drivers/net/mlx5/mlx5_flow_dv.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2018 Mellanox Technologies, Ltd
   3  */
   4
   5 #include <sys/queue.h>
   6 #include <stdalign.h>
   7 #include <stdint.h>
   8 #include <string.h>
   9 #include <unistd.h>
  10
  11 /* Verbs header. */
  12 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
  13 #ifdef PEDANTIC
  14 #pragma GCC diagnostic ignored "-Wpedantic"
  15 #endif
  16 #include <infiniband/verbs.h>
  17 #ifdef PEDANTIC
  18 #pragma GCC diagnostic error "-Wpedantic"
  19 #endif
  20
  21 #include <rte_common.h>
  22 #include <rte_ether.h>
  23 #include <rte_ethdev_driver.h>
  24 #include <rte_flow.h>
  25 #include <rte_flow_driver.h>
  26 #include <rte_malloc.h>
  27 #include <rte_ip.h>
  28 #include <rte_gre.h>
  29
  30 #include "mlx5.h"
  31 #include "mlx5_defs.h"
  32 #include "mlx5_glue.h"
  33 #include "mlx5_flow.h"
  34 #include "mlx5_prm.h"
  35 #include "mlx5_rxtx.h"
  36
  37 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
  38
  39 #ifndef HAVE_IBV_FLOW_DEVX_COUNTERS
  40 #define MLX5DV_FLOW_ACTION_COUNTERS_DEVX 0
  41 #endif
  42
  43 #ifndef HAVE_MLX5DV_DR_ESWITCH
  44 #ifndef MLX5DV_FLOW_TABLE_TYPE_FDB
  45 #define MLX5DV_FLOW_TABLE_TYPE_FDB 0
  46 #endif
  47 #endif
  48
  49 #ifndef HAVE_MLX5DV_DR
  50 #define MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL 1
  51 #endif
  52
  53 union flow_dv_attr {
  54         struct {
  55                 uint32_t valid:1;
  56                 uint32_t ipv4:1;
  57                 uint32_t ipv6:1;
  58                 uint32_t tcp:1;
  59                 uint32_t udp:1;
  60                 uint32_t reserved:27;
  61         };
  62         uint32_t attr;
  63 };
  64
  65 /**
  66  * Initialize flow attributes structure according to flow items' types.
  67  *
  68  * @param[in] item
  69  *   Pointer to item specification.
  70  * @param[out] attr
  71  *   Pointer to flow attributes structure.
  72  */
  73 static void
  74 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr)
  75 {
  76         for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
  77                 switch (item->type) {
  78                 case RTE_FLOW_ITEM_TYPE_IPV4:
  79                         attr->ipv4 = 1;
  80                         break;
  81                 case RTE_FLOW_ITEM_TYPE_IPV6:
  82                         attr->ipv6 = 1;
  83                         break;
  84                 case RTE_FLOW_ITEM_TYPE_UDP:
  85                         attr->udp = 1;
  86                         break;
  87                 case RTE_FLOW_ITEM_TYPE_TCP:
  88                         attr->tcp = 1;
  89                         break;
  90                 default:
  91                         break;
  92                 }
  93         }
  94         attr->valid = 1;
  95 }
  96
  97 struct field_modify_info {
  98         uint32_t size; /* Size of field in protocol header, in bytes. */
  99         uint32_t offset; /* Offset of field in protocol header, in bytes. */
 100         enum mlx5_modification_field id;
 101 };
 102
 103 struct field_modify_info modify_eth[] = {
 104         {4,  0, MLX5_MODI_OUT_DMAC_47_16},
 105         {2,  4, MLX5_MODI_OUT_DMAC_15_0},
 106         {4,  6, MLX5_MODI_OUT_SMAC_47_16},
 107         {2, 10, MLX5_MODI_OUT_SMAC_15_0},
 108         {0, 0, 0},
 109 };
 110
 111 struct field_modify_info modify_ipv4[] = {
 112         {1,  8, MLX5_MODI_OUT_IPV4_TTL},
 113         {4, 12, MLX5_MODI_OUT_SIPV4},
 114         {4, 16, MLX5_MODI_OUT_DIPV4},
 115         {0, 0, 0},
 116 };
 117
 118 struct field_modify_info modify_ipv6[] = {
 119         {1,  7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
 120         {4,  8, MLX5_MODI_OUT_SIPV6_127_96},
 121         {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
 122         {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
 123         {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
 124         {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
 125         {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
 126         {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
 127         {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
 128         {0, 0, 0},
 129 };
 130
 131 struct field_modify_info modify_udp[] = {
 132         {2, 0, MLX5_MODI_OUT_UDP_SPORT},
 133         {2, 2, MLX5_MODI_OUT_UDP_DPORT},
 134         {0, 0, 0},
 135 };
 136
 137 struct field_modify_info modify_tcp[] = {
 138         {2, 0, MLX5_MODI_OUT_TCP_SPORT},
 139         {2, 2, MLX5_MODI_OUT_TCP_DPORT},
 140         {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
 141         {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
 142         {0, 0, 0},
 143 };
 144
 145 static void
 146 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
 147                           uint8_t next_protocol, uint64_t *item_flags,
 148                           int *tunnel)
 149 {
 150         assert(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
 151                item->type == RTE_FLOW_ITEM_TYPE_IPV6);
 152         if (next_protocol == IPPROTO_IPIP) {
 153                 *item_flags |= MLX5_FLOW_LAYER_IPIP;
 154                 *tunnel = 1;
 155         }
 156         if (next_protocol == IPPROTO_IPV6) {
 157                 *item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
 158                 *tunnel = 1;
 159         }
 160 }
 161
 162 /**
 163  * Acquire the synchronizing object to protect multithreaded access
 164  * to shared dv context. Lock occurs only if context is actually
 165  * shared, i.e. we have multiport IB device and representors are
 166  * created.
 167  *
 168  * @param[in] dev
 169  *   Pointer to the rte_eth_dev structure.
 170  */
 171 static void
 172 flow_d_shared_lock(struct rte_eth_dev *dev)
 173 {
 174         struct mlx5_priv *priv = dev->data->dev_private;
 175         struct mlx5_ibv_shared *sh = priv->sh;
 176
 177         if (sh->dv_refcnt > 1) {
 178                 int ret;
 179
 180                 ret = pthread_mutex_lock(&sh->dv_mutex);
 181                 assert(!ret);
 182                 (void)ret;
 183         }
 184 }
 185
 186 static void
 187 flow_d_shared_unlock(struct rte_eth_dev *dev)
 188 {
 189         struct mlx5_priv *priv = dev->data->dev_private;
 190         struct mlx5_ibv_shared *sh = priv->sh;
 191
 192         if (sh->dv_refcnt > 1) {
 193                 int ret;
 194
 195                 ret = pthread_mutex_unlock(&sh->dv_mutex);
 196                 assert(!ret);
 197                 (void)ret;
 198         }
 199 }
 200
 201 /**
 202  * Convert modify-header action to DV specification.
 203  *
 204  * @param[in] item
 205  *   Pointer to item specification.
 206  * @param[in] field
 207  *   Pointer to field modification information.
 208  * @param[in,out] resource
 209  *   Pointer to the modify-header resource.
 210  * @param[in] type
 211  *   Type of modification.
 212  * @param[out] error
 213  *   Pointer to the error structure.
 214  *
 215  * @return
 216  *   0 on success, a negative errno value otherwise and rte_errno is set.
 217  */
 218 static int
 219 flow_dv_convert_modify_action(struct rte_flow_item *item,
 220                               struct field_modify_info *field,
 221                               struct mlx5_flow_dv_modify_hdr_resource *resource,
 222                               uint32_t type,
 223                               struct rte_flow_error *error)
 224 {
 225         uint32_t i = resource->actions_num;
 226         struct mlx5_modification_cmd *actions = resource->actions;
 227         const uint8_t *spec = item->spec;
 228         const uint8_t *mask = item->mask;
 229         uint32_t set;
 230
 231         while (field->size) {
 232                 set = 0;
 233                 /* Generate modify command for each mask segment. */
 234                 memcpy(&set, &mask[field->offset], field->size);
 235                 if (set) {
 236                         if (i >= MLX5_MODIFY_NUM)
 237                                 return rte_flow_error_set(error, EINVAL,
 238                                          RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 239                                          "too many items to modify");
 240                         actions[i].action_type = type;
 241                         actions[i].field = field->id;
 242                         actions[i].length = field->size ==
 243                                         4 ? 0 : field->size * 8;
 244                         rte_memcpy(&actions[i].data[4 - field->size],
 245                                    &spec[field->offset], field->size);
 246                         actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
 247                         ++i;
 248                 }
 249                 if (resource->actions_num != i)
 250                         resource->actions_num = i;
 251                 field++;
 252         }
 253         if (!resource->actions_num)
 254                 return rte_flow_error_set(error, EINVAL,
 255                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 256                                           "invalid modification flow item");
 257         return 0;
 258 }
 259
 260 /**
 261  * Convert modify-header set IPv4 address action to DV specification.
 262  *
 263  * @param[in,out] resource
 264  *   Pointer to the modify-header resource.
 265  * @param[in] action
 266  *   Pointer to action specification.
 267  * @param[out] error
 268  *   Pointer to the error structure.
 269  *
 270  * @return
 271  *   0 on success, a negative errno value otherwise and rte_errno is set.
 272  */
 273 static int
 274 flow_dv_convert_action_modify_ipv4
 275                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 276                          const struct rte_flow_action *action,
 277                          struct rte_flow_error *error)
 278 {
 279         const struct rte_flow_action_set_ipv4 *conf =
 280                 (const struct rte_flow_action_set_ipv4 *)(action->conf);
 281         struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
 282         struct rte_flow_item_ipv4 ipv4;
 283         struct rte_flow_item_ipv4 ipv4_mask;
 284
 285         memset(&ipv4, 0, sizeof(ipv4));
 286         memset(&ipv4_mask, 0, sizeof(ipv4_mask));
 287         if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
 288                 ipv4.hdr.src_addr = conf->ipv4_addr;
 289                 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
 290         } else {
 291                 ipv4.hdr.dst_addr = conf->ipv4_addr;
 292                 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
 293         }
 294         item.spec = &ipv4;
 295         item.mask = &ipv4_mask;
 296         return flow_dv_convert_modify_action(&item, modify_ipv4, resource,
 297                                              MLX5_MODIFICATION_TYPE_SET, error);
 298 }
 299
 300 /**
 301  * Convert modify-header set IPv6 address action to DV specification.
 302  *
 303  * @param[in,out] resource
 304  *   Pointer to the modify-header resource.
 305  * @param[in] action
 306  *   Pointer to action specification.
 307  * @param[out] error
 308  *   Pointer to the error structure.
 309  *
 310  * @return
 311  *   0 on success, a negative errno value otherwise and rte_errno is set.
 312  */
 313 static int
 314 flow_dv_convert_action_modify_ipv6
 315                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 316                          const struct rte_flow_action *action,
 317                          struct rte_flow_error *error)
 318 {
 319         const struct rte_flow_action_set_ipv6 *conf =
 320                 (const struct rte_flow_action_set_ipv6 *)(action->conf);
 321         struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
 322         struct rte_flow_item_ipv6 ipv6;
 323         struct rte_flow_item_ipv6 ipv6_mask;
 324
 325         memset(&ipv6, 0, sizeof(ipv6));
 326         memset(&ipv6_mask, 0, sizeof(ipv6_mask));
 327         if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
 328                 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
 329                        sizeof(ipv6.hdr.src_addr));
 330                 memcpy(&ipv6_mask.hdr.src_addr,
 331                        &rte_flow_item_ipv6_mask.hdr.src_addr,
 332                        sizeof(ipv6.hdr.src_addr));
 333         } else {
 334                 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
 335                        sizeof(ipv6.hdr.dst_addr));
 336                 memcpy(&ipv6_mask.hdr.dst_addr,
 337                        &rte_flow_item_ipv6_mask.hdr.dst_addr,
 338                        sizeof(ipv6.hdr.dst_addr));
 339         }
 340         item.spec = &ipv6;
 341         item.mask = &ipv6_mask;
 342         return flow_dv_convert_modify_action(&item, modify_ipv6, resource,
 343                                              MLX5_MODIFICATION_TYPE_SET, error);
 344 }
 345
 346 /**
 347  * Convert modify-header set MAC address action to DV specification.
