1512 {
1513         const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1514         const struct rte_flow_item_gre *mask = item->mask;
1515         int ret;
1516
1517         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1518                 return rte_flow_error_set(error, EINVAL,
1519                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1520                                           "protocol filtering not compatible"
1521                                           " with this GRE layer");
1522         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1523                 return rte_flow_error_set(error, ENOTSUP,
1524                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1525                                           "a tunnel is already present");
1526         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1527                 return rte_flow_error_set(error, ENOTSUP,
1528                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1529                                           "L3 Layer is missing");
1530         if (!mask)
1531                 mask = &rte_flow_item_gre_mask;
1532         ret = mlx5_flow_item_acceptable
1533                 (item, (const uint8_t *)mask,
1534                  (const uint8_t *)&rte_flow_item_gre_mask,
1535                  sizeof(struct rte_flow_item_gre), error);
1536         if (ret < 0)
1537                 return ret;
1538 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1539         if (spec && (spec->protocol & mask->protocol))
1540                 return rte_flow_error_set(error, ENOTSUP,
1541                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1542                                           "without MPLS support the"
1543                                           " specification cannot be used for"
1544                                           " filtering");
1545 #endif
1546         return 0;
1547 }
1548
1549 /**
1550  * Validate MPLS item.
1551  *
1552  * @param[in] item
1553  *   Item specification.
1554  * @param[in] item_flags
1555  *   Bit-fields that holds the items detected until now.
1556  * @param[in] target_protocol
1557  *   The next protocol in the previous item.
1558  * @param[out] error
1559  *   Pointer to error structure.
1560  *
1561  * @return
1562  *   0 on success, a negative errno value otherwise and rte_errno is set.
1563  */
1564 int
1565 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1566                              uint64_t item_flags __rte_unused,
1567                              uint8_t target_protocol __rte_unused,
1568                              struct rte_flow_error *error)
1569 {
1570 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1571         const struct rte_flow_item_mpls *mask = item->mask;
1572         int ret;
1573
1574         if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1575                 return rte_flow_error_set(error, EINVAL,
1576                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1577                                           "protocol filtering not compatible"
1578                                           " with MPLS layer");
1579         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1580                 return rte_flow_error_set(error, ENOTSUP,
1581                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1582                                           "a tunnel is already"
1583                                           " present");
1584         if (!mask)
1585                 mask = &rte_flow_item_mpls_mask;
1586         ret = mlx5_flow_item_acceptable
1587                 (item, (const uint8_t *)mask,
1588                  (const uint8_t *)&rte_flow_item_mpls_mask,
1589                  sizeof(struct rte_flow_item_mpls), error);
1590         if (ret < 0)
1591                 return ret;
1592         return 0;
1593 #endif
1594         return rte_flow_error_set(error, ENOTSUP,
1595                                   RTE_FLOW_ERROR_TYPE_ITEM, item,
1596                                   "MPLS is not supported by Verbs, please"
1597                                   " update.");
1598 }
1599
1600 static int
1601 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1602                    const struct rte_flow_attr *attr __rte_unused,
1603                    const struct rte_flow_item items[] __rte_unused,
1604                    const struct rte_flow_action actions[] __rte_unused,
1605                    struct rte_flow_error *error __rte_unused)
1606 {
1607         rte_errno = ENOTSUP;
1608         return -rte_errno;
1609 }
1610
1611 static struct mlx5_flow *
1612 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1613                   const struct rte_flow_item items[] __rte_unused,
1614                   const struct rte_flow_action actions[] __rte_unused,
1615                   uint64_t *item_flags __rte_unused,
1616                   uint64_t *action_flags __rte_unused,
1617                   struct rte_flow_error *error __rte_unused)
1618 {
1619         rte_errno = ENOTSUP;
1620         return NULL;
1621 }
1622
1623 static int
1624 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1625                     struct mlx5_flow *dev_flow __rte_unused,
1626                     const struct rte_flow_attr *attr __rte_unused,
1627                     const struct rte_flow_item items[] __rte_unused,
1628                     const struct rte_flow_action actions[] __rte_unused,
1629                     struct rte_flow_error *error __rte_unused)
1630 {
1631         rte_errno = ENOTSUP;
1632         return -rte_errno;
1633 }
1634
1635 static int
1636 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1637                 struct rte_flow *flow __rte_unused,
1638                 struct rte_flow_error *error __rte_unused)
1639 {
1640         rte_errno = ENOTSUP;
1641         return -rte_errno;
1642 }
1643
1644 static void
1645 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1646                  struct rte_flow *flow __rte_unused)
1647 {
1648 }
1649
1650 static void
1651 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1652                   struct rte_flow *flow __rte_unused)
1653 {
1654 }
1655
1656 static int
1657 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1658                 struct rte_flow *flow __rte_unused,
1659                 const struct rte_flow_action *actions __rte_unused,
1660                 void *data __rte_unused,
1661                 struct rte_flow_error *error __rte_unused)
1662 {
1663         rte_errno = ENOTSUP;
1664         return -rte_errno;
1665 }
1666
1667 /* Void driver to protect from null pointer reference. */
1668 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1669         .validate = flow_null_validate,
1670         .prepare = flow_null_prepare,
1671         .translate = flow_null_translate,
1672         .apply = flow_null_apply,
1673         .remove = flow_null_remove,
1674         .destroy = flow_null_destroy,
1675         .query = flow_null_query,
1676 };
1677
1678 /**
1679  * Select flow driver type according to flow attributes and device
1680  * configuration.
