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_ethdev_driver.h>
  25 #include <rte_flow.h>
  26 #include <rte_flow_driver.h>
  27 #include <rte_malloc.h>
  28 #include <rte_ip.h>
  29
  30 #include "mlx5.h"
  31 #include "mlx5_defs.h"
  32 #include "mlx5_flow.h"
  33 #include "mlx5_glue.h"
  34 #include "mlx5_prm.h"
  35 #include "mlx5_rxtx.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_verbs_drv_ops;
  46
  47 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
  48
  49 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
  50         [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
  51 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
  52         [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
  53 #endif
  54         [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
  55         [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
  56 };
  57
  58 enum mlx5_expansion {
  59         MLX5_EXPANSION_ROOT,
  60         MLX5_EXPANSION_ROOT_OUTER,
  61         MLX5_EXPANSION_ROOT_ETH_VLAN,
  62         MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
  63         MLX5_EXPANSION_OUTER_ETH,
  64         MLX5_EXPANSION_OUTER_ETH_VLAN,
  65         MLX5_EXPANSION_OUTER_VLAN,
  66         MLX5_EXPANSION_OUTER_IPV4,
  67         MLX5_EXPANSION_OUTER_IPV4_UDP,
  68         MLX5_EXPANSION_OUTER_IPV4_TCP,
  69         MLX5_EXPANSION_OUTER_IPV6,
  70         MLX5_EXPANSION_OUTER_IPV6_UDP,
  71         MLX5_EXPANSION_OUTER_IPV6_TCP,
  72         MLX5_EXPANSION_VXLAN,
  73         MLX5_EXPANSION_VXLAN_GPE,
  74         MLX5_EXPANSION_GRE,
  75         MLX5_EXPANSION_MPLS,
  76         MLX5_EXPANSION_ETH,
  77         MLX5_EXPANSION_ETH_VLAN,
  78         MLX5_EXPANSION_VLAN,
  79         MLX5_EXPANSION_IPV4,
  80         MLX5_EXPANSION_IPV4_UDP,
  81         MLX5_EXPANSION_IPV4_TCP,
  82         MLX5_EXPANSION_IPV6,
  83         MLX5_EXPANSION_IPV6_UDP,
  84         MLX5_EXPANSION_IPV6_TCP,
  85 };
  86
  87 /** Supported expansion of items. */
  88 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
  89         [MLX5_EXPANSION_ROOT] = {
  90                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
  91                                                  MLX5_EXPANSION_IPV4,
  92                                                  MLX5_EXPANSION_IPV6),
  93                 .type = RTE_FLOW_ITEM_TYPE_END,
  94         },
  95         [MLX5_EXPANSION_ROOT_OUTER] = {
  96                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
  97                                                  MLX5_EXPANSION_OUTER_IPV4,
  98                                                  MLX5_EXPANSION_OUTER_IPV6),
  99                 .type = RTE_FLOW_ITEM_TYPE_END,
 100         },
 101         [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
 102                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
 103                 .type = RTE_FLOW_ITEM_TYPE_END,
 104         },
 105         [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
 106                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
 107                 .type = RTE_FLOW_ITEM_TYPE_END,
 108         },
 109         [MLX5_EXPANSION_OUTER_ETH] = {
 110                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
 111                                                  MLX5_EXPANSION_OUTER_IPV6,
 112                                                  MLX5_EXPANSION_MPLS),
 113                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 114                 .rss_types = 0,
 115         },
 116         [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
 117                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
 118                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 119                 .rss_types = 0,
 120         },
 121         [MLX5_EXPANSION_OUTER_VLAN] = {
 122                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
 123                                                  MLX5_EXPANSION_OUTER_IPV6),
 124                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
 125         },
 126         [MLX5_EXPANSION_OUTER_IPV4] = {
 127                 .next = RTE_FLOW_EXPAND_RSS_NEXT
 128                         (MLX5_EXPANSION_OUTER_IPV4_UDP,
 129                          MLX5_EXPANSION_OUTER_IPV4_TCP,
 130                          MLX5_EXPANSION_GRE,
 131                          MLX5_EXPANSION_IPV4,
 132                          MLX5_EXPANSION_IPV6),
 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                          MLX5_EXPANSION_IPV4,
 152                          MLX5_EXPANSION_IPV6),
 153                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
 154                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
 155                         ETH_RSS_NONFRAG_IPV6_OTHER,
 156         },
 157         [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
 158                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
 159                                                  MLX5_EXPANSION_VXLAN_GPE),
 160                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 161                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
 162         },
 163         [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
 164                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 165                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
 166         },
 167         [MLX5_EXPANSION_VXLAN] = {
 168                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
 169                 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
 170         },
 171         [MLX5_EXPANSION_VXLAN_GPE] = {
 172                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
 173                                                  MLX5_EXPANSION_IPV4,
 174                                                  MLX5_EXPANSION_IPV6),
 175                 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
 176         },
 177         [MLX5_EXPANSION_GRE] = {
 178                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
 179                 .type = RTE_FLOW_ITEM_TYPE_GRE,
 180         },
 181         [MLX5_EXPANSION_MPLS] = {
 182                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 183                                                  MLX5_EXPANSION_IPV6),
 184                 .type = RTE_FLOW_ITEM_TYPE_MPLS,
 185         },
 186         [MLX5_EXPANSION_ETH] = {
 187                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 188                                                  MLX5_EXPANSION_IPV6),
 189                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 190         },
 191         [MLX5_EXPANSION_ETH_VLAN] = {
 192                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
 193                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 194         },
 195         [MLX5_EXPANSION_VLAN] = {
 196                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 197                                                  MLX5_EXPANSION_IPV6),
 198                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
 199         },
 200         [MLX5_EXPANSION_IPV4] = {
 201                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
 202                                                  MLX5_EXPANSION_IPV4_TCP),
 203                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
 204                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
 205                         ETH_RSS_NONFRAG_IPV4_OTHER,
 206         },
 207         [MLX5_EXPANSION_IPV4_UDP] = {
 208                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 209                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
 210         },
 211         [MLX5_EXPANSION_IPV4_TCP] = {
 212                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 213                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
 214         },
 215         [MLX5_EXPANSION_IPV6] = {
 216                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
 217                                                  MLX5_EXPANSION_IPV6_TCP),
 218                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
 219                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
 220                         ETH_RSS_NONFRAG_IPV6_OTHER,
 221         },
 222         [MLX5_EXPANSION_IPV6_UDP] = {
 223                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 224                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
 225         },
 226         [MLX5_EXPANSION_IPV6_TCP] = {
 227                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 228                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
 229         },
 230 };
 231
 232 static const struct rte_flow_ops mlx5_flow_ops = {
 233         .validate = mlx5_flow_validate,
 234         .create = mlx5_flow_create,
 235         .destroy = mlx5_flow_destroy,
 236         .flush = mlx5_flow_flush,
 237         .isolate = mlx5_flow_isolate,
 238         .query = mlx5_flow_query,
 239         .dev_dump = mlx5_flow_dev_dump,
 240 };
 241
 242 /* Convert FDIR request to Generic flow. */
 243 struct mlx5_fdir {
 244         struct rte_flow_attr attr;
 245         struct rte_flow_item items[4];
 246         struct rte_flow_item_eth l2;
 247         struct rte_flow_item_eth l2_mask;
 248         union {
 249                 struct rte_flow_item_ipv4 ipv4;
 250                 struct rte_flow_item_ipv6 ipv6;
 251         } l3;
 252         union {
 253                 struct rte_flow_item_ipv4 ipv4;
 254                 struct rte_flow_item_ipv6 ipv6;
 255         } l3_mask;
 256         union {
 257                 struct rte_flow_item_udp udp;
 258                 struct rte_flow_item_tcp tcp;
 259         } l4;
 260         union {
 261                 struct rte_flow_item_udp udp;
 262                 struct rte_flow_item_tcp tcp;
 263         } l4_mask;
 264         struct rte_flow_action actions[2];
 265         struct rte_flow_action_queue queue;
 266 };
 267
 268 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
 269 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
 270         { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
 271 };
 272
 273 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
 274 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
 275         { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
 276         { 9, 10, 11 }, { 12, 13, 14 },
 277 };
 278
 279 /* Tunnel information. */
 280 struct mlx5_flow_tunnel_info {
 281         uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
 282         uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
 283 };
 284
 285 static struct mlx5_flow_tunnel_info tunnels_info[] = {
 286         {
 287                 .tunnel = MLX5_FLOW_LAYER_VXLAN,
 288                 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
 289         },
 290         {
 291                 .tunnel = MLX5_FLOW_LAYER_GENEVE,
 292                 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
 293         },
 294         {
 295                 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
 296                 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
 297         },
 298         {
 299                 .tunnel = MLX5_FLOW_LAYER_GRE,
 300                 .ptype = RTE_PTYPE_TUNNEL_GRE,
 301         },
 302         {
 303                 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
 304                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
 305         },
 306         {
 307                 .tunnel = MLX5_FLOW_LAYER_MPLS,
 308                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
 309         },
 310         {
 311                 .tunnel = MLX5_FLOW_LAYER_NVGRE,
 312                 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
 313         },
 314         {
 315                 .tunnel = MLX5_FLOW_LAYER_IPIP,
 316                 .ptype = RTE_PTYPE_TUNNEL_IP,
 317         },
 318         {
 319                 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
 320                 .ptype = RTE_PTYPE_TUNNEL_IP,
 321         },
 322         {
 323                 .tunnel = MLX5_FLOW_LAYER_GTP,
 324                 .ptype = RTE_PTYPE_TUNNEL_GTPU,
 325         },
 326 };
 327
 328 /**
 329  * Translate tag ID to register.
 330  *
 331  * @param[in] dev
 332  *   Pointer to the Ethernet device structure.
 333  * @param[in] feature
 334  *   The feature that request the register.
 335  * @param[in] id
 336  *   The request register ID.
 337  * @param[out] error
 338  *   Error description in case of any.
 339  *
 340  * @return
 341  *   The request register on success, a negative errno
 342  *   value otherwise and rte_errno is set.
 343  */
 344 enum modify_reg
 345 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
 346                      enum mlx5_feature_name feature,
 347                      uint32_t id,
 348                      struct rte_flow_error *error)
 349 {
 350         struct mlx5_priv *priv = dev->data->dev_private;
 351         struct mlx5_dev_config *config = &priv->config;
 352         enum modify_reg start_reg;
 353
 354         switch (feature) {
 355         case MLX5_HAIRPIN_RX:
 356                 return REG_B;
 357         case MLX5_HAIRPIN_TX:
 358                 return REG_A;
 359         case MLX5_METADATA_RX:
 360                 switch (config->dv_xmeta_en) {
 361                 case MLX5_XMETA_MODE_LEGACY:
 362                         return REG_B;
 363                 case MLX5_XMETA_MODE_META16:
 364                         return REG_C_0;
 365                 case MLX5_XMETA_MODE_META32:
 366                         return REG_C_1;
 367                 }
 368                 break;
 369         case MLX5_METADATA_TX:
 370                 return REG_A;
 371         case MLX5_METADATA_FDB:
 372                 switch (config->dv_xmeta_en) {
 373                 case MLX5_XMETA_MODE_LEGACY:
 374                         return REG_NONE;
 375                 case MLX5_XMETA_MODE_META16:
 376                         return REG_C_0;
 377                 case MLX5_XMETA_MODE_META32:
 378                         return REG_C_1;
 379                 }
 380                 break;
 381         case MLX5_FLOW_MARK:
 382                 switch (config->dv_xmeta_en) {
 383                 case MLX5_XMETA_MODE_LEGACY:
 384                         return REG_NONE;
 385                 case MLX5_XMETA_MODE_META16:
 386                         return REG_C_1;
 387                 case MLX5_XMETA_MODE_META32:
 388                         return REG_C_0;
 389                 }
 390                 break;
 391         case MLX5_COPY_MARK:
 392         case MLX5_MTR_SFX:
 393                 /*
 394                  * Metadata COPY_MARK register using is in meter suffix sub
 395                  * flow while with meter. It's safe to share the same register.
 396                  */
 397                 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
 398         case MLX5_MTR_COLOR:
 399                 RTE_ASSERT(priv->mtr_color_reg != REG_NONE);
 400                 return priv->mtr_color_reg;
 401         case MLX5_APP_TAG:
 402                 /*
 403                  * If meter is enable, it will engage two registers for color
 404                  * match and flow match. If meter color match is not using the
 405                  * REG_C_2, need to skip the REG_C_x be used by meter color
 406                  * match.
 407                  * If meter is disable, free to use all available registers.
 408                  */
 409                 if (priv->mtr_color_reg != REG_NONE)
 410                         start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_3 :
 411                                     REG_C_4;
 412                 else
 413                         start_reg = REG_C_2;
 414                 if (id > (REG_C_7 - start_reg))
 415                         return rte_flow_error_set(error, EINVAL,
 416                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 417                                                   NULL, "invalid tag id");
 418                 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
 419                         return rte_flow_error_set(error, ENOTSUP,
 420                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 421                                                   NULL, "unsupported tag id");
 422                 /*
 423                  * This case means meter is using the REG_C_x great than 2.
 424                  * Take care not to conflict with meter color REG_C_x.
 425                  * If the available index REG_C_y >= REG_C_x, skip the
 426                  * color register.
 427                  */
 428                 if (start_reg == REG_C_3 && config->flow_mreg_c
 429                     [id + REG_C_3 - REG_C_0] >= priv->mtr_color_reg) {
 430                         if (config->flow_mreg_c[id + 1 + REG_C_3 - REG_C_0] !=
 431                             REG_NONE)
 432                                 return config->flow_mreg_c
 433                                                 [id + 1 + REG_C_3 - REG_C_0];
 434                         return rte_flow_error_set(error, ENOTSUP,
 435                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 436                                                   NULL, "unsupported tag id");
 437                 }
 438                 return config->flow_mreg_c[id + start_reg - REG_C_0];
 439         }
 440         assert(false);
 441         return rte_flow_error_set(error, EINVAL,
 442                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
 443                                   NULL, "invalid feature name");
 444 }
 445
 446 /**
 447  * Check extensive flow metadata register support.
 448  *
 449  * @param dev
 450  *   Pointer to rte_eth_dev structure.
 451  *
 452  * @return
 453  *   True if device supports extensive flow metadata register, otherwise false.
 454  */
 455 bool
 456 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
 457 {
 458         struct mlx5_priv *priv = dev->data->dev_private;
 459         struct mlx5_dev_config *config = &priv->config;
 460
 461         /*
 462          * Having available reg_c can be regarded inclusively as supporting
 463          * extensive flow metadata register, which could mean,
 464          * - metadata register copy action by modify header.
 465          * - 16 modify header actions is supported.
 466          * - reg_c's are preserved across different domain (FDB and NIC) on
 467          *   packet loopback by flow lookup miss.
 468          */
 469         return config->flow_mreg_c[2] != REG_NONE;
 470 }
 471
 472 /**
 473  * Discover the maximum number of priority available.
 474  *
 475  * @param[in] dev
 476  *   Pointer to the Ethernet device structure.
 477  *
 478  * @return
 479  *   number of supported flow priority on success, a negative errno
 480  *   value otherwise and rte_errno is set.
 481  */
 482 int
 483 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 484 {
 485         struct mlx5_priv *priv = dev->data->dev_private;
 486         struct {
 487                 struct ibv_flow_attr attr;
 488                 struct ibv_flow_spec_eth eth;
 489                 struct ibv_flow_spec_action_drop drop;
 490         } flow_attr = {
 491                 .attr = {
 492                         .num_of_specs = 2,
 493                         .port = (uint8_t)priv->ibv_port,
 494                 },
 495                 .eth = {
 496                         .type = IBV_FLOW_SPEC_ETH,
 497                         .size = sizeof(struct ibv_flow_spec_eth),
 498                 },
 499                 .drop = {
 500                         .size = sizeof(struct ibv_flow_spec_action_drop),
 501                         .type = IBV_FLOW_SPEC_ACTION_DROP,
 502                 },
 503         };
 504         struct ibv_flow *flow;
 505         struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
 506         uint16_t vprio[] = { 8, 16 };
 507         int i;
 508         int priority = 0;
 509
 510         if (!drop) {
 511                 rte_errno = ENOTSUP;
 512                 return -rte_errno;
 513         }
 514         for (i = 0; i != RTE_DIM(vprio); i++) {
 515                 flow_attr.attr.priority = vprio[i] - 1;
 516                 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 517                 if (!flow)
 518                         break;
 519                 claim_zero(mlx5_glue->destroy_flow(flow));
 520                 priority = vprio[i];
 521         }
 522         mlx5_hrxq_drop_release(dev);
 523         switch (priority) {
 524         case 8:
 525                 priority = RTE_DIM(priority_map_3);
 526                 break;
 527         case 16:
 528                 priority = RTE_DIM(priority_map_5);
 529                 break;
 530         default:
 531                 rte_errno = ENOTSUP;
 532                 DRV_LOG(ERR,
 533                         "port %u verbs maximum priority: %d expected 8/16",
 534                         dev->data->port_id, priority);
 535                 return -rte_errno;
 536         }
 537         DRV_LOG(INFO, "port %u flow maximum priority: %d",
 538                 dev->data->port_id, priority);
 539         return priority;
 540 }
 541
 542 /**
 543  * Adjust flow priority based on the highest layer and the request priority.
 544  *
 545  * @param[in] dev
 546  *   Pointer to the Ethernet device structure.
 547  * @param[in] priority
 548  *   The rule base priority.
 549  * @param[in] subpriority
 550  *   The priority based on the items.
 551  *
 552  * @return
 553  *   The new priority.
 554  */
 555 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
 556                                    uint32_t subpriority)
 557 {
 558         uint32_t res = 0;
 559         struct mlx5_priv *priv = dev->data->dev_private;
 560
 561         switch (priv->config.flow_prio) {
 562         case RTE_DIM(priority_map_3):
 563                 res = priority_map_3[priority][subpriority];
 564                 break;
 565         case RTE_DIM(priority_map_5):
 566                 res = priority_map_5[priority][subpriority];
 567                 break;
 568         }
 569         return  res;
 570 }
 571
 572 /**
 573  * Verify the @p item specifications (spec, last, mask) are compatible with the
 574  * NIC capabilities.
 575  *
 576  * @param[in] item
 577  *   Item specification.
 578  * @param[in] mask
 579  *   @p item->mask or flow default bit-masks.
 580  * @param[in] nic_mask
 581  *   Bit-masks covering supported fields by the NIC to compare with user mask.
 582  * @param[in] size
 583  *   Bit-masks size in bytes.
 584  * @param[out] error
 585  *   Pointer to error structure.
 586  *
 587  * @return
 588  *   0 on success, a negative errno value otherwise and rte_errno is set.
 589  */
 590 int
 591 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
 592                           const uint8_t *mask,
 593                           const uint8_t *nic_mask,
 594                           unsigned int size,
 595                           struct rte_flow_error *error)
 596 {
 597         unsigned int i;
 598
 599         assert(nic_mask);
 600         for (i = 0; i < size; ++i)
 601                 if ((nic_mask[i] | mask[i]) != nic_mask[i])
 602                         return rte_flow_error_set(error, ENOTSUP,
 603                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 604                                                   item,
 605                                                   "mask enables non supported"
 606                                                   " bits");
 607         if (!item->spec && (item->mask || item->last))
 608                 return rte_flow_error_set(error, EINVAL,
 609                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
 610                                           "mask/last without a spec is not"
 611                                           " supported");
 612         if (item->spec && item->last) {
 613                 uint8_t spec[size];
 614                 uint8_t last[size];
 615                 unsigned int i;
 616                 int ret;
 617
 618                 for (i = 0; i < size; ++i) {
 619                         spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
 620                         last[i] = ((const uint8_t *)item->last)[i] & mask[i];
 621                 }
 622                 ret = memcmp(spec, last, size);
 623                 if (ret != 0)
 624                         return rte_flow_error_set(error, EINVAL,
 625                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 626                                                   item,
 627                                                   "range is not valid");
 628         }
 629         return 0;
 630 }
 631
 632 /**
 633  * Adjust the hash fields according to the @p flow information.
 634  *
 635  * @param[in] dev_flow.
 636  *   Pointer to the mlx5_flow.
 637  * @param[in] tunnel
 638  *   1 when the hash field is for a tunnel item.
 639  * @param[in] layer_types
 640  *   ETH_RSS_* types.
 641  * @param[in] hash_fields
 642  *   Item hash fields.
 643  *
 644  * @return
 645  *   The hash fields that should be used.
 646  */
 647 uint64_t
 648 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
 649                             int tunnel __rte_unused, uint64_t layer_types,
 650                             uint64_t hash_fields)
 651 {
 652         struct rte_flow *flow = dev_flow->flow;
 653 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
 654         int rss_request_inner = flow->rss.level >= 2;
 655
 656         /* Check RSS hash level for tunnel. */
 657         if (tunnel && rss_request_inner)
 658                 hash_fields |= IBV_RX_HASH_INNER;
 659         else if (tunnel || rss_request_inner)
 660                 return 0;
 661 #endif
 662         /* Check if requested layer matches RSS hash fields. */
 663         if (!(flow->rss.types & layer_types))
 664                 return 0;
 665         return hash_fields;
 666 }
 667
 668 /**
 669  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
 670  * if several tunnel rules are used on this queue, the tunnel ptype will be
 671  * cleared.
 672  *
 673  * @param rxq_ctrl
 674  *   Rx queue to update.
 675  */
 676 static void
 677 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
 678 {
 679         unsigned int i;
 680         uint32_t tunnel_ptype = 0;
 681
 682         /* Look up for the ptype to use. */
 683         for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
 684                 if (!rxq_ctrl->flow_tunnels_n[i])
 685                         continue;
 686                 if (!tunnel_ptype) {
 687                         tunnel_ptype = tunnels_info[i].ptype;
 688                 } else {
 689                         tunnel_ptype = 0;
 690                         break;
 691                 }
 692         }
 693         rxq_ctrl->rxq.tunnel = tunnel_ptype;
 694 }
 695
 696 /**
 697  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
 698  * flow.
 699  *
 700  * @param[in] dev
 701  *   Pointer to the Ethernet device structure.
 702  * @param[in] dev_flow
 703  *   Pointer to device flow structure.
 704  */
 705 static void
 706 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
 707 {
 708         struct mlx5_priv *priv = dev->data->dev_private;
 709         struct rte_flow *flow = dev_flow->flow;
 710         const int mark = !!(dev_flow->actions &
 711                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
 712         const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
 713         unsigned int i;
 714
 715         for (i = 0; i != flow->rss.queue_num; ++i) {
 716                 int idx = (*flow->rss.queue)[i];
 717                 struct mlx5_rxq_ctrl *rxq_ctrl =
 718                         container_of((*priv->rxqs)[idx],
 719                                      struct mlx5_rxq_ctrl, rxq);
 720
 721                 /*
 722                  * To support metadata register copy on Tx loopback,
 723                  * this must be always enabled (metadata may arive
 724                  * from other port - not from local flows only.