 348  *
 349  * @param[in,out] resource
 350  *   Pointer to the modify-header resource.
 351  * @param[in] action
 352  *   Pointer to action specification.
 353  * @param[out] error
 354  *   Pointer to the error structure.
 355  *
 356  * @return
 357  *   0 on success, a negative errno value otherwise and rte_errno is set.
 358  */
 359 static int
 360 flow_dv_convert_action_modify_mac
 361                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 362                          const struct rte_flow_action *action,
 363                          struct rte_flow_error *error)
 364 {
 365         const struct rte_flow_action_set_mac *conf =
 366                 (const struct rte_flow_action_set_mac *)(action->conf);
 367         struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
 368         struct rte_flow_item_eth eth;
 369         struct rte_flow_item_eth eth_mask;
 370
 371         memset(&eth, 0, sizeof(eth));
 372         memset(&eth_mask, 0, sizeof(eth_mask));
 373         if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
 374                 memcpy(&eth.src.addr_bytes, &conf->mac_addr,
 375                        sizeof(eth.src.addr_bytes));
 376                 memcpy(&eth_mask.src.addr_bytes,
 377                        &rte_flow_item_eth_mask.src.addr_bytes,
 378                        sizeof(eth_mask.src.addr_bytes));
 379         } else {
 380                 memcpy(&eth.dst.addr_bytes, &conf->mac_addr,
 381                        sizeof(eth.dst.addr_bytes));
 382                 memcpy(&eth_mask.dst.addr_bytes,
 383                        &rte_flow_item_eth_mask.dst.addr_bytes,
 384                        sizeof(eth_mask.dst.addr_bytes));
 385         }
 386         item.spec = &eth;
 387         item.mask = &eth_mask;
 388         return flow_dv_convert_modify_action(&item, modify_eth, resource,
 389                                              MLX5_MODIFICATION_TYPE_SET, error);
 390 }
 391
 392 /**
 393  * Convert modify-header set TP action to DV specification.
 394  *
 395  * @param[in,out] resource
 396  *   Pointer to the modify-header resource.
 397  * @param[in] action
 398  *   Pointer to action specification.
 399  * @param[in] items
 400  *   Pointer to rte_flow_item objects list.
 401  * @param[in] attr
 402  *   Pointer to flow attributes structure.
 403  * @param[out] error
 404  *   Pointer to the error structure.
 405  *
 406  * @return
 407  *   0 on success, a negative errno value otherwise and rte_errno is set.
 408  */
 409 static int
 410 flow_dv_convert_action_modify_tp
 411                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 412                          const struct rte_flow_action *action,
 413                          const struct rte_flow_item *items,
 414                          union flow_dv_attr *attr,
 415                          struct rte_flow_error *error)
 416 {
 417         const struct rte_flow_action_set_tp *conf =
 418                 (const struct rte_flow_action_set_tp *)(action->conf);
 419         struct rte_flow_item item;
 420         struct rte_flow_item_udp udp;
 421         struct rte_flow_item_udp udp_mask;
 422         struct rte_flow_item_tcp tcp;
 423         struct rte_flow_item_tcp tcp_mask;
 424         struct field_modify_info *field;
 425
 426         if (!attr->valid)
 427                 flow_dv_attr_init(items, attr);
 428         if (attr->udp) {
 429                 memset(&udp, 0, sizeof(udp));
 430                 memset(&udp_mask, 0, sizeof(udp_mask));
 431                 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
 432                         udp.hdr.src_port = conf->port;
 433                         udp_mask.hdr.src_port =
 434                                         rte_flow_item_udp_mask.hdr.src_port;
 435                 } else {
 436                         udp.hdr.dst_port = conf->port;
 437                         udp_mask.hdr.dst_port =
 438                                         rte_flow_item_udp_mask.hdr.dst_port;
 439                 }
 440                 item.type = RTE_FLOW_ITEM_TYPE_UDP;
 441                 item.spec = &udp;
 442                 item.mask = &udp_mask;
 443                 field = modify_udp;
 444         }
 445         if (attr->tcp) {
 446                 memset(&tcp, 0, sizeof(tcp));
 447                 memset(&tcp_mask, 0, sizeof(tcp_mask));
 448                 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
 449                         tcp.hdr.src_port = conf->port;
 450                         tcp_mask.hdr.src_port =
 451                                         rte_flow_item_tcp_mask.hdr.src_port;
 452                 } else {
 453                         tcp.hdr.dst_port = conf->port;
 454                         tcp_mask.hdr.dst_port =
 455                                         rte_flow_item_tcp_mask.hdr.dst_port;
 456                 }
 457                 item.type = RTE_FLOW_ITEM_TYPE_TCP;
 458                 item.spec = &tcp;
 459                 item.mask = &tcp_mask;
 460                 field = modify_tcp;
 461         }
 462         return flow_dv_convert_modify_action(&item, field, resource,
 463                                              MLX5_MODIFICATION_TYPE_SET, error);
 464 }
 465
 466 /**
 467  * Convert modify-header set TTL action to DV specification.
 468  *
 469  * @param[in,out] resource
 470  *   Pointer to the modify-header resource.
 471  * @param[in] action
 472  *   Pointer to action specification.
 473  * @param[in] items
 474  *   Pointer to rte_flow_item objects list.
 475  * @param[in] attr
 476  *   Pointer to flow attributes structure.
 477  * @param[out] error
 478  *   Pointer to the error structure.
 479  *
 480  * @return
 481  *   0 on success, a negative errno value otherwise and rte_errno is set.
 482  */
 483 static int
 484 flow_dv_convert_action_modify_ttl
 485                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 486                          const struct rte_flow_action *action,
 487                          const struct rte_flow_item *items,
 488                          union flow_dv_attr *attr,
 489                          struct rte_flow_error *error)
 490 {
 491         const struct rte_flow_action_set_ttl *conf =
 492                 (const struct rte_flow_action_set_ttl *)(action->conf);
 493         struct rte_flow_item item;
 494         struct rte_flow_item_ipv4 ipv4;
 495         struct rte_flow_item_ipv4 ipv4_mask;
 496         struct rte_flow_item_ipv6 ipv6;
 497         struct rte_flow_item_ipv6 ipv6_mask;
 498         struct field_modify_info *field;
 499
 500         if (!attr->valid)
 501                 flow_dv_attr_init(items, attr);
 502         if (attr->ipv4) {
 503                 memset(&ipv4, 0, sizeof(ipv4));
 504                 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
 505                 ipv4.hdr.time_to_live = conf->ttl_value;
 506                 ipv4_mask.hdr.time_to_live = 0xFF;
 507                 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
 508                 item.spec = &ipv4;
 509                 item.mask = &ipv4_mask;
 510                 field = modify_ipv4;
 511         }
 512         if (attr->ipv6) {
 513                 memset(&ipv6, 0, sizeof(ipv6));
 514                 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
 515                 ipv6.hdr.hop_limits = conf->ttl_value;
 516                 ipv6_mask.hdr.hop_limits = 0xFF;
 517                 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
 518                 item.spec = &ipv6;
 519                 item.mask = &ipv6_mask;
 520                 field = modify_ipv6;
 521         }
 522         return flow_dv_convert_modify_action(&item, field, resource,
 523                                              MLX5_MODIFICATION_TYPE_SET, error);
 524 }
 525
 526 /**
 527  * Convert modify-header decrement TTL action to DV specification.
 528  *
 529  * @param[in,out] resource
 530  *   Pointer to the modify-header resource.
 531  * @param[in] action
 532  *   Pointer to action specification.
 533  * @param[in] items
 534  *   Pointer to rte_flow_item objects list.
 535  * @param[in] attr
 536  *   Pointer to flow attributes structure.
 537  * @param[out] error
 538  *   Pointer to the error structure.
 539  *
 540  * @return
 541  *   0 on success, a negative errno value otherwise and rte_errno is set.
 542  */
 543 static int
 544 flow_dv_convert_action_modify_dec_ttl
 545                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 546                          const struct rte_flow_item *items,
 547                          union flow_dv_attr *attr,
 548                          struct rte_flow_error *error)
 549 {
 550         struct rte_flow_item item;
 551         struct rte_flow_item_ipv4 ipv4;
 552         struct rte_flow_item_ipv4 ipv4_mask;
 553         struct rte_flow_item_ipv6 ipv6;
 554         struct rte_flow_item_ipv6 ipv6_mask;
 555         struct field_modify_info *field;
 556
 557         if (!attr->valid)
 558                 flow_dv_attr_init(items, attr);
 559         if (attr->ipv4) {
 560                 memset(&ipv4, 0, sizeof(ipv4));
 561                 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
 562                 ipv4.hdr.time_to_live = 0xFF;
 563                 ipv4_mask.hdr.time_to_live = 0xFF;
 564                 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
 565                 item.spec = &ipv4;
 566                 item.mask = &ipv4_mask;
 567                 field = modify_ipv4;
 568         }
 569         if (attr->ipv6) {
 570                 memset(&ipv6, 0, sizeof(ipv6));
 571                 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
 572                 ipv6.hdr.hop_limits = 0xFF;
 573                 ipv6_mask.hdr.hop_limits = 0xFF;
 574                 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
 575                 item.spec = &ipv6;
 576                 item.mask = &ipv6_mask;
 577                 field = modify_ipv6;
 578         }
 579         return flow_dv_convert_modify_action(&item, field, resource,
 580                                              MLX5_MODIFICATION_TYPE_ADD, error);
 581 }
 582
 583 /**
 584  * Convert modify-header increment/decrement TCP Sequence number
 585  * to DV specification.
 586  *
 587  * @param[in,out] resource
 588  *   Pointer to the modify-header resource.
 589  * @param[in] action
 590  *   Pointer to action specification.
 591  * @param[out] error
 592  *   Pointer to the error structure.
 593  *
 594  * @return
 595  *   0 on success, a negative errno value otherwise and rte_errno is set.
 596  */
 597 static int
 598 flow_dv_convert_action_modify_tcp_seq
 599                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 600                          const struct rte_flow_action *action,
 601                          struct rte_flow_error *error)
 602 {
 603         const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
 604         uint64_t value = rte_be_to_cpu_32(*conf);
 605         struct rte_flow_item item;
 606         struct rte_flow_item_tcp tcp;
 607         struct rte_flow_item_tcp tcp_mask;
 608
 609         memset(&tcp, 0, sizeof(tcp));
 610         memset(&tcp_mask, 0, sizeof(tcp_mask));
 611         if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
 612                 /*
 613                  * The HW has no decrement operation, only increment operation.
 614                  * To simulate decrement X from Y using increment operation
 615                  * we need to add UINT32_MAX X times to Y.
 616                  * Each adding of UINT32_MAX decrements Y by 1.
 617                  */
 618                 value *= UINT32_MAX;
 619         tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
 620         tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
 621         item.type = RTE_FLOW_ITEM_TYPE_TCP;
 622         item.spec = &tcp;
 623         item.mask = &tcp_mask;
 624         return flow_dv_convert_modify_action(&item, modify_tcp, resource,
 625                                              MLX5_MODIFICATION_TYPE_ADD, error);
 626 }
 627
 628 /**
 629  * Convert modify-header increment/decrement TCP Acknowledgment number
 630  * to DV specification.
 631  *
 632  * @param[in,out] resource
 633  *   Pointer to the modify-header resource.
 634  * @param[in] action
 635  *   Pointer to action specification.
 636  * @param[out] error
 637  *   Pointer to the error structure.
 638  *
 639  * @return
 640  *   0 on success, a negative errno value otherwise and rte_errno is set.
 641  */
 642 static int
 643 flow_dv_convert_action_modify_tcp_ack
 644                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 645                          const struct rte_flow_action *action,
 646                          struct rte_flow_error *error)
 647 {
 648         const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
 649         uint64_t value = rte_be_to_cpu_32(*conf);
 650         struct rte_flow_item item;
 651         struct rte_flow_item_tcp tcp;
 652         struct rte_flow_item_tcp tcp_mask;
 653
 654         memset(&tcp, 0, sizeof(tcp));
 655         memset(&tcp_mask, 0, sizeof(tcp_mask));
 656         if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
 657                 /*
 658                  * The HW has no decrement operation, only increment operation.
