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 #include <rte_common.h>
  14 #include <rte_ether.h>
  15 #include <rte_ethdev_driver.h>
  16 #include <rte_flow.h>
  17 #include <rte_cycles.h>
  18 #include <rte_flow_driver.h>
  19 #include <rte_malloc.h>
  20 #include <rte_ip.h>
  21
  22 #include <mlx5_glue.h>
  23 #include <mlx5_devx_cmds.h>
  24 #include <mlx5_prm.h>
  25 #include <mlx5_malloc.h>
  26
  27 #include "mlx5_defs.h"
  28 #include "mlx5.h"
  29 #include "mlx5_flow.h"
  30 #include "mlx5_flow_os.h"
  31 #include "mlx5_rxtx.h"
  32
  33 /** Device flow drivers. */
  34 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
  35
  36 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
  37
  38 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
  39         [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
  40 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
  41         [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
  42 #endif
  43         [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
  44         [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
  45 };
  46
  47 enum mlx5_expansion {
  48         MLX5_EXPANSION_ROOT,
  49         MLX5_EXPANSION_ROOT_OUTER,
  50         MLX5_EXPANSION_ROOT_ETH_VLAN,
  51         MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
  52         MLX5_EXPANSION_OUTER_ETH,
  53         MLX5_EXPANSION_OUTER_ETH_VLAN,
  54         MLX5_EXPANSION_OUTER_VLAN,
  55         MLX5_EXPANSION_OUTER_IPV4,
  56         MLX5_EXPANSION_OUTER_IPV4_UDP,
  57         MLX5_EXPANSION_OUTER_IPV4_TCP,
  58         MLX5_EXPANSION_OUTER_IPV6,
  59         MLX5_EXPANSION_OUTER_IPV6_UDP,
  60         MLX5_EXPANSION_OUTER_IPV6_TCP,
  61         MLX5_EXPANSION_VXLAN,
  62         MLX5_EXPANSION_VXLAN_GPE,
  63         MLX5_EXPANSION_GRE,
  64         MLX5_EXPANSION_MPLS,
  65         MLX5_EXPANSION_ETH,
  66         MLX5_EXPANSION_ETH_VLAN,
  67         MLX5_EXPANSION_VLAN,
  68         MLX5_EXPANSION_IPV4,
  69         MLX5_EXPANSION_IPV4_UDP,
  70         MLX5_EXPANSION_IPV4_TCP,
  71         MLX5_EXPANSION_IPV6,
  72         MLX5_EXPANSION_IPV6_UDP,
  73         MLX5_EXPANSION_IPV6_TCP,
  74 };
  75
  76 /** Supported expansion of items. */
  77 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
  78         [MLX5_EXPANSION_ROOT] = {
  79                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
  80                                                  MLX5_EXPANSION_IPV4,
  81                                                  MLX5_EXPANSION_IPV6),
  82                 .type = RTE_FLOW_ITEM_TYPE_END,
  83         },
  84         [MLX5_EXPANSION_ROOT_OUTER] = {
  85                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
  86                                                  MLX5_EXPANSION_OUTER_IPV4,
  87                                                  MLX5_EXPANSION_OUTER_IPV6),
  88                 .type = RTE_FLOW_ITEM_TYPE_END,
  89         },
  90         [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
  91                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
  92                 .type = RTE_FLOW_ITEM_TYPE_END,
  93         },
  94         [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
  95                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
  96                 .type = RTE_FLOW_ITEM_TYPE_END,
  97         },
  98         [MLX5_EXPANSION_OUTER_ETH] = {
  99                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
 100                                                  MLX5_EXPANSION_OUTER_IPV6,
 101                                                  MLX5_EXPANSION_MPLS),
 102                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 103                 .rss_types = 0,
 104         },
 105         [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
 106                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
 107                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 108                 .rss_types = 0,
 109         },
 110         [MLX5_EXPANSION_OUTER_VLAN] = {
 111                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
 112                                                  MLX5_EXPANSION_OUTER_IPV6),
 113                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
 114         },
 115         [MLX5_EXPANSION_OUTER_IPV4] = {
 116                 .next = RTE_FLOW_EXPAND_RSS_NEXT
 117                         (MLX5_EXPANSION_OUTER_IPV4_UDP,
 118                          MLX5_EXPANSION_OUTER_IPV4_TCP,
 119                          MLX5_EXPANSION_GRE,
 120                          MLX5_EXPANSION_IPV4,
 121                          MLX5_EXPANSION_IPV6),
 122                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
 123                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
 124                         ETH_RSS_NONFRAG_IPV4_OTHER,
 125         },
 126         [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
 127                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
 128                                                  MLX5_EXPANSION_VXLAN_GPE),
 129                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 130                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
 131         },
 132         [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
 133                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 134                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
 135         },
 136         [MLX5_EXPANSION_OUTER_IPV6] = {
 137                 .next = RTE_FLOW_EXPAND_RSS_NEXT
 138                         (MLX5_EXPANSION_OUTER_IPV6_UDP,
 139                          MLX5_EXPANSION_OUTER_IPV6_TCP,
 140                          MLX5_EXPANSION_IPV4,
 141                          MLX5_EXPANSION_IPV6),
 142                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
 143                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
 144                         ETH_RSS_NONFRAG_IPV6_OTHER,
 145         },
 146         [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
 147                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
 148                                                  MLX5_EXPANSION_VXLAN_GPE),
 149                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 150                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
 151         },
 152         [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
 153                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 154                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
 155         },
 156         [MLX5_EXPANSION_VXLAN] = {
 157                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
 158                                                  MLX5_EXPANSION_IPV4,
 159                                                  MLX5_EXPANSION_IPV6),
 160                 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
 161         },
 162         [MLX5_EXPANSION_VXLAN_GPE] = {
 163                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
 164                                                  MLX5_EXPANSION_IPV4,
 165                                                  MLX5_EXPANSION_IPV6),
 166                 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
 167         },
 168         [MLX5_EXPANSION_GRE] = {
 169                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
 170                 .type = RTE_FLOW_ITEM_TYPE_GRE,
 171         },
 172         [MLX5_EXPANSION_MPLS] = {
 173                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 174                                                  MLX5_EXPANSION_IPV6),
 175                 .type = RTE_FLOW_ITEM_TYPE_MPLS,
 176         },
 177         [MLX5_EXPANSION_ETH] = {
 178                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 179                                                  MLX5_EXPANSION_IPV6),
 180                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 181         },
 182         [MLX5_EXPANSION_ETH_VLAN] = {
 183                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
 184                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 185         },
 186         [MLX5_EXPANSION_VLAN] = {
 187                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 188                                                  MLX5_EXPANSION_IPV6),
 189                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
 190         },
 191         [MLX5_EXPANSION_IPV4] = {
 192                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
 193                                                  MLX5_EXPANSION_IPV4_TCP),
 194                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
 195                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
 196                         ETH_RSS_NONFRAG_IPV4_OTHER,
 197         },
 198         [MLX5_EXPANSION_IPV4_UDP] = {
 199                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 200                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
 201         },
 202         [MLX5_EXPANSION_IPV4_TCP] = {
 203                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 204                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
 205         },
 206         [MLX5_EXPANSION_IPV6] = {
 207                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
 208                                                  MLX5_EXPANSION_IPV6_TCP),
 209                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
 210                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
 211                         ETH_RSS_NONFRAG_IPV6_OTHER,
 212         },
 213         [MLX5_EXPANSION_IPV6_UDP] = {
 214                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 215                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
 216         },
 217         [MLX5_EXPANSION_IPV6_TCP] = {
 218                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 219                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
 220         },
 221 };
 222
 223 static const struct rte_flow_ops mlx5_flow_ops = {
 224         .validate = mlx5_flow_validate,
 225         .create = mlx5_flow_create,
 226         .destroy = mlx5_flow_destroy,
 227         .flush = mlx5_flow_flush,
 228         .isolate = mlx5_flow_isolate,
 229         .query = mlx5_flow_query,
 230         .dev_dump = mlx5_flow_dev_dump,
 231         .get_aged_flows = mlx5_flow_get_aged_flows,
 232 };
 233
 234 /* Convert FDIR request to Generic flow. */
 235 struct mlx5_fdir {
 236         struct rte_flow_attr attr;
 237         struct rte_flow_item items[4];
 238         struct rte_flow_item_eth l2;
 239         struct rte_flow_item_eth l2_mask;
 240         union {
 241                 struct rte_flow_item_ipv4 ipv4;
 242                 struct rte_flow_item_ipv6 ipv6;
 243         } l3;
 244         union {
 245                 struct rte_flow_item_ipv4 ipv4;
 246                 struct rte_flow_item_ipv6 ipv6;
 247         } l3_mask;
 248         union {
 249                 struct rte_flow_item_udp udp;
 250                 struct rte_flow_item_tcp tcp;
 251         } l4;
 252         union {
 253                 struct rte_flow_item_udp udp;
 254                 struct rte_flow_item_tcp tcp;
 255         } l4_mask;
 256         struct rte_flow_action actions[2];
 257         struct rte_flow_action_queue queue;
 258 };
 259
 260 /* Tunnel information. */
 261 struct mlx5_flow_tunnel_info {
 262         uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
 263         uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
 264 };
 265
 266 static struct mlx5_flow_tunnel_info tunnels_info[] = {
 267         {
 268                 .tunnel = MLX5_FLOW_LAYER_VXLAN,
 269                 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
 270         },
 271         {
 272                 .tunnel = MLX5_FLOW_LAYER_GENEVE,
 273                 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
 274         },
 275         {
 276                 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
 277                 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
 278         },
 279         {
 280                 .tunnel = MLX5_FLOW_LAYER_GRE,
 281                 .ptype = RTE_PTYPE_TUNNEL_GRE,
 282         },
 283         {
 284                 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
 285                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
 286         },
 287         {
 288                 .tunnel = MLX5_FLOW_LAYER_MPLS,
 289                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
 290         },
 291         {
 292                 .tunnel = MLX5_FLOW_LAYER_NVGRE,
 293                 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
 294         },
 295         {
 296                 .tunnel = MLX5_FLOW_LAYER_IPIP,
 297                 .ptype = RTE_PTYPE_TUNNEL_IP,
 298         },
 299         {
 300                 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
 301                 .ptype = RTE_PTYPE_TUNNEL_IP,
 302         },
 303         {
 304                 .tunnel = MLX5_FLOW_LAYER_GTP,
 305                 .ptype = RTE_PTYPE_TUNNEL_GTPU,
 306         },
 307 };
 308
 309 /**
 310  * Translate tag ID to register.
 311  *
 312  * @param[in] dev
 313  *   Pointer to the Ethernet device structure.
 314  * @param[in] feature
 315  *   The feature that request the register.
 316  * @param[in] id
 317  *   The request register ID.
 318  * @param[out] error
 319  *   Error description in case of any.
 320  *
 321  * @return
 322  *   The request register on success, a negative errno
 323  *   value otherwise and rte_errno is set.
 324  */
 325 int
 326 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
 327                      enum mlx5_feature_name feature,
 328                      uint32_t id,
 329                      struct rte_flow_error *error)
 330 {
 331         struct mlx5_priv *priv = dev->data->dev_private;
 332         struct mlx5_dev_config *config = &priv->config;
 333         enum modify_reg start_reg;
 334         bool skip_mtr_reg = false;
 335
 336         switch (feature) {
 337         case MLX5_HAIRPIN_RX:
 338                 return REG_B;
 339         case MLX5_HAIRPIN_TX:
 340                 return REG_A;
 341         case MLX5_METADATA_RX:
 342                 switch (config->dv_xmeta_en) {
 343                 case MLX5_XMETA_MODE_LEGACY:
 344                         return REG_B;
 345                 case MLX5_XMETA_MODE_META16:
 346                         return REG_C_0;
 347                 case MLX5_XMETA_MODE_META32:
 348                         return REG_C_1;
 349                 }
 350                 break;
 351         case MLX5_METADATA_TX:
 352                 return REG_A;
 353         case MLX5_METADATA_FDB:
 354                 switch (config->dv_xmeta_en) {
 355                 case MLX5_XMETA_MODE_LEGACY:
 356                         return REG_NON;
 357                 case MLX5_XMETA_MODE_META16:
 358                         return REG_C_0;
 359                 case MLX5_XMETA_MODE_META32:
 360                         return REG_C_1;
 361                 }
 362                 break;
 363         case MLX5_FLOW_MARK:
 364                 switch (config->dv_xmeta_en) {
 365                 case MLX5_XMETA_MODE_LEGACY:
 366                         return REG_NON;
 367                 case MLX5_XMETA_MODE_META16:
 368                         return REG_C_1;
 369                 case MLX5_XMETA_MODE_META32:
 370                         return REG_C_0;
 371                 }
 372                 break;
 373         case MLX5_MTR_SFX:
 374                 /*
 375                  * If meter color and flow match share one register, flow match
 376                  * should use the meter color register for match.
 377                  */
 378                 if (priv->mtr_reg_share)
 379                         return priv->mtr_color_reg;
 380                 else
 381                         return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
 382                                REG_C_3;
 383         case MLX5_MTR_COLOR:
 384                 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
 385                 return priv->mtr_color_reg;
 386         case MLX5_COPY_MARK:
 387                 /*
 388                  * Metadata COPY_MARK register using is in meter suffix sub
 389                  * flow while with meter. It's safe to share the same register.
 390                  */
 391                 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
 392         case MLX5_APP_TAG:
 393                 /*
 394                  * If meter is enable, it will engage the register for color
 395                  * match and flow match. If meter color match is not using the
 396                  * REG_C_2, need to skip the REG_C_x be used by meter color
 397                  * match.
 398                  * If meter is disable, free to use all available registers.
 399                  */
 400                 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
 401                             (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
 402                 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
 403                 if (id > (REG_C_7 - start_reg))
 404                         return rte_flow_error_set(error, EINVAL,
 405                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 406                                                   NULL, "invalid tag id");
 407                 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
 408                         return rte_flow_error_set(error, ENOTSUP,
 409                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 410                                                   NULL, "unsupported tag id");
 411                 /*
 412                  * This case means meter is using the REG_C_x great than 2.
 413                  * Take care not to conflict with meter color REG_C_x.
 414                  * If the available index REG_C_y >= REG_C_x, skip the
 415                  * color register.
 416                  */
 417                 if (skip_mtr_reg && config->flow_mreg_c
 418                     [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
 419                         if (id >= (REG_C_7 - start_reg))
 420                                 return rte_flow_error_set(error, EINVAL,
 421                                                        RTE_FLOW_ERROR_TYPE_ITEM,
 422                                                         NULL, "invalid tag id");
 423                         if (config->flow_mreg_c
 424                             [id + 1 + start_reg - REG_C_0] != REG_NON)
 425                                 return config->flow_mreg_c
 426                                                [id + 1 + start_reg - REG_C_0];
 427                         return rte_flow_error_set(error, ENOTSUP,
 428                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 429                                                   NULL, "unsupported tag id");
 430                 }
 431                 return config->flow_mreg_c[id + start_reg - REG_C_0];
 432         }
 433         MLX5_ASSERT(false);
 434         return rte_flow_error_set(error, EINVAL,
 435                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
 436                                   NULL, "invalid feature name");
 437 }
 438
 439 /**
 440  * Check extensive flow metadata register support.
 441  *
 442  * @param dev
 443  *   Pointer to rte_eth_dev structure.
 444  *
 445  * @return
 446  *   True if device supports extensive flow metadata register, otherwise false.
 447  */
 448 bool
 449 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
 450 {
 451         struct mlx5_priv *priv = dev->data->dev_private;
 452         struct mlx5_dev_config *config = &priv->config;
 453
 454         /*
 455          * Having available reg_c can be regarded inclusively as supporting
 456          * extensive flow metadata register, which could mean,
 457          * - metadata register copy action by modify header.
 458          * - 16 modify header actions is supported.
 459          * - reg_c's are preserved across different domain (FDB and NIC) on
 460          *   packet loopback by flow lookup miss.
 461          */
 462         return config->flow_mreg_c[2] != REG_NON;
 463 }
 464
 465 /**
 466  * Verify the @p item specifications (spec, last, mask) are compatible with the
 467  * NIC capabilities.
 468  *
 469  * @param[in] item
 470  *   Item specification.
 471  * @param[in] mask
 472  *   @p item->mask or flow default bit-masks.
 473  * @param[in] nic_mask
 474  *   Bit-masks covering supported fields by the NIC to compare with user mask.
 475  * @param[in] size
 476  *   Bit-masks size in bytes.
 477  * @param[out] error
 478  *   Pointer to error structure.
 479  *
 480  * @return
 481  *   0 on success, a negative errno value otherwise and rte_errno is set.
 482  */
 483 int
 484 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
 485                           const uint8_t *mask,
 486                           const uint8_t *nic_mask,
 487                           unsigned int size,
 488                           struct rte_flow_error *error)
 489 {
 490         unsigned int i;
 491
 492         MLX5_ASSERT(nic_mask);
 493         for (i = 0; i < size; ++i)
 494                 if ((nic_mask[i] | mask[i]) != nic_mask[i])
 495                         return rte_flow_error_set(error, ENOTSUP,
 496                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 497                                                   item,
 498                                                   "mask enables non supported"
 499                                                   " bits");
 500         if (!item->spec && (item->mask || item->last))
 501                 return rte_flow_error_set(error, EINVAL,
 502                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
 503                                           "mask/last without a spec is not"
 504                                           " supported");
 505         if (item->spec && item->last) {
 506                 uint8_t spec[size];
 507                 uint8_t last[size];
 508                 unsigned int i;
 509                 int ret;
 510
 511                 for (i = 0; i < size; ++i) {
 512                         spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
 513                         last[i] = ((const uint8_t *)item->last)[i] & mask[i];
 514                 }
 515                 ret = memcmp(spec, last, size);
 516                 if (ret != 0)
 517                         return rte_flow_error_set(error, EINVAL,
 518                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 519                                                   item,
 520                                                   "range is not valid");
 521         }
 522         return 0;
 523 }
 524
 525 /**
 526  * Adjust the hash fields according to the @p flow information.
 527  *
 528  * @param[in] dev_flow.
 529  *   Pointer to the mlx5_flow.
 530  * @param[in] tunnel
 531  *   1 when the hash field is for a tunnel item.
 532  * @param[in] layer_types
 533  *   ETH_RSS_* types.
 534  * @param[in] hash_fields
 535  *   Item hash fields.
 536  *
 537  * @return
 538  *   The hash fields that should be used.
 539  */
 540 uint64_t
 541 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
 542                             int tunnel __rte_unused, uint64_t layer_types,
 543                             uint64_t hash_fields)
 544 {
 545 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
 546         int rss_request_inner = rss_desc->level >= 2;
 547
 548         /* Check RSS hash level for tunnel. */
 549         if (tunnel && rss_request_inner)
 550                 hash_fields |= IBV_RX_HASH_INNER;
 551         else if (tunnel || rss_request_inner)
 552                 return 0;
 553 #endif
 554         /* Check if requested layer matches RSS hash fields. */
 555         if (!(rss_desc->types & layer_types))
 556                 return 0;
 557         return hash_fields;
 558 }
 559
 560 /**
 561  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
 562  * if several tunnel rules are used on this queue, the tunnel ptype will be
 563  * cleared.
 564  *
 565  * @param rxq_ctrl
 566  *   Rx queue to update.
 567  */
 568 static void
 569 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
 570 {
 571         unsigned int i;
 572         uint32_t tunnel_ptype = 0;
 573
 574         /* Look up for the ptype to use. */
 575         for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
 576                 if (!rxq_ctrl->flow_tunnels_n[i])
 577                         continue;
 578                 if (!tunnel_ptype) {
 579                         tunnel_ptype = tunnels_info[i].ptype;
 580                 } else {
 581                         tunnel_ptype = 0;
 582                         break;
 583                 }
 584         }
 585         rxq_ctrl->rxq.tunnel = tunnel_ptype;
 586 }
 587
 588 /**
 589  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
 590  * flow.
 591  *
 592  * @param[in] dev
 593  *   Pointer to the Ethernet device structure.
 594  * @param[in] dev_handle
 595  *   Pointer to device flow handle structure.
 596  */
 597 static void
 598 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
 599                        struct mlx5_flow_handle *dev_handle)
 600 {
 601         struct mlx5_priv *priv = dev->data->dev_private;
 602         const int mark = dev_handle->mark;
 603         const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
 604         struct mlx5_hrxq *hrxq;
 605         unsigned int i;
 606
 607         if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
 608                 return;
 609         hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
 610                               dev_handle->rix_hrxq);
 611         if (!hrxq)
 612                 return;
 613         for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
 614                 int idx = hrxq->ind_table->queues[i];
 615                 struct mlx5_rxq_ctrl *rxq_ctrl =
 616                         container_of((*priv->rxqs)[idx],
 617                                      struct mlx5_rxq_ctrl, rxq);
 618
 619                 /*
 620                  * To support metadata register copy on Tx loopback,
 621                  * this must be always enabled (metadata may arive
 622                  * from other port - not from local flows only.
 623                  */
 624                 if (priv->config.dv_flow_en &&
 625                     priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
 626                     mlx5_flow_ext_mreg_supported(dev)) {
 627                         rxq_ctrl->rxq.mark = 1;
 628                         rxq_ctrl->flow_mark_n = 1;
 629                 } else if (mark) {
 630                         rxq_ctrl->rxq.mark = 1;
 631                         rxq_ctrl->flow_mark_n++;
 632                 }
 633                 if (tunnel) {
 634                         unsigned int j;
 635
 636                         /* Increase the counter matching the flow. */
 637                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 638                                 if ((tunnels_info[j].tunnel &
 639                                      dev_handle->layers) ==
 640                                     tunnels_info[j].tunnel) {
 641                                         rxq_ctrl->flow_tunnels_n[j]++;
 642                                         break;
 643                                 }
 644                         }
 645                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
 646                 }
 647         }
 648 }
 649
 650 /**
 651  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
 652  *
 653  * @param[in] dev
 654  *   Pointer to the Ethernet device structure.
 655  * @param[in] flow
 656  *   Pointer to flow structure.
 657  */
 658 static void
 659 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
 660 {
 661         struct mlx5_priv *priv = dev->data->dev_private;
 662         uint32_t handle_idx;
 663         struct mlx5_flow_handle *dev_handle;
 664
 665         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
 666                        handle_idx, dev_handle, next)
 667                 flow_drv_rxq_flags_set(dev, dev_handle);
 668 }
 669
 670 /**
 671  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
 672  * device flow if no other flow uses it with the same kind of request.
 673  *
 674  * @param dev
 675  *   Pointer to Ethernet device.
 676  * @param[in] dev_handle
 677  *   Pointer to the device flow handle structure.
 678  */
 679 static void
 680 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
 681                         struct mlx5_flow_handle *dev_handle)
 682 {
 683         struct mlx5_priv *priv = dev->data->dev_private;
 684         const int mark = dev_handle->mark;
 685         const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
 686         struct mlx5_hrxq *hrxq;
 687         unsigned int i;
 688
 689         if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
 690                 return;
 691         hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
 692                               dev_handle->rix_hrxq);
 693         if (!hrxq)
 694                 return;
 695         MLX5_ASSERT(dev->data->dev_started);
 696         for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
 697                 int idx = hrxq->ind_table->queues[i];
 698                 struct mlx5_rxq_ctrl *rxq_ctrl =
 699                         container_of((*priv->rxqs)[idx],
 700                                      struct mlx5_rxq_ctrl, rxq);
 701
 702                 if (priv->config.dv_flow_en &&
 703                     priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
 704                     mlx5_flow_ext_mreg_supported(dev)) {
 705                         rxq_ctrl->rxq.mark = 1;
 706                         rxq_ctrl->flow_mark_n = 1;
 707                 } else if (mark) {
 708                         rxq_ctrl->flow_mark_n--;
 709                         rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
 710                 }
 711                 if (tunnel) {
 712                         unsigned int j;
 713
 714                         /* Decrease the counter matching the flow. */
 715                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
 716                                 if ((tunnels_info[j].tunnel &
 717                                      dev_handle->layers) ==
 718                                     tunnels_info[j].tunnel) {
 719                                         rxq_ctrl->flow_tunnels_n[j]--;
 720                                         break;
 721                                 }
 722                         }
 723                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
 724                 }
 725         }
 726 }
 727
 728 /**
 729  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
 730  * @p flow if no other flow uses it with the same kind of request.
 731  *
 732  * @param dev
 733  *   Pointer to Ethernet device.
 734  * @param[in] flow
 735  *   Pointer to the flow.
 736  */
 737 static void
 738 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
 739 {
 740         struct mlx5_priv *priv = dev->data->dev_private;
 741         uint32_t handle_idx;
 742         struct mlx5_flow_handle *dev_handle;
 743
 744         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
 745                        handle_idx, dev_handle, next)
 746                 flow_drv_rxq_flags_trim(dev, dev_handle);
 747 }
 748
 749 /**
 750  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
 751  *
 752  * @param dev
 753  *   Pointer to Ethernet device.
 754  */
 755 static void
 756 flow_rxq_flags_clear(struct rte_eth_dev *dev)
 757 {
 758         struct mlx5_priv *priv = dev->data->dev_private;
 759         unsigned int i;
 760
 761         for (i = 0; i != priv->rxqs_n; ++i) {
 762                 struct mlx5_rxq_ctrl *rxq_ctrl;
 763                 unsigned int j;
 764
 765                 if (!(*priv->rxqs)[i])
 766                         continue;
 767                 rxq_ctrl = container_of((*priv->rxqs)[i],
 768                                         struct mlx5_rxq_ctrl, rxq);
 769                 rxq_ctrl->flow_mark_n = 0;
 770                 rxq_ctrl->rxq.mark = 0;
 771                 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
 772                         rxq_ctrl->flow_tunnels_n[j] = 0;
 773                 rxq_ctrl->rxq.tunnel = 0;
 774         }
 775 }
 776
 777 /**
 778  * Set the Rx queue dynamic metadata (mask and offset) for a flow
 779  *
 780  * @param[in] dev
 781  *   Pointer to the Ethernet device structure.
