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