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 #include <stdbool.h>
  12
  13 /* Verbs header. */
  14 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
  15 #ifdef PEDANTIC
  16 #pragma GCC diagnostic ignored "-Wpedantic"
  17 #endif
  18 #include <infiniband/verbs.h>
  19 #ifdef PEDANTIC
  20 #pragma GCC diagnostic error "-Wpedantic"
  21 #endif
  22
  23 #include <rte_common.h>
  24 #include <rte_ether.h>
  25 #include <rte_ethdev_driver.h>
  26 #include <rte_flow.h>
  27 #include <rte_flow_driver.h>
  28 #include <rte_malloc.h>
  29 #include <rte_ip.h>
  30
  31 #include <mlx5_glue.h>
  32 #include <mlx5_devx_cmds.h>
  33 #include <mlx5_prm.h>
  34
  35 #include "mlx5_defs.h"
  36 #include "mlx5.h"
  37 #include "mlx5_flow.h"
  38 #include "mlx5_rxtx.h"
  39
  40 /* Dev ops structure defined in mlx5.c */
  41 extern const struct eth_dev_ops mlx5_dev_ops;
  42 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
  43
  44 /** Device flow drivers. */
  45 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
  46 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
  47 #endif
  48 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
  49
  50 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
  51
  52 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
  53         [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
  54 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
  55         [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
  56 #endif
  57         [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
  58         [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
  59 };
  60
  61 enum mlx5_expansion {
  62         MLX5_EXPANSION_ROOT,
  63         MLX5_EXPANSION_ROOT_OUTER,
  64         MLX5_EXPANSION_ROOT_ETH_VLAN,
  65         MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
  66         MLX5_EXPANSION_OUTER_ETH,
  67         MLX5_EXPANSION_OUTER_ETH_VLAN,
  68         MLX5_EXPANSION_OUTER_VLAN,
  69         MLX5_EXPANSION_OUTER_IPV4,
  70         MLX5_EXPANSION_OUTER_IPV4_UDP,
  71         MLX5_EXPANSION_OUTER_IPV4_TCP,
  72         MLX5_EXPANSION_OUTER_IPV6,
  73         MLX5_EXPANSION_OUTER_IPV6_UDP,
  74         MLX5_EXPANSION_OUTER_IPV6_TCP,
  75         MLX5_EXPANSION_VXLAN,
  76         MLX5_EXPANSION_VXLAN_GPE,
  77         MLX5_EXPANSION_GRE,
  78         MLX5_EXPANSION_MPLS,
  79         MLX5_EXPANSION_ETH,
  80         MLX5_EXPANSION_ETH_VLAN,
  81         MLX5_EXPANSION_VLAN,
  82         MLX5_EXPANSION_IPV4,
  83         MLX5_EXPANSION_IPV4_UDP,
  84         MLX5_EXPANSION_IPV4_TCP,
  85         MLX5_EXPANSION_IPV6,
  86         MLX5_EXPANSION_IPV6_UDP,
  87         MLX5_EXPANSION_IPV6_TCP,
  88 };
  89
  90 /** Supported expansion of items. */
  91 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
  92         [MLX5_EXPANSION_ROOT] = {
  93                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
  94                                                  MLX5_EXPANSION_IPV4,
  95                                                  MLX5_EXPANSION_IPV6),
  96                 .type = RTE_FLOW_ITEM_TYPE_END,
  97         },
  98         [MLX5_EXPANSION_ROOT_OUTER] = {
  99                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
 100                                                  MLX5_EXPANSION_OUTER_IPV4,
 101                                                  MLX5_EXPANSION_OUTER_IPV6),
 102                 .type = RTE_FLOW_ITEM_TYPE_END,
 103         },
 104         [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
 105                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
 106                 .type = RTE_FLOW_ITEM_TYPE_END,
 107         },
 108         [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
 109                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
 110                 .type = RTE_FLOW_ITEM_TYPE_END,
 111         },
 112         [MLX5_EXPANSION_OUTER_ETH] = {
 113                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
 114                                                  MLX5_EXPANSION_OUTER_IPV6,
 115                                                  MLX5_EXPANSION_MPLS),
 116                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 117                 .rss_types = 0,
 118         },
 119         [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
 120                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
 121                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 122                 .rss_types = 0,
 123         },
 124         [MLX5_EXPANSION_OUTER_VLAN] = {
 125                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
 126                                                  MLX5_EXPANSION_OUTER_IPV6),
 127                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
 128         },
 129         [MLX5_EXPANSION_OUTER_IPV4] = {
 130                 .next = RTE_FLOW_EXPAND_RSS_NEXT
 131                         (MLX5_EXPANSION_OUTER_IPV4_UDP,
 132                          MLX5_EXPANSION_OUTER_IPV4_TCP,
 133                          MLX5_EXPANSION_GRE,
 134                          MLX5_EXPANSION_IPV4,
 135                          MLX5_EXPANSION_IPV6),
 136                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
 137                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
 138                         ETH_RSS_NONFRAG_IPV4_OTHER,
 139         },
 140         [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
 141                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
 142                                                  MLX5_EXPANSION_VXLAN_GPE),
 143                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 144                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
 145         },
 146         [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
 147                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 148                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
 149         },
 150         [MLX5_EXPANSION_OUTER_IPV6] = {
 151                 .next = RTE_FLOW_EXPAND_RSS_NEXT
 152                         (MLX5_EXPANSION_OUTER_IPV6_UDP,
 153                          MLX5_EXPANSION_OUTER_IPV6_TCP,
 154                          MLX5_EXPANSION_IPV4,
 155                          MLX5_EXPANSION_IPV6),
 156                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
 157                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
 158                         ETH_RSS_NONFRAG_IPV6_OTHER,
 159         },
 160         [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
 161                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
 162                                                  MLX5_EXPANSION_VXLAN_GPE),
 163                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 164                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
 165         },
 166         [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
 167                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 168                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
 169         },
 170         [MLX5_EXPANSION_VXLAN] = {
 171                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
 172                                                  MLX5_EXPANSION_IPV4,
 173                                                  MLX5_EXPANSION_IPV6),
 174                 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
 175         },
 176         [MLX5_EXPANSION_VXLAN_GPE] = {
 177                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
 178                                                  MLX5_EXPANSION_IPV4,
 179                                                  MLX5_EXPANSION_IPV6),
 180                 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
 181         },
 182         [MLX5_EXPANSION_GRE] = {
 183                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
 184                 .type = RTE_FLOW_ITEM_TYPE_GRE,
 185         },
 186         [MLX5_EXPANSION_MPLS] = {
 187                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 188                                                  MLX5_EXPANSION_IPV6),
 189                 .type = RTE_FLOW_ITEM_TYPE_MPLS,
 190         },
 191         [MLX5_EXPANSION_ETH] = {
 192                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 193                                                  MLX5_EXPANSION_IPV6),
 194                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 195         },
 196         [MLX5_EXPANSION_ETH_VLAN] = {
 197                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
 198                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 199         },
 200         [MLX5_EXPANSION_VLAN] = {
 201                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 202                                                  MLX5_EXPANSION_IPV6),
 203                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
 204         },
 205         [MLX5_EXPANSION_IPV4] = {
 206                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
 207                                                  MLX5_EXPANSION_IPV4_TCP),
 208                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
 209                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
 210                         ETH_RSS_NONFRAG_IPV4_OTHER,
 211         },
 212         [MLX5_EXPANSION_IPV4_UDP] = {
 213                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 214                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
 215         },
 216         [MLX5_EXPANSION_IPV4_TCP] = {
 217                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 218                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
 219         },
 220         [MLX5_EXPANSION_IPV6] = {
 221                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
 222                                                  MLX5_EXPANSION_IPV6_TCP),
 223                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
 224                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
 225                         ETH_RSS_NONFRAG_IPV6_OTHER,
 226         },
 227         [MLX5_EXPANSION_IPV6_UDP] = {
 228                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 229                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
 230         },
 231         [MLX5_EXPANSION_IPV6_TCP] = {
 232                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 233                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
 234         },
 235 };
 236
 237 static const struct rte_flow_ops mlx5_flow_ops = {
 238         .validate = mlx5_flow_validate,
 239         .create = mlx5_flow_create,
 240         .destroy = mlx5_flow_destroy,
 241         .flush = mlx5_flow_flush,
 242         .isolate = mlx5_flow_isolate,
 243         .query = mlx5_flow_query,
 244         .dev_dump = mlx5_flow_dev_dump,
 245 };
 246
 247 /* Convert FDIR request to Generic flow. */
 248 struct mlx5_fdir {
 249         struct rte_flow_attr attr;
 250         struct rte_flow_item items[4];
 251         struct rte_flow_item_eth l2;
 252         struct rte_flow_item_eth l2_mask;
 253         union {
 254                 struct rte_flow_item_ipv4 ipv4;
 255                 struct rte_flow_item_ipv6 ipv6;
 256         } l3;
 257         union {
 258                 struct rte_flow_item_ipv4 ipv4;
 259                 struct rte_flow_item_ipv6 ipv6;
 260         } l3_mask;
 261         union {
 262                 struct rte_flow_item_udp udp;
 263                 struct rte_flow_item_tcp tcp;
 264         } l4;
 265         union {
 266                 struct rte_flow_item_udp udp;
 267                 struct rte_flow_item_tcp tcp;
 268         } l4_mask;
 269         struct rte_flow_action actions[2];
 270         struct rte_flow_action_queue queue;
 271 };
 272
 273 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
 274 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
 275         { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
 276 };
 277
 278 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
 279 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
 280         { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
 281         { 9, 10, 11 }, { 12, 13, 14 },
 282 };
 283
 284 /* Tunnel information. */
 285 struct mlx5_flow_tunnel_info {
 286         uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
 287         uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
 288 };
 289
 290 static struct mlx5_flow_tunnel_info tunnels_info[] = {
 291         {
 292                 .tunnel = MLX5_FLOW_LAYER_VXLAN,
 293                 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
 294         },
 295         {
 296                 .tunnel = MLX5_FLOW_LAYER_GENEVE,
 297                 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
 298         },
 299         {
 300                 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
 301                 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
 302         },
 303         {
 304                 .tunnel = MLX5_FLOW_LAYER_GRE,
 305                 .ptype = RTE_PTYPE_TUNNEL_GRE,
 306         },
 307         {
 308                 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
 309                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
 310         },
 311         {
 312                 .tunnel = MLX5_FLOW_LAYER_MPLS,
 313                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
 314         },
 315         {
 316                 .tunnel = MLX5_FLOW_LAYER_NVGRE,
 317                 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
 318         },
 319         {
 320                 .tunnel = MLX5_FLOW_LAYER_IPIP,
 321                 .ptype = RTE_PTYPE_TUNNEL_IP,
 322         },
 323         {
 324                 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
 325                 .ptype = RTE_PTYPE_TUNNEL_IP,
 326         },
 327         {
 328                 .tunnel = MLX5_FLOW_LAYER_GTP,
 329                 .ptype = RTE_PTYPE_TUNNEL_GTPU,
 330         },
 331 };
 332
 333 /**
 334  * Translate tag ID to register.
 335  *
 336  * @param[in] dev
 337  *   Pointer to the Ethernet device structure.
 338  * @param[in] feature
 339  *   The feature that request the register.
 340  * @param[in] id
 341  *   The request register ID.
 342  * @param[out] error
 343  *   Error description in case of any.
 344  *
 345  * @return
 346  *   The request register on success, a negative errno
 347  *   value otherwise and rte_errno is set.
 348  */
 349 int
 350 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
 351                      enum mlx5_feature_name feature,
 352                      uint32_t id,
 353                      struct rte_flow_error *error)
 354 {
 355         struct mlx5_priv *priv = dev->data->dev_private;
 356         struct mlx5_dev_config *config = &priv->config;
 357         enum modify_reg start_reg;
 358         bool skip_mtr_reg = false;
 359
 360         switch (feature) {
 361         case MLX5_HAIRPIN_RX:
 362                 return REG_B;
 363         case MLX5_HAIRPIN_TX:
 364                 return REG_A;
 365         case MLX5_METADATA_RX:
 366                 switch (config->dv_xmeta_en) {
 367                 case MLX5_XMETA_MODE_LEGACY:
 368                         return REG_B;
 369                 case MLX5_XMETA_MODE_META16:
 370                         return REG_C_0;
 371                 case MLX5_XMETA_MODE_META32:
 372                         return REG_C_1;
 373                 }
 374                 break;
 375         case MLX5_METADATA_TX:
 376                 return REG_A;
 377         case MLX5_METADATA_FDB:
 378                 switch (config->dv_xmeta_en) {
 379                 case MLX5_XMETA_MODE_LEGACY:
 380                         return REG_NONE;
 381                 case MLX5_XMETA_MODE_META16:
 382                         return REG_C_0;
 383                 case MLX5_XMETA_MODE_META32:
 384                         return REG_C_1;
 385                 }
 386                 break;
 387         case MLX5_FLOW_MARK:
 388                 switch (config->dv_xmeta_en) {
 389                 case MLX5_XMETA_MODE_LEGACY:
 390                         return REG_NONE;
 391                 case MLX5_XMETA_MODE_META16:
 392                         return REG_C_1;
 393                 case MLX5_XMETA_MODE_META32:
 394                         return REG_C_0;
 395                 }
 396                 break;
 397         case MLX5_MTR_SFX:
 398                 /*
 399                  * If meter color and flow match share one register, flow match
 400                  * should use the meter color register for match.
 401                  */
 402                 if (priv->mtr_reg_share)
 403                         return priv->mtr_color_reg;
 404                 else
 405                         return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
 406                                REG_C_3;
 407         case MLX5_MTR_COLOR:
 408                 MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
 409                 return priv->mtr_color_reg;
 410         case MLX5_COPY_MARK:
 411                 /*
 412                  * Metadata COPY_MARK register using is in meter suffix sub
 413                  * flow while with meter. It's safe to share the same register.
 414                  */
 415                 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
 416         case MLX5_APP_TAG:
 417                 /*
 418                  * If meter is enable, it will engage the register for color
 419                  * match and flow match. If meter color match is not using the
 420                  * REG_C_2, need to skip the REG_C_x be used by meter color
 421                  * match.
 422                  * If meter is disable, free to use all available registers.
 423                  */
 424                 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
 425                             (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
 426                 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
 427                 if (id > (REG_C_7 - start_reg))
 428                         return rte_flow_error_set(error, EINVAL,
 429                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 430                                                   NULL, "invalid tag id");
 431                 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
 432                         return rte_flow_error_set(error, ENOTSUP,
 433                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 434                                                   NULL, "unsupported tag id");
 435                 /*
 436                  * This case means meter is using the REG_C_x great than 2.
 437                  * Take care not to conflict with meter color REG_C_x.
 438                  * If the available index REG_C_y >= REG_C_x, skip the
 439                  * color register.
 440                  */
 441                 if (skip_mtr_reg && config->flow_mreg_c
 442                     [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
 443                         if (config->flow_mreg_c
 444                             [id + 1 + start_reg - REG_C_0] != REG_NONE)
 445                                 return config->flow_mreg_c
 446                                                [id + 1 + start_reg - REG_C_0];
 447                         return rte_flow_error_set(error, ENOTSUP,
 448                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 449                                                   NULL, "unsupported tag id");
 450                 }
 451                 return config->flow_mreg_c[id + start_reg - REG_C_0];
 452         }
 453         MLX5_ASSERT(false);
 454         return rte_flow_error_set(error, EINVAL,
 455                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
 456                                   NULL, "invalid feature name");
 457 }
 458
 459 /**
 460  * Check extensive flow metadata register support.
 461  *
 462  * @param dev
 463  *   Pointer to rte_eth_dev structure.
 464  *
 465  * @return
 466  *   True if device supports extensive flow metadata register, otherwise false.
 467  */
 468 bool
 469 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
 470 {
 471         struct mlx5_priv *priv = dev->data->dev_private;
 472         struct mlx5_dev_config *config = &priv->config;
 473
 474         /*
 475          * Having available reg_c can be regarded inclusively as supporting
 476          * extensive flow metadata register, which could mean,
 477          * - metadata register copy action by modify header.
 478          * - 16 modify header actions is supported.
 479          * - reg_c's are preserved across different domain (FDB and NIC) on
 480          *   packet loopback by flow lookup miss.
 481          */
 482         return config->flow_mreg_c[2] != REG_NONE;
 483 }
 484
 485 /**
 486  * Discover the maximum number of priority available.
 487  *
 488  * @param[in] dev
 489  *   Pointer to the Ethernet device structure.
 490  *
 491  * @return
 492  *   number of supported flow priority on success, a negative errno
 493  *   value otherwise and rte_errno is set.
 494  */
 495 int
 496 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 497 {
 498         struct mlx5_priv *priv = dev->data->dev_private;
 499         struct {
 500                 struct ibv_flow_attr attr;
 501                 struct ibv_flow_spec_eth eth;
 502                 struct ibv_flow_spec_action_drop drop;
 503         } flow_attr = {
 504                 .attr = {
 505                         .num_of_specs = 2,
 506                         .port = (uint8_t)priv->ibv_port,
 507                 },
 508                 .eth = {
 509                         .type = IBV_FLOW_SPEC_ETH,
 510                         .size = sizeof(struct ibv_flow_spec_eth),
 511                 },
 512                 .drop = {
 513                         .size = sizeof(struct ibv_flow_spec_action_drop),
 514                         .type = IBV_FLOW_SPEC_ACTION_DROP,
 515                 },
 516         };
 517         struct ibv_flow *flow;
 518         struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
 519         uint16_t vprio[] = { 8, 16 };
 520         int i;
 521         int priority = 0;
 522
 523         if (!drop) {
 524                 rte_errno = ENOTSUP;
 525                 return -rte_errno;
 526         }
 527         for (i = 0; i != RTE_DIM(vprio); i++) {
 528                 flow_attr.attr.priority = vprio[i] - 1;
 529                 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 530                 if (!flow)
 531                         break;
 532                 claim_zero(mlx5_glue->destroy_flow(flow));
 533                 priority = vprio[i];
 534         }
 535         mlx5_hrxq_drop_release(dev);
 536         switch (priority) {
 537         case 8:
 538                 priority = RTE_DIM(priority_map_3);
 539                 break;
 540         case 16:
 541                 priority = RTE_DIM(priority_map_5);
 542                 break;
 543         default:
 544                 rte_errno = ENOTSUP;
 545                 DRV_LOG(ERR,
 546                         "port %u verbs maximum priority: %d expected 8/16",
 547                         dev->data->port_id, priority);
 548                 return -rte_errno;
 549         }
 550         DRV_LOG(INFO, "port %u flow maximum priority: %d",
 551                 dev->data->port_id, priority);
 552         return priority;
 553 }
 554
 555 /**
 556  * Adjust flow priority based on the highest layer and the request priority.
 557  *
 558  * @param[in] dev
 559  *   Pointer to the Ethernet device structure.
 560  * @param[in] priority
 561  *   The rule base priority.
 562  * @param[in] subpriority
 563  *   The priority based on the items.
 564  *
 565  * @return
 566  *   The new priority.
 567  */
 568 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
 569                                    uint32_t subpriority)
 570 {
 571         uint32_t res = 0;
 572         struct mlx5_priv *priv = dev->data->dev_private;
 573
 574         switch (priv->config.flow_prio) {
 575         case RTE_DIM(priority_map_3):
 576                 res = priority_map_3[priority][subpriority];
 577                 break;
 578         case RTE_DIM(priority_map_5):
 579                 res = priority_map_5[priority][subpriority];
 580                 break;
 581         }
 582         return  res;
 583 }
 584
 585 /**
 586  * Verify the @p item specifications (spec, last, mask) are compatible with the
 587  * NIC capabilities.
 588  *
 589  * @param[in] item
 590  *   Item specification.
 591  * @param[in] mask
 592  *   @p item->mask or flow default bit-masks.
 593  * @param[in] nic_mask
 594  *   Bit-masks covering supported fields by the NIC to compare with user mask.
 595  * @param[in] size
 596  *   Bit-masks size in bytes.
 597  * @param[out] error
 598  *   Pointer to error structure.
 599  *
 600  * @return
 601  *   0 on success, a negative errno value otherwise and rte_errno is set.
 602  */
 603 int
 604 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
 605                           const uint8_t *mask,
 606                           const uint8_t *nic_mask,
 607                           unsigned int size,
 608                           struct rte_flow_error *error)
 609 {
 610         unsigned int i;
 611
 612         MLX5_ASSERT(nic_mask);
 613         for (i = 0; i < size; ++i)
 614                 if ((nic_mask[i] | mask[i]) != nic_mask[i])
 615                         return rte_flow_error_set(error, ENOTSUP,
 616                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 617                                                   item,
 618                                                   "mask enables non supported"
 619                                                   " bits");
 620         if (!item->spec && (item->mask || item->last))
 621                 return rte_flow_error_set(error, EINVAL,
 622                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
 623                                           "mask/last without a spec is not"
 624                                           " supported");
 625         if (item->spec && item->last) {
 626                 uint8_t spec[size];
 627                 uint8_t last[size];
 628                 unsigned int i;
 629                 int ret;
 630
 631                 for (i = 0; i < size; ++i) {
 632                         spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
 633                         last[i] = ((const uint8_t *)item->last)[i] & mask[i];
 634                 }
 635                 ret = memcmp(spec, last, size);
 636                 if (ret != 0)
 637                         return rte_flow_error_set(error, EINVAL,
 638                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 639                                                   item,
 640                                                   "range is not valid");
 641         }
 642         return 0;
 643 }
 644
 645 /**
 646  * Adjust the hash fields according to the @p flow information.
 647  *
 648  * @param[in] dev_flow.
 649  *   Pointer to the mlx5_flow.
 650  * @param[in] tunnel
 651  *   1 when the hash field is for a tunnel item.
 652  * @param[in] layer_types
 653  *   ETH_RSS_* types.
 654  * @param[in] hash_fields
 655  *   Item hash fields.
 656  *
 657  * @return
 658  *   The hash fields that should be used.
 659  */
 660 uint64_t
 661 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
 662                             int tunnel __rte_unused, uint64_t layer_types,
 663                             uint64_t hash_fields)
 664 {
 665 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
 666         int rss_request_inner = rss_desc->level >= 2;
 667
 668         /* Check RSS hash level for tunnel. */
 669         if (tunnel && rss_request_inner)
 670                 hash_fields |= IBV_RX_HASH_INNER;
 671         else if (tunnel || rss_request_inner)
 672                 return 0;
 673 #endif
 674         /* Check if requested layer matches RSS hash fields. */
 675         if (!(rss_desc->types & layer_types))
 676                 return 0;
 677         return hash_fields;
 678 }
 679
 680 /**
 681  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
 682  * if several tunnel rules are used on this queue, the tunnel ptype will be
 683  * cleared.
 684  *
 685  * @param rxq_ctrl
 686  *   Rx queue to update.
 687  */
 688 static void
 689 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
 690 {
 691         unsigned int i;
 692         uint32_t tunnel_ptype = 0;
 693
 694         /* Look up for the ptype to use. */
 695         for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
 696                 if (!rxq_ctrl->flow_tunnels_n[i])
 697                         continue;
 698                 if (!tunnel_ptype) {
 699                         tunnel_ptype = tunnels_info[i].ptype;
 700                 } else {
 701                         tunnel_ptype = 0;
 702                         break;
 703                 }
 704         }
 705         rxq_ctrl->rxq.tunnel = tunnel_ptype;
 706 }
 707
 708 /**
 709  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
 710  * flow.
 711  *
 712  * @param[in] dev
 713  *   Pointer to the Ethernet device structure.
 714  * @param[in] dev_handle
 715  *   Pointer to device flow handle structure.
 716  */
 717 static void
 718 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
 719                        struct mlx5_flow_handle *dev_handle)
 720 {
 721         struct mlx5_priv *priv = dev->data->dev_private;
 722         const int mark = dev_handle->mark;
 723         const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
 724         struct mlx5_hrxq *hrxq;
 725         unsigned int i;
 726
 727         if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
 728                 return;
 729         hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
 730                               dev_handle->rix_hrxq);
 731         if (!hrxq)
 732                 return;
 733         for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
 734                 int idx = hrxq->ind_table->queues[i];
 735                 struct mlx5_rxq_ctrl *rxq_ctrl =
 736                         container_of((*priv->rxqs)[idx],
 737                                      struct mlx5_rxq_ctrl, rxq);
 738
 739                 /*
 740                  * To support metadata register copy on Tx loopback,
 741                  * this must be always enabled (metadata may arive
 742                  * from other port - not from local flows only.