 782  */
 783 void
 784 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
 785 {
 786         struct mlx5_priv *priv = dev->data->dev_private;
 787         struct mlx5_rxq_data *data;
 788         unsigned int i;
 789
 790         for (i = 0; i != priv->rxqs_n; ++i) {
 791                 if (!(*priv->rxqs)[i])
 792                         continue;
 793                 data = (*priv->rxqs)[i];
 794                 if (!rte_flow_dynf_metadata_avail()) {
 795                         data->dynf_meta = 0;
 796                         data->flow_meta_mask = 0;
 797                         data->flow_meta_offset = -1;
 798                 } else {
 799                         data->dynf_meta = 1;
 800                         data->flow_meta_mask = rte_flow_dynf_metadata_mask;
 801                         data->flow_meta_offset = rte_flow_dynf_metadata_offs;
 802                 }
 803         }
 804 }
 805
 806 /*
 807  * return a pointer to the desired action in the list of actions.
 808  *
 809  * @param[in] actions
 810  *   The list of actions to search the action in.
 811  * @param[in] action
 812  *   The action to find.
 813  *
 814  * @return
 815  *   Pointer to the action in the list, if found. NULL otherwise.
 816  */
 817 const struct rte_flow_action *
 818 mlx5_flow_find_action(const struct rte_flow_action *actions,
 819                       enum rte_flow_action_type action)
 820 {
 821         if (actions == NULL)
 822                 return NULL;
 823         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
 824                 if (actions->type == action)
 825                         return actions;
 826         return NULL;
 827 }
 828
 829 /*
 830  * Validate the flag action.
 831  *
 832  * @param[in] action_flags
 833  *   Bit-fields that holds the actions detected until now.
 834  * @param[in] attr
 835  *   Attributes of flow that includes this action.
 836  * @param[out] error
 837  *   Pointer to error structure.
 838  *
 839  * @return
 840  *   0 on success, a negative errno value otherwise and rte_errno is set.
 841  */
 842 int
 843 mlx5_flow_validate_action_flag(uint64_t action_flags,
 844                                const struct rte_flow_attr *attr,
 845                                struct rte_flow_error *error)
 846 {
 847         if (action_flags & MLX5_FLOW_ACTION_MARK)
 848                 return rte_flow_error_set(error, EINVAL,
 849                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 850                                           "can't mark and flag in same flow");
 851         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 852                 return rte_flow_error_set(error, EINVAL,
 853                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 854                                           "can't have 2 flag"
 855                                           " actions in same flow");
 856         if (attr->egress)
 857                 return rte_flow_error_set(error, ENOTSUP,
 858                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 859                                           "flag action not supported for "
 860                                           "egress");
 861         return 0;
 862 }
 863
 864 /*
 865  * Validate the mark action.
 866  *
 867  * @param[in] action
 868  *   Pointer to the queue action.
 869  * @param[in] action_flags
 870  *   Bit-fields that holds the actions detected until now.
 871  * @param[in] attr
 872  *   Attributes of flow that includes this action.
 873  * @param[out] error
 874  *   Pointer to error structure.
 875  *
 876  * @return
 877  *   0 on success, a negative errno value otherwise and rte_errno is set.
 878  */
 879 int
 880 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
 881                                uint64_t action_flags,
 882                                const struct rte_flow_attr *attr,
 883                                struct rte_flow_error *error)
 884 {
 885         const struct rte_flow_action_mark *mark = action->conf;
 886
 887         if (!mark)
 888                 return rte_flow_error_set(error, EINVAL,
 889                                           RTE_FLOW_ERROR_TYPE_ACTION,
 890                                           action,
 891                                           "configuration cannot be null");
 892         if (mark->id >= MLX5_FLOW_MARK_MAX)
 893                 return rte_flow_error_set(error, EINVAL,
 894                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 895                                           &mark->id,
 896                                           "mark id must in 0 <= id < "
 897                                           RTE_STR(MLX5_FLOW_MARK_MAX));
 898         if (action_flags & MLX5_FLOW_ACTION_FLAG)
 899                 return rte_flow_error_set(error, EINVAL,
 900                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 901                                           "can't flag and mark in same flow");
 902         if (action_flags & MLX5_FLOW_ACTION_MARK)
 903                 return rte_flow_error_set(error, EINVAL,
 904                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 905                                           "can't have 2 mark actions in same"
 906                                           " flow");
 907         if (attr->egress)
 908                 return rte_flow_error_set(error, ENOTSUP,
 909                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 910                                           "mark action not supported for "
 911                                           "egress");
 912         return 0;
 913 }
 914
 915 /*
 916  * Validate the drop action.
 917  *
 918  * @param[in] action_flags
 919  *   Bit-fields that holds the actions detected until now.
 920  * @param[in] attr
 921  *   Attributes of flow that includes this action.
 922  * @param[out] error
 923  *   Pointer to error structure.
 924  *
 925  * @return
 926  *   0 on success, a negative errno value otherwise and rte_errno is set.
 927  */
 928 int
 929 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
 930                                const struct rte_flow_attr *attr,
 931                                struct rte_flow_error *error)
 932 {
 933         if (attr->egress)
 934                 return rte_flow_error_set(error, ENOTSUP,
 935                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 936                                           "drop action not supported for "
 937                                           "egress");
 938         return 0;
 939 }
 940
 941 /*
 942  * Validate the queue action.
 943  *
 944  * @param[in] action
 945  *   Pointer to the queue action.
 946  * @param[in] action_flags
 947  *   Bit-fields that holds the actions detected until now.
 948  * @param[in] dev
 949  *   Pointer to the Ethernet device structure.
 950  * @param[in] attr
 951  *   Attributes of flow that includes this action.
 952  * @param[out] error
 953  *   Pointer to error structure.
 954  *
 955  * @return
 956  *   0 on success, a negative errno value otherwise and rte_errno is set.
 957  */
 958 int
 959 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 960                                 uint64_t action_flags,
 961                                 struct rte_eth_dev *dev,
 962                                 const struct rte_flow_attr *attr,
 963                                 struct rte_flow_error *error)
 964 {
 965         struct mlx5_priv *priv = dev->data->dev_private;
 966         const struct rte_flow_action_queue *queue = action->conf;
 967
 968         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
 969                 return rte_flow_error_set(error, EINVAL,
 970                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 971                                           "can't have 2 fate actions in"
 972                                           " same flow");
 973         if (!priv->rxqs_n)
 974                 return rte_flow_error_set(error, EINVAL,
 975                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 976                                           NULL, "No Rx queues configured");
 977         if (queue->index >= priv->rxqs_n)
 978                 return rte_flow_error_set(error, EINVAL,
 979                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 980                                           &queue->index,
 981                                           "queue index out of range");
 982         if (!(*priv->rxqs)[queue->index])
 983                 return rte_flow_error_set(error, EINVAL,
 984                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 985                                           &queue->index,
 986                                           "queue is not configured");
 987         if (attr->egress)
 988                 return rte_flow_error_set(error, ENOTSUP,
 989                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 990                                           "queue action not supported for "
 991                                           "egress");
 992         return 0;
 993 }
 994
 995 /*
 996  * Validate the rss action.
 997  *
 998  * @param[in] action
 999  *   Pointer to the queue action.
1000  * @param[in] action_flags
1001  *   Bit-fields that holds the actions detected until now.
1002  * @param[in] dev
1003  *   Pointer to the Ethernet device structure.
1004  * @param[in] attr
1005  *   Attributes of flow that includes this action.
1006  * @param[in] item_flags
1007  *   Items that were detected.
1008  * @param[out] error
1009  *   Pointer to error structure.
1010  *
1011  * @return
1012  *   0 on success, a negative errno value otherwise and rte_errno is set.
1013  */
1014 int
1015 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1016                               uint64_t action_flags,
1017                               struct rte_eth_dev *dev,
1018                               const struct rte_flow_attr *attr,
1019                               uint64_t item_flags,
1020                               struct rte_flow_error *error)
1021 {
1022         struct mlx5_priv *priv = dev->data->dev_private;
1023         const struct rte_flow_action_rss *rss = action->conf;
1024         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1025         unsigned int i;
1026
1027         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1028                 return rte_flow_error_set(error, EINVAL,
1029                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1030                                           "can't have 2 fate actions"
1031                                           " in same flow");
1032         if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1033             rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1034                 return rte_flow_error_set(error, ENOTSUP,
1035                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1036                                           &rss->func,
1037                                           "RSS hash function not supported");
1038 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1039         if (rss->level > 2)
1040 #else
1041         if (rss->level > 1)
1042 #endif
1043                 return rte_flow_error_set(error, ENOTSUP,
1044                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1045                                           &rss->level,
1046                                           "tunnel RSS is not supported");
1047         /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1048         if (rss->key_len == 0 && rss->key != NULL)
1049                 return rte_flow_error_set(error, ENOTSUP,
1050                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1051                                           &rss->key_len,
1052                                           "RSS hash key length 0");
1053         if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1054                 return rte_flow_error_set(error, ENOTSUP,
1055                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1056                                           &rss->key_len,
1057                                           "RSS hash key too small");
1058         if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1059                 return rte_flow_error_set(error, ENOTSUP,
1060                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1061                                           &rss->key_len,
1062                                           "RSS hash key too large");
1063         if (rss->queue_num > priv->config.ind_table_max_size)
1064                 return rte_flow_error_set(error, ENOTSUP,
1065                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1066                                           &rss->queue_num,
1067                                           "number of queues too large");
1068         if (rss->types & MLX5_RSS_HF_MASK)
1069                 return rte_flow_error_set(error, ENOTSUP,
1070                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1071                                           &rss->types,
1072                                           "some RSS protocols are not"
1073                                           " supported");
1074         if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1075             !(rss->types & ETH_RSS_IP))
1076                 return rte_flow_error_set(error, EINVAL,
1077                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1078                                           "L3 partial RSS requested but L3 RSS"
1079                                           " type not specified");
1080         if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1081             !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1082                 return rte_flow_error_set(error, EINVAL,
1083                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1084                                           "L4 partial RSS requested but L4 RSS"
1085                                           " type not specified");
1086         if (!priv->rxqs_n)
1087                 return rte_flow_error_set(error, EINVAL,
1088                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1089                                           NULL, "No Rx queues configured");
1090         if (!rss->queue_num)
1091                 return rte_flow_error_set(error, EINVAL,
1092                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1093                                           NULL, "No queues configured");
1094         for (i = 0; i != rss->queue_num; ++i) {
1095                 if (rss->queue[i] >= priv->rxqs_n)
1096                         return rte_flow_error_set
1097                                 (error, EINVAL,
1098                                  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1099                                  &rss->queue[i], "queue index out of range");
1100                 if (!(*priv->rxqs)[rss->queue[i]])
1101                         return rte_flow_error_set
1102                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1103                                  &rss->queue[i], "queue is not configured");
1104         }
1105         if (attr->egress)
1106                 return rte_flow_error_set(error, ENOTSUP,
1107                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1108                                           "rss action not supported for "
1109                                           "egress");
1110         if (rss->level > 1 && !tunnel)
1111                 return rte_flow_error_set(error, EINVAL,
1112                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1113                                           "inner RSS is not supported for "
1114                                           "non-tunnel flows");
1115         if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1116             !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1117                 return rte_flow_error_set(error, EINVAL,
1118                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1119                                           "RSS on eCPRI is not supported now");
1120         }
1121         return 0;
1122 }
1123
1124 /*
1125  * Validate the default miss action.
1126  *
1127  * @param[in] action_flags
1128  *   Bit-fields that holds the actions detected until now.
1129  * @param[out] error
1130  *   Pointer to error structure.
1131  *
1132  * @return
1133  *   0 on success, a negative errno value otherwise and rte_errno is set.
1134  */
1135 int
1136 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1137                                 const struct rte_flow_attr *attr,
1138                                 struct rte_flow_error *error)
1139 {
1140         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1141                 return rte_flow_error_set(error, EINVAL,
1142                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1143                                           "can't have 2 fate actions in"
1144                                           " same flow");
1145         if (attr->egress)
1146                 return rte_flow_error_set(error, ENOTSUP,
1147                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1148                                           "default miss action not supported "
1149                                           "for egress");
1150         if (attr->group)
1151                 return rte_flow_error_set(error, ENOTSUP,
1152                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1153                                           "only group 0 is supported");
1154         if (attr->transfer)
1155                 return rte_flow_error_set(error, ENOTSUP,
1156                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1157                                           NULL, "transfer is not supported");
1158         return 0;
1159 }
1160
1161 /*
1162  * Validate the count action.
1163  *
1164  * @param[in] dev
1165  *   Pointer to the Ethernet device structure.
1166  * @param[in] attr
1167  *   Attributes of flow that includes this action.
1168  * @param[out] error
1169  *   Pointer to error structure.
1170  *
1171  * @return
1172  *   0 on success, a negative errno value otherwise and rte_errno is set.
1173  */
1174 int
1175 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1176                                 const struct rte_flow_attr *attr,
1177                                 struct rte_flow_error *error)
1178 {
1179         if (attr->egress)
1180                 return rte_flow_error_set(error, ENOTSUP,
1181                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1182                                           "count action not supported for "
1183                                           "egress");
1184         return 0;
1185 }
1186
1187 /**
1188  * Verify the @p attributes will be correctly understood by the NIC and store
1189  * them in the @p flow if everything is correct.
1190  *
1191  * @param[in] dev
1192  *   Pointer to the Ethernet device structure.
1193  * @param[in] attributes
1194  *   Pointer to flow attributes
1195  * @param[out] error
1196  *   Pointer to error structure.
1197  *
1198  * @return
1199  *   0 on success, a negative errno value otherwise and rte_errno is set.
1200  */
1201 int
1202 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1203                               const struct rte_flow_attr *attributes,
1204                               struct rte_flow_error *error)
1205 {
1206         struct mlx5_priv *priv = dev->data->dev_private;
1207         uint32_t priority_max = priv->config.flow_prio - 1;
1208
1209         if (attributes->group)
1210                 return rte_flow_error_set(error, ENOTSUP,
1211                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1212                                           NULL, "groups is not supported");
1213         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1214             attributes->priority >= priority_max)
1215                 return rte_flow_error_set(error, ENOTSUP,
1216                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1217                                           NULL, "priority out of range");
1218         if (attributes->egress)
1219                 return rte_flow_error_set(error, ENOTSUP,
1220                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1221                                           "egress is not supported");
1222         if (attributes->transfer && !priv->config.dv_esw_en)
1223                 return rte_flow_error_set(error, ENOTSUP,
1224                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1225                                           NULL, "transfer is not supported");
1226         if (!attributes->ingress)
1227                 return rte_flow_error_set(error, EINVAL,
1228                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1229                                           NULL,
1230                                           "ingress attribute is mandatory");
1231         return 0;
1232 }
1233
1234 /**
1235  * Validate ICMP6 item.
1236  *
1237  * @param[in] item
1238  *   Item specification.
1239  * @param[in] item_flags
1240  *   Bit-fields that holds the items detected until now.
1241  * @param[out] error
1242  *   Pointer to error structure.
1243  *
1244  * @return
1245  *   0 on success, a negative errno value otherwise and rte_errno is set.
1246  */
1247 int
1248 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1249                                uint64_t item_flags,
1250                                uint8_t target_protocol,
1251                                struct rte_flow_error *error)
1252 {
1253         const struct rte_flow_item_icmp6 *mask = item->mask;
1254         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1255         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1256                                       MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1257         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1258                                       MLX5_FLOW_LAYER_OUTER_L4;
1259         int ret;
1260
1261         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1262                 return rte_flow_error_set(error, EINVAL,
1263                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1264                                           "protocol filtering not compatible"
1265                                           " with ICMP6 layer");
1266         if (!(item_flags & l3m))
1267                 return rte_flow_error_set(error, EINVAL,
1268                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1269                                           "IPv6 is mandatory to filter on"
1270                                           " ICMP6");
1271         if (item_flags & l4m)
1272                 return rte_flow_error_set(error, EINVAL,
1273                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1274                                           "multiple L4 layers not supported");
1275         if (!mask)
1276                 mask = &rte_flow_item_icmp6_mask;
1277         ret = mlx5_flow_item_acceptable
1278                 (item, (const uint8_t *)mask,
1279                  (const uint8_t *)&rte_flow_item_icmp6_mask,
1280                  sizeof(struct rte_flow_item_icmp6), error);
1281         if (ret < 0)
1282                 return ret;
1283         return 0;
1284 }
1285
1286 /**
1287  * Validate ICMP item.
1288  *
1289  * @param[in] item
1290  *   Item specification.
1291  * @param[in] item_flags
1292  *   Bit-fields that holds the items detected until now.
1293  * @param[out] error
1294  *   Pointer to error structure.
1295  *
1296  * @return
1297  *   0 on success, a negative errno value otherwise and rte_errno is set.
1298  */
1299 int
1300 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1301                              uint64_t item_flags,
1302                              uint8_t target_protocol,
1303                              struct rte_flow_error *error)
1304 {
1305         const struct rte_flow_item_icmp *mask = item->mask;
1306         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1307         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1308                                       MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1309         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1310                                       MLX5_FLOW_LAYER_OUTER_L4;
1311         int ret;
1312
1313         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1314                 return rte_flow_error_set(error, EINVAL,
1315                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1316                                           "protocol filtering not compatible"
1317                                           " with ICMP layer");
1318         if (!(item_flags & l3m))
1319                 return rte_flow_error_set(error, EINVAL,
1320                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1321                                           "IPv4 is mandatory to filter"
1322                                           " on ICMP");
1323         if (item_flags & l4m)
1324                 return rte_flow_error_set(error, EINVAL,
1325                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1326                                           "multiple L4 layers not supported");
1327         if (!mask)
1328                 mask = &rte_flow_item_icmp_mask;
1329         ret = mlx5_flow_item_acceptable
1330                 (item, (const uint8_t *)mask,
1331                  (const uint8_t *)&rte_flow_item_icmp_mask,
1332                  sizeof(struct rte_flow_item_icmp), error);
1333         if (ret < 0)
1334                 return ret;
1335         return 0;
1336 }
1337
1338 /**
1339  * Validate Ethernet item.
1340  *
1341  * @param[in] item
1342  *   Item specification.
1343  * @param[in] item_flags
1344  *   Bit-fields that holds the items detected until now.
1345  * @param[out] error
1346  *   Pointer to error structure.
1347  *
1348  * @return
1349  *   0 on success, a negative errno value otherwise and rte_errno is set.
1350  */
1351 int
1352 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1353                             uint64_t item_flags,
1354                             struct rte_flow_error *error)
1355 {
1356         const struct rte_flow_item_eth *mask = item->mask;
1357         const struct rte_flow_item_eth nic_mask = {
1358                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1359                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1360                 .type = RTE_BE16(0xffff),
1361         };
1362         int ret;
1363         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1364         const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1365                                        MLX5_FLOW_LAYER_OUTER_L2;
1366
1367         if (item_flags & ethm)
1368                 return rte_flow_error_set(error, ENOTSUP,
1369                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1370                                           "multiple L2 layers not supported");
1371         if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1372             (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1373                 return rte_flow_error_set(error, EINVAL,
1374                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1375                                           "L2 layer should not follow "
1376                                           "L3 layers");
1377         if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1378             (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1379                 return rte_flow_error_set(error, EINVAL,
1380                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1381                                           "L2 layer should not follow VLAN");
1382         if (!mask)
1383                 mask = &rte_flow_item_eth_mask;
1384         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1385                                         (const uint8_t *)&nic_mask,
1386                                         sizeof(struct rte_flow_item_eth),
1387                                         error);
1388         return ret;
1389 }
1390
1391 /**
1392  * Validate VLAN item.
1393  *
1394  * @param[in] item
1395  *   Item specification.
1396  * @param[in] item_flags
1397  *   Bit-fields that holds the items detected until now.
1398  * @param[in] dev
1399  *   Ethernet device flow is being created on.
1400  * @param[out] error
1401  *   Pointer to error structure.
1402  *
1403  * @return
1404  *   0 on success, a negative errno value otherwise and rte_errno is set.
1405  */
1406 int
1407 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1408                              uint64_t item_flags,
1409                              struct rte_eth_dev *dev,
1410                              struct rte_flow_error *error)
1411 {
1412         const struct rte_flow_item_vlan *spec = item->spec;
1413         const struct rte_flow_item_vlan *mask = item->mask;
1414         const struct rte_flow_item_vlan nic_mask = {
1415                 .tci = RTE_BE16(UINT16_MAX),
1416                 .inner_type = RTE_BE16(UINT16_MAX),
1417         };
1418         uint16_t vlan_tag = 0;
1419         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1420         int ret;
1421         const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1422                                         MLX5_FLOW_LAYER_INNER_L4) :
1423                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1424                                         MLX5_FLOW_LAYER_OUTER_L4);
1425         const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1426                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1427
1428         if (item_flags & vlanm)
1429                 return rte_flow_error_set(error, EINVAL,
1430                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1431                                           "multiple VLAN layers not supported");
1432         else if ((item_flags & l34m) != 0)
1433                 return rte_flow_error_set(error, EINVAL,
1434                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1435                                           "VLAN cannot follow L3/L4 layer");
1436         if (!mask)
1437                 mask = &rte_flow_item_vlan_mask;
1438         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1439                                         (const uint8_t *)&nic_mask,
1440                                         sizeof(struct rte_flow_item_vlan),
1441                                         error);
1442         if (ret)
1443                 return ret;
1444         if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1445                 struct mlx5_priv *priv = dev->data->dev_private;
1446
1447                 if (priv->vmwa_context) {
1448                         /*
1449                          * Non-NULL context means we have a virtual machine
1450                          * and SR-IOV enabled, we have to create VLAN interface
1451                          * to make hypervisor to setup E-Switch vport
1452                          * context correctly. We avoid creating the multiple
1453                          * VLAN interfaces, so we cannot support VLAN tag mask.
1454                          */
1455                         return rte_flow_error_set(error, EINVAL,
1456                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1457                                                   item,
1458                                                   "VLAN tag mask is not"
1459                                                   " supported in virtual"
1460                                                   " environment");
1461                 }
1462         }
1463         if (spec) {
1464                 vlan_tag = spec->tci;
1465                 vlan_tag &= mask->tci;
1466         }
1467         /*
1468          * From verbs perspective an empty VLAN is equivalent
1469          * to a packet without VLAN layer.
1470          */
1471         if (!vlan_tag)
1472                 return rte_flow_error_set(error, EINVAL,
1473                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1474                                           item->spec,
1475                                           "VLAN cannot be empty");
1476         return 0;
1477 }
1478
1479 /**
1480  * Validate IPV4 item.
1481  *
1482  * @param[in] item
1483  *   Item specification.
1484  * @param[in] item_flags
1485  *   Bit-fields that holds the items detected until now.
1486  * @param[in] last_item
1487  *   Previous validated item in the pattern items.
1488  * @param[in] ether_type
1489  *   Type in the ethernet layer header (including dot1q).
1490  * @param[in] acc_mask
1491  *   Acceptable mask, if NULL default internal default mask
1492  *   will be used to check whether item fields are supported.
1493  * @param[out] error
1494  *   Pointer to error structure.
1495  *
1496  * @return
1497  *   0 on success, a negative errno value otherwise and rte_errno is set.
1498  */
1499 int
1500 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1501                              uint64_t item_flags,
1502                              uint64_t last_item,
1503                              uint16_t ether_type,
1504                              const struct rte_flow_item_ipv4 *acc_mask,
1505                              struct rte_flow_error *error)
1506 {
1507         const struct rte_flow_item_ipv4 *mask = item->mask;
1508         const struct rte_flow_item_ipv4 *spec = item->spec;
1509         const struct rte_flow_item_ipv4 nic_mask = {
1510                 .hdr = {
1511                         .src_addr = RTE_BE32(0xffffffff),
1512                         .dst_addr = RTE_BE32(0xffffffff),
1513                         .type_of_service = 0xff,
1514                         .next_proto_id = 0xff,
1515                 },
1516         };
1517         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1518         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1519                                       MLX5_FLOW_LAYER_OUTER_L3;
1520         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1521                                       MLX5_FLOW_LAYER_OUTER_L4;
1522         int ret;
1523         uint8_t next_proto = 0xFF;
1524         const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1525                                   MLX5_FLOW_LAYER_OUTER_VLAN |
1526                                   MLX5_FLOW_LAYER_INNER_VLAN);
1527
1528         if ((last_item & l2_vlan) && ether_type &&
1529             ether_type != RTE_ETHER_TYPE_IPV4)
1530                 return rte_flow_error_set(error, EINVAL,
1531                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1532                                           "IPv4 cannot follow L2/VLAN layer "
1533                                           "which ether type is not IPv4");
1534         if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1535                 if (mask && spec)
1536                         next_proto = mask->hdr.next_proto_id &
1537                                      spec->hdr.next_proto_id;
1538                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1539                         return rte_flow_error_set(error, EINVAL,
1540                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1541                                                   item,
1542                                                   "multiple tunnel "
1543                                                   "not supported");
1544         }
1545         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1546                 return rte_flow_error_set(error, EINVAL,
1547                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1548                                           "wrong tunnel type - IPv6 specified "
1549                                           "but IPv4 item provided");
1550         if (item_flags & l3m)
1551                 return rte_flow_error_set(error, ENOTSUP,
1552                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1553                                           "multiple L3 layers not supported");
1554         else if (item_flags & l4m)
1555                 return rte_flow_error_set(error, EINVAL,
1556                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1557                                           "L3 cannot follow an L4 layer.");
1558         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1559                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1560                 return rte_flow_error_set(error, EINVAL,
1561                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1562                                           "L3 cannot follow an NVGRE layer.");
1563         if (!mask)
1564                 mask = &rte_flow_item_ipv4_mask;
1565         else if (mask->hdr.next_proto_id != 0 &&
1566                  mask->hdr.next_proto_id != 0xff)
1567                 return rte_flow_error_set(error, EINVAL,
1568                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1569                                           "partial mask is not supported"
1570                                           " for protocol");
1571         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1572                                         acc_mask ? (const uint8_t *)acc_mask
1573                                                  : (const uint8_t *)&nic_mask,
1574                                         sizeof(struct rte_flow_item_ipv4),
1575                                         error);
1576         if (ret < 0)
1577                 return ret;
1578         return 0;
1579 }
1580
1581 /**
1582  * Validate IPV6 item.