 725                  */
 726                 if (priv->config.dv_flow_en &&
 727                     priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
 728                     mlx5_flow_ext_mreg_supported(dev)) {
 729                         rxq_ctrl->rxq.mark = 1;
 730                         rxq_ctrl->flow_mark_n = 1;
 731                 } else if (mark) {
 732                         rxq_ctrl->rxq.mark = 1;
 733                         rxq_ctrl->flow_mark_n++;
 734                 }
 735                 if (tunnel) {
 736                         unsigned int j;
 737
 738                         /* Increase the counter matching the flow. */
 739                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 740                                 if ((tunnels_info[j].tunnel &
 741                                      dev_flow->layers) ==
 742                                     tunnels_info[j].tunnel) {
 743                                         rxq_ctrl->flow_tunnels_n[j]++;
 744                                         break;
 745                                 }
 746                         }
 747                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
 748                 }
 749         }
 750 }
 751
 752 /**
 753  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
 754  *
 755  * @param[in] dev
 756  *   Pointer to the Ethernet device structure.
 757  * @param[in] flow
 758  *   Pointer to flow structure.
 759  */
 760 static void
 761 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
 762 {
 763         struct mlx5_flow *dev_flow;
 764
 765         LIST_FOREACH(dev_flow, &flow->dev_flows, next)
 766                 flow_drv_rxq_flags_set(dev, dev_flow);
 767 }
 768
 769 /**
 770  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
 771  * device flow if no other flow uses it with the same kind of request.
 772  *
 773  * @param dev
 774  *   Pointer to Ethernet device.
 775  * @param[in] dev_flow
 776  *   Pointer to the device flow.
 777  */
 778 static void
 779 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
 780 {
 781         struct mlx5_priv *priv = dev->data->dev_private;
 782         struct rte_flow *flow = dev_flow->flow;
 783         const int mark = !!(dev_flow->actions &
 784                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
 785         const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
 786         unsigned int i;
 787
 788         assert(dev->data->dev_started);
 789         for (i = 0; i != flow->rss.queue_num; ++i) {
 790                 int idx = (*flow->rss.queue)[i];
 791                 struct mlx5_rxq_ctrl *rxq_ctrl =
 792                         container_of((*priv->rxqs)[idx],
 793                                      struct mlx5_rxq_ctrl, rxq);
 794
 795                 if (priv->config.dv_flow_en &&
 796                     priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
 797                     mlx5_flow_ext_mreg_supported(dev)) {
 798                         rxq_ctrl->rxq.mark = 1;
 799                         rxq_ctrl->flow_mark_n = 1;
 800                 } else if (mark) {
 801                         rxq_ctrl->flow_mark_n--;
 802                         rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
 803                 }
 804                 if (tunnel) {
 805                         unsigned int j;
 806
 807                         /* Decrease the counter matching the flow. */
 808                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 809                                 if ((tunnels_info[j].tunnel &
 810                                      dev_flow->layers) ==
 811                                     tunnels_info[j].tunnel) {
 812                                         rxq_ctrl->flow_tunnels_n[j]--;
 813                                         break;
 814                                 }
 815                         }
 816                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
 817                 }
 818         }
 819 }
 820
 821 /**
 822  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
 823  * @p flow if no other flow uses it with the same kind of request.
 824  *
 825  * @param dev
 826  *   Pointer to Ethernet device.
 827  * @param[in] flow
 828  *   Pointer to the flow.
 829  */
 830 static void
 831 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
 832 {
 833         struct mlx5_flow *dev_flow;
 834
 835         LIST_FOREACH(dev_flow, &flow->dev_flows, next)
 836                 flow_drv_rxq_flags_trim(dev, dev_flow);
 837 }
 838
 839 /**
 840  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
 841  *
 842  * @param dev
 843  *   Pointer to Ethernet device.
 844  */
 845 static void
 846 flow_rxq_flags_clear(struct rte_eth_dev *dev)
 847 {
 848         struct mlx5_priv *priv = dev->data->dev_private;
 849         unsigned int i;
 850
 851         for (i = 0; i != priv->rxqs_n; ++i) {
 852                 struct mlx5_rxq_ctrl *rxq_ctrl;
 853                 unsigned int j;
 854
 855                 if (!(*priv->rxqs)[i])
 856                         continue;
 857                 rxq_ctrl = container_of((*priv->rxqs)[i],
 858                                         struct mlx5_rxq_ctrl, rxq);
 859                 rxq_ctrl->flow_mark_n = 0;
 860                 rxq_ctrl->rxq.mark = 0;
 861                 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
 862                         rxq_ctrl->flow_tunnels_n[j] = 0;
 863                 rxq_ctrl->rxq.tunnel = 0;
 864         }
 865 }
 866
 867 /*
 868  * return a pointer to the desired action in the list of actions.
 869  *
 870  * @param[in] actions
 871  *   The list of actions to search the action in.
 872  * @param[in] action
 873  *   The action to find.
 874  *
 875  * @return
 876  *   Pointer to the action in the list, if found. NULL otherwise.
 877  */
 878 const struct rte_flow_action *
 879 mlx5_flow_find_action(const struct rte_flow_action *actions,
 880                       enum rte_flow_action_type action)
 881 {
 882         if (actions == NULL)
 883                 return NULL;
 884         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
 885                 if (actions->type == action)
 886                         return actions;
 887         return NULL;
 888 }
 889
 890 /*
 891  * Validate the flag action.
 892  *
 893  * @param[in] action_flags
 894  *   Bit-fields that holds the actions detected until now.
 895  * @param[in] attr
 896  *   Attributes of flow that includes this action.
 897  * @param[out] error
 898  *   Pointer to error structure.
 899  *
 900  * @return
 901  *   0 on success, a negative errno value otherwise and rte_errno is set.
 902  */
 903 int
 904 mlx5_flow_validate_action_flag(uint64_t action_flags,
 905                                const struct rte_flow_attr *attr,
 906                                struct rte_flow_error *error)
 907 {
 908
 909         if (action_flags & MLX5_FLOW_ACTION_DROP)
 910                 return rte_flow_error_set(error, EINVAL,
 911                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 912                                           "can't drop and flag in same flow");
 913         if (action_flags & MLX5_FLOW_ACTION_MARK)
 914                 return rte_flow_error_set(error, EINVAL,
 915                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 916                                           "can't mark and flag in same flow");
 917         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 918                 return rte_flow_error_set(error, EINVAL,
 919                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 920                                           "can't have 2 flag"
 921                                           " actions in same flow");
 922         if (attr->egress)
 923                 return rte_flow_error_set(error, ENOTSUP,
 924                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 925                                           "flag action not supported for "
 926                                           "egress");
 927         return 0;
 928 }
 929
 930 /*
 931  * Validate the mark action.
 932  *
 933  * @param[in] action
 934  *   Pointer to the queue action.
 935  * @param[in] action_flags
 936  *   Bit-fields that holds the actions detected until now.
 937  * @param[in] attr
 938  *   Attributes of flow that includes this action.
 939  * @param[out] error
 940  *   Pointer to error structure.
 941  *
 942  * @return
 943  *   0 on success, a negative errno value otherwise and rte_errno is set.
 944  */
 945 int
 946 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
 947                                uint64_t action_flags,
 948                                const struct rte_flow_attr *attr,
 949                                struct rte_flow_error *error)
 950 {
 951         const struct rte_flow_action_mark *mark = action->conf;
 952
 953         if (!mark)
 954                 return rte_flow_error_set(error, EINVAL,
 955                                           RTE_FLOW_ERROR_TYPE_ACTION,
 956                                           action,
 957                                           "configuration cannot be null");
 958         if (mark->id >= MLX5_FLOW_MARK_MAX)
 959                 return rte_flow_error_set(error, EINVAL,
 960                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 961                                           &mark->id,
 962                                           "mark id must in 0 <= id < "
 963                                           RTE_STR(MLX5_FLOW_MARK_MAX));
 964         if (action_flags & MLX5_FLOW_ACTION_DROP)
 965                 return rte_flow_error_set(error, EINVAL,
 966                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 967                                           "can't drop and mark in same flow");
 968         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 969                 return rte_flow_error_set(error, EINVAL,
 970                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 971                                           "can't flag and mark in same flow");
 972         if (action_flags & MLX5_FLOW_ACTION_MARK)
 973                 return rte_flow_error_set(error, EINVAL,
 974                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 975                                           "can't have 2 mark actions in same"
 976                                           " flow");
 977         if (attr->egress)
 978                 return rte_flow_error_set(error, ENOTSUP,
 979                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 980                                           "mark action not supported for "
 981                                           "egress");
 982         return 0;
 983 }
 984
 985 /*
 986  * Validate the drop action.
 987  *
 988  * @param[in] action_flags
 989  *   Bit-fields that holds the actions detected until now.
 990  * @param[in] attr
 991  *   Attributes of flow that includes this action.
 992  * @param[out] error
 993  *   Pointer to error structure.
 994  *
 995  * @return
 996  *   0 on success, a negative errno value otherwise and rte_errno is set.
 997  */
 998 int
 999 mlx5_flow_validate_action_drop(uint64_t action_flags,
1000                                const struct rte_flow_attr *attr,
1001                                struct rte_flow_error *error)
1002 {
1003         if (action_flags & MLX5_FLOW_ACTION_FLAG)
1004                 return rte_flow_error_set(error, EINVAL,
1005                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1006                                           "can't drop and flag in same flow");
1007         if (action_flags & MLX5_FLOW_ACTION_MARK)
1008                 return rte_flow_error_set(error, EINVAL,
1009                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1010                                           "can't drop and mark in same flow");
1011         if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
1012                             MLX5_FLOW_FATE_ESWITCH_ACTIONS))
1013                 return rte_flow_error_set(error, EINVAL,
1014                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1015                                           "can't have 2 fate actions in"
1016                                           " same flow");
1017         if (attr->egress)
1018                 return rte_flow_error_set(error, ENOTSUP,
1019                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1020                                           "drop action not supported for "
1021                                           "egress");
1022         return 0;
1023 }
1024
1025 /*
1026  * Validate the queue action.
1027  *
1028  * @param[in] action
1029  *   Pointer to the queue action.
1030  * @param[in] action_flags
1031  *   Bit-fields that holds the actions detected until now.
1032  * @param[in] dev
1033  *   Pointer to the Ethernet device structure.
1034  * @param[in] attr
1035  *   Attributes of flow that includes this action.
1036  * @param[out] error
1037  *   Pointer to error structure.
1038  *
1039  * @return
1040  *   0 on success, a negative errno value otherwise and rte_errno is set.
1041  */
1042 int
1043 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1044                                 uint64_t action_flags,
1045                                 struct rte_eth_dev *dev,
1046                                 const struct rte_flow_attr *attr,
1047                                 struct rte_flow_error *error)
1048 {
1049         struct mlx5_priv *priv = dev->data->dev_private;
1050         const struct rte_flow_action_queue *queue = action->conf;
1051
1052         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1053                 return rte_flow_error_set(error, EINVAL,
1054                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1055                                           "can't have 2 fate actions in"
1056                                           " same flow");
1057         if (!priv->rxqs_n)
1058                 return rte_flow_error_set(error, EINVAL,
1059                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1060                                           NULL, "No Rx queues configured");
1061         if (queue->index >= priv->rxqs_n)
1062                 return rte_flow_error_set(error, EINVAL,
1063                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1064                                           &queue->index,
1065                                           "queue index out of range");
1066         if (!(*priv->rxqs)[queue->index])
1067                 return rte_flow_error_set(error, EINVAL,
1068                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1069                                           &queue->index,
1070                                           "queue is not configured");
1071         if (attr->egress)
1072                 return rte_flow_error_set(error, ENOTSUP,
1073                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1074                                           "queue action not supported for "
1075                                           "egress");
1076         return 0;
1077 }
1078
1079 /*
1080  * Validate the rss action.
1081  *
1082  * @param[in] action
1083  *   Pointer to the queue action.
1084  * @param[in] action_flags
1085  *   Bit-fields that holds the actions detected until now.
1086  * @param[in] dev
1087  *   Pointer to the Ethernet device structure.
1088  * @param[in] attr
1089  *   Attributes of flow that includes this action.
1090  * @param[in] item_flags
1091  *   Items that were detected.
1092  * @param[out] error
1093  *   Pointer to error structure.
1094  *
1095  * @return
1096  *   0 on success, a negative errno value otherwise and rte_errno is set.
1097  */
1098 int
1099 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1100                               uint64_t action_flags,
1101                               struct rte_eth_dev *dev,
1102                               const struct rte_flow_attr *attr,
1103                               uint64_t item_flags,
1104                               struct rte_flow_error *error)
1105 {
1106         struct mlx5_priv *priv = dev->data->dev_private;
1107         const struct rte_flow_action_rss *rss = action->conf;
1108         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1109         unsigned int i;
1110
1111         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1112                 return rte_flow_error_set(error, EINVAL,
1113                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1114                                           "can't have 2 fate actions"
1115                                           " in same flow");
1116         if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1117             rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1118                 return rte_flow_error_set(error, ENOTSUP,
1119                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1120                                           &rss->func,
1121                                           "RSS hash function not supported");
1122 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1123         if (rss->level > 2)
1124 #else
1125         if (rss->level > 1)
1126 #endif
1127                 return rte_flow_error_set(error, ENOTSUP,
1128                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1129                                           &rss->level,
1130                                           "tunnel RSS is not supported");
1131         /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1132         if (rss->key_len == 0 && rss->key != NULL)
1133                 return rte_flow_error_set(error, ENOTSUP,
1134                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1135                                           &rss->key_len,
1136                                           "RSS hash key length 0");
1137         if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1138                 return rte_flow_error_set(error, ENOTSUP,
1139                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1140                                           &rss->key_len,
1141                                           "RSS hash key too small");
1142         if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1143                 return rte_flow_error_set(error, ENOTSUP,
1144                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1145                                           &rss->key_len,
1146                                           "RSS hash key too large");
1147         if (rss->queue_num > priv->config.ind_table_max_size)
1148                 return rte_flow_error_set(error, ENOTSUP,
1149                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1150                                           &rss->queue_num,
1151                                           "number of queues too large");
1152         if (rss->types & MLX5_RSS_HF_MASK)
1153                 return rte_flow_error_set(error, ENOTSUP,
1154                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1155                                           &rss->types,
1156                                           "some RSS protocols are not"
1157                                           " supported");
1158         if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1159             !(rss->types & ETH_RSS_IP))
1160                 return rte_flow_error_set(error, EINVAL,
1161                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1162                                           "L3 partial RSS requested but L3 RSS"
1163                                           " type not specified");
1164         if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1165             !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1166                 return rte_flow_error_set(error, EINVAL,
1167                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1168                                           "L4 partial RSS requested but L4 RSS"
1169                                           " type not specified");
1170         if (!priv->rxqs_n)
1171                 return rte_flow_error_set(error, EINVAL,
1172                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1173                                           NULL, "No Rx queues configured");
1174         if (!rss->queue_num)
1175                 return rte_flow_error_set(error, EINVAL,
1176                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1177                                           NULL, "No queues configured");
1178         for (i = 0; i != rss->queue_num; ++i) {
1179                 if (rss->queue[i] >= priv->rxqs_n)
1180                         return rte_flow_error_set
1181                                 (error, EINVAL,
1182                                  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1183                                  &rss->queue[i], "queue index out of range");
1184                 if (!(*priv->rxqs)[rss->queue[i]])
1185                         return rte_flow_error_set
1186                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1187                                  &rss->queue[i], "queue is not configured");
1188         }
1189         if (attr->egress)
1190                 return rte_flow_error_set(error, ENOTSUP,
1191                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1192                                           "rss action not supported for "
1193                                           "egress");
1194         if (rss->level > 1 &&  !tunnel)
1195                 return rte_flow_error_set(error, EINVAL,
1196                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1197                                           "inner RSS is not supported for "
1198                                           "non-tunnel flows");
1199         return 0;
1200 }
1201
1202 /*
1203  * Validate the count action.
1204  *
1205  * @param[in] dev
1206  *   Pointer to the Ethernet device structure.
1207  * @param[in] attr
1208  *   Attributes of flow that includes this action.
1209  * @param[out] error
1210  *   Pointer to error structure.
1211  *
1212  * @return
1213  *   0 on success, a negative errno value otherwise and rte_errno is set.
1214  */
1215 int
1216 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1217                                 const struct rte_flow_attr *attr,
1218                                 struct rte_flow_error *error)
1219 {
1220         if (attr->egress)
1221                 return rte_flow_error_set(error, ENOTSUP,
1222                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1223                                           "count action not supported for "
1224                                           "egress");
1225         return 0;
1226 }
1227
1228 /**
1229  * Verify the @p attributes will be correctly understood by the NIC and store
1230  * them in the @p flow if everything is correct.
1231  *
1232  * @param[in] dev
1233  *   Pointer to the Ethernet device structure.
1234  * @param[in] attributes
1235  *   Pointer to flow attributes
1236  * @param[out] error
1237  *   Pointer to error structure.
1238  *
1239  * @return
1240  *   0 on success, a negative errno value otherwise and rte_errno is set.
1241  */
1242 int
1243 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1244                               const struct rte_flow_attr *attributes,
1245                               struct rte_flow_error *error)
1246 {
1247         struct mlx5_priv *priv = dev->data->dev_private;
1248         uint32_t priority_max = priv->config.flow_prio - 1;
1249
1250         if (attributes->group)
1251                 return rte_flow_error_set(error, ENOTSUP,
1252                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1253                                           NULL, "groups is not supported");
1254         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1255             attributes->priority >= priority_max)
1256                 return rte_flow_error_set(error, ENOTSUP,
1257                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1258                                           NULL, "priority out of range");
1259         if (attributes->egress)
1260                 return rte_flow_error_set(error, ENOTSUP,
1261                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1262                                           "egress is not supported");
1263         if (attributes->transfer && !priv->config.dv_esw_en)
1264                 return rte_flow_error_set(error, ENOTSUP,
1265                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1266                                           NULL, "transfer is not supported");
1267         if (!attributes->ingress)
1268                 return rte_flow_error_set(error, EINVAL,
1269                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1270                                           NULL,
1271                                           "ingress attribute is mandatory");
1272         return 0;
1273 }
1274
1275 /**
1276  * Validate ICMP6 item.
1277  *
1278  * @param[in] item
1279  *   Item specification.
1280  * @param[in] item_flags
1281  *   Bit-fields that holds the items detected until now.
1282  * @param[out] error
1283  *   Pointer to error structure.
1284  *
1285  * @return
1286  *   0 on success, a negative errno value otherwise and rte_errno is set.
1287  */
1288 int
1289 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1290                                uint64_t item_flags,
1291                                uint8_t target_protocol,
1292                                struct rte_flow_error *error)
1293 {
1294         const struct rte_flow_item_icmp6 *mask = item->mask;
1295         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1296         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1297                                       MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1298         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1299                                       MLX5_FLOW_LAYER_OUTER_L4;
1300         int ret;
1301
1302         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1303                 return rte_flow_error_set(error, EINVAL,
1304                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1305                                           "protocol filtering not compatible"
1306                                           " with ICMP6 layer");
1307         if (!(item_flags & l3m))
1308                 return rte_flow_error_set(error, EINVAL,
1309                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1310                                           "IPv6 is mandatory to filter on"
1311                                           " ICMP6");
1312         if (item_flags & l4m)
1313                 return rte_flow_error_set(error, EINVAL,
1314                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1315                                           "multiple L4 layers not supported");
1316         if (!mask)
1317                 mask = &rte_flow_item_icmp6_mask;
1318         ret = mlx5_flow_item_acceptable
1319                 (item, (const uint8_t *)mask,
1320                  (const uint8_t *)&rte_flow_item_icmp6_mask,
1321                  sizeof(struct rte_flow_item_icmp6), error);
1322         if (ret < 0)
1323                 return ret;
1324         return 0;
1325 }
1326
1327 /**
1328  * Validate ICMP item.
1329  *
1330  * @param[in] item
1331  *   Item specification.
1332  * @param[in] item_flags
1333  *   Bit-fields that holds the items detected until now.
1334  * @param[out] error
1335  *   Pointer to error structure.
1336  *
1337  * @return
1338  *   0 on success, a negative errno value otherwise and rte_errno is set.
1339  */
1340 int
1341 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1342                              uint64_t item_flags,
1343                              uint8_t target_protocol,
1344                              struct rte_flow_error *error)
1345 {
1346         const struct rte_flow_item_icmp *mask = item->mask;
1347         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1348         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1349                                       MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1350         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1351                                       MLX5_FLOW_LAYER_OUTER_L4;
1352         int ret;
1353
1354         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1355                 return rte_flow_error_set(error, EINVAL,
1356                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1357                                           "protocol filtering not compatible"
1358                                           " with ICMP layer");
1359         if (!(item_flags & l3m))
1360                 return rte_flow_error_set(error, EINVAL,
1361                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1362                                           "IPv4 is mandatory to filter"
1363                                           " on ICMP");
1364         if (item_flags & l4m)
1365                 return rte_flow_error_set(error, EINVAL,
1366                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1367                                           "multiple L4 layers not supported");
1368         if (!mask)
1369                 mask = &rte_flow_item_icmp_mask;
1370         ret = mlx5_flow_item_acceptable
1371                 (item, (const uint8_t *)mask,
1372                  (const uint8_t *)&rte_flow_item_icmp_mask,
1373                  sizeof(struct rte_flow_item_icmp), error);
1374         if (ret < 0)
1375                 return ret;
1376         return 0;
1377 }
1378
1379 /**
1380  * Validate Ethernet item.
1381  *
1382  * @param[in] item
1383  *   Item specification.
1384  * @param[in] item_flags
1385  *   Bit-fields that holds the items detected until now.
1386  * @param[out] error
1387  *   Pointer to error structure.
1388  *
1389  * @return
1390  *   0 on success, a negative errno value otherwise and rte_errno is set.
1391  */
1392 int
1393 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1394                             uint64_t item_flags,
1395                             struct rte_flow_error *error)
1396 {
1397         const struct rte_flow_item_eth *mask = item->mask;
1398         const struct rte_flow_item_eth nic_mask = {
1399                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1400                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1401                 .type = RTE_BE16(0xffff),
1402         };
1403         int ret;
1404         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1405         const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1406                                        MLX5_FLOW_LAYER_OUTER_L2;
1407
1408         if (item_flags & ethm)
1409                 return rte_flow_error_set(error, ENOTSUP,
1410                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1411                                           "multiple L2 layers not supported");
1412         if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1413             (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1414                 return rte_flow_error_set(error, EINVAL,
1415                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1416                                           "L2 layer should not follow "
1417                                           "L3 layers");
1418         if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1419             (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1420                 return rte_flow_error_set(error, EINVAL,
1421                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1422                                           "L2 layer should not follow VLAN");
1423         if (!mask)
1424                 mask = &rte_flow_item_eth_mask;
1425         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1426                                         (const uint8_t *)&nic_mask,
1427                                         sizeof(struct rte_flow_item_eth),
1428                                         error);
1429         return ret;
1430 }
1431
1432 /**
1433  * Validate VLAN item.
1434  *
1435  * @param[in] item
1436  *   Item specification.
1437  * @param[in] item_flags
1438  *   Bit-fields that holds the items detected until now.
1439  * @param[in] dev
1440  *   Ethernet device flow is being created on.
1441  * @param[out] error
1442  *   Pointer to error structure.
1443  *
1444  * @return
1445  *   0 on success, a negative errno value otherwise and rte_errno is set.