 743                  */
 744                 if (priv->config.dv_flow_en &&
 745                     priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
 746                     mlx5_flow_ext_mreg_supported(dev)) {
 747                         rxq_ctrl->rxq.mark = 1;
 748                         rxq_ctrl->flow_mark_n = 1;
 749                 } else if (mark) {
 750                         rxq_ctrl->rxq.mark = 1;
 751                         rxq_ctrl->flow_mark_n++;
 752                 }
 753                 if (tunnel) {
 754                         unsigned int j;
 755
 756                         /* Increase the counter matching the flow. */
 757                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 758                                 if ((tunnels_info[j].tunnel &
 759                                      dev_handle->layers) ==
 760                                     tunnels_info[j].tunnel) {
 761                                         rxq_ctrl->flow_tunnels_n[j]++;
 762                                         break;
 763                                 }
 764                         }
 765                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
 766                 }
 767         }
 768 }
 769
 770 /**
 771  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
 772  *
 773  * @param[in] dev
 774  *   Pointer to the Ethernet device structure.
 775  * @param[in] flow
 776  *   Pointer to flow structure.
 777  */
 778 static void
 779 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
 780 {
 781         struct mlx5_priv *priv = dev->data->dev_private;
 782         uint32_t handle_idx;
 783         struct mlx5_flow_handle *dev_handle;
 784
 785         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
 786                        handle_idx, dev_handle, next)
 787                 flow_drv_rxq_flags_set(dev, dev_handle);
 788 }
 789
 790 /**
 791  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
 792  * device flow if no other flow uses it with the same kind of request.
 793  *
 794  * @param dev
 795  *   Pointer to Ethernet device.
 796  * @param[in] dev_handle
 797  *   Pointer to the device flow handle structure.
 798  */
 799 static void
 800 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
 801                         struct mlx5_flow_handle *dev_handle)
 802 {
 803         struct mlx5_priv *priv = dev->data->dev_private;
 804         const int mark = dev_handle->mark;
 805         const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
 806         struct mlx5_hrxq *hrxq;
 807         unsigned int i;
 808
 809         if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
 810                 return;
 811         hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
 812                               dev_handle->rix_hrxq);
 813         if (!hrxq)
 814                 return;
 815         MLX5_ASSERT(dev->data->dev_started);
 816         for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
 817                 int idx = hrxq->ind_table->queues[i];
 818                 struct mlx5_rxq_ctrl *rxq_ctrl =
 819                         container_of((*priv->rxqs)[idx],
 820                                      struct mlx5_rxq_ctrl, rxq);
 821
 822                 if (priv->config.dv_flow_en &&
 823                     priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
 824                     mlx5_flow_ext_mreg_supported(dev)) {
 825                         rxq_ctrl->rxq.mark = 1;
 826                         rxq_ctrl->flow_mark_n = 1;
 827                 } else if (mark) {
 828                         rxq_ctrl->flow_mark_n--;
 829                         rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
 830                 }
 831                 if (tunnel) {
 832                         unsigned int j;
 833
 834                         /* Decrease the counter matching the flow. */
 835                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 836                                 if ((tunnels_info[j].tunnel &
 837                                      dev_handle->layers) ==
 838                                     tunnels_info[j].tunnel) {
 839                                         rxq_ctrl->flow_tunnels_n[j]--;
 840                                         break;
 841                                 }
 842                         }
 843                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
 844                 }
 845         }
 846 }
 847
 848 /**
 849  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
 850  * @p flow if no other flow uses it with the same kind of request.
 851  *
 852  * @param dev
 853  *   Pointer to Ethernet device.
 854  * @param[in] flow
 855  *   Pointer to the flow.
 856  */
 857 static void
 858 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
 859 {
 860         struct mlx5_priv *priv = dev->data->dev_private;
 861         uint32_t handle_idx;
 862         struct mlx5_flow_handle *dev_handle;
 863
 864         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
 865                        handle_idx, dev_handle, next)
 866                 flow_drv_rxq_flags_trim(dev, dev_handle);
 867 }
 868
 869 /**
 870  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
 871  *
 872  * @param dev
 873  *   Pointer to Ethernet device.
 874  */
 875 static void
 876 flow_rxq_flags_clear(struct rte_eth_dev *dev)
 877 {
 878         struct mlx5_priv *priv = dev->data->dev_private;
 879         unsigned int i;
 880
 881         for (i = 0; i != priv->rxqs_n; ++i) {
 882                 struct mlx5_rxq_ctrl *rxq_ctrl;
 883                 unsigned int j;
 884
 885                 if (!(*priv->rxqs)[i])
 886                         continue;
 887                 rxq_ctrl = container_of((*priv->rxqs)[i],
 888                                         struct mlx5_rxq_ctrl, rxq);
 889                 rxq_ctrl->flow_mark_n = 0;
 890                 rxq_ctrl->rxq.mark = 0;
 891                 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
 892                         rxq_ctrl->flow_tunnels_n[j] = 0;
 893                 rxq_ctrl->rxq.tunnel = 0;
 894         }
 895 }
 896
 897 /*
 898  * return a pointer to the desired action in the list of actions.
 899  *
 900  * @param[in] actions
 901  *   The list of actions to search the action in.
 902  * @param[in] action
 903  *   The action to find.
 904  *
 905  * @return
 906  *   Pointer to the action in the list, if found. NULL otherwise.
 907  */
 908 const struct rte_flow_action *
 909 mlx5_flow_find_action(const struct rte_flow_action *actions,
 910                       enum rte_flow_action_type action)
 911 {
 912         if (actions == NULL)
 913                 return NULL;
 914         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
 915                 if (actions->type == action)
 916                         return actions;
 917         return NULL;
 918 }
 919
 920 /*
 921  * Validate the flag action.
 922  *
 923  * @param[in] action_flags
 924  *   Bit-fields that holds the actions detected until now.
 925  * @param[in] attr
 926  *   Attributes of flow that includes this action.
 927  * @param[out] error
 928  *   Pointer to error structure.
 929  *
 930  * @return
 931  *   0 on success, a negative errno value otherwise and rte_errno is set.
 932  */
 933 int
 934 mlx5_flow_validate_action_flag(uint64_t action_flags,
 935                                const struct rte_flow_attr *attr,
 936                                struct rte_flow_error *error)
 937 {
 938         if (action_flags & MLX5_FLOW_ACTION_MARK)
 939                 return rte_flow_error_set(error, EINVAL,
 940                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 941                                           "can't mark and flag in same flow");
 942         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 943                 return rte_flow_error_set(error, EINVAL,
 944                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 945                                           "can't have 2 flag"
 946                                           " actions in same flow");
 947         if (attr->egress)
 948                 return rte_flow_error_set(error, ENOTSUP,
 949                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 950                                           "flag action not supported for "
 951                                           "egress");
 952         return 0;
 953 }
 954
 955 /*
 956  * Validate the mark action.
 957  *
 958  * @param[in] action
 959  *   Pointer to the queue action.
 960  * @param[in] action_flags
 961  *   Bit-fields that holds the actions detected until now.
 962  * @param[in] attr
 963  *   Attributes of flow that includes this action.
 964  * @param[out] error
 965  *   Pointer to error structure.
 966  *
 967  * @return
 968  *   0 on success, a negative errno value otherwise and rte_errno is set.
 969  */
 970 int
 971 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
 972                                uint64_t action_flags,
 973                                const struct rte_flow_attr *attr,
 974                                struct rte_flow_error *error)
 975 {
 976         const struct rte_flow_action_mark *mark = action->conf;
 977
 978         if (!mark)
 979                 return rte_flow_error_set(error, EINVAL,
 980                                           RTE_FLOW_ERROR_TYPE_ACTION,
 981                                           action,
 982                                           "configuration cannot be null");
 983         if (mark->id >= MLX5_FLOW_MARK_MAX)
 984                 return rte_flow_error_set(error, EINVAL,
 985                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 986                                           &mark->id,
 987                                           "mark id must in 0 <= id < "
 988                                           RTE_STR(MLX5_FLOW_MARK_MAX));
 989         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 990                 return rte_flow_error_set(error, EINVAL,
 991                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 992                                           "can't flag and mark in same flow");
 993         if (action_flags & MLX5_FLOW_ACTION_MARK)
 994                 return rte_flow_error_set(error, EINVAL,
 995                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 996                                           "can't have 2 mark actions in same"
 997                                           " flow");
 998         if (attr->egress)
 999                 return rte_flow_error_set(error, ENOTSUP,
1000                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1001                                           "mark action not supported for "
1002                                           "egress");
1003         return 0;
1004 }
1005
1006 /*
1007  * Validate the drop action.
1008  *
1009  * @param[in] action_flags
1010  *   Bit-fields that holds the actions detected until now.
1011  * @param[in] attr
1012  *   Attributes of flow that includes this action.
1013  * @param[out] error
1014  *   Pointer to error structure.
1015  *
1016  * @return
1017  *   0 on success, a negative errno value otherwise and rte_errno is set.
1018  */
1019 int
1020 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1021                                const struct rte_flow_attr *attr,
1022                                struct rte_flow_error *error)
1023 {
1024         if (attr->egress)
1025                 return rte_flow_error_set(error, ENOTSUP,
1026                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1027                                           "drop action not supported for "
1028                                           "egress");
1029         return 0;
1030 }
1031
1032 /*
1033  * Validate the queue action.
1034  *
1035  * @param[in] action
1036  *   Pointer to the queue action.
1037  * @param[in] action_flags
1038  *   Bit-fields that holds the actions detected until now.
1039  * @param[in] dev
1040  *   Pointer to the Ethernet device structure.
1041  * @param[in] attr
1042  *   Attributes of flow that includes this action.
1043  * @param[out] error
1044  *   Pointer to error structure.
1045  *
1046  * @return
1047  *   0 on success, a negative errno value otherwise and rte_errno is set.
1048  */
1049 int
1050 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1051                                 uint64_t action_flags,
1052                                 struct rte_eth_dev *dev,
1053                                 const struct rte_flow_attr *attr,
1054                                 struct rte_flow_error *error)
1055 {
1056         struct mlx5_priv *priv = dev->data->dev_private;
1057         const struct rte_flow_action_queue *queue = action->conf;
1058
1059         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1060                 return rte_flow_error_set(error, EINVAL,
1061                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1062                                           "can't have 2 fate actions in"
1063                                           " same flow");
1064         if (!priv->rxqs_n)
1065                 return rte_flow_error_set(error, EINVAL,
1066                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1067                                           NULL, "No Rx queues configured");
1068         if (queue->index >= priv->rxqs_n)
1069                 return rte_flow_error_set(error, EINVAL,
1070                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1071                                           &queue->index,
1072                                           "queue index out of range");
1073         if (!(*priv->rxqs)[queue->index])
1074                 return rte_flow_error_set(error, EINVAL,
1075                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1076                                           &queue->index,
1077                                           "queue is not configured");
1078         if (attr->egress)
1079                 return rte_flow_error_set(error, ENOTSUP,
1080                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1081                                           "queue action not supported for "
1082                                           "egress");
1083         return 0;
1084 }
1085
1086 /*
1087  * Validate the rss action.
1088  *
1089  * @param[in] action
1090  *   Pointer to the queue action.
1091  * @param[in] action_flags
1092  *   Bit-fields that holds the actions detected until now.
1093  * @param[in] dev
1094  *   Pointer to the Ethernet device structure.
1095  * @param[in] attr
1096  *   Attributes of flow that includes this action.
1097  * @param[in] item_flags
1098  *   Items that were detected.
1099  * @param[out] error
1100  *   Pointer to error structure.
1101  *
1102  * @return
1103  *   0 on success, a negative errno value otherwise and rte_errno is set.
1104  */
1105 int
1106 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1107                               uint64_t action_flags,
1108                               struct rte_eth_dev *dev,
1109                               const struct rte_flow_attr *attr,
1110                               uint64_t item_flags,
1111                               struct rte_flow_error *error)
1112 {
1113         struct mlx5_priv *priv = dev->data->dev_private;
1114         const struct rte_flow_action_rss *rss = action->conf;
1115         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1116         unsigned int i;
1117
1118         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1119                 return rte_flow_error_set(error, EINVAL,
1120                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1121                                           "can't have 2 fate actions"
1122                                           " in same flow");
1123         if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1124             rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1125                 return rte_flow_error_set(error, ENOTSUP,
1126                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1127                                           &rss->func,
1128                                           "RSS hash function not supported");
1129 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1130         if (rss->level > 2)
1131 #else
1132         if (rss->level > 1)
1133 #endif
1134                 return rte_flow_error_set(error, ENOTSUP,
1135                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1136                                           &rss->level,
1137                                           "tunnel RSS is not supported");
1138         /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1139         if (rss->key_len == 0 && rss->key != NULL)
1140                 return rte_flow_error_set(error, ENOTSUP,
1141                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1142                                           &rss->key_len,
1143                                           "RSS hash key length 0");
1144         if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1145                 return rte_flow_error_set(error, ENOTSUP,
1146                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1147                                           &rss->key_len,
1148                                           "RSS hash key too small");
1149         if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1150                 return rte_flow_error_set(error, ENOTSUP,
1151                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1152                                           &rss->key_len,
1153                                           "RSS hash key too large");
1154         if (rss->queue_num > priv->config.ind_table_max_size)
1155                 return rte_flow_error_set(error, ENOTSUP,
1156                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1157                                           &rss->queue_num,
1158                                           "number of queues too large");
1159         if (rss->types & MLX5_RSS_HF_MASK)
1160                 return rte_flow_error_set(error, ENOTSUP,
1161                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1162                                           &rss->types,
1163                                           "some RSS protocols are not"
1164                                           " supported");
1165         if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1166             !(rss->types & ETH_RSS_IP))
1167                 return rte_flow_error_set(error, EINVAL,
1168                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1169                                           "L3 partial RSS requested but L3 RSS"
1170                                           " type not specified");
1171         if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1172             !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1173                 return rte_flow_error_set(error, EINVAL,
1174                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1175                                           "L4 partial RSS requested but L4 RSS"
1176                                           " type not specified");
1177         if (!priv->rxqs_n)
1178                 return rte_flow_error_set(error, EINVAL,
1179                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1180                                           NULL, "No Rx queues configured");
1181         if (!rss->queue_num)
1182                 return rte_flow_error_set(error, EINVAL,
1183                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1184                                           NULL, "No queues configured");
1185         for (i = 0; i != rss->queue_num; ++i) {
1186                 if (rss->queue[i] >= priv->rxqs_n)
1187                         return rte_flow_error_set
1188                                 (error, EINVAL,
1189                                  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1190                                  &rss->queue[i], "queue index out of range");
1191                 if (!(*priv->rxqs)[rss->queue[i]])
1192                         return rte_flow_error_set
1193                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1194                                  &rss->queue[i], "queue is not configured");
1195         }
1196         if (attr->egress)
1197                 return rte_flow_error_set(error, ENOTSUP,
1198                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1199                                           "rss action not supported for "
1200                                           "egress");
1201         if (rss->level > 1 &&  !tunnel)
1202                 return rte_flow_error_set(error, EINVAL,
1203                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1204                                           "inner RSS is not supported for "
1205                                           "non-tunnel flows");
1206         return 0;
1207 }
1208
1209 /*
1210  * Validate the count action.
1211  *
1212  * @param[in] dev
1213  *   Pointer to the Ethernet device structure.
1214  * @param[in] attr
1215  *   Attributes of flow that includes this action.
1216  * @param[out] error
1217  *   Pointer to error structure.
1218  *
1219  * @return
1220  *   0 on success, a negative errno value otherwise and rte_errno is set.
1221  */
1222 int
1223 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1224                                 const struct rte_flow_attr *attr,
1225                                 struct rte_flow_error *error)
1226 {
1227         if (attr->egress)
1228                 return rte_flow_error_set(error, ENOTSUP,
1229                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1230                                           "count action not supported for "
1231                                           "egress");
1232         return 0;
1233 }
1234
1235 /**
1236  * Verify the @p attributes will be correctly understood by the NIC and store
1237  * them in the @p flow if everything is correct.
1238  *
1239  * @param[in] dev
1240  *   Pointer to the Ethernet device structure.
1241  * @param[in] attributes
1242  *   Pointer to flow attributes
1243  * @param[out] error
1244  *   Pointer to error structure.
1245  *
1246  * @return
1247  *   0 on success, a negative errno value otherwise and rte_errno is set.
1248  */
1249 int
1250 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1251                               const struct rte_flow_attr *attributes,
1252                               struct rte_flow_error *error)
1253 {
1254         struct mlx5_priv *priv = dev->data->dev_private;
1255         uint32_t priority_max = priv->config.flow_prio - 1;
1256
1257         if (attributes->group)
1258                 return rte_flow_error_set(error, ENOTSUP,
1259                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1260                                           NULL, "groups is not supported");
1261         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1262             attributes->priority >= priority_max)
1263                 return rte_flow_error_set(error, ENOTSUP,
1264                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1265                                           NULL, "priority out of range");
1266         if (attributes->egress)
1267                 return rte_flow_error_set(error, ENOTSUP,
1268                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1269                                           "egress is not supported");
1270         if (attributes->transfer && !priv->config.dv_esw_en)
1271                 return rte_flow_error_set(error, ENOTSUP,
1272                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1273                                           NULL, "transfer is not supported");
1274         if (!attributes->ingress)
1275                 return rte_flow_error_set(error, EINVAL,
1276                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1277                                           NULL,
1278                                           "ingress attribute is mandatory");
1279         return 0;
1280 }
1281
1282 /**
1283  * Validate ICMP6 item.
1284  *
1285  * @param[in] item
1286  *   Item specification.
1287  * @param[in] item_flags
1288  *   Bit-fields that holds the items detected until now.
1289  * @param[out] error
1290  *   Pointer to error structure.
1291  *
1292  * @return
1293  *   0 on success, a negative errno value otherwise and rte_errno is set.
1294  */
1295 int
1296 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1297                                uint64_t item_flags,
1298                                uint8_t target_protocol,
1299                                struct rte_flow_error *error)
1300 {
1301         const struct rte_flow_item_icmp6 *mask = item->mask;
1302         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1303         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1304                                       MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1305         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1306                                       MLX5_FLOW_LAYER_OUTER_L4;
1307         int ret;
1308
1309         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1310                 return rte_flow_error_set(error, EINVAL,
1311                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1312                                           "protocol filtering not compatible"
1313                                           " with ICMP6 layer");
1314         if (!(item_flags & l3m))
1315                 return rte_flow_error_set(error, EINVAL,
1316                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1317                                           "IPv6 is mandatory to filter on"
1318                                           " ICMP6");
1319         if (item_flags & l4m)
1320                 return rte_flow_error_set(error, EINVAL,
1321                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1322                                           "multiple L4 layers not supported");
1323         if (!mask)
1324                 mask = &rte_flow_item_icmp6_mask;
1325         ret = mlx5_flow_item_acceptable
1326                 (item, (const uint8_t *)mask,
1327                  (const uint8_t *)&rte_flow_item_icmp6_mask,
1328                  sizeof(struct rte_flow_item_icmp6), error);
1329         if (ret < 0)
1330                 return ret;
1331         return 0;
1332 }
1333
1334 /**
1335  * Validate ICMP item.
1336  *
1337  * @param[in] item
1338  *   Item specification.
1339  * @param[in] item_flags
1340  *   Bit-fields that holds the items detected until now.
1341  * @param[out] error
1342  *   Pointer to error structure.
1343  *
1344  * @return
1345  *   0 on success, a negative errno value otherwise and rte_errno is set.
1346  */
1347 int
1348 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1349                              uint64_t item_flags,
1350                              uint8_t target_protocol,
1351                              struct rte_flow_error *error)
1352 {
1353         const struct rte_flow_item_icmp *mask = item->mask;
1354         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1355         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1356                                       MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1357         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1358                                       MLX5_FLOW_LAYER_OUTER_L4;
1359         int ret;
1360
1361         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1362                 return rte_flow_error_set(error, EINVAL,
1363                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1364                                           "protocol filtering not compatible"
1365                                           " with ICMP layer");
1366         if (!(item_flags & l3m))
1367                 return rte_flow_error_set(error, EINVAL,
1368                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1369                                           "IPv4 is mandatory to filter"
1370                                           " on ICMP");
1371         if (item_flags & l4m)
1372                 return rte_flow_error_set(error, EINVAL,
1373                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1374                                           "multiple L4 layers not supported");
1375         if (!mask)
1376                 mask = &rte_flow_item_icmp_mask;
1377         ret = mlx5_flow_item_acceptable
1378                 (item, (const uint8_t *)mask,
1379                  (const uint8_t *)&rte_flow_item_icmp_mask,
1380                  sizeof(struct rte_flow_item_icmp), error);
1381         if (ret < 0)
1382                 return ret;
1383         return 0;
1384 }
1385
1386 /**
1387  * Validate Ethernet item.
1388  *
1389  * @param[in] item
1390  *   Item specification.
1391  * @param[in] item_flags
1392  *   Bit-fields that holds the items detected until now.
1393  * @param[out] error
1394  *   Pointer to error structure.
1395  *
1396  * @return
1397  *   0 on success, a negative errno value otherwise and rte_errno is set.
1398  */
1399 int
1400 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1401                             uint64_t item_flags,
1402                             struct rte_flow_error *error)
1403 {
1404         const struct rte_flow_item_eth *mask = item->mask;
1405         const struct rte_flow_item_eth nic_mask = {
1406                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1407                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1408                 .type = RTE_BE16(0xffff),
1409         };
1410         int ret;
1411         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1412         const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1413                                        MLX5_FLOW_LAYER_OUTER_L2;
1414
1415         if (item_flags & ethm)
1416                 return rte_flow_error_set(error, ENOTSUP,
1417                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1418                                           "multiple L2 layers not supported");
1419         if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1420             (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1421                 return rte_flow_error_set(error, EINVAL,
1422                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1423                                           "L2 layer should not follow "
1424                                           "L3 layers");
1425         if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1426             (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1427                 return rte_flow_error_set(error, EINVAL,
1428                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1429                                           "L2 layer should not follow VLAN");
1430         if (!mask)
1431                 mask = &rte_flow_item_eth_mask;
1432         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1433                                         (const uint8_t *)&nic_mask,
1434                                         sizeof(struct rte_flow_item_eth),
1435                                         error);
1436         return ret;
1437 }
1438
1439 /**
1440  * Validate VLAN item.
1441  *
1442  * @param[in] item
1443  *   Item specification.
1444  * @param[in] item_flags
1445  *   Bit-fields that holds the items detected until now.
1446  * @param[in] dev
1447  *   Ethernet device flow is being created on.
1448  * @param[out] error
1449  *   Pointer to error structure.
1450  *
1451  * @return
1452  *   0 on success, a negative errno value otherwise and rte_errno is set.
1453  */
1454 int
1455 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1456                              uint64_t item_flags,
1457                              struct rte_eth_dev *dev,
1458                              struct rte_flow_error *error)
1459 {
1460         const struct rte_flow_item_vlan *spec = item->spec;
1461         const struct rte_flow_item_vlan *mask = item->mask;
1462         const struct rte_flow_item_vlan nic_mask = {
1463                 .tci = RTE_BE16(UINT16_MAX),
1464                 .inner_type = RTE_BE16(UINT16_MAX),
1465         };
1466         uint16_t vlan_tag = 0;
1467         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1468         int ret;
1469         const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1470                                         MLX5_FLOW_LAYER_INNER_L4) :
1471                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1472                                         MLX5_FLOW_LAYER_OUTER_L4);
1473         const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1474                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1475
1476         if (item_flags & vlanm)
1477                 return rte_flow_error_set(error, EINVAL,
1478                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1479                                           "multiple VLAN layers not supported");
1480         else if ((item_flags & l34m) != 0)
1481                 return rte_flow_error_set(error, EINVAL,
1482                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1483                                           "VLAN cannot follow L3/L4 layer");
1484         if (!mask)
1485                 mask = &rte_flow_item_vlan_mask;
1486         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1487                                         (const uint8_t *)&nic_mask,
1488                                         sizeof(struct rte_flow_item_vlan),
1489                                         error);
1490         if (ret)
1491                 return ret;
1492         if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1493                 struct mlx5_priv *priv = dev->data->dev_private;
1494
1495                 if (priv->vmwa_context) {
1496                         /*
1497                          * Non-NULL context means we have a virtual machine
1498                          * and SR-IOV enabled, we have to create VLAN interface
1499                          * to make hypervisor to setup E-Switch vport
1500                          * context correctly. We avoid creating the multiple
1501                          * VLAN interfaces, so we cannot support VLAN tag mask.