1583  *
1584  * @param[in] item
1585  *   Item specification.
1586  * @param[in] item_flags
1587  *   Bit-fields that holds the items detected until now.
1588  * @param[in] last_item
1589  *   Previous validated item in the pattern items.
1590  * @param[in] ether_type
1591  *   Type in the ethernet layer header (including dot1q).
1592  * @param[in] acc_mask
1593  *   Acceptable mask, if NULL default internal default mask
1594  *   will be used to check whether item fields are supported.
1595  * @param[out] error
1596  *   Pointer to error structure.
1597  *
1598  * @return
1599  *   0 on success, a negative errno value otherwise and rte_errno is set.
1600  */
1601 int
1602 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1603                              uint64_t item_flags,
1604                              uint64_t last_item,
1605                              uint16_t ether_type,
1606                              const struct rte_flow_item_ipv6 *acc_mask,
1607                              struct rte_flow_error *error)
1608 {
1609         const struct rte_flow_item_ipv6 *mask = item->mask;
1610         const struct rte_flow_item_ipv6 *spec = item->spec;
1611         const struct rte_flow_item_ipv6 nic_mask = {
1612                 .hdr = {
1613                         .src_addr =
1614                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1615                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1616                         .dst_addr =
1617                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1618                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1619                         .vtc_flow = RTE_BE32(0xffffffff),
1620                         .proto = 0xff,
1621                 },
1622         };
1623         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1624         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1625                                       MLX5_FLOW_LAYER_OUTER_L3;
1626         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1627                                       MLX5_FLOW_LAYER_OUTER_L4;
1628         int ret;
1629         uint8_t next_proto = 0xFF;
1630         const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1631                                   MLX5_FLOW_LAYER_OUTER_VLAN |
1632                                   MLX5_FLOW_LAYER_INNER_VLAN);
1633
1634         if ((last_item & l2_vlan) && ether_type &&
1635             ether_type != RTE_ETHER_TYPE_IPV6)
1636                 return rte_flow_error_set(error, EINVAL,
1637                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1638                                           "IPv6 cannot follow L2/VLAN layer "
1639                                           "which ether type is not IPv6");
1640         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1641                 if (mask && spec)
1642                         next_proto = mask->hdr.proto & spec->hdr.proto;
1643                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1644                         return rte_flow_error_set(error, EINVAL,
1645                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1646                                                   item,
1647                                                   "multiple tunnel "
1648                                                   "not supported");
1649         }
1650         if (item_flags & MLX5_FLOW_LAYER_IPIP)
1651                 return rte_flow_error_set(error, EINVAL,
1652                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1653                                           "wrong tunnel type - IPv4 specified "
1654                                           "but IPv6 item provided");
1655         if (item_flags & l3m)
1656                 return rte_flow_error_set(error, ENOTSUP,
1657                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1658                                           "multiple L3 layers not supported");
1659         else if (item_flags & l4m)
1660                 return rte_flow_error_set(error, EINVAL,
1661                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1662                                           "L3 cannot follow an L4 layer.");
1663         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1664                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1665                 return rte_flow_error_set(error, EINVAL,
1666                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1667                                           "L3 cannot follow an NVGRE layer.");
1668         if (!mask)
1669                 mask = &rte_flow_item_ipv6_mask;
1670         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1671                                         acc_mask ? (const uint8_t *)acc_mask
1672                                                  : (const uint8_t *)&nic_mask,
1673                                         sizeof(struct rte_flow_item_ipv6),
1674                                         error);
1675         if (ret < 0)
1676                 return ret;
1677         return 0;
1678 }
1679
1680 /**
1681  * Validate UDP item.
1682  *
1683  * @param[in] item
1684  *   Item specification.
1685  * @param[in] item_flags
1686  *   Bit-fields that holds the items detected until now.
1687  * @param[in] target_protocol
1688  *   The next protocol in the previous item.
1689  * @param[in] flow_mask
1690  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1691  * @param[out] error
1692  *   Pointer to error structure.
1693  *
1694  * @return
1695  *   0 on success, a negative errno value otherwise and rte_errno is set.
1696  */
1697 int
1698 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1699                             uint64_t item_flags,
1700                             uint8_t target_protocol,
1701                             struct rte_flow_error *error)
1702 {
1703         const struct rte_flow_item_udp *mask = item->mask;
1704         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1705         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1706                                       MLX5_FLOW_LAYER_OUTER_L3;
1707         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1708                                       MLX5_FLOW_LAYER_OUTER_L4;
1709         int ret;
1710
1711         if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1712                 return rte_flow_error_set(error, EINVAL,
1713                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1714                                           "protocol filtering not compatible"
1715                                           " with UDP layer");
1716         if (!(item_flags & l3m))
1717                 return rte_flow_error_set(error, EINVAL,
1718                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1719                                           "L3 is mandatory to filter on L4");
1720         if (item_flags & l4m)
1721                 return rte_flow_error_set(error, EINVAL,
1722                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1723                                           "multiple L4 layers not supported");
1724         if (!mask)
1725                 mask = &rte_flow_item_udp_mask;
1726         ret = mlx5_flow_item_acceptable
1727                 (item, (const uint8_t *)mask,
1728                  (const uint8_t *)&rte_flow_item_udp_mask,
1729                  sizeof(struct rte_flow_item_udp), error);
1730         if (ret < 0)
1731                 return ret;
1732         return 0;
1733 }
1734
1735 /**
1736  * Validate TCP item.
1737  *
1738  * @param[in] item
1739  *   Item specification.
1740  * @param[in] item_flags
1741  *   Bit-fields that holds the items detected until now.
1742  * @param[in] target_protocol
1743  *   The next protocol in the previous item.
1744  * @param[out] error
1745  *   Pointer to error structure.
1746  *
1747  * @return
1748  *   0 on success, a negative errno value otherwise and rte_errno is set.
1749  */
1750 int
1751 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1752                             uint64_t item_flags,
1753                             uint8_t target_protocol,
1754                             const struct rte_flow_item_tcp *flow_mask,
1755                             struct rte_flow_error *error)
1756 {
1757         const struct rte_flow_item_tcp *mask = item->mask;
1758         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1759         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1760                                       MLX5_FLOW_LAYER_OUTER_L3;
1761         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1762                                       MLX5_FLOW_LAYER_OUTER_L4;
1763         int ret;
1764
1765         MLX5_ASSERT(flow_mask);
1766         if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1767                 return rte_flow_error_set(error, EINVAL,
1768                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1769                                           "protocol filtering not compatible"
1770                                           " with TCP layer");
1771         if (!(item_flags & l3m))
1772                 return rte_flow_error_set(error, EINVAL,
1773                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1774                                           "L3 is mandatory to filter on L4");
1775         if (item_flags & l4m)
1776                 return rte_flow_error_set(error, EINVAL,
1777                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1778                                           "multiple L4 layers not supported");
1779         if (!mask)
1780                 mask = &rte_flow_item_tcp_mask;
1781         ret = mlx5_flow_item_acceptable
1782                 (item, (const uint8_t *)mask,
1783                  (const uint8_t *)flow_mask,
1784                  sizeof(struct rte_flow_item_tcp), error);
1785         if (ret < 0)
1786                 return ret;
1787         return 0;
1788 }
1789
1790 /**
1791  * Validate VXLAN item.
1792  *
1793  * @param[in] item
1794  *   Item specification.
1795  * @param[in] item_flags
1796  *   Bit-fields that holds the items detected until now.
1797  * @param[in] target_protocol
1798  *   The next protocol in the previous item.
1799  * @param[out] error
1800  *   Pointer to error structure.
1801  *
1802  * @return
1803  *   0 on success, a negative errno value otherwise and rte_errno is set.
1804  */
1805 int
1806 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1807                               uint64_t item_flags,
1808                               struct rte_flow_error *error)
1809 {
1810         const struct rte_flow_item_vxlan *spec = item->spec;
1811         const struct rte_flow_item_vxlan *mask = item->mask;
1812         int ret;
1813         union vni {
1814                 uint32_t vlan_id;
1815                 uint8_t vni[4];
1816         } id = { .vlan_id = 0, };
1817
1818
1819         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1820                 return rte_flow_error_set(error, ENOTSUP,
1821                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1822                                           "multiple tunnel layers not"
1823                                           " supported");
1824         /*
1825          * Verify only UDPv4 is present as defined in
1826          * https://tools.ietf.org/html/rfc7348
1827          */
1828         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1829                 return rte_flow_error_set(error, EINVAL,
1830                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1831                                           "no outer UDP layer found");
1832         if (!mask)
1833                 mask = &rte_flow_item_vxlan_mask;
1834         ret = mlx5_flow_item_acceptable
1835                 (item, (const uint8_t *)mask,
1836                  (const uint8_t *)&rte_flow_item_vxlan_mask,
1837                  sizeof(struct rte_flow_item_vxlan),
1838                  error);
1839         if (ret < 0)
1840                 return ret;
1841         if (spec) {
1842                 memcpy(&id.vni[1], spec->vni, 3);
1843                 memcpy(&id.vni[1], mask->vni, 3);
1844         }
1845         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1846                 return rte_flow_error_set(error, ENOTSUP,
1847                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1848                                           "VXLAN tunnel must be fully defined");
1849         return 0;
1850 }
1851
1852 /**
1853  * Validate VXLAN_GPE item.
1854  *
1855  * @param[in] item
1856  *   Item specification.
1857  * @param[in] item_flags
1858  *   Bit-fields that holds the items detected until now.
1859  * @param[in] priv
1860  *   Pointer to the private data structure.
1861  * @param[in] target_protocol
1862  *   The next protocol in the previous item.
1863  * @param[out] error
1864  *   Pointer to error structure.
1865  *
1866  * @return
1867  *   0 on success, a negative errno value otherwise and rte_errno is set.
1868  */
1869 int
1870 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1871                                   uint64_t item_flags,
1872                                   struct rte_eth_dev *dev,
1873                                   struct rte_flow_error *error)
1874 {
1875         struct mlx5_priv *priv = dev->data->dev_private;
1876         const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1877         const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1878         int ret;
1879         union vni {
1880                 uint32_t vlan_id;
1881                 uint8_t vni[4];
1882         } id = { .vlan_id = 0, };
1883
1884         if (!priv->config.l3_vxlan_en)
1885                 return rte_flow_error_set(error, ENOTSUP,
1886                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1887                                           "L3 VXLAN is not enabled by device"
1888                                           " parameter and/or not configured in"
1889                                           " firmware");
1890         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1891                 return rte_flow_error_set(error, ENOTSUP,
1892                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1893                                           "multiple tunnel layers not"
1894                                           " supported");
1895         /*
1896          * Verify only UDPv4 is present as defined in
1897          * https://tools.ietf.org/html/rfc7348
1898          */
1899         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1900                 return rte_flow_error_set(error, EINVAL,
1901                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1902                                           "no outer UDP layer found");
1903         if (!mask)
1904                 mask = &rte_flow_item_vxlan_gpe_mask;
1905         ret = mlx5_flow_item_acceptable
1906                 (item, (const uint8_t *)mask,
1907                  (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1908                  sizeof(struct rte_flow_item_vxlan_gpe),
1909                  error);
1910         if (ret < 0)
1911                 return ret;
1912         if (spec) {
1913                 if (spec->protocol)
1914                         return rte_flow_error_set(error, ENOTSUP,
1915                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1916                                                   item,
1917                                                   "VxLAN-GPE protocol"
1918                                                   " not supported");
1919                 memcpy(&id.vni[1], spec->vni, 3);
1920                 memcpy(&id.vni[1], mask->vni, 3);
1921         }
1922         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1923                 return rte_flow_error_set(error, ENOTSUP,
1924                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1925                                           "VXLAN-GPE tunnel must be fully"
1926                                           " defined");
1927         return 0;
1928 }
1929 /**
1930  * Validate GRE Key item.
1931  *
1932  * @param[in] item
1933  *   Item specification.
1934  * @param[in] item_flags
1935  *   Bit flags to mark detected items.
1936  * @param[in] gre_item
1937  *   Pointer to gre_item
1938  * @param[out] error
1939  *   Pointer to error structure.
1940  *
1941  * @return
1942  *   0 on success, a negative errno value otherwise and rte_errno is set.
1943  */
1944 int
1945 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1946                                 uint64_t item_flags,
1947                                 const struct rte_flow_item *gre_item,
1948                                 struct rte_flow_error *error)
1949 {
1950         const rte_be32_t *mask = item->mask;
1951         int ret = 0;
1952         rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1953         const struct rte_flow_item_gre *gre_spec;
1954         const struct rte_flow_item_gre *gre_mask;
1955
1956         if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1957                 return rte_flow_error_set(error, ENOTSUP,
1958                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1959                                           "Multiple GRE key not support");
1960         if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1961                 return rte_flow_error_set(error, ENOTSUP,
1962                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1963                                           "No preceding GRE header");
1964         if (item_flags & MLX5_FLOW_LAYER_INNER)
1965                 return rte_flow_error_set(error, ENOTSUP,
1966                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1967                                           "GRE key following a wrong item");
1968         gre_mask = gre_item->mask;
1969         if (!gre_mask)
1970                 gre_mask = &rte_flow_item_gre_mask;
1971         gre_spec = gre_item->spec;
1972         if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
1973                          !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
1974                 return rte_flow_error_set(error, EINVAL,
1975                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1976                                           "Key bit must be on");
1977
1978         if (!mask)
1979                 mask = &gre_key_default_mask;
1980         ret = mlx5_flow_item_acceptable
1981                 (item, (const uint8_t *)mask,
1982                  (const uint8_t *)&gre_key_default_mask,
1983                  sizeof(rte_be32_t), error);
1984         return ret;
1985 }
1986
1987 /**
1988  * Validate GRE item.
1989  *
1990  * @param[in] item
1991  *   Item specification.
1992  * @param[in] item_flags
1993  *   Bit flags to mark detected items.
1994  * @param[in] target_protocol
1995  *   The next protocol in the previous item.
1996  * @param[out] error
1997  *   Pointer to error structure.
1998  *
1999  * @return
2000  *   0 on success, a negative errno value otherwise and rte_errno is set.
2001  */
2002 int
2003 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2004                             uint64_t item_flags,
2005                             uint8_t target_protocol,
2006                             struct rte_flow_error *error)
2007 {
2008         const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2009         const struct rte_flow_item_gre *mask = item->mask;
2010         int ret;
2011         const struct rte_flow_item_gre nic_mask = {
2012                 .c_rsvd0_ver = RTE_BE16(0xB000),
2013                 .protocol = RTE_BE16(UINT16_MAX),
2014         };
2015
2016         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2017                 return rte_flow_error_set(error, EINVAL,
2018                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2019                                           "protocol filtering not compatible"
2020                                           " with this GRE layer");
2021         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2022                 return rte_flow_error_set(error, ENOTSUP,
2023                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2024                                           "multiple tunnel layers not"
2025                                           " supported");
2026         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2027                 return rte_flow_error_set(error, ENOTSUP,
2028                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2029                                           "L3 Layer is missing");
2030         if (!mask)
2031                 mask = &rte_flow_item_gre_mask;
2032         ret = mlx5_flow_item_acceptable
2033                 (item, (const uint8_t *)mask,
2034                  (const uint8_t *)&nic_mask,
2035                  sizeof(struct rte_flow_item_gre), error);
2036         if (ret < 0)
2037                 return ret;
2038 #ifndef HAVE_MLX5DV_DR
2039 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2040         if (spec && (spec->protocol & mask->protocol))
2041                 return rte_flow_error_set(error, ENOTSUP,
2042                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2043                                           "without MPLS support the"
2044                                           " specification cannot be used for"
2045                                           " filtering");
2046 #endif
2047 #endif
2048         return 0;
2049 }
2050
2051 /**
2052  * Validate Geneve item.
2053  *
2054  * @param[in] item
2055  *   Item specification.
2056  * @param[in] itemFlags
2057  *   Bit-fields that holds the items detected until now.
2058  * @param[in] enPriv
2059  *   Pointer to the private data structure.
2060  * @param[out] error
2061  *   Pointer to error structure.
2062  *
2063  * @return
2064  *   0 on success, a negative errno value otherwise and rte_errno is set.
2065  */
2066
2067 int
2068 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2069                                uint64_t item_flags,
2070                                struct rte_eth_dev *dev,
2071                                struct rte_flow_error *error)
2072 {
2073         struct mlx5_priv *priv = dev->data->dev_private;
2074         const struct rte_flow_item_geneve *spec = item->spec;
2075         const struct rte_flow_item_geneve *mask = item->mask;
2076         int ret;
2077         uint16_t gbhdr;
2078         uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2079                           MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2080         const struct rte_flow_item_geneve nic_mask = {
2081                 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2082                 .vni = "\xff\xff\xff",
2083                 .protocol = RTE_BE16(UINT16_MAX),
2084         };
2085
2086         if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2087                 return rte_flow_error_set(error, ENOTSUP,
2088                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2089                                           "L3 Geneve is not enabled by device"
2090                                           " parameter and/or not configured in"
2091                                           " firmware");
2092         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2093                 return rte_flow_error_set(error, ENOTSUP,
2094                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2095                                           "multiple tunnel layers not"
2096                                           " supported");
2097         /*
2098          * Verify only UDPv4 is present as defined in
2099          * https://tools.ietf.org/html/rfc7348
2100          */
2101         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2102                 return rte_flow_error_set(error, EINVAL,
2103                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2104                                           "no outer UDP layer found");
2105         if (!mask)
2106                 mask = &rte_flow_item_geneve_mask;
2107         ret = mlx5_flow_item_acceptable
2108                                   (item, (const uint8_t *)mask,
2109                                    (const uint8_t *)&nic_mask,
2110                                    sizeof(struct rte_flow_item_geneve), error);
2111         if (ret)
2112                 return ret;
2113         if (spec) {
2114                 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2115                 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2116                      MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2117                      MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2118                         return rte_flow_error_set(error, ENOTSUP,
2119                                                   RTE_FLOW_ERROR_TYPE_ITEM,
2120                                                   item,
2121                                                   "Geneve protocol unsupported"
2122                                                   " fields are being used");
2123                 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2124                         return rte_flow_error_set
2125                                         (error, ENOTSUP,
2126                                          RTE_FLOW_ERROR_TYPE_ITEM,
2127                                          item,
2128                                          "Unsupported Geneve options length");
2129         }
2130         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2131                 return rte_flow_error_set
2132                                     (error, ENOTSUP,
2133                                      RTE_FLOW_ERROR_TYPE_ITEM, item,
2134                                      "Geneve tunnel must be fully defined");
2135         return 0;
2136 }
2137
2138 /**
2139  * Validate MPLS item.
2140  *
2141  * @param[in] dev
2142  *   Pointer to the rte_eth_dev structure.
2143  * @param[in] item
2144  *   Item specification.
2145  * @param[in] item_flags
2146  *   Bit-fields that holds the items detected until now.
2147  * @param[in] prev_layer
2148  *   The protocol layer indicated in previous item.
2149  * @param[out] error
2150  *   Pointer to error structure.
2151  *
2152  * @return
2153  *   0 on success, a negative errno value otherwise and rte_errno is set.
2154  */
2155 int
2156 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2157                              const struct rte_flow_item *item __rte_unused,
2158                              uint64_t item_flags __rte_unused,
2159                              uint64_t prev_layer __rte_unused,
2160                              struct rte_flow_error *error)
2161 {
2162 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2163         const struct rte_flow_item_mpls *mask = item->mask;
2164         struct mlx5_priv *priv = dev->data->dev_private;
2165         int ret;
2166
2167         if (!priv->config.mpls_en)
2168                 return rte_flow_error_set(error, ENOTSUP,
2169                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2170                                           "MPLS not supported or"
2171                                           " disabled in firmware"
2172                                           " configuration.");
2173         /* MPLS over IP, UDP, GRE is allowed */
2174         if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2175                             MLX5_FLOW_LAYER_OUTER_L4_UDP |
2176                             MLX5_FLOW_LAYER_GRE)))
2177                 return rte_flow_error_set(error, EINVAL,
2178                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2179                                           "protocol filtering not compatible"
2180                                           " with MPLS layer");
2181         /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2182         if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2183             !(item_flags & MLX5_FLOW_LAYER_GRE))
2184                 return rte_flow_error_set(error, ENOTSUP,
2185                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2186                                           "multiple tunnel layers not"
2187                                           " supported");
2188         if (!mask)
2189                 mask = &rte_flow_item_mpls_mask;
2190         ret = mlx5_flow_item_acceptable
2191                 (item, (const uint8_t *)mask,
2192                  (const uint8_t *)&rte_flow_item_mpls_mask,
2193                  sizeof(struct rte_flow_item_mpls), error);
2194         if (ret < 0)
2195                 return ret;
2196         return 0;
2197 #else
2198         return rte_flow_error_set(error, ENOTSUP,
2199                                   RTE_FLOW_ERROR_TYPE_ITEM, item,
2200                                   "MPLS is not supported by Verbs, please"
2201                                   " update.");
2202 #endif
2203 }
2204
2205 /**
2206  * Validate NVGRE item.
2207  *
2208  * @param[in] item
2209  *   Item specification.
2210  * @param[in] item_flags
2211  *   Bit flags to mark detected items.
2212  * @param[in] target_protocol
2213  *   The next protocol in the previous item.
2214  * @param[out] error
2215  *   Pointer to error structure.
2216  *
2217  * @return
2218  *   0 on success, a negative errno value otherwise and rte_errno is set.
2219  */
2220 int
2221 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2222                               uint64_t item_flags,
2223                               uint8_t target_protocol,
2224                               struct rte_flow_error *error)
2225 {
2226         const struct rte_flow_item_nvgre *mask = item->mask;
2227         int ret;
2228
2229         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2230                 return rte_flow_error_set(error, EINVAL,
2231                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2232                                           "protocol filtering not compatible"
2233                                           " with this GRE layer");
2234         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2235                 return rte_flow_error_set(error, ENOTSUP,
2236                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2237                                           "multiple tunnel layers not"
2238                                           " supported");
2239         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2240                 return rte_flow_error_set(error, ENOTSUP,
2241                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2242                                           "L3 Layer is missing");
2243         if (!mask)
2244                 mask = &rte_flow_item_nvgre_mask;
2245         ret = mlx5_flow_item_acceptable
2246                 (item, (const uint8_t *)mask,
2247                  (const uint8_t *)&rte_flow_item_nvgre_mask,
2248                  sizeof(struct rte_flow_item_nvgre), error);
2249         if (ret < 0)
2250                 return ret;
2251         return 0;
2252 }
2253
2254 /**
2255  * Validate eCPRI item.
2256  *
2257  * @param[in] item
2258  *   Item specification.
2259  * @param[in] item_flags
2260  *   Bit-fields that holds the items detected until now.
2261  * @param[in] last_item
2262  *   Previous validated item in the pattern items.
2263  * @param[in] ether_type
2264  *   Type in the ethernet layer header (including dot1q).
2265  * @param[in] acc_mask
2266  *   Acceptable mask, if NULL default internal default mask
2267  *   will be used to check whether item fields are supported.
2268  * @param[out] error
2269  *   Pointer to error structure.
2270  *
2271  * @return
2272  *   0 on success, a negative errno value otherwise and rte_errno is set.
2273  */
2274 int
2275 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2276                               uint64_t item_flags,
2277                               uint64_t last_item,
2278                               uint16_t ether_type,
2279                               const struct rte_flow_item_ecpri *acc_mask,
2280                               struct rte_flow_error *error)
2281 {
2282         const struct rte_flow_item_ecpri *mask = item->mask;
2283         const struct rte_flow_item_ecpri nic_mask = {
2284                 .hdr = {
2285                         .common = {
2286                                 .u32 =
2287                                 RTE_BE32(((const struct rte_ecpri_common_hdr) {
2288                                         .type = 0xFF,
2289                                         }).u32),
2290                         },
2291                         .dummy[0] = 0xFFFFFFFF,
2292                 },
2293         };
2294         const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2295                                         MLX5_FLOW_LAYER_OUTER_VLAN);
2296         struct rte_flow_item_ecpri mask_lo;
2297
2298         if ((last_item & outer_l2_vlan) && ether_type &&
2299             ether_type != RTE_ETHER_TYPE_ECPRI)
2300                 return rte_flow_error_set(error, EINVAL,
2301                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2302                                           "eCPRI cannot follow L2/VLAN layer "
2303                                           "which ether type is not 0xAEFE.");
2304         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2305                 return rte_flow_error_set(error, EINVAL,
2306                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2307                                           "eCPRI with tunnel is not supported "
2308                                           "right now.");
2309         if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2310                 return rte_flow_error_set(error, ENOTSUP,
2311                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2312                                           "multiple L3 layers not supported");
2313         else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2314                 return rte_flow_error_set(error, EINVAL,
2315                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2316                                           "eCPRI cannot follow a TCP layer.");
2317         /* In specification, eCPRI could be over UDP layer. */
2318         else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2319                 return rte_flow_error_set(error, EINVAL,
2320                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2321                                           "eCPRI over UDP layer is not yet "
2322                                           "supported right now.");
2323         /* Mask for type field in common header could be zero. */
2324         if (!mask)
2325                 mask = &rte_flow_item_ecpri_mask;
2326         mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2327         /* Input mask is in big-endian format. */
2328         if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2329                 return rte_flow_error_set(error, EINVAL,
2330                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2331                                           "partial mask is not supported "
2332                                           "for protocol");
2333         else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2334                 return rte_flow_error_set(error, EINVAL,
2335                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2336                                           "message header mask must be after "
2337                                           "a type mask");
2338         return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2339                                          acc_mask ? (const uint8_t *)acc_mask
2340                                                   : (const uint8_t *)&nic_mask,
2341                                          sizeof(struct rte_flow_item_ecpri),
2342                                          error);
2343 }
2344
2345 /* Allocate unique ID for the split Q/RSS subflows. */
2346 static uint32_t
2347 flow_qrss_get_id(struct rte_eth_dev *dev)
2348 {
2349         struct mlx5_priv *priv = dev->data->dev_private;
2350         uint32_t qrss_id, ret;
2351
2352         ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2353         if (ret)
2354                 return 0;
2355         MLX5_ASSERT(qrss_id);
2356         return qrss_id;
2357 }
2358
2359 /* Free unique ID for the split Q/RSS subflows. */
2360 static void
2361 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2362 {
2363         struct mlx5_priv *priv = dev->data->dev_private;
2364
2365         if (qrss_id)
2366                 mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2367 }
2368
2369 /**
2370  * Release resource related QUEUE/RSS action split.