1446  */
1447 int
1448 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1449                              uint64_t item_flags,
1450                              struct rte_eth_dev *dev,
1451                              struct rte_flow_error *error)
1452 {
1453         const struct rte_flow_item_vlan *spec = item->spec;
1454         const struct rte_flow_item_vlan *mask = item->mask;
1455         const struct rte_flow_item_vlan nic_mask = {
1456                 .tci = RTE_BE16(UINT16_MAX),
1457                 .inner_type = RTE_BE16(UINT16_MAX),
1458         };
1459         uint16_t vlan_tag = 0;
1460         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1461         int ret;
1462         const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1463                                         MLX5_FLOW_LAYER_INNER_L4) :
1464                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1465                                         MLX5_FLOW_LAYER_OUTER_L4);
1466         const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1467                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1468
1469         if (item_flags & vlanm)
1470                 return rte_flow_error_set(error, EINVAL,
1471                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1472                                           "multiple VLAN layers not supported");
1473         else if ((item_flags & l34m) != 0)
1474                 return rte_flow_error_set(error, EINVAL,
1475                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1476                                           "VLAN cannot follow L3/L4 layer");
1477         if (!mask)
1478                 mask = &rte_flow_item_vlan_mask;
1479         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1480                                         (const uint8_t *)&nic_mask,
1481                                         sizeof(struct rte_flow_item_vlan),
1482                                         error);
1483         if (ret)
1484                 return ret;
1485         if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1486                 struct mlx5_priv *priv = dev->data->dev_private;
1487
1488                 if (priv->vmwa_context) {
1489                         /*
1490                          * Non-NULL context means we have a virtual machine
1491                          * and SR-IOV enabled, we have to create VLAN interface
1492                          * to make hypervisor to setup E-Switch vport
1493                          * context correctly. We avoid creating the multiple
1494                          * VLAN interfaces, so we cannot support VLAN tag mask.
1495                          */
1496                         return rte_flow_error_set(error, EINVAL,
1497                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1498                                                   item,
1499                                                   "VLAN tag mask is not"
1500                                                   " supported in virtual"
1501                                                   " environment");
1502                 }
1503         }
1504         if (spec) {
1505                 vlan_tag = spec->tci;
1506                 vlan_tag &= mask->tci;
1507         }
1508         /*
1509          * From verbs perspective an empty VLAN is equivalent
1510          * to a packet without VLAN layer.
1511          */
1512         if (!vlan_tag)
1513                 return rte_flow_error_set(error, EINVAL,
1514                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1515                                           item->spec,
1516                                           "VLAN cannot be empty");
1517         return 0;
1518 }
1519
1520 /**
1521  * Validate IPV4 item.
1522  *
1523  * @param[in] item
1524  *   Item specification.
1525  * @param[in] item_flags
1526  *   Bit-fields that holds the items detected until now.
1527  * @param[in] acc_mask
1528  *   Acceptable mask, if NULL default internal default mask
1529  *   will be used to check whether item fields are supported.
1530  * @param[out] error
1531  *   Pointer to error structure.
1532  *
1533  * @return
1534  *   0 on success, a negative errno value otherwise and rte_errno is set.
1535  */
1536 int
1537 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1538                              uint64_t item_flags,
1539                              uint64_t last_item,
1540                              uint16_t ether_type,
1541                              const struct rte_flow_item_ipv4 *acc_mask,
1542                              struct rte_flow_error *error)
1543 {
1544         const struct rte_flow_item_ipv4 *mask = item->mask;
1545         const struct rte_flow_item_ipv4 *spec = item->spec;
1546         const struct rte_flow_item_ipv4 nic_mask = {
1547                 .hdr = {
1548                         .src_addr = RTE_BE32(0xffffffff),
1549                         .dst_addr = RTE_BE32(0xffffffff),
1550                         .type_of_service = 0xff,
1551                         .next_proto_id = 0xff,
1552                 },
1553         };
1554         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1555         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1556                                       MLX5_FLOW_LAYER_OUTER_L3;
1557         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1558                                       MLX5_FLOW_LAYER_OUTER_L4;
1559         int ret;
1560         uint8_t next_proto = 0xFF;
1561         const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1562                                   MLX5_FLOW_LAYER_OUTER_VLAN |
1563                                   MLX5_FLOW_LAYER_INNER_VLAN);
1564
1565         if ((last_item & l2_vlan) && ether_type &&
1566             ether_type != RTE_ETHER_TYPE_IPV4)
1567                 return rte_flow_error_set(error, EINVAL,
1568                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1569                                           "IPv4 cannot follow L2/VLAN layer "
1570                                           "which ether type is not IPv4");
1571         if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1572                 if (mask && spec)
1573                         next_proto = mask->hdr.next_proto_id &
1574                                      spec->hdr.next_proto_id;
1575                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1576                         return rte_flow_error_set(error, EINVAL,
1577                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1578                                                   item,
1579                                                   "multiple tunnel "
1580                                                   "not supported");
1581         }
1582         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1583                 return rte_flow_error_set(error, EINVAL,
1584                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1585                                           "wrong tunnel type - IPv6 specified "
1586                                           "but IPv4 item provided");
1587         if (item_flags & l3m)
1588                 return rte_flow_error_set(error, ENOTSUP,
1589                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1590                                           "multiple L3 layers not supported");
1591         else if (item_flags & l4m)
1592                 return rte_flow_error_set(error, EINVAL,
1593                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1594                                           "L3 cannot follow an L4 layer.");
1595         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1596                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1597                 return rte_flow_error_set(error, EINVAL,
1598                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1599                                           "L3 cannot follow an NVGRE layer.");
1600         if (!mask)
1601                 mask = &rte_flow_item_ipv4_mask;
1602         else if (mask->hdr.next_proto_id != 0 &&
1603                  mask->hdr.next_proto_id != 0xff)
1604                 return rte_flow_error_set(error, EINVAL,
1605                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1606                                           "partial mask is not supported"
1607                                           " for protocol");
1608         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1609                                         acc_mask ? (const uint8_t *)acc_mask
1610                                                  : (const uint8_t *)&nic_mask,
1611                                         sizeof(struct rte_flow_item_ipv4),
1612                                         error);
1613         if (ret < 0)
1614                 return ret;
1615         return 0;
1616 }
1617
1618 /**
1619  * Validate IPV6 item.
1620  *
1621  * @param[in] item
1622  *   Item specification.
1623  * @param[in] item_flags
1624  *   Bit-fields that holds the items detected until now.
1625  * @param[in] acc_mask
1626  *   Acceptable mask, if NULL default internal default mask
1627  *   will be used to check whether item fields are supported.
1628  * @param[out] error
1629  *   Pointer to error structure.
1630  *
1631  * @return
1632  *   0 on success, a negative errno value otherwise and rte_errno is set.
1633  */
1634 int
1635 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1636                              uint64_t item_flags,
1637                              uint64_t last_item,
1638                              uint16_t ether_type,
1639                              const struct rte_flow_item_ipv6 *acc_mask,
1640                              struct rte_flow_error *error)
1641 {
1642         const struct rte_flow_item_ipv6 *mask = item->mask;
1643         const struct rte_flow_item_ipv6 *spec = item->spec;
1644         const struct rte_flow_item_ipv6 nic_mask = {
1645                 .hdr = {
1646                         .src_addr =
1647                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1648                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1649                         .dst_addr =
1650                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1651                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1652                         .vtc_flow = RTE_BE32(0xffffffff),
1653                         .proto = 0xff,
1654                         .hop_limits = 0xff,
1655                 },
1656         };
1657         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1658         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1659                                       MLX5_FLOW_LAYER_OUTER_L3;
1660         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1661                                       MLX5_FLOW_LAYER_OUTER_L4;
1662         int ret;
1663         uint8_t next_proto = 0xFF;
1664         const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1665                                   MLX5_FLOW_LAYER_OUTER_VLAN |
1666                                   MLX5_FLOW_LAYER_INNER_VLAN);
1667
1668         if ((last_item & l2_vlan) && ether_type &&
1669             ether_type != RTE_ETHER_TYPE_IPV6)
1670                 return rte_flow_error_set(error, EINVAL,
1671                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1672                                           "IPv6 cannot follow L2/VLAN layer "
1673                                           "which ether type is not IPv6");
1674         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1675                 if (mask && spec)
1676                         next_proto = mask->hdr.proto & spec->hdr.proto;
1677                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1678                         return rte_flow_error_set(error, EINVAL,
1679                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1680                                                   item,
1681                                                   "multiple tunnel "
1682                                                   "not supported");
1683         }
1684         if (item_flags & MLX5_FLOW_LAYER_IPIP)
1685                 return rte_flow_error_set(error, EINVAL,
1686                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1687                                           "wrong tunnel type - IPv4 specified "
1688                                           "but IPv6 item provided");
1689         if (item_flags & l3m)
1690                 return rte_flow_error_set(error, ENOTSUP,
1691                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1692                                           "multiple L3 layers not supported");
1693         else if (item_flags & l4m)
1694                 return rte_flow_error_set(error, EINVAL,
1695                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1696                                           "L3 cannot follow an L4 layer.");
1697         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1698                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1699                 return rte_flow_error_set(error, EINVAL,
1700                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1701                                           "L3 cannot follow an NVGRE layer.");
1702         if (!mask)
1703                 mask = &rte_flow_item_ipv6_mask;
1704         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1705                                         acc_mask ? (const uint8_t *)acc_mask
1706                                                  : (const uint8_t *)&nic_mask,
1707                                         sizeof(struct rte_flow_item_ipv6),
1708                                         error);
1709         if (ret < 0)
1710                 return ret;
1711         return 0;
1712 }
1713
1714 /**
1715  * Validate UDP item.
1716  *
1717  * @param[in] item
1718  *   Item specification.
1719  * @param[in] item_flags
1720  *   Bit-fields that holds the items detected until now.
1721  * @param[in] target_protocol
1722  *   The next protocol in the previous item.
1723  * @param[in] flow_mask
1724  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1725  * @param[out] error
1726  *   Pointer to error structure.
1727  *
1728  * @return
1729  *   0 on success, a negative errno value otherwise and rte_errno is set.
1730  */
1731 int
1732 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1733                             uint64_t item_flags,
1734                             uint8_t target_protocol,
1735                             struct rte_flow_error *error)
1736 {
1737         const struct rte_flow_item_udp *mask = item->mask;
1738         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1739         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1740                                       MLX5_FLOW_LAYER_OUTER_L3;
1741         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1742                                       MLX5_FLOW_LAYER_OUTER_L4;
1743         int ret;
1744
1745         if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1746                 return rte_flow_error_set(error, EINVAL,
1747                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1748                                           "protocol filtering not compatible"
1749                                           " with UDP layer");
1750         if (!(item_flags & l3m))
1751                 return rte_flow_error_set(error, EINVAL,
1752                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1753                                           "L3 is mandatory to filter on L4");
1754         if (item_flags & l4m)
1755                 return rte_flow_error_set(error, EINVAL,
1756                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1757                                           "multiple L4 layers not supported");
1758         if (!mask)
1759                 mask = &rte_flow_item_udp_mask;
1760         ret = mlx5_flow_item_acceptable
1761                 (item, (const uint8_t *)mask,
1762                  (const uint8_t *)&rte_flow_item_udp_mask,
1763                  sizeof(struct rte_flow_item_udp), error);
1764         if (ret < 0)
1765                 return ret;
1766         return 0;
1767 }
1768
1769 /**
1770  * Validate TCP item.
1771  *
1772  * @param[in] item
1773  *   Item specification.
1774  * @param[in] item_flags
1775  *   Bit-fields that holds the items detected until now.
1776  * @param[in] target_protocol
1777  *   The next protocol in the previous item.
1778  * @param[out] error
1779  *   Pointer to error structure.
1780  *
1781  * @return
1782  *   0 on success, a negative errno value otherwise and rte_errno is set.
1783  */
1784 int
1785 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1786                             uint64_t item_flags,
1787                             uint8_t target_protocol,
1788                             const struct rte_flow_item_tcp *flow_mask,
1789                             struct rte_flow_error *error)
1790 {
1791         const struct rte_flow_item_tcp *mask = item->mask;
1792         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1793         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1794                                       MLX5_FLOW_LAYER_OUTER_L3;
1795         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1796                                       MLX5_FLOW_LAYER_OUTER_L4;
1797         int ret;
1798
1799         assert(flow_mask);
1800         if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1801                 return rte_flow_error_set(error, EINVAL,
1802                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1803                                           "protocol filtering not compatible"
1804                                           " with TCP layer");
1805         if (!(item_flags & l3m))
1806                 return rte_flow_error_set(error, EINVAL,
1807                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1808                                           "L3 is mandatory to filter on L4");
1809         if (item_flags & l4m)
1810                 return rte_flow_error_set(error, EINVAL,
1811                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1812                                           "multiple L4 layers not supported");
1813         if (!mask)
1814                 mask = &rte_flow_item_tcp_mask;
1815         ret = mlx5_flow_item_acceptable
1816                 (item, (const uint8_t *)mask,
1817                  (const uint8_t *)flow_mask,
1818                  sizeof(struct rte_flow_item_tcp), error);
1819         if (ret < 0)
1820                 return ret;
1821         return 0;
1822 }
1823
1824 /**
1825  * Validate VXLAN item.
1826  *
1827  * @param[in] item
1828  *   Item specification.
1829  * @param[in] item_flags
1830  *   Bit-fields that holds the items detected until now.
1831  * @param[in] target_protocol
1832  *   The next protocol in the previous item.
1833  * @param[out] error
1834  *   Pointer to error structure.
1835  *
1836  * @return
1837  *   0 on success, a negative errno value otherwise and rte_errno is set.
1838  */
1839 int
1840 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1841                               uint64_t item_flags,
1842                               struct rte_flow_error *error)
1843 {
1844         const struct rte_flow_item_vxlan *spec = item->spec;
1845         const struct rte_flow_item_vxlan *mask = item->mask;
1846         int ret;
1847         union vni {
1848                 uint32_t vlan_id;
1849                 uint8_t vni[4];
1850         } id = { .vlan_id = 0, };
1851         uint32_t vlan_id = 0;
1852
1853
1854         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1855                 return rte_flow_error_set(error, ENOTSUP,
1856                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1857                                           "multiple tunnel layers not"
1858                                           " supported");
1859         /*
1860          * Verify only UDPv4 is present as defined in
1861          * https://tools.ietf.org/html/rfc7348
1862          */
1863         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1864                 return rte_flow_error_set(error, EINVAL,
1865                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1866                                           "no outer UDP layer found");
1867         if (!mask)
1868                 mask = &rte_flow_item_vxlan_mask;
1869         ret = mlx5_flow_item_acceptable
1870                 (item, (const uint8_t *)mask,
1871                  (const uint8_t *)&rte_flow_item_vxlan_mask,
1872                  sizeof(struct rte_flow_item_vxlan),
1873                  error);
1874         if (ret < 0)
1875                 return ret;
1876         if (spec) {
1877                 memcpy(&id.vni[1], spec->vni, 3);
1878                 vlan_id = id.vlan_id;
1879                 memcpy(&id.vni[1], mask->vni, 3);
1880                 vlan_id &= id.vlan_id;
1881         }
1882         /*
1883          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1884          * only this layer is defined in the Verbs specification it is
1885          * interpreted as wildcard and all packets will match this
1886          * rule, if it follows a full stack layer (ex: eth / ipv4 /
1887          * udp), all packets matching the layers before will also
1888          * match this rule.  To avoid such situation, VNI 0 is
1889          * currently refused.
1890          */
1891         if (!vlan_id)
1892                 return rte_flow_error_set(error, ENOTSUP,
1893                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1894                                           "VXLAN vni cannot be 0");
1895         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1896                 return rte_flow_error_set(error, ENOTSUP,
1897                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1898                                           "VXLAN tunnel must be fully defined");
1899         return 0;
1900 }
1901
1902 /**
1903  * Validate VXLAN_GPE item.
1904  *
1905  * @param[in] item
1906  *   Item specification.
1907  * @param[in] item_flags
1908  *   Bit-fields that holds the items detected until now.
1909  * @param[in] priv
1910  *   Pointer to the private data structure.
1911  * @param[in] target_protocol
1912  *   The next protocol in the previous item.
1913  * @param[out] error
1914  *   Pointer to error structure.
1915  *
1916  * @return
1917  *   0 on success, a negative errno value otherwise and rte_errno is set.
1918  */
1919 int
1920 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1921                                   uint64_t item_flags,
1922                                   struct rte_eth_dev *dev,
1923                                   struct rte_flow_error *error)
1924 {
1925         struct mlx5_priv *priv = dev->data->dev_private;
1926         const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1927         const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1928         int ret;
1929         union vni {
1930                 uint32_t vlan_id;
1931                 uint8_t vni[4];
1932         } id = { .vlan_id = 0, };
1933         uint32_t vlan_id = 0;
1934
1935         if (!priv->config.l3_vxlan_en)
1936                 return rte_flow_error_set(error, ENOTSUP,
1937                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1938                                           "L3 VXLAN is not enabled by device"
1939                                           " parameter and/or not configured in"
1940                                           " firmware");
1941         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1942                 return rte_flow_error_set(error, ENOTSUP,
1943                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1944                                           "multiple tunnel layers not"
1945                                           " supported");
1946         /*
1947          * Verify only UDPv4 is present as defined in
1948          * https://tools.ietf.org/html/rfc7348
1949          */
1950         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1951                 return rte_flow_error_set(error, EINVAL,
1952                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1953                                           "no outer UDP layer found");
1954         if (!mask)
1955                 mask = &rte_flow_item_vxlan_gpe_mask;
1956         ret = mlx5_flow_item_acceptable
1957                 (item, (const uint8_t *)mask,
1958                  (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1959                  sizeof(struct rte_flow_item_vxlan_gpe),
1960                  error);
1961         if (ret < 0)
1962                 return ret;
1963         if (spec) {
1964                 if (spec->protocol)
1965                         return rte_flow_error_set(error, ENOTSUP,
1966                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1967                                                   item,
1968                                                   "VxLAN-GPE protocol"
1969                                                   " not supported");
1970                 memcpy(&id.vni[1], spec->vni, 3);
1971                 vlan_id = id.vlan_id;
1972                 memcpy(&id.vni[1], mask->vni, 3);
1973                 vlan_id &= id.vlan_id;
1974         }
1975         /*
1976          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1977          * layer is defined in the Verbs specification it is interpreted as
1978          * wildcard and all packets will match this rule, if it follows a full
1979          * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1980          * before will also match this rule.  To avoid such situation, VNI 0
1981          * is currently refused.
1982          */
1983         if (!vlan_id)
1984                 return rte_flow_error_set(error, ENOTSUP,
1985                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1986                                           "VXLAN-GPE vni cannot be 0");
1987         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1988                 return rte_flow_error_set(error, ENOTSUP,
1989                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1990                                           "VXLAN-GPE tunnel must be fully"
1991                                           " defined");
1992         return 0;
1993 }
1994 /**
1995  * Validate GRE Key item.
1996  *
1997  * @param[in] item
1998  *   Item specification.
1999  * @param[in] item_flags
2000  *   Bit flags to mark detected items.
2001  * @param[in] gre_item
2002  *   Pointer to gre_item
2003  * @param[out] error
2004  *   Pointer to error structure.
2005  *
2006  * @return
2007  *   0 on success, a negative errno value otherwise and rte_errno is set.
2008  */
2009 int
2010 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2011                                 uint64_t item_flags,
2012                                 const struct rte_flow_item *gre_item,
2013                                 struct rte_flow_error *error)
2014 {
2015         const rte_be32_t *mask = item->mask;
2016         int ret = 0;
2017         rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2018         const struct rte_flow_item_gre *gre_spec;
2019         const struct rte_flow_item_gre *gre_mask;
2020
2021         if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2022                 return rte_flow_error_set(error, ENOTSUP,
2023                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2024                                           "Multiple GRE key not support");
2025         if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2026                 return rte_flow_error_set(error, ENOTSUP,
2027                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2028                                           "No preceding GRE header");
2029         if (item_flags & MLX5_FLOW_LAYER_INNER)
2030                 return rte_flow_error_set(error, ENOTSUP,
2031                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2032                                           "GRE key following a wrong item");
2033         gre_mask = gre_item->mask;
2034         if (!gre_mask)
2035                 gre_mask = &rte_flow_item_gre_mask;
2036         gre_spec = gre_item->spec;
2037         if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2038                          !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2039                 return rte_flow_error_set(error, EINVAL,
2040                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2041                                           "Key bit must be on");
2042
2043         if (!mask)
2044                 mask = &gre_key_default_mask;
2045         ret = mlx5_flow_item_acceptable
2046                 (item, (const uint8_t *)mask,
2047                  (const uint8_t *)&gre_key_default_mask,
2048                  sizeof(rte_be32_t), error);
2049         return ret;
2050 }
2051
2052 /**
2053  * Validate GRE item.
2054  *
2055  * @param[in] item
2056  *   Item specification.
2057  * @param[in] item_flags
2058  *   Bit flags to mark detected items.
2059  * @param[in] target_protocol
2060  *   The next protocol in the previous item.
2061  * @param[out] error
2062  *   Pointer to error structure.
2063  *
2064  * @return
2065  *   0 on success, a negative errno value otherwise and rte_errno is set.
2066  */
2067 int
2068 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2069                             uint64_t item_flags,
2070                             uint8_t target_protocol,
2071                             struct rte_flow_error *error)
2072 {
2073         const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2074         const struct rte_flow_item_gre *mask = item->mask;
2075         int ret;
2076         const struct rte_flow_item_gre nic_mask = {
2077                 .c_rsvd0_ver = RTE_BE16(0xB000),
2078                 .protocol = RTE_BE16(UINT16_MAX),
2079         };
2080
2081         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2082                 return rte_flow_error_set(error, EINVAL,
2083                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2084                                           "protocol filtering not compatible"
2085                                           " with this GRE layer");
2086         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2087                 return rte_flow_error_set(error, ENOTSUP,
2088                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2089                                           "multiple tunnel layers not"
2090                                           " supported");
2091         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2092                 return rte_flow_error_set(error, ENOTSUP,
2093                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2094                                           "L3 Layer is missing");
2095         if (!mask)
2096                 mask = &rte_flow_item_gre_mask;
2097         ret = mlx5_flow_item_acceptable
2098                 (item, (const uint8_t *)mask,
2099                  (const uint8_t *)&nic_mask,
2100                  sizeof(struct rte_flow_item_gre), error);
2101         if (ret < 0)
2102                 return ret;
2103 #ifndef HAVE_MLX5DV_DR
2104 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2105         if (spec && (spec->protocol & mask->protocol))
2106                 return rte_flow_error_set(error, ENOTSUP,
2107                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2108                                           "without MPLS support the"
2109                                           " specification cannot be used for"
2110                                           " filtering");
2111 #endif
2112 #endif
2113         return 0;
2114 }
2115
2116 /**
2117  * Validate Geneve item.
2118  *
2119  * @param[in] item
2120  *   Item specification.
2121  * @param[in] itemFlags
2122  *   Bit-fields that holds the items detected until now.
2123  * @param[in] enPriv
2124  *   Pointer to the private data structure.
2125  * @param[out] error
2126  *   Pointer to error structure.
2127  *
2128  * @return
2129  *   0 on success, a negative errno value otherwise and rte_errno is set.