1681  *
1682  * @param[in] dev
1683  *   Pointer to the dev structure.
1684  * @param[in] attr
1685  *   Pointer to the flow attributes.
1686  *
1687  * @return
1688  *   flow driver type if supported, MLX5_FLOW_TYPE_MAX otherwise.
1689  */
1690 static enum mlx5_flow_drv_type
1691 flow_get_drv_type(struct rte_eth_dev *dev __rte_unused,
1692                   const struct rte_flow_attr *attr)
1693 {
1694         struct priv *priv __rte_unused = dev->data->dev_private;
1695         enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1696
1697         if (attr->transfer) {
1698                 type = MLX5_FLOW_TYPE_TCF;
1699         } else {
1700 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
1701                 type = priv->config.dv_flow_en ?  MLX5_FLOW_TYPE_DV :
1702                                                   MLX5_FLOW_TYPE_VERBS;
1703 #else
1704                 type = MLX5_FLOW_TYPE_VERBS;
1705 #endif
1706         }
1707         return type;
1708 }
1709
1710 #define flow_get_drv_ops(type) flow_drv_ops[type]
1711
1712 /**
1713  * Flow driver validation API. This abstracts calling driver specific functions.
1714  * The type of flow driver is determined according to flow attributes.
1715  *
1716  * @param[in] dev
1717  *   Pointer to the dev structure.
1718  * @param[in] attr
1719  *   Pointer to the flow attributes.
1720  * @param[in] items
1721  *   Pointer to the list of items.
1722  * @param[in] actions
1723  *   Pointer to the list of actions.
1724  * @param[out] error
1725  *   Pointer to the error structure.
1726  *
1727  * @return
1728  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1729  */
1730 static inline int
1731 flow_drv_validate(struct rte_eth_dev *dev,
1732                   const struct rte_flow_attr *attr,
1733                   const struct rte_flow_item items[],
1734                   const struct rte_flow_action actions[],
1735                   struct rte_flow_error *error)
1736 {
1737         const struct mlx5_flow_driver_ops *fops;
1738         enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1739
1740         fops = flow_get_drv_ops(type);
1741         return fops->validate(dev, attr, items, actions, error);
1742 }
1743
1744 /**
1745  * Flow driver preparation API. This abstracts calling driver specific
1746  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1747  * calculates the size of memory required for device flow, allocates the memory,
1748  * initializes the device flow and returns the pointer.
1749  *
1750  * @param[in] attr
1751  *   Pointer to the flow attributes.
1752  * @param[in] items
1753  *   Pointer to the list of items.
1754  * @param[in] actions
1755  *   Pointer to the list of actions.
1756  * @param[out] item_flags
1757  *   Pointer to bit mask of all items detected.
1758  * @param[out] action_flags
1759  *   Pointer to bit mask of all actions detected.
1760  * @param[out] error
1761  *   Pointer to the error structure.
1762  *
1763  * @return
1764  *   Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1765  */
1766 static inline struct mlx5_flow *
1767 flow_drv_prepare(struct rte_flow *flow,
1768                  const struct rte_flow_attr *attr,
1769                  const struct rte_flow_item items[],
1770                  const struct rte_flow_action actions[],
1771                  uint64_t *item_flags,
1772                  uint64_t *action_flags,
1773                  struct rte_flow_error *error)
1774 {
1775         const struct mlx5_flow_driver_ops *fops;
1776         enum mlx5_flow_drv_type type = flow->drv_type;
1777
1778         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1779         fops = flow_get_drv_ops(type);
1780         return fops->prepare(attr, items, actions, item_flags, action_flags,
1781                              error);
1782 }
1783
1784 /**
1785  * Flow driver translation API. This abstracts calling driver specific
1786  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1787  * translates a generic flow into a driver flow. flow_drv_prepare() must
1788  * precede.
1789  *
1790  *
1791  * @param[in] dev
1792  *   Pointer to the rte dev structure.
1793  * @param[in, out] dev_flow
1794  *   Pointer to the mlx5 flow.
1795  * @param[in] attr
1796  *   Pointer to the flow attributes.
1797  * @param[in] items
1798  *   Pointer to the list of items.
1799  * @param[in] actions
1800  *   Pointer to the list of actions.
1801  * @param[out] error
1802  *   Pointer to the error structure.
1803  *
1804  * @return
1805  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1806  */
1807 static inline int
1808 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1809                    const struct rte_flow_attr *attr,
1810                    const struct rte_flow_item items[],
1811                    const struct rte_flow_action actions[],
1812                    struct rte_flow_error *error)
1813 {
1814         const struct mlx5_flow_driver_ops *fops;
1815         enum mlx5_flow_drv_type type = dev_flow->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->translate(dev, dev_flow, attr, items, actions, error);
1820 }
1821
1822 /**
1823  * Flow driver apply API. This abstracts calling driver specific functions.
1824  * Parent flow (rte_flow) should have driver type (drv_type). It applies
1825  * translated driver flows on to device. flow_drv_translate() must precede.
1826  *
1827  * @param[in] dev
1828  *   Pointer to Ethernet device structure.
1829  * @param[in, out] flow
1830  *   Pointer to flow structure.
1831  * @param[out] error
1832  *   Pointer to error structure.
1833  *
1834  * @return
1835  *   0 on success, a negative errno value otherwise and rte_errno is set.