2371  *
2372  * @param dev
2373  *   Pointer to Ethernet device.
2374  * @param flow
2375  *   Flow to release id's from.
2376  */
2377 static void
2378 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2379                              struct rte_flow *flow)
2380 {
2381         struct mlx5_priv *priv = dev->data->dev_private;
2382         uint32_t handle_idx;
2383         struct mlx5_flow_handle *dev_handle;
2384
2385         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2386                        handle_idx, dev_handle, next)
2387                 if (dev_handle->split_flow_id)
2388                         flow_qrss_free_id(dev, dev_handle->split_flow_id);
2389 }
2390
2391 static int
2392 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2393                    const struct rte_flow_attr *attr __rte_unused,
2394                    const struct rte_flow_item items[] __rte_unused,
2395                    const struct rte_flow_action actions[] __rte_unused,
2396                    bool external __rte_unused,
2397                    int hairpin __rte_unused,
2398                    struct rte_flow_error *error)
2399 {
2400         return rte_flow_error_set(error, ENOTSUP,
2401                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2402 }
2403
2404 static struct mlx5_flow *
2405 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2406                   const struct rte_flow_attr *attr __rte_unused,
2407                   const struct rte_flow_item items[] __rte_unused,
2408                   const struct rte_flow_action actions[] __rte_unused,
2409                   struct rte_flow_error *error)
2410 {
2411         rte_flow_error_set(error, ENOTSUP,
2412                            RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2413         return NULL;
2414 }
2415
2416 static int
2417 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2418                     struct mlx5_flow *dev_flow __rte_unused,
2419                     const struct rte_flow_attr *attr __rte_unused,
2420                     const struct rte_flow_item items[] __rte_unused,
2421                     const struct rte_flow_action actions[] __rte_unused,
2422                     struct rte_flow_error *error)
2423 {
2424         return rte_flow_error_set(error, ENOTSUP,
2425                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2426 }
2427
2428 static int
2429 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2430                 struct rte_flow *flow __rte_unused,
2431                 struct rte_flow_error *error)
2432 {
2433         return rte_flow_error_set(error, ENOTSUP,
2434                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2435 }
2436
2437 static void
2438 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2439                  struct rte_flow *flow __rte_unused)
2440 {
2441 }
2442
2443 static void
2444 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2445                   struct rte_flow *flow __rte_unused)
2446 {
2447 }
2448
2449 static int
2450 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2451                 struct rte_flow *flow __rte_unused,
2452                 const struct rte_flow_action *actions __rte_unused,
2453                 void *data __rte_unused,
2454                 struct rte_flow_error *error)
2455 {
2456         return rte_flow_error_set(error, ENOTSUP,
2457                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2458 }
2459
2460 /* Void driver to protect from null pointer reference. */
2461 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2462         .validate = flow_null_validate,
2463         .prepare = flow_null_prepare,
2464         .translate = flow_null_translate,
2465         .apply = flow_null_apply,
2466         .remove = flow_null_remove,
2467         .destroy = flow_null_destroy,
2468         .query = flow_null_query,
2469 };
2470
2471 /**
2472  * Select flow driver type according to flow attributes and device
2473  * configuration.
2474  *
2475  * @param[in] dev
2476  *   Pointer to the dev structure.
2477  * @param[in] attr
2478  *   Pointer to the flow attributes.
2479  *
2480  * @return
2481  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2482  */
2483 static enum mlx5_flow_drv_type
2484 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2485 {
2486         struct mlx5_priv *priv = dev->data->dev_private;
2487         /* The OS can determine first a specific flow type (DV, VERBS) */
2488         enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
2489
2490         if (type != MLX5_FLOW_TYPE_MAX)
2491                 return type;
2492         /* If no OS specific type - continue with DV/VERBS selection */
2493         if (attr->transfer && priv->config.dv_esw_en)
2494                 type = MLX5_FLOW_TYPE_DV;
2495         if (!attr->transfer)
2496                 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2497                                                  MLX5_FLOW_TYPE_VERBS;
2498         return type;
2499 }
2500
2501 #define flow_get_drv_ops(type) flow_drv_ops[type]
2502
2503 /**
2504  * Flow driver validation API. This abstracts calling driver specific functions.
2505  * The type of flow driver is determined according to flow attributes.
2506  *
2507  * @param[in] dev
2508  *   Pointer to the dev structure.
2509  * @param[in] attr
2510  *   Pointer to the flow attributes.
2511  * @param[in] items
2512  *   Pointer to the list of items.
2513  * @param[in] actions
2514  *   Pointer to the list of actions.
2515  * @param[in] external
2516  *   This flow rule is created by request external to PMD.
2517  * @param[in] hairpin
2518  *   Number of hairpin TX actions, 0 means classic flow.
2519  * @param[out] error
2520  *   Pointer to the error structure.
2521  *
2522  * @return
2523  *   0 on success, a negative errno value otherwise and rte_errno is set.
2524  */
2525 static inline int
2526 flow_drv_validate(struct rte_eth_dev *dev,
2527                   const struct rte_flow_attr *attr,
2528                   const struct rte_flow_item items[],
2529                   const struct rte_flow_action actions[],
2530                   bool external, int hairpin, struct rte_flow_error *error)
2531 {
2532         const struct mlx5_flow_driver_ops *fops;
2533         enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2534
2535         fops = flow_get_drv_ops(type);
2536         return fops->validate(dev, attr, items, actions, external,
2537                               hairpin, error);
2538 }
2539
2540 /**
2541  * Flow driver preparation API. This abstracts calling driver specific
2542  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2543  * calculates the size of memory required for device flow, allocates the memory,
2544  * initializes the device flow and returns the pointer.
2545  *
2546  * @note
2547  *   This function initializes device flow structure such as dv or verbs in
2548  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2549  *   rest. For example, adding returning device flow to flow->dev_flow list and
2550  *   setting backward reference to the flow should be done out of this function.
2551  *   layers field is not filled either.
2552  *
2553  * @param[in] dev
2554  *   Pointer to the dev structure.
2555  * @param[in] attr
2556  *   Pointer to the flow attributes.
2557  * @param[in] items
2558  *   Pointer to the list of items.
2559  * @param[in] actions
2560  *   Pointer to the list of actions.
2561  * @param[in] flow_idx
2562  *   This memory pool index to the flow.
2563  * @param[out] error
2564  *   Pointer to the error structure.
2565  *
2566  * @return
2567  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2568  */
2569 static inline struct mlx5_flow *
2570 flow_drv_prepare(struct rte_eth_dev *dev,
2571                  const struct rte_flow *flow,
2572                  const struct rte_flow_attr *attr,
2573                  const struct rte_flow_item items[],
2574                  const struct rte_flow_action actions[],
2575                  uint32_t flow_idx,
2576                  struct rte_flow_error *error)
2577 {
2578         const struct mlx5_flow_driver_ops *fops;
2579         enum mlx5_flow_drv_type type = flow->drv_type;
2580         struct mlx5_flow *mlx5_flow = NULL;
2581
2582         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2583         fops = flow_get_drv_ops(type);
2584         mlx5_flow = fops->prepare(dev, attr, items, actions, error);
2585         if (mlx5_flow)
2586                 mlx5_flow->flow_idx = flow_idx;
2587         return mlx5_flow;
2588 }
2589
2590 /**
2591  * Flow driver translation API. This abstracts calling driver specific
2592  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2593  * translates a generic flow into a driver flow. flow_drv_prepare() must
2594  * precede.
2595  *
2596  * @note
2597  *   dev_flow->layers could be filled as a result of parsing during translation
2598  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2599  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2600  *   flow->actions could be overwritten even though all the expanded dev_flows
2601  *   have the same actions.
2602  *
2603  * @param[in] dev
2604  *   Pointer to the rte dev structure.
2605  * @param[in, out] dev_flow
2606  *   Pointer to the mlx5 flow.
2607  * @param[in] attr
2608  *   Pointer to the flow attributes.
2609  * @param[in] items
2610  *   Pointer to the list of items.
2611  * @param[in] actions
2612  *   Pointer to the list of actions.
2613  * @param[out] error
2614  *   Pointer to the error structure.
2615  *
2616  * @return
2617  *   0 on success, a negative errno value otherwise and rte_errno is set.
2618  */
2619 static inline int
2620 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2621                    const struct rte_flow_attr *attr,
2622                    const struct rte_flow_item items[],
2623                    const struct rte_flow_action actions[],
2624                    struct rte_flow_error *error)
2625 {
2626         const struct mlx5_flow_driver_ops *fops;
2627         enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2628
2629         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2630         fops = flow_get_drv_ops(type);
2631         return fops->translate(dev, dev_flow, attr, items, actions, error);
2632 }
2633
2634 /**
2635  * Flow driver apply API. This abstracts calling driver specific functions.
2636  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2637  * translated driver flows on to device. flow_drv_translate() must precede.
2638  *
2639  * @param[in] dev
2640  *   Pointer to Ethernet device structure.
2641  * @param[in, out] flow
2642  *   Pointer to flow structure.
2643  * @param[out] error
2644  *   Pointer to error structure.
2645  *
2646  * @return
2647  *   0 on success, a negative errno value otherwise and rte_errno is set.
2648  */
2649 static inline int
2650 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2651                struct rte_flow_error *error)
2652 {
2653         const struct mlx5_flow_driver_ops *fops;
2654         enum mlx5_flow_drv_type type = flow->drv_type;
2655
2656         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2657         fops = flow_get_drv_ops(type);
2658         return fops->apply(dev, flow, error);
2659 }
2660
2661 /**
2662  * Flow driver remove API. This abstracts calling driver specific functions.
2663  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2664  * on device. All the resources of the flow should be freed by calling
2665  * flow_drv_destroy().
2666  *
2667  * @param[in] dev
2668  *   Pointer to Ethernet device.
2669  * @param[in, out] flow
2670  *   Pointer to flow structure.
2671  */
2672 static inline void
2673 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2674 {
2675         const struct mlx5_flow_driver_ops *fops;
2676         enum mlx5_flow_drv_type type = flow->drv_type;
2677
2678         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2679         fops = flow_get_drv_ops(type);
2680         fops->remove(dev, flow);
2681 }
2682
2683 /**
2684  * Flow driver destroy API. This abstracts calling driver specific functions.
2685  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2686  * on device and releases resources of the flow.
2687  *
2688  * @param[in] dev
2689  *   Pointer to Ethernet device.
2690  * @param[in, out] flow
2691  *   Pointer to flow structure.
2692  */
2693 static inline void
2694 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2695 {
2696         const struct mlx5_flow_driver_ops *fops;
2697         enum mlx5_flow_drv_type type = flow->drv_type;
2698
2699         flow_mreg_split_qrss_release(dev, flow);
2700         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2701         fops = flow_get_drv_ops(type);
2702         fops->destroy(dev, flow);
2703 }
2704
2705 /**
2706  * Get RSS action from the action list.
2707  *
2708  * @param[in] actions
2709  *   Pointer to the list of actions.
2710  *
2711  * @return
2712  *   Pointer to the RSS action if exist, else return NULL.
2713  */
2714 static const struct rte_flow_action_rss*
2715 flow_get_rss_action(const struct rte_flow_action actions[])
2716 {
2717         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2718                 switch (actions->type) {
2719                 case RTE_FLOW_ACTION_TYPE_RSS:
2720                         return (const struct rte_flow_action_rss *)
2721                                actions->conf;
2722                 default:
2723                         break;
2724                 }
2725         }
2726         return NULL;
2727 }
2728
2729 static unsigned int
2730 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2731 {
2732         const struct rte_flow_item *item;
2733         unsigned int has_vlan = 0;
2734
2735         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2736                 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2737                         has_vlan = 1;
2738                         break;
2739                 }
2740         }
2741         if (has_vlan)
2742                 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2743                                        MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2744         return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2745                                MLX5_EXPANSION_ROOT_OUTER;
2746 }
2747
2748 /**
2749  *  Get layer flags from the prefix flow.
2750  *
2751  *  Some flows may be split to several subflows, the prefix subflow gets the
2752  *  match items and the suffix sub flow gets the actions.
2753  *  Some actions need the user defined match item flags to get the detail for
2754  *  the action.
2755  *  This function helps the suffix flow to get the item layer flags from prefix
2756  *  subflow.
2757  *
2758  * @param[in] dev_flow
2759  *   Pointer the created preifx subflow.
2760  *
2761  * @return
2762  *   The layers get from prefix subflow.
2763  */
2764 static inline uint64_t
2765 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2766 {
2767         uint64_t layers = 0;
2768
2769         /*
2770          * Layers bits could be localization, but usually the compiler will
2771          * help to do the optimization work for source code.
2772          * If no decap actions, use the layers directly.
2773          */
2774         if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2775                 return dev_flow->handle->layers;
2776         /* Convert L3 layers with decap action. */
2777         if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2778                 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2779         else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2780                 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2781         /* Convert L4 layers with decap action.  */
2782         if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2783                 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2784         else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2785                 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2786         return layers;
2787 }
2788
2789 /**
2790  * Get metadata split action information.
2791  *
2792  * @param[in] actions
2793  *   Pointer to the list of actions.
2794  * @param[out] qrss
2795  *   Pointer to the return pointer.
2796  * @param[out] qrss_type
2797  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2798  *   if no QUEUE/RSS is found.
2799  * @param[out] encap_idx
2800  *   Pointer to the index of the encap action if exists, otherwise the last
2801  *   action index.
2802  *
2803  * @return
2804  *   Total number of actions.
2805  */
2806 static int
2807 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2808                                        const struct rte_flow_action **qrss,
2809                                        int *encap_idx)
2810 {
2811         const struct rte_flow_action_raw_encap *raw_encap;
2812         int actions_n = 0;
2813         int raw_decap_idx = -1;
2814
2815         *encap_idx = -1;
2816         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2817                 switch (actions->type) {
2818                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2819                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2820                         *encap_idx = actions_n;
2821                         break;
2822                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2823                         raw_decap_idx = actions_n;
2824                         break;
2825                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2826                         raw_encap = actions->conf;
2827                         if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2828                                 *encap_idx = raw_decap_idx != -1 ?
2829                                                       raw_decap_idx : actions_n;
2830                         break;
2831                 case RTE_FLOW_ACTION_TYPE_QUEUE:
2832                 case RTE_FLOW_ACTION_TYPE_RSS:
2833                         *qrss = actions;
2834                         break;
2835                 default:
2836                         break;
2837                 }
2838                 actions_n++;
2839         }
2840         if (*encap_idx == -1)
2841                 *encap_idx = actions_n;
2842         /* Count RTE_FLOW_ACTION_TYPE_END. */
2843         return actions_n + 1;
2844 }
2845
2846 /**
2847  * Check meter action from the action list.
2848  *
2849  * @param[in] actions
2850  *   Pointer to the list of actions.
2851  * @param[out] mtr
2852  *   Pointer to the meter exist flag.
2853  *
2854  * @return
2855  *   Total number of actions.
2856  */
2857 static int
2858 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2859 {
2860         int actions_n = 0;
2861
2862         MLX5_ASSERT(mtr);
2863         *mtr = 0;
2864         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2865                 switch (actions->type) {
2866                 case RTE_FLOW_ACTION_TYPE_METER:
2867                         *mtr = 1;
2868                         break;
2869                 default:
2870                         break;
2871                 }
2872                 actions_n++;
2873         }
2874         /* Count RTE_FLOW_ACTION_TYPE_END. */
2875         return actions_n + 1;
2876 }
2877
2878 /**
2879  * Check if the flow should be split due to hairpin.
2880  * The reason for the split is that in current HW we can't
2881  * support encap and push-vlan on Rx, so if a flow contains
2882  * these actions we move it to Tx.
2883  *
2884  * @param dev
2885  *   Pointer to Ethernet device.
2886  * @param[in] attr
2887  *   Flow rule attributes.
2888  * @param[in] actions
2889  *   Associated actions (list terminated by the END action).
2890  *
2891  * @return
2892  *   > 0 the number of actions and the flow should be split,
2893  *   0 when no split required.
2894  */
2895 static int
2896 flow_check_hairpin_split(struct rte_eth_dev *dev,
2897                          const struct rte_flow_attr *attr,
2898                          const struct rte_flow_action actions[])
2899 {
2900         int queue_action = 0;
2901         int action_n = 0;
2902         int split = 0;
2903         const struct rte_flow_action_queue *queue;
2904         const struct rte_flow_action_rss *rss;
2905         const struct rte_flow_action_raw_encap *raw_encap;
2906
2907         if (!attr->ingress)
2908                 return 0;
2909         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2910                 switch (actions->type) {
2911                 case RTE_FLOW_ACTION_TYPE_QUEUE:
2912                         queue = actions->conf;
2913                         if (queue == NULL)
2914                                 return 0;
2915                         if (mlx5_rxq_get_type(dev, queue->index) !=
2916                             MLX5_RXQ_TYPE_HAIRPIN)
2917                                 return 0;
2918                         queue_action = 1;
2919                         action_n++;
2920                         break;
2921                 case RTE_FLOW_ACTION_TYPE_RSS:
2922                         rss = actions->conf;
2923                         if (rss == NULL || rss->queue_num == 0)
2924                                 return 0;
2925                         if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2926                             MLX5_RXQ_TYPE_HAIRPIN)
2927                                 return 0;
2928                         queue_action = 1;
2929                         action_n++;
2930                         break;
2931                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2932                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2933                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
2934                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
2935                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
2936                         split++;
2937                         action_n++;
2938                         break;
2939                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2940                         raw_encap = actions->conf;
2941                         if (raw_encap->size >
2942                             (sizeof(struct rte_flow_item_eth) +
2943                              sizeof(struct rte_flow_item_ipv4)))
2944                                 split++;
2945                         action_n++;
2946                         break;
2947                 default:
2948                         action_n++;
2949                         break;
2950                 }
2951         }
2952         if (split && queue_action)
2953                 return action_n;
2954         return 0;
2955 }
2956
2957 /* Declare flow create/destroy prototype in advance. */
2958 static uint32_t
2959 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
2960                  const struct rte_flow_attr *attr,
2961                  const struct rte_flow_item items[],
2962                  const struct rte_flow_action actions[],
2963                  bool external, struct rte_flow_error *error);
2964
2965 static void
2966 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
2967                   uint32_t flow_idx);
2968
2969 /**
2970  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2971  *
2972  * As mark_id is unique, if there's already a registered flow for the mark_id,
2973  * return by increasing the reference counter of the resource. Otherwise, create
2974  * the resource (mcp_res) and flow.
2975  *
2976  * Flow looks like,
2977  *   - If ingress port is ANY and reg_c[1] is mark_id,
2978  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2979  *
2980  * For default flow (zero mark_id), flow is like,
2981  *   - If ingress port is ANY,
2982  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2983  *
2984  * @param dev
2985  *   Pointer to Ethernet device.
2986  * @param mark_id
2987  *   ID of MARK action, zero means default flow for META.
2988  * @param[out] error
2989  *   Perform verbose error reporting if not NULL.
2990  *
2991  * @return
2992  *   Associated resource on success, NULL otherwise and rte_errno is set.
2993  */
2994 static struct mlx5_flow_mreg_copy_resource *
2995 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2996                           struct rte_flow_error *error)
2997 {
2998         struct mlx5_priv *priv = dev->data->dev_private;
2999         struct rte_flow_attr attr = {
3000                 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3001                 .ingress = 1,
3002         };
3003         struct mlx5_rte_flow_item_tag tag_spec = {
3004                 .data = mark_id,
3005         };
3006         struct rte_flow_item items[] = {
3007                 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3008         };
3009         struct rte_flow_action_mark ftag = {
3010                 .id = mark_id,
3011         };
3012         struct mlx5_flow_action_copy_mreg cp_mreg = {
3013                 .dst = REG_B,
3014                 .src = REG_NON,
3015         };
3016         struct rte_flow_action_jump jump = {
3017                 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3018         };
3019         struct rte_flow_action actions[] = {
3020                 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3021         };
3022         struct mlx5_flow_mreg_copy_resource *mcp_res;
3023         uint32_t idx = 0;
3024         int ret;
3025
3026         /* Fill the register fileds in the flow. */
3027         ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3028         if (ret < 0)
3029                 return NULL;
3030         tag_spec.id = ret;
3031         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3032         if (ret < 0)
3033                 return NULL;
3034         cp_mreg.src = ret;
3035         /* Check if already registered. */
3036         MLX5_ASSERT(priv->mreg_cp_tbl);
3037         mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3038         if (mcp_res) {
3039                 /* For non-default rule. */
3040                 if (mark_id != MLX5_DEFAULT_COPY_ID)
3041                         mcp_res->refcnt++;
3042                 MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3043                             mcp_res->refcnt == 1);
3044                 return mcp_res;
3045         }
3046         /* Provide the full width of FLAG specific value. */
3047         if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3048                 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3049         /* Build a new flow. */
3050         if (mark_id != MLX5_DEFAULT_COPY_ID) {
3051                 items[0] = (struct rte_flow_item){
3052                         .type = (enum rte_flow_item_type)
3053                                 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3054                         .spec = &tag_spec,
3055                 };
3056                 items[1] = (struct rte_flow_item){
3057                         .type = RTE_FLOW_ITEM_TYPE_END,
3058                 };
3059                 actions[0] = (struct rte_flow_action){
3060                         .type = (enum rte_flow_action_type)
3061                                 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3062                         .conf = &ftag,
3063                 };
3064                 actions[1] = (struct rte_flow_action){
3065                         .type = (enum rte_flow_action_type)
3066                                 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3067                         .conf = &cp_mreg,
3068                 };
3069                 actions[2] = (struct rte_flow_action){
3070                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
3071                         .conf = &jump,
3072                 };
3073                 actions[3] = (struct rte_flow_action){
3074                         .type = RTE_FLOW_ACTION_TYPE_END,
3075                 };
3076         } else {
3077                 /* Default rule, wildcard match. */
3078                 attr.priority = MLX5_FLOW_PRIO_RSVD;
3079                 items[0] = (struct rte_flow_item){
3080                         .type = RTE_FLOW_ITEM_TYPE_END,
3081                 };
3082                 actions[0] = (struct rte_flow_action){
3083                         .type = (enum rte_flow_action_type)
3084                                 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3085                         .conf = &cp_mreg,
3086                 };
3087                 actions[1] = (struct rte_flow_action){
3088                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
3089                         .conf = &jump,
3090                 };
3091                 actions[2] = (struct rte_flow_action){
3092                         .type = RTE_FLOW_ACTION_TYPE_END,
3093                 };
3094         }
3095         /* Build a new entry. */
3096         mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3097         if (!mcp_res) {
3098                 rte_errno = ENOMEM;
3099                 return NULL;
3100         }
3101         mcp_res->idx = idx;
3102         /*
3103          * The copy Flows are not included in any list. There
3104          * ones are referenced from other Flows and can not
3105          * be applied, removed, deleted in ardbitrary order
3106          * by list traversing.
3107          */
3108         mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3109                                          actions, false, error);
3110         if (!mcp_res->rix_flow)
3111                 goto error;
3112         mcp_res->refcnt++;
3113         mcp_res->hlist_ent.key = mark_id;
3114         ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3115                                 &mcp_res->hlist_ent);
3116         MLX5_ASSERT(!ret);
3117         if (ret)
3118                 goto error;
3119         return mcp_res;
3120 error:
3121         if (mcp_res->rix_flow)
3122                 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3123         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3124         return NULL;
3125 }
3126
3127 /**
3128  * Release flow in RX_CP_TBL.
3129  *
3130  * @param dev
3131  *   Pointer to Ethernet device.
3132  * @flow
3133  *   Parent flow for wich copying is provided.
3134  */
3135 static void
3136 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3137                           struct rte_flow *flow)
3138 {
3139         struct mlx5_flow_mreg_copy_resource *mcp_res;
3140         struct mlx5_priv *priv = dev->data->dev_private;
3141
3142         if (!flow->rix_mreg_copy)
3143                 return;
3144         mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3145                                  flow->rix_mreg_copy);
3146         if (!mcp_res || !priv->mreg_cp_tbl)
3147                 return;
3148         if (flow->copy_applied) {
3149                 MLX5_ASSERT(mcp_res->appcnt);
3150                 flow->copy_applied = 0;
3151                 --mcp_res->appcnt;
3152                 if (!mcp_res->appcnt) {
3153                         struct rte_flow *mcp_flow = mlx5_ipool_get
3154                                         (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3155                                         mcp_res->rix_flow);
3156
3157                         if (mcp_flow)
3158                                 flow_drv_remove(dev, mcp_flow);
3159                 }
3160         }
3161         /*
3162          * We do not check availability of metadata registers here,
3163          * because copy resources are not allocated in this case.
3164          */
3165         if (--mcp_res->refcnt)
3166                 return;
3167         MLX5_ASSERT(mcp_res->rix_flow);
3168         flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3169         mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3170         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3171         flow->rix_mreg_copy = 0;
3172 }
3173
3174 /**
3175  * Start flow in RX_CP_TBL.
3176  *
3177  * @param dev
3178  *   Pointer to Ethernet device.
3179  * @flow
3180  *   Parent flow for wich copying is provided.
3181  *
3182  * @return
3183  *   0 on success, a negative errno value otherwise and rte_errno is set.