2130  */
2131
2132 int
2133 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2134                                uint64_t item_flags,
2135                                struct rte_eth_dev *dev,
2136                                struct rte_flow_error *error)
2137 {
2138         struct mlx5_priv *priv = dev->data->dev_private;
2139         const struct rte_flow_item_geneve *spec = item->spec;
2140         const struct rte_flow_item_geneve *mask = item->mask;
2141         int ret;
2142         uint16_t gbhdr;
2143         uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2144                           MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2145         const struct rte_flow_item_geneve nic_mask = {
2146                 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2147                 .vni = "\xff\xff\xff",
2148                 .protocol = RTE_BE16(UINT16_MAX),
2149         };
2150
2151         if (!(priv->config.hca_attr.flex_parser_protocols &
2152               MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2153             !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2154                 return rte_flow_error_set(error, ENOTSUP,
2155                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2156                                           "L3 Geneve is not enabled by device"
2157                                           " parameter and/or not configured in"
2158                                           " firmware");
2159         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2160                 return rte_flow_error_set(error, ENOTSUP,
2161                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2162                                           "multiple tunnel layers not"
2163                                           " supported");
2164         /*
2165          * Verify only UDPv4 is present as defined in
2166          * https://tools.ietf.org/html/rfc7348
2167          */
2168         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2169                 return rte_flow_error_set(error, EINVAL,
2170                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2171                                           "no outer UDP layer found");
2172         if (!mask)
2173                 mask = &rte_flow_item_geneve_mask;
2174         ret = mlx5_flow_item_acceptable
2175                                   (item, (const uint8_t *)mask,
2176                                    (const uint8_t *)&nic_mask,
2177                                    sizeof(struct rte_flow_item_geneve), error);
2178         if (ret)
2179                 return ret;
2180         if (spec) {
2181                 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2182                 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2183                      MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2184                      MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2185                         return rte_flow_error_set(error, ENOTSUP,
2186                                                   RTE_FLOW_ERROR_TYPE_ITEM,
2187                                                   item,
2188                                                   "Geneve protocol unsupported"
2189                                                   " fields are being used");
2190                 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2191                         return rte_flow_error_set
2192                                         (error, ENOTSUP,
2193                                          RTE_FLOW_ERROR_TYPE_ITEM,
2194                                          item,
2195                                          "Unsupported Geneve options length");
2196         }
2197         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2198                 return rte_flow_error_set
2199                                     (error, ENOTSUP,
2200                                      RTE_FLOW_ERROR_TYPE_ITEM, item,
2201                                      "Geneve tunnel must be fully defined");
2202         return 0;
2203 }
2204
2205 /**
2206  * Validate MPLS item.
2207  *
2208  * @param[in] dev
2209  *   Pointer to the rte_eth_dev structure.
2210  * @param[in] item
2211  *   Item specification.
2212  * @param[in] item_flags
2213  *   Bit-fields that holds the items detected until now.
2214  * @param[in] prev_layer
2215  *   The protocol layer indicated in previous item.
2216  * @param[out] error
2217  *   Pointer to error structure.
2218  *
2219  * @return
2220  *   0 on success, a negative errno value otherwise and rte_errno is set.
2221  */
2222 int
2223 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2224                              const struct rte_flow_item *item __rte_unused,
2225                              uint64_t item_flags __rte_unused,
2226                              uint64_t prev_layer __rte_unused,
2227                              struct rte_flow_error *error)
2228 {
2229 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2230         const struct rte_flow_item_mpls *mask = item->mask;
2231         struct mlx5_priv *priv = dev->data->dev_private;
2232         int ret;
2233
2234         if (!priv->config.mpls_en)
2235                 return rte_flow_error_set(error, ENOTSUP,
2236                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2237                                           "MPLS not supported or"
2238                                           " disabled in firmware"
2239                                           " configuration.");
2240         /* MPLS over IP, UDP, GRE is allowed */
2241         if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2242                             MLX5_FLOW_LAYER_OUTER_L4_UDP |
2243                             MLX5_FLOW_LAYER_GRE)))
2244                 return rte_flow_error_set(error, EINVAL,
2245                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2246                                           "protocol filtering not compatible"
2247                                           " with MPLS layer");
2248         /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2249         if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2250             !(item_flags & MLX5_FLOW_LAYER_GRE))
2251                 return rte_flow_error_set(error, ENOTSUP,
2252                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2253                                           "multiple tunnel layers not"
2254                                           " supported");
2255         if (!mask)
2256                 mask = &rte_flow_item_mpls_mask;
2257         ret = mlx5_flow_item_acceptable
2258                 (item, (const uint8_t *)mask,
2259                  (const uint8_t *)&rte_flow_item_mpls_mask,
2260                  sizeof(struct rte_flow_item_mpls), error);
2261         if (ret < 0)
2262                 return ret;
2263         return 0;
2264 #endif
2265         return rte_flow_error_set(error, ENOTSUP,
2266                                   RTE_FLOW_ERROR_TYPE_ITEM, item,
2267                                   "MPLS is not supported by Verbs, please"
2268                                   " update.");
2269 }
2270
2271 /**
2272  * Validate NVGRE item.
2273  *
2274  * @param[in] item
2275  *   Item specification.
2276  * @param[in] item_flags
2277  *   Bit flags to mark detected items.
2278  * @param[in] target_protocol
2279  *   The next protocol in the previous item.
2280  * @param[out] error
2281  *   Pointer to error structure.
2282  *
2283  * @return
2284  *   0 on success, a negative errno value otherwise and rte_errno is set.
2285  */
2286 int
2287 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2288                               uint64_t item_flags,
2289                               uint8_t target_protocol,
2290                               struct rte_flow_error *error)
2291 {
2292         const struct rte_flow_item_nvgre *mask = item->mask;
2293         int ret;
2294
2295         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2296                 return rte_flow_error_set(error, EINVAL,
2297                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2298                                           "protocol filtering not compatible"
2299                                           " with this GRE layer");
2300         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2301                 return rte_flow_error_set(error, ENOTSUP,
2302                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2303                                           "multiple tunnel layers not"
2304                                           " supported");
2305         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2306                 return rte_flow_error_set(error, ENOTSUP,
2307                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2308                                           "L3 Layer is missing");
2309         if (!mask)
2310                 mask = &rte_flow_item_nvgre_mask;
2311         ret = mlx5_flow_item_acceptable
2312                 (item, (const uint8_t *)mask,
2313                  (const uint8_t *)&rte_flow_item_nvgre_mask,
2314                  sizeof(struct rte_flow_item_nvgre), error);
2315         if (ret < 0)
2316                 return ret;
2317         return 0;
2318 }
2319
2320 /* Allocate unique ID for the split Q/RSS subflows. */
2321 static uint32_t
2322 flow_qrss_get_id(struct rte_eth_dev *dev)
2323 {
2324         struct mlx5_priv *priv = dev->data->dev_private;
2325         uint32_t qrss_id, ret;
2326
2327         ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2328         if (ret)
2329                 return 0;
2330         assert(qrss_id);
2331         return qrss_id;
2332 }
2333
2334 /* Free unique ID for the split Q/RSS subflows. */
2335 static void
2336 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2337 {
2338         struct mlx5_priv *priv = dev->data->dev_private;
2339
2340         if (qrss_id)
2341                 mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2342 }
2343
2344 /**
2345  * Release resource related QUEUE/RSS action split.
2346  *
2347  * @param dev
2348  *   Pointer to Ethernet device.
2349  * @param flow
2350  *   Flow to release id's from.
2351  */
2352 static void
2353 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2354                              struct rte_flow *flow)
2355 {
2356         struct mlx5_flow *dev_flow;
2357
2358         LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2359                 if (dev_flow->qrss_id)
2360                         flow_qrss_free_id(dev, dev_flow->qrss_id);
2361 }
2362
2363 static int
2364 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2365                    const struct rte_flow_attr *attr __rte_unused,
2366                    const struct rte_flow_item items[] __rte_unused,
2367                    const struct rte_flow_action actions[] __rte_unused,
2368                    bool external __rte_unused,
2369                    struct rte_flow_error *error)
2370 {
2371         return rte_flow_error_set(error, ENOTSUP,
2372                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2373 }
2374
2375 static struct mlx5_flow *
2376 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2377                   const struct rte_flow_item items[] __rte_unused,
2378                   const struct rte_flow_action actions[] __rte_unused,
2379                   struct rte_flow_error *error)
2380 {
2381         rte_flow_error_set(error, ENOTSUP,
2382                            RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2383         return NULL;
2384 }
2385
2386 static int
2387 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2388                     struct mlx5_flow *dev_flow __rte_unused,
2389                     const struct rte_flow_attr *attr __rte_unused,
2390                     const struct rte_flow_item items[] __rte_unused,
2391                     const struct rte_flow_action actions[] __rte_unused,
2392                     struct rte_flow_error *error)
2393 {
2394         return rte_flow_error_set(error, ENOTSUP,
2395                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2396 }
2397
2398 static int
2399 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2400                 struct rte_flow *flow __rte_unused,
2401                 struct rte_flow_error *error)
2402 {
2403         return rte_flow_error_set(error, ENOTSUP,
2404                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2405 }
2406
2407 static void
2408 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2409                  struct rte_flow *flow __rte_unused)
2410 {
2411 }
2412
2413 static void
2414 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2415                   struct rte_flow *flow __rte_unused)
2416 {
2417 }
2418
2419 static int
2420 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2421                 struct rte_flow *flow __rte_unused,
2422                 const struct rte_flow_action *actions __rte_unused,
2423                 void *data __rte_unused,
2424                 struct rte_flow_error *error)
2425 {
2426         return rte_flow_error_set(error, ENOTSUP,
2427                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2428 }
2429
2430 /* Void driver to protect from null pointer reference. */
2431 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2432         .validate = flow_null_validate,
2433         .prepare = flow_null_prepare,
2434         .translate = flow_null_translate,
2435         .apply = flow_null_apply,
2436         .remove = flow_null_remove,
2437         .destroy = flow_null_destroy,
2438         .query = flow_null_query,
2439 };
2440
2441 /**
2442  * Select flow driver type according to flow attributes and device
2443  * configuration.
2444  *
2445  * @param[in] dev
2446  *   Pointer to the dev structure.
2447  * @param[in] attr
2448  *   Pointer to the flow attributes.
2449  *
2450  * @return
2451  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2452  */
2453 static enum mlx5_flow_drv_type
2454 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2455 {
2456         struct mlx5_priv *priv = dev->data->dev_private;
2457         enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2458
2459         if (attr->transfer && priv->config.dv_esw_en)
2460                 type = MLX5_FLOW_TYPE_DV;
2461         if (!attr->transfer)
2462                 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2463                                                  MLX5_FLOW_TYPE_VERBS;
2464         return type;
2465 }
2466
2467 #define flow_get_drv_ops(type) flow_drv_ops[type]
2468
2469 /**
2470  * Flow driver validation API. This abstracts calling driver specific functions.
2471  * The type of flow driver is determined according to flow attributes.
2472  *
2473  * @param[in] dev
2474  *   Pointer to the dev structure.
2475  * @param[in] attr
2476  *   Pointer to the flow attributes.
2477  * @param[in] items
2478  *   Pointer to the list of items.
2479  * @param[in] actions
2480  *   Pointer to the list of actions.
2481  * @param[in] external
2482  *   This flow rule is created by request external to PMD.
2483  * @param[out] error
2484  *   Pointer to the error structure.
2485  *
2486  * @return
2487  *   0 on success, a negative errno value otherwise and rte_errno is set.
2488  */
2489 static inline int
2490 flow_drv_validate(struct rte_eth_dev *dev,
2491                   const struct rte_flow_attr *attr,
2492                   const struct rte_flow_item items[],
2493                   const struct rte_flow_action actions[],
2494                   bool external, struct rte_flow_error *error)
2495 {
2496         const struct mlx5_flow_driver_ops *fops;
2497         enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2498
2499         fops = flow_get_drv_ops(type);
2500         return fops->validate(dev, attr, items, actions, external, error);
2501 }
2502
2503 /**
2504  * Flow driver preparation API. This abstracts calling driver specific
2505  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2506  * calculates the size of memory required for device flow, allocates the memory,
2507  * initializes the device flow and returns the pointer.
2508  *
2509  * @note
2510  *   This function initializes device flow structure such as dv or verbs in
2511  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2512  *   rest. For example, adding returning device flow to flow->dev_flow list and
2513  *   setting backward reference to the flow should be done out of this function.
2514  *   layers field is not filled either.
2515  *
2516  * @param[in] attr
2517  *   Pointer to the flow attributes.
2518  * @param[in] items
2519  *   Pointer to the list of items.
2520  * @param[in] actions
2521  *   Pointer to the list of actions.
2522  * @param[out] error
2523  *   Pointer to the error structure.
2524  *
2525  * @return
2526  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2527  */
2528 static inline struct mlx5_flow *
2529 flow_drv_prepare(const struct rte_flow *flow,
2530                  const struct rte_flow_attr *attr,
2531                  const struct rte_flow_item items[],
2532                  const struct rte_flow_action actions[],
2533                  struct rte_flow_error *error)
2534 {
2535         const struct mlx5_flow_driver_ops *fops;
2536         enum mlx5_flow_drv_type type = flow->drv_type;
2537
2538         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2539         fops = flow_get_drv_ops(type);
2540         return fops->prepare(attr, items, actions, error);
2541 }
2542
2543 /**
2544  * Flow driver translation API. This abstracts calling driver specific
2545  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2546  * translates a generic flow into a driver flow. flow_drv_prepare() must
2547  * precede.
2548  *
2549  * @note
2550  *   dev_flow->layers could be filled as a result of parsing during translation
2551  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2552  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2553  *   flow->actions could be overwritten even though all the expanded dev_flows
2554  *   have the same actions.
2555  *
2556  * @param[in] dev
2557  *   Pointer to the rte dev structure.
2558  * @param[in, out] dev_flow
2559  *   Pointer to the mlx5 flow.
2560  * @param[in] attr
2561  *   Pointer to the flow attributes.
2562  * @param[in] items
2563  *   Pointer to the list of items.
2564  * @param[in] actions
2565  *   Pointer to the list of actions.
2566  * @param[out] error
2567  *   Pointer to the error structure.
2568  *
2569  * @return
2570  *   0 on success, a negative errno value otherwise and rte_errno is set.
2571  */
2572 static inline int
2573 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2574                    const struct rte_flow_attr *attr,
2575                    const struct rte_flow_item items[],
2576                    const struct rte_flow_action actions[],
2577                    struct rte_flow_error *error)
2578 {
2579         const struct mlx5_flow_driver_ops *fops;
2580         enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2581
2582         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2583         fops = flow_get_drv_ops(type);
2584         return fops->translate(dev, dev_flow, attr, items, actions, error);
2585 }
2586
2587 /**
2588  * Flow driver apply API. This abstracts calling driver specific functions.
2589  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2590  * translated driver flows on to device. flow_drv_translate() must precede.
2591  *
2592  * @param[in] dev
2593  *   Pointer to Ethernet device structure.
2594  * @param[in, out] flow
2595  *   Pointer to flow structure.
2596  * @param[out] error
2597  *   Pointer to error structure.
2598  *
2599  * @return
2600  *   0 on success, a negative errno value otherwise and rte_errno is set.
2601  */
2602 static inline int
2603 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2604                struct rte_flow_error *error)
2605 {
2606         const struct mlx5_flow_driver_ops *fops;
2607         enum mlx5_flow_drv_type type = flow->drv_type;
2608
2609         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2610         fops = flow_get_drv_ops(type);
2611         return fops->apply(dev, flow, error);
2612 }
2613
2614 /**
2615  * Flow driver remove API. This abstracts calling driver specific functions.
2616  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2617  * on device. All the resources of the flow should be freed by calling
2618  * flow_drv_destroy().
2619  *
2620  * @param[in] dev
2621  *   Pointer to Ethernet device.
2622  * @param[in, out] flow
2623  *   Pointer to flow structure.
2624  */
2625 static inline void
2626 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2627 {
2628         const struct mlx5_flow_driver_ops *fops;
2629         enum mlx5_flow_drv_type type = flow->drv_type;
2630
2631         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2632         fops = flow_get_drv_ops(type);
2633         fops->remove(dev, flow);
2634 }
2635
2636 /**
2637  * Flow driver destroy API. This abstracts calling driver specific functions.
2638  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2639  * on device and releases resources of the flow.
2640  *
2641  * @param[in] dev
2642  *   Pointer to Ethernet device.
2643  * @param[in, out] flow
2644  *   Pointer to flow structure.
2645  */
2646 static inline void
2647 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2648 {
2649         const struct mlx5_flow_driver_ops *fops;
2650         enum mlx5_flow_drv_type type = flow->drv_type;
2651
2652         flow_mreg_split_qrss_release(dev, flow);
2653         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2654         fops = flow_get_drv_ops(type);
2655         fops->destroy(dev, flow);
2656 }
2657
2658 /**
2659  * Validate a flow supported by the NIC.
2660  *
2661  * @see rte_flow_validate()
2662  * @see rte_flow_ops
2663  */
2664 int
2665 mlx5_flow_validate(struct rte_eth_dev *dev,
2666                    const struct rte_flow_attr *attr,
2667                    const struct rte_flow_item items[],
2668                    const struct rte_flow_action actions[],
2669                    struct rte_flow_error *error)
2670 {
2671         int ret;
2672
2673         ret = flow_drv_validate(dev, attr, items, actions, true, error);
2674         if (ret < 0)
2675                 return ret;
2676         return 0;
2677 }
2678
2679 /**
2680  * Get port id item from the item list.
2681  *
2682  * @param[in] item
2683  *   Pointer to the list of items.
2684  *
2685  * @return
2686  *   Pointer to the port id item if exist, else return NULL.
2687  */
2688 static const struct rte_flow_item *
2689 find_port_id_item(const struct rte_flow_item *item)
2690 {
2691         assert(item);
2692         for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2693                 if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2694                         return item;
2695         }
2696         return NULL;
2697 }
2698
2699 /**
2700  * Get RSS action from the action list.
2701  *
2702  * @param[in] actions
2703  *   Pointer to the list of actions.
2704  *
2705  * @return
2706  *   Pointer to the RSS action if exist, else return NULL.
2707  */
2708 static const struct rte_flow_action_rss*
2709 flow_get_rss_action(const struct rte_flow_action actions[])
2710 {
2711         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2712                 switch (actions->type) {
2713                 case RTE_FLOW_ACTION_TYPE_RSS:
2714                         return (const struct rte_flow_action_rss *)
2715                                actions->conf;
2716                 default:
2717                         break;
2718                 }
2719         }
2720         return NULL;
2721 }
2722
2723 static unsigned int
2724 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2725 {
2726         const struct rte_flow_item *item;
2727         unsigned int has_vlan = 0;
2728
2729         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2730                 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2731                         has_vlan = 1;
2732                         break;
2733                 }
2734         }
2735         if (has_vlan)
2736                 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2737                                        MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2738         return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2739                                MLX5_EXPANSION_ROOT_OUTER;
2740 }
2741
2742 /**
2743  * Get QUEUE/RSS action from the action list.
2744  *
2745  * @param[in] actions
2746  *   Pointer to the list of actions.
2747  * @param[out] qrss
2748  *   Pointer to the return pointer.
2749  * @param[out] qrss_type
2750  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2751  *   if no QUEUE/RSS is found.
2752  *
2753  * @return
2754  *   Total number of actions.
2755  */
2756 static int
2757 flow_parse_qrss_action(const struct rte_flow_action actions[],
2758                        const struct rte_flow_action **qrss)
2759 {
2760         int actions_n = 0;
2761
2762         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2763                 switch (actions->type) {
2764                 case RTE_FLOW_ACTION_TYPE_QUEUE:
2765                 case RTE_FLOW_ACTION_TYPE_RSS:
2766                         *qrss = actions;
2767                         break;
2768                 default:
2769                         break;
2770                 }
2771                 actions_n++;
2772         }
2773         /* Count RTE_FLOW_ACTION_TYPE_END. */
2774         return actions_n + 1;
2775 }
2776
2777 /**
2778  * Check meter action from the action list.
2779  *
2780  * @param[in] actions
2781  *   Pointer to the list of actions.
2782  * @param[out] mtr
2783  *   Pointer to the meter exist flag.
2784  *
2785  * @return
2786  *   Total number of actions.
2787  */
2788 static int
2789 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2790 {
2791         int actions_n = 0;
2792
2793         assert(mtr);
2794         *mtr = 0;
2795         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2796                 switch (actions->type) {
2797                 case RTE_FLOW_ACTION_TYPE_METER:
2798                         *mtr = 1;
2799                         break;
2800                 default:
2801                         break;
2802                 }
2803                 actions_n++;
2804         }
2805         /* Count RTE_FLOW_ACTION_TYPE_END. */
2806         return actions_n + 1;
2807 }
2808
2809 /**
2810  * Check if the flow should be splited due to hairpin.
2811  * The reason for the split is that in current HW we can't
2812  * support encap on Rx, so if a flow have encap we move it
2813  * to Tx.
2814  *
2815  * @param dev
2816  *   Pointer to Ethernet device.
2817  * @param[in] attr
2818  *   Flow rule attributes.
2819  * @param[in] actions
2820  *   Associated actions (list terminated by the END action).
2821  *
2822  * @return
2823  *   > 0 the number of actions and the flow should be split,
2824  *   0 when no split required.
2825  */
2826 static int
2827 flow_check_hairpin_split(struct rte_eth_dev *dev,
2828                          const struct rte_flow_attr *attr,
2829                          const struct rte_flow_action actions[])
2830 {
2831         int queue_action = 0;
2832         int action_n = 0;
2833         int encap = 0;
2834         const struct rte_flow_action_queue *queue;
2835         const struct rte_flow_action_rss *rss;
2836         const struct rte_flow_action_raw_encap *raw_encap;
2837
2838         if (!attr->ingress)
2839                 return 0;
2840         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2841                 switch (actions->type) {
2842                 case RTE_FLOW_ACTION_TYPE_QUEUE:
2843                         queue = actions->conf;
2844                         if (queue == NULL)
2845                                 return 0;
2846                         if (mlx5_rxq_get_type(dev, queue->index) !=
2847                             MLX5_RXQ_TYPE_HAIRPIN)
2848                                 return 0;
2849                         queue_action = 1;
2850                         action_n++;
2851                         break;
2852                 case RTE_FLOW_ACTION_TYPE_RSS:
2853                         rss = actions->conf;
2854                         if (rss == NULL || rss->queue_num == 0)
2855                                 return 0;
2856                         if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2857                             MLX5_RXQ_TYPE_HAIRPIN)
2858                                 return 0;
2859                         queue_action = 1;
2860                         action_n++;
2861                         break;
2862                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2863                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2864                         encap = 1;
2865                         action_n++;
2866                         break;
2867                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2868                         raw_encap = actions->conf;
2869                         if (raw_encap->size >
2870                             (sizeof(struct rte_flow_item_eth) +
2871                              sizeof(struct rte_flow_item_ipv4)))
2872                                 encap = 1;
2873                         action_n++;
2874                         break;
2875                 default:
2876                         action_n++;
2877                         break;
2878                 }
2879         }
2880         if (encap == 1 && queue_action)
2881                 return action_n;
2882         return 0;
2883 }
2884
2885 /* Declare flow create/destroy prototype in advance. */
2886 static struct rte_flow *
2887 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2888                  const struct rte_flow_attr *attr,
2889                  const struct rte_flow_item items[],
2890                  const struct rte_flow_action actions[],
2891                  bool external, struct rte_flow_error *error);
2892
2893 static void
2894 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2895                   struct rte_flow *flow);
2896
2897 /**
2898  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2899  *
2900  * As mark_id is unique, if there's already a registered flow for the mark_id,
2901  * return by increasing the reference counter of the resource. Otherwise, create
2902  * the resource (mcp_res) and flow.