 659                  * To simulate decrement X from Y using increment operation
 660                  * we need to add UINT32_MAX X times to Y.
 661                  * Each adding of UINT32_MAX decrements Y by 1.
 662                  */
 663                 value *= UINT32_MAX;
 664         tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
 665         tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
 666         item.type = RTE_FLOW_ITEM_TYPE_TCP;
 667         item.spec = &tcp;
 668         item.mask = &tcp_mask;
 669         return flow_dv_convert_modify_action(&item, modify_tcp, resource,
 670                                              MLX5_MODIFICATION_TYPE_ADD, error);
 671 }
 672
 673 /**
 674  * Validate META item.
 675  *
 676  * @param[in] dev
 677  *   Pointer to the rte_eth_dev structure.
 678  * @param[in] item
 679  *   Item specification.
 680  * @param[in] attr
 681  *   Attributes of flow that includes this item.
 682  * @param[out] error
 683  *   Pointer to error structure.
 684  *
 685  * @return
 686  *   0 on success, a negative errno value otherwise and rte_errno is set.
 687  */
 688 static int
 689 flow_dv_validate_item_meta(struct rte_eth_dev *dev,
 690                            const struct rte_flow_item *item,
 691                            const struct rte_flow_attr *attr,
 692                            struct rte_flow_error *error)
 693 {
 694         const struct rte_flow_item_meta *spec = item->spec;
 695         const struct rte_flow_item_meta *mask = item->mask;
 696         const struct rte_flow_item_meta nic_mask = {
 697                 .data = RTE_BE32(UINT32_MAX)
 698         };
 699         int ret;
 700         uint64_t offloads = dev->data->dev_conf.txmode.offloads;
 701
 702         if (!(offloads & DEV_TX_OFFLOAD_MATCH_METADATA))
 703                 return rte_flow_error_set(error, EPERM,
 704                                           RTE_FLOW_ERROR_TYPE_ITEM,
 705                                           NULL,
 706                                           "match on metadata offload "
 707                                           "configuration is off for this port");
 708         if (!spec)
 709                 return rte_flow_error_set(error, EINVAL,
 710                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
 711                                           item->spec,
 712                                           "data cannot be empty");
 713         if (!spec->data)
 714                 return rte_flow_error_set(error, EINVAL,
 715                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
 716                                           NULL,
 717                                           "data cannot be zero");
 718         if (!mask)
 719                 mask = &rte_flow_item_meta_mask;
 720         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
 721                                         (const uint8_t *)&nic_mask,
 722                                         sizeof(struct rte_flow_item_meta),
 723                                         error);
 724         if (ret < 0)
 725                 return ret;
 726         if (attr->ingress)
 727                 return rte_flow_error_set(error, ENOTSUP,
 728                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
 729                                           NULL,
 730                                           "pattern not supported for ingress");
 731         return 0;
 732 }
 733
 734 /**
 735  * Validate vport item.
 736  *
 737  * @param[in] dev
 738  *   Pointer to the rte_eth_dev structure.
 739  * @param[in] item
 740  *   Item specification.
 741  * @param[in] attr
 742  *   Attributes of flow that includes this item.
 743  * @param[in] item_flags
 744  *   Bit-fields that holds the items detected until now.
 745  * @param[out] error
 746  *   Pointer to error structure.
 747  *
 748  * @return
 749  *   0 on success, a negative errno value otherwise and rte_errno is set.
 750  */
 751 static int
 752 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
 753                               const struct rte_flow_item *item,
 754                               const struct rte_flow_attr *attr,
 755                               uint64_t item_flags,
 756                               struct rte_flow_error *error)
 757 {
 758         const struct rte_flow_item_port_id *spec = item->spec;
 759         const struct rte_flow_item_port_id *mask = item->mask;
 760         const struct rte_flow_item_port_id switch_mask = {
 761                         .id = 0xffffffff,
 762         };
 763         uint16_t esw_domain_id;
 764         uint16_t item_port_esw_domain_id;
 765         int ret;
 766
 767         if (!attr->transfer)
 768                 return rte_flow_error_set(error, EINVAL,
 769                                           RTE_FLOW_ERROR_TYPE_ITEM,
 770                                           NULL,
 771                                           "match on port id is valid only"
 772                                           " when transfer flag is enabled");
 773         if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
 774                 return rte_flow_error_set(error, ENOTSUP,
 775                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
 776                                           "multiple source ports are not"
 777                                           " supported");
 778         if (!mask)
 779                 mask = &switch_mask;
 780         if (mask->id != 0xffffffff)
 781                 return rte_flow_error_set(error, ENOTSUP,
 782                                            RTE_FLOW_ERROR_TYPE_ITEM_MASK,
 783                                            mask,
 784                                            "no support for partial mask on"
 785                                            " \"id\" field");
 786         ret = mlx5_flow_item_acceptable
 787                                 (item, (const uint8_t *)mask,
 788                                  (const uint8_t *)&rte_flow_item_port_id_mask,
 789                                  sizeof(struct rte_flow_item_port_id),
 790                                  error);
 791         if (ret)
 792                 return ret;
 793         if (!spec)
 794                 return 0;
 795         ret = mlx5_port_to_eswitch_info(spec->id, &item_port_esw_domain_id,
 796                                         NULL);
 797         if (ret)
 798                 return rte_flow_error_set(error, -ret,
 799                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
 800                                           "failed to obtain E-Switch info for"
 801                                           " port");
 802         ret = mlx5_port_to_eswitch_info(dev->data->port_id,
 803                                         &esw_domain_id, NULL);
 804         if (ret < 0)
 805                 return rte_flow_error_set(error, -ret,
 806                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
 807                                           NULL,
 808                                           "failed to obtain E-Switch info");
 809         if (item_port_esw_domain_id != esw_domain_id)
 810                 return rte_flow_error_set(error, -ret,
 811                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
 812                                           "cannot match on a port from a"
 813                                           " different E-Switch");
 814         return 0;
 815 }
 816
 817 /**
 818  * Validate count action.
 819  *
 820  * @param[in] dev
 821  *   device otr.
 822  * @param[out] error
 823  *   Pointer to error structure.
 824  *
 825  * @return
 826  *   0 on success, a negative errno value otherwise and rte_errno is set.
 827  */
 828 static int
 829 flow_dv_validate_action_count(struct rte_eth_dev *dev,
 830                               struct rte_flow_error *error)
 831 {
 832         struct mlx5_priv *priv = dev->data->dev_private;
 833
 834         if (!priv->config.devx)
 835                 goto notsup_err;
 836 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
 837         return 0;
 838 #endif
 839 notsup_err:
 840         return rte_flow_error_set
 841                       (error, ENOTSUP,
 842                        RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
 843                        NULL,
 844                        "count action not supported");
 845 }
 846
 847 /**
 848  * Validate the L2 encap action.
 849  *
 850  * @param[in] action_flags
 851  *   Holds the actions detected until now.
 852  * @param[in] action
 853  *   Pointer to the encap action.
 854  * @param[in] attr
 855  *   Pointer to flow attributes
 856  * @param[out] error
 857  *   Pointer to error structure.
 858  *
 859  * @return
 860  *   0 on success, a negative errno value otherwise and rte_errno is set.
 861  */
 862 static int
 863 flow_dv_validate_action_l2_encap(uint64_t action_flags,
 864                                  const struct rte_flow_action *action,
 865                                  const struct rte_flow_attr *attr,
 866                                  struct rte_flow_error *error)
 867 {
 868         if (!(action->conf))
 869                 return rte_flow_error_set(error, EINVAL,
 870                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
 871                                           "configuration cannot be null");
 872         if (action_flags & MLX5_FLOW_ACTION_DROP)
 873                 return rte_flow_error_set(error, EINVAL,
 874                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 875                                           "can't drop and encap in same flow");
 876         if (action_flags & (MLX5_FLOW_ENCAP_ACTIONS | MLX5_FLOW_DECAP_ACTIONS))
 877                 return rte_flow_error_set(error, EINVAL,
 878                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 879                                           "can only have a single encap or"
 880                                           " decap action in a flow");
 881         if (!attr->transfer && attr->ingress)
 882                 return rte_flow_error_set(error, ENOTSUP,
 883                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
 884                                           NULL,
 885                                           "encap action not supported for "
 886                                           "ingress");
 887         return 0;
 888 }
 889
 890 /**
 891  * Validate the L2 decap action.
 892  *
 893  * @param[in] action_flags
 894  *   Holds the actions detected until now.
 895  * @param[in] attr
 896  *   Pointer to flow attributes
 897  * @param[out] error
 898  *   Pointer to error structure.
 899  *
 900  * @return
 901  *   0 on success, a negative errno value otherwise and rte_errno is set.
 902  */
 903 static int
 904 flow_dv_validate_action_l2_decap(uint64_t action_flags,
 905                                  const struct rte_flow_attr *attr,
 906                                  struct rte_flow_error *error)
 907 {
 908         if (action_flags & MLX5_FLOW_ACTION_DROP)
 909                 return rte_flow_error_set(error, EINVAL,
 910                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 911                                           "can't drop and decap in same flow");
 912         if (action_flags & (MLX5_FLOW_ENCAP_ACTIONS | MLX5_FLOW_DECAP_ACTIONS))
 913                 return rte_flow_error_set(error, EINVAL,
 914                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 915                                           "can only have a single encap or"
 916                                           " decap action in a flow");
 917         if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
 918                 return rte_flow_error_set(error, EINVAL,
 919                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 920                                           "can't have decap action after"
 921                                           " modify action");
 922         if (attr->egress)
 923                 return rte_flow_error_set(error, ENOTSUP,
 924                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
 925                                           NULL,
 926                                           "decap action not supported for "
 927                                           "egress");
 928         return 0;
 929 }
 930
 931 /**
 932  * Validate the raw encap action.
 933  *
 934  * @param[in] action_flags
 935  *   Holds the actions detected until now.
 936  * @param[in] action
 937  *   Pointer to the encap action.
 938  * @param[in] attr
 939  *   Pointer to flow attributes
 940  * @param[out] error
 941  *   Pointer to error structure.
 942  *
 943  * @return
 944  *   0 on success, a negative errno value otherwise and rte_errno is set.
 945  */
 946 static int
 947 flow_dv_validate_action_raw_encap(uint64_t action_flags,
 948                                   const struct rte_flow_action *action,
 949                                   const struct rte_flow_attr *attr,
 950                                   struct rte_flow_error *error)
 951 {
 952         const struct rte_flow_action_raw_encap *raw_encap =
 953                 (const struct rte_flow_action_raw_encap *)action->conf;
 954         if (!(action->conf))
 955                 return rte_flow_error_set(error, EINVAL,
 956                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
 957                                           "configuration cannot be null");
 958         if (action_flags & MLX5_FLOW_ACTION_DROP)
 959                 return rte_flow_error_set(error, EINVAL,
 960                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 961                                           "can't drop and encap in same flow");
 962         if (action_flags & MLX5_FLOW_ENCAP_ACTIONS)
 963                 return rte_flow_error_set(error, EINVAL,
 964                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 965                                           "can only have a single encap"
 966                                           " action in a flow");
 967         /* encap without preceding decap is not supported for ingress */
 968         if (!attr->transfer &&  attr->ingress &&
 969             !(action_flags & MLX5_FLOW_ACTION_RAW_DECAP))
 970                 return rte_flow_error_set(error, ENOTSUP,
 971                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
 972                                           NULL,
 973                                           "encap action not supported for "
 974                                           "ingress");
 975         if (!raw_encap->size || !raw_encap->data)
 976                 return rte_flow_error_set(error, EINVAL,
 977                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
 978                                           "raw encap data cannot be empty");
 979         return 0;
 980 }
 981
 982 /**
 983  * Validate the raw decap action.
 984  *
 985  * @param[in] action_flags
 986  *   Holds the actions detected until now.
 987  * @param[in] action
 988  *   Pointer to the encap action.
 989  * @param[in] attr
 990  *   Pointer to flow attributes
 991  * @param[out] error
 992  *   Pointer to error structure.