1502                          */
1503                         return rte_flow_error_set(error, EINVAL,
1504                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1505                                                   item,
1506                                                   "VLAN tag mask is not"
1507                                                   " supported in virtual"
1508                                                   " environment");
1509                 }
1510         }
1511         if (spec) {
1512                 vlan_tag = spec->tci;
1513                 vlan_tag &= mask->tci;
1514         }
1515         /*
1516          * From verbs perspective an empty VLAN is equivalent
1517          * to a packet without VLAN layer.
1518          */
1519         if (!vlan_tag)
1520                 return rte_flow_error_set(error, EINVAL,
1521                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1522                                           item->spec,
1523                                           "VLAN cannot be empty");
1524         return 0;
1525 }
1526
1527 /**
1528  * Validate IPV4 item.
1529  *
1530  * @param[in] item
1531  *   Item specification.
1532  * @param[in] item_flags
1533  *   Bit-fields that holds the items detected until now.
1534  * @param[in] acc_mask
1535  *   Acceptable mask, if NULL default internal default mask
1536  *   will be used to check whether item fields are supported.
1537  * @param[out] error
1538  *   Pointer to error structure.
1539  *
1540  * @return
1541  *   0 on success, a negative errno value otherwise and rte_errno is set.
1542  */
1543 int
1544 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1545                              uint64_t item_flags,
1546                              uint64_t last_item,
1547                              uint16_t ether_type,
1548                              const struct rte_flow_item_ipv4 *acc_mask,
1549                              struct rte_flow_error *error)
1550 {
1551         const struct rte_flow_item_ipv4 *mask = item->mask;
1552         const struct rte_flow_item_ipv4 *spec = item->spec;
1553         const struct rte_flow_item_ipv4 nic_mask = {
1554                 .hdr = {
1555                         .src_addr = RTE_BE32(0xffffffff),
1556                         .dst_addr = RTE_BE32(0xffffffff),
1557                         .type_of_service = 0xff,
1558                         .next_proto_id = 0xff,
1559                 },
1560         };
1561         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1562         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1563                                       MLX5_FLOW_LAYER_OUTER_L3;
1564         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1565                                       MLX5_FLOW_LAYER_OUTER_L4;
1566         int ret;
1567         uint8_t next_proto = 0xFF;
1568         const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1569                                   MLX5_FLOW_LAYER_OUTER_VLAN |
1570                                   MLX5_FLOW_LAYER_INNER_VLAN);
1571
1572         if ((last_item & l2_vlan) && ether_type &&
1573             ether_type != RTE_ETHER_TYPE_IPV4)
1574                 return rte_flow_error_set(error, EINVAL,
1575                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1576                                           "IPv4 cannot follow L2/VLAN layer "
1577                                           "which ether type is not IPv4");
1578         if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1579                 if (mask && spec)
1580                         next_proto = mask->hdr.next_proto_id &
1581                                      spec->hdr.next_proto_id;
1582                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1583                         return rte_flow_error_set(error, EINVAL,
1584                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1585                                                   item,
1586                                                   "multiple tunnel "
1587                                                   "not supported");
1588         }
1589         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1590                 return rte_flow_error_set(error, EINVAL,
1591                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1592                                           "wrong tunnel type - IPv6 specified "
1593                                           "but IPv4 item provided");
1594         if (item_flags & l3m)
1595                 return rte_flow_error_set(error, ENOTSUP,
1596                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1597                                           "multiple L3 layers not supported");
1598         else if (item_flags & l4m)
1599                 return rte_flow_error_set(error, EINVAL,
1600                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1601                                           "L3 cannot follow an L4 layer.");
1602         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1603                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1604                 return rte_flow_error_set(error, EINVAL,
1605                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1606                                           "L3 cannot follow an NVGRE layer.");
1607         if (!mask)
1608                 mask = &rte_flow_item_ipv4_mask;
1609         else if (mask->hdr.next_proto_id != 0 &&
1610                  mask->hdr.next_proto_id != 0xff)
1611                 return rte_flow_error_set(error, EINVAL,
1612                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1613                                           "partial mask is not supported"
1614                                           " for protocol");
1615         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1616                                         acc_mask ? (const uint8_t *)acc_mask
1617                                                  : (const uint8_t *)&nic_mask,
1618                                         sizeof(struct rte_flow_item_ipv4),
1619                                         error);
1620         if (ret < 0)
1621                 return ret;
1622         return 0;
1623 }
1624
1625 /**
1626  * Validate IPV6 item.
1627  *
1628  * @param[in] item
1629  *   Item specification.
1630  * @param[in] item_flags
1631  *   Bit-fields that holds the items detected until now.
1632  * @param[in] acc_mask
1633  *   Acceptable mask, if NULL default internal default mask
1634  *   will be used to check whether item fields are supported.
1635  * @param[out] error
1636  *   Pointer to error structure.
1637  *
1638  * @return
1639  *   0 on success, a negative errno value otherwise and rte_errno is set.
1640  */
1641 int
1642 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1643                              uint64_t item_flags,
1644                              uint64_t last_item,
1645                              uint16_t ether_type,
1646                              const struct rte_flow_item_ipv6 *acc_mask,
1647                              struct rte_flow_error *error)
1648 {
1649         const struct rte_flow_item_ipv6 *mask = item->mask;
1650         const struct rte_flow_item_ipv6 *spec = item->spec;
1651         const struct rte_flow_item_ipv6 nic_mask = {
1652                 .hdr = {
1653                         .src_addr =
1654                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1655                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1656                         .dst_addr =
1657                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1658                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1659                         .vtc_flow = RTE_BE32(0xffffffff),
1660                         .proto = 0xff,
1661                 },
1662         };
1663         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1664         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1665                                       MLX5_FLOW_LAYER_OUTER_L3;
1666         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1667                                       MLX5_FLOW_LAYER_OUTER_L4;
1668         int ret;
1669         uint8_t next_proto = 0xFF;
1670         const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1671                                   MLX5_FLOW_LAYER_OUTER_VLAN |
1672                                   MLX5_FLOW_LAYER_INNER_VLAN);
1673
1674         if ((last_item & l2_vlan) && ether_type &&
1675             ether_type != RTE_ETHER_TYPE_IPV6)
1676                 return rte_flow_error_set(error, EINVAL,
1677                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1678                                           "IPv6 cannot follow L2/VLAN layer "
1679                                           "which ether type is not IPv6");
1680         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1681                 if (mask && spec)
1682                         next_proto = mask->hdr.proto & spec->hdr.proto;
1683                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1684                         return rte_flow_error_set(error, EINVAL,
1685                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1686                                                   item,
1687                                                   "multiple tunnel "
1688                                                   "not supported");
1689         }
1690         if (item_flags & MLX5_FLOW_LAYER_IPIP)
1691                 return rte_flow_error_set(error, EINVAL,
1692                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1693                                           "wrong tunnel type - IPv4 specified "
1694                                           "but IPv6 item provided");
1695         if (item_flags & l3m)
1696                 return rte_flow_error_set(error, ENOTSUP,
1697                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1698                                           "multiple L3 layers not supported");
1699         else if (item_flags & l4m)
1700                 return rte_flow_error_set(error, EINVAL,
1701                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1702                                           "L3 cannot follow an L4 layer.");
1703         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1704                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1705                 return rte_flow_error_set(error, EINVAL,
1706                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1707                                           "L3 cannot follow an NVGRE layer.");
1708         if (!mask)
1709                 mask = &rte_flow_item_ipv6_mask;
1710         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1711                                         acc_mask ? (const uint8_t *)acc_mask
1712                                                  : (const uint8_t *)&nic_mask,
1713                                         sizeof(struct rte_flow_item_ipv6),
1714                                         error);
1715         if (ret < 0)
1716                 return ret;
1717         return 0;
1718 }
1719
1720 /**
1721  * Validate UDP item.
1722  *
1723  * @param[in] item
1724  *   Item specification.
1725  * @param[in] item_flags
1726  *   Bit-fields that holds the items detected until now.
1727  * @param[in] target_protocol
1728  *   The next protocol in the previous item.
1729  * @param[in] flow_mask
1730  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1731  * @param[out] error
1732  *   Pointer to error structure.
1733  *
1734  * @return
1735  *   0 on success, a negative errno value otherwise and rte_errno is set.
1736  */
1737 int
1738 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1739                             uint64_t item_flags,
1740                             uint8_t target_protocol,
1741                             struct rte_flow_error *error)
1742 {
1743         const struct rte_flow_item_udp *mask = item->mask;
1744         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1745         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1746                                       MLX5_FLOW_LAYER_OUTER_L3;
1747         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1748                                       MLX5_FLOW_LAYER_OUTER_L4;
1749         int ret;
1750
1751         if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1752                 return rte_flow_error_set(error, EINVAL,
1753                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1754                                           "protocol filtering not compatible"
1755                                           " with UDP layer");
1756         if (!(item_flags & l3m))
1757                 return rte_flow_error_set(error, EINVAL,
1758                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1759                                           "L3 is mandatory to filter on L4");
1760         if (item_flags & l4m)
1761                 return rte_flow_error_set(error, EINVAL,
1762                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1763                                           "multiple L4 layers not supported");
1764         if (!mask)
1765                 mask = &rte_flow_item_udp_mask;
1766         ret = mlx5_flow_item_acceptable
1767                 (item, (const uint8_t *)mask,
1768                  (const uint8_t *)&rte_flow_item_udp_mask,
1769                  sizeof(struct rte_flow_item_udp), error);
1770         if (ret < 0)
1771                 return ret;
1772         return 0;
1773 }
1774
1775 /**
1776  * Validate TCP item.
1777  *
1778  * @param[in] item
1779  *   Item specification.
1780  * @param[in] item_flags
1781  *   Bit-fields that holds the items detected until now.
1782  * @param[in] target_protocol
1783  *   The next protocol in the previous item.
1784  * @param[out] error
1785  *   Pointer to error structure.
1786  *
1787  * @return
1788  *   0 on success, a negative errno value otherwise and rte_errno is set.
1789  */
1790 int
1791 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1792                             uint64_t item_flags,
1793                             uint8_t target_protocol,
1794                             const struct rte_flow_item_tcp *flow_mask,
1795                             struct rte_flow_error *error)
1796 {
1797         const struct rte_flow_item_tcp *mask = item->mask;
1798         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1799         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1800                                       MLX5_FLOW_LAYER_OUTER_L3;
1801         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1802                                       MLX5_FLOW_LAYER_OUTER_L4;
1803         int ret;
1804
1805         MLX5_ASSERT(flow_mask);
1806         if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1807                 return rte_flow_error_set(error, EINVAL,
1808                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1809                                           "protocol filtering not compatible"
1810                                           " with TCP layer");
1811         if (!(item_flags & l3m))
1812                 return rte_flow_error_set(error, EINVAL,
1813                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1814                                           "L3 is mandatory to filter on L4");
1815         if (item_flags & l4m)
1816                 return rte_flow_error_set(error, EINVAL,
1817                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1818                                           "multiple L4 layers not supported");
1819         if (!mask)
1820                 mask = &rte_flow_item_tcp_mask;
1821         ret = mlx5_flow_item_acceptable
1822                 (item, (const uint8_t *)mask,
1823                  (const uint8_t *)flow_mask,
1824                  sizeof(struct rte_flow_item_tcp), error);
1825         if (ret < 0)
1826                 return ret;
1827         return 0;
1828 }
1829
1830 /**
1831  * Validate VXLAN item.
1832  *
1833  * @param[in] item
1834  *   Item specification.
1835  * @param[in] item_flags
1836  *   Bit-fields that holds the items detected until now.
1837  * @param[in] target_protocol
1838  *   The next protocol in the previous item.
1839  * @param[out] error
1840  *   Pointer to error structure.
1841  *
1842  * @return
1843  *   0 on success, a negative errno value otherwise and rte_errno is set.
1844  */
1845 int
1846 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1847                               uint64_t item_flags,
1848                               struct rte_flow_error *error)
1849 {
1850         const struct rte_flow_item_vxlan *spec = item->spec;
1851         const struct rte_flow_item_vxlan *mask = item->mask;
1852         int ret;
1853         union vni {
1854                 uint32_t vlan_id;
1855                 uint8_t vni[4];
1856         } id = { .vlan_id = 0, };
1857
1858
1859         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1860                 return rte_flow_error_set(error, ENOTSUP,
1861                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1862                                           "multiple tunnel layers not"
1863                                           " supported");
1864         /*
1865          * Verify only UDPv4 is present as defined in
1866          * https://tools.ietf.org/html/rfc7348
1867          */
1868         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1869                 return rte_flow_error_set(error, EINVAL,
1870                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1871                                           "no outer UDP layer found");
1872         if (!mask)
1873                 mask = &rte_flow_item_vxlan_mask;
1874         ret = mlx5_flow_item_acceptable
1875                 (item, (const uint8_t *)mask,
1876                  (const uint8_t *)&rte_flow_item_vxlan_mask,
1877                  sizeof(struct rte_flow_item_vxlan),
1878                  error);
1879         if (ret < 0)
1880                 return ret;
1881         if (spec) {
1882                 memcpy(&id.vni[1], spec->vni, 3);
1883                 memcpy(&id.vni[1], mask->vni, 3);
1884         }
1885         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1886                 return rte_flow_error_set(error, ENOTSUP,
1887                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1888                                           "VXLAN tunnel must be fully defined");
1889         return 0;
1890 }
1891
1892 /**
1893  * Validate VXLAN_GPE item.
1894  *
1895  * @param[in] item
1896  *   Item specification.
1897  * @param[in] item_flags
1898  *   Bit-fields that holds the items detected until now.
1899  * @param[in] priv
1900  *   Pointer to the private data structure.
1901  * @param[in] target_protocol
1902  *   The next protocol in the previous item.
1903  * @param[out] error
1904  *   Pointer to error structure.
1905  *
1906  * @return
1907  *   0 on success, a negative errno value otherwise and rte_errno is set.
1908  */
1909 int
1910 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1911                                   uint64_t item_flags,
1912                                   struct rte_eth_dev *dev,
1913                                   struct rte_flow_error *error)
1914 {
1915         struct mlx5_priv *priv = dev->data->dev_private;
1916         const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1917         const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1918         int ret;
1919         union vni {
1920                 uint32_t vlan_id;
1921                 uint8_t vni[4];
1922         } id = { .vlan_id = 0, };
1923
1924         if (!priv->config.l3_vxlan_en)
1925                 return rte_flow_error_set(error, ENOTSUP,
1926                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1927                                           "L3 VXLAN is not enabled by device"
1928                                           " parameter and/or not configured in"
1929                                           " firmware");
1930         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1931                 return rte_flow_error_set(error, ENOTSUP,
1932                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1933                                           "multiple tunnel layers not"
1934                                           " supported");
1935         /*
1936          * Verify only UDPv4 is present as defined in
1937          * https://tools.ietf.org/html/rfc7348
1938          */
1939         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1940                 return rte_flow_error_set(error, EINVAL,
1941                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1942                                           "no outer UDP layer found");
1943         if (!mask)
1944                 mask = &rte_flow_item_vxlan_gpe_mask;
1945         ret = mlx5_flow_item_acceptable
1946                 (item, (const uint8_t *)mask,
1947                  (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1948                  sizeof(struct rte_flow_item_vxlan_gpe),
1949                  error);
1950         if (ret < 0)
1951                 return ret;
1952         if (spec) {
1953                 if (spec->protocol)
1954                         return rte_flow_error_set(error, ENOTSUP,
1955                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1956                                                   item,
1957                                                   "VxLAN-GPE protocol"
1958                                                   " not supported");
1959                 memcpy(&id.vni[1], spec->vni, 3);
1960                 memcpy(&id.vni[1], mask->vni, 3);
1961         }
1962         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1963                 return rte_flow_error_set(error, ENOTSUP,
1964                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1965                                           "VXLAN-GPE tunnel must be fully"
1966                                           " defined");
1967         return 0;
1968 }
1969 /**
1970  * Validate GRE Key item.
1971  *
1972  * @param[in] item
1973  *   Item specification.
1974  * @param[in] item_flags
1975  *   Bit flags to mark detected items.
1976  * @param[in] gre_item
1977  *   Pointer to gre_item
1978  * @param[out] error
1979  *   Pointer to error structure.
1980  *
1981  * @return
1982  *   0 on success, a negative errno value otherwise and rte_errno is set.
1983  */
1984 int
1985 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1986                                 uint64_t item_flags,
1987                                 const struct rte_flow_item *gre_item,
1988                                 struct rte_flow_error *error)
1989 {
1990         const rte_be32_t *mask = item->mask;
1991         int ret = 0;
1992         rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1993         const struct rte_flow_item_gre *gre_spec;
1994         const struct rte_flow_item_gre *gre_mask;
1995
1996         if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1997                 return rte_flow_error_set(error, ENOTSUP,
1998                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1999                                           "Multiple GRE key not support");
2000         if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2001                 return rte_flow_error_set(error, ENOTSUP,
2002                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2003                                           "No preceding GRE header");
2004         if (item_flags & MLX5_FLOW_LAYER_INNER)
2005                 return rte_flow_error_set(error, ENOTSUP,
2006                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2007                                           "GRE key following a wrong item");
2008         gre_mask = gre_item->mask;
2009         if (!gre_mask)
2010                 gre_mask = &rte_flow_item_gre_mask;
2011         gre_spec = gre_item->spec;
2012         if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2013                          !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2014                 return rte_flow_error_set(error, EINVAL,
2015                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2016                                           "Key bit must be on");
2017
2018         if (!mask)
2019                 mask = &gre_key_default_mask;
2020         ret = mlx5_flow_item_acceptable
2021                 (item, (const uint8_t *)mask,
2022                  (const uint8_t *)&gre_key_default_mask,
2023                  sizeof(rte_be32_t), error);
2024         return ret;
2025 }
2026
2027 /**
2028  * Validate GRE item.
2029  *
2030  * @param[in] item
2031  *   Item specification.
2032  * @param[in] item_flags
2033  *   Bit flags to mark detected items.
2034  * @param[in] target_protocol
2035  *   The next protocol in the previous item.
2036  * @param[out] error
2037  *   Pointer to error structure.
2038  *
2039  * @return
2040  *   0 on success, a negative errno value otherwise and rte_errno is set.
2041  */
2042 int
2043 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2044                             uint64_t item_flags,
2045                             uint8_t target_protocol,
2046                             struct rte_flow_error *error)
2047 {
2048         const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2049         const struct rte_flow_item_gre *mask = item->mask;
2050         int ret;
2051         const struct rte_flow_item_gre nic_mask = {
2052                 .c_rsvd0_ver = RTE_BE16(0xB000),
2053                 .protocol = RTE_BE16(UINT16_MAX),
2054         };
2055
2056         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2057                 return rte_flow_error_set(error, EINVAL,
2058                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2059                                           "protocol filtering not compatible"
2060                                           " with this GRE layer");
2061         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2062                 return rte_flow_error_set(error, ENOTSUP,
2063                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2064                                           "multiple tunnel layers not"
2065                                           " supported");
2066         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2067                 return rte_flow_error_set(error, ENOTSUP,
2068                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2069                                           "L3 Layer is missing");
2070         if (!mask)
2071                 mask = &rte_flow_item_gre_mask;
2072         ret = mlx5_flow_item_acceptable
2073                 (item, (const uint8_t *)mask,
2074                  (const uint8_t *)&nic_mask,
2075                  sizeof(struct rte_flow_item_gre), error);
2076         if (ret < 0)
2077                 return ret;
2078 #ifndef HAVE_MLX5DV_DR
2079 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2080         if (spec && (spec->protocol & mask->protocol))
2081                 return rte_flow_error_set(error, ENOTSUP,
2082                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2083                                           "without MPLS support the"
2084                                           " specification cannot be used for"
2085                                           " filtering");
2086 #endif
2087 #endif
2088         return 0;
2089 }
2090
2091 /**
2092  * Validate Geneve item.
2093  *
2094  * @param[in] item
2095  *   Item specification.
2096  * @param[in] itemFlags
2097  *   Bit-fields that holds the items detected until now.
2098  * @param[in] enPriv
2099  *   Pointer to the private data structure.
2100  * @param[out] error
2101  *   Pointer to error structure.
2102  *
2103  * @return
2104  *   0 on success, a negative errno value otherwise and rte_errno is set.
2105  */
2106
2107 int
2108 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2109                                uint64_t item_flags,
2110                                struct rte_eth_dev *dev,
2111                                struct rte_flow_error *error)
2112 {
2113         struct mlx5_priv *priv = dev->data->dev_private;
2114         const struct rte_flow_item_geneve *spec = item->spec;
2115         const struct rte_flow_item_geneve *mask = item->mask;
2116         int ret;
2117         uint16_t gbhdr;
2118         uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2119                           MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2120         const struct rte_flow_item_geneve nic_mask = {
2121                 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2122                 .vni = "\xff\xff\xff",
2123                 .protocol = RTE_BE16(UINT16_MAX),
2124         };
2125
2126         if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2127                 return rte_flow_error_set(error, ENOTSUP,
2128                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2129                                           "L3 Geneve is not enabled by device"
2130                                           " parameter and/or not configured in"
2131                                           " firmware");
2132         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2133                 return rte_flow_error_set(error, ENOTSUP,
2134                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2135                                           "multiple tunnel layers not"
2136                                           " supported");
2137         /*
2138          * Verify only UDPv4 is present as defined in
2139          * https://tools.ietf.org/html/rfc7348
2140          */
2141         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2142                 return rte_flow_error_set(error, EINVAL,
2143                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2144                                           "no outer UDP layer found");
2145         if (!mask)
2146                 mask = &rte_flow_item_geneve_mask;
2147         ret = mlx5_flow_item_acceptable
2148                                   (item, (const uint8_t *)mask,
2149                                    (const uint8_t *)&nic_mask,
2150                                    sizeof(struct rte_flow_item_geneve), error);
2151         if (ret)
2152                 return ret;
2153         if (spec) {
2154                 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2155                 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2156                      MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2157                      MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2158                         return rte_flow_error_set(error, ENOTSUP,
2159                                                   RTE_FLOW_ERROR_TYPE_ITEM,
2160                                                   item,
2161                                                   "Geneve protocol unsupported"
2162                                                   " fields are being used");
2163                 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2164                         return rte_flow_error_set
2165                                         (error, ENOTSUP,
2166                                          RTE_FLOW_ERROR_TYPE_ITEM,
2167                                          item,
2168                                          "Unsupported Geneve options length");
2169         }
2170         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2171                 return rte_flow_error_set
2172                                     (error, ENOTSUP,
2173                                      RTE_FLOW_ERROR_TYPE_ITEM, item,
2174                                      "Geneve tunnel must be fully defined");
2175         return 0;
2176 }
2177
2178 /**
2179  * Validate MPLS item.
2180  *
2181  * @param[in] dev
2182  *   Pointer to the rte_eth_dev structure.
2183  * @param[in] item
2184  *   Item specification.
2185  * @param[in] item_flags
2186  *   Bit-fields that holds the items detected until now.
2187  * @param[in] prev_layer
2188  *   The protocol layer indicated in previous item.
2189  * @param[out] error
2190  *   Pointer to error structure.
2191  *
2192  * @return
2193  *   0 on success, a negative errno value otherwise and rte_errno is set.