2903  *
2904  * Flow looks like,
2905  *   - If ingress port is ANY and reg_c[1] is mark_id,
2906  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2907  *
2908  * For default flow (zero mark_id), flow is like,
2909  *   - If ingress port is ANY,
2910  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2911  *
2912  * @param dev
2913  *   Pointer to Ethernet device.
2914  * @param mark_id
2915  *   ID of MARK action, zero means default flow for META.
2916  * @param[out] error
2917  *   Perform verbose error reporting if not NULL.
2918  *
2919  * @return
2920  *   Associated resource on success, NULL otherwise and rte_errno is set.
2921  */
2922 static struct mlx5_flow_mreg_copy_resource *
2923 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2924                           struct rte_flow_error *error)
2925 {
2926         struct mlx5_priv *priv = dev->data->dev_private;
2927         struct rte_flow_attr attr = {
2928                 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2929                 .ingress = 1,
2930         };
2931         struct mlx5_rte_flow_item_tag tag_spec = {
2932                 .data = mark_id,
2933         };
2934         struct rte_flow_item items[] = {
2935                 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2936         };
2937         struct rte_flow_action_mark ftag = {
2938                 .id = mark_id,
2939         };
2940         struct mlx5_flow_action_copy_mreg cp_mreg = {
2941                 .dst = REG_B,
2942                 .src = 0,
2943         };
2944         struct rte_flow_action_jump jump = {
2945                 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2946         };
2947         struct rte_flow_action actions[] = {
2948                 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2949         };
2950         struct mlx5_flow_mreg_copy_resource *mcp_res;
2951         int ret;
2952
2953         /* Fill the register fileds in the flow. */
2954         ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2955         if (ret < 0)
2956                 return NULL;
2957         tag_spec.id = ret;
2958         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2959         if (ret < 0)
2960                 return NULL;
2961         cp_mreg.src = ret;
2962         /* Check if already registered. */
2963         assert(priv->mreg_cp_tbl);
2964         mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2965         if (mcp_res) {
2966                 /* For non-default rule. */
2967                 if (mark_id != MLX5_DEFAULT_COPY_ID)
2968                         mcp_res->refcnt++;
2969                 assert(mark_id != MLX5_DEFAULT_COPY_ID || mcp_res->refcnt == 1);
2970                 return mcp_res;
2971         }
2972         /* Provide the full width of FLAG specific value. */
2973         if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2974                 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2975         /* Build a new flow. */
2976         if (mark_id != MLX5_DEFAULT_COPY_ID) {
2977                 items[0] = (struct rte_flow_item){
2978                         .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2979                         .spec = &tag_spec,
2980                 };
2981                 items[1] = (struct rte_flow_item){
2982                         .type = RTE_FLOW_ITEM_TYPE_END,
2983                 };
2984                 actions[0] = (struct rte_flow_action){
2985                         .type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2986                         .conf = &ftag,
2987                 };
2988                 actions[1] = (struct rte_flow_action){
2989                         .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2990                         .conf = &cp_mreg,
2991                 };
2992                 actions[2] = (struct rte_flow_action){
2993                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
2994                         .conf = &jump,
2995                 };
2996                 actions[3] = (struct rte_flow_action){
2997                         .type = RTE_FLOW_ACTION_TYPE_END,
2998                 };
2999         } else {
3000                 /* Default rule, wildcard match. */
3001                 attr.priority = MLX5_FLOW_PRIO_RSVD;
3002                 items[0] = (struct rte_flow_item){
3003                         .type = RTE_FLOW_ITEM_TYPE_END,
3004                 };
3005                 actions[0] = (struct rte_flow_action){
3006                         .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3007                         .conf = &cp_mreg,
3008                 };
3009                 actions[1] = (struct rte_flow_action){
3010                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
3011                         .conf = &jump,
3012                 };
3013                 actions[2] = (struct rte_flow_action){
3014                         .type = RTE_FLOW_ACTION_TYPE_END,
3015                 };
3016         }
3017         /* Build a new entry. */
3018         mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3019         if (!mcp_res) {
3020                 rte_errno = ENOMEM;
3021                 return NULL;
3022         }
3023         /*
3024          * The copy Flows are not included in any list. There
3025          * ones are referenced from other Flows and can not
3026          * be applied, removed, deleted in ardbitrary order
3027          * by list traversing.
3028          */
3029         mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3030                                          actions, false, error);
3031         if (!mcp_res->flow)
3032                 goto error;
3033         mcp_res->refcnt++;
3034         mcp_res->hlist_ent.key = mark_id;
3035         ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3036                                 &mcp_res->hlist_ent);
3037         assert(!ret);
3038         if (ret)
3039                 goto error;
3040         return mcp_res;
3041 error:
3042         if (mcp_res->flow)
3043                 flow_list_destroy(dev, NULL, mcp_res->flow);
3044         rte_free(mcp_res);
3045         return NULL;
3046 }
3047
3048 /**
3049  * Release flow in RX_CP_TBL.
3050  *
3051  * @param dev
3052  *   Pointer to Ethernet device.
3053  * @flow
3054  *   Parent flow for wich copying is provided.
3055  */
3056 static void
3057 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3058                           struct rte_flow *flow)
3059 {
3060         struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3061         struct mlx5_priv *priv = dev->data->dev_private;
3062
3063         if (!mcp_res || !priv->mreg_cp_tbl)
3064                 return;
3065         if (flow->copy_applied) {
3066                 assert(mcp_res->appcnt);
3067                 flow->copy_applied = 0;
3068                 --mcp_res->appcnt;
3069                 if (!mcp_res->appcnt)
3070                         flow_drv_remove(dev, mcp_res->flow);
3071         }
3072         /*
3073          * We do not check availability of metadata registers here,
3074          * because copy resources are not allocated in this case.
3075          */
3076         if (--mcp_res->refcnt)
3077                 return;
3078         assert(mcp_res->flow);
3079         flow_list_destroy(dev, NULL, mcp_res->flow);
3080         mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3081         rte_free(mcp_res);
3082         flow->mreg_copy = NULL;
3083 }
3084
3085 /**
3086  * Start flow in RX_CP_TBL.
3087  *
3088  * @param dev
3089  *   Pointer to Ethernet device.
3090  * @flow
3091  *   Parent flow for wich copying is provided.
3092  *
3093  * @return
3094  *   0 on success, a negative errno value otherwise and rte_errno is set.
3095  */
3096 static int
3097 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3098                             struct rte_flow *flow)
3099 {
3100         struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3101         int ret;
3102
3103         if (!mcp_res || flow->copy_applied)
3104                 return 0;
3105         if (!mcp_res->appcnt) {
3106                 ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3107                 if (ret)
3108                         return ret;
3109         }
3110         ++mcp_res->appcnt;
3111         flow->copy_applied = 1;
3112         return 0;
3113 }
3114
3115 /**
3116  * Stop flow in RX_CP_TBL.
3117  *
3118  * @param dev
3119  *   Pointer to Ethernet device.
3120  * @flow
3121  *   Parent flow for wich copying is provided.
3122  */
3123 static void
3124 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3125                            struct rte_flow *flow)
3126 {
3127         struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3128
3129         if (!mcp_res || !flow->copy_applied)
3130                 return;
3131         assert(mcp_res->appcnt);
3132         --mcp_res->appcnt;
3133         flow->copy_applied = 0;
3134         if (!mcp_res->appcnt)
3135                 flow_drv_remove(dev, mcp_res->flow);
3136 }
3137
3138 /**
3139  * Remove the default copy action from RX_CP_TBL.
3140  *
3141  * @param dev
3142  *   Pointer to Ethernet device.
3143  */
3144 static void
3145 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3146 {
3147         struct mlx5_flow_mreg_copy_resource *mcp_res;
3148         struct mlx5_priv *priv = dev->data->dev_private;
3149
3150         /* Check if default flow is registered. */
3151         if (!priv->mreg_cp_tbl)
3152                 return;
3153         mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3154                                             MLX5_DEFAULT_COPY_ID);
3155         if (!mcp_res)
3156                 return;
3157         assert(mcp_res->flow);
3158         flow_list_destroy(dev, NULL, mcp_res->flow);
3159         mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3160         rte_free(mcp_res);
3161 }
3162
3163 /**
3164  * Add the default copy action in in RX_CP_TBL.
3165  *
3166  * @param dev
3167  *   Pointer to Ethernet device.
3168  * @param[out] error
3169  *   Perform verbose error reporting if not NULL.
3170  *
3171  * @return
3172  *   0 for success, negative value otherwise and rte_errno is set.
3173  */
3174 static int
3175 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3176                                   struct rte_flow_error *error)
3177 {
3178         struct mlx5_priv *priv = dev->data->dev_private;
3179         struct mlx5_flow_mreg_copy_resource *mcp_res;
3180
3181         /* Check whether extensive metadata feature is engaged. */
3182         if (!priv->config.dv_flow_en ||
3183             priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3184             !mlx5_flow_ext_mreg_supported(dev) ||
3185             !priv->sh->dv_regc0_mask)
3186                 return 0;
3187         mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3188         if (!mcp_res)
3189                 return -rte_errno;
3190         return 0;
3191 }
3192
3193 /**
3194  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3195  *
3196  * All the flow having Q/RSS action should be split by
3197  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3198  * performs the following,
3199  *   - CQE->flow_tag := reg_c[1] (MARK)
3200  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3201  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3202  * but there should be a flow per each MARK ID set by MARK action.
3203  *
3204  * For the aforementioned reason, if there's a MARK action in flow's action
3205  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3206  * the MARK ID to CQE's flow_tag like,
3207  *   - If reg_c[1] is mark_id,
3208  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3209  *
3210  * For SET_META action which stores value in reg_c[0], as the destination is
3211  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3212  * MARK ID means the default flow. The default flow looks like,
3213  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3214  *
3215  * @param dev
3216  *   Pointer to Ethernet device.
3217  * @param flow
3218  *   Pointer to flow structure.
3219  * @param[in] actions
3220  *   Pointer to the list of actions.
3221  * @param[out] error
3222  *   Perform verbose error reporting if not NULL.
3223  *
3224  * @return
3225  *   0 on success, negative value otherwise and rte_errno is set.
3226  */
3227 static int
3228 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3229                             struct rte_flow *flow,
3230                             const struct rte_flow_action *actions,
3231                             struct rte_flow_error *error)
3232 {
3233         struct mlx5_priv *priv = dev->data->dev_private;
3234         struct mlx5_dev_config *config = &priv->config;
3235         struct mlx5_flow_mreg_copy_resource *mcp_res;
3236         const struct rte_flow_action_mark *mark;
3237
3238         /* Check whether extensive metadata feature is engaged. */
3239         if (!config->dv_flow_en ||
3240             config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3241             !mlx5_flow_ext_mreg_supported(dev) ||
3242             !priv->sh->dv_regc0_mask)
3243                 return 0;
3244         /* Find MARK action. */
3245         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3246                 switch (actions->type) {
3247                 case RTE_FLOW_ACTION_TYPE_FLAG:
3248                         mcp_res = flow_mreg_add_copy_action
3249                                 (dev, MLX5_FLOW_MARK_DEFAULT, error);
3250                         if (!mcp_res)
3251                                 return -rte_errno;
3252                         flow->mreg_copy = mcp_res;
3253                         if (dev->data->dev_started) {
3254                                 mcp_res->appcnt++;
3255                                 flow->copy_applied = 1;
3256                         }
3257                         return 0;
3258                 case RTE_FLOW_ACTION_TYPE_MARK:
3259                         mark = (const struct rte_flow_action_mark *)
3260                                 actions->conf;
3261                         mcp_res =
3262                                 flow_mreg_add_copy_action(dev, mark->id, error);
3263                         if (!mcp_res)
3264                                 return -rte_errno;
3265                         flow->mreg_copy = mcp_res;
3266                         if (dev->data->dev_started) {
3267                                 mcp_res->appcnt++;
3268                                 flow->copy_applied = 1;
3269                         }
3270                         return 0;
3271                 default:
3272                         break;
3273                 }
3274         }
3275         return 0;
3276 }
3277
3278 #define MLX5_MAX_SPLIT_ACTIONS 24
3279 #define MLX5_MAX_SPLIT_ITEMS 24
3280
3281 /**
3282  * Split the hairpin flow.
3283  * Since HW can't support encap on Rx we move the encap to Tx.
3284  * If the count action is after the encap then we also
3285  * move the count action. in this case the count will also measure
3286  * the outer bytes.
3287  *
3288  * @param dev
3289  *   Pointer to Ethernet device.
3290  * @param[in] actions
3291  *   Associated actions (list terminated by the END action).
3292  * @param[out] actions_rx
3293  *   Rx flow actions.
3294  * @param[out] actions_tx
3295  *   Tx flow actions..
3296  * @param[out] pattern_tx
3297  *   The pattern items for the Tx flow.
3298  * @param[out] flow_id
3299  *   The flow ID connected to this flow.
3300  *
3301  * @return
3302  *   0 on success.
3303  */
3304 static int
3305 flow_hairpin_split(struct rte_eth_dev *dev,
3306                    const struct rte_flow_action actions[],
3307                    struct rte_flow_action actions_rx[],
3308                    struct rte_flow_action actions_tx[],
3309                    struct rte_flow_item pattern_tx[],
3310                    uint32_t *flow_id)
3311 {
3312         struct mlx5_priv *priv = dev->data->dev_private;
3313         const struct rte_flow_action_raw_encap *raw_encap;
3314         const struct rte_flow_action_raw_decap *raw_decap;
3315         struct mlx5_rte_flow_action_set_tag *set_tag;
3316         struct rte_flow_action *tag_action;
3317         struct mlx5_rte_flow_item_tag *tag_item;
3318         struct rte_flow_item *item;
3319         char *addr;
3320         int encap = 0;
3321
3322         mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3323         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3324                 switch (actions->type) {
3325                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3326                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3327                         rte_memcpy(actions_tx, actions,
3328                                sizeof(struct rte_flow_action));
3329                         actions_tx++;
3330                         break;
3331                 case RTE_FLOW_ACTION_TYPE_COUNT:
3332                         if (encap) {
3333                                 rte_memcpy(actions_tx, actions,
3334                                            sizeof(struct rte_flow_action));
3335                                 actions_tx++;
3336                         } else {
3337                                 rte_memcpy(actions_rx, actions,
3338                                            sizeof(struct rte_flow_action));
3339                                 actions_rx++;
3340                         }
3341                         break;
3342                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3343                         raw_encap = actions->conf;
3344                         if (raw_encap->size >
3345                             (sizeof(struct rte_flow_item_eth) +
3346                              sizeof(struct rte_flow_item_ipv4))) {
3347                                 memcpy(actions_tx, actions,
3348                                        sizeof(struct rte_flow_action));
3349                                 actions_tx++;
3350                                 encap = 1;
3351                         } else {
3352                                 rte_memcpy(actions_rx, actions,
3353                                            sizeof(struct rte_flow_action));
3354                                 actions_rx++;
3355                         }
3356                         break;
3357                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3358                         raw_decap = actions->conf;
3359                         if (raw_decap->size <
3360                             (sizeof(struct rte_flow_item_eth) +
3361                              sizeof(struct rte_flow_item_ipv4))) {
3362                                 memcpy(actions_tx, actions,
3363                                        sizeof(struct rte_flow_action));
3364                                 actions_tx++;
3365                         } else {
3366                                 rte_memcpy(actions_rx, actions,
3367                                            sizeof(struct rte_flow_action));
3368                                 actions_rx++;
3369                         }
3370                         break;
3371                 default:
3372                         rte_memcpy(actions_rx, actions,
3373                                    sizeof(struct rte_flow_action));
3374                         actions_rx++;
3375                         break;
3376                 }
3377         }
3378         /* Add set meta action and end action for the Rx flow. */
3379         tag_action = actions_rx;
3380         tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3381         actions_rx++;
3382         rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3383         actions_rx++;
3384         set_tag = (void *)actions_rx;
3385         set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3386         assert(set_tag->id > REG_NONE);
3387         set_tag->data = *flow_id;
3388         tag_action->conf = set_tag;
3389         /* Create Tx item list. */
3390         rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3391         addr = (void *)&pattern_tx[2];
3392         item = pattern_tx;
3393         item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3394         tag_item = (void *)addr;
3395         tag_item->data = *flow_id;
3396         tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3397         assert(set_tag->id > REG_NONE);
3398         item->spec = tag_item;
3399         addr += sizeof(struct mlx5_rte_flow_item_tag);
3400         tag_item = (void *)addr;
3401         tag_item->data = UINT32_MAX;
3402         tag_item->id = UINT16_MAX;
3403         item->mask = tag_item;
3404         addr += sizeof(struct mlx5_rte_flow_item_tag);
3405         item->last = NULL;
3406         item++;
3407         item->type = RTE_FLOW_ITEM_TYPE_END;
3408         return 0;
3409 }
3410
3411 /**
3412  * The last stage of splitting chain, just creates the subflow
3413  * without any modification.
3414  *
3415  * @param dev
3416  *   Pointer to Ethernet device.
3417  * @param[in] flow
3418  *   Parent flow structure pointer.
3419  * @param[in, out] sub_flow
3420  *   Pointer to return the created subflow, may be NULL.
3421  * @param[in] attr
3422  *   Flow rule attributes.
3423  * @param[in] items
3424  *   Pattern specification (list terminated by the END pattern item).
3425  * @param[in] actions
3426  *   Associated actions (list terminated by the END action).
3427  * @param[in] external
3428  *   This flow rule is created by request external to PMD.
3429  * @param[out] error
3430  *   Perform verbose error reporting if not NULL.
3431  * @return
3432  *   0 on success, negative value otherwise
3433  */
3434 static int
3435 flow_create_split_inner(struct rte_eth_dev *dev,
3436                         struct rte_flow *flow,
3437                         struct mlx5_flow **sub_flow,
3438                         const struct rte_flow_attr *attr,
3439                         const struct rte_flow_item items[],
3440                         const struct rte_flow_action actions[],
3441                         bool external, struct rte_flow_error *error)
3442 {
3443         struct mlx5_flow *dev_flow;
3444
3445         dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3446         if (!dev_flow)
3447                 return -rte_errno;
3448         dev_flow->flow = flow;
3449         dev_flow->external = external;
3450         /* Subflow object was created, we must include one in the list. */
3451         LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3452         if (sub_flow)
3453                 *sub_flow = dev_flow;
3454         return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3455 }
3456
3457 /**
3458  * Split the meter flow.
3459  *
3460  * As meter flow will split to three sub flow, other than meter
3461  * action, the other actions make sense to only meter accepts
3462  * the packet. If it need to be dropped, no other additional
3463  * actions should be take.
3464  *
3465  * One kind of special action which decapsulates the L3 tunnel
3466  * header will be in the prefix sub flow, as not to take the
3467  * L3 tunnel header into account.
3468  *
3469  * @param dev
3470  *   Pointer to Ethernet device.
3471  * @param[in] actions
3472  *   Associated actions (list terminated by the END action).
3473  * @param[out] actions_sfx
3474  *   Suffix flow actions.
3475  * @param[out] actions_pre
3476  *   Prefix flow actions.
3477  * @param[out] pattern_sfx
3478  *   The pattern items for the suffix flow.
3479  * @param[out] tag_sfx
3480  *   Pointer to suffix flow tag.
3481  *
3482  * @return
3483  *   0 on success.
3484  */
3485 static int
3486 flow_meter_split_prep(struct rte_eth_dev *dev,
3487                  const struct rte_flow_action actions[],
3488                  struct rte_flow_action actions_sfx[],
3489                  struct rte_flow_action actions_pre[])
3490 {
3491         struct rte_flow_action *tag_action;
3492         struct mlx5_rte_flow_action_set_tag *set_tag;
3493         struct rte_flow_error error;
3494         const struct rte_flow_action_raw_encap *raw_encap;
3495         const struct rte_flow_action_raw_decap *raw_decap;
3496         uint32_t tag_id;
3497
3498         /* Add the extra tag action first. */
3499         tag_action = actions_pre;
3500         tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3501         actions_pre++;
3502         /* Prepare the actions for prefix and suffix flow. */
3503         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3504                 switch (actions->type) {
3505                 case RTE_FLOW_ACTION_TYPE_METER:
3506                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3507                 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3508                         memcpy(actions_pre, actions,
3509                                sizeof(struct rte_flow_action));
3510                         actions_pre++;
3511                         break;
3512                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3513                         raw_encap = actions->conf;
3514                         if (raw_encap->size >
3515                             (sizeof(struct rte_flow_item_eth) +
3516                              sizeof(struct rte_flow_item_ipv4))) {
3517                                 memcpy(actions_sfx, actions,
3518                                        sizeof(struct rte_flow_action));
3519                                 actions_sfx++;
3520                         } else {
3521                                 rte_memcpy(actions_pre, actions,
3522                                            sizeof(struct rte_flow_action));
3523                                 actions_pre++;
3524                         }
3525                         break;
3526                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3527                         raw_decap = actions->conf;
3528                         /* Size 0 decap means 50 bytes as vxlan decap. */
3529                         if (raw_decap->size && (raw_decap->size <
3530                             (sizeof(struct rte_flow_item_eth) +
3531                              sizeof(struct rte_flow_item_ipv4)))) {
3532                                 memcpy(actions_sfx, actions,
3533                                        sizeof(struct rte_flow_action));
3534                                 actions_sfx++;
3535                         } else {
3536                                 rte_memcpy(actions_pre, actions,
3537                                            sizeof(struct rte_flow_action));
3538                                 actions_pre++;
3539                         }
3540                         break;
3541                 default:
3542                         memcpy(actions_sfx, actions,
3543                                 sizeof(struct rte_flow_action));
3544                         actions_sfx++;
3545                         break;
3546                 }
3547         }
3548         /* Add end action to the actions. */
3549         actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3550         actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3551         actions_pre++;
3552         /* Set the tag. */
3553         set_tag = (void *)actions_pre;
3554         set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3555         /*
3556          * Get the id from the qrss_pool to make qrss share the id with meter.
3557          */
3558         tag_id = flow_qrss_get_id(dev);
3559         set_tag->data = rte_cpu_to_be_32(tag_id);
3560         tag_action->conf = set_tag;
3561         return tag_id;
3562 }
3563
3564 /**
3565  * Split action list having QUEUE/RSS for metadata register copy.
3566  *
3567  * Once Q/RSS action is detected in user's action list, the flow action
3568  * should be split in order to copy metadata registers, which will happen in
3569  * RX_CP_TBL like,
3570  *   - CQE->flow_tag := reg_c[1] (MARK)
3571  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3572  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3573  * This is because the last action of each flow must be a terminal action
3574  * (QUEUE, RSS or DROP).
3575  *
3576  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3577  * stored and kept in the mlx5_flow structure per each sub_flow.
3578  *
3579  * The Q/RSS action is replaced with,
3580  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3581  * And the following JUMP action is added at the end,
3582  *   - JUMP, to RX_CP_TBL.
3583  *
3584  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3585  * flow_create_split_metadata() routine. The flow will look like,
3586  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3587  *
3588  * @param dev
3589  *   Pointer to Ethernet device.
3590  * @param[out] split_actions
3591  *   Pointer to store split actions to jump to CP_TBL.
3592  * @param[in] actions
3593  *   Pointer to the list of original flow actions.