 993  *
 994  * @return
 995  *   0 on success, a negative errno value otherwise and rte_errno is set.
 996  */
 997 static int
 998 flow_dv_validate_action_raw_decap(uint64_t action_flags,
 999                                   const struct rte_flow_action *action,
1000                                   const struct rte_flow_attr *attr,
1001                                   struct rte_flow_error *error)
1002 {
1003         if (action_flags & MLX5_FLOW_ACTION_DROP)
1004                 return rte_flow_error_set(error, EINVAL,
1005                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1006                                           "can't drop and decap in same flow");
1007         if (action_flags & MLX5_FLOW_ENCAP_ACTIONS)
1008                 return rte_flow_error_set(error, EINVAL,
1009                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1010                                           "can't have encap action before"
1011                                           " decap action");
1012         if (action_flags & MLX5_FLOW_DECAP_ACTIONS)
1013                 return rte_flow_error_set(error, EINVAL,
1014                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1015                                           "can only have a single decap"
1016                                           " action in a flow");
1017         if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
1018                 return rte_flow_error_set(error, EINVAL,
1019                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1020                                           "can't have decap action after"
1021                                           " modify action");
1022         /* decap action is valid on egress only if it is followed by encap */
1023         if (attr->egress) {
1024                 for (; action->type != RTE_FLOW_ACTION_TYPE_END &&
1025                        action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP;
1026                        action++) {
1027                 }
1028                 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP)
1029                         return rte_flow_error_set
1030                                         (error, ENOTSUP,
1031                                          RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
1032                                          NULL, "decap action not supported"
1033                                          " for egress");
1034         }
1035         return 0;
1036 }
1037
1038 /**
1039  * Find existing encap/decap resource or create and register a new one.
1040  *
1041  * @param dev[in, out]
1042  *   Pointer to rte_eth_dev structure.
1043  * @param[in, out] resource
1044  *   Pointer to encap/decap resource.
1045  * @parm[in, out] dev_flow
1046  *   Pointer to the dev_flow.
1047  * @param[out] error
1048  *   pointer to error structure.
1049  *
1050  * @return
1051  *   0 on success otherwise -errno and errno is set.
1052  */
1053 static int
1054 flow_dv_encap_decap_resource_register
1055                         (struct rte_eth_dev *dev,
1056                          struct mlx5_flow_dv_encap_decap_resource *resource,
1057                          struct mlx5_flow *dev_flow,
1058                          struct rte_flow_error *error)
1059 {
1060         struct mlx5_priv *priv = dev->data->dev_private;
1061         struct mlx5_ibv_shared *sh = priv->sh;
1062         struct mlx5_flow_dv_encap_decap_resource *cache_resource;
1063         struct rte_flow *flow = dev_flow->flow;
1064         struct mlx5dv_dr_domain *domain;
1065
1066         resource->flags = flow->group ? 0 : 1;
1067         if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
1068                 domain = sh->fdb_domain;
1069         else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
1070                 domain = sh->rx_domain;
1071         else
1072                 domain = sh->tx_domain;
1073
1074         /* Lookup a matching resource from cache. */
1075         LIST_FOREACH(cache_resource, &sh->encaps_decaps, next) {
1076                 if (resource->reformat_type == cache_resource->reformat_type &&
1077                     resource->ft_type == cache_resource->ft_type &&
1078                     resource->flags == cache_resource->flags &&
1079                     resource->size == cache_resource->size &&
1080                     !memcmp((const void *)resource->buf,
1081                             (const void *)cache_resource->buf,
1082                             resource->size)) {
1083                         DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
1084                                 (void *)cache_resource,
1085                                 rte_atomic32_read(&cache_resource->refcnt));
1086                         rte_atomic32_inc(&cache_resource->refcnt);
1087                         dev_flow->dv.encap_decap = cache_resource;
1088                         return 0;
1089                 }
1090         }
1091         /* Register new encap/decap resource. */
1092         cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
1093         if (!cache_resource)
1094                 return rte_flow_error_set(error, ENOMEM,
1095                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1096                                           "cannot allocate resource memory");
1097         *cache_resource = *resource;
1098         cache_resource->verbs_action =
1099                 mlx5_glue->dv_create_flow_action_packet_reformat
1100                         (sh->ctx, cache_resource->reformat_type,
1101                          cache_resource->ft_type, domain, cache_resource->flags,
1102                          cache_resource->size,
1103                          (cache_resource->size ? cache_resource->buf : NULL));
1104         if (!cache_resource->verbs_action) {
1105                 rte_free(cache_resource);
1106                 return rte_flow_error_set(error, ENOMEM,
1107                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1108                                           NULL, "cannot create action");
1109         }
1110         rte_atomic32_init(&cache_resource->refcnt);
1111         rte_atomic32_inc(&cache_resource->refcnt);
1112         LIST_INSERT_HEAD(&sh->encaps_decaps, cache_resource, next);
1113         dev_flow->dv.encap_decap = cache_resource;
1114         DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
1115                 (void *)cache_resource,
1116                 rte_atomic32_read(&cache_resource->refcnt));
1117         return 0;
1118 }
1119
1120 /**
1121  * Find existing table jump resource or create and register a new one.
1122  *
1123  * @param dev[in, out]
1124  *   Pointer to rte_eth_dev structure.
1125  * @param[in, out] resource
1126  *   Pointer to jump table resource.
1127  * @parm[in, out] dev_flow
1128  *   Pointer to the dev_flow.
1129  * @param[out] error
1130  *   pointer to error structure.
1131  *
1132  * @return
1133  *   0 on success otherwise -errno and errno is set.
1134  */
1135 static int
1136 flow_dv_jump_tbl_resource_register
1137                         (struct rte_eth_dev *dev,
1138                          struct mlx5_flow_dv_jump_tbl_resource *resource,
1139                          struct mlx5_flow *dev_flow,
1140                          struct rte_flow_error *error)
1141 {
1142         struct mlx5_priv *priv = dev->data->dev_private;
1143         struct mlx5_ibv_shared *sh = priv->sh;
1144         struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
1145
1146         /* Lookup a matching resource from cache. */
1147         LIST_FOREACH(cache_resource, &sh->jump_tbl, next) {
1148                 if (resource->tbl == cache_resource->tbl) {
1149                         DRV_LOG(DEBUG, "jump table resource resource %p: refcnt %d++",
1150                                 (void *)cache_resource,
1151                                 rte_atomic32_read(&cache_resource->refcnt));
1152                         rte_atomic32_inc(&cache_resource->refcnt);
1153                         dev_flow->dv.jump = cache_resource;
1154                         return 0;
1155                 }
1156         }
1157         /* Register new jump table resource. */
1158         cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
1159         if (!cache_resource)
1160                 return rte_flow_error_set(error, ENOMEM,
1161                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1162                                           "cannot allocate resource memory");
1163         *cache_resource = *resource;
1164         cache_resource->action =
1165                 mlx5_glue->dr_create_flow_action_dest_flow_tbl
1166                 (resource->tbl->obj);
1167         if (!cache_resource->action) {
1168                 rte_free(cache_resource);
1169                 return rte_flow_error_set(error, ENOMEM,
1170                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1171                                           NULL, "cannot create action");
1172         }
1173         rte_atomic32_init(&cache_resource->refcnt);
1174         rte_atomic32_inc(&cache_resource->refcnt);
1175         LIST_INSERT_HEAD(&sh->jump_tbl, cache_resource, next);
1176         dev_flow->dv.jump = cache_resource;
1177         DRV_LOG(DEBUG, "new jump table  resource %p: refcnt %d++",
1178                 (void *)cache_resource,
1179                 rte_atomic32_read(&cache_resource->refcnt));
1180         return 0;
1181 }
1182
1183 /**
1184  * Find existing table port ID resource or create and register a new one.
1185  *
1186  * @param dev[in, out]
1187  *   Pointer to rte_eth_dev structure.
1188  * @param[in, out] resource
1189  *   Pointer to port ID action resource.
1190  * @parm[in, out] dev_flow
1191  *   Pointer to the dev_flow.
1192  * @param[out] error
1193  *   pointer to error structure.
1194  *
1195  * @return
1196  *   0 on success otherwise -errno and errno is set.
1197  */
1198 static int
1199 flow_dv_port_id_action_resource_register
1200                         (struct rte_eth_dev *dev,
1201                          struct mlx5_flow_dv_port_id_action_resource *resource,
1202                          struct mlx5_flow *dev_flow,
1203                          struct rte_flow_error *error)
1204 {
1205         struct mlx5_priv *priv = dev->data->dev_private;
1206         struct mlx5_ibv_shared *sh = priv->sh;
1207         struct mlx5_flow_dv_port_id_action_resource *cache_resource;
1208
1209         /* Lookup a matching resource from cache. */
1210         LIST_FOREACH(cache_resource, &sh->port_id_action_list, next) {
1211                 if (resource->port_id == cache_resource->port_id) {
1212                         DRV_LOG(DEBUG, "port id action resource resource %p: "
1213                                 "refcnt %d++",
1214                                 (void *)cache_resource,
1215                                 rte_atomic32_read(&cache_resource->refcnt));
1216                         rte_atomic32_inc(&cache_resource->refcnt);
1217                         dev_flow->dv.port_id_action = cache_resource;
1218                         return 0;
1219                 }
1220         }
1221         /* Register new port id action resource. */
1222         cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
1223         if (!cache_resource)
1224                 return rte_flow_error_set(error, ENOMEM,
1225                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1226                                           "cannot allocate resource memory");
1227         *cache_resource = *resource;
1228         cache_resource->action =
1229                 mlx5_glue->dr_create_flow_action_dest_vport
1230                         (priv->sh->fdb_domain, resource->port_id);
1231         if (!cache_resource->action) {
1232                 rte_free(cache_resource);
1233                 return rte_flow_error_set(error, ENOMEM,
1234                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1235                                           NULL, "cannot create action");
1236         }
1237         rte_atomic32_init(&cache_resource->refcnt);
1238         rte_atomic32_inc(&cache_resource->refcnt);
1239         LIST_INSERT_HEAD(&sh->port_id_action_list, cache_resource, next);
1240         dev_flow->dv.port_id_action = cache_resource;
1241         DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
1242                 (void *)cache_resource,
1243                 rte_atomic32_read(&cache_resource->refcnt));
1244         return 0;
1245 }
1246
1247 /**
1248  * Get the size of specific rte_flow_item_type
1249  *
1250  * @param[in] item_type
1251  *   Tested rte_flow_item_type.
1252  *
1253  * @return
1254  *   sizeof struct item_type, 0 if void or irrelevant.
1255  */
1256 static size_t
1257 flow_dv_get_item_len(const enum rte_flow_item_type item_type)
1258 {
1259         size_t retval;
1260
1261         switch (item_type) {
1262         case RTE_FLOW_ITEM_TYPE_ETH:
1263                 retval = sizeof(struct rte_flow_item_eth);
1264                 break;
1265         case RTE_FLOW_ITEM_TYPE_VLAN:
1266                 retval = sizeof(struct rte_flow_item_vlan);
1267                 break;
1268         case RTE_FLOW_ITEM_TYPE_IPV4:
1269                 retval = sizeof(struct rte_flow_item_ipv4);
1270                 break;
1271         case RTE_FLOW_ITEM_TYPE_IPV6:
1272                 retval = sizeof(struct rte_flow_item_ipv6);
1273                 break;
1274         case RTE_FLOW_ITEM_TYPE_UDP:
1275                 retval = sizeof(struct rte_flow_item_udp);
1276                 break;
1277         case RTE_FLOW_ITEM_TYPE_TCP:
1278                 retval = sizeof(struct rte_flow_item_tcp);
1279                 break;
1280         case RTE_FLOW_ITEM_TYPE_VXLAN:
1281                 retval = sizeof(struct rte_flow_item_vxlan);
1282                 break;
1283         case RTE_FLOW_ITEM_TYPE_GRE:
1284                 retval = sizeof(struct rte_flow_item_gre);
1285                 break;
1286         case RTE_FLOW_ITEM_TYPE_NVGRE:
1287                 retval = sizeof(struct rte_flow_item_nvgre);
1288                 break;
1289         case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1290                 retval = sizeof(struct rte_flow_item_vxlan_gpe);
1291                 break;
1292         case RTE_FLOW_ITEM_TYPE_MPLS:
1293                 retval = sizeof(struct rte_flow_item_mpls);
1294                 break;
1295         case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
1296         default:
1297                 retval = 0;
1298                 break;
1299         }
1300         return retval;
1301 }
1302
1303 #define MLX5_ENCAP_IPV4_VERSION         0x40
1304 #define MLX5_ENCAP_IPV4_IHL_MIN         0x05
1305 #define MLX5_ENCAP_IPV4_TTL_DEF         0x40
1306 #define MLX5_ENCAP_IPV6_VTC_FLOW        0x60000000
1307 #define MLX5_ENCAP_IPV6_HOP_LIMIT       0xff
1308 #define MLX5_ENCAP_VXLAN_FLAGS          0x08000000
1309 #define MLX5_ENCAP_VXLAN_GPE_FLAGS      0x04
1310
1311 /**
1312  * Convert the encap action data from list of rte_flow_item to raw buffer
1313  *
1314  * @param[in] items
1315  *   Pointer to rte_flow_item objects list.