1836  */
1837 static inline int
1838 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1839                struct rte_flow_error *error)
1840 {
1841         const struct mlx5_flow_driver_ops *fops;
1842         enum mlx5_flow_drv_type type = flow->drv_type;
1843
1844         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1845         fops = flow_get_drv_ops(type);
1846         return fops->apply(dev, flow, error);
1847 }
1848
1849 /**
1850  * Flow driver remove API. This abstracts calling driver specific functions.
1851  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1852  * on device. All the resources of the flow should be freed by calling
1853  * flow_dv_destroy().
1854  *
1855  * @param[in] dev
1856  *   Pointer to Ethernet device.
1857  * @param[in, out] flow
1858  *   Pointer to flow structure.
1859  */
1860 static inline void
1861 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1862 {
1863         const struct mlx5_flow_driver_ops *fops;
1864         enum mlx5_flow_drv_type type = flow->drv_type;
1865
1866         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1867         fops = flow_get_drv_ops(type);
1868         fops->remove(dev, flow);
1869 }
1870
1871 /**
1872  * Flow driver destroy API. This abstracts calling driver specific functions.
1873  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1874  * on device and releases resources of the flow.
1875  *
1876  * @param[in] dev
1877  *   Pointer to Ethernet device.
1878  * @param[in, out] flow
1879  *   Pointer to flow structure.
1880  */
1881 static inline void
1882 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1883 {
1884         const struct mlx5_flow_driver_ops *fops;
1885         enum mlx5_flow_drv_type type = flow->drv_type;
1886
1887         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1888         fops = flow_get_drv_ops(type);
1889         fops->destroy(dev, flow);
1890 }
1891
1892 /**
1893  * Validate a flow supported by the NIC.
1894  *
1895  * @see rte_flow_validate()
1896  * @see rte_flow_ops
1897  */
1898 int
1899 mlx5_flow_validate(struct rte_eth_dev *dev,
1900                    const struct rte_flow_attr *attr,
1901                    const struct rte_flow_item items[],
1902                    const struct rte_flow_action actions[],
1903                    struct rte_flow_error *error)
1904 {
1905         int ret;
1906
1907         ret = flow_drv_validate(dev, attr, items, actions, error);
1908         if (ret < 0)
1909                 return ret;
1910         return 0;
1911 }
1912
1913 /**
1914  * Get RSS action from the action list.
1915  *
1916  * @param[in] actions
1917  *   Pointer to the list of actions.
1918  *
1919  * @return
1920  *   Pointer to the RSS action if exist, else return NULL.
1921  */
1922 static const struct rte_flow_action_rss*
1923 mlx5_flow_get_rss_action(const struct rte_flow_action actions[])
1924 {
1925         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1926                 switch (actions->type) {
1927                 case RTE_FLOW_ACTION_TYPE_RSS:
1928                         return (const struct rte_flow_action_rss *)
1929                                actions->conf;
1930                 default:
1931                         break;
1932                 }
1933         }
1934         return NULL;
1935 }
1936
1937 static unsigned int
1938 mlx5_find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
1939 {
1940         const struct rte_flow_item *item;
1941         unsigned int has_vlan = 0;
1942
1943         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
1944                 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
1945                         has_vlan = 1;
1946                         break;
1947                 }
1948         }
1949         if (has_vlan)
1950                 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
1951                                        MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
1952         return rss_level < 2 ? MLX5_EXPANSION_ROOT :
1953                                MLX5_EXPANSION_ROOT_OUTER;
1954 }
1955
1956 /**
1957  * Create a flow and add it to @p list.
1958  *
1959  * @param dev
1960  *   Pointer to Ethernet device.
1961  * @param list
1962  *   Pointer to a TAILQ flow list.
1963  * @param[in] attr
1964  *   Flow rule attributes.
1965  * @param[in] items
1966  *   Pattern specification (list terminated by the END pattern item).
1967  * @param[in] actions
1968  *   Associated actions (list terminated by the END action).
1969  * @param[out] error
1970  *   Perform verbose error reporting if not NULL.
1971  *
1972  * @return
1973  *   A flow on success, NULL otherwise and rte_errno is set.