3594  * @param[in] qrss
3595  *   Pointer to the Q/RSS action.
3596  * @param[in] actions_n
3597  *   Number of original actions.
3598  * @param[out] error
3599  *   Perform verbose error reporting if not NULL.
3600  *
3601  * @return
3602  *   non-zero unique flow_id on success, otherwise 0 and
3603  *   error/rte_error are set.
3604  */
3605 static uint32_t
3606 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3607                           struct rte_flow_action *split_actions,
3608                           const struct rte_flow_action *actions,
3609                           const struct rte_flow_action *qrss,
3610                           int actions_n, struct rte_flow_error *error)
3611 {
3612         struct mlx5_rte_flow_action_set_tag *set_tag;
3613         struct rte_flow_action_jump *jump;
3614         const int qrss_idx = qrss - actions;
3615         uint32_t flow_id = 0;
3616         int ret = 0;
3617
3618         /*
3619          * Given actions will be split
3620          * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3621          * - Add jump to mreg CP_TBL.
3622          * As a result, there will be one more action.
3623          */
3624         ++actions_n;
3625         memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3626         set_tag = (void *)(split_actions + actions_n);
3627         /*
3628          * If tag action is not set to void(it means we are not the meter
3629          * suffix flow), add the tag action. Since meter suffix flow already
3630          * has the tag added.
3631          */
3632         if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3633                 /*
3634                  * Allocate the new subflow ID. This one is unique within
3635                  * device and not shared with representors. Otherwise,
3636                  * we would have to resolve multi-thread access synch
3637                  * issue. Each flow on the shared device is appended
3638                  * with source vport identifier, so the resulting
3639                  * flows will be unique in the shared (by master and
3640                  * representors) domain even if they have coinciding
3641                  * IDs.
3642                  */
3643                 flow_id = flow_qrss_get_id(dev);
3644                 if (!flow_id)
3645                         return rte_flow_error_set(error, ENOMEM,
3646                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3647                                                   NULL, "can't allocate id "
3648                                                   "for split Q/RSS subflow");
3649                 /* Internal SET_TAG action to set flow ID. */
3650                 *set_tag = (struct mlx5_rte_flow_action_set_tag){
3651                         .data = flow_id,
3652                 };
3653                 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3654                 if (ret < 0)
3655                         return ret;
3656                 set_tag->id = ret;
3657                 /* Construct new actions array. */
3658                 /* Replace QUEUE/RSS action. */
3659                 split_actions[qrss_idx] = (struct rte_flow_action){
3660                         .type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3661                         .conf = set_tag,
3662                 };
3663         }
3664         /* JUMP action to jump to mreg copy table (CP_TBL). */
3665         jump = (void *)(set_tag + 1);
3666         *jump = (struct rte_flow_action_jump){
3667                 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3668         };
3669         split_actions[actions_n - 2] = (struct rte_flow_action){
3670                 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3671                 .conf = jump,
3672         };
3673         split_actions[actions_n - 1] = (struct rte_flow_action){
3674                 .type = RTE_FLOW_ACTION_TYPE_END,
3675         };
3676         return flow_id;
3677 }
3678
3679 /**
3680  * Extend the given action list for Tx metadata copy.
3681  *
3682  * Copy the given action list to the ext_actions and add flow metadata register
3683  * copy action in order to copy reg_a set by WQE to reg_c[0].
3684  *
3685  * @param[out] ext_actions
3686  *   Pointer to the extended action list.
3687  * @param[in] actions
3688  *   Pointer to the list of actions.
3689  * @param[in] actions_n
3690  *   Number of actions in the list.
3691  * @param[out] error
3692  *   Perform verbose error reporting if not NULL.
3693  *
3694  * @return
3695  *   0 on success, negative value otherwise
3696  */
3697 static int
3698 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3699                        struct rte_flow_action *ext_actions,
3700                        const struct rte_flow_action *actions,
3701                        int actions_n, struct rte_flow_error *error)
3702 {
3703         struct mlx5_flow_action_copy_mreg *cp_mreg =
3704                 (struct mlx5_flow_action_copy_mreg *)
3705                         (ext_actions + actions_n + 1);
3706         int ret;
3707
3708         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3709         if (ret < 0)
3710                 return ret;
3711         cp_mreg->dst = ret;
3712         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3713         if (ret < 0)
3714                 return ret;
3715         cp_mreg->src = ret;
3716         memcpy(ext_actions, actions,
3717                         sizeof(*ext_actions) * actions_n);
3718         ext_actions[actions_n - 1] = (struct rte_flow_action){
3719                 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3720                 .conf = cp_mreg,
3721         };
3722         ext_actions[actions_n] = (struct rte_flow_action){
3723                 .type = RTE_FLOW_ACTION_TYPE_END,
3724         };
3725         return 0;
3726 }
3727
3728 /**
3729  * The splitting for metadata feature.
3730  *
3731  * - Q/RSS action on NIC Rx should be split in order to pass by
3732  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3733  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3734  *
3735  * - All the actions on NIC Tx should have a mreg copy action to
3736  *   copy reg_a from WQE to reg_c[0].
3737  *
3738  * @param dev
3739  *   Pointer to Ethernet device.
3740  * @param[in] flow
3741  *   Parent flow structure pointer.
3742  * @param[in] attr
3743  *   Flow rule attributes.
3744  * @param[in] items
3745  *   Pattern specification (list terminated by the END pattern item).
3746  * @param[in] actions
3747  *   Associated actions (list terminated by the END action).
3748  * @param[in] external
3749  *   This flow rule is created by request external to PMD.
3750  * @param[out] error
3751  *   Perform verbose error reporting if not NULL.
3752  * @return
3753  *   0 on success, negative value otherwise
3754  */
3755 static int
3756 flow_create_split_metadata(struct rte_eth_dev *dev,
3757                            struct rte_flow *flow,
3758                            const struct rte_flow_attr *attr,
3759                            const struct rte_flow_item items[],
3760                            const struct rte_flow_action actions[],
3761                            bool external, struct rte_flow_error *error)
3762 {
3763         struct mlx5_priv *priv = dev->data->dev_private;
3764         struct mlx5_dev_config *config = &priv->config;
3765         const struct rte_flow_action *qrss = NULL;
3766         struct rte_flow_action *ext_actions = NULL;
3767         struct mlx5_flow *dev_flow = NULL;
3768         uint32_t qrss_id = 0;
3769         int mtr_sfx = 0;
3770         size_t act_size;
3771         int actions_n;
3772         int ret;
3773
3774         /* Check whether extensive metadata feature is engaged. */
3775         if (!config->dv_flow_en ||
3776             config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3777             !mlx5_flow_ext_mreg_supported(dev))
3778                 return flow_create_split_inner(dev, flow, NULL, attr, items,
3779                                                actions, external, error);
3780         actions_n = flow_parse_qrss_action(actions, &qrss);
3781         if (qrss) {
3782                 /* Exclude hairpin flows from splitting. */
3783                 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3784                         const struct rte_flow_action_queue *queue;
3785
3786                         queue = qrss->conf;
3787                         if (mlx5_rxq_get_type(dev, queue->index) ==
3788                             MLX5_RXQ_TYPE_HAIRPIN)
3789                                 qrss = NULL;
3790                 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3791                         const struct rte_flow_action_rss *rss;
3792
3793                         rss = qrss->conf;
3794                         if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3795                             MLX5_RXQ_TYPE_HAIRPIN)
3796                                 qrss = NULL;
3797                 }
3798         }
3799         if (qrss) {
3800                 /* Check if it is in meter suffix table. */
3801                 mtr_sfx = attr->group == (attr->transfer ?
3802                           (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3803                           MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3804                 /*
3805                  * Q/RSS action on NIC Rx should be split in order to pass by
3806                  * the mreg copy table (RX_CP_TBL) and then it jumps to the
3807                  * action table (RX_ACT_TBL) which has the split Q/RSS action.
3808                  */
3809                 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3810                            sizeof(struct rte_flow_action_set_tag) +
3811                            sizeof(struct rte_flow_action_jump);
3812                 ext_actions = rte_zmalloc(__func__, act_size, 0);
3813                 if (!ext_actions)
3814                         return rte_flow_error_set(error, ENOMEM,
3815                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3816                                                   NULL, "no memory to split "
3817                                                   "metadata flow");
3818                 /*
3819                  * If we are the suffix flow of meter, tag already exist.
3820                  * Set the tag action to void.
3821                  */
3822                 if (mtr_sfx)
3823                         ext_actions[qrss - actions].type =
3824                                                 RTE_FLOW_ACTION_TYPE_VOID;
3825                 else
3826                         ext_actions[qrss - actions].type =
3827                                                 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3828                 /*
3829                  * Create the new actions list with removed Q/RSS action
3830                  * and appended set tag and jump to register copy table
3831                  * (RX_CP_TBL). We should preallocate unique tag ID here
3832                  * in advance, because it is needed for set tag action.
3833                  */
3834                 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3835                                                     qrss, actions_n, error);
3836                 if (!mtr_sfx && !qrss_id) {
3837                         ret = -rte_errno;
3838                         goto exit;
3839                 }
3840         } else if (attr->egress && !attr->transfer) {
3841                 /*
3842                  * All the actions on NIC Tx should have a metadata register
3843                  * copy action to copy reg_a from WQE to reg_c[meta]
3844                  */
3845                 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3846                            sizeof(struct mlx5_flow_action_copy_mreg);
3847                 ext_actions = rte_zmalloc(__func__, act_size, 0);
3848                 if (!ext_actions)
3849                         return rte_flow_error_set(error, ENOMEM,
3850                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3851                                                   NULL, "no memory to split "
3852                                                   "metadata flow");
3853                 /* Create the action list appended with copy register. */
3854                 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3855                                              actions_n, error);
3856                 if (ret < 0)
3857                         goto exit;
3858         }
3859         /* Add the unmodified original or prefix subflow. */
3860         ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3861                                       ext_actions ? ext_actions : actions,
3862                                       external, error);
3863         if (ret < 0)
3864                 goto exit;
3865         assert(dev_flow);
3866         if (qrss) {
3867                 const struct rte_flow_attr q_attr = {
3868                         .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3869                         .ingress = 1,
3870                 };
3871                 /* Internal PMD action to set register. */
3872                 struct mlx5_rte_flow_item_tag q_tag_spec = {
3873                         .data = qrss_id,
3874                         .id = 0,
3875                 };
3876                 struct rte_flow_item q_items[] = {
3877                         {
3878                                 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3879                                 .spec = &q_tag_spec,
3880                                 .last = NULL,
3881                                 .mask = NULL,
3882                         },
3883                         {
3884                                 .type = RTE_FLOW_ITEM_TYPE_END,
3885                         },
3886                 };
3887                 struct rte_flow_action q_actions[] = {
3888                         {
3889                                 .type = qrss->type,
3890                                 .conf = qrss->conf,
3891                         },
3892                         {
3893                                 .type = RTE_FLOW_ACTION_TYPE_END,
3894                         },
3895                 };
3896                 uint64_t hash_fields = dev_flow->hash_fields;
3897
3898                 /*
3899                  * Configure the tag item only if there is no meter subflow.
3900                  * Since tag is already marked in the meter suffix subflow
3901                  * we can just use the meter suffix items as is.
3902                  */
3903                 if (qrss_id) {
3904                         /* Not meter subflow. */
3905                         assert(!mtr_sfx);
3906                         /*
3907                          * Put unique id in prefix flow due to it is destroyed
3908                          * after suffix flow and id will be freed after there
3909                          * is no actual flows with this id and identifier
3910                          * reallocation becomes possible (for example, for
3911                          * other flows in other threads).
3912                          */
3913                         dev_flow->qrss_id = qrss_id;
3914                         qrss_id = 0;
3915                         ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3916                                                    error);
3917                         if (ret < 0)
3918                                 goto exit;
3919                         q_tag_spec.id = ret;
3920                 }
3921                 dev_flow = NULL;
3922                 /* Add suffix subflow to execute Q/RSS. */
3923                 ret = flow_create_split_inner(dev, flow, &dev_flow,
3924                                               &q_attr, mtr_sfx ? items :
3925                                               q_items, q_actions,
3926                                               external, error);
3927                 if (ret < 0)
3928                         goto exit;
3929                 assert(dev_flow);
3930                 dev_flow->hash_fields = hash_fields;
3931         }
3932
3933 exit:
3934         /*
3935          * We do not destroy the partially created sub_flows in case of error.
3936          * These ones are included into parent flow list and will be destroyed
3937          * by flow_drv_destroy.
3938          */
3939         flow_qrss_free_id(dev, qrss_id);
3940         rte_free(ext_actions);
3941         return ret;
3942 }
3943
3944 /**
3945  * The splitting for meter feature.
3946  *
3947  * - The meter flow will be split to two flows as prefix and
3948  *   suffix flow. The packets make sense only it pass the prefix
3949  *   meter action.
3950  *
3951  * - Reg_C_5 is used for the packet to match betweend prefix and
3952  *   suffix flow.
3953  *
3954  * @param dev
3955  *   Pointer to Ethernet device.
3956  * @param[in] flow
3957  *   Parent flow structure pointer.
3958  * @param[in] attr
3959  *   Flow rule attributes.
3960  * @param[in] items
3961  *   Pattern specification (list terminated by the END pattern item).
3962  * @param[in] actions
3963  *   Associated actions (list terminated by the END action).
3964  * @param[in] external
3965  *   This flow rule is created by request external to PMD.
3966  * @param[out] error
3967  *   Perform verbose error reporting if not NULL.
3968  * @return
3969  *   0 on success, negative value otherwise
3970  */
3971 static int
3972 flow_create_split_meter(struct rte_eth_dev *dev,
3973                            struct rte_flow *flow,
3974                            const struct rte_flow_attr *attr,
3975                            const struct rte_flow_item items[],
3976                            const struct rte_flow_action actions[],
3977                            bool external, struct rte_flow_error *error)
3978 {
3979         struct mlx5_priv *priv = dev->data->dev_private;
3980         struct rte_flow_action *sfx_actions = NULL;
3981         struct rte_flow_action *pre_actions = NULL;
3982         struct rte_flow_item *sfx_items = NULL;
3983         const  struct rte_flow_item *sfx_port_id_item;
3984         struct mlx5_flow *dev_flow = NULL;
3985         struct rte_flow_attr sfx_attr = *attr;
3986         uint32_t mtr = 0;
3987         uint32_t mtr_tag_id = 0;
3988         size_t act_size;
3989         size_t item_size;
3990         int actions_n = 0;
3991         int ret;
3992
3993         if (priv->mtr_en)
3994                 actions_n = flow_check_meter_action(actions, &mtr);
3995         if (mtr) {
3996                 struct mlx5_rte_flow_item_tag *tag_spec;
3997                 /* The five prefix actions: meter, decap, encap, tag, end. */
3998                 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3999                            sizeof(struct rte_flow_action_set_tag);
4000                 /* tag, end. */
4001 #define METER_SUFFIX_ITEM 3
4002                 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4003                             sizeof(struct mlx5_rte_flow_item_tag);
4004                 sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4005                 if (!sfx_actions)
4006                         return rte_flow_error_set(error, ENOMEM,
4007                                                   RTE_FLOW_ERROR_TYPE_ACTION,
4008                                                   NULL, "no memory to split "
4009                                                   "meter flow");
4010                 pre_actions = sfx_actions + actions_n;
4011                 mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
4012                                                      pre_actions);
4013                 if (!mtr_tag_id) {
4014                         ret = -rte_errno;
4015                         goto exit;
4016                 }
4017                 /* Add the prefix subflow. */
4018                 ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
4019                                                   pre_actions, external, error);
4020                 if (ret) {
4021                         ret = -rte_errno;
4022                         goto exit;
4023                 }
4024                 dev_flow->mtr_flow_id = mtr_tag_id;
4025                 /* Prepare the suffix flow match pattern. */
4026                 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4027                              act_size);
4028                 tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
4029                             METER_SUFFIX_ITEM);
4030                 tag_spec->data = rte_cpu_to_be_32(dev_flow->mtr_flow_id);
4031                 tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
4032                                                     error);
4033                 sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4034                 sfx_items->spec = tag_spec;
4035                 sfx_items->last = NULL;
4036                 sfx_items->mask = NULL;
4037                 sfx_items++;
4038                 sfx_port_id_item = find_port_id_item(items);
4039                 if (sfx_port_id_item) {
4040                         memcpy(sfx_items, sfx_port_id_item,
4041                                sizeof(*sfx_items));
4042                         sfx_items++;
4043                 }
4044                 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4045                 sfx_items -= METER_SUFFIX_ITEM;
4046                 /* Setting the sfx group atrr. */
4047                 sfx_attr.group = sfx_attr.transfer ?
4048                                 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4049                                  MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4050         }
4051         /* Add the prefix subflow. */
4052         ret = flow_create_split_metadata(dev, flow, &sfx_attr,
4053                                          sfx_items ? sfx_items : items,
4054                                          sfx_actions ? sfx_actions : actions,
4055                                          external, error);
4056 exit:
4057         if (sfx_actions)
4058                 rte_free(sfx_actions);
4059         return ret;
4060 }
4061
4062 /**
4063  * Split the flow to subflow set. The splitters might be linked
4064  * in the chain, like this:
4065  * flow_create_split_outer() calls:
4066  *   flow_create_split_meter() calls:
4067  *     flow_create_split_metadata(meter_subflow_0) calls:
4068  *       flow_create_split_inner(metadata_subflow_0)
4069  *       flow_create_split_inner(metadata_subflow_1)
4070  *       flow_create_split_inner(metadata_subflow_2)
4071  *     flow_create_split_metadata(meter_subflow_1) calls:
4072  *       flow_create_split_inner(metadata_subflow_0)
4073  *       flow_create_split_inner(metadata_subflow_1)
4074  *       flow_create_split_inner(metadata_subflow_2)
4075  *
4076  * This provide flexible way to add new levels of flow splitting.
4077  * The all of successfully created subflows are included to the
4078  * parent flow dev_flow list.
4079  *
4080  * @param dev
4081  *   Pointer to Ethernet device.
4082  * @param[in] flow
4083  *   Parent flow structure pointer.
4084  * @param[in] attr
4085  *   Flow rule attributes.
4086  * @param[in] items
4087  *   Pattern specification (list terminated by the END pattern item).
4088  * @param[in] actions
4089  *   Associated actions (list terminated by the END action).
4090  * @param[in] external
4091  *   This flow rule is created by request external to PMD.
4092  * @param[out] error
4093  *   Perform verbose error reporting if not NULL.
4094  * @return
4095  *   0 on success, negative value otherwise
4096  */
4097 static int
4098 flow_create_split_outer(struct rte_eth_dev *dev,
4099                         struct rte_flow *flow,
4100                         const struct rte_flow_attr *attr,
4101                         const struct rte_flow_item items[],
4102                         const struct rte_flow_action actions[],
4103                         bool external, struct rte_flow_error *error)
4104 {
4105         int ret;
4106
4107         ret = flow_create_split_meter(dev, flow, attr, items,
4108                                          actions, external, error);
4109         assert(ret <= 0);
4110         return ret;
4111 }
4112
4113 /**
4114  * Create a flow and add it to @p list.
4115  *
4116  * @param dev
4117  *   Pointer to Ethernet device.
4118  * @param list
4119  *   Pointer to a TAILQ flow list. If this parameter NULL,
4120  *   no list insertion occurred, flow is just created,
4121  *   this is caller's responsibility to track the
4122  *   created flow.
4123  * @param[in] attr
4124  *   Flow rule attributes.
4125  * @param[in] items
4126  *   Pattern specification (list terminated by the END pattern item).
4127  * @param[in] actions
4128  *   Associated actions (list terminated by the END action).
4129  * @param[in] external
4130  *   This flow rule is created by request external to PMD.
4131  * @param[out] error
4132  *   Perform verbose error reporting if not NULL.
4133  *
4134  * @return
4135  *   A flow on success, NULL otherwise and rte_errno is set.
4136  */
4137 static struct rte_flow *
4138 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4139                  const struct rte_flow_attr *attr,
4140                  const struct rte_flow_item items[],
4141                  const struct rte_flow_action actions[],
4142                  bool external, struct rte_flow_error *error)
4143 {
4144         struct mlx5_priv *priv = dev->data->dev_private;
4145         struct rte_flow *flow = NULL;
4146         struct mlx5_flow *dev_flow;
4147         const struct rte_flow_action_rss *rss;
4148         union {
4149                 struct rte_flow_expand_rss buf;
4150                 uint8_t buffer[2048];
4151         } expand_buffer;
4152         union {
4153                 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4154                 uint8_t buffer[2048];
4155         } actions_rx;
4156         union {
4157                 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4158                 uint8_t buffer[2048];
4159         } actions_hairpin_tx;
4160         union {
4161                 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4162                 uint8_t buffer[2048];
4163         } items_tx;
4164         struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4165         const struct rte_flow_action *p_actions_rx = actions;
4166         int ret;
4167         uint32_t i;
4168         uint32_t flow_size;
4169         int hairpin_flow = 0;
4170         uint32_t hairpin_id = 0;
4171         struct rte_flow_attr attr_tx = { .priority = 0 };
4172
4173         hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4174         if (hairpin_flow > 0) {
4175                 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4176                         rte_errno = EINVAL;
4177                         return NULL;
4178                 }
4179                 flow_hairpin_split(dev, actions, actions_rx.actions,
4180                                    actions_hairpin_tx.actions, items_tx.items,
4181                                    &hairpin_id);
4182                 p_actions_rx = actions_rx.actions;
4183         }
4184         ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4185                                 error);
4186         if (ret < 0)
4187                 goto error_before_flow;
4188         flow_size = sizeof(struct rte_flow);
4189         rss = flow_get_rss_action(p_actions_rx);
4190         if (rss)
4191                 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4192                                             sizeof(void *));
4193         else
4194                 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4195         flow = rte_calloc(__func__, 1, flow_size, 0);
4196         if (!flow) {
4197                 rte_errno = ENOMEM;
4198                 goto error_before_flow;
4199         }
4200         flow->drv_type = flow_get_drv_type(dev, attr);
4201         if (hairpin_id != 0)
4202                 flow->hairpin_flow_id = hairpin_id;
4203         assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4204                flow->drv_type < MLX5_FLOW_TYPE_MAX);
4205         flow->rss.queue = (void *)(flow + 1);
4206         if (rss) {
4207                 /*
4208                  * The following information is required by
4209                  * mlx5_flow_hashfields_adjust() in advance.
4210                  */
4211                 flow->rss.level = rss->level;
4212                 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4213                 flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4214         }
4215         LIST_INIT(&flow->dev_flows);
4216         if (rss && rss->types) {
4217                 unsigned int graph_root;
4218
4219                 graph_root = find_graph_root(items, rss->level);
4220                 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4221                                           items, rss->types,
4222                                           mlx5_support_expansion,
4223                                           graph_root);
4224                 assert(ret > 0 &&
4225                        (unsigned int)ret < sizeof(expand_buffer.buffer));
4226         } else {
4227                 buf->entries = 1;
4228                 buf->entry[0].pattern = (void *)(uintptr_t)items;
4229         }
4230         for (i = 0; i < buf->entries; ++i) {
4231                 /*
4232                  * The splitter may create multiple dev_flows,
4233                  * depending on configuration. In the simplest
4234                  * case it just creates unmodified original flow.