1316  * @param[out] buf
1317  *   Pointer to the output buffer.
1318  * @param[out] size
1319  *   Pointer to the output buffer size.
1320  * @param[out] error
1321  *   Pointer to the error structure.
1322  *
1323  * @return
1324  *   0 on success, a negative errno value otherwise and rte_errno is set.
1325  */
1326 static int
1327 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
1328                            size_t *size, struct rte_flow_error *error)
1329 {
1330         struct rte_ether_hdr *eth = NULL;
1331         struct rte_vlan_hdr *vlan = NULL;
1332         struct rte_ipv4_hdr *ipv4 = NULL;
1333         struct rte_ipv6_hdr *ipv6 = NULL;
1334         struct rte_udp_hdr *udp = NULL;
1335         struct rte_vxlan_hdr *vxlan = NULL;
1336         struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
1337         struct rte_gre_hdr *gre = NULL;
1338         size_t len;
1339         size_t temp_size = 0;
1340
1341         if (!items)
1342                 return rte_flow_error_set(error, EINVAL,
1343                                           RTE_FLOW_ERROR_TYPE_ACTION,
1344                                           NULL, "invalid empty data");
1345         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1346                 len = flow_dv_get_item_len(items->type);
1347                 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
1348                         return rte_flow_error_set(error, EINVAL,
1349                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1350                                                   (void *)items->type,
1351                                                   "items total size is too big"
1352                                                   " for encap action");
1353                 rte_memcpy((void *)&buf[temp_size], items->spec, len);
1354                 switch (items->type) {
1355                 case RTE_FLOW_ITEM_TYPE_ETH:
1356                         eth = (struct rte_ether_hdr *)&buf[temp_size];
1357                         break;
1358                 case RTE_FLOW_ITEM_TYPE_VLAN:
1359                         vlan = (struct rte_vlan_hdr *)&buf[temp_size];
1360                         if (!eth)
1361                                 return rte_flow_error_set(error, EINVAL,
1362                                                 RTE_FLOW_ERROR_TYPE_ACTION,
1363                                                 (void *)items->type,
1364                                                 "eth header not found");
1365                         if (!eth->ether_type)
1366                                 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
1367                         break;
1368                 case RTE_FLOW_ITEM_TYPE_IPV4:
1369                         ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
1370                         if (!vlan && !eth)
1371                                 return rte_flow_error_set(error, EINVAL,
1372                                                 RTE_FLOW_ERROR_TYPE_ACTION,
1373                                                 (void *)items->type,
1374                                                 "neither eth nor vlan"
1375                                                 " header found");
1376                         if (vlan && !vlan->eth_proto)
1377                                 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
1378                         else if (eth && !eth->ether_type)
1379                                 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
1380                         if (!ipv4->version_ihl)
1381                                 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
1382                                                     MLX5_ENCAP_IPV4_IHL_MIN;
1383                         if (!ipv4->time_to_live)
1384                                 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
1385                         break;
1386                 case RTE_FLOW_ITEM_TYPE_IPV6:
1387                         ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
1388                         if (!vlan && !eth)
1389                                 return rte_flow_error_set(error, EINVAL,
1390                                                 RTE_FLOW_ERROR_TYPE_ACTION,
1391                                                 (void *)items->type,
1392                                                 "neither eth nor vlan"
1393                                                 " header found");
1394                         if (vlan && !vlan->eth_proto)
1395                                 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
1396                         else if (eth && !eth->ether_type)
1397                                 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
1398                         if (!ipv6->vtc_flow)
1399                                 ipv6->vtc_flow =
1400                                         RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
1401                         if (!ipv6->hop_limits)
1402                                 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
1403                         break;
1404                 case RTE_FLOW_ITEM_TYPE_UDP:
1405                         udp = (struct rte_udp_hdr *)&buf[temp_size];
1406                         if (!ipv4 && !ipv6)
1407                                 return rte_flow_error_set(error, EINVAL,
1408                                                 RTE_FLOW_ERROR_TYPE_ACTION,
1409                                                 (void *)items->type,
1410                                                 "ip header not found");
1411                         if (ipv4 && !ipv4->next_proto_id)
1412                                 ipv4->next_proto_id = IPPROTO_UDP;
1413                         else if (ipv6 && !ipv6->proto)
1414                                 ipv6->proto = IPPROTO_UDP;
1415                         break;
1416                 case RTE_FLOW_ITEM_TYPE_VXLAN:
1417                         vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
1418                         if (!udp)
1419                                 return rte_flow_error_set(error, EINVAL,
1420                                                 RTE_FLOW_ERROR_TYPE_ACTION,
1421                                                 (void *)items->type,
1422                                                 "udp header not found");
1423                         if (!udp->dst_port)
1424                                 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
1425                         if (!vxlan->vx_flags)
1426                                 vxlan->vx_flags =
1427                                         RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
1428                         break;
1429                 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1430                         vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
1431                         if (!udp)
1432                                 return rte_flow_error_set(error, EINVAL,
1433                                                 RTE_FLOW_ERROR_TYPE_ACTION,
1434                                                 (void *)items->type,
1435                                                 "udp header not found");
1436                         if (!vxlan_gpe->proto)
1437                                 return rte_flow_error_set(error, EINVAL,
1438                                                 RTE_FLOW_ERROR_TYPE_ACTION,
1439                                                 (void *)items->type,
1440                                                 "next protocol not found");
1441                         if (!udp->dst_port)
1442                                 udp->dst_port =
1443                                         RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
1444                         if (!vxlan_gpe->vx_flags)
1445                                 vxlan_gpe->vx_flags =
1446                                                 MLX5_ENCAP_VXLAN_GPE_FLAGS;
1447                         break;
1448                 case RTE_FLOW_ITEM_TYPE_GRE:
1449                 case RTE_FLOW_ITEM_TYPE_NVGRE:
1450                         gre = (struct rte_gre_hdr *)&buf[temp_size];
1451                         if (!gre->proto)
1452                                 return rte_flow_error_set(error, EINVAL,
1453                                                 RTE_FLOW_ERROR_TYPE_ACTION,
1454                                                 (void *)items->type,
1455                                                 "next protocol not found");
1456                         if (!ipv4 && !ipv6)
1457                                 return rte_flow_error_set(error, EINVAL,
1458                                                 RTE_FLOW_ERROR_TYPE_ACTION,
1459                                                 (void *)items->type,
1460                                                 "ip header not found");
1461                         if (ipv4 && !ipv4->next_proto_id)
1462                                 ipv4->next_proto_id = IPPROTO_GRE;
1463                         else if (ipv6 && !ipv6->proto)
1464                                 ipv6->proto = IPPROTO_GRE;
1465                         break;
1466                 case RTE_FLOW_ITEM_TYPE_VOID:
1467                         break;
1468                 default:
1469                         return rte_flow_error_set(error, EINVAL,
1470                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1471                                                   (void *)items->type,
1472                                                   "unsupported item type");
1473                         break;
1474                 }
1475                 temp_size += len;
1476         }
1477         *size = temp_size;
1478         return 0;
1479 }
1480
1481 static int
1482 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
1483 {
1484         struct rte_ether_hdr *eth = NULL;
1485         struct rte_vlan_hdr *vlan = NULL;
1486         struct rte_ipv6_hdr *ipv6 = NULL;
1487         struct rte_udp_hdr *udp = NULL;
1488         char *next_hdr;
1489         uint16_t proto;
1490
1491         eth = (struct rte_ether_hdr *)data;
1492         next_hdr = (char *)(eth + 1);
1493         proto = RTE_BE16(eth->ether_type);
1494
1495         /* VLAN skipping */
1496         while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
1497                 vlan = (struct rte_vlan_hdr *)next_hdr;
1498                 proto = RTE_BE16(vlan->eth_proto);
1499                 next_hdr += sizeof(struct rte_vlan_hdr);
1500         }
1501
1502         /* HW calculates IPv4 csum. no need to proceed */
1503         if (proto == RTE_ETHER_TYPE_IPV4)
1504                 return 0;
1505
1506         /* non IPv4/IPv6 header. not supported */
1507         if (proto != RTE_ETHER_TYPE_IPV6) {
1508                 return rte_flow_error_set(error, ENOTSUP,
1509                                           RTE_FLOW_ERROR_TYPE_ACTION,
1510                                           NULL, "Cannot offload non IPv4/IPv6");
1511         }
1512
1513         ipv6 = (struct rte_ipv6_hdr *)next_hdr;
1514
1515         /* ignore non UDP */
1516         if (ipv6->proto != IPPROTO_UDP)
1517                 return 0;
1518
1519         udp = (struct rte_udp_hdr *)(ipv6 + 1);
1520         udp->dgram_cksum = 0;
1521
1522         return 0;
1523 }
1524
1525 /**
1526  * Convert L2 encap action to DV specification.
1527  *
1528  * @param[in] dev
1529  *   Pointer to rte_eth_dev structure.
1530  * @param[in] action
1531  *   Pointer to action structure.
1532  * @param[in, out] dev_flow
1533  *   Pointer to the mlx5_flow.
1534  * @param[in] transfer
1535  *   Mark if the flow is E-Switch flow.
1536  * @param[out] error
1537  *   Pointer to the error structure.
1538  *
1539  * @return
1540  *   0 on success, a negative errno value otherwise and rte_errno is set.
1541  */
1542 static int
1543 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
1544                                const struct rte_flow_action *action,
1545                                struct mlx5_flow *dev_flow,
1546                                uint8_t transfer,
1547                                struct rte_flow_error *error)
1548 {
1549         const struct rte_flow_item *encap_data;
1550         const struct rte_flow_action_raw_encap *raw_encap_data;
1551         struct mlx5_flow_dv_encap_decap_resource res = {
1552                 .reformat_type =
1553                         MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
1554                 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
1555                                       MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
1556         };
1557
1558         if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
1559                 raw_encap_data =
1560                         (const struct rte_flow_action_raw_encap *)action->conf;
1561                 res.size = raw_encap_data->size;
1562                 memcpy(res.buf, raw_encap_data->data, res.size);
1563                 if (flow_dv_zero_encap_udp_csum(res.buf, error))
1564                         return -rte_errno;
1565         } else {
1566                 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
1567                         encap_data =
1568                                 ((const struct rte_flow_action_vxlan_encap *)
1569                                                 action->conf)->definition;
1570                 else
1571                         encap_data =
1572                                 ((const struct rte_flow_action_nvgre_encap *)
1573                                                 action->conf)->definition;
1574                 if (flow_dv_convert_encap_data(encap_data, res.buf,
1575                                                &res.size, error))
1576                         return -rte_errno;
1577         }
1578         if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
1579                 return rte_flow_error_set(error, EINVAL,
1580                                           RTE_FLOW_ERROR_TYPE_ACTION,
1581                                           NULL, "can't create L2 encap action");
1582         return 0;
1583 }
1584
1585 /**
1586  * Convert L2 decap action to DV specification.
1587  *
1588  * @param[in] dev
1589  *   Pointer to rte_eth_dev structure.
1590  * @param[in, out] dev_flow
1591  *   Pointer to the mlx5_flow.
1592  * @param[in] transfer
1593  *   Mark if the flow is E-Switch flow.