1974  */
1975 static struct rte_flow *
1976 mlx5_flow_list_create(struct rte_eth_dev *dev,
1977                       struct mlx5_flows *list,
1978                       const struct rte_flow_attr *attr,
1979                       const struct rte_flow_item items[],
1980                       const struct rte_flow_action actions[],
1981                       struct rte_flow_error *error)
1982 {
1983         struct rte_flow *flow = NULL;
1984         struct mlx5_flow *dev_flow;
1985         uint64_t action_flags = 0;
1986         uint64_t item_flags = 0;
1987         const struct rte_flow_action_rss *rss;
1988         union {
1989                 struct rte_flow_expand_rss buf;
1990                 uint8_t buffer[2048];
1991         } expand_buffer;
1992         struct rte_flow_expand_rss *buf = &expand_buffer.buf;
1993         int ret;
1994         uint32_t i;
1995         uint32_t flow_size;
1996
1997         ret = flow_drv_validate(dev, attr, items, actions, error);
1998         if (ret < 0)
1999                 return NULL;
2000         flow_size = sizeof(struct rte_flow);
2001         rss = mlx5_flow_get_rss_action(actions);
2002         if (rss)
2003                 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2004                                             sizeof(void *));
2005         else
2006                 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2007         flow = rte_calloc(__func__, 1, flow_size, 0);
2008         flow->drv_type = flow_get_drv_type(dev, attr);
2009         assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2010                flow->drv_type < MLX5_FLOW_TYPE_MAX);
2011         flow->queue = (void *)(flow + 1);
2012         LIST_INIT(&flow->dev_flows);
2013         if (rss && rss->types) {
2014                 unsigned int graph_root;
2015
2016                 graph_root = mlx5_find_graph_root(items, rss->level);
2017                 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2018                                           items, rss->types,
2019                                           mlx5_support_expansion,
2020                                           graph_root);
2021                 assert(ret > 0 &&
2022                        (unsigned int)ret < sizeof(expand_buffer.buffer));
2023         } else {
2024                 buf->entries = 1;
2025                 buf->entry[0].pattern = (void *)(uintptr_t)items;
2026         }
2027         for (i = 0; i < buf->entries; ++i) {
2028                 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2029                                             actions, &item_flags, &action_flags,
2030                                             error);
2031                 if (!dev_flow)
2032                         goto error;
2033                 dev_flow->flow = flow;
2034                 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2035                 ret = flow_drv_translate(dev, dev_flow, attr,
2036                                          buf->entry[i].pattern,
2037                                          actions, error);
2038                 if (ret < 0)
2039                         goto error;
2040         }
2041         if (dev->data->dev_started) {
2042                 ret = flow_drv_apply(dev, flow, error);
2043                 if (ret < 0)
2044                         goto error;
2045         }
2046         TAILQ_INSERT_TAIL(list, flow, next);
2047         mlx5_flow_rxq_flags_set(dev, flow);
2048         return flow;
2049 error:
2050         ret = rte_errno; /* Save rte_errno before cleanup. */
2051         assert(flow);
2052         flow_drv_destroy(dev, flow);
2053         rte_free(flow);
2054         rte_errno = ret; /* Restore rte_errno. */
2055         return NULL;
2056 }
2057
2058 /**
2059  * Create a flow.
2060  *
2061  * @see rte_flow_create()
2062  * @see rte_flow_ops
2063  */
2064 struct rte_flow *
2065 mlx5_flow_create(struct rte_eth_dev *dev,
2066                  const struct rte_flow_attr *attr,
2067                  const struct rte_flow_item items[],
2068                  const struct rte_flow_action actions[],
2069                  struct rte_flow_error *error)
2070 {
2071         return mlx5_flow_list_create
2072                 (dev, &((struct priv *)dev->data->dev_private)->flows,
2073                  attr, items, actions, error);
2074 }
2075
2076 /**
2077  * Destroy a flow in a list.
2078  *
2079  * @param dev
2080  *   Pointer to Ethernet device.
2081  * @param list
2082  *   Pointer to a TAILQ flow list.
2083  * @param[in] flow
2084  *   Flow to destroy.
2085  */
2086 static void
2087 mlx5_flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2088                        struct rte_flow *flow)
2089 {
2090         flow_drv_destroy(dev, flow);
2091         TAILQ_REMOVE(list, flow, next);
2092         /*
2093          * Update RX queue flags only if port is started, otherwise it is
2094          * already clean.
2095          */
2096         if (dev->data->dev_started)
2097                 mlx5_flow_rxq_flags_trim(dev, flow);
2098         rte_free(flow);
2099 }
2100
2101 /**
2102  * Destroy all flows.
2103  *
2104  * @param dev
2105  *   Pointer to Ethernet device.
2106  * @param list
2107  *   Pointer to a TAILQ flow list.
2108  */
2109 void
2110 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2111 {
2112         while (!TAILQ_EMPTY(list)) {
2113                 struct rte_flow *flow;
2114
2115                 flow = TAILQ_FIRST(list);
2116                 mlx5_flow_list_destroy(dev, list, flow);
2117         }
2118 }
2119
2120 /**
2121  * Remove all flows.
2122  *
2123  * @param dev
2124  *   Pointer to Ethernet device.
2125  * @param list
2126  *   Pointer to a TAILQ flow list.
2127  */
2128 void
2129 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2130 {
2131         struct rte_flow *flow;
2132
2133         TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2134                 flow_drv_remove(dev, flow);
2135         mlx5_flow_rxq_flags_clear(dev);
2136 }
2137
2138 /**
2139  * Add all flows.
2140  *
2141  * @param dev
2142  *   Pointer to Ethernet device.
2143  * @param list
2144  *   Pointer to a TAILQ flow list.
2145  *
2146  * @return
2147  *   0 on success, a negative errno value otherwise and rte_errno is set.
2148  */
2149 int
2150 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2151 {
2152         struct rte_flow *flow;
2153         struct rte_flow_error error;
2154         int ret = 0;
2155
2156         TAILQ_FOREACH(flow, list, next) {
2157                 ret = flow_drv_apply(dev, flow, &error);
2158                 if (ret < 0)
2159                         goto error;
2160                 mlx5_flow_rxq_flags_set(dev, flow);
2161         }
2162         return 0;
2163 error:
2164         ret = rte_errno; /* Save rte_errno before cleanup. */
2165         mlx5_flow_stop(dev, list);
2166         rte_errno = ret; /* Restore rte_errno. */
2167         return -rte_errno;
2168 }
2169
2170 /**
2171  * Verify the flow list is empty
2172  *
2173  * @param dev
2174  *  Pointer to Ethernet device.
2175  *
2176  * @return the number of flows not released.