4235                  */
4236                 ret = flow_create_split_outer(dev, flow, attr,
4237                                               buf->entry[i].pattern,
4238                                               p_actions_rx, external,
4239                                               error);
4240                 if (ret < 0)
4241                         goto error;
4242         }
4243         /* Create the tx flow. */
4244         if (hairpin_flow) {
4245                 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4246                 attr_tx.ingress = 0;
4247                 attr_tx.egress = 1;
4248                 dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4249                                             actions_hairpin_tx.actions, error);
4250                 if (!dev_flow)
4251                         goto error;
4252                 dev_flow->flow = flow;
4253                 dev_flow->external = 0;
4254                 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4255                 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4256                                          items_tx.items,
4257                                          actions_hairpin_tx.actions, error);
4258                 if (ret < 0)
4259                         goto error;
4260         }
4261         /*
4262          * Update the metadata register copy table. If extensive
4263          * metadata feature is enabled and registers are supported
4264          * we might create the extra rte_flow for each unique
4265          * MARK/FLAG action ID.
4266          *
4267          * The table is updated for ingress Flows only, because
4268          * the egress Flows belong to the different device and
4269          * copy table should be updated in peer NIC Rx domain.
4270          */
4271         if (attr->ingress &&
4272             (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4273                 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4274                 if (ret)
4275                         goto error;
4276         }
4277         if (dev->data->dev_started) {
4278                 ret = flow_drv_apply(dev, flow, error);
4279                 if (ret < 0)
4280                         goto error;
4281         }
4282         if (list)
4283                 TAILQ_INSERT_TAIL(list, flow, next);
4284         flow_rxq_flags_set(dev, flow);
4285         return flow;
4286 error_before_flow:
4287         if (hairpin_id)
4288                 mlx5_flow_id_release(priv->sh->flow_id_pool,
4289                                      hairpin_id);
4290         return NULL;
4291 error:
4292         assert(flow);
4293         flow_mreg_del_copy_action(dev, flow);
4294         ret = rte_errno; /* Save rte_errno before cleanup. */
4295         if (flow->hairpin_flow_id)
4296                 mlx5_flow_id_release(priv->sh->flow_id_pool,
4297                                      flow->hairpin_flow_id);
4298         assert(flow);
4299         flow_drv_destroy(dev, flow);
4300         rte_free(flow);
4301         rte_errno = ret; /* Restore rte_errno. */
4302         return NULL;
4303 }
4304
4305 /**
4306  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4307  * incoming packets to table 1.
4308  *
4309  * Other flow rules, requested for group n, will be created in
4310  * e-switch table n+1.
4311  * Jump action to e-switch group n will be created to group n+1.
4312  *
4313  * Used when working in switchdev mode, to utilise advantages of table 1
4314  * and above.
4315  *
4316  * @param dev
4317  *   Pointer to Ethernet device.
4318  *
4319  * @return
4320  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4321  */
4322 struct rte_flow *
4323 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4324 {
4325         const struct rte_flow_attr attr = {
4326                 .group = 0,
4327                 .priority = 0,
4328                 .ingress = 1,
4329                 .egress = 0,
4330                 .transfer = 1,
4331         };
4332         const struct rte_flow_item pattern = {
4333                 .type = RTE_FLOW_ITEM_TYPE_END,
4334         };
4335         struct rte_flow_action_jump jump = {
4336                 .group = 1,
4337         };
4338         const struct rte_flow_action actions[] = {
4339                 {
4340                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
4341                         .conf = &jump,
4342                 },
4343                 {
4344                         .type = RTE_FLOW_ACTION_TYPE_END,
4345                 },
4346         };
4347         struct mlx5_priv *priv = dev->data->dev_private;
4348         struct rte_flow_error error;
4349
4350         return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4351                                 actions, false, &error);
4352 }
4353
4354 /**
4355  * Create a flow.
4356  *
4357  * @see rte_flow_create()
4358  * @see rte_flow_ops
4359  */
4360 struct rte_flow *
4361 mlx5_flow_create(struct rte_eth_dev *dev,
4362                  const struct rte_flow_attr *attr,
4363                  const struct rte_flow_item items[],
4364                  const struct rte_flow_action actions[],
4365                  struct rte_flow_error *error)
4366 {
4367         struct mlx5_priv *priv = dev->data->dev_private;
4368
4369         return flow_list_create(dev, &priv->flows,
4370                                 attr, items, actions, true, error);
4371 }
4372
4373 /**
4374  * Destroy a flow in a list.
4375  *
4376  * @param dev
4377  *   Pointer to Ethernet device.
4378  * @param list
4379  *   Pointer to a TAILQ flow list. If this parameter NULL,
4380  *   there is no flow removal from the list.
4381  * @param[in] flow
4382  *   Flow to destroy.
4383  */
4384 static void
4385 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4386                   struct rte_flow *flow)
4387 {
4388         struct mlx5_priv *priv = dev->data->dev_private;
4389
4390         /*
4391          * Update RX queue flags only if port is started, otherwise it is
4392          * already clean.
4393          */
4394         if (dev->data->dev_started)
4395                 flow_rxq_flags_trim(dev, flow);
4396         if (flow->hairpin_flow_id)
4397                 mlx5_flow_id_release(priv->sh->flow_id_pool,
4398                                      flow->hairpin_flow_id);
4399         flow_drv_destroy(dev, flow);
4400         if (list)
4401                 TAILQ_REMOVE(list, flow, next);
4402         flow_mreg_del_copy_action(dev, flow);
4403         rte_free(flow->fdir);
4404         rte_free(flow);
4405 }
4406
4407 /**
4408  * Destroy all flows.
4409  *
4410  * @param dev
4411  *   Pointer to Ethernet device.
4412  * @param list
4413  *   Pointer to a TAILQ flow list.
4414  */
4415 void
4416 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4417 {
4418         while (!TAILQ_EMPTY(list)) {
4419                 struct rte_flow *flow;
4420
4421                 flow = TAILQ_FIRST(list);
4422                 flow_list_destroy(dev, list, flow);
4423         }
4424 }
4425
4426 /**
4427  * Remove all flows.
4428  *
4429  * @param dev
4430  *   Pointer to Ethernet device.
4431  * @param list
4432  *   Pointer to a TAILQ flow list.
4433  */
4434 void
4435 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4436 {
4437         struct rte_flow *flow;
4438
4439         TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4440                 flow_drv_remove(dev, flow);
4441                 flow_mreg_stop_copy_action(dev, flow);
4442         }
4443         flow_mreg_del_default_copy_action(dev);
4444         flow_rxq_flags_clear(dev);
4445 }
4446
4447 /**
4448  * Add all flows.
4449  *
4450  * @param dev
4451  *   Pointer to Ethernet device.
4452  * @param list
4453  *   Pointer to a TAILQ flow list.
4454  *
4455  * @return
4456  *   0 on success, a negative errno value otherwise and rte_errno is set.
4457  */
4458 int
4459 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4460 {
4461         struct rte_flow *flow;
4462         struct rte_flow_error error;
4463         int ret = 0;
4464
4465         /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4466         ret = flow_mreg_add_default_copy_action(dev, &error);
4467         if (ret < 0)
4468                 return -rte_errno;
4469         /* Apply Flows created by application. */
4470         TAILQ_FOREACH(flow, list, next) {
4471                 ret = flow_mreg_start_copy_action(dev, flow);
4472                 if (ret < 0)
4473                         goto error;
4474                 ret = flow_drv_apply(dev, flow, &error);
4475                 if (ret < 0)
4476                         goto error;
4477                 flow_rxq_flags_set(dev, flow);
4478         }
4479         return 0;
4480 error:
4481         ret = rte_errno; /* Save rte_errno before cleanup. */
4482         mlx5_flow_stop(dev, list);
4483         rte_errno = ret; /* Restore rte_errno. */
4484         return -rte_errno;
4485 }
4486
4487 /**
4488  * Verify the flow list is empty
4489  *
4490  * @param dev
4491  *  Pointer to Ethernet device.
4492  *
4493  * @return the number of flows not released.
4494  */
4495 int
4496 mlx5_flow_verify(struct rte_eth_dev *dev)
4497 {
4498         struct mlx5_priv *priv = dev->data->dev_private;
4499         struct rte_flow *flow;
4500         int ret = 0;
4501
4502         TAILQ_FOREACH(flow, &priv->flows, next) {
4503                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
4504                         dev->data->port_id, (void *)flow);
4505                 ++ret;
4506         }
4507         return ret;
4508 }
4509
4510 /**
4511  * Enable default hairpin egress flow.
4512  *
4513  * @param dev
4514  *   Pointer to Ethernet device.
4515  * @param queue
4516  *   The queue index.
4517  *
4518  * @return
4519  *   0 on success, a negative errno value otherwise and rte_errno is set.
4520  */
4521 int
4522 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4523                             uint32_t queue)
4524 {
4525         struct mlx5_priv *priv = dev->data->dev_private;
4526         const struct rte_flow_attr attr = {
4527                 .egress = 1,
4528                 .priority = 0,
4529         };
4530         struct mlx5_rte_flow_item_tx_queue queue_spec = {
4531                 .queue = queue,
4532         };
4533         struct mlx5_rte_flow_item_tx_queue queue_mask = {
4534                 .queue = UINT32_MAX,
4535         };
4536         struct rte_flow_item items[] = {
4537                 {
4538                         .type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4539                         .spec = &queue_spec,
4540                         .last = NULL,
4541                         .mask = &queue_mask,
4542                 },
4543                 {
4544                         .type = RTE_FLOW_ITEM_TYPE_END,
4545                 },
4546         };
4547         struct rte_flow_action_jump jump = {
4548                 .group = MLX5_HAIRPIN_TX_TABLE,
4549         };
4550         struct rte_flow_action actions[2];
4551         struct rte_flow *flow;
4552         struct rte_flow_error error;
4553
4554         actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4555         actions[0].conf = &jump;
4556         actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4557         flow = flow_list_create(dev, &priv->ctrl_flows,
4558                                 &attr, items, actions, false, &error);
4559         if (!flow) {
4560                 DRV_LOG(DEBUG,
4561                         "Failed to create ctrl flow: rte_errno(%d),"
4562                         " type(%d), message(%s)",
4563                         rte_errno, error.type,
4564                         error.message ? error.message : " (no stated reason)");
4565                 return -rte_errno;
4566         }
4567         return 0;
4568 }
4569
4570 /**
4571  * Enable a control flow configured from the control plane.
4572  *
4573  * @param dev
4574  *   Pointer to Ethernet device.
4575  * @param eth_spec
4576  *   An Ethernet flow spec to apply.
4577  * @param eth_mask
4578  *   An Ethernet flow mask to apply.
4579  * @param vlan_spec
4580  *   A VLAN flow spec to apply.
4581  * @param vlan_mask
4582  *   A VLAN flow mask to apply.
4583  *
4584  * @return
4585  *   0 on success, a negative errno value otherwise and rte_errno is set.
4586  */
4587 int
4588 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4589                     struct rte_flow_item_eth *eth_spec,
4590                     struct rte_flow_item_eth *eth_mask,
4591                     struct rte_flow_item_vlan *vlan_spec,
4592                     struct rte_flow_item_vlan *vlan_mask)
4593 {
4594         struct mlx5_priv *priv = dev->data->dev_private;
4595         const struct rte_flow_attr attr = {
4596                 .ingress = 1,
4597                 .priority = MLX5_FLOW_PRIO_RSVD,
4598         };
4599         struct rte_flow_item items[] = {
4600                 {
4601                         .type = RTE_FLOW_ITEM_TYPE_ETH,
4602                         .spec = eth_spec,
4603                         .last = NULL,
4604                         .mask = eth_mask,
4605                 },
4606                 {
4607                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4608                                               RTE_FLOW_ITEM_TYPE_END,
4609                         .spec = vlan_spec,
4610                         .last = NULL,
4611                         .mask = vlan_mask,
4612                 },
4613                 {
4614                         .type = RTE_FLOW_ITEM_TYPE_END,
4615                 },
4616         };
4617         uint16_t queue[priv->reta_idx_n];
4618         struct rte_flow_action_rss action_rss = {
4619                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4620                 .level = 0,
4621                 .types = priv->rss_conf.rss_hf,
4622                 .key_len = priv->rss_conf.rss_key_len,
4623                 .queue_num = priv->reta_idx_n,
4624                 .key = priv->rss_conf.rss_key,
4625                 .queue = queue,
4626         };
4627         struct rte_flow_action actions[] = {
4628                 {
4629                         .type = RTE_FLOW_ACTION_TYPE_RSS,
4630                         .conf = &action_rss,
4631                 },
4632                 {
4633                         .type = RTE_FLOW_ACTION_TYPE_END,
4634                 },
4635         };
4636         struct rte_flow *flow;
4637         struct rte_flow_error error;
4638         unsigned int i;
4639
4640         if (!priv->reta_idx_n || !priv->rxqs_n) {
4641                 return 0;
4642         }
4643         for (i = 0; i != priv->reta_idx_n; ++i)
4644                 queue[i] = (*priv->reta_idx)[i];
4645         flow = flow_list_create(dev, &priv->ctrl_flows,
4646                                 &attr, items, actions, false, &error);
4647         if (!flow)
4648                 return -rte_errno;
4649         return 0;
4650 }
4651
4652 /**
4653  * Enable a flow control configured from the control plane.
4654  *
4655  * @param dev
4656  *   Pointer to Ethernet device.
4657  * @param eth_spec
4658  *   An Ethernet flow spec to apply.
4659  * @param eth_mask
4660  *   An Ethernet flow mask to apply.
4661  *
4662  * @return
4663  *   0 on success, a negative errno value otherwise and rte_errno is set.
4664  */
4665 int
4666 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4667                struct rte_flow_item_eth *eth_spec,
4668                struct rte_flow_item_eth *eth_mask)
4669 {
4670         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4671 }
4672
4673 /**
4674  * Destroy a flow.
4675  *
4676  * @see rte_flow_destroy()
4677  * @see rte_flow_ops
4678  */
4679 int
4680 mlx5_flow_destroy(struct rte_eth_dev *dev,
4681                   struct rte_flow *flow,
4682                   struct rte_flow_error *error __rte_unused)
4683 {
4684         struct mlx5_priv *priv = dev->data->dev_private;
4685
4686         flow_list_destroy(dev, &priv->flows, flow);
4687         return 0;
4688 }
4689
4690 /**
4691  * Destroy all flows.
4692  *
4693  * @see rte_flow_flush()
4694  * @see rte_flow_ops
4695  */
4696 int
4697 mlx5_flow_flush(struct rte_eth_dev *dev,
4698                 struct rte_flow_error *error __rte_unused)
4699 {
4700         struct mlx5_priv *priv = dev->data->dev_private;
4701
4702         mlx5_flow_list_flush(dev, &priv->flows);
4703         return 0;
4704 }
4705
4706 /**
4707  * Isolated mode.
4708  *
4709  * @see rte_flow_isolate()
4710  * @see rte_flow_ops
4711  */
4712 int
4713 mlx5_flow_isolate(struct rte_eth_dev *dev,
4714                   int enable,
4715                   struct rte_flow_error *error)
4716 {
4717         struct mlx5_priv *priv = dev->data->dev_private;
4718
4719         if (dev->data->dev_started) {
4720                 rte_flow_error_set(error, EBUSY,
4721                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4722                                    NULL,
4723                                    "port must be stopped first");
4724                 return -rte_errno;
4725         }
4726         priv->isolated = !!enable;
4727         if (enable)
4728                 dev->dev_ops = &mlx5_dev_ops_isolate;
4729         else
4730                 dev->dev_ops = &mlx5_dev_ops;
4731         return 0;
4732 }
4733
4734 /**
4735  * Query a flow.
4736  *
4737  * @see rte_flow_query()
4738  * @see rte_flow_ops
4739  */
4740 static int
4741 flow_drv_query(struct rte_eth_dev *dev,
4742                struct rte_flow *flow,
4743                const struct rte_flow_action *actions,
4744                void *data,
4745                struct rte_flow_error *error)
4746 {
4747         const struct mlx5_flow_driver_ops *fops;
4748         enum mlx5_flow_drv_type ftype = flow->drv_type;
4749
4750         assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4751         fops = flow_get_drv_ops(ftype);
4752
4753         return fops->query(dev, flow, actions, data, error);
4754 }
4755
4756 /**
4757  * Query a flow.
4758  *
4759  * @see rte_flow_query()
4760  * @see rte_flow_ops
4761  */
4762 int
4763 mlx5_flow_query(struct rte_eth_dev *dev,
4764                 struct rte_flow *flow,
4765                 const struct rte_flow_action *actions,
4766                 void *data,
4767                 struct rte_flow_error *error)
4768 {
4769         int ret;
4770
4771         ret = flow_drv_query(dev, flow, actions, data, error);
4772         if (ret < 0)
4773                 return ret;
4774         return 0;
4775 }
4776
4777 /**
4778  * Convert a flow director filter to a generic flow.
4779  *
4780  * @param dev
4781  *   Pointer to Ethernet device.
4782  * @param fdir_filter
4783  *   Flow director filter to add.
4784  * @param attributes
4785  *   Generic flow parameters structure.
4786  *
4787  * @return
4788  *   0 on success, a negative errno value otherwise and rte_errno is set.
4789  */
4790 static int
4791 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4792                          const struct rte_eth_fdir_filter *fdir_filter,
4793                          struct mlx5_fdir *attributes)
4794 {
4795         struct mlx5_priv *priv = dev->data->dev_private;
4796         const struct rte_eth_fdir_input *input = &fdir_filter->input;
4797         const struct rte_eth_fdir_masks *mask =
4798                 &dev->data->dev_conf.fdir_conf.mask;
4799
4800         /* Validate queue number. */
4801         if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4802                 DRV_LOG(ERR, "port %u invalid queue number %d",
4803                         dev->data->port_id, fdir_filter->action.rx_queue);
4804                 rte_errno = EINVAL;
4805                 return -rte_errno;
4806         }
4807         attributes->attr.ingress = 1;
4808         attributes->items[0] = (struct rte_flow_item) {
4809                 .type = RTE_FLOW_ITEM_TYPE_ETH,
4810                 .spec = &attributes->l2,
4811                 .mask = &attributes->l2_mask,
4812         };
4813         switch (fdir_filter->action.behavior) {
4814         case RTE_ETH_FDIR_ACCEPT:
4815                 attributes->actions[0] = (struct rte_flow_action){
4816                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
4817                         .conf = &attributes->queue,
4818                 };
4819                 break;
4820         case RTE_ETH_FDIR_REJECT:
4821                 attributes->actions[0] = (struct rte_flow_action){
4822                         .type = RTE_FLOW_ACTION_TYPE_DROP,
4823                 };
4824                 break;
4825         default:
4826                 DRV_LOG(ERR, "port %u invalid behavior %d",
4827                         dev->data->port_id,
4828                         fdir_filter->action.behavior);
4829                 rte_errno = ENOTSUP;
4830                 return -rte_errno;
4831         }
4832         attributes->queue.index = fdir_filter->action.rx_queue;
4833         /* Handle L3. */
4834         switch (fdir_filter->input.flow_type) {
4835         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4836         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4837         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4838                 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4839                         .src_addr = input->flow.ip4_flow.src_ip,
4840                         .dst_addr = input->flow.ip4_flow.dst_ip,
4841                         .time_to_live = input->flow.ip4_flow.ttl,
4842                         .type_of_service = input->flow.ip4_flow.tos,
4843                 };
4844                 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4845                         .src_addr = mask->ipv4_mask.src_ip,
4846                         .dst_addr = mask->ipv4_mask.dst_ip,
4847                         .time_to_live = mask->ipv4_mask.ttl,
4848                         .type_of_service = mask->ipv4_mask.tos,
4849                         .next_proto_id = mask->ipv4_mask.proto,
4850                 };
4851                 attributes->items[1] = (struct rte_flow_item){
4852                         .type = RTE_FLOW_ITEM_TYPE_IPV4,
4853                         .spec = &attributes->l3,
4854                         .mask = &attributes->l3_mask,
4855                 };
4856                 break;
4857         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4858         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4859         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4860                 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4861                         .hop_limits = input->flow.ipv6_flow.hop_limits,
4862                         .proto = input->flow.ipv6_flow.proto,
4863                 };
4864
4865                 memcpy(attributes->l3.ipv6.hdr.src_addr,
4866                        input->flow.ipv6_flow.src_ip,
4867                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4868                 memcpy(attributes->l3.ipv6.hdr.dst_addr,
4869                        input->flow.ipv6_flow.dst_ip,
4870                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4871                 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4872                        mask->ipv6_mask.src_ip,
4873                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4874                 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4875                        mask->ipv6_mask.dst_ip,
4876                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4877                 attributes->items[1] = (struct rte_flow_item){
4878                         .type = RTE_FLOW_ITEM_TYPE_IPV6,
4879                         .spec = &attributes->l3,
4880                         .mask = &attributes->l3_mask,
4881                 };
4882                 break;
4883         default:
4884                 DRV_LOG(ERR, "port %u invalid flow type%d",
4885                         dev->data->port_id, fdir_filter->input.flow_type);
4886                 rte_errno = ENOTSUP;
4887                 return -rte_errno;
4888         }
4889         /* Handle L4. */
4890         switch (fdir_filter->input.flow_type) {
4891         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4892                 attributes->l4.udp.hdr = (struct rte_udp_hdr){
4893                         .src_port = input->flow.udp4_flow.src_port,
4894                         .dst_port = input->flow.udp4_flow.dst_port,
4895                 };
4896                 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4897                         .src_port = mask->src_port_mask,
4898                         .dst_port = mask->dst_port_mask,
4899                 };
4900                 attributes->items[2] = (struct rte_flow_item){
4901                         .type = RTE_FLOW_ITEM_TYPE_UDP,
4902                         .spec = &attributes->l4,
4903                         .mask = &attributes->l4_mask,
4904                 };
4905                 break;
4906         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4907                 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4908                         .src_port = input->flow.tcp4_flow.src_port,
4909                         .dst_port = input->flow.tcp4_flow.dst_port,
4910                 };
4911                 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4912                         .src_port = mask->src_port_mask,
4913                         .dst_port = mask->dst_port_mask,
4914                 };
4915                 attributes->items[2] = (struct rte_flow_item){
4916                         .type = RTE_FLOW_ITEM_TYPE_TCP,
4917                         .spec = &attributes->l4,
4918                         .mask = &attributes->l4_mask,
4919                 };
4920                 break;
4921         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4922                 attributes->l4.udp.hdr = (struct rte_udp_hdr){
4923                         .src_port = input->flow.udp6_flow.src_port,
4924                         .dst_port = input->flow.udp6_flow.dst_port,
4925                 };
4926                 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4927                         .src_port = mask->src_port_mask,
4928                         .dst_port = mask->dst_port_mask,
4929                 };
4930                 attributes->items[2] = (struct rte_flow_item){
4931                         .type = RTE_FLOW_ITEM_TYPE_UDP,
4932                         .spec = &attributes->l4,
4933                         .mask = &attributes->l4_mask,
4934                 };
4935                 break;
4936         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4937                 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4938                         .src_port = input->flow.tcp6_flow.src_port,
4939                         .dst_port = input->flow.tcp6_flow.dst_port,
4940                 };
4941                 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4942                         .src_port = mask->src_port_mask,
4943                         .dst_port = mask->dst_port_mask,
4944                 };
4945                 attributes->items[2] = (struct rte_flow_item){
4946                         .type = RTE_FLOW_ITEM_TYPE_TCP,
4947                         .spec = &attributes->l4,
4948                         .mask = &attributes->l4_mask,
4949                 };
4950                 break;
4951         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4952         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4953                 break;
4954         default:
4955                 DRV_LOG(ERR, "port %u invalid flow type%d",
4956                         dev->data->port_id, fdir_filter->input.flow_type);
4957                 rte_errno = ENOTSUP;
4958                 return -rte_errno;
4959         }
4960         return 0;
4961 }
4962
4963 #define FLOW_FDIR_CMP(f1, f2, fld) \
4964         memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4965
4966 /**
4967  * Compare two FDIR flows. If items and actions are identical, the two flows are
4968  * regarded as same.