3184  */
3185 static int
3186 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3187                             struct rte_flow *flow)
3188 {
3189         struct mlx5_flow_mreg_copy_resource *mcp_res;
3190         struct mlx5_priv *priv = dev->data->dev_private;
3191         int ret;
3192
3193         if (!flow->rix_mreg_copy || flow->copy_applied)
3194                 return 0;
3195         mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3196                                  flow->rix_mreg_copy);
3197         if (!mcp_res)
3198                 return 0;
3199         if (!mcp_res->appcnt) {
3200                 struct rte_flow *mcp_flow = mlx5_ipool_get
3201                                 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3202                                 mcp_res->rix_flow);
3203
3204                 if (mcp_flow) {
3205                         ret = flow_drv_apply(dev, mcp_flow, NULL);
3206                         if (ret)
3207                                 return ret;
3208                 }
3209         }
3210         ++mcp_res->appcnt;
3211         flow->copy_applied = 1;
3212         return 0;
3213 }
3214
3215 /**
3216  * Stop flow in RX_CP_TBL.
3217  *
3218  * @param dev
3219  *   Pointer to Ethernet device.
3220  * @flow
3221  *   Parent flow for wich copying is provided.
3222  */
3223 static void
3224 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3225                            struct rte_flow *flow)
3226 {
3227         struct mlx5_flow_mreg_copy_resource *mcp_res;
3228         struct mlx5_priv *priv = dev->data->dev_private;
3229
3230         if (!flow->rix_mreg_copy || !flow->copy_applied)
3231                 return;
3232         mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3233                                  flow->rix_mreg_copy);
3234         if (!mcp_res)
3235                 return;
3236         MLX5_ASSERT(mcp_res->appcnt);
3237         --mcp_res->appcnt;
3238         flow->copy_applied = 0;
3239         if (!mcp_res->appcnt) {
3240                 struct rte_flow *mcp_flow = mlx5_ipool_get
3241                                 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3242                                 mcp_res->rix_flow);
3243
3244                 if (mcp_flow)
3245                         flow_drv_remove(dev, mcp_flow);
3246         }
3247 }
3248
3249 /**
3250  * Remove the default copy action from RX_CP_TBL.
3251  *
3252  * @param dev
3253  *   Pointer to Ethernet device.
3254  */
3255 static void
3256 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3257 {
3258         struct mlx5_flow_mreg_copy_resource *mcp_res;
3259         struct mlx5_priv *priv = dev->data->dev_private;
3260
3261         /* Check if default flow is registered. */
3262         if (!priv->mreg_cp_tbl)
3263                 return;
3264         mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3265                                             MLX5_DEFAULT_COPY_ID);
3266         if (!mcp_res)
3267                 return;
3268         MLX5_ASSERT(mcp_res->rix_flow);
3269         flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3270         mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3271         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3272 }
3273
3274 /**
3275  * Add the default copy action in in RX_CP_TBL.
3276  *
3277  * @param dev
3278  *   Pointer to Ethernet device.
3279  * @param[out] error
3280  *   Perform verbose error reporting if not NULL.
3281  *
3282  * @return
3283  *   0 for success, negative value otherwise and rte_errno is set.
3284  */
3285 static int
3286 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3287                                   struct rte_flow_error *error)
3288 {
3289         struct mlx5_priv *priv = dev->data->dev_private;
3290         struct mlx5_flow_mreg_copy_resource *mcp_res;
3291
3292         /* Check whether extensive metadata feature is engaged. */
3293         if (!priv->config.dv_flow_en ||
3294             priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3295             !mlx5_flow_ext_mreg_supported(dev) ||
3296             !priv->sh->dv_regc0_mask)
3297                 return 0;
3298         mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3299         if (!mcp_res)
3300                 return -rte_errno;
3301         return 0;
3302 }
3303
3304 /**
3305  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3306  *
3307  * All the flow having Q/RSS action should be split by
3308  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3309  * performs the following,
3310  *   - CQE->flow_tag := reg_c[1] (MARK)
3311  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3312  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3313  * but there should be a flow per each MARK ID set by MARK action.
3314  *
3315  * For the aforementioned reason, if there's a MARK action in flow's action
3316  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3317  * the MARK ID to CQE's flow_tag like,
3318  *   - If reg_c[1] is mark_id,
3319  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3320  *
3321  * For SET_META action which stores value in reg_c[0], as the destination is
3322  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3323  * MARK ID means the default flow. The default flow looks like,
3324  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3325  *
3326  * @param dev
3327  *   Pointer to Ethernet device.
3328  * @param flow
3329  *   Pointer to flow structure.
3330  * @param[in] actions
3331  *   Pointer to the list of actions.
3332  * @param[out] error
3333  *   Perform verbose error reporting if not NULL.
3334  *
3335  * @return
3336  *   0 on success, negative value otherwise and rte_errno is set.
3337  */
3338 static int
3339 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3340                             struct rte_flow *flow,
3341                             const struct rte_flow_action *actions,
3342                             struct rte_flow_error *error)
3343 {
3344         struct mlx5_priv *priv = dev->data->dev_private;
3345         struct mlx5_dev_config *config = &priv->config;
3346         struct mlx5_flow_mreg_copy_resource *mcp_res;
3347         const struct rte_flow_action_mark *mark;
3348
3349         /* Check whether extensive metadata feature is engaged. */
3350         if (!config->dv_flow_en ||
3351             config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3352             !mlx5_flow_ext_mreg_supported(dev) ||
3353             !priv->sh->dv_regc0_mask)
3354                 return 0;
3355         /* Find MARK action. */
3356         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3357                 switch (actions->type) {
3358                 case RTE_FLOW_ACTION_TYPE_FLAG:
3359                         mcp_res = flow_mreg_add_copy_action
3360                                 (dev, MLX5_FLOW_MARK_DEFAULT, error);
3361                         if (!mcp_res)
3362                                 return -rte_errno;
3363                         flow->rix_mreg_copy = mcp_res->idx;
3364                         if (dev->data->dev_started) {
3365                                 mcp_res->appcnt++;
3366                                 flow->copy_applied = 1;
3367                         }
3368                         return 0;
3369                 case RTE_FLOW_ACTION_TYPE_MARK:
3370                         mark = (const struct rte_flow_action_mark *)
3371                                 actions->conf;
3372                         mcp_res =
3373                                 flow_mreg_add_copy_action(dev, mark->id, error);
3374                         if (!mcp_res)
3375                                 return -rte_errno;
3376                         flow->rix_mreg_copy = mcp_res->idx;
3377                         if (dev->data->dev_started) {
3378                                 mcp_res->appcnt++;
3379                                 flow->copy_applied = 1;
3380                         }
3381                         return 0;
3382                 default:
3383                         break;
3384                 }
3385         }
3386         return 0;
3387 }
3388
3389 #define MLX5_MAX_SPLIT_ACTIONS 24
3390 #define MLX5_MAX_SPLIT_ITEMS 24
3391
3392 /**
3393  * Split the hairpin flow.
3394  * Since HW can't support encap and push-vlan on Rx, we move these
3395  * actions to Tx.
3396  * If the count action is after the encap then we also
3397  * move the count action. in this case the count will also measure
3398  * the outer bytes.
3399  *
3400  * @param dev
3401  *   Pointer to Ethernet device.
3402  * @param[in] actions
3403  *   Associated actions (list terminated by the END action).
3404  * @param[out] actions_rx
3405  *   Rx flow actions.
3406  * @param[out] actions_tx
3407  *   Tx flow actions..
3408  * @param[out] pattern_tx
3409  *   The pattern items for the Tx flow.
3410  * @param[out] flow_id
3411  *   The flow ID connected to this flow.
3412  *
3413  * @return
3414  *   0 on success.
3415  */
3416 static int
3417 flow_hairpin_split(struct rte_eth_dev *dev,
3418                    const struct rte_flow_action actions[],
3419                    struct rte_flow_action actions_rx[],
3420                    struct rte_flow_action actions_tx[],
3421                    struct rte_flow_item pattern_tx[],
3422                    uint32_t *flow_id)
3423 {
3424         struct mlx5_priv *priv = dev->data->dev_private;
3425         const struct rte_flow_action_raw_encap *raw_encap;
3426         const struct rte_flow_action_raw_decap *raw_decap;
3427         struct mlx5_rte_flow_action_set_tag *set_tag;
3428         struct rte_flow_action *tag_action;
3429         struct mlx5_rte_flow_item_tag *tag_item;
3430         struct rte_flow_item *item;
3431         char *addr;
3432         int encap = 0;
3433
3434         mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3435         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3436                 switch (actions->type) {
3437                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3438                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3439                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3440                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3441                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3442                         rte_memcpy(actions_tx, actions,
3443                                sizeof(struct rte_flow_action));
3444                         actions_tx++;
3445                         break;
3446                 case RTE_FLOW_ACTION_TYPE_COUNT:
3447                         if (encap) {
3448                                 rte_memcpy(actions_tx, actions,
3449                                            sizeof(struct rte_flow_action));
3450                                 actions_tx++;
3451                         } else {
3452                                 rte_memcpy(actions_rx, actions,
3453                                            sizeof(struct rte_flow_action));
3454                                 actions_rx++;
3455                         }
3456                         break;
3457                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3458                         raw_encap = actions->conf;
3459                         if (raw_encap->size >
3460                             (sizeof(struct rte_flow_item_eth) +
3461                              sizeof(struct rte_flow_item_ipv4))) {
3462                                 memcpy(actions_tx, actions,
3463                                        sizeof(struct rte_flow_action));
3464                                 actions_tx++;
3465                                 encap = 1;
3466                         } else {
3467                                 rte_memcpy(actions_rx, actions,
3468                                            sizeof(struct rte_flow_action));
3469                                 actions_rx++;
3470                         }
3471                         break;
3472                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3473                         raw_decap = actions->conf;
3474                         if (raw_decap->size <
3475                             (sizeof(struct rte_flow_item_eth) +
3476                              sizeof(struct rte_flow_item_ipv4))) {
3477                                 memcpy(actions_tx, actions,
3478                                        sizeof(struct rte_flow_action));
3479                                 actions_tx++;
3480                         } else {
3481                                 rte_memcpy(actions_rx, actions,
3482                                            sizeof(struct rte_flow_action));
3483                                 actions_rx++;
3484                         }
3485                         break;
3486                 default:
3487                         rte_memcpy(actions_rx, actions,
3488                                    sizeof(struct rte_flow_action));
3489                         actions_rx++;
3490                         break;
3491                 }
3492         }
3493         /* Add set meta action and end action for the Rx flow. */
3494         tag_action = actions_rx;
3495         tag_action->type = (enum rte_flow_action_type)
3496                            MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3497         actions_rx++;
3498         rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3499         actions_rx++;
3500         set_tag = (void *)actions_rx;
3501         set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3502         MLX5_ASSERT(set_tag->id > REG_NON);
3503         set_tag->data = *flow_id;
3504         tag_action->conf = set_tag;
3505         /* Create Tx item list. */
3506         rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3507         addr = (void *)&pattern_tx[2];
3508         item = pattern_tx;
3509         item->type = (enum rte_flow_item_type)
3510                      MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3511         tag_item = (void *)addr;
3512         tag_item->data = *flow_id;
3513         tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3514         MLX5_ASSERT(set_tag->id > REG_NON);
3515         item->spec = tag_item;
3516         addr += sizeof(struct mlx5_rte_flow_item_tag);
3517         tag_item = (void *)addr;
3518         tag_item->data = UINT32_MAX;
3519         tag_item->id = UINT16_MAX;
3520         item->mask = tag_item;
3521         item->last = NULL;
3522         item++;
3523         item->type = RTE_FLOW_ITEM_TYPE_END;
3524         return 0;
3525 }
3526
3527 /**
3528  * The last stage of splitting chain, just creates the subflow
3529  * without any modification.
3530  *
3531  * @param[in] dev
3532  *   Pointer to Ethernet device.
3533  * @param[in] flow
3534  *   Parent flow structure pointer.
3535  * @param[in, out] sub_flow
3536  *   Pointer to return the created subflow, may be NULL.
3537  * @param[in] prefix_layers
3538  *   Prefix subflow layers, may be 0.
3539  * @param[in] attr
3540  *   Flow rule attributes.
3541  * @param[in] items
3542  *   Pattern specification (list terminated by the END pattern item).
3543  * @param[in] actions
3544  *   Associated actions (list terminated by the END action).
3545  * @param[in] external
3546  *   This flow rule is created by request external to PMD.
3547  * @param[in] flow_idx
3548  *   This memory pool index to the flow.
3549  * @param[out] error
3550  *   Perform verbose error reporting if not NULL.
3551  * @return
3552  *   0 on success, negative value otherwise
3553  */
3554 static int
3555 flow_create_split_inner(struct rte_eth_dev *dev,
3556                         struct rte_flow *flow,
3557                         struct mlx5_flow **sub_flow,
3558                         uint64_t prefix_layers,
3559                         const struct rte_flow_attr *attr,
3560                         const struct rte_flow_item items[],
3561                         const struct rte_flow_action actions[],
3562                         bool external, uint32_t flow_idx,
3563                         struct rte_flow_error *error)
3564 {
3565         struct mlx5_flow *dev_flow;
3566
3567         dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
3568                 flow_idx, error);
3569         if (!dev_flow)
3570                 return -rte_errno;
3571         dev_flow->flow = flow;
3572         dev_flow->external = external;
3573         /* Subflow object was created, we must include one in the list. */
3574         SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3575                       dev_flow->handle, next);
3576         /*
3577          * If dev_flow is as one of the suffix flow, some actions in suffix
3578          * flow may need some user defined item layer flags.
3579          */
3580         if (prefix_layers)
3581                 dev_flow->handle->layers = prefix_layers;
3582         if (sub_flow)
3583                 *sub_flow = dev_flow;
3584         return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3585 }
3586
3587 /**
3588  * Split the meter flow.
3589  *
3590  * As meter flow will split to three sub flow, other than meter
3591  * action, the other actions make sense to only meter accepts
3592  * the packet. If it need to be dropped, no other additional
3593  * actions should be take.
3594  *
3595  * One kind of special action which decapsulates the L3 tunnel
3596  * header will be in the prefix sub flow, as not to take the
3597  * L3 tunnel header into account.
3598  *
3599  * @param dev
3600  *   Pointer to Ethernet device.
3601  * @param[in] items
3602  *   Pattern specification (list terminated by the END pattern item).
3603  * @param[out] sfx_items
3604  *   Suffix flow match items (list terminated by the END pattern item).
3605  * @param[in] actions
3606  *   Associated actions (list terminated by the END action).
3607  * @param[out] actions_sfx
3608  *   Suffix flow actions.
3609  * @param[out] actions_pre
3610  *   Prefix flow actions.
3611  * @param[out] pattern_sfx
3612  *   The pattern items for the suffix flow.
3613  * @param[out] tag_sfx
3614  *   Pointer to suffix flow tag.
3615  *
3616  * @return
3617  *   0 on success.
3618  */
3619 static int
3620 flow_meter_split_prep(struct rte_eth_dev *dev,
3621                  const struct rte_flow_item items[],
3622                  struct rte_flow_item sfx_items[],
3623                  const struct rte_flow_action actions[],
3624                  struct rte_flow_action actions_sfx[],
3625                  struct rte_flow_action actions_pre[])
3626 {
3627         struct rte_flow_action *tag_action = NULL;
3628         struct rte_flow_item *tag_item;
3629         struct mlx5_rte_flow_action_set_tag *set_tag;
3630         struct rte_flow_error error;
3631         const struct rte_flow_action_raw_encap *raw_encap;
3632         const struct rte_flow_action_raw_decap *raw_decap;
3633         struct mlx5_rte_flow_item_tag *tag_spec;
3634         struct mlx5_rte_flow_item_tag *tag_mask;
3635         uint32_t tag_id;
3636         bool copy_vlan = false;
3637
3638         /* Prepare the actions for prefix and suffix flow. */
3639         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3640                 struct rte_flow_action **action_cur = NULL;
3641
3642                 switch (actions->type) {
3643                 case RTE_FLOW_ACTION_TYPE_METER:
3644                         /* Add the extra tag action first. */
3645                         tag_action = actions_pre;
3646                         tag_action->type = (enum rte_flow_action_type)
3647                                            MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3648                         actions_pre++;
3649                         action_cur = &actions_pre;
3650                         break;
3651                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3652                 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3653                         action_cur = &actions_pre;
3654                         break;
3655                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3656                         raw_encap = actions->conf;
3657                         if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3658                                 action_cur = &actions_pre;
3659                         break;
3660                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3661                         raw_decap = actions->conf;
3662                         if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3663                                 action_cur = &actions_pre;
3664                         break;
3665                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3666                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3667                         copy_vlan = true;
3668                         break;
3669                 default:
3670                         break;
3671                 }
3672                 if (!action_cur)
3673                         action_cur = &actions_sfx;
3674                 memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3675                 (*action_cur)++;
3676         }
3677         /* Add end action to the actions. */
3678         actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3679         actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3680         actions_pre++;
3681         /* Set the tag. */
3682         set_tag = (void *)actions_pre;
3683         set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3684         /*
3685          * Get the id from the qrss_pool to make qrss share the id with meter.
3686          */
3687         tag_id = flow_qrss_get_id(dev);
3688         set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3689         assert(tag_action);
3690         tag_action->conf = set_tag;
3691         /* Prepare the suffix subflow items. */
3692         tag_item = sfx_items++;
3693         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3694                 int item_type = items->type;
3695
3696                 switch (item_type) {
3697                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
3698                         memcpy(sfx_items, items, sizeof(*sfx_items));
3699                         sfx_items++;
3700                         break;
3701                 case RTE_FLOW_ITEM_TYPE_VLAN:
3702                         if (copy_vlan) {
3703                                 memcpy(sfx_items, items, sizeof(*sfx_items));
3704                                 /*
3705                                  * Convert to internal match item, it is used
3706                                  * for vlan push and set vid.
3707                                  */
3708                                 sfx_items->type = (enum rte_flow_item_type)
3709                                                   MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3710                                 sfx_items++;
3711                         }
3712                         break;
3713                 default:
3714                         break;
3715                 }
3716         }
3717         sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3718         sfx_items++;
3719         tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3720         tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3721         tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3722         tag_mask = tag_spec + 1;
3723         tag_mask->data = 0xffffff00;
3724         tag_item->type = (enum rte_flow_item_type)
3725                          MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3726         tag_item->spec = tag_spec;
3727         tag_item->last = NULL;
3728         tag_item->mask = tag_mask;
3729         return tag_id;
3730 }
3731
3732 /**
3733  * Split action list having QUEUE/RSS for metadata register copy.
3734  *
3735  * Once Q/RSS action is detected in user's action list, the flow action
3736  * should be split in order to copy metadata registers, which will happen in
3737  * RX_CP_TBL like,
3738  *   - CQE->flow_tag := reg_c[1] (MARK)
3739  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3740  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3741  * This is because the last action of each flow must be a terminal action
3742  * (QUEUE, RSS or DROP).
3743  *
3744  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3745  * stored and kept in the mlx5_flow structure per each sub_flow.
3746  *
3747  * The Q/RSS action is replaced with,
3748  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3749  * And the following JUMP action is added at the end,
3750  *   - JUMP, to RX_CP_TBL.
3751  *
3752  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3753  * flow_create_split_metadata() routine. The flow will look like,
3754  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3755  *
3756  * @param dev
3757  *   Pointer to Ethernet device.
3758  * @param[out] split_actions
3759  *   Pointer to store split actions to jump to CP_TBL.
3760  * @param[in] actions
3761  *   Pointer to the list of original flow actions.
3762  * @param[in] qrss
3763  *   Pointer to the Q/RSS action.
3764  * @param[in] actions_n
3765  *   Number of original actions.
3766  * @param[out] error
3767  *   Perform verbose error reporting if not NULL.
3768  *
3769  * @return
3770  *   non-zero unique flow_id on success, otherwise 0 and
3771  *   error/rte_error are set.
3772  */
3773 static uint32_t
3774 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3775                           struct rte_flow_action *split_actions,
3776                           const struct rte_flow_action *actions,
3777                           const struct rte_flow_action *qrss,
3778                           int actions_n, struct rte_flow_error *error)
3779 {
3780         struct mlx5_rte_flow_action_set_tag *set_tag;
3781         struct rte_flow_action_jump *jump;
3782         const int qrss_idx = qrss - actions;
3783         uint32_t flow_id = 0;
3784         int ret = 0;
3785
3786         /*
3787          * Given actions will be split
3788          * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3789          * - Add jump to mreg CP_TBL.
3790          * As a result, there will be one more action.
3791          */
3792         ++actions_n;
3793         memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3794         set_tag = (void *)(split_actions + actions_n);
3795         /*
3796          * If tag action is not set to void(it means we are not the meter
3797          * suffix flow), add the tag action. Since meter suffix flow already
3798          * has the tag added.
3799          */
3800         if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3801                 /*
3802                  * Allocate the new subflow ID. This one is unique within
3803                  * device and not shared with representors. Otherwise,
3804                  * we would have to resolve multi-thread access synch
3805                  * issue. Each flow on the shared device is appended
3806                  * with source vport identifier, so the resulting
3807                  * flows will be unique in the shared (by master and
3808                  * representors) domain even if they have coinciding
3809                  * IDs.
3810                  */
3811                 flow_id = flow_qrss_get_id(dev);
3812                 if (!flow_id)
3813                         return rte_flow_error_set(error, ENOMEM,
3814                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3815                                                   NULL, "can't allocate id "
3816                                                   "for split Q/RSS subflow");
3817                 /* Internal SET_TAG action to set flow ID. */
3818                 *set_tag = (struct mlx5_rte_flow_action_set_tag){
3819                         .data = flow_id,
3820                 };
3821                 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3822                 if (ret < 0)
3823                         return ret;
3824                 set_tag->id = ret;
3825                 /* Construct new actions array. */
3826                 /* Replace QUEUE/RSS action. */
3827                 split_actions[qrss_idx] = (struct rte_flow_action){
3828                         .type = (enum rte_flow_action_type)
3829                                 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3830                         .conf = set_tag,
3831                 };
3832         }
3833         /* JUMP action to jump to mreg copy table (CP_TBL). */
3834         jump = (void *)(set_tag + 1);
3835         *jump = (struct rte_flow_action_jump){
3836                 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3837         };
3838         split_actions[actions_n - 2] = (struct rte_flow_action){
3839                 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3840                 .conf = jump,
3841         };
3842         split_actions[actions_n - 1] = (struct rte_flow_action){
3843                 .type = RTE_FLOW_ACTION_TYPE_END,
3844         };
3845         return flow_id;
3846 }
3847
3848 /**
3849  * Extend the given action list for Tx metadata copy.
3850  *
3851  * Copy the given action list to the ext_actions and add flow metadata register
3852  * copy action in order to copy reg_a set by WQE to reg_c[0].
3853  *
3854  * @param[out] ext_actions
3855  *   Pointer to the extended action list.
3856  * @param[in] actions
3857  *   Pointer to the list of actions.
3858  * @param[in] actions_n
3859  *   Number of actions in the list.
3860  * @param[out] error
3861  *   Perform verbose error reporting if not NULL.
3862  * @param[in] encap_idx
3863  *   The encap action inndex.
3864  *
3865  * @return
3866  *   0 on success, negative value otherwise
3867  */
3868 static int
3869 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3870                        struct rte_flow_action *ext_actions,
3871                        const struct rte_flow_action *actions,
3872                        int actions_n, struct rte_flow_error *error,
3873                        int encap_idx)
3874 {
3875         struct mlx5_flow_action_copy_mreg *cp_mreg =
3876                 (struct mlx5_flow_action_copy_mreg *)
3877                         (ext_actions + actions_n + 1);
3878         int ret;
3879
3880         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3881         if (ret < 0)
3882                 return ret;
3883         cp_mreg->dst = ret;
3884         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3885         if (ret < 0)
3886                 return ret;
3887         cp_mreg->src = ret;
3888         if (encap_idx != 0)
3889                 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3890         if (encap_idx == actions_n - 1) {
3891                 ext_actions[actions_n - 1] = (struct rte_flow_action){
3892                         .type = (enum rte_flow_action_type)
3893                                 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3894                         .conf = cp_mreg,
3895                 };
3896                 ext_actions[actions_n] = (struct rte_flow_action){
3897                         .type = RTE_FLOW_ACTION_TYPE_END,
3898                 };
3899         } else {
3900                 ext_actions[encap_idx] = (struct rte_flow_action){
3901                         .type = (enum rte_flow_action_type)
3902                                 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3903                         .conf = cp_mreg,
3904                 };
3905                 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3906                                 sizeof(*ext_actions) * (actions_n - encap_idx));
3907         }
3908         return 0;
3909 }
3910
3911 /**
3912  * The splitting for metadata feature.
3913  *
3914  * - Q/RSS action on NIC Rx should be split in order to pass by
3915  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3916  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3917  *
3918  * - All the actions on NIC Tx should have a mreg copy action to
3919  *   copy reg_a from WQE to reg_c[0].
3920  *
3921  * @param dev
3922  *   Pointer to Ethernet device.
3923  * @param[in] flow
3924  *   Parent flow structure pointer.
3925  * @param[in] prefix_layers
3926  *   Prefix flow layer flags.
3927  * @param[in] attr
3928  *   Flow rule attributes.
3929  * @param[in] items
3930  *   Pattern specification (list terminated by the END pattern item).
3931  * @param[in] actions
3932  *   Associated actions (list terminated by the END action).
3933  * @param[in] external
3934  *   This flow rule is created by request external to PMD.
3935  * @param[in] flow_idx
3936  *   This memory pool index to the flow.
3937  * @param[out] error
3938  *   Perform verbose error reporting if not NULL.