2194  */
2195 int
2196 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2197                              const struct rte_flow_item *item __rte_unused,
2198                              uint64_t item_flags __rte_unused,
2199                              uint64_t prev_layer __rte_unused,
2200                              struct rte_flow_error *error)
2201 {
2202 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2203         const struct rte_flow_item_mpls *mask = item->mask;
2204         struct mlx5_priv *priv = dev->data->dev_private;
2205         int ret;
2206
2207         if (!priv->config.mpls_en)
2208                 return rte_flow_error_set(error, ENOTSUP,
2209                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2210                                           "MPLS not supported or"
2211                                           " disabled in firmware"
2212                                           " configuration.");
2213         /* MPLS over IP, UDP, GRE is allowed */
2214         if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2215                             MLX5_FLOW_LAYER_OUTER_L4_UDP |
2216                             MLX5_FLOW_LAYER_GRE)))
2217                 return rte_flow_error_set(error, EINVAL,
2218                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2219                                           "protocol filtering not compatible"
2220                                           " with MPLS layer");
2221         /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2222         if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2223             !(item_flags & MLX5_FLOW_LAYER_GRE))
2224                 return rte_flow_error_set(error, ENOTSUP,
2225                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2226                                           "multiple tunnel layers not"
2227                                           " supported");
2228         if (!mask)
2229                 mask = &rte_flow_item_mpls_mask;
2230         ret = mlx5_flow_item_acceptable
2231                 (item, (const uint8_t *)mask,
2232                  (const uint8_t *)&rte_flow_item_mpls_mask,
2233                  sizeof(struct rte_flow_item_mpls), error);
2234         if (ret < 0)
2235                 return ret;
2236         return 0;
2237 #endif
2238         return rte_flow_error_set(error, ENOTSUP,
2239                                   RTE_FLOW_ERROR_TYPE_ITEM, item,
2240                                   "MPLS is not supported by Verbs, please"
2241                                   " update.");
2242 }
2243
2244 /**
2245  * Validate NVGRE item.
2246  *
2247  * @param[in] item
2248  *   Item specification.
2249  * @param[in] item_flags
2250  *   Bit flags to mark detected items.
2251  * @param[in] target_protocol
2252  *   The next protocol in the previous item.
2253  * @param[out] error
2254  *   Pointer to error structure.
2255  *
2256  * @return
2257  *   0 on success, a negative errno value otherwise and rte_errno is set.
2258  */
2259 int
2260 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2261                               uint64_t item_flags,
2262                               uint8_t target_protocol,
2263                               struct rte_flow_error *error)
2264 {
2265         const struct rte_flow_item_nvgre *mask = item->mask;
2266         int ret;
2267
2268         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2269                 return rte_flow_error_set(error, EINVAL,
2270                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2271                                           "protocol filtering not compatible"
2272                                           " with this GRE layer");
2273         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2274                 return rte_flow_error_set(error, ENOTSUP,
2275                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2276                                           "multiple tunnel layers not"
2277                                           " supported");
2278         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2279                 return rte_flow_error_set(error, ENOTSUP,
2280                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2281                                           "L3 Layer is missing");
2282         if (!mask)
2283                 mask = &rte_flow_item_nvgre_mask;
2284         ret = mlx5_flow_item_acceptable
2285                 (item, (const uint8_t *)mask,
2286                  (const uint8_t *)&rte_flow_item_nvgre_mask,
2287                  sizeof(struct rte_flow_item_nvgre), error);
2288         if (ret < 0)
2289                 return ret;
2290         return 0;
2291 }
2292
2293 /* Allocate unique ID for the split Q/RSS subflows. */
2294 static uint32_t
2295 flow_qrss_get_id(struct rte_eth_dev *dev)
2296 {
2297         struct mlx5_priv *priv = dev->data->dev_private;
2298         uint32_t qrss_id, ret;
2299
2300         ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2301         if (ret)
2302                 return 0;
2303         MLX5_ASSERT(qrss_id);
2304         return qrss_id;
2305 }
2306
2307 /* Free unique ID for the split Q/RSS subflows. */
2308 static void
2309 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2310 {
2311         struct mlx5_priv *priv = dev->data->dev_private;
2312
2313         if (qrss_id)
2314                 mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2315 }
2316
2317 /**
2318  * Release resource related QUEUE/RSS action split.
2319  *
2320  * @param dev
2321  *   Pointer to Ethernet device.
2322  * @param flow
2323  *   Flow to release id's from.
2324  */
2325 static void
2326 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2327                              struct rte_flow *flow)
2328 {
2329         struct mlx5_priv *priv = dev->data->dev_private;
2330         uint32_t handle_idx;
2331         struct mlx5_flow_handle *dev_handle;
2332
2333         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2334                        handle_idx, dev_handle, next)
2335                 if (dev_handle->split_flow_id)
2336                         flow_qrss_free_id(dev, dev_handle->split_flow_id);
2337 }
2338
2339 static int
2340 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2341                    const struct rte_flow_attr *attr __rte_unused,
2342                    const struct rte_flow_item items[] __rte_unused,
2343                    const struct rte_flow_action actions[] __rte_unused,
2344                    bool external __rte_unused,
2345                    struct rte_flow_error *error)
2346 {
2347         return rte_flow_error_set(error, ENOTSUP,
2348                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2349 }
2350
2351 static struct mlx5_flow *
2352 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2353                   const struct rte_flow_attr *attr __rte_unused,
2354                   const struct rte_flow_item items[] __rte_unused,
2355                   const struct rte_flow_action actions[] __rte_unused,
2356                   struct rte_flow_error *error)
2357 {
2358         rte_flow_error_set(error, ENOTSUP,
2359                            RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2360         return NULL;
2361 }
2362
2363 static int
2364 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2365                     struct mlx5_flow *dev_flow __rte_unused,
2366                     const struct rte_flow_attr *attr __rte_unused,
2367                     const struct rte_flow_item items[] __rte_unused,
2368                     const struct rte_flow_action actions[] __rte_unused,
2369                     struct rte_flow_error *error)
2370 {
2371         return rte_flow_error_set(error, ENOTSUP,
2372                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2373 }
2374
2375 static int
2376 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2377                 struct rte_flow *flow __rte_unused,
2378                 struct rte_flow_error *error)
2379 {
2380         return rte_flow_error_set(error, ENOTSUP,
2381                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2382 }
2383
2384 static void
2385 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2386                  struct rte_flow *flow __rte_unused)
2387 {
2388 }
2389
2390 static void
2391 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2392                   struct rte_flow *flow __rte_unused)
2393 {
2394 }
2395
2396 static int
2397 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2398                 struct rte_flow *flow __rte_unused,
2399                 const struct rte_flow_action *actions __rte_unused,
2400                 void *data __rte_unused,
2401                 struct rte_flow_error *error)
2402 {
2403         return rte_flow_error_set(error, ENOTSUP,
2404                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2405 }
2406
2407 /* Void driver to protect from null pointer reference. */
2408 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2409         .validate = flow_null_validate,
2410         .prepare = flow_null_prepare,
2411         .translate = flow_null_translate,
2412         .apply = flow_null_apply,
2413         .remove = flow_null_remove,
2414         .destroy = flow_null_destroy,
2415         .query = flow_null_query,
2416 };
2417
2418 /**
2419  * Select flow driver type according to flow attributes and device
2420  * configuration.
2421  *
2422  * @param[in] dev
2423  *   Pointer to the dev structure.
2424  * @param[in] attr
2425  *   Pointer to the flow attributes.
2426  *
2427  * @return
2428  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2429  */
2430 static enum mlx5_flow_drv_type
2431 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2432 {
2433         struct mlx5_priv *priv = dev->data->dev_private;
2434         enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2435
2436         if (attr->transfer && priv->config.dv_esw_en)
2437                 type = MLX5_FLOW_TYPE_DV;
2438         if (!attr->transfer)
2439                 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2440                                                  MLX5_FLOW_TYPE_VERBS;
2441         return type;
2442 }
2443
2444 #define flow_get_drv_ops(type) flow_drv_ops[type]
2445
2446 /**
2447  * Flow driver validation API. This abstracts calling driver specific functions.
2448  * The type of flow driver is determined according to flow attributes.
2449  *
2450  * @param[in] dev
2451  *   Pointer to the dev structure.
2452  * @param[in] attr
2453  *   Pointer to the flow attributes.
2454  * @param[in] items
2455  *   Pointer to the list of items.
2456  * @param[in] actions
2457  *   Pointer to the list of actions.
2458  * @param[in] external
2459  *   This flow rule is created by request external to PMD.
2460  * @param[out] error
2461  *   Pointer to the error structure.
2462  *
2463  * @return
2464  *   0 on success, a negative errno value otherwise and rte_errno is set.
2465  */
2466 static inline int
2467 flow_drv_validate(struct rte_eth_dev *dev,
2468                   const struct rte_flow_attr *attr,
2469                   const struct rte_flow_item items[],
2470                   const struct rte_flow_action actions[],
2471                   bool external, struct rte_flow_error *error)
2472 {
2473         const struct mlx5_flow_driver_ops *fops;
2474         enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2475
2476         fops = flow_get_drv_ops(type);
2477         return fops->validate(dev, attr, items, actions, external, error);
2478 }
2479
2480 /**
2481  * Flow driver preparation API. This abstracts calling driver specific
2482  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2483  * calculates the size of memory required for device flow, allocates the memory,
2484  * initializes the device flow and returns the pointer.
2485  *
2486  * @note
2487  *   This function initializes device flow structure such as dv or verbs in
2488  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2489  *   rest. For example, adding returning device flow to flow->dev_flow list and
2490  *   setting backward reference to the flow should be done out of this function.
2491  *   layers field is not filled either.
2492  *
2493  * @param[in] dev
2494  *   Pointer to the dev structure.
2495  * @param[in] attr
2496  *   Pointer to the flow attributes.
2497  * @param[in] items
2498  *   Pointer to the list of items.
2499  * @param[in] actions
2500  *   Pointer to the list of actions.
2501  * @param[out] error
2502  *   Pointer to the error structure.
2503  *
2504  * @return
2505  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2506  */
2507 static inline struct mlx5_flow *
2508 flow_drv_prepare(struct rte_eth_dev *dev,
2509                  const struct rte_flow *flow,
2510                  const struct rte_flow_attr *attr,
2511                  const struct rte_flow_item items[],
2512                  const struct rte_flow_action actions[],
2513                  struct rte_flow_error *error)
2514 {
2515         const struct mlx5_flow_driver_ops *fops;
2516         enum mlx5_flow_drv_type type = flow->drv_type;
2517
2518         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2519         fops = flow_get_drv_ops(type);
2520         return fops->prepare(dev, attr, items, actions, error);
2521 }
2522
2523 /**
2524  * Flow driver translation API. This abstracts calling driver specific
2525  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2526  * translates a generic flow into a driver flow. flow_drv_prepare() must
2527  * precede.
2528  *
2529  * @note
2530  *   dev_flow->layers could be filled as a result of parsing during translation
2531  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2532  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2533  *   flow->actions could be overwritten even though all the expanded dev_flows
2534  *   have the same actions.
2535  *
2536  * @param[in] dev
2537  *   Pointer to the rte dev structure.
2538  * @param[in, out] dev_flow
2539  *   Pointer to the mlx5 flow.
2540  * @param[in] attr
2541  *   Pointer to the flow attributes.
2542  * @param[in] items
2543  *   Pointer to the list of items.
2544  * @param[in] actions
2545  *   Pointer to the list of actions.
2546  * @param[out] error
2547  *   Pointer to the error structure.
2548  *
2549  * @return
2550  *   0 on success, a negative errno value otherwise and rte_errno is set.
2551  */
2552 static inline int
2553 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2554                    const struct rte_flow_attr *attr,
2555                    const struct rte_flow_item items[],
2556                    const struct rte_flow_action actions[],
2557                    struct rte_flow_error *error)
2558 {
2559         const struct mlx5_flow_driver_ops *fops;
2560         enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2561
2562         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2563         fops = flow_get_drv_ops(type);
2564         return fops->translate(dev, dev_flow, attr, items, actions, error);
2565 }
2566
2567 /**
2568  * Flow driver apply API. This abstracts calling driver specific functions.
2569  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2570  * translated driver flows on to device. flow_drv_translate() must precede.
2571  *
2572  * @param[in] dev
2573  *   Pointer to Ethernet device structure.
2574  * @param[in, out] flow
2575  *   Pointer to flow structure.
2576  * @param[out] error
2577  *   Pointer to error structure.
2578  *
2579  * @return
2580  *   0 on success, a negative errno value otherwise and rte_errno is set.
2581  */
2582 static inline int
2583 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2584                struct rte_flow_error *error)
2585 {
2586         const struct mlx5_flow_driver_ops *fops;
2587         enum mlx5_flow_drv_type type = flow->drv_type;
2588
2589         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2590         fops = flow_get_drv_ops(type);
2591         return fops->apply(dev, flow, error);
2592 }
2593
2594 /**
2595  * Flow driver remove API. This abstracts calling driver specific functions.
2596  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2597  * on device. All the resources of the flow should be freed by calling
2598  * flow_drv_destroy().
2599  *
2600  * @param[in] dev
2601  *   Pointer to Ethernet device.
2602  * @param[in, out] flow
2603  *   Pointer to flow structure.
2604  */
2605 static inline void
2606 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2607 {
2608         const struct mlx5_flow_driver_ops *fops;
2609         enum mlx5_flow_drv_type type = flow->drv_type;
2610
2611         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2612         fops = flow_get_drv_ops(type);
2613         fops->remove(dev, flow);
2614 }
2615
2616 /**
2617  * Flow driver destroy API. This abstracts calling driver specific functions.
2618  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2619  * on device and releases resources of the flow.
2620  *
2621  * @param[in] dev
2622  *   Pointer to Ethernet device.
2623  * @param[in, out] flow
2624  *   Pointer to flow structure.
2625  */
2626 static inline void
2627 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2628 {
2629         const struct mlx5_flow_driver_ops *fops;
2630         enum mlx5_flow_drv_type type = flow->drv_type;
2631
2632         flow_mreg_split_qrss_release(dev, flow);
2633         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2634         fops = flow_get_drv_ops(type);
2635         fops->destroy(dev, flow);
2636 }
2637
2638 /**
2639  * Validate a flow supported by the NIC.
2640  *
2641  * @see rte_flow_validate()
2642  * @see rte_flow_ops
2643  */
2644 int
2645 mlx5_flow_validate(struct rte_eth_dev *dev,
2646                    const struct rte_flow_attr *attr,
2647                    const struct rte_flow_item items[],
2648                    const struct rte_flow_action actions[],
2649                    struct rte_flow_error *error)
2650 {
2651         int ret;
2652
2653         ret = flow_drv_validate(dev, attr, items, actions, true, error);
2654         if (ret < 0)
2655                 return ret;
2656         return 0;
2657 }
2658
2659 /**
2660  * Get RSS action from the action list.
2661  *
2662  * @param[in] actions
2663  *   Pointer to the list of actions.
2664  *
2665  * @return
2666  *   Pointer to the RSS action if exist, else return NULL.
2667  */
2668 static const struct rte_flow_action_rss*
2669 flow_get_rss_action(const struct rte_flow_action actions[])
2670 {
2671         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2672                 switch (actions->type) {
2673                 case RTE_FLOW_ACTION_TYPE_RSS:
2674                         return (const struct rte_flow_action_rss *)
2675                                actions->conf;
2676                 default:
2677                         break;
2678                 }
2679         }
2680         return NULL;
2681 }
2682
2683 static unsigned int
2684 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2685 {
2686         const struct rte_flow_item *item;
2687         unsigned int has_vlan = 0;
2688
2689         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2690                 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2691                         has_vlan = 1;
2692                         break;
2693                 }
2694         }
2695         if (has_vlan)
2696                 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2697                                        MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2698         return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2699                                MLX5_EXPANSION_ROOT_OUTER;
2700 }
2701
2702 /**
2703  *  Get layer flags from the prefix flow.
2704  *
2705  *  Some flows may be split to several subflows, the prefix subflow gets the
2706  *  match items and the suffix sub flow gets the actions.
2707  *  Some actions need the user defined match item flags to get the detail for
2708  *  the action.
2709  *  This function helps the suffix flow to get the item layer flags from prefix
2710  *  subflow.
2711  *
2712  * @param[in] dev_flow
2713  *   Pointer the created preifx subflow.
2714  *
2715  * @return
2716  *   The layers get from prefix subflow.
2717  */
2718 static inline uint64_t
2719 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2720 {
2721         uint64_t layers = 0;
2722
2723         /*
2724          * Layers bits could be localization, but usually the compiler will
2725          * help to do the optimization work for source code.
2726          * If no decap actions, use the layers directly.
2727          */
2728         if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2729                 return dev_flow->handle->layers;
2730         /* Convert L3 layers with decap action. */
2731         if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2732                 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2733         else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2734                 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2735         /* Convert L4 layers with decap action.  */
2736         if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2737                 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2738         else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2739                 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2740         return layers;
2741 }
2742
2743 /**
2744  * Get metadata split action information.
2745  *
2746  * @param[in] actions
2747  *   Pointer to the list of actions.
2748  * @param[out] qrss
2749  *   Pointer to the return pointer.
2750  * @param[out] qrss_type
2751  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2752  *   if no QUEUE/RSS is found.
2753  * @param[out] encap_idx
2754  *   Pointer to the index of the encap action if exists, otherwise the last
2755  *   action index.
2756  *
2757  * @return
2758  *   Total number of actions.
2759  */
2760 static int
2761 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2762                                        const struct rte_flow_action **qrss,
2763                                        int *encap_idx)
2764 {
2765         const struct rte_flow_action_raw_encap *raw_encap;
2766         int actions_n = 0;
2767         int raw_decap_idx = -1;
2768
2769         *encap_idx = -1;
2770         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2771                 switch (actions->type) {
2772                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2773                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2774                         *encap_idx = actions_n;
2775                         break;
2776                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2777                         raw_decap_idx = actions_n;
2778                         break;
2779                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2780                         raw_encap = actions->conf;
2781                         if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2782                                 *encap_idx = raw_decap_idx != -1 ?
2783                                                       raw_decap_idx : actions_n;
2784                         break;
2785                 case RTE_FLOW_ACTION_TYPE_QUEUE:
2786                 case RTE_FLOW_ACTION_TYPE_RSS:
2787                         *qrss = actions;
2788                         break;
2789                 default:
2790                         break;
2791                 }
2792                 actions_n++;
2793         }
2794         if (*encap_idx == -1)
2795                 *encap_idx = actions_n;
2796         /* Count RTE_FLOW_ACTION_TYPE_END. */
2797         return actions_n + 1;
2798 }
2799
2800 /**
2801  * Check meter action from the action list.
2802  *
2803  * @param[in] actions
2804  *   Pointer to the list of actions.
2805  * @param[out] mtr
2806  *   Pointer to the meter exist flag.
2807  *
2808  * @return
2809  *   Total number of actions.
2810  */
2811 static int
2812 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2813 {
2814         int actions_n = 0;
2815
2816         MLX5_ASSERT(mtr);
2817         *mtr = 0;
2818         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2819                 switch (actions->type) {
2820                 case RTE_FLOW_ACTION_TYPE_METER:
2821                         *mtr = 1;
2822                         break;
2823                 default:
2824                         break;
2825                 }
2826                 actions_n++;
2827         }
2828         /* Count RTE_FLOW_ACTION_TYPE_END. */
2829         return actions_n + 1;
2830 }
2831
2832 /**
2833  * Check if the flow should be splited due to hairpin.
2834  * The reason for the split is that in current HW we can't
2835  * support encap on Rx, so if a flow have encap we move it
2836  * to Tx.
2837  *
2838  * @param dev
2839  *   Pointer to Ethernet device.
2840  * @param[in] attr
2841  *   Flow rule attributes.
2842  * @param[in] actions
2843  *   Associated actions (list terminated by the END action).
2844  *
2845  * @return
2846  *   > 0 the number of actions and the flow should be split,
2847  *   0 when no split required.
2848  */
2849 static int
2850 flow_check_hairpin_split(struct rte_eth_dev *dev,
2851                          const struct rte_flow_attr *attr,
2852                          const struct rte_flow_action actions[])
2853 {
2854         int queue_action = 0;
2855         int action_n = 0;
2856         int encap = 0;
2857         const struct rte_flow_action_queue *queue;
2858         const struct rte_flow_action_rss *rss;
2859         const struct rte_flow_action_raw_encap *raw_encap;
2860
2861         if (!attr->ingress)
2862                 return 0;
2863         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2864                 switch (actions->type) {
2865                 case RTE_FLOW_ACTION_TYPE_QUEUE:
2866                         queue = actions->conf;
2867                         if (queue == NULL)
2868                                 return 0;
2869                         if (mlx5_rxq_get_type(dev, queue->index) !=
2870                             MLX5_RXQ_TYPE_HAIRPIN)
2871                                 return 0;
2872                         queue_action = 1;
2873                         action_n++;
2874                         break;
2875                 case RTE_FLOW_ACTION_TYPE_RSS:
2876                         rss = actions->conf;
2877                         if (rss == NULL || rss->queue_num == 0)
2878                                 return 0;
2879                         if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2880                             MLX5_RXQ_TYPE_HAIRPIN)
2881                                 return 0;
2882                         queue_action = 1;
2883                         action_n++;
2884                         break;
2885                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2886                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2887                         encap = 1;
2888                         action_n++;
2889                         break;
2890                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2891                         raw_encap = actions->conf;
2892                         if (raw_encap->size >
2893                             (sizeof(struct rte_flow_item_eth) +
2894                              sizeof(struct rte_flow_item_ipv4)))
2895                                 encap = 1;
2896                         action_n++;
2897                         break;
2898                 default:
2899                         action_n++;
2900                         break;
2901                 }
2902         }
2903         if (encap == 1 && queue_action)
2904                 return action_n;
2905         return 0;
2906 }
2907
2908 /* Declare flow create/destroy prototype in advance. */
2909 static uint32_t
2910 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
2911                  const struct rte_flow_attr *attr,
2912                  const struct rte_flow_item items[],
2913                  const struct rte_flow_action actions[],
2914                  bool external, struct rte_flow_error *error);
2915
2916 static void
2917 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
2918                   uint32_t flow_idx);
2919
2920 /**
2921  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2922  *
2923  * As mark_id is unique, if there's already a registered flow for the mark_id,
2924  * return by increasing the reference counter of the resource. Otherwise, create
2925  * the resource (mcp_res) and flow.
2926  *
2927  * Flow looks like,
2928  *   - If ingress port is ANY and reg_c[1] is mark_id,
2929  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2930  *
2931  * For default flow (zero mark_id), flow is like,
2932  *   - If ingress port is ANY,
2933  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2934  *
2935  * @param dev
2936  *   Pointer to Ethernet device.
2937  * @param mark_id
2938  *   ID of MARK action, zero means default flow for META.
2939  * @param[out] error
2940  *   Perform verbose error reporting if not NULL.
2941  *
2942  * @return
2943  *   Associated resource on success, NULL otherwise and rte_errno is set.
2944  */
2945 static struct mlx5_flow_mreg_copy_resource *
2946 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2947                           struct rte_flow_error *error)
2948 {
2949         struct mlx5_priv *priv = dev->data->dev_private;
2950         struct rte_flow_attr attr = {
2951                 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2952                 .ingress = 1,
2953         };
2954         struct mlx5_rte_flow_item_tag tag_spec = {
2955                 .data = mark_id,
2956         };
2957         struct rte_flow_item items[] = {
2958                 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2959         };
2960         struct rte_flow_action_mark ftag = {
2961                 .id = mark_id,
2962         };
2963         struct mlx5_flow_action_copy_mreg cp_mreg = {
2964                 .dst = REG_B,
2965                 .src = 0,
2966         };
2967         struct rte_flow_action_jump jump = {
2968                 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2969         };
2970         struct rte_flow_action actions[] = {
2971                 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2972         };
2973         struct mlx5_flow_mreg_copy_resource *mcp_res;
2974         uint32_t idx = 0;
2975         int ret;
2976
2977         /* Fill the register fileds in the flow. */
2978         ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2979         if (ret < 0)
2980                 return NULL;
2981         tag_spec.id = ret;
2982         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2983         if (ret < 0)
2984                 return NULL;
2985         cp_mreg.src = ret;
2986         /* Check if already registered. */
2987         MLX5_ASSERT(priv->mreg_cp_tbl);
2988         mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2989         if (mcp_res) {
2990                 /* For non-default rule. */
2991                 if (mark_id != MLX5_DEFAULT_COPY_ID)
2992                         mcp_res->refcnt++;
2993                 MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2994                             mcp_res->refcnt == 1);
2995                 return mcp_res;
2996         }
2997         /* Provide the full width of FLAG specific value. */
2998         if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2999                 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3000         /* Build a new flow. */
3001         if (mark_id != MLX5_DEFAULT_COPY_ID) {
3002                 items[0] = (struct rte_flow_item){
3003                         .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3004                         .spec = &tag_spec,
3005                 };
3006                 items[1] = (struct rte_flow_item){
3007                         .type = RTE_FLOW_ITEM_TYPE_END,
3008                 };
3009                 actions[0] = (struct rte_flow_action){
3010                         .type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3011                         .conf = &ftag,
3012                 };
3013                 actions[1] = (struct rte_flow_action){
3014                         .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3015                         .conf = &cp_mreg,
3016                 };
3017                 actions[2] = (struct rte_flow_action){
3018                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
3019                         .conf = &jump,
3020                 };
3021                 actions[3] = (struct rte_flow_action){
3022                         .type = RTE_FLOW_ACTION_TYPE_END,
3023                 };
3024         } else {
3025                 /* Default rule, wildcard match. */
3026                 attr.priority = MLX5_FLOW_PRIO_RSVD;
3027                 items[0] = (struct rte_flow_item){
3028                         .type = RTE_FLOW_ITEM_TYPE_END,
3029                 };
3030                 actions[0] = (struct rte_flow_action){
3031                         .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3032                         .conf = &cp_mreg,
3033                 };
3034                 actions[1] = (struct rte_flow_action){
3035                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
3036                         .conf = &jump,
3037                 };
3038                 actions[2] = (struct rte_flow_action){
3039                         .type = RTE_FLOW_ACTION_TYPE_END,
3040                 };
3041         }
3042         /* Build a new entry. */
3043         mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3044         if (!mcp_res) {
3045                 rte_errno = ENOMEM;
3046                 return NULL;
3047         }
3048         mcp_res->idx = idx;
3049         /*
3050          * The copy Flows are not included in any list. There
3051          * ones are referenced from other Flows and can not
3052          * be applied, removed, deleted in ardbitrary order
3053          * by list traversing.