4969  *
4970  * @param dev
4971  *   Pointer to Ethernet device.
4972  * @param f1
4973  *   FDIR flow to compare.
4974  * @param f2
4975  *   FDIR flow to compare.
4976  *
4977  * @return
4978  *   Zero on match, 1 otherwise.
4979  */
4980 static int
4981 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4982 {
4983         if (FLOW_FDIR_CMP(f1, f2, attr) ||
4984             FLOW_FDIR_CMP(f1, f2, l2) ||
4985             FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4986             FLOW_FDIR_CMP(f1, f2, l3) ||
4987             FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4988             FLOW_FDIR_CMP(f1, f2, l4) ||
4989             FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4990             FLOW_FDIR_CMP(f1, f2, actions[0].type))
4991                 return 1;
4992         if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4993             FLOW_FDIR_CMP(f1, f2, queue))
4994                 return 1;
4995         return 0;
4996 }
4997
4998 /**
4999  * Search device flow list to find out a matched FDIR flow.
5000  *
5001  * @param dev
5002  *   Pointer to Ethernet device.
5003  * @param fdir_flow
5004  *   FDIR flow to lookup.
5005  *
5006  * @return
5007  *   Pointer of flow if found, NULL otherwise.
5008  */
5009 static struct rte_flow *
5010 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5011 {
5012         struct mlx5_priv *priv = dev->data->dev_private;
5013         struct rte_flow *flow = NULL;
5014
5015         assert(fdir_flow);
5016         TAILQ_FOREACH(flow, &priv->flows, next) {
5017                 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5018                         DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5019                                 dev->data->port_id, (void *)flow);
5020                         break;
5021                 }
5022         }
5023         return flow;
5024 }
5025
5026 /**
5027  * Add new flow director filter and store it in list.
5028  *
5029  * @param dev
5030  *   Pointer to Ethernet device.
5031  * @param fdir_filter
5032  *   Flow director filter to add.
5033  *
5034  * @return
5035  *   0 on success, a negative errno value otherwise and rte_errno is set.
5036  */
5037 static int
5038 flow_fdir_filter_add(struct rte_eth_dev *dev,
5039                      const struct rte_eth_fdir_filter *fdir_filter)
5040 {
5041         struct mlx5_priv *priv = dev->data->dev_private;
5042         struct mlx5_fdir *fdir_flow;
5043         struct rte_flow *flow;
5044         int ret;
5045
5046         fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5047         if (!fdir_flow) {
5048                 rte_errno = ENOMEM;
5049                 return -rte_errno;
5050         }
5051         ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5052         if (ret)
5053                 goto error;
5054         flow = flow_fdir_filter_lookup(dev, fdir_flow);
5055         if (flow) {
5056                 rte_errno = EEXIST;
5057                 goto error;
5058         }
5059         flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5060                                 fdir_flow->items, fdir_flow->actions, true,
5061                                 NULL);
5062         if (!flow)
5063                 goto error;
5064         assert(!flow->fdir);
5065         flow->fdir = fdir_flow;
5066         DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5067                 dev->data->port_id, (void *)flow);
5068         return 0;
5069 error:
5070         rte_free(fdir_flow);
5071         return -rte_errno;
5072 }
5073
5074 /**
5075  * Delete specific filter.
5076  *
5077  * @param dev
5078  *   Pointer to Ethernet device.
5079  * @param fdir_filter
5080  *   Filter to be deleted.
5081  *
5082  * @return
5083  *   0 on success, a negative errno value otherwise and rte_errno is set.
5084  */
5085 static int
5086 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5087                         const struct rte_eth_fdir_filter *fdir_filter)
5088 {
5089         struct mlx5_priv *priv = dev->data->dev_private;
5090         struct rte_flow *flow;
5091         struct mlx5_fdir fdir_flow = {
5092                 .attr.group = 0,
5093         };
5094         int ret;
5095
5096         ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5097         if (ret)
5098                 return -rte_errno;
5099         flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5100         if (!flow) {
5101                 rte_errno = ENOENT;
5102                 return -rte_errno;
5103         }
5104         flow_list_destroy(dev, &priv->flows, flow);
5105         DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5106                 dev->data->port_id, (void *)flow);
5107         return 0;
5108 }
5109
5110 /**
5111  * Update queue for specific filter.
5112  *
5113  * @param dev
5114  *   Pointer to Ethernet device.
5115  * @param fdir_filter
5116  *   Filter to be updated.
5117  *
5118  * @return
5119  *   0 on success, a negative errno value otherwise and rte_errno is set.
5120  */
5121 static int
5122 flow_fdir_filter_update(struct rte_eth_dev *dev,
5123                         const struct rte_eth_fdir_filter *fdir_filter)
5124 {
5125         int ret;
5126
5127         ret = flow_fdir_filter_delete(dev, fdir_filter);
5128         if (ret)
5129                 return ret;
5130         return flow_fdir_filter_add(dev, fdir_filter);
5131 }
5132
5133 /**
5134  * Flush all filters.
5135  *
5136  * @param dev
5137  *   Pointer to Ethernet device.
5138  */
5139 static void
5140 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5141 {
5142         struct mlx5_priv *priv = dev->data->dev_private;
5143
5144         mlx5_flow_list_flush(dev, &priv->flows);
5145 }
5146
5147 /**
5148  * Get flow director information.
5149  *
5150  * @param dev
5151  *   Pointer to Ethernet device.
5152  * @param[out] fdir_info
5153  *   Resulting flow director information.
5154  */
5155 static void
5156 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5157 {
5158         struct rte_eth_fdir_masks *mask =
5159                 &dev->data->dev_conf.fdir_conf.mask;
5160
5161         fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5162         fdir_info->guarant_spc = 0;
5163         rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5164         fdir_info->max_flexpayload = 0;
5165         fdir_info->flow_types_mask[0] = 0;
5166         fdir_info->flex_payload_unit = 0;
5167         fdir_info->max_flex_payload_segment_num = 0;
5168         fdir_info->flex_payload_limit = 0;
5169         memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5170 }
5171
5172 /**
5173  * Deal with flow director operations.
5174  *
5175  * @param dev
5176  *   Pointer to Ethernet device.
5177  * @param filter_op
5178  *   Operation to perform.
5179  * @param arg
5180  *   Pointer to operation-specific structure.
5181  *
5182  * @return
5183  *   0 on success, a negative errno value otherwise and rte_errno is set.
5184  */
5185 static int
5186 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5187                     void *arg)
5188 {
5189         enum rte_fdir_mode fdir_mode =
5190                 dev->data->dev_conf.fdir_conf.mode;
5191
5192         if (filter_op == RTE_ETH_FILTER_NOP)
5193                 return 0;
5194         if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5195             fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5196                 DRV_LOG(ERR, "port %u flow director mode %d not supported",
5197                         dev->data->port_id, fdir_mode);
5198                 rte_errno = EINVAL;
5199                 return -rte_errno;
5200         }
5201         switch (filter_op) {
5202         case RTE_ETH_FILTER_ADD:
5203                 return flow_fdir_filter_add(dev, arg);
5204         case RTE_ETH_FILTER_UPDATE:
5205                 return flow_fdir_filter_update(dev, arg);
5206         case RTE_ETH_FILTER_DELETE:
5207                 return flow_fdir_filter_delete(dev, arg);
5208         case RTE_ETH_FILTER_FLUSH:
5209                 flow_fdir_filter_flush(dev);
5210                 break;
5211         case RTE_ETH_FILTER_INFO:
5212                 flow_fdir_info_get(dev, arg);
5213                 break;
5214         default:
5215                 DRV_LOG(DEBUG, "port %u unknown operation %u",
5216                         dev->data->port_id, filter_op);
5217                 rte_errno = EINVAL;
5218                 return -rte_errno;
5219         }
5220         return 0;
5221 }
5222
5223 /**
5224  * Manage filter operations.
5225  *
5226  * @param dev
5227  *   Pointer to Ethernet device structure.
5228  * @param filter_type
5229  *   Filter type.
5230  * @param filter_op
5231  *   Operation to perform.
5232  * @param arg
5233  *   Pointer to operation-specific structure.
5234  *
5235  * @return
5236  *   0 on success, a negative errno value otherwise and rte_errno is set.
5237  */
5238 int
5239 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5240                      enum rte_filter_type filter_type,
5241                      enum rte_filter_op filter_op,
5242                      void *arg)
5243 {
5244         switch (filter_type) {
5245         case RTE_ETH_FILTER_GENERIC:
5246                 if (filter_op != RTE_ETH_FILTER_GET) {
5247                         rte_errno = EINVAL;
5248                         return -rte_errno;
5249                 }
5250                 *(const void **)arg = &mlx5_flow_ops;
5251                 return 0;
5252         case RTE_ETH_FILTER_FDIR:
5253                 return flow_fdir_ctrl_func(dev, filter_op, arg);
5254         default:
5255                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
5256                         dev->data->port_id, filter_type);
5257                 rte_errno = ENOTSUP;
5258                 return -rte_errno;
5259         }
5260         return 0;
5261 }
5262
5263 /**
5264  * Create the needed meter and suffix tables.
5265  *
5266  * @param[in] dev
5267  *   Pointer to Ethernet device.
5268  * @param[in] fm
5269  *   Pointer to the flow meter.
5270  *
5271  * @return
5272  *   Pointer to table set on success, NULL otherwise.
5273  */
5274 struct mlx5_meter_domains_infos *
5275 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5276                           const struct mlx5_flow_meter *fm)
5277 {
5278         const struct mlx5_flow_driver_ops *fops;
5279
5280         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5281         return fops->create_mtr_tbls(dev, fm);
5282 }
5283
5284 /**
5285  * Destroy the meter table set.
5286  *
5287  * @param[in] dev
5288  *   Pointer to Ethernet device.
5289  * @param[in] tbl
5290  *   Pointer to the meter table set.
5291  *
5292  * @return
5293  *   0 on success.
5294  */
5295 int
5296 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5297                            struct mlx5_meter_domains_infos *tbls)
5298 {
5299         const struct mlx5_flow_driver_ops *fops;
5300
5301         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5302         return fops->destroy_mtr_tbls(dev, tbls);
5303 }
5304
5305 /**
5306  * Create policer rules.
5307  *
5308  * @param[in] dev
5309  *   Pointer to Ethernet device.
5310  * @param[in] fm
5311  *   Pointer to flow meter structure.
5312  * @param[in] attr
5313  *   Pointer to flow attributes.
5314  *
5315  * @return
5316  *   0 on success, -1 otherwise.
5317  */
5318 int
5319 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5320                                struct mlx5_flow_meter *fm,
5321                                const struct rte_flow_attr *attr)
5322 {
5323         const struct mlx5_flow_driver_ops *fops;
5324
5325         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5326         return fops->create_policer_rules(dev, fm, attr);
5327 }
5328
5329 /**
5330  * Destroy policer rules.
5331  *
5332  * @param[in] fm
5333  *   Pointer to flow meter structure.
5334  * @param[in] attr
5335  *   Pointer to flow attributes.
5336  *
5337  * @return
5338  *   0 on success, -1 otherwise.
5339  */
5340 int
5341 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5342                                 struct mlx5_flow_meter *fm,
5343                                 const struct rte_flow_attr *attr)
5344 {
5345         const struct mlx5_flow_driver_ops *fops;
5346
5347         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5348         return fops->destroy_policer_rules(dev, fm, attr);
5349 }
5350
5351 /**
5352  * Allocate a counter.
5353  *
5354  * @param[in] dev
5355  *   Pointer to Ethernet device structure.
5356  *
5357  * @return
5358  *   Pointer to allocated counter  on success, NULL otherwise.
5359  */
5360 struct mlx5_flow_counter *
5361 mlx5_counter_alloc(struct rte_eth_dev *dev)
5362 {
5363         const struct mlx5_flow_driver_ops *fops;
5364         struct rte_flow_attr attr = { .transfer = 0 };
5365
5366         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5367                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5368                 return fops->counter_alloc(dev);
5369         }
5370         DRV_LOG(ERR,
5371                 "port %u counter allocate is not supported.",
5372                  dev->data->port_id);
5373         return NULL;
5374 }
5375
5376 /**
5377  * Free a counter.
5378  *
5379  * @param[in] dev
5380  *   Pointer to Ethernet device structure.
5381  * @param[in] cnt
5382  *   Pointer to counter to be free.
5383  */
5384 void
5385 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5386 {
5387         const struct mlx5_flow_driver_ops *fops;
5388         struct rte_flow_attr attr = { .transfer = 0 };
5389
5390         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5391                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5392                 fops->counter_free(dev, cnt);
5393                 return;
5394         }
5395         DRV_LOG(ERR,
5396                 "port %u counter free is not supported.",
5397                  dev->data->port_id);
5398 }
5399
5400 /**
5401  * Query counter statistics.
5402  *
5403  * @param[in] dev
5404  *   Pointer to Ethernet device structure.
5405  * @param[in] cnt
5406  *   Pointer to counter to query.
5407  * @param[in] clear
5408  *   Set to clear counter statistics.
5409  * @param[out] pkts
5410  *   The counter hits packets number to save.
5411  * @param[out] bytes
5412  *   The counter hits bytes number to save.
5413  *
5414  * @return
5415  *   0 on success, a negative errno value otherwise.
5416  */
5417 int
5418 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5419                    bool clear, uint64_t *pkts, uint64_t *bytes)
5420 {
5421         const struct mlx5_flow_driver_ops *fops;
5422         struct rte_flow_attr attr = { .transfer = 0 };
5423
5424         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5425                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5426                 return fops->counter_query(dev, cnt, clear, pkts, bytes);
5427         }
5428         DRV_LOG(ERR,
5429                 "port %u counter query is not supported.",
5430                  dev->data->port_id);
5431         return -ENOTSUP;
5432 }
5433
5434 #define MLX5_POOL_QUERY_FREQ_US 1000000
5435
5436 /**
5437  * Set the periodic procedure for triggering asynchronous batch queries for all
5438  * the counter pools.
5439  *
5440  * @param[in] sh
5441  *   Pointer to mlx5_ibv_shared object.
5442  */
5443 void
5444 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5445 {
5446         struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5447         uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5448         uint32_t us;
5449
5450         cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5451         pools_n += rte_atomic16_read(&cont->n_valid);
5452         us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5453         DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5454         if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5455                 sh->cmng.query_thread_on = 0;
5456                 DRV_LOG(ERR, "Cannot reinitialize query alarm");
5457         } else {
5458                 sh->cmng.query_thread_on = 1;
5459         }
5460 }
5461
5462 /**
5463  * The periodic procedure for triggering asynchronous batch queries for all the
5464  * counter pools. This function is probably called by the host thread.
5465  *
5466  * @param[in] arg
5467  *   The parameter for the alarm process.
5468  */
5469 void
5470 mlx5_flow_query_alarm(void *arg)
5471 {
5472         struct mlx5_ibv_shared *sh = arg;
5473         struct mlx5_devx_obj *dcs;
5474         uint16_t offset;
5475         int ret;
5476         uint8_t batch = sh->cmng.batch;
5477         uint16_t pool_index = sh->cmng.pool_index;
5478         struct mlx5_pools_container *cont;
5479         struct mlx5_pools_container *mcont;
5480         struct mlx5_flow_counter_pool *pool;
5481
5482         if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5483                 goto set_alarm;
5484 next_container:
5485         cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5486         mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5487         /* Check if resize was done and need to flip a container. */
5488         if (cont != mcont) {
5489                 if (cont->pools) {
5490                         /* Clean the old container. */
5491                         rte_free(cont->pools);
5492                         memset(cont, 0, sizeof(*cont));
5493                 }
5494                 rte_cio_wmb();
5495                  /* Flip the host container. */
5496                 sh->cmng.mhi[batch] ^= (uint8_t)2;
5497                 cont = mcont;
5498         }
5499         if (!cont->pools) {
5500                 /* 2 empty containers case is unexpected. */
5501                 if (unlikely(batch != sh->cmng.batch))
5502                         goto set_alarm;
5503                 batch ^= 0x1;
5504                 pool_index = 0;
5505                 goto next_container;
5506         }
5507         pool = cont->pools[pool_index];
5508         if (pool->raw_hw)
5509                 /* There is a pool query in progress. */
5510                 goto set_alarm;
5511         pool->raw_hw =
5512                 LIST_FIRST(&sh->cmng.free_stat_raws);
5513         if (!pool->raw_hw)
5514                 /* No free counter statistics raw memory. */
5515                 goto set_alarm;
5516         dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5517                                                               (&pool->a64_dcs);
5518         offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5519         ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5520                                                offset, NULL, NULL,
5521                                                pool->raw_hw->mem_mng->dm->id,
5522                                                (void *)(uintptr_t)
5523                                                (pool->raw_hw->data + offset),
5524                                                sh->devx_comp,
5525                                                (uint64_t)(uintptr_t)pool);
5526         if (ret) {
5527                 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5528                         " %d", pool->min_dcs->id);
5529                 pool->raw_hw = NULL;
5530                 goto set_alarm;
5531         }
5532         pool->raw_hw->min_dcs_id = dcs->id;
5533         LIST_REMOVE(pool->raw_hw, next);
5534         sh->cmng.pending_queries++;
5535         pool_index++;
5536         if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5537                 batch ^= 0x1;
5538                 pool_index = 0;
5539         }
5540 set_alarm:
5541         sh->cmng.batch = batch;
5542         sh->cmng.pool_index = pool_index;
5543         mlx5_set_query_alarm(sh);
5544 }
5545
5546 /**
5547  * Handler for the HW respond about ready values from an asynchronous batch
5548  * query. This function is probably called by the host thread.
5549  *
5550  * @param[in] sh
5551  *   The pointer to the shared IB device context.
5552  * @param[in] async_id
5553  *   The Devx async ID.
5554  * @param[in] status
5555  *   The status of the completion.
5556  */
5557 void
5558 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5559                                   uint64_t async_id, int status)
5560 {
5561         struct mlx5_flow_counter_pool *pool =
5562                 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5563         struct mlx5_counter_stats_raw *raw_to_free;
5564
5565         if (unlikely(status)) {
5566                 raw_to_free = pool->raw_hw;
5567         } else {
5568                 raw_to_free = pool->raw;
5569                 rte_spinlock_lock(&pool->sl);
5570                 pool->raw = pool->raw_hw;
5571                 rte_spinlock_unlock(&pool->sl);
5572                 rte_atomic64_add(&pool->query_gen, 1);
5573                 /* Be sure the new raw counters data is updated in memory. */
5574                 rte_cio_wmb();
5575         }
5576         LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5577         pool->raw_hw = NULL;
5578         sh->cmng.pending_queries--;
5579 }
5580
5581 /**
5582  * Translate the rte_flow group index to HW table value.
5583  *
5584  * @param[in] attributes
5585  *   Pointer to flow attributes
5586  * @param[in] external
5587  *   Value is part of flow rule created by request external to PMD.
5588  * @param[in] group
5589  *   rte_flow group index value.
5590  * @param[out] table
5591  *   HW table value.
5592  * @param[out] error
5593  *   Pointer to error structure.
5594  *
5595  * @return
5596  *   0 on success, a negative errno value otherwise and rte_errno is set.
5597  */
5598 int
5599 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5600                          uint32_t group, uint32_t *table,
5601                          struct rte_flow_error *error)
5602 {
5603         if (attributes->transfer && external) {
5604                 if (group == UINT32_MAX)
5605                         return rte_flow_error_set
5606                                                 (error, EINVAL,
5607                                                  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5608                                                  NULL,
5609                                                  "group index not supported");
5610                 *table = group + 1;
5611         } else {
5612                 *table = group;
5613         }
5614         return 0;
5615 }
5616
5617 /**
5618  * Discover availability of metadata reg_c's.
5619  *
5620  * Iteratively use test flows to check availability.
5621  *
5622  * @param[in] dev
5623  *   Pointer to the Ethernet device structure.
5624  *
5625  * @return
5626  *   0 on success, a negative errno value otherwise and rte_errno is set.
5627  */
5628 int
5629 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5630 {
5631         struct mlx5_priv *priv = dev->data->dev_private;
5632         struct mlx5_dev_config *config = &priv->config;
5633         enum modify_reg idx;
5634         int n = 0;
5635
5636         /* reg_c[0] and reg_c[1] are reserved. */
5637         config->flow_mreg_c[n++] = REG_C_0;
5638         config->flow_mreg_c[n++] = REG_C_1;
5639         /* Discover availability of other reg_c's. */
5640         for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5641                 struct rte_flow_attr attr = {
5642                         .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5643                         .priority = MLX5_FLOW_PRIO_RSVD,
5644                         .ingress = 1,
5645                 };
5646                 struct rte_flow_item items[] = {
5647                         [0] = {
5648                                 .type = RTE_FLOW_ITEM_TYPE_END,
5649                         },
5650                 };
5651                 struct rte_flow_action actions[] = {
5652                         [0] = {
5653                                 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5654                                 .conf = &(struct mlx5_flow_action_copy_mreg){
5655                                         .src = REG_C_1,
5656                                         .dst = idx,
5657                                 },
5658                         },
5659                         [1] = {
5660                                 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5661                                 .conf = &(struct rte_flow_action_jump){
5662                                         .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5663                                 },
5664                         },
5665                         [2] = {
5666                                 .type = RTE_FLOW_ACTION_TYPE_END,
5667                         },
5668                 };
5669                 struct rte_flow *flow;
5670                 struct rte_flow_error error;
5671
5672                 if (!config->dv_flow_en)
5673                         break;
5674                 /* Create internal flow, validation skips copy action. */
5675                 flow = flow_list_create(dev, NULL, &attr, items,
5676                                         actions, false, &error);
5677                 if (!flow)
5678                         continue;
5679                 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5680                         config->flow_mreg_c[n++] = idx;
5681                 flow_list_destroy(dev, NULL, flow);
5682         }
5683         for (; n < MLX5_MREG_C_NUM; ++n)
5684                 config->flow_mreg_c[n] = REG_NONE;
5685         return 0;
5686 }
5687
5688 /**
5689  * Dump flow raw hw data to file
5690  *
5691  * @param[in] dev
5692  *    The pointer to Ethernet device.
5693  * @param[in] file
5694  *   A pointer to a file for output.
5695  * @param[out] error
5696  *   Perform verbose error reporting if not NULL. PMDs initialize this
5697  *   structure in case of error only.
5698  * @return
5699  *   0 on success, a nagative value otherwise.
5700  */
5701 int
5702 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5703                    FILE *file,
5704                    struct rte_flow_error *error __rte_unused)
5705 {
5706         struct mlx5_priv *priv = dev->data->dev_private;
5707
5708         return mlx5_devx_cmd_flow_dump(priv->sh, file);
5709 }