3939  * @return
3940  *   0 on success, negative value otherwise
3941  */
3942 static int
3943 flow_create_split_metadata(struct rte_eth_dev *dev,
3944                            struct rte_flow *flow,
3945                            uint64_t prefix_layers,
3946                            const struct rte_flow_attr *attr,
3947                            const struct rte_flow_item items[],
3948                            const struct rte_flow_action actions[],
3949                            bool external, uint32_t flow_idx,
3950                            struct rte_flow_error *error)
3951 {
3952         struct mlx5_priv *priv = dev->data->dev_private;
3953         struct mlx5_dev_config *config = &priv->config;
3954         const struct rte_flow_action *qrss = NULL;
3955         struct rte_flow_action *ext_actions = NULL;
3956         struct mlx5_flow *dev_flow = NULL;
3957         uint32_t qrss_id = 0;
3958         int mtr_sfx = 0;
3959         size_t act_size;
3960         int actions_n;
3961         int encap_idx;
3962         int ret;
3963
3964         /* Check whether extensive metadata feature is engaged. */
3965         if (!config->dv_flow_en ||
3966             config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3967             !mlx5_flow_ext_mreg_supported(dev))
3968                 return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3969                                                attr, items, actions, external,
3970                                                flow_idx, error);
3971         actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3972                                                            &encap_idx);
3973         if (qrss) {
3974                 /* Exclude hairpin flows from splitting. */
3975                 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3976                         const struct rte_flow_action_queue *queue;
3977
3978                         queue = qrss->conf;
3979                         if (mlx5_rxq_get_type(dev, queue->index) ==
3980                             MLX5_RXQ_TYPE_HAIRPIN)
3981                                 qrss = NULL;
3982                 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3983                         const struct rte_flow_action_rss *rss;
3984
3985                         rss = qrss->conf;
3986                         if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3987                             MLX5_RXQ_TYPE_HAIRPIN)
3988                                 qrss = NULL;
3989                 }
3990         }
3991         if (qrss) {
3992                 /* Check if it is in meter suffix table. */
3993                 mtr_sfx = attr->group == (attr->transfer ?
3994                           (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3995                           MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3996                 /*
3997                  * Q/RSS action on NIC Rx should be split in order to pass by
3998                  * the mreg copy table (RX_CP_TBL) and then it jumps to the
3999                  * action table (RX_ACT_TBL) which has the split Q/RSS action.
4000                  */
4001                 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4002                            sizeof(struct rte_flow_action_set_tag) +
4003                            sizeof(struct rte_flow_action_jump);
4004                 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4005                                           SOCKET_ID_ANY);
4006                 if (!ext_actions)
4007                         return rte_flow_error_set(error, ENOMEM,
4008                                                   RTE_FLOW_ERROR_TYPE_ACTION,
4009                                                   NULL, "no memory to split "
4010                                                   "metadata flow");
4011                 /*
4012                  * If we are the suffix flow of meter, tag already exist.
4013                  * Set the tag action to void.
4014                  */
4015                 if (mtr_sfx)
4016                         ext_actions[qrss - actions].type =
4017                                                 RTE_FLOW_ACTION_TYPE_VOID;
4018                 else
4019                         ext_actions[qrss - actions].type =
4020                                                 (enum rte_flow_action_type)
4021                                                 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4022                 /*
4023                  * Create the new actions list with removed Q/RSS action
4024                  * and appended set tag and jump to register copy table
4025                  * (RX_CP_TBL). We should preallocate unique tag ID here
4026                  * in advance, because it is needed for set tag action.
4027                  */
4028                 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4029                                                     qrss, actions_n, error);
4030                 if (!mtr_sfx && !qrss_id) {
4031                         ret = -rte_errno;
4032                         goto exit;
4033                 }
4034         } else if (attr->egress && !attr->transfer) {
4035                 /*
4036                  * All the actions on NIC Tx should have a metadata register
4037                  * copy action to copy reg_a from WQE to reg_c[meta]
4038                  */
4039                 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4040                            sizeof(struct mlx5_flow_action_copy_mreg);
4041                 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4042                                           SOCKET_ID_ANY);
4043                 if (!ext_actions)
4044                         return rte_flow_error_set(error, ENOMEM,
4045                                                   RTE_FLOW_ERROR_TYPE_ACTION,
4046                                                   NULL, "no memory to split "
4047                                                   "metadata flow");
4048                 /* Create the action list appended with copy register. */
4049                 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4050                                              actions_n, error, encap_idx);
4051                 if (ret < 0)
4052                         goto exit;
4053         }
4054         /* Add the unmodified original or prefix subflow. */
4055         ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
4056                                       items, ext_actions ? ext_actions :
4057                                       actions, external, flow_idx, error);
4058         if (ret < 0)
4059                 goto exit;
4060         MLX5_ASSERT(dev_flow);
4061         if (qrss) {
4062                 const struct rte_flow_attr q_attr = {
4063                         .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4064                         .ingress = 1,
4065                 };
4066                 /* Internal PMD action to set register. */
4067                 struct mlx5_rte_flow_item_tag q_tag_spec = {
4068                         .data = qrss_id,
4069                         .id = REG_NON,
4070                 };
4071                 struct rte_flow_item q_items[] = {
4072                         {
4073                                 .type = (enum rte_flow_item_type)
4074                                         MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4075                                 .spec = &q_tag_spec,
4076                                 .last = NULL,
4077                                 .mask = NULL,
4078                         },
4079                         {
4080                                 .type = RTE_FLOW_ITEM_TYPE_END,
4081                         },
4082                 };
4083                 struct rte_flow_action q_actions[] = {
4084                         {
4085                                 .type = qrss->type,
4086                                 .conf = qrss->conf,
4087                         },
4088                         {
4089                                 .type = RTE_FLOW_ACTION_TYPE_END,
4090                         },
4091                 };
4092                 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4093
4094                 /*
4095                  * Configure the tag item only if there is no meter subflow.
4096                  * Since tag is already marked in the meter suffix subflow
4097                  * we can just use the meter suffix items as is.
4098                  */
4099                 if (qrss_id) {
4100                         /* Not meter subflow. */
4101                         MLX5_ASSERT(!mtr_sfx);
4102                         /*
4103                          * Put unique id in prefix flow due to it is destroyed
4104                          * after suffix flow and id will be freed after there
4105                          * is no actual flows with this id and identifier
4106                          * reallocation becomes possible (for example, for
4107                          * other flows in other threads).
4108                          */
4109                         dev_flow->handle->split_flow_id = qrss_id;
4110                         ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4111                                                    error);
4112                         if (ret < 0)
4113                                 goto exit;
4114                         q_tag_spec.id = ret;
4115                 }
4116                 dev_flow = NULL;
4117                 /* Add suffix subflow to execute Q/RSS. */
4118                 ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4119                                               &q_attr, mtr_sfx ? items :
4120                                               q_items, q_actions,
4121                                               external, flow_idx, error);
4122                 if (ret < 0)
4123                         goto exit;
4124                 /* qrss ID should be freed if failed. */
4125                 qrss_id = 0;
4126                 MLX5_ASSERT(dev_flow);
4127         }
4128
4129 exit:
4130         /*
4131          * We do not destroy the partially created sub_flows in case of error.
4132          * These ones are included into parent flow list and will be destroyed
4133          * by flow_drv_destroy.
4134          */
4135         flow_qrss_free_id(dev, qrss_id);
4136         mlx5_free(ext_actions);
4137         return ret;
4138 }
4139
4140 /**
4141  * The splitting for meter feature.
4142  *
4143  * - The meter flow will be split to two flows as prefix and
4144  *   suffix flow. The packets make sense only it pass the prefix
4145  *   meter action.
4146  *
4147  * - Reg_C_5 is used for the packet to match betweend prefix and
4148  *   suffix flow.
4149  *
4150  * @param dev
4151  *   Pointer to Ethernet device.
4152  * @param[in] flow
4153  *   Parent flow structure pointer.
4154  * @param[in] attr
4155  *   Flow rule attributes.
4156  * @param[in] items
4157  *   Pattern specification (list terminated by the END pattern item).
4158  * @param[in] actions
4159  *   Associated actions (list terminated by the END action).
4160  * @param[in] external
4161  *   This flow rule is created by request external to PMD.
4162  * @param[in] flow_idx
4163  *   This memory pool index to the flow.
4164  * @param[out] error
4165  *   Perform verbose error reporting if not NULL.
4166  * @return
4167  *   0 on success, negative value otherwise
4168  */
4169 static int
4170 flow_create_split_meter(struct rte_eth_dev *dev,
4171                            struct rte_flow *flow,
4172                            const struct rte_flow_attr *attr,
4173                            const struct rte_flow_item items[],
4174                            const struct rte_flow_action actions[],
4175                            bool external, uint32_t flow_idx,
4176                            struct rte_flow_error *error)
4177 {
4178         struct mlx5_priv *priv = dev->data->dev_private;
4179         struct rte_flow_action *sfx_actions = NULL;
4180         struct rte_flow_action *pre_actions = NULL;
4181         struct rte_flow_item *sfx_items = NULL;
4182         struct mlx5_flow *dev_flow = NULL;
4183         struct rte_flow_attr sfx_attr = *attr;
4184         uint32_t mtr = 0;
4185         uint32_t mtr_tag_id = 0;
4186         size_t act_size;
4187         size_t item_size;
4188         int actions_n = 0;
4189         int ret;
4190
4191         if (priv->mtr_en)
4192                 actions_n = flow_check_meter_action(actions, &mtr);
4193         if (mtr) {
4194                 /* The five prefix actions: meter, decap, encap, tag, end. */
4195                 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4196                            sizeof(struct mlx5_rte_flow_action_set_tag);
4197                 /* tag, vlan, port id, end. */
4198 #define METER_SUFFIX_ITEM 4
4199                 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4200                             sizeof(struct mlx5_rte_flow_item_tag) * 2;
4201                 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
4202                                           0, SOCKET_ID_ANY);
4203                 if (!sfx_actions)
4204                         return rte_flow_error_set(error, ENOMEM,
4205                                                   RTE_FLOW_ERROR_TYPE_ACTION,
4206                                                   NULL, "no memory to split "
4207                                                   "meter flow");
4208                 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4209                              act_size);
4210                 pre_actions = sfx_actions + actions_n;
4211                 mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4212                                                    actions, sfx_actions,
4213                                                    pre_actions);
4214                 if (!mtr_tag_id) {
4215                         ret = -rte_errno;
4216                         goto exit;
4217                 }
4218                 /* Add the prefix subflow. */
4219                 ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4220                                               items, pre_actions, external,
4221                                               flow_idx, error);
4222                 if (ret) {
4223                         ret = -rte_errno;
4224                         goto exit;
4225                 }
4226                 dev_flow->handle->split_flow_id = mtr_tag_id;
4227                 /* Setting the sfx group atrr. */
4228                 sfx_attr.group = sfx_attr.transfer ?
4229                                 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4230                                  MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4231         }
4232         /* Add the prefix subflow. */
4233         ret = flow_create_split_metadata(dev, flow, dev_flow ?
4234                                          flow_get_prefix_layer_flags(dev_flow) :
4235                                          0, &sfx_attr,
4236                                          sfx_items ? sfx_items : items,
4237                                          sfx_actions ? sfx_actions : actions,
4238                                          external, flow_idx, error);
4239 exit:
4240         if (sfx_actions)
4241                 mlx5_free(sfx_actions);
4242         return ret;
4243 }
4244
4245 /**
4246  * Split the flow to subflow set. The splitters might be linked
4247  * in the chain, like this:
4248  * flow_create_split_outer() calls:
4249  *   flow_create_split_meter() calls:
4250  *     flow_create_split_metadata(meter_subflow_0) calls:
4251  *       flow_create_split_inner(metadata_subflow_0)
4252  *       flow_create_split_inner(metadata_subflow_1)
4253  *       flow_create_split_inner(metadata_subflow_2)
4254  *     flow_create_split_metadata(meter_subflow_1) calls:
4255  *       flow_create_split_inner(metadata_subflow_0)
4256  *       flow_create_split_inner(metadata_subflow_1)
4257  *       flow_create_split_inner(metadata_subflow_2)
4258  *
4259  * This provide flexible way to add new levels of flow splitting.
4260  * The all of successfully created subflows are included to the
4261  * parent flow dev_flow list.
4262  *
4263  * @param dev
4264  *   Pointer to Ethernet device.
4265  * @param[in] flow
4266  *   Parent flow structure pointer.
4267  * @param[in] attr
4268  *   Flow rule attributes.
4269  * @param[in] items
4270  *   Pattern specification (list terminated by the END pattern item).
4271  * @param[in] actions
4272  *   Associated actions (list terminated by the END action).
4273  * @param[in] external
4274  *   This flow rule is created by request external to PMD.
4275  * @param[in] flow_idx
4276  *   This memory pool index to the flow.
4277  * @param[out] error
4278  *   Perform verbose error reporting if not NULL.
4279  * @return
4280  *   0 on success, negative value otherwise
4281  */
4282 static int
4283 flow_create_split_outer(struct rte_eth_dev *dev,
4284                         struct rte_flow *flow,
4285                         const struct rte_flow_attr *attr,
4286                         const struct rte_flow_item items[],
4287                         const struct rte_flow_action actions[],
4288                         bool external, uint32_t flow_idx,
4289                         struct rte_flow_error *error)
4290 {
4291         int ret;
4292
4293         ret = flow_create_split_meter(dev, flow, attr, items,
4294                                          actions, external, flow_idx, error);
4295         MLX5_ASSERT(ret <= 0);
4296         return ret;
4297 }
4298
4299 /**
4300  * Create a flow and add it to @p list.
4301  *
4302  * @param dev
4303  *   Pointer to Ethernet device.
4304  * @param list
4305  *   Pointer to a TAILQ flow list. If this parameter NULL,
4306  *   no list insertion occurred, flow is just created,
4307  *   this is caller's responsibility to track the
4308  *   created flow.
4309  * @param[in] attr
4310  *   Flow rule attributes.
4311  * @param[in] items
4312  *   Pattern specification (list terminated by the END pattern item).
4313  * @param[in] actions
4314  *   Associated actions (list terminated by the END action).
4315  * @param[in] external
4316  *   This flow rule is created by request external to PMD.
4317  * @param[out] error
4318  *   Perform verbose error reporting if not NULL.
4319  *
4320  * @return
4321  *   A flow index on success, 0 otherwise and rte_errno is set.
4322  */
4323 static uint32_t
4324 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4325                  const struct rte_flow_attr *attr,
4326                  const struct rte_flow_item items[],
4327                  const struct rte_flow_action actions[],
4328                  bool external, struct rte_flow_error *error)
4329 {
4330         struct mlx5_priv *priv = dev->data->dev_private;
4331         struct rte_flow *flow = NULL;
4332         struct mlx5_flow *dev_flow;
4333         const struct rte_flow_action_rss *rss;
4334         union {
4335                 struct rte_flow_expand_rss buf;
4336                 uint8_t buffer[2048];
4337         } expand_buffer;
4338         union {
4339                 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4340                 uint8_t buffer[2048];
4341         } actions_rx;
4342         union {
4343                 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4344                 uint8_t buffer[2048];
4345         } actions_hairpin_tx;
4346         union {
4347                 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4348                 uint8_t buffer[2048];
4349         } items_tx;
4350         struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4351         struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4352                                               priv->rss_desc)[!!priv->flow_idx];
4353         const struct rte_flow_action *p_actions_rx = actions;
4354         uint32_t i;
4355         uint32_t idx = 0;
4356         int hairpin_flow;
4357         uint32_t hairpin_id = 0;
4358         struct rte_flow_attr attr_tx = { .priority = 0 };
4359         int ret;
4360
4361         hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4362         ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4363                                 external, hairpin_flow, error);
4364         if (ret < 0)
4365                 return 0;
4366         if (hairpin_flow > 0) {
4367                 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4368                         rte_errno = EINVAL;
4369                         return 0;
4370                 }
4371                 flow_hairpin_split(dev, actions, actions_rx.actions,
4372                                    actions_hairpin_tx.actions, items_tx.items,
4373                                    &hairpin_id);
4374                 p_actions_rx = actions_rx.actions;
4375         }
4376         flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4377         if (!flow) {
4378                 rte_errno = ENOMEM;
4379                 goto error_before_flow;
4380         }
4381         flow->drv_type = flow_get_drv_type(dev, attr);
4382         if (hairpin_id != 0)
4383                 flow->hairpin_flow_id = hairpin_id;
4384         MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4385                     flow->drv_type < MLX5_FLOW_TYPE_MAX);
4386         memset(rss_desc, 0, sizeof(*rss_desc));
4387         rss = flow_get_rss_action(p_actions_rx);
4388         if (rss) {
4389                 /*
4390                  * The following information is required by
4391                  * mlx5_flow_hashfields_adjust() in advance.
4392                  */
4393                 rss_desc->level = rss->level;
4394                 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4395                 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4396         }
4397         flow->dev_handles = 0;
4398         if (rss && rss->types) {
4399                 unsigned int graph_root;
4400
4401                 graph_root = find_graph_root(items, rss->level);
4402                 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4403                                           items, rss->types,
4404                                           mlx5_support_expansion,
4405                                           graph_root);
4406                 MLX5_ASSERT(ret > 0 &&
4407                        (unsigned int)ret < sizeof(expand_buffer.buffer));
4408         } else {
4409                 buf->entries = 1;
4410                 buf->entry[0].pattern = (void *)(uintptr_t)items;
4411         }
4412         /*
4413          * Record the start index when there is a nested call. All sub-flows
4414          * need to be translated before another calling.
4415          * No need to use ping-pong buffer to save memory here.
4416          */
4417         if (priv->flow_idx) {
4418                 MLX5_ASSERT(!priv->flow_nested_idx);
4419                 priv->flow_nested_idx = priv->flow_idx;
4420         }
4421         for (i = 0; i < buf->entries; ++i) {
4422                 /*
4423                  * The splitter may create multiple dev_flows,
4424                  * depending on configuration. In the simplest
4425                  * case it just creates unmodified original flow.
4426                  */
4427                 ret = flow_create_split_outer(dev, flow, attr,
4428                                               buf->entry[i].pattern,
4429                                               p_actions_rx, external, idx,
4430                                               error);
4431                 if (ret < 0)
4432                         goto error;
4433         }
4434         /* Create the tx flow. */
4435         if (hairpin_flow) {
4436                 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4437                 attr_tx.ingress = 0;
4438                 attr_tx.egress = 1;
4439                 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4440                                          actions_hairpin_tx.actions,
4441                                          idx, error);
4442                 if (!dev_flow)
4443                         goto error;
4444                 dev_flow->flow = flow;
4445                 dev_flow->external = 0;
4446                 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4447                               dev_flow->handle, next);
4448                 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4449                                          items_tx.items,
4450                                          actions_hairpin_tx.actions, error);
4451                 if (ret < 0)
4452                         goto error;
4453         }
4454         /*
4455          * Update the metadata register copy table. If extensive
4456          * metadata feature is enabled and registers are supported
4457          * we might create the extra rte_flow for each unique
4458          * MARK/FLAG action ID.
4459          *
4460          * The table is updated for ingress Flows only, because
4461          * the egress Flows belong to the different device and
4462          * copy table should be updated in peer NIC Rx domain.
4463          */
4464         if (attr->ingress &&
4465             (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4466                 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4467                 if (ret)
4468                         goto error;
4469         }
4470         /*
4471          * If the flow is external (from application) OR device is started, then
4472          * the flow will be applied immediately.
4473          */
4474         if (external || dev->data->dev_started) {
4475                 ret = flow_drv_apply(dev, flow, error);
4476                 if (ret < 0)
4477                         goto error;
4478         }
4479         if (list)
4480                 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4481                              flow, next);
4482         flow_rxq_flags_set(dev, flow);
4483         /* Nested flow creation index recovery. */
4484         priv->flow_idx = priv->flow_nested_idx;
4485         if (priv->flow_nested_idx)
4486                 priv->flow_nested_idx = 0;
4487         return idx;
4488 error:
4489         MLX5_ASSERT(flow);
4490         ret = rte_errno; /* Save rte_errno before cleanup. */
4491         flow_mreg_del_copy_action(dev, flow);
4492         flow_drv_destroy(dev, flow);
4493         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4494         rte_errno = ret; /* Restore rte_errno. */
4495 error_before_flow:
4496         ret = rte_errno;
4497         if (hairpin_id)
4498                 mlx5_flow_id_release(priv->sh->flow_id_pool,
4499                                      hairpin_id);
4500         rte_errno = ret;
4501         priv->flow_idx = priv->flow_nested_idx;
4502         if (priv->flow_nested_idx)
4503                 priv->flow_nested_idx = 0;
4504         return 0;
4505 }
4506
4507 /**
4508  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4509  * incoming packets to table 1.
4510  *
4511  * Other flow rules, requested for group n, will be created in
4512  * e-switch table n+1.
4513  * Jump action to e-switch group n will be created to group n+1.
4514  *
4515  * Used when working in switchdev mode, to utilise advantages of table 1
4516  * and above.
4517  *
4518  * @param dev
4519  *   Pointer to Ethernet device.
4520  *
4521  * @return
4522  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4523  */
4524 struct rte_flow *
4525 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4526 {
4527         const struct rte_flow_attr attr = {
4528                 .group = 0,
4529                 .priority = 0,
4530                 .ingress = 1,
4531                 .egress = 0,
4532                 .transfer = 1,
4533         };
4534         const struct rte_flow_item pattern = {
4535                 .type = RTE_FLOW_ITEM_TYPE_END,
4536         };
4537         struct rte_flow_action_jump jump = {
4538                 .group = 1,
4539         };
4540         const struct rte_flow_action actions[] = {
4541                 {
4542                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
4543                         .conf = &jump,
4544                 },
4545                 {
4546                         .type = RTE_FLOW_ACTION_TYPE_END,
4547                 },
4548         };
4549         struct mlx5_priv *priv = dev->data->dev_private;
4550         struct rte_flow_error error;
4551
4552         return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4553                                                    &attr, &pattern,
4554                                                    actions, false, &error);
4555 }
4556
4557 /**
4558  * Validate a flow supported by the NIC.
4559  *
4560  * @see rte_flow_validate()
4561  * @see rte_flow_ops
4562  */
4563 int
4564 mlx5_flow_validate(struct rte_eth_dev *dev,
4565                    const struct rte_flow_attr *attr,
4566                    const struct rte_flow_item items[],
4567                    const struct rte_flow_action actions[],
4568                    struct rte_flow_error *error)
4569 {
4570         int hairpin_flow;
4571
4572         hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4573         return flow_drv_validate(dev, attr, items, actions,
4574                                 true, hairpin_flow, error);
4575 }
4576
4577 /**
4578  * Create a flow.
4579  *
4580  * @see rte_flow_create()
4581  * @see rte_flow_ops
4582  */
4583 struct rte_flow *
4584 mlx5_flow_create(struct rte_eth_dev *dev,
4585                  const struct rte_flow_attr *attr,
4586                  const struct rte_flow_item items[],
4587                  const struct rte_flow_action actions[],
4588                  struct rte_flow_error *error)
4589 {
4590         struct mlx5_priv *priv = dev->data->dev_private;
4591
4592         /*
4593          * If the device is not started yet, it is not allowed to created a
4594          * flow from application. PMD default flows and traffic control flows
4595          * are not affected.
4596          */
4597         if (unlikely(!dev->data->dev_started)) {
4598                 DRV_LOG(DEBUG, "port %u is not started when "
4599                         "inserting a flow", dev->data->port_id);
4600                 rte_flow_error_set(error, ENODEV,
4601                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4602                                    NULL,
4603                                    "port not started");
4604                 return NULL;
4605         }
4606         return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4607                                   attr, items, actions, true, error);
4608 }
4609
4610 /**
4611  * Destroy a flow in a list.
4612  *
4613  * @param dev
4614  *   Pointer to Ethernet device.
4615  * @param list
4616  *   Pointer to the Indexed flow list. If this parameter NULL,
4617  *   there is no flow removal from the list. Be noted that as
4618  *   flow is add to the indexed list, memory of the indexed
4619  *   list points to maybe changed as flow destroyed.
4620  * @param[in] flow_idx
4621  *   Index of flow to destroy.
4622  */
4623 static void
4624 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4625                   uint32_t flow_idx)
4626 {
4627         struct mlx5_priv *priv = dev->data->dev_private;
4628         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4629         struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4630                                                [MLX5_IPOOL_RTE_FLOW], flow_idx);
4631
4632         if (!flow)
4633                 return;
4634         /*
4635          * Update RX queue flags only if port is started, otherwise it is
4636          * already clean.
4637          */
4638         if (dev->data->dev_started)
4639                 flow_rxq_flags_trim(dev, flow);
4640         if (flow->hairpin_flow_id)
4641                 mlx5_flow_id_release(priv->sh->flow_id_pool,
4642                                      flow->hairpin_flow_id);
4643         flow_drv_destroy(dev, flow);
4644         if (list)
4645                 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4646                              flow_idx, flow, next);
4647         flow_mreg_del_copy_action(dev, flow);
4648         if (flow->fdir) {
4649                 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4650                         if (priv_fdir_flow->rix_flow == flow_idx)
4651                                 break;
4652                 }
4653                 if (priv_fdir_flow) {
4654                         LIST_REMOVE(priv_fdir_flow, next);
4655                         mlx5_free(priv_fdir_flow->fdir);
4656                         mlx5_free(priv_fdir_flow);
4657                 }
4658         }
4659         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4660 }
4661
4662 /**
4663  * Destroy all flows.
4664  *
4665  * @param dev
4666  *   Pointer to Ethernet device.
4667  * @param list
4668  *   Pointer to the Indexed flow list.
4669  * @param active
4670  *   If flushing is called avtively.
4671  */
4672 void
4673 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4674 {
4675         uint32_t num_flushed = 0;
4676
4677         while (*list) {
4678                 flow_list_destroy(dev, list, *list);
4679                 num_flushed++;
4680         }
4681         if (active) {
4682                 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4683                         dev->data->port_id, num_flushed);
4684         }
4685 }
4686
4687 /**
4688  * Remove all flows.
4689  *
4690  * @param dev
4691  *   Pointer to Ethernet device.
4692  * @param list
4693  *   Pointer to the Indexed flow list.