3054          */
3055         mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3056                                          actions, false, error);
3057         if (!mcp_res->rix_flow)
3058                 goto error;
3059         mcp_res->refcnt++;
3060         mcp_res->hlist_ent.key = mark_id;
3061         ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3062                                 &mcp_res->hlist_ent);
3063         MLX5_ASSERT(!ret);
3064         if (ret)
3065                 goto error;
3066         return mcp_res;
3067 error:
3068         if (mcp_res->rix_flow)
3069                 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3070         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3071         return NULL;
3072 }
3073
3074 /**
3075  * Release flow in RX_CP_TBL.
3076  *
3077  * @param dev
3078  *   Pointer to Ethernet device.
3079  * @flow
3080  *   Parent flow for wich copying is provided.
3081  */
3082 static void
3083 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3084                           struct rte_flow *flow)
3085 {
3086         struct mlx5_flow_mreg_copy_resource *mcp_res;
3087         struct mlx5_priv *priv = dev->data->dev_private;
3088
3089         if (!flow->rix_mreg_copy)
3090                 return;
3091         mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3092                                  flow->rix_mreg_copy);
3093         if (!mcp_res || !priv->mreg_cp_tbl)
3094                 return;
3095         if (flow->copy_applied) {
3096                 MLX5_ASSERT(mcp_res->appcnt);
3097                 flow->copy_applied = 0;
3098                 --mcp_res->appcnt;
3099                 if (!mcp_res->appcnt) {
3100                         struct rte_flow *mcp_flow = mlx5_ipool_get
3101                                         (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3102                                         mcp_res->rix_flow);
3103
3104                         if (mcp_flow)
3105                                 flow_drv_remove(dev, mcp_flow);
3106                 }
3107         }
3108         /*
3109          * We do not check availability of metadata registers here,
3110          * because copy resources are not allocated in this case.
3111          */
3112         if (--mcp_res->refcnt)
3113                 return;
3114         MLX5_ASSERT(mcp_res->rix_flow);
3115         flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3116         mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3117         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3118         flow->rix_mreg_copy = 0;
3119 }
3120
3121 /**
3122  * Start flow in RX_CP_TBL.
3123  *
3124  * @param dev
3125  *   Pointer to Ethernet device.
3126  * @flow
3127  *   Parent flow for wich copying is provided.
3128  *
3129  * @return
3130  *   0 on success, a negative errno value otherwise and rte_errno is set.
3131  */
3132 static int
3133 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3134                             struct rte_flow *flow)
3135 {
3136         struct mlx5_flow_mreg_copy_resource *mcp_res;
3137         struct mlx5_priv *priv = dev->data->dev_private;
3138         int ret;
3139
3140         if (!flow->rix_mreg_copy || flow->copy_applied)
3141                 return 0;
3142         mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3143                                  flow->rix_mreg_copy);
3144         if (!mcp_res)
3145                 return 0;
3146         if (!mcp_res->appcnt) {
3147                 struct rte_flow *mcp_flow = mlx5_ipool_get
3148                                 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3149                                 mcp_res->rix_flow);
3150
3151                 if (mcp_flow) {
3152                         ret = flow_drv_apply(dev, mcp_flow, NULL);
3153                         if (ret)
3154                                 return ret;
3155                 }
3156         }
3157         ++mcp_res->appcnt;
3158         flow->copy_applied = 1;
3159         return 0;
3160 }
3161
3162 /**
3163  * Stop flow in RX_CP_TBL.
3164  *
3165  * @param dev
3166  *   Pointer to Ethernet device.
3167  * @flow
3168  *   Parent flow for wich copying is provided.
3169  */
3170 static void
3171 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3172                            struct rte_flow *flow)
3173 {
3174         struct mlx5_flow_mreg_copy_resource *mcp_res;
3175         struct mlx5_priv *priv = dev->data->dev_private;
3176
3177         if (!flow->rix_mreg_copy || !flow->copy_applied)
3178                 return;
3179         mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3180                                  flow->rix_mreg_copy);
3181         if (!mcp_res)
3182                 return;
3183         MLX5_ASSERT(mcp_res->appcnt);
3184         --mcp_res->appcnt;
3185         flow->copy_applied = 0;
3186         if (!mcp_res->appcnt) {
3187                 struct rte_flow *mcp_flow = mlx5_ipool_get
3188                                 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3189                                 mcp_res->rix_flow);
3190
3191                 if (mcp_flow)
3192                         flow_drv_remove(dev, mcp_flow);
3193         }
3194 }
3195
3196 /**
3197  * Remove the default copy action from RX_CP_TBL.
3198  *
3199  * @param dev
3200  *   Pointer to Ethernet device.
3201  */
3202 static void
3203 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3204 {
3205         struct mlx5_flow_mreg_copy_resource *mcp_res;
3206         struct mlx5_priv *priv = dev->data->dev_private;
3207
3208         /* Check if default flow is registered. */
3209         if (!priv->mreg_cp_tbl)
3210                 return;
3211         mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3212                                             MLX5_DEFAULT_COPY_ID);
3213         if (!mcp_res)
3214                 return;
3215         MLX5_ASSERT(mcp_res->rix_flow);
3216         flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3217         mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3218         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3219 }
3220
3221 /**
3222  * Add the default copy action in in RX_CP_TBL.
3223  *
3224  * @param dev
3225  *   Pointer to Ethernet device.
3226  * @param[out] error
3227  *   Perform verbose error reporting if not NULL.
3228  *
3229  * @return
3230  *   0 for success, negative value otherwise and rte_errno is set.
3231  */
3232 static int
3233 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3234                                   struct rte_flow_error *error)
3235 {
3236         struct mlx5_priv *priv = dev->data->dev_private;
3237         struct mlx5_flow_mreg_copy_resource *mcp_res;
3238
3239         /* Check whether extensive metadata feature is engaged. */
3240         if (!priv->config.dv_flow_en ||
3241             priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3242             !mlx5_flow_ext_mreg_supported(dev) ||
3243             !priv->sh->dv_regc0_mask)
3244                 return 0;
3245         mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3246         if (!mcp_res)
3247                 return -rte_errno;
3248         return 0;
3249 }
3250
3251 /**
3252  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3253  *
3254  * All the flow having Q/RSS action should be split by
3255  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3256  * performs the following,
3257  *   - CQE->flow_tag := reg_c[1] (MARK)
3258  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3259  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3260  * but there should be a flow per each MARK ID set by MARK action.
3261  *
3262  * For the aforementioned reason, if there's a MARK action in flow's action
3263  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3264  * the MARK ID to CQE's flow_tag like,
3265  *   - If reg_c[1] is mark_id,
3266  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3267  *
3268  * For SET_META action which stores value in reg_c[0], as the destination is
3269  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3270  * MARK ID means the default flow. The default flow looks like,
3271  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3272  *
3273  * @param dev
3274  *   Pointer to Ethernet device.
3275  * @param flow
3276  *   Pointer to flow structure.
3277  * @param[in] actions
3278  *   Pointer to the list of actions.
3279  * @param[out] error
3280  *   Perform verbose error reporting if not NULL.
3281  *
3282  * @return
3283  *   0 on success, negative value otherwise and rte_errno is set.
3284  */
3285 static int
3286 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3287                             struct rte_flow *flow,
3288                             const struct rte_flow_action *actions,
3289                             struct rte_flow_error *error)
3290 {
3291         struct mlx5_priv *priv = dev->data->dev_private;
3292         struct mlx5_dev_config *config = &priv->config;
3293         struct mlx5_flow_mreg_copy_resource *mcp_res;
3294         const struct rte_flow_action_mark *mark;
3295
3296         /* Check whether extensive metadata feature is engaged. */
3297         if (!config->dv_flow_en ||
3298             config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3299             !mlx5_flow_ext_mreg_supported(dev) ||
3300             !priv->sh->dv_regc0_mask)
3301                 return 0;
3302         /* Find MARK action. */
3303         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3304                 switch (actions->type) {
3305                 case RTE_FLOW_ACTION_TYPE_FLAG:
3306                         mcp_res = flow_mreg_add_copy_action
3307                                 (dev, MLX5_FLOW_MARK_DEFAULT, error);
3308                         if (!mcp_res)
3309                                 return -rte_errno;
3310                         flow->rix_mreg_copy = mcp_res->idx;
3311                         if (dev->data->dev_started) {
3312                                 mcp_res->appcnt++;
3313                                 flow->copy_applied = 1;
3314                         }
3315                         return 0;
3316                 case RTE_FLOW_ACTION_TYPE_MARK:
3317                         mark = (const struct rte_flow_action_mark *)
3318                                 actions->conf;
3319                         mcp_res =
3320                                 flow_mreg_add_copy_action(dev, mark->id, error);
3321                         if (!mcp_res)
3322                                 return -rte_errno;
3323                         flow->rix_mreg_copy = mcp_res->idx;
3324                         if (dev->data->dev_started) {
3325                                 mcp_res->appcnt++;
3326                                 flow->copy_applied = 1;
3327                         }
3328                         return 0;
3329                 default:
3330                         break;
3331                 }
3332         }
3333         return 0;
3334 }
3335
3336 #define MLX5_MAX_SPLIT_ACTIONS 24
3337 #define MLX5_MAX_SPLIT_ITEMS 24
3338
3339 /**
3340  * Split the hairpin flow.
3341  * Since HW can't support encap on Rx we move the encap to Tx.
3342  * If the count action is after the encap then we also
3343  * move the count action. in this case the count will also measure
3344  * the outer bytes.
3345  *
3346  * @param dev
3347  *   Pointer to Ethernet device.
3348  * @param[in] actions
3349  *   Associated actions (list terminated by the END action).
3350  * @param[out] actions_rx
3351  *   Rx flow actions.
3352  * @param[out] actions_tx
3353  *   Tx flow actions..
3354  * @param[out] pattern_tx
3355  *   The pattern items for the Tx flow.
3356  * @param[out] flow_id
3357  *   The flow ID connected to this flow.
3358  *
3359  * @return
3360  *   0 on success.
3361  */
3362 static int
3363 flow_hairpin_split(struct rte_eth_dev *dev,
3364                    const struct rte_flow_action actions[],
3365                    struct rte_flow_action actions_rx[],
3366                    struct rte_flow_action actions_tx[],
3367                    struct rte_flow_item pattern_tx[],
3368                    uint32_t *flow_id)
3369 {
3370         struct mlx5_priv *priv = dev->data->dev_private;
3371         const struct rte_flow_action_raw_encap *raw_encap;
3372         const struct rte_flow_action_raw_decap *raw_decap;
3373         struct mlx5_rte_flow_action_set_tag *set_tag;
3374         struct rte_flow_action *tag_action;
3375         struct mlx5_rte_flow_item_tag *tag_item;
3376         struct rte_flow_item *item;
3377         char *addr;
3378         int encap = 0;
3379
3380         mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3381         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3382                 switch (actions->type) {
3383                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3384                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3385                         rte_memcpy(actions_tx, actions,
3386                                sizeof(struct rte_flow_action));
3387                         actions_tx++;
3388                         break;
3389                 case RTE_FLOW_ACTION_TYPE_COUNT:
3390                         if (encap) {
3391                                 rte_memcpy(actions_tx, actions,
3392                                            sizeof(struct rte_flow_action));
3393                                 actions_tx++;
3394                         } else {
3395                                 rte_memcpy(actions_rx, actions,
3396                                            sizeof(struct rte_flow_action));
3397                                 actions_rx++;
3398                         }
3399                         break;
3400                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3401                         raw_encap = actions->conf;
3402                         if (raw_encap->size >
3403                             (sizeof(struct rte_flow_item_eth) +
3404                              sizeof(struct rte_flow_item_ipv4))) {
3405                                 memcpy(actions_tx, actions,
3406                                        sizeof(struct rte_flow_action));
3407                                 actions_tx++;
3408                                 encap = 1;
3409                         } else {
3410                                 rte_memcpy(actions_rx, actions,
3411                                            sizeof(struct rte_flow_action));
3412                                 actions_rx++;
3413                         }
3414                         break;
3415                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3416                         raw_decap = actions->conf;
3417                         if (raw_decap->size <
3418                             (sizeof(struct rte_flow_item_eth) +
3419                              sizeof(struct rte_flow_item_ipv4))) {
3420                                 memcpy(actions_tx, actions,
3421                                        sizeof(struct rte_flow_action));
3422                                 actions_tx++;
3423                         } else {
3424                                 rte_memcpy(actions_rx, actions,
3425                                            sizeof(struct rte_flow_action));
3426                                 actions_rx++;
3427                         }
3428                         break;
3429                 default:
3430                         rte_memcpy(actions_rx, actions,
3431                                    sizeof(struct rte_flow_action));
3432                         actions_rx++;
3433                         break;
3434                 }
3435         }
3436         /* Add set meta action and end action for the Rx flow. */
3437         tag_action = actions_rx;
3438         tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3439         actions_rx++;
3440         rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3441         actions_rx++;
3442         set_tag = (void *)actions_rx;
3443         set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3444         MLX5_ASSERT(set_tag->id > REG_NONE);
3445         set_tag->data = *flow_id;
3446         tag_action->conf = set_tag;
3447         /* Create Tx item list. */
3448         rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3449         addr = (void *)&pattern_tx[2];
3450         item = pattern_tx;
3451         item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3452         tag_item = (void *)addr;
3453         tag_item->data = *flow_id;
3454         tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3455         MLX5_ASSERT(set_tag->id > REG_NONE);
3456         item->spec = tag_item;
3457         addr += sizeof(struct mlx5_rte_flow_item_tag);
3458         tag_item = (void *)addr;
3459         tag_item->data = UINT32_MAX;
3460         tag_item->id = UINT16_MAX;
3461         item->mask = tag_item;
3462         addr += sizeof(struct mlx5_rte_flow_item_tag);
3463         item->last = NULL;
3464         item++;
3465         item->type = RTE_FLOW_ITEM_TYPE_END;
3466         return 0;
3467 }
3468
3469 /**
3470  * The last stage of splitting chain, just creates the subflow
3471  * without any modification.
3472  *
3473  * @param[in] dev
3474  *   Pointer to Ethernet device.
3475  * @param[in] flow
3476  *   Parent flow structure pointer.
3477  * @param[in, out] sub_flow
3478  *   Pointer to return the created subflow, may be NULL.
3479  * @param[in] prefix_layers
3480  *   Prefix subflow layers, may be 0.
3481  * @param[in] attr
3482  *   Flow rule attributes.
3483  * @param[in] items
3484  *   Pattern specification (list terminated by the END pattern item).
3485  * @param[in] actions
3486  *   Associated actions (list terminated by the END action).
3487  * @param[in] external
3488  *   This flow rule is created by request external to PMD.
3489  * @param[out] error
3490  *   Perform verbose error reporting if not NULL.
3491  * @return
3492  *   0 on success, negative value otherwise
3493  */
3494 static int
3495 flow_create_split_inner(struct rte_eth_dev *dev,
3496                         struct rte_flow *flow,
3497                         struct mlx5_flow **sub_flow,
3498                         uint64_t prefix_layers,
3499                         const struct rte_flow_attr *attr,
3500                         const struct rte_flow_item items[],
3501                         const struct rte_flow_action actions[],
3502                         bool external, struct rte_flow_error *error)
3503 {
3504         struct mlx5_flow *dev_flow;
3505
3506         dev_flow = flow_drv_prepare(dev, flow, attr, items, actions, error);
3507         if (!dev_flow)
3508                 return -rte_errno;
3509         dev_flow->flow = flow;
3510         dev_flow->external = external;
3511         /* Subflow object was created, we must include one in the list. */
3512         SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3513                       dev_flow->handle, next);
3514         /*
3515          * If dev_flow is as one of the suffix flow, some actions in suffix
3516          * flow may need some user defined item layer flags.
3517          */
3518         if (prefix_layers)
3519                 dev_flow->handle->layers = prefix_layers;
3520         if (sub_flow)
3521                 *sub_flow = dev_flow;
3522         return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3523 }
3524
3525 /**
3526  * Split the meter flow.
3527  *
3528  * As meter flow will split to three sub flow, other than meter
3529  * action, the other actions make sense to only meter accepts
3530  * the packet. If it need to be dropped, no other additional
3531  * actions should be take.
3532  *
3533  * One kind of special action which decapsulates the L3 tunnel
3534  * header will be in the prefix sub flow, as not to take the
3535  * L3 tunnel header into account.
3536  *
3537  * @param dev
3538  *   Pointer to Ethernet device.
3539  * @param[in] items
3540  *   Pattern specification (list terminated by the END pattern item).
3541  * @param[out] sfx_items
3542  *   Suffix flow match items (list terminated by the END pattern item).
3543  * @param[in] actions
3544  *   Associated actions (list terminated by the END action).
3545  * @param[out] actions_sfx
3546  *   Suffix flow actions.
3547  * @param[out] actions_pre
3548  *   Prefix flow actions.
3549  * @param[out] pattern_sfx
3550  *   The pattern items for the suffix flow.
3551  * @param[out] tag_sfx
3552  *   Pointer to suffix flow tag.
3553  *
3554  * @return
3555  *   0 on success.
3556  */
3557 static int
3558 flow_meter_split_prep(struct rte_eth_dev *dev,
3559                  const struct rte_flow_item items[],
3560                  struct rte_flow_item sfx_items[],
3561                  const struct rte_flow_action actions[],
3562                  struct rte_flow_action actions_sfx[],
3563                  struct rte_flow_action actions_pre[])
3564 {
3565         struct rte_flow_action *tag_action = NULL;
3566         struct rte_flow_item *tag_item;
3567         struct mlx5_rte_flow_action_set_tag *set_tag;
3568         struct rte_flow_error error;
3569         const struct rte_flow_action_raw_encap *raw_encap;
3570         const struct rte_flow_action_raw_decap *raw_decap;
3571         struct mlx5_rte_flow_item_tag *tag_spec;
3572         struct mlx5_rte_flow_item_tag *tag_mask;
3573         uint32_t tag_id;
3574         bool copy_vlan = false;
3575
3576         /* Prepare the actions for prefix and suffix flow. */
3577         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3578                 struct rte_flow_action **action_cur = NULL;
3579
3580                 switch (actions->type) {
3581                 case RTE_FLOW_ACTION_TYPE_METER:
3582                         /* Add the extra tag action first. */
3583                         tag_action = actions_pre;
3584                         tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3585                         actions_pre++;
3586                         action_cur = &actions_pre;
3587                         break;
3588                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3589                 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3590                         action_cur = &actions_pre;
3591                         break;
3592                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3593                         raw_encap = actions->conf;
3594                         if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3595                                 action_cur = &actions_pre;
3596                         break;
3597                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3598                         raw_decap = actions->conf;
3599                         if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3600                                 action_cur = &actions_pre;
3601                         break;
3602                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3603                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3604                         copy_vlan = true;
3605                         break;
3606                 default:
3607                         break;
3608                 }
3609                 if (!action_cur)
3610                         action_cur = &actions_sfx;
3611                 memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3612                 (*action_cur)++;
3613         }
3614         /* Add end action to the actions. */
3615         actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3616         actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3617         actions_pre++;
3618         /* Set the tag. */
3619         set_tag = (void *)actions_pre;
3620         set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3621         /*
3622          * Get the id from the qrss_pool to make qrss share the id with meter.
3623          */
3624         tag_id = flow_qrss_get_id(dev);
3625         set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3626         assert(tag_action);
3627         tag_action->conf = set_tag;
3628         /* Prepare the suffix subflow items. */
3629         tag_item = sfx_items++;
3630         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3631                 int item_type = items->type;
3632
3633                 switch (item_type) {
3634                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
3635                         memcpy(sfx_items, items, sizeof(*sfx_items));
3636                         sfx_items++;
3637                         break;
3638                 case RTE_FLOW_ITEM_TYPE_VLAN:
3639                         if (copy_vlan) {
3640                                 memcpy(sfx_items, items, sizeof(*sfx_items));
3641                                 /*
3642                                  * Convert to internal match item, it is used
3643                                  * for vlan push and set vid.
3644                                  */
3645                                 sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3646                                 sfx_items++;
3647                         }
3648                         break;
3649                 default:
3650                         break;
3651                 }
3652         }
3653         sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3654         sfx_items++;
3655         tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3656         tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3657         tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3658         tag_mask = tag_spec + 1;
3659         tag_mask->data = 0xffffff00;
3660         tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3661         tag_item->spec = tag_spec;
3662         tag_item->last = NULL;
3663         tag_item->mask = tag_mask;
3664         return tag_id;
3665 }
3666
3667 /**
3668  * Split action list having QUEUE/RSS for metadata register copy.
3669  *
3670  * Once Q/RSS action is detected in user's action list, the flow action
3671  * should be split in order to copy metadata registers, which will happen in
3672  * RX_CP_TBL like,
3673  *   - CQE->flow_tag := reg_c[1] (MARK)
3674  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3675  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3676  * This is because the last action of each flow must be a terminal action
3677  * (QUEUE, RSS or DROP).
3678  *
3679  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3680  * stored and kept in the mlx5_flow structure per each sub_flow.
3681  *
3682  * The Q/RSS action is replaced with,
3683  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3684  * And the following JUMP action is added at the end,
3685  *   - JUMP, to RX_CP_TBL.
3686  *
3687  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3688  * flow_create_split_metadata() routine. The flow will look like,
3689  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3690  *
3691  * @param dev
3692  *   Pointer to Ethernet device.
3693  * @param[out] split_actions
3694  *   Pointer to store split actions to jump to CP_TBL.
3695  * @param[in] actions
3696  *   Pointer to the list of original flow actions.
3697  * @param[in] qrss
3698  *   Pointer to the Q/RSS action.
3699  * @param[in] actions_n
3700  *   Number of original actions.
3701  * @param[out] error
3702  *   Perform verbose error reporting if not NULL.
3703  *
3704  * @return
3705  *   non-zero unique flow_id on success, otherwise 0 and
3706  *   error/rte_error are set.
3707  */
3708 static uint32_t
3709 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3710                           struct rte_flow_action *split_actions,
3711                           const struct rte_flow_action *actions,
3712                           const struct rte_flow_action *qrss,
3713                           int actions_n, struct rte_flow_error *error)
3714 {
3715         struct mlx5_rte_flow_action_set_tag *set_tag;
3716         struct rte_flow_action_jump *jump;
3717         const int qrss_idx = qrss - actions;
3718         uint32_t flow_id = 0;
3719         int ret = 0;
3720
3721         /*
3722          * Given actions will be split
3723          * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3724          * - Add jump to mreg CP_TBL.
3725          * As a result, there will be one more action.
3726          */
3727         ++actions_n;
3728         memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3729         set_tag = (void *)(split_actions + actions_n);
3730         /*
3731          * If tag action is not set to void(it means we are not the meter
3732          * suffix flow), add the tag action. Since meter suffix flow already
3733          * has the tag added.
3734          */
3735         if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3736                 /*
3737                  * Allocate the new subflow ID. This one is unique within
3738                  * device and not shared with representors. Otherwise,
3739                  * we would have to resolve multi-thread access synch
3740                  * issue. Each flow on the shared device is appended
3741                  * with source vport identifier, so the resulting
3742                  * flows will be unique in the shared (by master and
3743                  * representors) domain even if they have coinciding
3744                  * IDs.