1594  * @param[out] error
1595  *   Pointer to the error structure.
1596  *
1597  * @return
1598  *   0 on success, a negative errno value otherwise and rte_errno is set.
1599  */
1600 static int
1601 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
1602                                struct mlx5_flow *dev_flow,
1603                                uint8_t transfer,
1604                                struct rte_flow_error *error)
1605 {
1606         struct mlx5_flow_dv_encap_decap_resource res = {
1607                 .size = 0,
1608                 .reformat_type =
1609                         MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
1610                 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
1611                                       MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
1612         };
1613
1614         if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
1615                 return rte_flow_error_set(error, EINVAL,
1616                                           RTE_FLOW_ERROR_TYPE_ACTION,
1617                                           NULL, "can't create L2 decap action");
1618         return 0;
1619 }
1620
1621 /**
1622  * Convert raw decap/encap (L3 tunnel) action to DV specification.
1623  *
1624  * @param[in] dev
1625  *   Pointer to rte_eth_dev structure.
1626  * @param[in] action
1627  *   Pointer to action structure.
1628  * @param[in, out] dev_flow
1629  *   Pointer to the mlx5_flow.
1630  * @param[in] attr
1631  *   Pointer to the flow attributes.
1632  * @param[out] error
1633  *   Pointer to the error structure.
1634  *
1635  * @return
1636  *   0 on success, a negative errno value otherwise and rte_errno is set.
1637  */
1638 static int
1639 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
1640                                 const struct rte_flow_action *action,
1641                                 struct mlx5_flow *dev_flow,
1642                                 const struct rte_flow_attr *attr,
1643                                 struct rte_flow_error *error)
1644 {
1645         const struct rte_flow_action_raw_encap *encap_data;
1646         struct mlx5_flow_dv_encap_decap_resource res;
1647
1648         encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
1649         res.size = encap_data->size;
1650         memcpy(res.buf, encap_data->data, res.size);
1651         res.reformat_type = attr->egress ?
1652                 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL :
1653                 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2;
1654         if (attr->transfer)
1655                 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
1656         else
1657                 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
1658                                              MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
1659         if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
1660                 return rte_flow_error_set(error, EINVAL,
1661                                           RTE_FLOW_ERROR_TYPE_ACTION,
1662                                           NULL, "can't create encap action");
1663         return 0;
1664 }
1665
1666 /**
1667  * Validate the modify-header actions.
1668  *
1669  * @param[in] action_flags
1670  *   Holds the actions detected until now.
1671  * @param[in] action
1672  *   Pointer to the modify action.
1673  * @param[out] error
1674  *   Pointer to error structure.
1675  *
1676  * @return
1677  *   0 on success, a negative errno value otherwise and rte_errno is set.
1678  */
1679 static int
1680 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
1681                                    const struct rte_flow_action *action,
1682                                    struct rte_flow_error *error)
1683 {
1684         if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
1685                 return rte_flow_error_set(error, EINVAL,
1686                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1687                                           NULL, "action configuration not set");
1688         if (action_flags & MLX5_FLOW_ENCAP_ACTIONS)
1689                 return rte_flow_error_set(error, EINVAL,
1690                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1691                                           "can't have encap action before"
1692                                           " modify action");
1693         return 0;
1694 }
1695
1696 /**
1697  * Validate the modify-header MAC address actions.
1698  *
1699  * @param[in] action_flags
1700  *   Holds the actions detected until now.
1701  * @param[in] action
1702  *   Pointer to the modify action.
1703  * @param[in] item_flags
1704  *   Holds the items detected.
1705  * @param[out] error
1706  *   Pointer to error structure.
1707  *
1708  * @return
1709  *   0 on success, a negative errno value otherwise and rte_errno is set.
1710  */
1711 static int
1712 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
1713                                    const struct rte_flow_action *action,
1714                                    const uint64_t item_flags,
1715                                    struct rte_flow_error *error)
1716 {
1717         int ret = 0;
1718
1719         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1720         if (!ret) {
1721                 if (!(item_flags & MLX5_FLOW_LAYER_L2))
1722                         return rte_flow_error_set(error, EINVAL,
1723                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1724                                                   NULL,
1725                                                   "no L2 item in pattern");
1726         }
1727         return ret;
1728 }
1729
1730 /**
1731  * Validate the modify-header IPv4 address actions.
1732  *
1733  * @param[in] action_flags
1734  *   Holds the actions detected until now.
1735  * @param[in] action
1736  *   Pointer to the modify action.
1737  * @param[in] item_flags
1738  *   Holds the items detected.
1739  * @param[out] error
1740  *   Pointer to error structure.
1741  *
1742  * @return
1743  *   0 on success, a negative errno value otherwise and rte_errno is set.
1744  */
1745 static int
1746 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
1747                                     const struct rte_flow_action *action,
1748                                     const uint64_t item_flags,
1749                                     struct rte_flow_error *error)
1750 {
1751         int ret = 0;
1752
1753         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1754         if (!ret) {
1755                 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
1756                         return rte_flow_error_set(error, EINVAL,
1757                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1758                                                   NULL,
1759                                                   "no ipv4 item in pattern");
1760         }
1761         return ret;
1762 }
1763
1764 /**
1765  * Validate the modify-header IPv6 address actions.
1766  *
1767  * @param[in] action_flags
1768  *   Holds the actions detected until now.
1769  * @param[in] action
1770  *   Pointer to the modify action.
1771  * @param[in] item_flags
1772  *   Holds the items detected.
1773  * @param[out] error
1774  *   Pointer to error structure.
1775  *
1776  * @return
1777  *   0 on success, a negative errno value otherwise and rte_errno is set.
1778  */
1779 static int
1780 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
1781                                     const struct rte_flow_action *action,
1782                                     const uint64_t item_flags,
1783                                     struct rte_flow_error *error)
1784 {
1785         int ret = 0;
1786
1787         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1788         if (!ret) {
1789                 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
1790                         return rte_flow_error_set(error, EINVAL,
1791                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1792                                                   NULL,
1793                                                   "no ipv6 item in pattern");
1794         }
1795         return ret;
1796 }
1797
1798 /**
1799  * Validate the modify-header TP actions.
1800  *
1801  * @param[in] action_flags
1802  *   Holds the actions detected until now.
1803  * @param[in] action
1804  *   Pointer to the modify action.
1805  * @param[in] item_flags
1806  *   Holds the items detected.
1807  * @param[out] error
1808  *   Pointer to error structure.
1809  *
1810  * @return
1811  *   0 on success, a negative errno value otherwise and rte_errno is set.
1812  */
1813 static int
1814 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
1815                                   const struct rte_flow_action *action,
1816                                   const uint64_t item_flags,
1817                                   struct rte_flow_error *error)
1818 {
1819         int ret = 0;
1820
1821         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1822         if (!ret) {
1823                 if (!(item_flags & MLX5_FLOW_LAYER_L4))
1824                         return rte_flow_error_set(error, EINVAL,
1825                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1826                                                   NULL, "no transport layer "
1827                                                   "in pattern");
1828         }
1829         return ret;
1830 }
1831
1832 /**
1833  * Validate the modify-header actions of increment/decrement
1834  * TCP Sequence-number.
1835  *
1836  * @param[in] action_flags
1837  *   Holds the actions detected until now.
1838  * @param[in] action
1839  *   Pointer to the modify action.
1840  * @param[in] item_flags
1841  *   Holds the items detected.
1842  * @param[out] error
1843  *   Pointer to error structure.
1844  *
1845  * @return
1846  *   0 on success, a negative errno value otherwise and rte_errno is set.
1847  */
1848 static int
1849 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
1850                                        const struct rte_flow_action *action,
1851                                        const uint64_t item_flags,
1852                                        struct rte_flow_error *error)
1853 {
1854         int ret = 0;
1855
1856         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1857         if (!ret) {
1858                 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP))
1859                         return rte_flow_error_set(error, EINVAL,
1860                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1861                                                   NULL, "no TCP item in"
1862                                                   " pattern");
1863                 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
1864                         (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
1865                     (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
1866                         (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
1867                         return rte_flow_error_set(error, EINVAL,
1868                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1869                                                   NULL,
1870                                                   "cannot decrease and increase"
1871                                                   " TCP sequence number"
1872                                                   " at the same time");
1873         }
1874         return ret;
1875 }
1876
1877 /**
1878  * Validate the modify-header actions of increment/decrement
1879  * TCP Acknowledgment number.
1880  *
1881  * @param[in] action_flags
1882  *   Holds the actions detected until now.
1883  * @param[in] action
1884  *   Pointer to the modify action.
1885  * @param[in] item_flags
1886  *   Holds the items detected.
1887  * @param[out] error
1888  *   Pointer to error structure.
1889  *
1890  * @return
1891  *   0 on success, a negative errno value otherwise and rte_errno is set.
1892  */
1893 static int
1894 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
1895                                        const struct rte_flow_action *action,
1896                                        const uint64_t item_flags,
1897                                        struct rte_flow_error *error)
1898 {
1899         int ret = 0;
1900
1901         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1902         if (!ret) {
1903                 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP))
1904                         return rte_flow_error_set(error, EINVAL,
1905                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1906                                                   NULL, "no TCP item in"
1907                                                   " pattern");
1908                 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
1909                         (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
1910                     (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
1911                         (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
1912                         return rte_flow_error_set(error, EINVAL,
1913                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1914                                                   NULL,
1915                                                   "cannot decrease and increase"
1916                                                   " TCP acknowledgment number"
1917                                                   " at the same time");
1918         }
1919         return ret;
1920 }
1921
1922 /**
1923  * Validate the modify-header TTL actions.
1924  *
1925  * @param[in] action_flags
1926  *   Holds the actions detected until now.
1927  * @param[in] action
1928  *   Pointer to the modify action.
1929  * @param[in] item_flags
1930  *   Holds the items detected.
1931  * @param[out] error
1932  *   Pointer to error structure.
1933  *
1934  * @return
1935  *   0 on success, a negative errno value otherwise and rte_errno is set.
1936  */
1937 static int
1938 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
1939                                    const struct rte_flow_action *action,
1940                                    const uint64_t item_flags,
1941                                    struct rte_flow_error *error)
1942 {
1943         int ret = 0;
1944
1945         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1946         if (!ret) {
1947                 if (!(item_flags & MLX5_FLOW_LAYER_L3))
1948                         return rte_flow_error_set(error, EINVAL,
1949                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1950                                                   NULL,
1951                                                   "no IP protocol in pattern");
1952         }
1953         return ret;
1954 }
1955
1956 /**
1957  * Validate jump action.
1958  *
1959  * @param[in] action
1960  *   Pointer to the jump action.
1961  * @param[in] action_flags
1962  *   Holds the actions detected until now.
1963  * @param[in] group
1964  *   The group of the current flow.
1965  * @param[out] error
1966  *   Pointer to error structure.
1967  *
1968  * @return
1969  *   0 on success, a negative errno value otherwise and rte_errno is set.
1970  */
1971 static int
1972 flow_dv_validate_action_jump(const struct rte_flow_action *action,
1973                              uint64_t action_flags,
1974                              uint32_t group,
1975                              struct rte_flow_error *error)
1976 {
1977         if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
1978                             MLX5_FLOW_FATE_ESWITCH_ACTIONS))
1979                 return rte_flow_error_set(error, EINVAL,
1980                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1981                                           "can't have 2 fate actions in"
1982                                           " same flow");
1983         if (!action->conf)
1984                 return rte_flow_error_set(error, EINVAL,
1985                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1986                                           NULL, "action configuration not set");
1987         if (group >= ((const struct rte_flow_action_jump *)action->conf)->group)
1988                 return rte_flow_error_set(error, EINVAL,
1989                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1990                                           "target group must be higher then"
1991                                           " the current flow group");
1992         return 0;
1993 }
1994
1995 /*
1996  * Validate the port_id action.
1997  *
1998  * @param[in] dev
1999  *   Pointer to rte_eth_dev structure.
2000  * @param[in] action_flags
2001  *   Bit-fields that holds the actions detected until now.
2002  * @param[in] action
2003  *   Port_id RTE action structure.
2004  * @param[in] attr
2005  *   Attributes of flow that includes this action.
2006  * @param[out] error
2007  *   Pointer to error structure.