4694  */
4695 void
4696 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4697 {
4698         struct mlx5_priv *priv = dev->data->dev_private;
4699         struct rte_flow *flow = NULL;
4700         uint32_t idx;
4701
4702         ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4703                       flow, next) {
4704                 flow_drv_remove(dev, flow);
4705                 flow_mreg_stop_copy_action(dev, flow);
4706         }
4707         flow_mreg_del_default_copy_action(dev);
4708         flow_rxq_flags_clear(dev);
4709 }
4710
4711 /**
4712  * Add all flows.
4713  *
4714  * @param dev
4715  *   Pointer to Ethernet device.
4716  * @param list
4717  *   Pointer to the Indexed flow list.
4718  *
4719  * @return
4720  *   0 on success, a negative errno value otherwise and rte_errno is set.
4721  */
4722 int
4723 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4724 {
4725         struct mlx5_priv *priv = dev->data->dev_private;
4726         struct rte_flow *flow = NULL;
4727         struct rte_flow_error error;
4728         uint32_t idx;
4729         int ret = 0;
4730
4731         /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4732         ret = flow_mreg_add_default_copy_action(dev, &error);
4733         if (ret < 0)
4734                 return -rte_errno;
4735         /* Apply Flows created by application. */
4736         ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4737                       flow, next) {
4738                 ret = flow_mreg_start_copy_action(dev, flow);
4739                 if (ret < 0)
4740                         goto error;
4741                 ret = flow_drv_apply(dev, flow, &error);
4742                 if (ret < 0)
4743                         goto error;
4744                 flow_rxq_flags_set(dev, flow);
4745         }
4746         return 0;
4747 error:
4748         ret = rte_errno; /* Save rte_errno before cleanup. */
4749         mlx5_flow_stop(dev, list);
4750         rte_errno = ret; /* Restore rte_errno. */
4751         return -rte_errno;
4752 }
4753
4754 /**
4755  * Stop all default actions for flows.
4756  *
4757  * @param dev
4758  *   Pointer to Ethernet device.
4759  */
4760 void
4761 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4762 {
4763         flow_mreg_del_default_copy_action(dev);
4764         flow_rxq_flags_clear(dev);
4765 }
4766
4767 /**
4768  * Start all default actions for flows.
4769  *
4770  * @param dev
4771  *   Pointer to Ethernet device.
4772  * @return
4773  *   0 on success, a negative errno value otherwise and rte_errno is set.
4774  */
4775 int
4776 mlx5_flow_start_default(struct rte_eth_dev *dev)
4777 {
4778         struct rte_flow_error error;
4779
4780         /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4781         return flow_mreg_add_default_copy_action(dev, &error);
4782 }
4783
4784 /**
4785  * Allocate intermediate resources for flow creation.
4786  *
4787  * @param dev
4788  *   Pointer to Ethernet device.
4789  */
4790 void
4791 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4792 {
4793         struct mlx5_priv *priv = dev->data->dev_private;
4794
4795         if (!priv->inter_flows) {
4796                 priv->inter_flows = mlx5_malloc(MLX5_MEM_ZERO,
4797                                     MLX5_NUM_MAX_DEV_FLOWS *
4798                                     sizeof(struct mlx5_flow) +
4799                                     (sizeof(struct mlx5_flow_rss_desc) +
4800                                     sizeof(uint16_t) * UINT16_MAX) * 2, 0,
4801                                     SOCKET_ID_ANY);
4802                 if (!priv->inter_flows) {
4803                         DRV_LOG(ERR, "can't allocate intermediate memory.");
4804                         return;
4805                 }
4806         }
4807         priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4808                          [MLX5_NUM_MAX_DEV_FLOWS];
4809         /* Reset the index. */
4810         priv->flow_idx = 0;
4811         priv->flow_nested_idx = 0;
4812 }
4813
4814 /**
4815  * Free intermediate resources for flows.
4816  *
4817  * @param dev
4818  *   Pointer to Ethernet device.
4819  */
4820 void
4821 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4822 {
4823         struct mlx5_priv *priv = dev->data->dev_private;
4824
4825         mlx5_free(priv->inter_flows);
4826         priv->inter_flows = NULL;
4827 }
4828
4829 /**
4830  * Verify the flow list is empty
4831  *
4832  * @param dev
4833  *  Pointer to Ethernet device.
4834  *
4835  * @return the number of flows not released.
4836  */
4837 int
4838 mlx5_flow_verify(struct rte_eth_dev *dev)
4839 {
4840         struct mlx5_priv *priv = dev->data->dev_private;
4841         struct rte_flow *flow;
4842         uint32_t idx;
4843         int ret = 0;
4844
4845         ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4846                       flow, next) {
4847                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
4848                         dev->data->port_id, (void *)flow);
4849                 ++ret;
4850         }
4851         return ret;
4852 }
4853
4854 /**
4855  * Enable default hairpin egress flow.
4856  *
4857  * @param dev
4858  *   Pointer to Ethernet device.
4859  * @param queue
4860  *   The queue index.
4861  *
4862  * @return
4863  *   0 on success, a negative errno value otherwise and rte_errno is set.
4864  */
4865 int
4866 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4867                             uint32_t queue)
4868 {
4869         struct mlx5_priv *priv = dev->data->dev_private;
4870         const struct rte_flow_attr attr = {
4871                 .egress = 1,
4872                 .priority = 0,
4873         };
4874         struct mlx5_rte_flow_item_tx_queue queue_spec = {
4875                 .queue = queue,
4876         };
4877         struct mlx5_rte_flow_item_tx_queue queue_mask = {
4878                 .queue = UINT32_MAX,
4879         };
4880         struct rte_flow_item items[] = {
4881                 {
4882                         .type = (enum rte_flow_item_type)
4883                                 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4884                         .spec = &queue_spec,
4885                         .last = NULL,
4886                         .mask = &queue_mask,
4887                 },
4888                 {
4889                         .type = RTE_FLOW_ITEM_TYPE_END,
4890                 },
4891         };
4892         struct rte_flow_action_jump jump = {
4893                 .group = MLX5_HAIRPIN_TX_TABLE,
4894         };
4895         struct rte_flow_action actions[2];
4896         uint32_t flow_idx;
4897         struct rte_flow_error error;
4898
4899         actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4900         actions[0].conf = &jump;
4901         actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4902         flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4903                                 &attr, items, actions, false, &error);
4904         if (!flow_idx) {
4905                 DRV_LOG(DEBUG,
4906                         "Failed to create ctrl flow: rte_errno(%d),"
4907                         " type(%d), message(%s)",
4908                         rte_errno, error.type,
4909                         error.message ? error.message : " (no stated reason)");
4910                 return -rte_errno;
4911         }
4912         return 0;
4913 }
4914
4915 /**
4916  * Enable a control flow configured from the control plane.
4917  *
4918  * @param dev
4919  *   Pointer to Ethernet device.
4920  * @param eth_spec
4921  *   An Ethernet flow spec to apply.
4922  * @param eth_mask
4923  *   An Ethernet flow mask to apply.
4924  * @param vlan_spec
4925  *   A VLAN flow spec to apply.
4926  * @param vlan_mask
4927  *   A VLAN flow mask to apply.
4928  *
4929  * @return
4930  *   0 on success, a negative errno value otherwise and rte_errno is set.
4931  */
4932 int
4933 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4934                     struct rte_flow_item_eth *eth_spec,
4935                     struct rte_flow_item_eth *eth_mask,
4936                     struct rte_flow_item_vlan *vlan_spec,
4937                     struct rte_flow_item_vlan *vlan_mask)
4938 {
4939         struct mlx5_priv *priv = dev->data->dev_private;
4940         const struct rte_flow_attr attr = {
4941                 .ingress = 1,
4942                 .priority = MLX5_FLOW_PRIO_RSVD,
4943         };
4944         struct rte_flow_item items[] = {
4945                 {
4946                         .type = RTE_FLOW_ITEM_TYPE_ETH,
4947                         .spec = eth_spec,
4948                         .last = NULL,
4949                         .mask = eth_mask,
4950                 },
4951                 {
4952                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4953                                               RTE_FLOW_ITEM_TYPE_END,
4954                         .spec = vlan_spec,
4955                         .last = NULL,
4956                         .mask = vlan_mask,
4957                 },
4958                 {
4959                         .type = RTE_FLOW_ITEM_TYPE_END,
4960                 },
4961         };
4962         uint16_t queue[priv->reta_idx_n];
4963         struct rte_flow_action_rss action_rss = {
4964                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4965                 .level = 0,
4966                 .types = priv->rss_conf.rss_hf,
4967                 .key_len = priv->rss_conf.rss_key_len,
4968                 .queue_num = priv->reta_idx_n,
4969                 .key = priv->rss_conf.rss_key,
4970                 .queue = queue,
4971         };
4972         struct rte_flow_action actions[] = {
4973                 {
4974                         .type = RTE_FLOW_ACTION_TYPE_RSS,
4975                         .conf = &action_rss,
4976                 },
4977                 {
4978                         .type = RTE_FLOW_ACTION_TYPE_END,
4979                 },
4980         };
4981         uint32_t flow_idx;
4982         struct rte_flow_error error;
4983         unsigned int i;
4984
4985         if (!priv->reta_idx_n || !priv->rxqs_n) {
4986                 return 0;
4987         }
4988         if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4989                 action_rss.types = 0;
4990         for (i = 0; i != priv->reta_idx_n; ++i)
4991                 queue[i] = (*priv->reta_idx)[i];
4992         flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4993                                 &attr, items, actions, false, &error);
4994         if (!flow_idx)
4995                 return -rte_errno;
4996         return 0;
4997 }
4998
4999 /**
5000  * Enable a flow control configured from the control plane.
5001  *
5002  * @param dev
5003  *   Pointer to Ethernet device.
5004  * @param eth_spec
5005  *   An Ethernet flow spec to apply.
5006  * @param eth_mask
5007  *   An Ethernet flow mask to apply.
5008  *
5009  * @return
5010  *   0 on success, a negative errno value otherwise and rte_errno is set.
5011  */
5012 int
5013 mlx5_ctrl_flow(struct rte_eth_dev *dev,
5014                struct rte_flow_item_eth *eth_spec,
5015                struct rte_flow_item_eth *eth_mask)
5016 {
5017         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
5018 }
5019
5020 /**
5021  * Create default miss flow rule matching lacp traffic
5022  *
5023  * @param dev
5024  *   Pointer to Ethernet device.
5025  * @param eth_spec
5026  *   An Ethernet flow spec to apply.
5027  *
5028  * @return
5029  *   0 on success, a negative errno value otherwise and rte_errno is set.
5030  */
5031 int
5032 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
5033 {
5034         struct mlx5_priv *priv = dev->data->dev_private;
5035         /*
5036          * The LACP matching is done by only using ether type since using
5037          * a multicast dst mac causes kernel to give low priority to this flow.
5038          */
5039         static const struct rte_flow_item_eth lacp_spec = {
5040                 .type = RTE_BE16(0x8809),
5041         };
5042         static const struct rte_flow_item_eth lacp_mask = {
5043                 .type = 0xffff,
5044         };
5045         const struct rte_flow_attr attr = {
5046                 .ingress = 1,
5047         };
5048         struct rte_flow_item items[] = {
5049                 {
5050                         .type = RTE_FLOW_ITEM_TYPE_ETH,
5051                         .spec = &lacp_spec,
5052                         .mask = &lacp_mask,
5053                 },
5054                 {
5055                         .type = RTE_FLOW_ITEM_TYPE_END,
5056                 },
5057         };
5058         struct rte_flow_action actions[] = {
5059                 {
5060                         .type = (enum rte_flow_action_type)
5061                                 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
5062                 },
5063                 {
5064                         .type = RTE_FLOW_ACTION_TYPE_END,
5065                 },
5066         };
5067         struct rte_flow_error error;
5068         uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5069                                 &attr, items, actions, false, &error);
5070
5071         if (!flow_idx)
5072                 return -rte_errno;
5073         return 0;
5074 }
5075
5076 /**
5077  * Destroy a flow.
5078  *
5079  * @see rte_flow_destroy()
5080  * @see rte_flow_ops
5081  */
5082 int
5083 mlx5_flow_destroy(struct rte_eth_dev *dev,
5084                   struct rte_flow *flow,
5085                   struct rte_flow_error *error __rte_unused)
5086 {
5087         struct mlx5_priv *priv = dev->data->dev_private;
5088
5089         flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
5090         return 0;
5091 }
5092
5093 /**
5094  * Destroy all flows.
5095  *
5096  * @see rte_flow_flush()
5097  * @see rte_flow_ops
5098  */
5099 int
5100 mlx5_flow_flush(struct rte_eth_dev *dev,
5101                 struct rte_flow_error *error __rte_unused)
5102 {
5103         struct mlx5_priv *priv = dev->data->dev_private;
5104
5105         mlx5_flow_list_flush(dev, &priv->flows, false);
5106         return 0;
5107 }
5108
5109 /**
5110  * Isolated mode.
5111  *
5112  * @see rte_flow_isolate()
5113  * @see rte_flow_ops
5114  */
5115 int
5116 mlx5_flow_isolate(struct rte_eth_dev *dev,
5117                   int enable,
5118                   struct rte_flow_error *error)
5119 {
5120         struct mlx5_priv *priv = dev->data->dev_private;
5121
5122         if (dev->data->dev_started) {
5123                 rte_flow_error_set(error, EBUSY,
5124                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5125                                    NULL,
5126                                    "port must be stopped first");
5127                 return -rte_errno;
5128         }
5129         priv->isolated = !!enable;
5130         if (enable)
5131                 dev->dev_ops = &mlx5_os_dev_ops_isolate;
5132         else
5133                 dev->dev_ops = &mlx5_os_dev_ops;
5134         return 0;
5135 }
5136
5137 /**
5138  * Query a flow.
5139  *
5140  * @see rte_flow_query()
5141  * @see rte_flow_ops
5142  */
5143 static int
5144 flow_drv_query(struct rte_eth_dev *dev,
5145                uint32_t flow_idx,
5146                const struct rte_flow_action *actions,
5147                void *data,
5148                struct rte_flow_error *error)
5149 {
5150         struct mlx5_priv *priv = dev->data->dev_private;
5151         const struct mlx5_flow_driver_ops *fops;
5152         struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5153                                                [MLX5_IPOOL_RTE_FLOW],
5154                                                flow_idx);
5155         enum mlx5_flow_drv_type ftype;
5156
5157         if (!flow) {
5158                 return rte_flow_error_set(error, ENOENT,
5159                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5160                           NULL,
5161                           "invalid flow handle");
5162         }
5163         ftype = flow->drv_type;
5164         MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5165         fops = flow_get_drv_ops(ftype);
5166
5167         return fops->query(dev, flow, actions, data, error);
5168 }
5169
5170 /**
5171  * Query a flow.
5172  *
5173  * @see rte_flow_query()
5174  * @see rte_flow_ops
5175  */
5176 int
5177 mlx5_flow_query(struct rte_eth_dev *dev,
5178                 struct rte_flow *flow,
5179                 const struct rte_flow_action *actions,
5180                 void *data,
5181                 struct rte_flow_error *error)
5182 {
5183         int ret;
5184
5185         ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5186                              error);
5187         if (ret < 0)
5188                 return ret;
5189         return 0;
5190 }
5191
5192 /**
5193  * Convert a flow director filter to a generic flow.
5194  *
5195  * @param dev
5196  *   Pointer to Ethernet device.
5197  * @param fdir_filter
5198  *   Flow director filter to add.
5199  * @param attributes
5200  *   Generic flow parameters structure.
5201  *
5202  * @return
5203  *   0 on success, a negative errno value otherwise and rte_errno is set.
5204  */
5205 static int
5206 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5207                          const struct rte_eth_fdir_filter *fdir_filter,
5208                          struct mlx5_fdir *attributes)
5209 {
5210         struct mlx5_priv *priv = dev->data->dev_private;
5211         const struct rte_eth_fdir_input *input = &fdir_filter->input;
5212         const struct rte_eth_fdir_masks *mask =
5213                 &dev->data->dev_conf.fdir_conf.mask;
5214
5215         /* Validate queue number. */
5216         if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5217                 DRV_LOG(ERR, "port %u invalid queue number %d",
5218                         dev->data->port_id, fdir_filter->action.rx_queue);
5219                 rte_errno = EINVAL;
5220                 return -rte_errno;
5221         }
5222         attributes->attr.ingress = 1;
5223         attributes->items[0] = (struct rte_flow_item) {
5224                 .type = RTE_FLOW_ITEM_TYPE_ETH,
5225                 .spec = &attributes->l2,
5226                 .mask = &attributes->l2_mask,
5227         };
5228         switch (fdir_filter->action.behavior) {
5229         case RTE_ETH_FDIR_ACCEPT:
5230                 attributes->actions[0] = (struct rte_flow_action){
5231                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
5232                         .conf = &attributes->queue,
5233                 };
5234                 break;
5235         case RTE_ETH_FDIR_REJECT:
5236                 attributes->actions[0] = (struct rte_flow_action){
5237                         .type = RTE_FLOW_ACTION_TYPE_DROP,
5238                 };
5239                 break;
5240         default:
5241                 DRV_LOG(ERR, "port %u invalid behavior %d",
5242                         dev->data->port_id,
5243                         fdir_filter->action.behavior);
5244                 rte_errno = ENOTSUP;
5245                 return -rte_errno;
5246         }
5247         attributes->queue.index = fdir_filter->action.rx_queue;
5248         /* Handle L3. */
5249         switch (fdir_filter->input.flow_type) {
5250         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5251         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5252         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5253                 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5254                         .src_addr = input->flow.ip4_flow.src_ip,
5255                         .dst_addr = input->flow.ip4_flow.dst_ip,
5256                         .time_to_live = input->flow.ip4_flow.ttl,
5257                         .type_of_service = input->flow.ip4_flow.tos,
5258                 };
5259                 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5260                         .src_addr = mask->ipv4_mask.src_ip,
5261                         .dst_addr = mask->ipv4_mask.dst_ip,
5262                         .time_to_live = mask->ipv4_mask.ttl,
5263                         .type_of_service = mask->ipv4_mask.tos,
5264                         .next_proto_id = mask->ipv4_mask.proto,
5265                 };
5266                 attributes->items[1] = (struct rte_flow_item){
5267                         .type = RTE_FLOW_ITEM_TYPE_IPV4,
5268                         .spec = &attributes->l3,
5269                         .mask = &attributes->l3_mask,
5270                 };
5271                 break;
5272         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5273         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5274         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5275                 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5276                         .hop_limits = input->flow.ipv6_flow.hop_limits,
5277                         .proto = input->flow.ipv6_flow.proto,
5278                 };
5279
5280                 memcpy(attributes->l3.ipv6.hdr.src_addr,
5281                        input->flow.ipv6_flow.src_ip,
5282                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5283                 memcpy(attributes->l3.ipv6.hdr.dst_addr,
5284                        input->flow.ipv6_flow.dst_ip,
5285                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5286                 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5287                        mask->ipv6_mask.src_ip,
5288                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5289                 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5290                        mask->ipv6_mask.dst_ip,
5291                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5292                 attributes->items[1] = (struct rte_flow_item){
5293                         .type = RTE_FLOW_ITEM_TYPE_IPV6,
5294                         .spec = &attributes->l3,
5295                         .mask = &attributes->l3_mask,
5296                 };
5297                 break;
5298         default:
5299                 DRV_LOG(ERR, "port %u invalid flow type%d",
5300                         dev->data->port_id, fdir_filter->input.flow_type);
5301                 rte_errno = ENOTSUP;
5302                 return -rte_errno;
5303         }
5304         /* Handle L4. */
5305         switch (fdir_filter->input.flow_type) {
5306         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5307                 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5308                         .src_port = input->flow.udp4_flow.src_port,
5309                         .dst_port = input->flow.udp4_flow.dst_port,
5310                 };
5311                 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5312                         .src_port = mask->src_port_mask,
5313                         .dst_port = mask->dst_port_mask,
5314                 };
5315                 attributes->items[2] = (struct rte_flow_item){
5316                         .type = RTE_FLOW_ITEM_TYPE_UDP,
5317                         .spec = &attributes->l4,
5318                         .mask = &attributes->l4_mask,
5319                 };
5320                 break;
5321         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5322                 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5323                         .src_port = input->flow.tcp4_flow.src_port,
5324                         .dst_port = input->flow.tcp4_flow.dst_port,
5325                 };
5326                 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5327                         .src_port = mask->src_port_mask,
5328                         .dst_port = mask->dst_port_mask,
5329                 };
5330                 attributes->items[2] = (struct rte_flow_item){
5331                         .type = RTE_FLOW_ITEM_TYPE_TCP,
5332                         .spec = &attributes->l4,
5333                         .mask = &attributes->l4_mask,
5334                 };
5335                 break;
5336         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5337                 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5338                         .src_port = input->flow.udp6_flow.src_port,
5339                         .dst_port = input->flow.udp6_flow.dst_port,
5340                 };
5341                 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5342                         .src_port = mask->src_port_mask,
5343                         .dst_port = mask->dst_port_mask,
5344                 };
5345                 attributes->items[2] = (struct rte_flow_item){
5346                         .type = RTE_FLOW_ITEM_TYPE_UDP,
5347                         .spec = &attributes->l4,
5348                         .mask = &attributes->l4_mask,
5349                 };
5350                 break;
5351         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5352                 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5353                         .src_port = input->flow.tcp6_flow.src_port,
5354                         .dst_port = input->flow.tcp6_flow.dst_port,
5355                 };
5356                 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5357                         .src_port = mask->src_port_mask,
5358                         .dst_port = mask->dst_port_mask,
5359                 };
5360                 attributes->items[2] = (struct rte_flow_item){
5361                         .type = RTE_FLOW_ITEM_TYPE_TCP,
5362                         .spec = &attributes->l4,
5363                         .mask = &attributes->l4_mask,
5364                 };
5365                 break;
5366         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5367         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5368                 break;
5369         default:
5370                 DRV_LOG(ERR, "port %u invalid flow type%d",
5371                         dev->data->port_id, fdir_filter->input.flow_type);
5372                 rte_errno = ENOTSUP;
5373                 return -rte_errno;
5374         }
5375         return 0;
5376 }
5377
5378 #define FLOW_FDIR_CMP(f1, f2, fld) \
5379         memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5380
5381 /**
5382  * Compare two FDIR flows. If items and actions are identical, the two flows are
5383  * regarded as same.
5384  *
5385  * @param dev
5386  *   Pointer to Ethernet device.
5387  * @param f1
5388  *   FDIR flow to compare.
5389  * @param f2
5390  *   FDIR flow to compare.
5391  *
5392  * @return
5393  *   Zero on match, 1 otherwise.
5394  */
5395 static int
5396 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5397 {
5398         if (FLOW_FDIR_CMP(f1, f2, attr) ||
5399             FLOW_FDIR_CMP(f1, f2, l2) ||
5400             FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5401             FLOW_FDIR_CMP(f1, f2, l3) ||
5402             FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5403             FLOW_FDIR_CMP(f1, f2, l4) ||
5404             FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5405             FLOW_FDIR_CMP(f1, f2, actions[0].type))
5406                 return 1;
5407         if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5408             FLOW_FDIR_CMP(f1, f2, queue))
5409                 return 1;
5410         return 0;
5411 }
5412
5413 /**
5414  * Search device flow list to find out a matched FDIR flow.
5415  *
5416  * @param dev
5417  *   Pointer to Ethernet device.
5418  * @param fdir_flow
5419  *   FDIR flow to lookup.
5420  *
5421  * @return
5422  *   Index of flow if found, 0 otherwise.
5423  */
5424 static uint32_t
5425 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5426 {
5427         struct mlx5_priv *priv = dev->data->dev_private;
5428         uint32_t flow_idx = 0;
5429         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5430
5431         MLX5_ASSERT(fdir_flow);
5432         LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5433                 if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5434                         DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5435                                 dev->data->port_id, flow_idx);
5436                         flow_idx = priv_fdir_flow->rix_flow;
5437                         break;
5438                 }
5439         }
5440         return flow_idx;
5441 }
5442
5443 /**
5444  * Add new flow director filter and store it in list.
5445  *
5446  * @param dev
5447  *   Pointer to Ethernet device.
5448  * @param fdir_filter
5449  *   Flow director filter to add.
5450  *
5451  * @return
5452  *   0 on success, a negative errno value otherwise and rte_errno is set.
5453  */
5454 static int
5455 flow_fdir_filter_add(struct rte_eth_dev *dev,
5456                      const struct rte_eth_fdir_filter *fdir_filter)
5457 {
5458         struct mlx5_priv *priv = dev->data->dev_private;
5459         struct mlx5_fdir *fdir_flow;
5460         struct rte_flow *flow;
5461         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5462         uint32_t flow_idx;
5463         int ret;
5464
5465         fdir_flow = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*fdir_flow), 0,
5466                                 SOCKET_ID_ANY);
5467         if (!fdir_flow) {
5468                 rte_errno = ENOMEM;
5469                 return -rte_errno;
5470         }
5471         ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5472         if (ret)
5473                 goto error;
5474         flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5475         if (flow_idx) {
5476                 rte_errno = EEXIST;
5477                 goto error;
5478         }
5479         priv_fdir_flow = mlx5_malloc(MLX5_MEM_ZERO,
5480                                      sizeof(struct mlx5_fdir_flow),
5481                                      0, SOCKET_ID_ANY);
5482         if (!priv_fdir_flow) {
5483                 rte_errno = ENOMEM;
5484                 goto error;
5485         }
5486         flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5487                                     fdir_flow->items, fdir_flow->actions, true,
5488                                     NULL);
5489         flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5490         if (!flow)
5491                 goto error;
5492         flow->fdir = 1;
5493         priv_fdir_flow->fdir = fdir_flow;
5494         priv_fdir_flow->rix_flow = flow_idx;
5495         LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5496         DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5497                 dev->data->port_id, (void *)flow);
5498         return 0;
5499 error:
5500         mlx5_free(priv_fdir_flow);
5501         mlx5_free(fdir_flow);
5502         return -rte_errno;
5503 }
5504
5505 /**
5506  * Delete specific filter.