2177  */
2178 int
2179 mlx5_flow_verify(struct rte_eth_dev *dev)
2180 {
2181         struct priv *priv = dev->data->dev_private;
2182         struct rte_flow *flow;
2183         int ret = 0;
2184
2185         TAILQ_FOREACH(flow, &priv->flows, next) {
2186                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
2187                         dev->data->port_id, (void *)flow);
2188                 ++ret;
2189         }
2190         return ret;
2191 }
2192
2193 /**
2194  * Enable a control flow configured from the control plane.
2195  *
2196  * @param dev
2197  *   Pointer to Ethernet device.
2198  * @param eth_spec
2199  *   An Ethernet flow spec to apply.
2200  * @param eth_mask
2201  *   An Ethernet flow mask to apply.
2202  * @param vlan_spec
2203  *   A VLAN flow spec to apply.
2204  * @param vlan_mask
2205  *   A VLAN flow mask to apply.
2206  *
2207  * @return
2208  *   0 on success, a negative errno value otherwise and rte_errno is set.
2209  */
2210 int
2211 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2212                     struct rte_flow_item_eth *eth_spec,
2213                     struct rte_flow_item_eth *eth_mask,
2214                     struct rte_flow_item_vlan *vlan_spec,
2215                     struct rte_flow_item_vlan *vlan_mask)
2216 {
2217         struct priv *priv = dev->data->dev_private;
2218         const struct rte_flow_attr attr = {
2219                 .ingress = 1,
2220                 .priority = MLX5_FLOW_PRIO_RSVD,
2221         };
2222         struct rte_flow_item items[] = {
2223                 {
2224                         .type = RTE_FLOW_ITEM_TYPE_ETH,
2225                         .spec = eth_spec,
2226                         .last = NULL,
2227                         .mask = eth_mask,
2228                 },
2229                 {
2230                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2231                                               RTE_FLOW_ITEM_TYPE_END,
2232                         .spec = vlan_spec,
2233                         .last = NULL,
2234                         .mask = vlan_mask,
2235                 },
2236                 {
2237                         .type = RTE_FLOW_ITEM_TYPE_END,
2238                 },
2239         };
2240         uint16_t queue[priv->reta_idx_n];
2241         struct rte_flow_action_rss action_rss = {
2242                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2243                 .level = 0,
2244                 .types = priv->rss_conf.rss_hf,
2245                 .key_len = priv->rss_conf.rss_key_len,
2246                 .queue_num = priv->reta_idx_n,
2247                 .key = priv->rss_conf.rss_key,
2248                 .queue = queue,
2249         };
2250         struct rte_flow_action actions[] = {
2251                 {
2252                         .type = RTE_FLOW_ACTION_TYPE_RSS,
2253                         .conf = &action_rss,
2254                 },
2255                 {
2256                         .type = RTE_FLOW_ACTION_TYPE_END,
2257                 },
2258         };
2259         struct rte_flow *flow;
2260         struct rte_flow_error error;
2261         unsigned int i;
2262
2263         if (!priv->reta_idx_n) {
2264                 rte_errno = EINVAL;
2265                 return -rte_errno;
2266         }
2267         for (i = 0; i != priv->reta_idx_n; ++i)
2268                 queue[i] = (*priv->reta_idx)[i];
2269         flow = mlx5_flow_list_create(dev, &priv->ctrl_flows, &attr, items,
2270                                      actions, &error);
2271         if (!flow)
2272                 return -rte_errno;
2273         return 0;
2274 }
2275
2276 /**
2277  * Enable a flow control configured from the control plane.
2278  *
2279  * @param dev
2280  *   Pointer to Ethernet device.
2281  * @param eth_spec
2282  *   An Ethernet flow spec to apply.
2283  * @param eth_mask
2284  *   An Ethernet flow mask to apply.
2285  *
2286  * @return
2287  *   0 on success, a negative errno value otherwise and rte_errno is set.
2288  */
2289 int
2290 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2291                struct rte_flow_item_eth *eth_spec,
2292                struct rte_flow_item_eth *eth_mask)
2293 {
2294         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2295 }
2296
2297 /**
2298  * Destroy a flow.
2299  *
2300  * @see rte_flow_destroy()
2301  * @see rte_flow_ops
2302  */
2303 int
2304 mlx5_flow_destroy(struct rte_eth_dev *dev,
2305                   struct rte_flow *flow,
2306                   struct rte_flow_error *error __rte_unused)
2307 {
2308         struct priv *priv = dev->data->dev_private;
2309
2310         mlx5_flow_list_destroy(dev, &priv->flows, flow);
2311         return 0;
2312 }
2313
2314 /**
2315  * Destroy all flows.
2316  *
2317  * @see rte_flow_flush()
2318  * @see rte_flow_ops
2319  */
2320 int
2321 mlx5_flow_flush(struct rte_eth_dev *dev,
2322                 struct rte_flow_error *error __rte_unused)
2323 {
2324         struct priv *priv = dev->data->dev_private;
2325
2326         mlx5_flow_list_flush(dev, &priv->flows);
2327         return 0;
2328 }
2329
2330 /**
2331  * Isolated mode.