2008  *
2009  * @return
2010  *   0 on success, a negative errno value otherwise and rte_errno is set.
2011  */
2012 static int
2013 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
2014                                 uint64_t action_flags,
2015                                 const struct rte_flow_action *action,
2016                                 const struct rte_flow_attr *attr,
2017                                 struct rte_flow_error *error)
2018 {
2019         const struct rte_flow_action_port_id *port_id;
2020         uint16_t port;
2021         uint16_t esw_domain_id;
2022         uint16_t act_port_domain_id;
2023         int ret;
2024
2025         if (!attr->transfer)
2026                 return rte_flow_error_set(error, ENOTSUP,
2027                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2028                                           NULL,
2029                                           "port id action is valid in transfer"
2030                                           " mode only");
2031         if (!action || !action->conf)
2032                 return rte_flow_error_set(error, ENOTSUP,
2033                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2034                                           NULL,
2035                                           "port id action parameters must be"
2036                                           " specified");
2037         if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
2038                             MLX5_FLOW_FATE_ESWITCH_ACTIONS))
2039                 return rte_flow_error_set(error, EINVAL,
2040                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2041                                           "can have only one fate actions in"
2042                                           " a flow");
2043         ret = mlx5_port_to_eswitch_info(dev->data->port_id,
2044                                         &esw_domain_id, NULL);
2045         if (ret < 0)
2046                 return rte_flow_error_set(error, -ret,
2047                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2048                                           NULL,
2049                                           "failed to obtain E-Switch info");
2050         port_id = action->conf;
2051         port = port_id->original ? dev->data->port_id : port_id->id;
2052         ret = mlx5_port_to_eswitch_info(port, &act_port_domain_id, NULL);
2053         if (ret)
2054                 return rte_flow_error_set
2055                                 (error, -ret,
2056                                  RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
2057                                  "failed to obtain E-Switch port id for port");
2058         if (act_port_domain_id != esw_domain_id)
2059                 return rte_flow_error_set
2060                                 (error, -ret,
2061                                  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2062                                  "port does not belong to"
2063                                  " E-Switch being configured");
2064         return 0;
2065 }
2066
2067 /**
2068  * Find existing modify-header resource or create and register a new one.
2069  *
2070  * @param dev[in, out]
2071  *   Pointer to rte_eth_dev structure.
2072  * @param[in, out] resource
2073  *   Pointer to modify-header resource.
2074  * @parm[in, out] dev_flow
2075  *   Pointer to the dev_flow.
2076  * @param[out] error
2077  *   pointer to error structure.
2078  *
2079  * @return
2080  *   0 on success otherwise -errno and errno is set.
2081  */
2082 static int
2083 flow_dv_modify_hdr_resource_register
2084                         (struct rte_eth_dev *dev,
2085                          struct mlx5_flow_dv_modify_hdr_resource *resource,
2086                          struct mlx5_flow *dev_flow,
2087                          struct rte_flow_error *error)
2088 {
2089         struct mlx5_priv *priv = dev->data->dev_private;
2090         struct mlx5_ibv_shared *sh = priv->sh;
2091         struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
2092         struct mlx5dv_dr_domain *ns;
2093
2094         if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2095                 ns = sh->fdb_domain;
2096         else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
2097                 ns = sh->tx_domain;
2098         else
2099                 ns = sh->rx_domain;
2100         resource->flags =
2101                 dev_flow->flow->group ? 0 : MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
2102         /* Lookup a matching resource from cache. */
2103         LIST_FOREACH(cache_resource, &sh->modify_cmds, next) {
2104                 if (resource->ft_type == cache_resource->ft_type &&
2105                     resource->actions_num == cache_resource->actions_num &&
2106                     resource->flags == cache_resource->flags &&
2107                     !memcmp((const void *)resource->actions,
2108                             (const void *)cache_resource->actions,
2109                             (resource->actions_num *
2110                                             sizeof(resource->actions[0])))) {
2111                         DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
2112                                 (void *)cache_resource,
2113                                 rte_atomic32_read(&cache_resource->refcnt));
2114                         rte_atomic32_inc(&cache_resource->refcnt);
2115                         dev_flow->dv.modify_hdr = cache_resource;
2116                         return 0;
2117                 }
2118         }
2119         /* Register new modify-header resource. */
2120         cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
2121         if (!cache_resource)
2122                 return rte_flow_error_set(error, ENOMEM,
2123                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2124                                           "cannot allocate resource memory");
2125         *cache_resource = *resource;
2126         cache_resource->verbs_action =
2127                 mlx5_glue->dv_create_flow_action_modify_header
2128                                         (sh->ctx, cache_resource->ft_type,
2129                                          ns, cache_resource->flags,
2130                                          cache_resource->actions_num *
2131                                          sizeof(cache_resource->actions[0]),
2132                                          (uint64_t *)cache_resource->actions);
2133         if (!cache_resource->verbs_action) {
2134                 rte_free(cache_resource);
2135                 return rte_flow_error_set(error, ENOMEM,
2136                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2137                                           NULL, "cannot create action");
2138         }
2139         rte_atomic32_init(&cache_resource->refcnt);
2140         rte_atomic32_inc(&cache_resource->refcnt);
2141         LIST_INSERT_HEAD(&sh->modify_cmds, cache_resource, next);
2142         dev_flow->dv.modify_hdr = cache_resource;
2143         DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
2144                 (void *)cache_resource,
2145                 rte_atomic32_read(&cache_resource->refcnt));
2146         return 0;
2147 }
2148
2149 #define MLX5_CNT_CONTAINER_RESIZE 64
2150
2151 /**
2152  * Get or create a flow counter.
2153  *
2154  * @param[in] dev
2155  *   Pointer to the Ethernet device structure.
2156  * @param[in] shared
2157  *   Indicate if this counter is shared with other flows.
2158  * @param[in] id
2159  *   Counter identifier.
2160  *
2161  * @return
2162  *   pointer to flow counter on success, NULL otherwise and rte_errno is set.
2163  */
2164 static struct mlx5_flow_counter *
2165 flow_dv_counter_alloc_fallback(struct rte_eth_dev *dev, uint32_t shared,
2166                                uint32_t id)
2167 {
2168         struct mlx5_priv *priv = dev->data->dev_private;
2169         struct mlx5_flow_counter *cnt = NULL;
2170         struct mlx5_devx_obj *dcs = NULL;
2171
2172         if (!priv->config.devx) {
2173                 rte_errno = ENOTSUP;
2174                 return NULL;
2175         }
2176         if (shared) {
2177                 TAILQ_FOREACH(cnt, &priv->sh->cmng.flow_counters, next) {
2178                         if (cnt->shared && cnt->id == id) {
2179                                 cnt->ref_cnt++;
2180                                 return cnt;
2181                         }
2182                 }
2183         }
2184         dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
2185         if (!dcs)
2186                 return NULL;
2187         cnt = rte_calloc(__func__, 1, sizeof(*cnt), 0);
2188         if (!cnt) {
2189                 claim_zero(mlx5_devx_cmd_destroy(cnt->dcs));
2190                 rte_errno = ENOMEM;
2191                 return NULL;
2192         }
2193         struct mlx5_flow_counter tmpl = {
2194                 .shared = shared,
2195                 .ref_cnt = 1,
2196                 .id = id,
2197                 .dcs = dcs,
2198         };
2199         tmpl.action = mlx5_glue->dv_create_flow_action_counter(dcs->obj, 0);
2200         if (!tmpl.action) {
2201                 claim_zero(mlx5_devx_cmd_destroy(cnt->dcs));
2202                 rte_errno = errno;
2203                 rte_free(cnt);
2204                 return NULL;
2205         }
2206         *cnt = tmpl;
2207         TAILQ_INSERT_HEAD(&priv->sh->cmng.flow_counters, cnt, next);
2208         return cnt;
2209 }
2210
2211 /**
2212  * Release a flow counter.
2213  *
2214  * @param[in] dev
2215  *   Pointer to the Ethernet device structure.
2216  * @param[in] counter
2217  *   Pointer to the counter handler.
2218  */
2219 static void
2220 flow_dv_counter_release_fallback(struct rte_eth_dev *dev,
2221                                  struct mlx5_flow_counter *counter)
2222 {
2223         struct mlx5_priv *priv = dev->data->dev_private;
2224
2225         if (!counter)
2226                 return;
2227         if (--counter->ref_cnt == 0) {
2228                 TAILQ_REMOVE(&priv->sh->cmng.flow_counters, counter, next);
2229                 claim_zero(mlx5_devx_cmd_destroy(counter->dcs));
2230                 rte_free(counter);
2231         }
2232 }
2233
2234 /**
2235  * Query a devx flow counter.
2236  *
2237  * @param[in] dev
2238  *   Pointer to the Ethernet device structure.
2239  * @param[in] cnt
2240  *   Pointer to the flow counter.
2241  * @param[out] pkts
2242  *   The statistics value of packets.
2243  * @param[out] bytes
2244  *   The statistics value of bytes.
2245  *
2246  * @return
2247  *   0 on success, otherwise a negative errno value and rte_errno is set.
2248  */
2249 static inline int
2250 _flow_dv_query_count_fallback(struct rte_eth_dev *dev __rte_unused,
2251                      struct mlx5_flow_counter *cnt, uint64_t *pkts,
2252                      uint64_t *bytes)
2253 {
2254         return mlx5_devx_cmd_flow_counter_query(cnt->dcs, 0, 0, pkts, bytes,
2255                                                 0, NULL, NULL, 0);
2256 }
2257
2258 /**
2259  * Get a pool by a counter.
2260  *
2261  * @param[in] cnt
2262  *   Pointer to the counter.
2263  *
2264  * @return
2265  *   The counter pool.
2266  */
2267 static struct mlx5_flow_counter_pool *
2268 flow_dv_counter_pool_get(struct mlx5_flow_counter *cnt)
2269 {
2270         if (!cnt->batch) {
2271                 cnt -= cnt->dcs->id % MLX5_COUNTERS_PER_POOL;
2272                 return (struct mlx5_flow_counter_pool *)cnt - 1;
2273         }
2274         return cnt->pool;
2275 }
2276
2277 /**
2278  * Get a pool by devx counter ID.
2279  *
2280  * @param[in] cont
2281  *   Pointer to the counter container.
2282  * @param[in] id
2283  *   The counter devx ID.
2284  *
2285  * @return
2286  *   The counter pool pointer if exists, NULL otherwise,
2287  */
2288 static struct mlx5_flow_counter_pool *
2289 flow_dv_find_pool_by_id(struct mlx5_pools_container *cont, int id)
2290 {
2291         struct mlx5_flow_counter_pool *pool;
2292
2293         TAILQ_FOREACH(pool, &cont->pool_list, next) {
2294                 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
2295                                 MLX5_COUNTERS_PER_POOL;
2296
2297                 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
2298                         return pool;
2299         };
2300         return NULL;
2301 }
2302
2303 /**
2304  * Allocate a new memory for the counter values wrapped by all the needed
2305  * management.
2306  *
2307  * @param[in] dev
2308  *   Pointer to the Ethernet device structure.
2309  * @param[in] raws_n
2310  *   The raw memory areas - each one for MLX5_COUNTERS_PER_POOL counters.
2311  *
2312  * @return
2313  *   The new memory management pointer on success, otherwise NULL and rte_errno
2314  *   is set.