5507  *
5508  * @param dev
5509  *   Pointer to Ethernet device.
5510  * @param fdir_filter
5511  *   Filter to be deleted.
5512  *
5513  * @return
5514  *   0 on success, a negative errno value otherwise and rte_errno is set.
5515  */
5516 static int
5517 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5518                         const struct rte_eth_fdir_filter *fdir_filter)
5519 {
5520         struct mlx5_priv *priv = dev->data->dev_private;
5521         uint32_t flow_idx;
5522         struct mlx5_fdir fdir_flow = {
5523                 .attr.group = 0,
5524         };
5525         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5526         int ret;
5527
5528         ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5529         if (ret)
5530                 return -rte_errno;
5531         LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5532                 /* Find the fdir in priv list */
5533                 if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5534                         break;
5535         }
5536         if (!priv_fdir_flow)
5537                 return 0;
5538         LIST_REMOVE(priv_fdir_flow, next);
5539         flow_idx = priv_fdir_flow->rix_flow;
5540         flow_list_destroy(dev, &priv->flows, flow_idx);
5541         mlx5_free(priv_fdir_flow->fdir);
5542         mlx5_free(priv_fdir_flow);
5543         DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5544                 dev->data->port_id, flow_idx);
5545         return 0;
5546 }
5547
5548 /**
5549  * Update queue for specific filter.
5550  *
5551  * @param dev
5552  *   Pointer to Ethernet device.
5553  * @param fdir_filter
5554  *   Filter to be updated.
5555  *
5556  * @return
5557  *   0 on success, a negative errno value otherwise and rte_errno is set.
5558  */
5559 static int
5560 flow_fdir_filter_update(struct rte_eth_dev *dev,
5561                         const struct rte_eth_fdir_filter *fdir_filter)
5562 {
5563         int ret;
5564
5565         ret = flow_fdir_filter_delete(dev, fdir_filter);
5566         if (ret)
5567                 return ret;
5568         return flow_fdir_filter_add(dev, fdir_filter);
5569 }
5570
5571 /**
5572  * Flush all filters.
5573  *
5574  * @param dev
5575  *   Pointer to Ethernet device.
5576  */
5577 static void
5578 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5579 {
5580         struct mlx5_priv *priv = dev->data->dev_private;
5581         struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5582
5583         while (!LIST_EMPTY(&priv->fdir_flows)) {
5584                 priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5585                 LIST_REMOVE(priv_fdir_flow, next);
5586                 flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5587                 mlx5_free(priv_fdir_flow->fdir);
5588                 mlx5_free(priv_fdir_flow);
5589         }
5590 }
5591
5592 /**
5593  * Get flow director information.
5594  *
5595  * @param dev
5596  *   Pointer to Ethernet device.
5597  * @param[out] fdir_info
5598  *   Resulting flow director information.
5599  */
5600 static void
5601 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5602 {
5603         struct rte_eth_fdir_masks *mask =
5604                 &dev->data->dev_conf.fdir_conf.mask;
5605
5606         fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5607         fdir_info->guarant_spc = 0;
5608         rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5609         fdir_info->max_flexpayload = 0;
5610         fdir_info->flow_types_mask[0] = 0;
5611         fdir_info->flex_payload_unit = 0;
5612         fdir_info->max_flex_payload_segment_num = 0;
5613         fdir_info->flex_payload_limit = 0;
5614         memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5615 }
5616
5617 /**
5618  * Deal with flow director operations.
5619  *
5620  * @param dev
5621  *   Pointer to Ethernet device.
5622  * @param filter_op
5623  *   Operation to perform.
5624  * @param arg
5625  *   Pointer to operation-specific structure.
5626  *
5627  * @return
5628  *   0 on success, a negative errno value otherwise and rte_errno is set.
5629  */
5630 static int
5631 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5632                     void *arg)
5633 {
5634         enum rte_fdir_mode fdir_mode =
5635                 dev->data->dev_conf.fdir_conf.mode;
5636
5637         if (filter_op == RTE_ETH_FILTER_NOP)
5638                 return 0;
5639         if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5640             fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5641                 DRV_LOG(ERR, "port %u flow director mode %d not supported",
5642                         dev->data->port_id, fdir_mode);
5643                 rte_errno = EINVAL;
5644                 return -rte_errno;
5645         }
5646         switch (filter_op) {
5647         case RTE_ETH_FILTER_ADD:
5648                 return flow_fdir_filter_add(dev, arg);
5649         case RTE_ETH_FILTER_UPDATE:
5650                 return flow_fdir_filter_update(dev, arg);
5651         case RTE_ETH_FILTER_DELETE:
5652                 return flow_fdir_filter_delete(dev, arg);
5653         case RTE_ETH_FILTER_FLUSH:
5654                 flow_fdir_filter_flush(dev);
5655                 break;
5656         case RTE_ETH_FILTER_INFO:
5657                 flow_fdir_info_get(dev, arg);
5658                 break;
5659         default:
5660                 DRV_LOG(DEBUG, "port %u unknown operation %u",
5661                         dev->data->port_id, filter_op);
5662                 rte_errno = EINVAL;
5663                 return -rte_errno;
5664         }
5665         return 0;
5666 }
5667
5668 /**
5669  * Manage filter operations.
5670  *
5671  * @param dev
5672  *   Pointer to Ethernet device structure.
5673  * @param filter_type
5674  *   Filter type.
5675  * @param filter_op
5676  *   Operation to perform.
5677  * @param arg
5678  *   Pointer to operation-specific structure.
5679  *
5680  * @return
5681  *   0 on success, a negative errno value otherwise and rte_errno is set.
5682  */
5683 int
5684 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5685                      enum rte_filter_type filter_type,
5686                      enum rte_filter_op filter_op,
5687                      void *arg)
5688 {
5689         switch (filter_type) {
5690         case RTE_ETH_FILTER_GENERIC:
5691                 if (filter_op != RTE_ETH_FILTER_GET) {
5692                         rte_errno = EINVAL;
5693                         return -rte_errno;
5694                 }
5695                 *(const void **)arg = &mlx5_flow_ops;
5696                 return 0;
5697         case RTE_ETH_FILTER_FDIR:
5698                 return flow_fdir_ctrl_func(dev, filter_op, arg);
5699         default:
5700                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
5701                         dev->data->port_id, filter_type);
5702                 rte_errno = ENOTSUP;
5703                 return -rte_errno;
5704         }
5705         return 0;
5706 }
5707
5708 /**
5709  * Create the needed meter and suffix tables.
5710  *
5711  * @param[in] dev
5712  *   Pointer to Ethernet device.
5713  * @param[in] fm
5714  *   Pointer to the flow meter.
5715  *
5716  * @return
5717  *   Pointer to table set on success, NULL otherwise.
5718  */
5719 struct mlx5_meter_domains_infos *
5720 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5721                           const struct mlx5_flow_meter *fm)
5722 {
5723         const struct mlx5_flow_driver_ops *fops;
5724
5725         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5726         return fops->create_mtr_tbls(dev, fm);
5727 }
5728
5729 /**
5730  * Destroy the meter table set.
5731  *
5732  * @param[in] dev
5733  *   Pointer to Ethernet device.
5734  * @param[in] tbl
5735  *   Pointer to the meter table set.
5736  *
5737  * @return
5738  *   0 on success.
5739  */
5740 int
5741 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5742                            struct mlx5_meter_domains_infos *tbls)
5743 {
5744         const struct mlx5_flow_driver_ops *fops;
5745
5746         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5747         return fops->destroy_mtr_tbls(dev, tbls);
5748 }
5749
5750 /**
5751  * Create policer rules.
5752  *
5753  * @param[in] dev
5754  *   Pointer to Ethernet device.
5755  * @param[in] fm
5756  *   Pointer to flow meter structure.
5757  * @param[in] attr
5758  *   Pointer to flow attributes.
5759  *
5760  * @return
5761  *   0 on success, -1 otherwise.
5762  */
5763 int
5764 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5765                                struct mlx5_flow_meter *fm,
5766                                const struct rte_flow_attr *attr)
5767 {
5768         const struct mlx5_flow_driver_ops *fops;
5769
5770         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5771         return fops->create_policer_rules(dev, fm, attr);
5772 }
5773
5774 /**
5775  * Destroy policer rules.
5776  *
5777  * @param[in] fm
5778  *   Pointer to flow meter structure.
5779  * @param[in] attr
5780  *   Pointer to flow attributes.
5781  *
5782  * @return
5783  *   0 on success, -1 otherwise.
5784  */
5785 int
5786 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5787                                 struct mlx5_flow_meter *fm,
5788                                 const struct rte_flow_attr *attr)
5789 {
5790         const struct mlx5_flow_driver_ops *fops;
5791
5792         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5793         return fops->destroy_policer_rules(dev, fm, attr);
5794 }
5795
5796 /**
5797  * Allocate a counter.
5798  *
5799  * @param[in] dev
5800  *   Pointer to Ethernet device structure.
5801  *
5802  * @return
5803  *   Index to allocated counter  on success, 0 otherwise.
5804  */
5805 uint32_t
5806 mlx5_counter_alloc(struct rte_eth_dev *dev)
5807 {
5808         const struct mlx5_flow_driver_ops *fops;
5809         struct rte_flow_attr attr = { .transfer = 0 };
5810
5811         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5812                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5813                 return fops->counter_alloc(dev);
5814         }
5815         DRV_LOG(ERR,
5816                 "port %u counter allocate is not supported.",
5817                  dev->data->port_id);
5818         return 0;
5819 }
5820
5821 /**
5822  * Free a counter.
5823  *
5824  * @param[in] dev
5825  *   Pointer to Ethernet device structure.
5826  * @param[in] cnt
5827  *   Index to counter to be free.
5828  */
5829 void
5830 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5831 {
5832         const struct mlx5_flow_driver_ops *fops;
5833         struct rte_flow_attr attr = { .transfer = 0 };
5834
5835         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5836                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5837                 fops->counter_free(dev, cnt);
5838                 return;
5839         }
5840         DRV_LOG(ERR,
5841                 "port %u counter free is not supported.",
5842                  dev->data->port_id);
5843 }
5844
5845 /**
5846  * Query counter statistics.
5847  *
5848  * @param[in] dev
5849  *   Pointer to Ethernet device structure.
5850  * @param[in] cnt
5851  *   Index to counter to query.
5852  * @param[in] clear
5853  *   Set to clear counter statistics.
5854  * @param[out] pkts
5855  *   The counter hits packets number to save.
5856  * @param[out] bytes
5857  *   The counter hits bytes number to save.
5858  *
5859  * @return
5860  *   0 on success, a negative errno value otherwise.
5861  */
5862 int
5863 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5864                    bool clear, uint64_t *pkts, uint64_t *bytes)
5865 {
5866         const struct mlx5_flow_driver_ops *fops;
5867         struct rte_flow_attr attr = { .transfer = 0 };
5868
5869         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5870                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5871                 return fops->counter_query(dev, cnt, clear, pkts, bytes);
5872         }
5873         DRV_LOG(ERR,
5874                 "port %u counter query is not supported.",
5875                  dev->data->port_id);
5876         return -ENOTSUP;
5877 }
5878
5879 #define MLX5_POOL_QUERY_FREQ_US 1000000
5880
5881 /**
5882  * Get number of all validate pools.
5883  *
5884  * @param[in] sh
5885  *   Pointer to mlx5_dev_ctx_shared object.
5886  *
5887  * @return
5888  *   The number of all validate pools.
5889  */
5890 static uint32_t
5891 mlx5_get_all_valid_pool_count(struct mlx5_dev_ctx_shared *sh)
5892 {
5893         int i;
5894         uint32_t pools_n = 0;
5895
5896         for (i = 0; i < MLX5_CCONT_TYPE_MAX; ++i)
5897                 pools_n += rte_atomic16_read(&sh->cmng.ccont[i].n_valid);
5898         return pools_n;
5899 }
5900
5901 /**
5902  * Set the periodic procedure for triggering asynchronous batch queries for all
5903  * the counter pools.
5904  *
5905  * @param[in] sh
5906  *   Pointer to mlx5_dev_ctx_shared object.
5907  */
5908 void
5909 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
5910 {
5911         uint32_t pools_n, us;
5912
5913         pools_n = mlx5_get_all_valid_pool_count(sh);
5914         us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5915         DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5916         if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5917                 sh->cmng.query_thread_on = 0;
5918                 DRV_LOG(ERR, "Cannot reinitialize query alarm");
5919         } else {
5920                 sh->cmng.query_thread_on = 1;
5921         }
5922 }
5923
5924 /**
5925  * The periodic procedure for triggering asynchronous batch queries for all the
5926  * counter pools. This function is probably called by the host thread.
5927  *
5928  * @param[in] arg
5929  *   The parameter for the alarm process.
5930  */
5931 void
5932 mlx5_flow_query_alarm(void *arg)
5933 {
5934         struct mlx5_dev_ctx_shared *sh = arg;
5935         struct mlx5_devx_obj *dcs;
5936         uint16_t offset;
5937         int ret;
5938         uint8_t batch = sh->cmng.batch;
5939         uint8_t age = sh->cmng.age;
5940         uint16_t pool_index = sh->cmng.pool_index;
5941         struct mlx5_pools_container *cont;
5942         struct mlx5_flow_counter_pool *pool;
5943         int cont_loop = MLX5_CCONT_TYPE_MAX;
5944
5945         if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5946                 goto set_alarm;
5947 next_container:
5948         cont = MLX5_CNT_CONTAINER(sh, batch, age);
5949         rte_spinlock_lock(&cont->resize_sl);
5950         if (!cont->pools) {
5951                 rte_spinlock_unlock(&cont->resize_sl);
5952                 /* Check if all the containers are empty. */
5953                 if (unlikely(--cont_loop == 0))
5954                         goto set_alarm;
5955                 batch ^= 0x1;
5956                 pool_index = 0;
5957                 if (batch == 0 && pool_index == 0) {
5958                         age ^= 0x1;
5959                         sh->cmng.batch = batch;
5960                         sh->cmng.age = age;
5961                 }
5962                 goto next_container;
5963         }
5964         pool = cont->pools[pool_index];
5965         rte_spinlock_unlock(&cont->resize_sl);
5966         if (pool->raw_hw)
5967                 /* There is a pool query in progress. */
5968                 goto set_alarm;
5969         pool->raw_hw =
5970                 LIST_FIRST(&sh->cmng.free_stat_raws);
5971         if (!pool->raw_hw)
5972                 /* No free counter statistics raw memory. */
5973                 goto set_alarm;
5974         dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5975                                                               (&pool->a64_dcs);
5976         if (dcs->id & (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1)) {
5977                 /* Pool without valid counter. */
5978                 pool->raw_hw = NULL;
5979                 goto next_pool;
5980         }
5981         offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5982         /*
5983          * Identify the counters released between query trigger and query
5984          * handle more effiecntly. The counter released in this gap period
5985          * should wait for a new round of query as the new arrived packets
5986          * will not be taken into account.
5987          */
5988         pool->query_gen++;
5989         ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5990                                                offset, NULL, NULL,
5991                                                pool->raw_hw->mem_mng->dm->id,
5992                                                (void *)(uintptr_t)
5993                                                (pool->raw_hw->data + offset),
5994                                                sh->devx_comp,
5995                                                (uint64_t)(uintptr_t)pool);
5996         if (ret) {
5997                 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5998                         " %d", pool->min_dcs->id);
5999                 pool->raw_hw = NULL;
6000                 goto set_alarm;
6001         }
6002         pool->raw_hw->min_dcs_id = dcs->id;
6003         LIST_REMOVE(pool->raw_hw, next);
6004         sh->cmng.pending_queries++;
6005 next_pool:
6006         pool_index++;
6007         if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
6008                 batch ^= 0x1;
6009                 pool_index = 0;
6010                 if (batch == 0 && pool_index == 0)
6011                         age ^= 0x1;
6012         }
6013 set_alarm:
6014         sh->cmng.batch = batch;
6015         sh->cmng.pool_index = pool_index;
6016         sh->cmng.age = age;
6017         mlx5_set_query_alarm(sh);
6018 }
6019
6020 /**
6021  * Check and callback event for new aged flow in the counter pool
6022  *
6023  * @param[in] sh
6024  *   Pointer to mlx5_dev_ctx_shared object.
6025  * @param[in] pool
6026  *   Pointer to Current counter pool.
6027  */
6028 static void
6029 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6030                    struct mlx5_flow_counter_pool *pool)
6031 {
6032         struct mlx5_priv *priv;
6033         struct mlx5_flow_counter *cnt;
6034         struct mlx5_age_info *age_info;
6035         struct mlx5_age_param *age_param;
6036         struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6037         struct mlx5_counter_stats_raw *prev = pool->raw;
6038         uint16_t curr = rte_rdtsc() / (rte_get_tsc_hz() / 10);
6039         uint32_t i;
6040
6041         for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6042                 cnt = MLX5_POOL_GET_CNT(pool, i);
6043                 age_param = MLX5_CNT_TO_AGE(cnt);
6044                 if (rte_atomic16_read(&age_param->state) != AGE_CANDIDATE)
6045                         continue;
6046                 if (cur->data[i].hits != prev->data[i].hits) {
6047                         age_param->expire = curr + age_param->timeout;
6048                         continue;
6049                 }
6050                 if ((uint16_t)(curr - age_param->expire) >= (UINT16_MAX / 2))
6051                         continue;
6052                 /**
6053                  * Hold the lock first, or if between the
6054                  * state AGE_TMOUT and tailq operation the
6055                  * release happened, the release procedure
6056                  * may delete a non-existent tailq node.
6057                  */
6058                 priv = rte_eth_devices[age_param->port_id].data->dev_private;
6059                 age_info = GET_PORT_AGE_INFO(priv);
6060                 rte_spinlock_lock(&age_info->aged_sl);
6061                 /* If the cpmset fails, release happens. */
6062                 if (rte_atomic16_cmpset((volatile uint16_t *)
6063                                         &age_param->state,
6064                                         AGE_CANDIDATE,
6065                                         AGE_TMOUT) ==
6066                                         AGE_CANDIDATE) {
6067                         TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6068                         MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6069                 }
6070                 rte_spinlock_unlock(&age_info->aged_sl);
6071         }
6072         for (i = 0; i < sh->max_port; i++) {
6073                 age_info = &sh->port[i].age_info;
6074                 if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
6075                         continue;
6076                 if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
6077                         _rte_eth_dev_callback_process
6078                                 (&rte_eth_devices[sh->port[i].devx_ih_port_id],
6079                                 RTE_ETH_EVENT_FLOW_AGED, NULL);
6080                 age_info->flags = 0;
6081         }
6082 }
6083
6084 /**
6085  * Handler for the HW respond about ready values from an asynchronous batch
6086  * query. This function is probably called by the host thread.
6087  *
6088  * @param[in] sh
6089  *   The pointer to the shared device context.
6090  * @param[in] async_id
6091  *   The Devx async ID.
6092  * @param[in] status
6093  *   The status of the completion.
6094  */
6095 void
6096 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6097                                   uint64_t async_id, int status)
6098 {
6099         struct mlx5_flow_counter_pool *pool =
6100                 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6101         struct mlx5_counter_stats_raw *raw_to_free;
6102         uint8_t age = !!IS_AGE_POOL(pool);
6103         uint8_t query_gen = pool->query_gen ^ 1;
6104         struct mlx5_pools_container *cont =
6105                 MLX5_CNT_CONTAINER(sh, !IS_EXT_POOL(pool), age);
6106
6107         if (unlikely(status)) {
6108                 raw_to_free = pool->raw_hw;
6109         } else {
6110                 raw_to_free = pool->raw;
6111                 if (IS_AGE_POOL(pool))
6112                         mlx5_flow_aging_check(sh, pool);
6113                 rte_spinlock_lock(&pool->sl);
6114                 pool->raw = pool->raw_hw;
6115                 rte_spinlock_unlock(&pool->sl);
6116                 /* Be sure the new raw counters data is updated in memory. */
6117                 rte_cio_wmb();
6118                 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6119                         rte_spinlock_lock(&cont->csl);
6120                         TAILQ_CONCAT(&cont->counters,
6121                                      &pool->counters[query_gen], next);
6122                         rte_spinlock_unlock(&cont->csl);
6123                 }
6124         }
6125         LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6126         pool->raw_hw = NULL;
6127         sh->cmng.pending_queries--;
6128 }
6129
6130 /**
6131  * Translate the rte_flow group index to HW table value.
6132  *
6133  * @param[in] attributes
6134  *   Pointer to flow attributes
6135  * @param[in] external
6136  *   Value is part of flow rule created by request external to PMD.
6137  * @param[in] group
6138  *   rte_flow group index value.
6139  * @param[out] fdb_def_rule
6140  *   Whether fdb jump to table 1 is configured.
6141  * @param[out] table
6142  *   HW table value.
6143  * @param[out] error
6144  *   Pointer to error structure.
6145  *
6146  * @return
6147  *   0 on success, a negative errno value otherwise and rte_errno is set.
6148  */
6149 int
6150 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
6151                          uint32_t group, bool fdb_def_rule, uint32_t *table,
6152                          struct rte_flow_error *error)
6153 {
6154         if (attributes->transfer && external && fdb_def_rule) {
6155                 if (group == UINT32_MAX)
6156                         return rte_flow_error_set
6157                                                 (error, EINVAL,
6158                                                  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6159                                                  NULL,
6160                                                  "group index not supported");
6161                 *table = group + 1;
6162         } else {
6163                 *table = group;
6164         }
6165         return 0;
6166 }
6167
6168 /**
6169  * Discover availability of metadata reg_c's.
6170  *
6171  * Iteratively use test flows to check availability.
6172  *
6173  * @param[in] dev
6174  *   Pointer to the Ethernet device structure.
6175  *
6176  * @return
6177  *   0 on success, a negative errno value otherwise and rte_errno is set.
6178  */
6179 int
6180 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6181 {
6182         struct mlx5_priv *priv = dev->data->dev_private;
6183         struct mlx5_dev_config *config = &priv->config;
6184         enum modify_reg idx;
6185         int n = 0;
6186
6187         /* reg_c[0] and reg_c[1] are reserved. */
6188         config->flow_mreg_c[n++] = REG_C_0;
6189         config->flow_mreg_c[n++] = REG_C_1;
6190         /* Discover availability of other reg_c's. */
6191         for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6192                 struct rte_flow_attr attr = {
6193                         .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6194                         .priority = MLX5_FLOW_PRIO_RSVD,
6195                         .ingress = 1,
6196                 };
6197                 struct rte_flow_item items[] = {
6198                         [0] = {
6199                                 .type = RTE_FLOW_ITEM_TYPE_END,
6200                         },
6201                 };
6202                 struct rte_flow_action actions[] = {
6203                         [0] = {
6204                                 .type = (enum rte_flow_action_type)
6205                                         MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6206                                 .conf = &(struct mlx5_flow_action_copy_mreg){
6207                                         .src = REG_C_1,
6208                                         .dst = idx,
6209                                 },
6210                         },
6211                         [1] = {
6212                                 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6213                                 .conf = &(struct rte_flow_action_jump){
6214                                         .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6215                                 },
6216                         },
6217                         [2] = {
6218                                 .type = RTE_FLOW_ACTION_TYPE_END,
6219                         },
6220                 };
6221                 uint32_t flow_idx;
6222                 struct rte_flow *flow;
6223                 struct rte_flow_error error;
6224
6225                 if (!config->dv_flow_en)
6226                         break;
6227                 /* Create internal flow, validation skips copy action. */
6228                 flow_idx = flow_list_create(dev, NULL, &attr, items,
6229                                             actions, false, &error);
6230                 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6231                                       flow_idx);
6232                 if (!flow)
6233                         continue;
6234                 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6235                         config->flow_mreg_c[n++] = idx;
6236                 flow_list_destroy(dev, NULL, flow_idx);
6237         }
6238         for (; n < MLX5_MREG_C_NUM; ++n)
6239                 config->flow_mreg_c[n] = REG_NON;
6240         return 0;
6241 }
6242
6243 /**
6244  * Dump flow raw hw data to file
6245  *
6246  * @param[in] dev
6247  *    The pointer to Ethernet device.
6248  * @param[in] file
6249  *   A pointer to a file for output.
6250  * @param[out] error
6251  *   Perform verbose error reporting if not NULL. PMDs initialize this
6252  *   structure in case of error only.
6253  * @return
6254  *   0 on success, a nagative value otherwise.
6255  */
6256 int
6257 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6258                    FILE *file,
6259                    struct rte_flow_error *error __rte_unused)
6260 {
6261         struct mlx5_priv *priv = dev->data->dev_private;
6262         struct mlx5_dev_ctx_shared *sh = priv->sh;
6263
6264         return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6265                                        sh->tx_domain, file);
6266 }
6267
6268 /**
6269  * Get aged-out flows.
6270  *
6271  * @param[in] dev
6272  *   Pointer to the Ethernet device structure.
6273  * @param[in] context
6274  *   The address of an array of pointers to the aged-out flows contexts.
6275  * @param[in] nb_countexts
6276  *   The length of context array pointers.
6277  * @param[out] error
6278  *   Perform verbose error reporting if not NULL. Initialized in case of
6279  *   error only.
6280  *
6281  * @return
6282  *   how many contexts get in success, otherwise negative errno value.
6283  *   if nb_contexts is 0, return the amount of all aged contexts.
6284  *   if nb_contexts is not 0 , return the amount of aged flows reported
6285  *   in the context array.
6286  */
6287 int
6288 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6289                         uint32_t nb_contexts, struct rte_flow_error *error)
6290 {
6291         const struct mlx5_flow_driver_ops *fops;
6292         struct rte_flow_attr attr = { .transfer = 0 };
6293
6294         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6295                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6296                 return fops->get_aged_flows(dev, contexts, nb_contexts,
6297                                                     error);
6298         }
6299         DRV_LOG(ERR,
6300                 "port %u get aged flows is not supported.",
6301                  dev->data->port_id);
6302         return -ENOTSUP;
6303 }