2332  *
2333  * @see rte_flow_isolate()
2334  * @see rte_flow_ops
2335  */
2336 int
2337 mlx5_flow_isolate(struct rte_eth_dev *dev,
2338                   int enable,
2339                   struct rte_flow_error *error)
2340 {
2341         struct priv *priv = dev->data->dev_private;
2342
2343         if (dev->data->dev_started) {
2344                 rte_flow_error_set(error, EBUSY,
2345                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2346                                    NULL,
2347                                    "port must be stopped first");
2348                 return -rte_errno;
2349         }
2350         priv->isolated = !!enable;
2351         if (enable)
2352                 dev->dev_ops = &mlx5_dev_ops_isolate;
2353         else
2354                 dev->dev_ops = &mlx5_dev_ops;
2355         return 0;
2356 }
2357
2358 /**
2359  * Query a flow.
2360  *
2361  * @see rte_flow_query()
2362  * @see rte_flow_ops
2363  */
2364 static int
2365 flow_drv_query(struct rte_eth_dev *dev,
2366                struct rte_flow *flow,
2367                const struct rte_flow_action *actions,
2368                void *data,
2369                struct rte_flow_error *error)
2370 {
2371         const struct mlx5_flow_driver_ops *fops;
2372         enum mlx5_flow_drv_type ftype = flow->drv_type;
2373
2374         assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2375         fops = flow_get_drv_ops(ftype);
2376
2377         return fops->query(dev, flow, actions, data, error);
2378 }
2379
2380 /**
2381  * Query a flow.
2382  *
2383  * @see rte_flow_query()
2384  * @see rte_flow_ops
2385  */
2386 int
2387 mlx5_flow_query(struct rte_eth_dev *dev,
2388                 struct rte_flow *flow,
2389                 const struct rte_flow_action *actions,
2390                 void *data,
2391                 struct rte_flow_error *error)
2392 {
2393         int ret;
2394
2395         ret = flow_drv_query(dev, flow, actions, data, error);
2396         if (ret < 0)
2397                 return ret;
2398         return 0;
2399 }
2400
2401 /**
2402  * Convert a flow director filter to a generic flow.
2403  *
2404  * @param dev
2405  *   Pointer to Ethernet device.
2406  * @param fdir_filter
2407  *   Flow director filter to add.
2408  * @param attributes
2409  *   Generic flow parameters structure.
2410  *
2411  * @return
2412  *   0 on success, a negative errno value otherwise and rte_errno is set.
2413  */
2414 static int
2415 mlx5_fdir_filter_convert(struct rte_eth_dev *dev,
2416                          const struct rte_eth_fdir_filter *fdir_filter,
2417                          struct mlx5_fdir *attributes)
2418 {
2419         struct priv *priv = dev->data->dev_private;
2420         const struct rte_eth_fdir_input *input = &fdir_filter->input;
2421         const struct rte_eth_fdir_masks *mask =
2422                 &dev->data->dev_conf.fdir_conf.mask;
2423
2424         /* Validate queue number. */
2425         if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2426                 DRV_LOG(ERR, "port %u invalid queue number %d",
2427                         dev->data->port_id, fdir_filter->action.rx_queue);
2428                 rte_errno = EINVAL;
2429                 return -rte_errno;
2430         }
2431         attributes->attr.ingress = 1;
2432         attributes->items[0] = (struct rte_flow_item) {
2433                 .type = RTE_FLOW_ITEM_TYPE_ETH,
2434                 .spec = &attributes->l2,
2435                 .mask = &attributes->l2_mask,
2436         };
2437         switch (fdir_filter->action.behavior) {
2438         case RTE_ETH_FDIR_ACCEPT:
2439                 attributes->actions[0] = (struct rte_flow_action){
2440                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
2441                         .conf = &attributes->queue,
2442                 };
2443                 break;
2444         case RTE_ETH_FDIR_REJECT:
2445                 attributes->actions[0] = (struct rte_flow_action){
2446                         .type = RTE_FLOW_ACTION_TYPE_DROP,
2447                 };
2448                 break;
2449         default:
2450                 DRV_LOG(ERR, "port %u invalid behavior %d",
2451                         dev->data->port_id,
2452                         fdir_filter->action.behavior);
2453                 rte_errno = ENOTSUP;
2454                 return -rte_errno;
2455         }
2456         attributes->queue.index = fdir_filter->action.rx_queue;
2457         /* Handle L3. */
2458         switch (fdir_filter->input.flow_type) {
2459         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2460         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2461         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2462                 attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2463                         .src_addr = input->flow.ip4_flow.src_ip,
2464                         .dst_addr = input->flow.ip4_flow.dst_ip,
2465                         .time_to_live = input->flow.ip4_flow.ttl,
2466                         .type_of_service = input->flow.ip4_flow.tos,
2467                 };
2468                 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2469                         .src_addr = mask->ipv4_mask.src_ip,
2470                         .dst_addr = mask->ipv4_mask.dst_ip,
2471                         .time_to_live = mask->ipv4_mask.ttl,
2472                         .type_of_service = mask->ipv4_mask.tos,
2473                         .next_proto_id = mask->ipv4_mask.proto,
2474                 };
2475                 attributes->items[1] = (struct rte_flow_item){
2476                         .type = RTE_FLOW_ITEM_TYPE_IPV4,
2477                         .spec = &attributes->l3,
2478                         .mask = &attributes->l3_mask,
2479                 };
2480                 break;
2481         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2482         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2483         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2484                 attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2485                         .hop_limits = input->flow.ipv6_flow.hop_limits,
2486                         .proto = input->flow.ipv6_flow.proto,
2487                 };
2488
2489                 memcpy(attributes->l3.ipv6.hdr.src_addr,
2490                        input->flow.ipv6_flow.src_ip,
2491                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2492                 memcpy(attributes->l3.ipv6.hdr.dst_addr,
2493                        input->flow.ipv6_flow.dst_ip,