3745                  */
3746                 flow_id = flow_qrss_get_id(dev);
3747                 if (!flow_id)
3748                         return rte_flow_error_set(error, ENOMEM,
3749                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3750                                                   NULL, "can't allocate id "
3751                                                   "for split Q/RSS subflow");
3752                 /* Internal SET_TAG action to set flow ID. */
3753                 *set_tag = (struct mlx5_rte_flow_action_set_tag){
3754                         .data = flow_id,
3755                 };
3756                 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3757                 if (ret < 0)
3758                         return ret;
3759                 set_tag->id = ret;
3760                 /* Construct new actions array. */
3761                 /* Replace QUEUE/RSS action. */
3762                 split_actions[qrss_idx] = (struct rte_flow_action){
3763                         .type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3764                         .conf = set_tag,
3765                 };
3766         }
3767         /* JUMP action to jump to mreg copy table (CP_TBL). */
3768         jump = (void *)(set_tag + 1);
3769         *jump = (struct rte_flow_action_jump){
3770                 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3771         };
3772         split_actions[actions_n - 2] = (struct rte_flow_action){
3773                 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3774                 .conf = jump,
3775         };
3776         split_actions[actions_n - 1] = (struct rte_flow_action){
3777                 .type = RTE_FLOW_ACTION_TYPE_END,
3778         };
3779         return flow_id;
3780 }
3781
3782 /**
3783  * Extend the given action list for Tx metadata copy.
3784  *
3785  * Copy the given action list to the ext_actions and add flow metadata register
3786  * copy action in order to copy reg_a set by WQE to reg_c[0].
3787  *
3788  * @param[out] ext_actions
3789  *   Pointer to the extended action list.
3790  * @param[in] actions
3791  *   Pointer to the list of actions.
3792  * @param[in] actions_n
3793  *   Number of actions in the list.
3794  * @param[out] error
3795  *   Perform verbose error reporting if not NULL.
3796  * @param[in] encap_idx
3797  *   The encap action inndex.
3798  *
3799  * @return
3800  *   0 on success, negative value otherwise
3801  */
3802 static int
3803 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3804                        struct rte_flow_action *ext_actions,
3805                        const struct rte_flow_action *actions,
3806                        int actions_n, struct rte_flow_error *error,
3807                        int encap_idx)
3808 {
3809         struct mlx5_flow_action_copy_mreg *cp_mreg =
3810                 (struct mlx5_flow_action_copy_mreg *)
3811                         (ext_actions + actions_n + 1);
3812         int ret;
3813
3814         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3815         if (ret < 0)
3816                 return ret;
3817         cp_mreg->dst = ret;
3818         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3819         if (ret < 0)
3820                 return ret;
3821         cp_mreg->src = ret;
3822         if (encap_idx != 0)
3823                 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3824         if (encap_idx == actions_n - 1) {
3825                 ext_actions[actions_n - 1] = (struct rte_flow_action){
3826                         .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3827                         .conf = cp_mreg,
3828                 };
3829                 ext_actions[actions_n] = (struct rte_flow_action){
3830                         .type = RTE_FLOW_ACTION_TYPE_END,
3831                 };
3832         } else {
3833                 ext_actions[encap_idx] = (struct rte_flow_action){
3834                         .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3835                         .conf = cp_mreg,
3836                 };
3837                 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3838                                 sizeof(*ext_actions) * (actions_n - encap_idx));
3839         }
3840         return 0;
3841 }
3842
3843 /**
3844  * The splitting for metadata feature.
3845  *
3846  * - Q/RSS action on NIC Rx should be split in order to pass by
3847  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3848  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3849  *
3850  * - All the actions on NIC Tx should have a mreg copy action to
3851  *   copy reg_a from WQE to reg_c[0].
3852  *
3853  * @param dev
3854  *   Pointer to Ethernet device.
3855  * @param[in] flow
3856  *   Parent flow structure pointer.
3857  * @param[in] prefix_layers
3858  *   Prefix flow layer flags.
3859  * @param[in] attr
3860  *   Flow rule attributes.
3861  * @param[in] items
3862  *   Pattern specification (list terminated by the END pattern item).
3863  * @param[in] actions
3864  *   Associated actions (list terminated by the END action).
3865  * @param[in] external
3866  *   This flow rule is created by request external to PMD.
3867  * @param[out] error
3868  *   Perform verbose error reporting if not NULL.
3869  * @return
3870  *   0 on success, negative value otherwise
3871  */
3872 static int
3873 flow_create_split_metadata(struct rte_eth_dev *dev,
3874                            struct rte_flow *flow,
3875                            uint64_t prefix_layers,
3876                            const struct rte_flow_attr *attr,
3877                            const struct rte_flow_item items[],
3878                            const struct rte_flow_action actions[],
3879                            bool external, struct rte_flow_error *error)
3880 {
3881         struct mlx5_priv *priv = dev->data->dev_private;
3882         struct mlx5_dev_config *config = &priv->config;
3883         const struct rte_flow_action *qrss = NULL;
3884         struct rte_flow_action *ext_actions = NULL;
3885         struct mlx5_flow *dev_flow = NULL;
3886         uint32_t qrss_id = 0;
3887         int mtr_sfx = 0;
3888         size_t act_size;
3889         int actions_n;
3890         int encap_idx;
3891         int ret;
3892
3893         /* Check whether extensive metadata feature is engaged. */
3894         if (!config->dv_flow_en ||
3895             config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3896             !mlx5_flow_ext_mreg_supported(dev))
3897                 return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3898                                                attr, items, actions, external,
3899                                                error);
3900         actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3901                                                            &encap_idx);
3902         if (qrss) {
3903                 /* Exclude hairpin flows from splitting. */
3904                 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3905                         const struct rte_flow_action_queue *queue;
3906
3907                         queue = qrss->conf;
3908                         if (mlx5_rxq_get_type(dev, queue->index) ==
3909                             MLX5_RXQ_TYPE_HAIRPIN)
3910                                 qrss = NULL;
3911                 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3912                         const struct rte_flow_action_rss *rss;
3913
3914                         rss = qrss->conf;
3915                         if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3916                             MLX5_RXQ_TYPE_HAIRPIN)
3917                                 qrss = NULL;
3918                 }
3919         }
3920         if (qrss) {
3921                 /* Check if it is in meter suffix table. */
3922                 mtr_sfx = attr->group == (attr->transfer ?
3923                           (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3924                           MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3925                 /*
3926                  * Q/RSS action on NIC Rx should be split in order to pass by
3927                  * the mreg copy table (RX_CP_TBL) and then it jumps to the
3928                  * action table (RX_ACT_TBL) which has the split Q/RSS action.
3929                  */
3930                 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3931                            sizeof(struct rte_flow_action_set_tag) +
3932                            sizeof(struct rte_flow_action_jump);
3933                 ext_actions = rte_zmalloc(__func__, act_size, 0);
3934                 if (!ext_actions)
3935                         return rte_flow_error_set(error, ENOMEM,
3936                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3937                                                   NULL, "no memory to split "
3938                                                   "metadata flow");
3939                 /*
3940                  * If we are the suffix flow of meter, tag already exist.
3941                  * Set the tag action to void.
3942                  */
3943                 if (mtr_sfx)
3944                         ext_actions[qrss - actions].type =
3945                                                 RTE_FLOW_ACTION_TYPE_VOID;
3946                 else
3947                         ext_actions[qrss - actions].type =
3948                                                 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3949                 /*
3950                  * Create the new actions list with removed Q/RSS action
3951                  * and appended set tag and jump to register copy table
3952                  * (RX_CP_TBL). We should preallocate unique tag ID here
3953                  * in advance, because it is needed for set tag action.
3954                  */
3955                 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3956                                                     qrss, actions_n, error);
3957                 if (!mtr_sfx && !qrss_id) {
3958                         ret = -rte_errno;
3959                         goto exit;
3960                 }
3961         } else if (attr->egress && !attr->transfer) {
3962                 /*
3963                  * All the actions on NIC Tx should have a metadata register
3964                  * copy action to copy reg_a from WQE to reg_c[meta]
3965                  */
3966                 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3967                            sizeof(struct mlx5_flow_action_copy_mreg);
3968                 ext_actions = rte_zmalloc(__func__, act_size, 0);
3969                 if (!ext_actions)
3970                         return rte_flow_error_set(error, ENOMEM,
3971                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3972                                                   NULL, "no memory to split "
3973                                                   "metadata flow");
3974                 /* Create the action list appended with copy register. */
3975                 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3976                                              actions_n, error, encap_idx);
3977                 if (ret < 0)
3978                         goto exit;
3979         }
3980         /* Add the unmodified original or prefix subflow. */
3981         ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
3982                                       items, ext_actions ? ext_actions :
3983                                       actions, external, error);
3984         if (ret < 0)
3985                 goto exit;
3986         MLX5_ASSERT(dev_flow);
3987         if (qrss) {
3988                 const struct rte_flow_attr q_attr = {
3989                         .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3990                         .ingress = 1,
3991                 };
3992                 /* Internal PMD action to set register. */
3993                 struct mlx5_rte_flow_item_tag q_tag_spec = {
3994                         .data = qrss_id,
3995                         .id = 0,
3996                 };
3997                 struct rte_flow_item q_items[] = {
3998                         {
3999                                 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4000                                 .spec = &q_tag_spec,
4001                                 .last = NULL,
4002                                 .mask = NULL,
4003                         },
4004                         {
4005                                 .type = RTE_FLOW_ITEM_TYPE_END,
4006                         },
4007                 };
4008                 struct rte_flow_action q_actions[] = {
4009                         {
4010                                 .type = qrss->type,
4011                                 .conf = qrss->conf,
4012                         },
4013                         {
4014                                 .type = RTE_FLOW_ACTION_TYPE_END,
4015                         },
4016                 };
4017                 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4018
4019                 /*
4020                  * Configure the tag item only if there is no meter subflow.
4021                  * Since tag is already marked in the meter suffix subflow
4022                  * we can just use the meter suffix items as is.
4023                  */
4024                 if (qrss_id) {
4025                         /* Not meter subflow. */
4026                         MLX5_ASSERT(!mtr_sfx);
4027                         /*
4028                          * Put unique id in prefix flow due to it is destroyed
4029                          * after suffix flow and id will be freed after there
4030                          * is no actual flows with this id and identifier
4031                          * reallocation becomes possible (for example, for
4032                          * other flows in other threads).
4033                          */
4034                         dev_flow->handle->split_flow_id = qrss_id;
4035                         ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4036                                                    error);
4037                         if (ret < 0)
4038                                 goto exit;
4039                         q_tag_spec.id = ret;
4040                 }
4041                 dev_flow = NULL;
4042                 /* Add suffix subflow to execute Q/RSS. */
4043                 ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4044                                               &q_attr, mtr_sfx ? items :
4045                                               q_items, q_actions,
4046                                               external, error);
4047                 if (ret < 0)
4048                         goto exit;
4049                 /* qrss ID should be freed if failed. */
4050                 qrss_id = 0;
4051                 MLX5_ASSERT(dev_flow);
4052         }
4053
4054 exit:
4055         /*
4056          * We do not destroy the partially created sub_flows in case of error.
4057          * These ones are included into parent flow list and will be destroyed
4058          * by flow_drv_destroy.
4059          */
4060         flow_qrss_free_id(dev, qrss_id);
4061         rte_free(ext_actions);
4062         return ret;
4063 }
4064
4065 /**
4066  * The splitting for meter feature.
4067  *
4068  * - The meter flow will be split to two flows as prefix and
4069  *   suffix flow. The packets make sense only it pass the prefix
4070  *   meter action.
4071  *
4072  * - Reg_C_5 is used for the packet to match betweend prefix and
4073  *   suffix flow.
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_meter(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         struct mlx5_priv *priv = dev->data->dev_private;
4101         struct rte_flow_action *sfx_actions = NULL;
4102         struct rte_flow_action *pre_actions = NULL;
4103         struct rte_flow_item *sfx_items = NULL;
4104         struct mlx5_flow *dev_flow = NULL;
4105         struct rte_flow_attr sfx_attr = *attr;
4106         uint32_t mtr = 0;
4107         uint32_t mtr_tag_id = 0;
4108         size_t act_size;
4109         size_t item_size;
4110         int actions_n = 0;
4111         int ret;
4112
4113         if (priv->mtr_en)
4114                 actions_n = flow_check_meter_action(actions, &mtr);
4115         if (mtr) {
4116                 /* The five prefix actions: meter, decap, encap, tag, end. */
4117                 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4118                            sizeof(struct mlx5_rte_flow_action_set_tag);
4119                 /* tag, vlan, port id, end. */
4120 #define METER_SUFFIX_ITEM 4
4121                 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4122                             sizeof(struct mlx5_rte_flow_item_tag) * 2;
4123                 sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4124                 if (!sfx_actions)
4125                         return rte_flow_error_set(error, ENOMEM,
4126                                                   RTE_FLOW_ERROR_TYPE_ACTION,
4127                                                   NULL, "no memory to split "
4128                                                   "meter flow");
4129                 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4130                              act_size);
4131                 pre_actions = sfx_actions + actions_n;
4132                 mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4133                                                    actions, sfx_actions,
4134                                                    pre_actions);
4135                 if (!mtr_tag_id) {
4136                         ret = -rte_errno;
4137                         goto exit;
4138                 }
4139                 /* Add the prefix subflow. */
4140                 ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4141                                               items, pre_actions, external,
4142                                               error);
4143                 if (ret) {
4144                         ret = -rte_errno;
4145                         goto exit;
4146                 }
4147                 dev_flow->handle->split_flow_id = mtr_tag_id;
4148                 /* Setting the sfx group atrr. */
4149                 sfx_attr.group = sfx_attr.transfer ?
4150                                 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4151                                  MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4152         }
4153         /* Add the prefix subflow. */
4154         ret = flow_create_split_metadata(dev, flow, dev_flow ?
4155                                          flow_get_prefix_layer_flags(dev_flow) :
4156                                          0, &sfx_attr,
4157                                          sfx_items ? sfx_items : items,
4158                                          sfx_actions ? sfx_actions : actions,
4159                                          external, error);
4160 exit:
4161         if (sfx_actions)
4162                 rte_free(sfx_actions);
4163         return ret;
4164 }
4165
4166 /**
4167  * Split the flow to subflow set. The splitters might be linked
4168  * in the chain, like this:
4169  * flow_create_split_outer() calls:
4170  *   flow_create_split_meter() calls:
4171  *     flow_create_split_metadata(meter_subflow_0) calls:
4172  *       flow_create_split_inner(metadata_subflow_0)
4173  *       flow_create_split_inner(metadata_subflow_1)
4174  *       flow_create_split_inner(metadata_subflow_2)
4175  *     flow_create_split_metadata(meter_subflow_1) calls:
4176  *       flow_create_split_inner(metadata_subflow_0)
4177  *       flow_create_split_inner(metadata_subflow_1)
4178  *       flow_create_split_inner(metadata_subflow_2)
4179  *
4180  * This provide flexible way to add new levels of flow splitting.
4181  * The all of successfully created subflows are included to the
4182  * parent flow dev_flow list.
4183  *
4184  * @param dev
4185  *   Pointer to Ethernet device.
4186  * @param[in] flow
4187  *   Parent flow structure pointer.
4188  * @param[in] attr
4189  *   Flow rule attributes.
4190  * @param[in] items
4191  *   Pattern specification (list terminated by the END pattern item).
4192  * @param[in] actions
4193  *   Associated actions (list terminated by the END action).
4194  * @param[in] external
4195  *   This flow rule is created by request external to PMD.
4196  * @param[out] error
4197  *   Perform verbose error reporting if not NULL.
4198  * @return
4199  *   0 on success, negative value otherwise
4200  */
4201 static int
4202 flow_create_split_outer(struct rte_eth_dev *dev,
4203                         struct rte_flow *flow,
4204                         const struct rte_flow_attr *attr,
4205                         const struct rte_flow_item items[],
4206                         const struct rte_flow_action actions[],
4207                         bool external, struct rte_flow_error *error)
4208 {
4209         int ret;
4210
4211         ret = flow_create_split_meter(dev, flow, attr, items,
4212                                          actions, external, error);
4213         MLX5_ASSERT(ret <= 0);
4214         return ret;
4215 }
4216
4217 /**
4218  * Create a flow and add it to @p list.
4219  *
4220  * @param dev
4221  *   Pointer to Ethernet device.
4222  * @param list
4223  *   Pointer to a TAILQ flow list. If this parameter NULL,
4224  *   no list insertion occurred, flow is just created,
4225  *   this is caller's responsibility to track the
4226  *   created flow.
4227  * @param[in] attr
4228  *   Flow rule attributes.
4229  * @param[in] items
4230  *   Pattern specification (list terminated by the END pattern item).
4231  * @param[in] actions
4232  *   Associated actions (list terminated by the END action).
4233  * @param[in] external
4234  *   This flow rule is created by request external to PMD.
4235  * @param[out] error
4236  *   Perform verbose error reporting if not NULL.
4237  *
4238  * @return
4239  *   A flow index on success, 0 otherwise and rte_errno is set.
4240  */
4241 static uint32_t
4242 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4243                  const struct rte_flow_attr *attr,
4244                  const struct rte_flow_item items[],
4245                  const struct rte_flow_action actions[],
4246                  bool external, struct rte_flow_error *error)
4247 {
4248         struct mlx5_priv *priv = dev->data->dev_private;
4249         struct rte_flow *flow = NULL;
4250         struct mlx5_flow *dev_flow;
4251         const struct rte_flow_action_rss *rss;
4252         union {
4253                 struct rte_flow_expand_rss buf;
4254                 uint8_t buffer[2048];
4255         } expand_buffer;
4256         union {
4257                 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4258                 uint8_t buffer[2048];
4259         } actions_rx;
4260         union {
4261                 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4262                 uint8_t buffer[2048];
4263         } actions_hairpin_tx;
4264         union {
4265                 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4266                 uint8_t buffer[2048];
4267         } items_tx;
4268         struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4269         struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4270                                               priv->rss_desc)[!!priv->flow_idx];
4271         const struct rte_flow_action *p_actions_rx = actions;
4272         uint32_t i;
4273         uint32_t idx = 0;
4274         int hairpin_flow = 0;
4275         uint32_t hairpin_id = 0;
4276         struct rte_flow_attr attr_tx = { .priority = 0 };
4277         int ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4278                                     error);
4279
4280         if (ret < 0)
4281                 return 0;
4282         hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4283         if (hairpin_flow > 0) {
4284                 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4285                         rte_errno = EINVAL;
4286                         return 0;
4287                 }
4288                 flow_hairpin_split(dev, actions, actions_rx.actions,
4289                                    actions_hairpin_tx.actions, items_tx.items,
4290                                    &hairpin_id);
4291                 p_actions_rx = actions_rx.actions;
4292         }
4293         flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4294         if (!flow) {
4295                 rte_errno = ENOMEM;
4296                 goto error_before_flow;
4297         }
4298         flow->drv_type = flow_get_drv_type(dev, attr);
4299         if (hairpin_id != 0)
4300                 flow->hairpin_flow_id = hairpin_id;
4301         MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4302                     flow->drv_type < MLX5_FLOW_TYPE_MAX);
4303         memset(rss_desc, 0, sizeof(*rss_desc));
4304         rss = flow_get_rss_action(p_actions_rx);
4305         if (rss) {
4306                 /*
4307                  * The following information is required by
4308                  * mlx5_flow_hashfields_adjust() in advance.
4309                  */
4310                 rss_desc->level = rss->level;
4311                 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4312                 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4313         }
4314         flow->dev_handles = 0;
4315         if (rss && rss->types) {
4316                 unsigned int graph_root;
4317
4318                 graph_root = find_graph_root(items, rss->level);
4319                 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4320                                           items, rss->types,
4321                                           mlx5_support_expansion,
4322                                           graph_root);
4323                 MLX5_ASSERT(ret > 0 &&
4324                        (unsigned int)ret < sizeof(expand_buffer.buffer));
4325         } else {
4326                 buf->entries = 1;
4327                 buf->entry[0].pattern = (void *)(uintptr_t)items;
4328         }
4329         /*
4330          * Record the start index when there is a nested call. All sub-flows
4331          * need to be translated before another calling.
4332          * No need to use ping-pong buffer to save memory here.
4333          */
4334         if (priv->flow_idx) {
4335                 MLX5_ASSERT(!priv->flow_nested_idx);
4336                 priv->flow_nested_idx = priv->flow_idx;
4337         }
4338         for (i = 0; i < buf->entries; ++i) {
4339                 /*
4340                  * The splitter may create multiple dev_flows,
4341                  * depending on configuration. In the simplest
4342                  * case it just creates unmodified original flow.
4343                  */
4344                 ret = flow_create_split_outer(dev, flow, attr,
4345                                               buf->entry[i].pattern,
4346                                               p_actions_rx, external,
4347                                               error);
4348                 if (ret < 0)
4349                         goto error;
4350         }
4351         /* Create the tx flow. */
4352         if (hairpin_flow) {
4353                 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4354                 attr_tx.ingress = 0;
4355                 attr_tx.egress = 1;
4356                 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4357                                             actions_hairpin_tx.actions, error);
4358                 if (!dev_flow)
4359                         goto error;
4360                 dev_flow->flow = flow;
4361                 dev_flow->external = 0;
4362                 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4363                               dev_flow->handle, next);
4364                 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4365                                          items_tx.items,
4366                                          actions_hairpin_tx.actions, error);
4367                 if (ret < 0)
4368                         goto error;
4369         }
4370         /*
4371          * Update the metadata register copy table. If extensive
4372          * metadata feature is enabled and registers are supported
4373          * we might create the extra rte_flow for each unique
4374          * MARK/FLAG action ID.
4375          *
4376          * The table is updated for ingress Flows only, because
4377          * the egress Flows belong to the different device and
4378          * copy table should be updated in peer NIC Rx domain.
4379          */
4380         if (attr->ingress &&
4381             (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4382                 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4383                 if (ret)
4384                         goto error;
4385         }
4386         /*
4387          * If the flow is external (from application) OR device is started, then
4388          * the flow will be applied immediately.
4389          */
4390         if (external || dev->data->dev_started) {
4391                 ret = flow_drv_apply(dev, flow, error);
4392                 if (ret < 0)
4393                         goto error;
4394         }
4395         if (list)
4396                 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4397                              flow, next);
4398         flow_rxq_flags_set(dev, flow);
4399         /* Nested flow creation index recovery. */
4400         priv->flow_idx = priv->flow_nested_idx;
4401         if (priv->flow_nested_idx)
4402                 priv->flow_nested_idx = 0;
4403         return idx;
4404 error:
4405         MLX5_ASSERT(flow);
4406         ret = rte_errno; /* Save rte_errno before cleanup. */
4407         flow_mreg_del_copy_action(dev, flow);
4408         flow_drv_destroy(dev, flow);
4409         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4410         rte_errno = ret; /* Restore rte_errno. */
4411 error_before_flow:
4412         ret = rte_errno;
4413         if (hairpin_id)
4414                 mlx5_flow_id_release(priv->sh->flow_id_pool,
4415                                      hairpin_id);
4416         rte_errno = ret;
4417         priv->flow_idx = priv->flow_nested_idx;
4418         if (priv->flow_nested_idx)
4419                 priv->flow_nested_idx = 0;
4420         return 0;
4421 }
4422
4423 /**
4424  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4425  * incoming packets to table 1.
4426  *
4427  * Other flow rules, requested for group n, will be created in
4428  * e-switch table n+1.
4429  * Jump action to e-switch group n will be created to group n+1.
4430  *
4431  * Used when working in switchdev mode, to utilise advantages of table 1
4432  * and above.
4433  *
4434  * @param dev
4435  *   Pointer to Ethernet device.
4436  *
4437  * @return
4438  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4439  */
4440 struct rte_flow *
4441 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4442 {
4443         const struct rte_flow_attr attr = {
4444                 .group = 0,
4445                 .priority = 0,
4446                 .ingress = 1,
4447                 .egress = 0,
4448                 .transfer = 1,
4449         };
4450         const struct rte_flow_item pattern = {
4451                 .type = RTE_FLOW_ITEM_TYPE_END,
4452         };
4453         struct rte_flow_action_jump jump = {
4454                 .group = 1,
4455         };
4456         const struct rte_flow_action actions[] = {
4457                 {
4458                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
4459                         .conf = &jump,
4460                 },
4461                 {
4462                         .type = RTE_FLOW_ACTION_TYPE_END,
4463                 },
4464         };
4465         struct mlx5_priv *priv = dev->data->dev_private;
4466         struct rte_flow_error error;
4467
4468         return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4469                                                    &attr, &pattern,
4470                                                    actions, false, &error);
4471 }
4472
4473 /**
4474  * Create a flow.