2494                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2495                 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2496                        mask->ipv6_mask.src_ip,
2497                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2498                 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2499                        mask->ipv6_mask.dst_ip,
2500                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2501                 attributes->items[1] = (struct rte_flow_item){
2502                         .type = RTE_FLOW_ITEM_TYPE_IPV6,
2503                         .spec = &attributes->l3,
2504                         .mask = &attributes->l3_mask,
2505                 };
2506                 break;
2507         default:
2508                 DRV_LOG(ERR, "port %u invalid flow type%d",
2509                         dev->data->port_id, fdir_filter->input.flow_type);
2510                 rte_errno = ENOTSUP;
2511                 return -rte_errno;
2512         }
2513         /* Handle L4. */
2514         switch (fdir_filter->input.flow_type) {
2515         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2516                 attributes->l4.udp.hdr = (struct udp_hdr){
2517                         .src_port = input->flow.udp4_flow.src_port,
2518                         .dst_port = input->flow.udp4_flow.dst_port,
2519                 };
2520                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2521                         .src_port = mask->src_port_mask,
2522                         .dst_port = mask->dst_port_mask,
2523                 };
2524                 attributes->items[2] = (struct rte_flow_item){
2525                         .type = RTE_FLOW_ITEM_TYPE_UDP,
2526                         .spec = &attributes->l4,
2527                         .mask = &attributes->l4_mask,
2528                 };
2529                 break;
2530         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2531                 attributes->l4.tcp.hdr = (struct tcp_hdr){
2532                         .src_port = input->flow.tcp4_flow.src_port,
2533                         .dst_port = input->flow.tcp4_flow.dst_port,
2534                 };
2535                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2536                         .src_port = mask->src_port_mask,
2537                         .dst_port = mask->dst_port_mask,
2538                 };
2539                 attributes->items[2] = (struct rte_flow_item){
2540                         .type = RTE_FLOW_ITEM_TYPE_TCP,
2541                         .spec = &attributes->l4,
2542                         .mask = &attributes->l4_mask,
2543                 };
2544                 break;
2545         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2546                 attributes->l4.udp.hdr = (struct udp_hdr){
2547                         .src_port = input->flow.udp6_flow.src_port,
2548                         .dst_port = input->flow.udp6_flow.dst_port,
2549                 };
2550                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2551                         .src_port = mask->src_port_mask,
2552                         .dst_port = mask->dst_port_mask,
2553                 };
2554                 attributes->items[2] = (struct rte_flow_item){
2555                         .type = RTE_FLOW_ITEM_TYPE_UDP,
2556                         .spec = &attributes->l4,
2557                         .mask = &attributes->l4_mask,
2558                 };
2559                 break;
2560         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2561                 attributes->l4.tcp.hdr = (struct tcp_hdr){
2562                         .src_port = input->flow.tcp6_flow.src_port,
2563                         .dst_port = input->flow.tcp6_flow.dst_port,
2564                 };
2565                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2566                         .src_port = mask->src_port_mask,
2567                         .dst_port = mask->dst_port_mask,
2568                 };
2569                 attributes->items[2] = (struct rte_flow_item){
2570                         .type = RTE_FLOW_ITEM_TYPE_TCP,
2571                         .spec = &attributes->l4,
2572                         .mask = &attributes->l4_mask,
2573                 };
2574                 break;
2575         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2576         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2577                 break;
2578         default:
2579                 DRV_LOG(ERR, "port %u invalid flow type%d",
2580                         dev->data->port_id, fdir_filter->input.flow_type);
2581                 rte_errno = ENOTSUP;
2582                 return -rte_errno;
2583         }
2584         return 0;
2585 }
2586
2587 /**
2588  * Add new flow director filter and store it in list.
2589  *
2590  * @param dev
2591  *   Pointer to Ethernet device.
2592  * @param fdir_filter
2593  *   Flow director filter to add.
2594  *
2595  * @return
2596  *   0 on success, a negative errno value otherwise and rte_errno is set.
2597  */
2598 static int
2599 mlx5_fdir_filter_add(struct rte_eth_dev *dev,
2600                      const struct rte_eth_fdir_filter *fdir_filter)
2601 {
2602         struct priv *priv = dev->data->dev_private;
2603         struct mlx5_fdir attributes = {
2604                 .attr.group = 0,
2605                 .l2_mask = {
2606                         .dst.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2607                         .src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2608                         .type = 0,
2609                 },
2610         };
2611         struct rte_flow_error error;
2612         struct rte_flow *flow;
2613         int ret;
2614
2615         ret = mlx5_fdir_filter_convert(dev, fdir_filter, &attributes);
2616         if (ret)
2617                 return ret;
2618         flow = mlx5_flow_list_create(dev, &priv->flows, &attributes.attr,
2619                                      attributes.items, attributes.actions,
2620                                      &error);
2621         if (flow) {
2622                 DRV_LOG(DEBUG, "port %u FDIR created %p", dev->data->port_id,
2623                         (void *)flow);
2624                 return 0;
2625         }
2626         return -rte_errno;
2627 }
2628
2629 /**
2630  * Delete specific filter.
2631  *
2632  * @param dev
2633  *   Pointer to Ethernet device.
2634  * @param fdir_filter
2635  *   Filter to be deleted.
2636  *
2637  * @return
2638  *   0 on success, a negative errno value otherwise and rte_errno is set.
2639  */
2640 static int
2641 mlx5_fdir_filter_delete(struct rte_eth_dev *dev __rte_unused,
2642                         const struct rte_eth_fdir_filter *fdir_filter
2643                         __rte_unused)
2644 {
2645         rte_errno = ENOTSUP;
2646         return -rte_errno;
2647 }
2648
2649 /**
2650  * Update queue for specific filter.
2651  *
2652  * @param dev
2653  *   Pointer to Ethernet device.
2654  * @param fdir_filter
2655  *   Filter to be updated.
2656  *
2657  * @return
2658  *   0 on success, a negative errno value otherwise and rte_errno is set.
2659  */
2660 static int
2661 mlx5_fdir_filter_update(struct rte_eth_dev *dev,
2662                         const struct rte_eth_fdir_filter *fdir_filter)
2663 {
2664         int ret;
2665
2666         ret = mlx5_fdir_filter_delete(dev, fdir_filter);
2667         if (ret)
2668                 return ret;
2669         return mlx5_fdir_filter_add(dev, fdir_filter);
2670 }
2671
2672 /**
2673  * Flush all filters.
2674  *
2675  * @param dev
2676  *   Pointer to Ethernet device.
2677  */
2678 static void
2679 mlx5_fdir_filter_flush(struct rte_eth_dev *dev)
2680 {
2681         struct priv *priv = dev->data->dev_private;
2682
2683         mlx5_flow_list_flush(dev, &priv->flows);
2684 }
2685
2686 /**
2687  * Get flow director information.
2688  *
2689  * @param dev
2690  *   Pointer to Ethernet device.
2691  * @param[out] fdir_info
2692  *   Resulting flow director information.
2693  */
2694 static void
2695 mlx5_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2696 {
2697         struct rte_eth_fdir_masks *mask =
2698                 &dev->data->dev_conf.fdir_conf.mask;
2699
2700         fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2701         fdir_info->guarant_spc = 0;
2702         rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2703         fdir_info->max_flexpayload = 0;
2704         fdir_info->flow_types_mask[0] = 0;
2705         fdir_info->flex_payload_unit = 0;
2706         fdir_info->max_flex_payload_segment_num = 0;
2707         fdir_info->flex_payload_limit = 0;
2708         memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2709 }
2710
2711 /**
2712  * Deal with flow director operations.
2713  *
2714  * @param dev
2715  *   Pointer to Ethernet device.
2716  * @param filter_op
2717  *   Operation to perform.
2718  * @param arg
2719  *   Pointer to operation-specific structure.
2720  *
2721  * @return
2722  *   0 on success, a negative errno value otherwise and rte_errno is set.
2723  */
2724 static int
2725 mlx5_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2726                     void *arg)
2727 {
2728         enum rte_fdir_mode fdir_mode =
2729                 dev->data->dev_conf.fdir_conf.mode;
2730
2731         if (filter_op == RTE_ETH_FILTER_NOP)
2732                 return 0;
2733         if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2734             fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2735                 DRV_LOG(ERR, "port %u flow director mode %d not supported",
2736                         dev->data->port_id, fdir_mode);
2737                 rte_errno = EINVAL;
2738                 return -rte_errno;
2739         }
2740         switch (filter_op) {
2741         case RTE_ETH_FILTER_ADD:
2742                 return mlx5_fdir_filter_add(dev, arg);
2743         case RTE_ETH_FILTER_UPDATE:
2744                 return mlx5_fdir_filter_update(dev, arg);
2745         case RTE_ETH_FILTER_DELETE:
2746                 return mlx5_fdir_filter_delete(dev, arg);
2747         case RTE_ETH_FILTER_FLUSH:
2748                 mlx5_fdir_filter_flush(dev);
2749                 break;
2750         case RTE_ETH_FILTER_INFO:
2751                 mlx5_fdir_info_get(dev, arg);
2752                 break;
2753         default:
2754                 DRV_LOG(DEBUG, "port %u unknown operation %u",
2755                         dev->data->port_id, filter_op);
2756                 rte_errno = EINVAL;
2757                 return -rte_errno;
2758         }
2759         return 0;
2760 }
2761
2762 /**
2763  * Manage filter operations.
2764  *
2765  * @param dev
2766  *   Pointer to Ethernet device structure.
2767  * @param filter_type
2768  *   Filter type.
2769  * @param filter_op
2770  *   Operation to perform.
2771  * @param arg
2772  *   Pointer to operation-specific structure.
2773  *
2774  * @return
2775  *   0 on success, a negative errno value otherwise and rte_errno is set.
2776  */
2777 int
2778 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2779                      enum rte_filter_type filter_type,
2780                      enum rte_filter_op filter_op,
2781                      void *arg)
2782 {
2783         switch (filter_type) {
2784         case RTE_ETH_FILTER_GENERIC:
2785                 if (filter_op != RTE_ETH_FILTER_GET) {
2786                         rte_errno = EINVAL;
2787                         return -rte_errno;
2788                 }
2789                 *(const void **)arg = &mlx5_flow_ops;
2790                 return 0;
2791         case RTE_ETH_FILTER_FDIR:
2792                 return mlx5_fdir_ctrl_func(dev, filter_op, arg);
2793         default:
2794                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
2795                         dev->data->port_id, filter_type);
2796                 rte_errno = ENOTSUP;
2797                 return -rte_errno;
2798         }
2799         return 0;
2800 }