4475  *
4476  * @see rte_flow_create()
4477  * @see rte_flow_ops
4478  */
4479 struct rte_flow *
4480 mlx5_flow_create(struct rte_eth_dev *dev,
4481                  const struct rte_flow_attr *attr,
4482                  const struct rte_flow_item items[],
4483                  const struct rte_flow_action actions[],
4484                  struct rte_flow_error *error)
4485 {
4486         struct mlx5_priv *priv = dev->data->dev_private;
4487
4488         /*
4489          * If the device is not started yet, it is not allowed to created a
4490          * flow from application. PMD default flows and traffic control flows
4491          * are not affected.
4492          */
4493         if (unlikely(!dev->data->dev_started)) {
4494                 rte_errno = ENODEV;
4495                 DRV_LOG(DEBUG, "port %u is not started when "
4496                         "inserting a flow", dev->data->port_id);
4497                 return NULL;
4498         }
4499         return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4500                                   attr, items, actions, true, error);
4501 }
4502
4503 /**
4504  * Destroy a flow in a list.
4505  *
4506  * @param dev
4507  *   Pointer to Ethernet device.
4508  * @param list
4509  *   Pointer to the Indexed flow list. If this parameter NULL,
4510  *   there is no flow removal from the list. Be noted that as
4511  *   flow is add to the indexed list, memory of the indexed
4512  *   list points to maybe changed as flow destroyed.
4513  * @param[in] flow_idx
4514  *   Index of flow to destroy.
4515  */
4516 static void
4517 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4518                   uint32_t flow_idx)
4519 {
4520         struct mlx5_priv *priv = dev->data->dev_private;
4521         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4522         struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4523                                                [MLX5_IPOOL_RTE_FLOW], flow_idx);
4524
4525         if (!flow)
4526                 return;
4527         /*
4528          * Update RX queue flags only if port is started, otherwise it is
4529          * already clean.
4530          */
4531         if (dev->data->dev_started)
4532                 flow_rxq_flags_trim(dev, flow);
4533         if (flow->hairpin_flow_id)
4534                 mlx5_flow_id_release(priv->sh->flow_id_pool,
4535                                      flow->hairpin_flow_id);
4536         flow_drv_destroy(dev, flow);
4537         if (list)
4538                 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4539                              flow_idx, flow, next);
4540         flow_mreg_del_copy_action(dev, flow);
4541         if (flow->fdir) {
4542                 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4543                         if (priv_fdir_flow->rix_flow == flow_idx)
4544                                 break;
4545                 }
4546                 if (priv_fdir_flow) {
4547                         LIST_REMOVE(priv_fdir_flow, next);
4548                         rte_free(priv_fdir_flow->fdir);
4549                         rte_free(priv_fdir_flow);
4550                 }
4551         }
4552         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4553 }
4554
4555 /**
4556  * Destroy all flows.
4557  *
4558  * @param dev
4559  *   Pointer to Ethernet device.
4560  * @param list
4561  *   Pointer to the Indexed flow list.
4562  * @param active
4563  *   If flushing is called avtively.
4564  */
4565 void
4566 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4567 {
4568         uint32_t num_flushed = 0;
4569
4570         while (*list) {
4571                 flow_list_destroy(dev, list, *list);
4572                 num_flushed++;
4573         }
4574         if (active) {
4575                 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4576                         dev->data->port_id, num_flushed);
4577         }
4578 }
4579
4580 /**
4581  * Remove all flows.
4582  *
4583  * @param dev
4584  *   Pointer to Ethernet device.
4585  * @param list
4586  *   Pointer to the Indexed flow list.
4587  */
4588 void
4589 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4590 {
4591         struct mlx5_priv *priv = dev->data->dev_private;
4592         struct rte_flow *flow = NULL;
4593         uint32_t idx;
4594
4595         ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4596                       flow, next) {
4597                 flow_drv_remove(dev, flow);
4598                 flow_mreg_stop_copy_action(dev, flow);
4599         }
4600         flow_mreg_del_default_copy_action(dev);
4601         flow_rxq_flags_clear(dev);
4602 }
4603
4604 /**
4605  * Add all flows.
4606  *
4607  * @param dev
4608  *   Pointer to Ethernet device.
4609  * @param list
4610  *   Pointer to the Indexed flow list.
4611  *
4612  * @return
4613  *   0 on success, a negative errno value otherwise and rte_errno is set.
4614  */
4615 int
4616 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4617 {
4618         struct mlx5_priv *priv = dev->data->dev_private;
4619         struct rte_flow *flow = NULL;
4620         struct rte_flow_error error;
4621         uint32_t idx;
4622         int ret = 0;
4623
4624         /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4625         ret = flow_mreg_add_default_copy_action(dev, &error);
4626         if (ret < 0)
4627                 return -rte_errno;
4628         /* Apply Flows created by application. */
4629         ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4630                       flow, next) {
4631                 ret = flow_mreg_start_copy_action(dev, flow);
4632                 if (ret < 0)
4633                         goto error;
4634                 ret = flow_drv_apply(dev, flow, &error);
4635                 if (ret < 0)
4636                         goto error;
4637                 flow_rxq_flags_set(dev, flow);
4638         }
4639         return 0;
4640 error:
4641         ret = rte_errno; /* Save rte_errno before cleanup. */
4642         mlx5_flow_stop(dev, list);
4643         rte_errno = ret; /* Restore rte_errno. */
4644         return -rte_errno;
4645 }
4646
4647 /**
4648  * Stop all default actions for flows.
4649  *
4650  * @param dev
4651  *   Pointer to Ethernet device.
4652  */
4653 void
4654 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4655 {
4656         flow_mreg_del_default_copy_action(dev);
4657 }
4658
4659 /**
4660  * Start all default actions for flows.
4661  *
4662  * @param dev
4663  *   Pointer to Ethernet device.
4664  * @return
4665  *   0 on success, a negative errno value otherwise and rte_errno is set.
4666  */
4667 int
4668 mlx5_flow_start_default(struct rte_eth_dev *dev)
4669 {
4670         struct rte_flow_error error;
4671
4672         /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4673         return flow_mreg_add_default_copy_action(dev, &error);
4674 }
4675
4676 /**
4677  * Allocate intermediate resources for flow creation.
4678  *
4679  * @param dev
4680  *   Pointer to Ethernet device.
4681  */
4682 void
4683 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4684 {
4685         struct mlx5_priv *priv = dev->data->dev_private;
4686
4687         if (!priv->inter_flows) {
4688                 priv->inter_flows = rte_calloc(__func__, 1,
4689                                     MLX5_NUM_MAX_DEV_FLOWS *
4690                                     sizeof(struct mlx5_flow) +
4691                                     (sizeof(struct mlx5_flow_rss_desc) +
4692                                     sizeof(uint16_t) * UINT16_MAX) * 2, 0);
4693                 if (!priv->inter_flows) {
4694                         DRV_LOG(ERR, "can't allocate intermediate memory.");
4695                         return;
4696                 }
4697         }
4698         priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4699                          [MLX5_NUM_MAX_DEV_FLOWS];
4700         /* Reset the index. */
4701         priv->flow_idx = 0;
4702         priv->flow_nested_idx = 0;
4703 }
4704
4705 /**
4706  * Free intermediate resources for flows.
4707  *
4708  * @param dev
4709  *   Pointer to Ethernet device.
4710  */
4711 void
4712 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4713 {
4714         struct mlx5_priv *priv = dev->data->dev_private;
4715
4716         rte_free(priv->inter_flows);
4717         priv->inter_flows = NULL;
4718 }
4719
4720 /**
4721  * Verify the flow list is empty
4722  *
4723  * @param dev
4724  *  Pointer to Ethernet device.
4725  *
4726  * @return the number of flows not released.
4727  */
4728 int
4729 mlx5_flow_verify(struct rte_eth_dev *dev)
4730 {
4731         struct mlx5_priv *priv = dev->data->dev_private;
4732         struct rte_flow *flow;
4733         uint32_t idx;
4734         int ret = 0;
4735
4736         ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4737                       flow, next) {
4738                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
4739                         dev->data->port_id, (void *)flow);
4740                 ++ret;
4741         }
4742         return ret;
4743 }
4744
4745 /**
4746  * Enable default hairpin egress flow.
4747  *
4748  * @param dev
4749  *   Pointer to Ethernet device.
4750  * @param queue
4751  *   The queue index.
4752  *
4753  * @return
4754  *   0 on success, a negative errno value otherwise and rte_errno is set.
4755  */
4756 int
4757 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4758                             uint32_t queue)
4759 {
4760         struct mlx5_priv *priv = dev->data->dev_private;
4761         const struct rte_flow_attr attr = {
4762                 .egress = 1,
4763                 .priority = 0,
4764         };
4765         struct mlx5_rte_flow_item_tx_queue queue_spec = {
4766                 .queue = queue,
4767         };
4768         struct mlx5_rte_flow_item_tx_queue queue_mask = {
4769                 .queue = UINT32_MAX,
4770         };
4771         struct rte_flow_item items[] = {
4772                 {
4773                         .type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4774                         .spec = &queue_spec,
4775                         .last = NULL,
4776                         .mask = &queue_mask,
4777                 },
4778                 {
4779                         .type = RTE_FLOW_ITEM_TYPE_END,
4780                 },
4781         };
4782         struct rte_flow_action_jump jump = {
4783                 .group = MLX5_HAIRPIN_TX_TABLE,
4784         };
4785         struct rte_flow_action actions[2];
4786         uint32_t flow_idx;
4787         struct rte_flow_error error;
4788
4789         actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4790         actions[0].conf = &jump;
4791         actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4792         flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4793                                 &attr, items, actions, false, &error);
4794         if (!flow_idx) {
4795                 DRV_LOG(DEBUG,
4796                         "Failed to create ctrl flow: rte_errno(%d),"
4797                         " type(%d), message(%s)",
4798                         rte_errno, error.type,
4799                         error.message ? error.message : " (no stated reason)");
4800                 return -rte_errno;
4801         }
4802         return 0;
4803 }
4804
4805 /**
4806  * Enable a control flow configured from the control plane.
4807  *
4808  * @param dev
4809  *   Pointer to Ethernet device.
4810  * @param eth_spec
4811  *   An Ethernet flow spec to apply.
4812  * @param eth_mask
4813  *   An Ethernet flow mask to apply.
4814  * @param vlan_spec
4815  *   A VLAN flow spec to apply.
4816  * @param vlan_mask
4817  *   A VLAN flow mask to apply.
4818  *
4819  * @return
4820  *   0 on success, a negative errno value otherwise and rte_errno is set.
4821  */
4822 int
4823 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4824                     struct rte_flow_item_eth *eth_spec,
4825                     struct rte_flow_item_eth *eth_mask,
4826                     struct rte_flow_item_vlan *vlan_spec,
4827                     struct rte_flow_item_vlan *vlan_mask)
4828 {
4829         struct mlx5_priv *priv = dev->data->dev_private;
4830         const struct rte_flow_attr attr = {
4831                 .ingress = 1,
4832                 .priority = MLX5_FLOW_PRIO_RSVD,
4833         };
4834         struct rte_flow_item items[] = {
4835                 {
4836                         .type = RTE_FLOW_ITEM_TYPE_ETH,
4837                         .spec = eth_spec,
4838                         .last = NULL,
4839                         .mask = eth_mask,
4840                 },
4841                 {
4842                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4843                                               RTE_FLOW_ITEM_TYPE_END,
4844                         .spec = vlan_spec,
4845                         .last = NULL,
4846                         .mask = vlan_mask,
4847                 },
4848                 {
4849                         .type = RTE_FLOW_ITEM_TYPE_END,
4850                 },
4851         };
4852         uint16_t queue[priv->reta_idx_n];
4853         struct rte_flow_action_rss action_rss = {
4854                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4855                 .level = 0,
4856                 .types = priv->rss_conf.rss_hf,
4857                 .key_len = priv->rss_conf.rss_key_len,
4858                 .queue_num = priv->reta_idx_n,
4859                 .key = priv->rss_conf.rss_key,
4860                 .queue = queue,
4861         };
4862         struct rte_flow_action actions[] = {
4863                 {
4864                         .type = RTE_FLOW_ACTION_TYPE_RSS,
4865                         .conf = &action_rss,
4866                 },
4867                 {
4868                         .type = RTE_FLOW_ACTION_TYPE_END,
4869                 },
4870         };
4871         uint32_t flow_idx;
4872         struct rte_flow_error error;
4873         unsigned int i;
4874
4875         if (!priv->reta_idx_n || !priv->rxqs_n) {
4876                 return 0;
4877         }
4878         for (i = 0; i != priv->reta_idx_n; ++i)
4879                 queue[i] = (*priv->reta_idx)[i];
4880         flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4881                                 &attr, items, actions, false, &error);
4882         if (!flow_idx)
4883                 return -rte_errno;
4884         return 0;
4885 }
4886
4887 /**
4888  * Enable a flow control configured from the control plane.
4889  *
4890  * @param dev
4891  *   Pointer to Ethernet device.
4892  * @param eth_spec
4893  *   An Ethernet flow spec to apply.
4894  * @param eth_mask
4895  *   An Ethernet flow mask to apply.
4896  *
4897  * @return
4898  *   0 on success, a negative errno value otherwise and rte_errno is set.
4899  */
4900 int
4901 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4902                struct rte_flow_item_eth *eth_spec,
4903                struct rte_flow_item_eth *eth_mask)
4904 {
4905         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4906 }
4907
4908 /**
4909  * Destroy a flow.
4910  *
4911  * @see rte_flow_destroy()
4912  * @see rte_flow_ops
4913  */
4914 int
4915 mlx5_flow_destroy(struct rte_eth_dev *dev,
4916                   struct rte_flow *flow,
4917                   struct rte_flow_error *error __rte_unused)
4918 {
4919         struct mlx5_priv *priv = dev->data->dev_private;
4920
4921         flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
4922         return 0;
4923 }
4924
4925 /**
4926  * Destroy all flows.
4927  *
4928  * @see rte_flow_flush()
4929  * @see rte_flow_ops
4930  */
4931 int
4932 mlx5_flow_flush(struct rte_eth_dev *dev,
4933                 struct rte_flow_error *error __rte_unused)
4934 {
4935         struct mlx5_priv *priv = dev->data->dev_private;
4936
4937         mlx5_flow_list_flush(dev, &priv->flows, false);
4938         return 0;
4939 }
4940
4941 /**
4942  * Isolated mode.
4943  *
4944  * @see rte_flow_isolate()
4945  * @see rte_flow_ops
4946  */
4947 int
4948 mlx5_flow_isolate(struct rte_eth_dev *dev,
4949                   int enable,
4950                   struct rte_flow_error *error)
4951 {
4952         struct mlx5_priv *priv = dev->data->dev_private;
4953
4954         if (dev->data->dev_started) {
4955                 rte_flow_error_set(error, EBUSY,
4956                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4957                                    NULL,
4958                                    "port must be stopped first");
4959                 return -rte_errno;
4960         }
4961         priv->isolated = !!enable;
4962         if (enable)
4963                 dev->dev_ops = &mlx5_dev_ops_isolate;
4964         else
4965                 dev->dev_ops = &mlx5_dev_ops;
4966         return 0;
4967 }
4968
4969 /**
4970  * Query a flow.
4971  *
4972  * @see rte_flow_query()
4973  * @see rte_flow_ops
4974  */
4975 static int
4976 flow_drv_query(struct rte_eth_dev *dev,
4977                uint32_t flow_idx,
4978                const struct rte_flow_action *actions,
4979                void *data,
4980                struct rte_flow_error *error)
4981 {
4982         struct mlx5_priv *priv = dev->data->dev_private;
4983         const struct mlx5_flow_driver_ops *fops;
4984         struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4985                                                [MLX5_IPOOL_RTE_FLOW],
4986                                                flow_idx);
4987         enum mlx5_flow_drv_type ftype;
4988
4989         if (!flow) {
4990                 return rte_flow_error_set(error, ENOENT,
4991                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4992                           NULL,
4993                           "invalid flow handle");
4994         }
4995         ftype = flow->drv_type;
4996         MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4997         fops = flow_get_drv_ops(ftype);
4998
4999         return fops->query(dev, flow, actions, data, error);
5000 }
5001
5002 /**
5003  * Query a flow.
5004  *
5005  * @see rte_flow_query()
5006  * @see rte_flow_ops
5007  */
5008 int
5009 mlx5_flow_query(struct rte_eth_dev *dev,
5010                 struct rte_flow *flow,
5011                 const struct rte_flow_action *actions,
5012                 void *data,
5013                 struct rte_flow_error *error)
5014 {
5015         int ret;
5016
5017         ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5018                              error);
5019         if (ret < 0)
5020                 return ret;
5021         return 0;
5022 }
5023
5024 /**
5025  * Convert a flow director filter to a generic flow.
5026  *
5027  * @param dev
5028  *   Pointer to Ethernet device.
5029  * @param fdir_filter
5030  *   Flow director filter to add.
5031  * @param attributes
5032  *   Generic flow parameters structure.
5033  *
5034  * @return
5035  *   0 on success, a negative errno value otherwise and rte_errno is set.
5036  */
5037 static int
5038 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5039                          const struct rte_eth_fdir_filter *fdir_filter,
5040                          struct mlx5_fdir *attributes)
5041 {
5042         struct mlx5_priv *priv = dev->data->dev_private;
5043         const struct rte_eth_fdir_input *input = &fdir_filter->input;
5044         const struct rte_eth_fdir_masks *mask =
5045                 &dev->data->dev_conf.fdir_conf.mask;
5046
5047         /* Validate queue number. */
5048         if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5049                 DRV_LOG(ERR, "port %u invalid queue number %d",
5050                         dev->data->port_id, fdir_filter->action.rx_queue);
5051                 rte_errno = EINVAL;
5052                 return -rte_errno;
5053         }
5054         attributes->attr.ingress = 1;
5055         attributes->items[0] = (struct rte_flow_item) {
5056                 .type = RTE_FLOW_ITEM_TYPE_ETH,
5057                 .spec = &attributes->l2,
5058                 .mask = &attributes->l2_mask,
5059         };
5060         switch (fdir_filter->action.behavior) {
5061         case RTE_ETH_FDIR_ACCEPT:
5062                 attributes->actions[0] = (struct rte_flow_action){
5063                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
5064                         .conf = &attributes->queue,
5065                 };
5066                 break;
5067         case RTE_ETH_FDIR_REJECT:
5068                 attributes->actions[0] = (struct rte_flow_action){
5069                         .type = RTE_FLOW_ACTION_TYPE_DROP,
5070                 };
5071                 break;
5072         default:
5073                 DRV_LOG(ERR, "port %u invalid behavior %d",
5074                         dev->data->port_id,
5075                         fdir_filter->action.behavior);
5076                 rte_errno = ENOTSUP;
5077                 return -rte_errno;
5078         }
5079         attributes->queue.index = fdir_filter->action.rx_queue;
5080         /* Handle L3. */
5081         switch (fdir_filter->input.flow_type) {
5082         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5083         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5084         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5085                 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5086                         .src_addr = input->flow.ip4_flow.src_ip,
5087                         .dst_addr = input->flow.ip4_flow.dst_ip,
5088                         .time_to_live = input->flow.ip4_flow.ttl,
5089                         .type_of_service = input->flow.ip4_flow.tos,
5090                 };
5091                 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5092                         .src_addr = mask->ipv4_mask.src_ip,
5093                         .dst_addr = mask->ipv4_mask.dst_ip,
5094                         .time_to_live = mask->ipv4_mask.ttl,
5095                         .type_of_service = mask->ipv4_mask.tos,
5096                         .next_proto_id = mask->ipv4_mask.proto,
5097                 };
5098                 attributes->items[1] = (struct rte_flow_item){
5099                         .type = RTE_FLOW_ITEM_TYPE_IPV4,
5100                         .spec = &attributes->l3,
5101                         .mask = &attributes->l3_mask,
5102                 };
5103                 break;
5104         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5105         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5106         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5107                 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5108                         .hop_limits = input->flow.ipv6_flow.hop_limits,
5109                         .proto = input->flow.ipv6_flow.proto,
5110                 };
5111
5112                 memcpy(attributes->l3.ipv6.hdr.src_addr,
5113                        input->flow.ipv6_flow.src_ip,
5114                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5115                 memcpy(attributes->l3.ipv6.hdr.dst_addr,
5116                        input->flow.ipv6_flow.dst_ip,
5117                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5118                 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5119                        mask->ipv6_mask.src_ip,
5120                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5121                 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5122                        mask->ipv6_mask.dst_ip,
5123                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5124                 attributes->items[1] = (struct rte_flow_item){
5125                         .type = RTE_FLOW_ITEM_TYPE_IPV6,
5126                         .spec = &attributes->l3,
5127                         .mask = &attributes->l3_mask,
5128                 };
5129                 break;
5130         default:
5131                 DRV_LOG(ERR, "port %u invalid flow type%d",
5132                         dev->data->port_id, fdir_filter->input.flow_type);
5133                 rte_errno = ENOTSUP;
5134                 return -rte_errno;
5135         }
5136         /* Handle L4. */
5137         switch (fdir_filter->input.flow_type) {
5138         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5139                 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5140                         .src_port = input->flow.udp4_flow.src_port,
5141                         .dst_port = input->flow.udp4_flow.dst_port,
5142                 };
5143                 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5144                         .src_port = mask->src_port_mask,
5145                         .dst_port = mask->dst_port_mask,
5146                 };
5147                 attributes->items[2] = (struct rte_flow_item){
5148                         .type = RTE_FLOW_ITEM_TYPE_UDP,
5149                         .spec = &attributes->l4,
5150                         .mask = &attributes->l4_mask,
5151                 };
5152                 break;
5153         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5154                 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5155                         .src_port = input->flow.tcp4_flow.src_port,
5156                         .dst_port = input->flow.tcp4_flow.dst_port,
5157                 };
5158                 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5159                         .src_port = mask->src_port_mask,
5160                         .dst_port = mask->dst_port_mask,
5161                 };
5162                 attributes->items[2] = (struct rte_flow_item){
5163                         .type = RTE_FLOW_ITEM_TYPE_TCP,
5164                         .spec = &attributes->l4,
5165                         .mask = &attributes->l4_mask,
5166                 };
5167                 break;
5168         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5169                 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5170                         .src_port = input->flow.udp6_flow.src_port,
5171                         .dst_port = input->flow.udp6_flow.dst_port,
5172                 };
5173                 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5174                         .src_port = mask->src_port_mask,
5175                         .dst_port = mask->dst_port_mask,
5176                 };
5177                 attributes->items[2] = (struct rte_flow_item){
5178                         .type = RTE_FLOW_ITEM_TYPE_UDP,
5179                         .spec = &attributes->l4,
5180                         .mask = &attributes->l4_mask,
5181                 };
5182                 break;
5183         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5184                 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5185                         .src_port = input->flow.tcp6_flow.src_port,
5186                         .dst_port = input->flow.tcp6_flow.dst_port,
5187                 };
5188                 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5189                         .src_port = mask->src_port_mask,
5190                         .dst_port = mask->dst_port_mask,
5191                 };
5192                 attributes->items[2] = (struct rte_flow_item){
5193                         .type = RTE_FLOW_ITEM_TYPE_TCP,
5194                         .spec = &attributes->l4,
5195                         .mask = &attributes->l4_mask,
5196                 };
5197                 break;
5198         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5199         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5200                 break;
5201         default:
5202                 DRV_LOG(ERR, "port %u invalid flow type%d",
5203                         dev->data->port_id, fdir_filter->input.flow_type);
5204                 rte_errno = ENOTSUP;
5205                 return -rte_errno;
5206         }
5207         return 0;
5208 }
5209
5210 #define FLOW_FDIR_CMP(f1, f2, fld) \
5211         memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5212
5213 /**
5214  * Compare two FDIR flows. If items and actions are identical, the two flows are
5215  * regarded as same.
5216  *
5217  * @param dev
5218  *   Pointer to Ethernet device.
5219  * @param f1
5220  *   FDIR flow to compare.
5221  * @param f2
5222  *   FDIR flow to compare.
5223  *
5224  * @return
5225  *   Zero on match, 1 otherwise.
5226  */
5227 static int
5228 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5229 {
5230         if (FLOW_FDIR_CMP(f1, f2, attr) ||
5231             FLOW_FDIR_CMP(f1, f2, l2) ||
5232             FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5233             FLOW_FDIR_CMP(f1, f2, l3) ||
5234             FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5235             FLOW_FDIR_CMP(f1, f2, l4) ||
5236             FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5237             FLOW_FDIR_CMP(f1, f2, actions[0].type))
5238                 return 1;
5239         if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5240             FLOW_FDIR_CMP(f1, f2, queue))
5241                 return 1;
5242         return 0;
5243 }
5244
5245 /**
5246  * Search device flow list to find out a matched FDIR flow.
5247  *
5248  * @param dev
5249  *   Pointer to Ethernet device.
5250  * @param fdir_flow
5251  *   FDIR flow to lookup.
5252  *
5253  * @return
5254  *   Index of flow if found, 0 otherwise.
5255  */
5256 static uint32_t
5257 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5258 {
5259         struct mlx5_priv *priv = dev->data->dev_private;
5260         uint32_t flow_idx = 0;
5261         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5262
5263         MLX5_ASSERT(fdir_flow);
5264         LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5265                 if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5266                         DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5267                                 dev->data->port_id, flow_idx);
5268                         flow_idx = priv_fdir_flow->rix_flow;
5269                         break;
5270                 }
5271         }
5272         return flow_idx;
5273 }
5274
5275 /**
5276  * Add new flow director filter and store it in list.
5277  *
5278  * @param dev
5279  *   Pointer to Ethernet device.
5280  * @param fdir_filter
5281  *   Flow director filter to add.
5282  *
5283  * @return
5284  *   0 on success, a negative errno value otherwise and rte_errno is set.
5285  */
5286 static int
5287 flow_fdir_filter_add(struct rte_eth_dev *dev,
5288                      const struct rte_eth_fdir_filter *fdir_filter)
5289 {
5290         struct mlx5_priv *priv = dev->data->dev_private;
5291         struct mlx5_fdir *fdir_flow;
5292         struct rte_flow *flow;
5293         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5294         uint32_t flow_idx;
5295         int ret;
5296
5297         fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5298         if (!fdir_flow) {
5299                 rte_errno = ENOMEM;
5300                 return -rte_errno;
5301         }
5302         ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5303         if (ret)
5304                 goto error;
5305         flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5306         if (flow_idx) {
5307                 rte_errno = EEXIST;
5308                 goto error;
5309         }
5310         priv_fdir_flow = rte_zmalloc(__func__, sizeof(struct mlx5_fdir_flow),
5311                                      0);
5312         if (!priv_fdir_flow) {
5313                 rte_errno = ENOMEM;
5314                 goto error;
5315         }
5316         flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5317                                     fdir_flow->items, fdir_flow->actions, true,
5318                                     NULL);
5319         flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5320         if (!flow)
5321                 goto error;
5322         flow->fdir = 1;
5323         priv_fdir_flow->fdir = fdir_flow;
5324         priv_fdir_flow->rix_flow = flow_idx;
5325         LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5326         DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5327                 dev->data->port_id, (void *)flow);
5328         return 0;
5329 error:
5330         rte_free(priv_fdir_flow);
5331         rte_free(fdir_flow);
5332         return -rte_errno;
5333 }
5334
5335 /**
5336  * Delete specific filter.
5337  *
5338  * @param dev
5339  *   Pointer to Ethernet device.
5340  * @param fdir_filter
5341  *   Filter to be deleted.
5342  *
5343  * @return
5344  *   0 on success, a negative errno value otherwise and rte_errno is set.
5345  */
5346 static int
5347 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5348                         const struct rte_eth_fdir_filter *fdir_filter)
5349 {
5350         struct mlx5_priv *priv = dev->data->dev_private;
5351         uint32_t flow_idx;
5352         struct mlx5_fdir fdir_flow = {
5353                 .attr.group = 0,
5354         };
5355         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5356         int ret;
5357
5358         ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5359         if (ret)
5360                 return -rte_errno;
5361         LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5362                 /* Find the fdir in priv list */
5363                 if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5364                         break;
5365         }
5366         if (!priv_fdir_flow)
5367                 return 0;
5368         LIST_REMOVE(priv_fdir_flow, next);
5369         flow_idx = priv_fdir_flow->rix_flow;
5370         flow_list_destroy(dev, &priv->flows, flow_idx);
5371         rte_free(priv_fdir_flow->fdir);
5372         rte_free(priv_fdir_flow);
5373         DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5374                 dev->data->port_id, flow_idx);
5375         return 0;
5376 }
5377
5378 /**
5379  * Update queue for specific filter.
5380  *
5381  * @param dev
5382  *   Pointer to Ethernet device.
5383  * @param fdir_filter
5384  *   Filter to be updated.
5385  *
5386  * @return
5387  *   0 on success, a negative errno value otherwise and rte_errno is set.
5388  */
5389 static int
5390 flow_fdir_filter_update(struct rte_eth_dev *dev,
5391                         const struct rte_eth_fdir_filter *fdir_filter)
5392 {
5393         int ret;
5394
5395         ret = flow_fdir_filter_delete(dev, fdir_filter);
5396         if (ret)
5397                 return ret;
5398         return flow_fdir_filter_add(dev, fdir_filter);
5399 }
5400
5401 /**
5402  * Flush all filters.
5403  *
5404  * @param dev
5405  *   Pointer to Ethernet device.
5406  */
5407 static void
5408 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5409 {
5410         struct mlx5_priv *priv = dev->data->dev_private;
5411         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5412
5413         while (!LIST_EMPTY(&priv->fdir_flows)) {
5414                 priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5415                 LIST_REMOVE(priv_fdir_flow, next);
5416                 flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5417                 rte_free(priv_fdir_flow->fdir);
5418                 rte_free(priv_fdir_flow);
5419         }
5420 }
5421
5422 /**
5423  * Get flow director information.
5424  *
5425  * @param dev
5426  *   Pointer to Ethernet device.
5427  * @param[out] fdir_info
5428  *   Resulting flow director information.
5429  */
5430 static void
5431 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5432 {
5433         struct rte_eth_fdir_masks *mask =
5434                 &dev->data->dev_conf.fdir_conf.mask;
5435
5436         fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5437         fdir_info->guarant_spc = 0;
5438         rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5439         fdir_info->max_flexpayload = 0;
5440         fdir_info->flow_types_mask[0] = 0;
5441         fdir_info->flex_payload_unit = 0;
5442         fdir_info->max_flex_payload_segment_num = 0;
5443         fdir_info->flex_payload_limit = 0;
5444         memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5445 }
5446
5447 /**
5448  * Deal with flow director operations.
5449  *
5450  * @param dev
5451  *   Pointer to Ethernet device.
5452  * @param filter_op
5453  *   Operation to perform.
5454  * @param arg
5455  *   Pointer to operation-specific structure.
5456  *
5457  * @return
5458  *   0 on success, a negative errno value otherwise and rte_errno is set.
5459  */
5460 static int
5461 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5462                     void *arg)
5463 {
5464         enum rte_fdir_mode fdir_mode =
5465                 dev->data->dev_conf.fdir_conf.mode;
5466
5467         if (filter_op == RTE_ETH_FILTER_NOP)
5468                 return 0;
5469         if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5470             fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5471                 DRV_LOG(ERR, "port %u flow director mode %d not supported",
5472                         dev->data->port_id, fdir_mode);
5473                 rte_errno = EINVAL;
5474                 return -rte_errno;
5475         }
5476         switch (filter_op) {
5477         case RTE_ETH_FILTER_ADD:
5478                 return flow_fdir_filter_add(dev, arg);
5479         case RTE_ETH_FILTER_UPDATE:
5480                 return flow_fdir_filter_update(dev, arg);
5481         case RTE_ETH_FILTER_DELETE:
5482                 return flow_fdir_filter_delete(dev, arg);
5483         case RTE_ETH_FILTER_FLUSH:
5484                 flow_fdir_filter_flush(dev);
5485                 break;
5486         case RTE_ETH_FILTER_INFO:
5487                 flow_fdir_info_get(dev, arg);
5488                 break;
5489         default:
5490                 DRV_LOG(DEBUG, "port %u unknown operation %u",
5491                         dev->data->port_id, filter_op);
5492                 rte_errno = EINVAL;
5493                 return -rte_errno;
5494         }
5495         return 0;
5496 }
5497
5498 /**
5499  * Manage filter operations.
5500  *
5501  * @param dev
5502  *   Pointer to Ethernet device structure.
5503  * @param filter_type
5504  *   Filter type.
5505  * @param filter_op
5506  *   Operation to perform.
5507  * @param arg
5508  *   Pointer to operation-specific structure.
5509  *
5510  * @return
5511  *   0 on success, a negative errno value otherwise and rte_errno is set.
5512  */
5513 int
5514 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5515                      enum rte_filter_type filter_type,
5516                      enum rte_filter_op filter_op,
5517                      void *arg)
5518 {
5519         switch (filter_type) {
5520         case RTE_ETH_FILTER_GENERIC:
5521                 if (filter_op != RTE_ETH_FILTER_GET) {
5522                         rte_errno = EINVAL;
5523                         return -rte_errno;
5524                 }
5525                 *(const void **)arg = &mlx5_flow_ops;
5526                 return 0;
5527         case RTE_ETH_FILTER_FDIR:
5528                 return flow_fdir_ctrl_func(dev, filter_op, arg);
5529         default:
5530                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
5531                         dev->data->port_id, filter_type);
5532                 rte_errno = ENOTSUP;
5533                 return -rte_errno;
5534         }
5535         return 0;
5536 }
5537
5538 /**
5539  * Create the needed meter and suffix tables.
5540  *
5541  * @param[in] dev
5542  *   Pointer to Ethernet device.
5543  * @param[in] fm
5544  *   Pointer to the flow meter.
5545  *
5546  * @return
5547  *   Pointer to table set on success, NULL otherwise.
5548  */
5549 struct mlx5_meter_domains_infos *
5550 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5551                           const struct mlx5_flow_meter *fm)
5552 {
5553         const struct mlx5_flow_driver_ops *fops;
5554
5555         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5556         return fops->create_mtr_tbls(dev, fm);
5557 }
5558
5559 /**
5560  * Destroy the meter table set.
5561  *
5562  * @param[in] dev
5563  *   Pointer to Ethernet device.
5564  * @param[in] tbl
5565  *   Pointer to the meter table set.
5566  *
5567  * @return
5568  *   0 on success.
5569  */
5570 int
5571 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5572                            struct mlx5_meter_domains_infos *tbls)
5573 {
5574         const struct mlx5_flow_driver_ops *fops;
5575
5576         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5577         return fops->destroy_mtr_tbls(dev, tbls);
5578 }
5579
5580 /**
5581  * Create policer rules.
5582  *
5583  * @param[in] dev
5584  *   Pointer to Ethernet device.
5585  * @param[in] fm
5586  *   Pointer to flow meter structure.
5587  * @param[in] attr
5588  *   Pointer to flow attributes.
5589  *
5590  * @return
5591  *   0 on success, -1 otherwise.
5592  */
5593 int
5594 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5595                                struct mlx5_flow_meter *fm,
5596                                const struct rte_flow_attr *attr)
5597 {
5598         const struct mlx5_flow_driver_ops *fops;
5599
5600         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5601         return fops->create_policer_rules(dev, fm, attr);
5602 }
5603
5604 /**
5605  * Destroy policer rules.
5606  *
5607  * @param[in] fm
5608  *   Pointer to flow meter structure.
5609  * @param[in] attr
5610  *   Pointer to flow attributes.
5611  *
5612  * @return
5613  *   0 on success, -1 otherwise.
5614  */
5615 int
5616 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5617                                 struct mlx5_flow_meter *fm,
5618                                 const struct rte_flow_attr *attr)
5619 {
5620         const struct mlx5_flow_driver_ops *fops;
5621
5622         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5623         return fops->destroy_policer_rules(dev, fm, attr);
5624 }
5625
5626 /**
5627  * Allocate a counter.
5628  *
5629  * @param[in] dev
5630  *   Pointer to Ethernet device structure.
5631  *
5632  * @return
5633  *   Index to allocated counter  on success, 0 otherwise.
5634  */
5635 uint32_t
5636 mlx5_counter_alloc(struct rte_eth_dev *dev)
5637 {
5638         const struct mlx5_flow_driver_ops *fops;
5639         struct rte_flow_attr attr = { .transfer = 0 };
5640
5641         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5642                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5643                 return fops->counter_alloc(dev);
5644         }
5645         DRV_LOG(ERR,
5646                 "port %u counter allocate is not supported.",
5647                  dev->data->port_id);
5648         return 0;
5649 }
5650
5651 /**
5652  * Free a counter.
5653  *
5654  * @param[in] dev
5655  *   Pointer to Ethernet device structure.
5656  * @param[in] cnt
5657  *   Index to counter to be free.
5658  */
5659 void
5660 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5661 {
5662         const struct mlx5_flow_driver_ops *fops;
5663         struct rte_flow_attr attr = { .transfer = 0 };
5664
5665         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5666                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5667                 fops->counter_free(dev, cnt);
5668                 return;
5669         }
5670         DRV_LOG(ERR,
5671                 "port %u counter free is not supported.",
5672                  dev->data->port_id);
5673 }
5674
5675 /**
5676  * Query counter statistics.
5677  *
5678  * @param[in] dev
5679  *   Pointer to Ethernet device structure.
5680  * @param[in] cnt
5681  *   Index to counter to query.
5682  * @param[in] clear
5683  *   Set to clear counter statistics.
5684  * @param[out] pkts
5685  *   The counter hits packets number to save.
5686  * @param[out] bytes
5687  *   The counter hits bytes number to save.
5688  *
5689  * @return
5690  *   0 on success, a negative errno value otherwise.
5691  */
5692 int
5693 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5694                    bool clear, uint64_t *pkts, uint64_t *bytes)
5695 {
5696         const struct mlx5_flow_driver_ops *fops;
5697         struct rte_flow_attr attr = { .transfer = 0 };
5698
5699         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5700                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5701                 return fops->counter_query(dev, cnt, clear, pkts, bytes);
5702         }
5703         DRV_LOG(ERR,
5704                 "port %u counter query is not supported.",
5705                  dev->data->port_id);
5706         return -ENOTSUP;
5707 }
5708
5709 #define MLX5_POOL_QUERY_FREQ_US 1000000
5710
5711 /**
5712  * Set the periodic procedure for triggering asynchronous batch queries for all
5713  * the counter pools.
5714  *
5715  * @param[in] sh
5716  *   Pointer to mlx5_ibv_shared object.
5717  */
5718 void
5719 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5720 {
5721         struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5722         uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5723         uint32_t us;
5724
5725         cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5726         pools_n += rte_atomic16_read(&cont->n_valid);
5727         us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5728         DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5729         if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5730                 sh->cmng.query_thread_on = 0;
5731                 DRV_LOG(ERR, "Cannot reinitialize query alarm");
5732         } else {
5733                 sh->cmng.query_thread_on = 1;
5734         }
5735 }
5736
5737 /**
5738  * The periodic procedure for triggering asynchronous batch queries for all the
5739  * counter pools. This function is probably called by the host thread.
5740  *
5741  * @param[in] arg
5742  *   The parameter for the alarm process.
5743  */
5744 void
5745 mlx5_flow_query_alarm(void *arg)
5746 {
5747         struct mlx5_ibv_shared *sh = arg;
5748         struct mlx5_devx_obj *dcs;
5749         uint16_t offset;
5750         int ret;
5751         uint8_t batch = sh->cmng.batch;
5752         uint16_t pool_index = sh->cmng.pool_index;
5753         struct mlx5_pools_container *cont;
5754         struct mlx5_pools_container *mcont;
5755         struct mlx5_flow_counter_pool *pool;
5756
5757         if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5758                 goto set_alarm;
5759 next_container:
5760         cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5761         mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5762         /* Check if resize was done and need to flip a container. */
5763         if (cont != mcont) {
5764                 if (cont->pools) {
5765                         /* Clean the old container. */
5766                         rte_free(cont->pools);
5767                         memset(cont, 0, sizeof(*cont));
5768                 }
5769                 rte_cio_wmb();
5770                  /* Flip the host container. */
5771                 sh->cmng.mhi[batch] ^= (uint8_t)2;
5772                 cont = mcont;
5773         }
5774         if (!cont->pools) {
5775                 /* 2 empty containers case is unexpected. */
5776                 if (unlikely(batch != sh->cmng.batch))
5777                         goto set_alarm;
5778                 batch ^= 0x1;
5779                 pool_index = 0;
5780                 goto next_container;
5781         }
5782         pool = cont->pools[pool_index];
5783         if (pool->raw_hw)
5784                 /* There is a pool query in progress. */
5785                 goto set_alarm;
5786         pool->raw_hw =
5787                 LIST_FIRST(&sh->cmng.free_stat_raws);
5788         if (!pool->raw_hw)
5789                 /* No free counter statistics raw memory. */
5790                 goto set_alarm;
5791         dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5792                                                               (&pool->a64_dcs);
5793         offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5794         /*
5795          * Identify the counters released between query trigger and query
5796          * handle more effiecntly. The counter released in this gap period
5797          * should wait for a new round of query as the new arrived packets
5798          * will not be taken into account.
5799          */
5800         rte_atomic64_add(&pool->start_query_gen, 1);
5801         ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5802                                                offset, NULL, NULL,
5803                                                pool->raw_hw->mem_mng->dm->id,
5804                                                (void *)(uintptr_t)
5805                                                (pool->raw_hw->data + offset),
5806                                                sh->devx_comp,
5807                                                (uint64_t)(uintptr_t)pool);
5808         if (ret) {
5809                 rte_atomic64_sub(&pool->start_query_gen, 1);
5810                 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5811                         " %d", pool->min_dcs->id);
5812                 pool->raw_hw = NULL;
5813                 goto set_alarm;
5814         }
5815         pool->raw_hw->min_dcs_id = dcs->id;
5816         LIST_REMOVE(pool->raw_hw, next);
5817         sh->cmng.pending_queries++;
5818         pool_index++;
5819         if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5820                 batch ^= 0x1;
5821                 pool_index = 0;
5822         }
5823 set_alarm:
5824         sh->cmng.batch = batch;
5825         sh->cmng.pool_index = pool_index;
5826         mlx5_set_query_alarm(sh);
5827 }
5828
5829 /**
5830  * Handler for the HW respond about ready values from an asynchronous batch
5831  * query. This function is probably called by the host thread.
5832  *
5833  * @param[in] sh
5834  *   The pointer to the shared IB device context.
5835  * @param[in] async_id
5836  *   The Devx async ID.
5837  * @param[in] status
5838  *   The status of the completion.
5839  */
5840 void
5841 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5842                                   uint64_t async_id, int status)
5843 {
5844         struct mlx5_flow_counter_pool *pool =
5845                 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5846         struct mlx5_counter_stats_raw *raw_to_free;
5847
5848         if (unlikely(status)) {
5849                 rte_atomic64_sub(&pool->start_query_gen, 1);
5850                 raw_to_free = pool->raw_hw;
5851         } else {
5852                 raw_to_free = pool->raw;
5853                 rte_spinlock_lock(&pool->sl);
5854                 pool->raw = pool->raw_hw;
5855                 rte_spinlock_unlock(&pool->sl);
5856                 MLX5_ASSERT(rte_atomic64_read(&pool->end_query_gen) + 1 ==
5857                             rte_atomic64_read(&pool->start_query_gen));
5858                 rte_atomic64_set(&pool->end_query_gen,
5859                                  rte_atomic64_read(&pool->start_query_gen));
5860                 /* Be sure the new raw counters data is updated in memory. */
5861                 rte_cio_wmb();
5862         }
5863         LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5864         pool->raw_hw = NULL;
5865         sh->cmng.pending_queries--;
5866 }
5867
5868 /**
5869  * Translate the rte_flow group index to HW table value.
5870  *
5871  * @param[in] attributes
5872  *   Pointer to flow attributes
5873  * @param[in] external
5874  *   Value is part of flow rule created by request external to PMD.
5875  * @param[in] group
5876  *   rte_flow group index value.
5877  * @param[out] fdb_def_rule
5878  *   Whether fdb jump to table 1 is configured.
5879  * @param[out] table
5880  *   HW table value.
5881  * @param[out] error
5882  *   Pointer to error structure.
5883  *
5884  * @return
5885  *   0 on success, a negative errno value otherwise and rte_errno is set.
5886  */
5887 int
5888 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5889                          uint32_t group, bool fdb_def_rule, uint32_t *table,
5890                          struct rte_flow_error *error)
5891 {
5892         if (attributes->transfer && external && fdb_def_rule) {
5893                 if (group == UINT32_MAX)
5894                         return rte_flow_error_set
5895                                                 (error, EINVAL,
5896                                                  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5897                                                  NULL,
5898                                                  "group index not supported");
5899                 *table = group + 1;
5900         } else {
5901                 *table = group;
5902         }
5903         return 0;
5904 }
5905
5906 /**
5907  * Discover availability of metadata reg_c's.
5908  *
5909  * Iteratively use test flows to check availability.
5910  *
5911  * @param[in] dev
5912  *   Pointer to the Ethernet device structure.
5913  *
5914  * @return
5915  *   0 on success, a negative errno value otherwise and rte_errno is set.
5916  */
5917 int
5918 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5919 {
5920         struct mlx5_priv *priv = dev->data->dev_private;
5921         struct mlx5_dev_config *config = &priv->config;
5922         enum modify_reg idx;
5923         int n = 0;
5924
5925         /* reg_c[0] and reg_c[1] are reserved. */
5926         config->flow_mreg_c[n++] = REG_C_0;
5927         config->flow_mreg_c[n++] = REG_C_1;
5928         /* Discover availability of other reg_c's. */
5929         for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5930                 struct rte_flow_attr attr = {
5931                         .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5932                         .priority = MLX5_FLOW_PRIO_RSVD,
5933                         .ingress = 1,
5934                 };
5935                 struct rte_flow_item items[] = {
5936                         [0] = {
5937                                 .type = RTE_FLOW_ITEM_TYPE_END,
5938                         },
5939                 };
5940                 struct rte_flow_action actions[] = {
5941                         [0] = {
5942                                 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5943                                 .conf = &(struct mlx5_flow_action_copy_mreg){
5944                                         .src = REG_C_1,
5945                                         .dst = idx,
5946                                 },
5947                         },
5948                         [1] = {
5949                                 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5950                                 .conf = &(struct rte_flow_action_jump){
5951                                         .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5952                                 },
5953                         },
5954                         [2] = {
5955                                 .type = RTE_FLOW_ACTION_TYPE_END,
5956                         },
5957                 };
5958                 uint32_t flow_idx;
5959                 struct rte_flow *flow;
5960                 struct rte_flow_error error;
5961
5962                 if (!config->dv_flow_en)
5963                         break;
5964                 /* Create internal flow, validation skips copy action. */
5965                 flow_idx = flow_list_create(dev, NULL, &attr, items,
5966                                             actions, false, &error);
5967                 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
5968                                       flow_idx);
5969                 if (!flow)
5970                         continue;
5971                 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5972                         config->flow_mreg_c[n++] = idx;
5973                 flow_list_destroy(dev, NULL, flow_idx);
5974         }
5975         for (; n < MLX5_MREG_C_NUM; ++n)
5976                 config->flow_mreg_c[n] = REG_NONE;
5977         return 0;
5978 }
5979
5980 /**
5981  * Dump flow raw hw data to file
5982  *
5983  * @param[in] dev
5984  *    The pointer to Ethernet device.
5985  * @param[in] file
5986  *   A pointer to a file for output.
5987  * @param[out] error
5988  *   Perform verbose error reporting if not NULL. PMDs initialize this
5989  *   structure in case of error only.
5990  * @return
5991  *   0 on success, a nagative value otherwise.
5992  */
5993 int
5994 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5995                    FILE *file,
5996                    struct rte_flow_error *error __rte_unused)
5997 {
5998         struct mlx5_priv *priv = dev->data->dev_private;
5999         struct mlx5_ibv_shared *sh = priv->sh;
6000
6001         return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6002                                        sh->tx_domain, file);
6003 }