2315  */
2316 static struct mlx5_counter_stats_mem_mng *
2317 flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
2318 {
2319         struct mlx5_ibv_shared *sh = ((struct mlx5_priv *)
2320                                         (dev->data->dev_private))->sh;
2321         struct mlx5_devx_mkey_attr mkey_attr;
2322         struct mlx5_counter_stats_mem_mng *mem_mng;
2323         volatile struct flow_counter_stats *raw_data;
2324         int size = (sizeof(struct flow_counter_stats) *
2325                         MLX5_COUNTERS_PER_POOL +
2326                         sizeof(struct mlx5_counter_stats_raw)) * raws_n +
2327                         sizeof(struct mlx5_counter_stats_mem_mng);
2328         uint8_t *mem = rte_calloc(__func__, 1, size, sysconf(_SC_PAGESIZE));
2329         int i;
2330
2331         if (!mem) {
2332                 rte_errno = ENOMEM;
2333                 return NULL;
2334         }
2335         mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
2336         size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
2337         mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
2338                                                  IBV_ACCESS_LOCAL_WRITE);
2339         if (!mem_mng->umem) {
2340                 rte_errno = errno;
2341                 rte_free(mem);
2342                 return NULL;
2343         }
2344         mkey_attr.addr = (uintptr_t)mem;
2345         mkey_attr.size = size;
2346         mkey_attr.umem_id = mem_mng->umem->umem_id;
2347         mkey_attr.pd = sh->pdn;
2348         mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
2349         if (!mem_mng->dm) {
2350                 mlx5_glue->devx_umem_dereg(mem_mng->umem);
2351                 rte_errno = errno;
2352                 rte_free(mem);
2353                 return NULL;
2354         }
2355         mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
2356         raw_data = (volatile struct flow_counter_stats *)mem;
2357         for (i = 0; i < raws_n; ++i) {
2358                 mem_mng->raws[i].mem_mng = mem_mng;
2359                 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
2360         }
2361         LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
2362         return mem_mng;
2363 }
2364
2365 /**
2366  * Resize a counter container.
2367  *
2368  * @param[in] dev
2369  *   Pointer to the Ethernet device structure.
2370  * @param[in] batch
2371  *   Whether the pool is for counter that was allocated by batch command.
2372  *
2373  * @return
2374  *   The new container pointer on success, otherwise NULL and rte_errno is set.
2375  */
2376 static struct mlx5_pools_container *
2377 flow_dv_container_resize(struct rte_eth_dev *dev, uint32_t batch)
2378 {
2379         struct mlx5_priv *priv = dev->data->dev_private;
2380         struct mlx5_pools_container *cont =
2381                         MLX5_CNT_CONTAINER(priv->sh, batch, 0);
2382         struct mlx5_pools_container *new_cont =
2383                         MLX5_CNT_CONTAINER_UNUSED(priv->sh, batch, 0);
2384         struct mlx5_counter_stats_mem_mng *mem_mng;
2385         uint32_t resize = cont->n + MLX5_CNT_CONTAINER_RESIZE;
2386         uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
2387         int i;
2388
2389         if (cont != MLX5_CNT_CONTAINER(priv->sh, batch, 1)) {
2390                 /* The last resize still hasn't detected by the host thread. */
2391                 rte_errno = EAGAIN;
2392                 return NULL;
2393         }
2394         new_cont->pools = rte_calloc(__func__, 1, mem_size, 0);
2395         if (!new_cont->pools) {
2396                 rte_errno = ENOMEM;
2397                 return NULL;
2398         }
2399         if (cont->n)
2400                 memcpy(new_cont->pools, cont->pools, cont->n *
2401                        sizeof(struct mlx5_flow_counter_pool *));
2402         mem_mng = flow_dv_create_counter_stat_mem_mng(dev,
2403                 MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES);
2404         if (!mem_mng) {
2405                 rte_free(new_cont->pools);
2406                 return NULL;
2407         }
2408         for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
2409                 LIST_INSERT_HEAD(&priv->sh->cmng.free_stat_raws,
2410                                  mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE +
2411                                  i, next);
2412         new_cont->n = resize;
2413         rte_atomic16_set(&new_cont->n_valid, rte_atomic16_read(&cont->n_valid));
2414         TAILQ_INIT(&new_cont->pool_list);
2415         TAILQ_CONCAT(&new_cont->pool_list, &cont->pool_list, next);
2416         new_cont->init_mem_mng = mem_mng;
2417         rte_cio_wmb();
2418          /* Flip the master container. */
2419         priv->sh->cmng.mhi[batch] ^= (uint8_t)1;
2420         return new_cont;
2421 }
2422
2423 /**
2424  * Query a devx flow counter.
2425  *
2426  * @param[in] dev
2427  *   Pointer to the Ethernet device structure.
2428  * @param[in] cnt
2429  *   Pointer to the flow counter.
2430  * @param[out] pkts
2431  *   The statistics value of packets.
2432  * @param[out] bytes
2433  *   The statistics value of bytes.
2434  *
2435  * @return
2436  *   0 on success, otherwise a negative errno value and rte_errno is set.
2437  */
2438 static inline int
2439 _flow_dv_query_count(struct rte_eth_dev *dev,
2440                      struct mlx5_flow_counter *cnt, uint64_t *pkts,
2441                      uint64_t *bytes)
2442 {
2443         struct mlx5_priv *priv = dev->data->dev_private;
2444         struct mlx5_flow_counter_pool *pool =
2445                         flow_dv_counter_pool_get(cnt);
2446         int offset = cnt - &pool->counters_raw[0];
2447
2448         if (priv->counter_fallback)
2449                 return _flow_dv_query_count_fallback(dev, cnt, pkts, bytes);
2450
2451         rte_spinlock_lock(&pool->sl);
2452         /*
2453          * The single counters allocation may allocate smaller ID than the
2454          * current allocated in parallel to the host reading.
2455          * In this case the new counter values must be reported as 0.
2456          */
2457         if (unlikely(!cnt->batch && cnt->dcs->id < pool->raw->min_dcs_id)) {
2458                 *pkts = 0;
2459                 *bytes = 0;
2460         } else {
2461                 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
2462                 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
2463         }
2464         rte_spinlock_unlock(&pool->sl);
2465         return 0;
2466 }
2467
2468 /**
2469  * Create and initialize a new counter pool.
2470  *
2471  * @param[in] dev
2472  *   Pointer to the Ethernet device structure.
2473  * @param[out] dcs
2474  *   The devX counter handle.
2475  * @param[in] batch
2476  *   Whether the pool is for counter that was allocated by batch command.
2477  *
2478  * @return
2479  *   A new pool pointer on success, NULL otherwise and rte_errno is set.
2480  */
2481 static struct mlx5_flow_counter_pool *
2482 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
2483                     uint32_t batch)
2484 {
2485         struct mlx5_priv *priv = dev->data->dev_private;
2486         struct mlx5_flow_counter_pool *pool;
2487         struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
2488                                                                0);
2489         int16_t n_valid = rte_atomic16_read(&cont->n_valid);
2490         uint32_t size;
2491
2492         if (cont->n == n_valid) {
2493                 cont = flow_dv_container_resize(dev, batch);
2494                 if (!cont)
2495                         return NULL;
2496         }
2497         size = sizeof(*pool) + MLX5_COUNTERS_PER_POOL *
2498                         sizeof(struct mlx5_flow_counter);
2499         pool = rte_calloc(__func__, 1, size, 0);
2500         if (!pool) {
2501                 rte_errno = ENOMEM;
2502                 return NULL;
2503         }
2504         pool->min_dcs = dcs;
2505         pool->raw = cont->init_mem_mng->raws + n_valid %
2506                                                      MLX5_CNT_CONTAINER_RESIZE;
2507         pool->raw_hw = NULL;
2508         rte_spinlock_init(&pool->sl);
2509         /*
2510          * The generation of the new allocated counters in this pool is 0, 2 in
2511          * the pool generation makes all the counters valid for allocation.
2512          */
2513         rte_atomic64_set(&pool->query_gen, 0x2);
2514         TAILQ_INIT(&pool->counters);
2515         TAILQ_INSERT_TAIL(&cont->pool_list, pool, next);
2516         cont->pools[n_valid] = pool;
2517         /* Pool initialization must be updated before host thread access. */
2518         rte_cio_wmb();
2519         rte_atomic16_add(&cont->n_valid, 1);
2520         return pool;
2521 }
2522
2523 /**
2524  * Prepare a new counter and/or a new counter pool.
2525  *
2526  * @param[in] dev
2527  *   Pointer to the Ethernet device structure.
2528  * @param[out] cnt_free
2529  *   Where to put the pointer of a new counter.
2530  * @param[in] batch
2531  *   Whether the pool is for counter that was allocated by batch command.
2532  *
2533  * @return
2534  *   The free counter pool pointer and @p cnt_free is set on success,
2535  *   NULL otherwise and rte_errno is set.
2536  */
2537 static struct mlx5_flow_counter_pool *
2538 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
2539                              struct mlx5_flow_counter **cnt_free,
2540                              uint32_t batch)
2541 {
2542         struct mlx5_priv *priv = dev->data->dev_private;
2543         struct mlx5_flow_counter_pool *pool;
2544         struct mlx5_devx_obj *dcs = NULL;
2545         struct mlx5_flow_counter *cnt;
2546         uint32_t i;
2547
2548         if (!batch) {
2549                 /* bulk_bitmap must be 0 for single counter allocation. */
2550                 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
2551                 if (!dcs)
2552                         return NULL;
2553                 pool = flow_dv_find_pool_by_id
2554                         (MLX5_CNT_CONTAINER(priv->sh, batch, 0), dcs->id);
2555                 if (!pool) {
2556                         pool = flow_dv_pool_create(dev, dcs, batch);
2557                         if (!pool) {
2558                                 mlx5_devx_cmd_destroy(dcs);
2559                                 return NULL;
2560                         }
2561                 } else if (dcs->id < pool->min_dcs->id) {
2562                         rte_atomic64_set(&pool->a64_dcs,
2563                                          (int64_t)(uintptr_t)dcs);
2564                 }
2565                 cnt = &pool->counters_raw[dcs->id % MLX5_COUNTERS_PER_POOL];
2566                 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
2567                 cnt->dcs = dcs;
2568                 *cnt_free = cnt;
2569                 return pool;
2570         }
2571         /* bulk_bitmap is in 128 counters units. */
2572         if (priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4)
2573                 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
2574         if (!dcs) {
2575                 rte_errno = ENODATA;
2576                 return NULL;
2577         }
2578         pool = flow_dv_pool_create(dev, dcs, batch);
2579         if (!pool) {
2580                 mlx5_devx_cmd_destroy(dcs);
2581                 return NULL;
2582         }
2583         for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
2584                 cnt = &pool->counters_raw[i];
2585                 cnt->pool = pool;
2586                 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
2587         }
2588         *cnt_free = &pool->counters_raw[0];
2589         return pool;
2590 }
2591
2592 /**
2593  * Search for existed shared counter.
2594  *
2595  * @param[in] cont
2596  *   Pointer to the relevant counter pool container.
2597  * @param[in] id
2598  *   The shared counter ID to search.
2599  *
2600  * @return
2601  *   NULL if not existed, otherwise pointer to the shared counter.
2602  */
2603 static struct mlx5_flow_counter *
2604 flow_dv_counter_shared_search(struct mlx5_pools_container *cont,
2605                               uint32_t id)
2606 {
2607         static struct mlx5_flow_counter *cnt;
2608         struct mlx5_flow_counter_pool *pool;
2609         int i;
2610
2611         TAILQ_FOREACH(pool, &cont->pool_list, next) {
2612                 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
2613                         cnt = &pool->counters_raw[i];
2614                         if (cnt->ref_cnt && cnt->shared && cnt->id == id)
2615                                 return cnt;
2616                 }
2617         }
2618         return NULL;
2619 }
2620
2621 /**
2622  * Allocate a flow counter.
2623  *
2624  * @param[in] dev
2625  *   Pointer to the Ethernet device structure.
2626  * @param[in] shared
2627  *   Indicate if this counter is shared with other flows.
2628  * @param[in] id
2629  *   Counter identifier.
2630  * @param[in] group
2631  *   Counter flow group.
2632  *
2633  * @return
2634  *   pointer to flow counter on success, NULL otherwise and rte_errno is set.
2635  */
2636 static struct mlx5_flow_counter *
2637 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t shared, uint32_t id,
2638                       uint16_t group)
2639 {
2640         struct mlx5_priv *priv = dev->data->dev_private;
2641         struct mlx5_flow_counter_pool *pool = NULL;
2642         struct mlx5_flow_counter *cnt_free = NULL;
2643         /*
2644          * Currently group 0 flow counter cannot be assigned to a flow if it is
2645          * not the first one in the batch counter allocation, so it is better
2646          * to allocate counters one by one for these flows in a separate
2647          * container.
2648          * A counter can be shared between different groups so need to take
2649          * shared counters from the single container.
2650          */
2651         uint32_t batch = (group && !shared) ? 1 : 0;
2652         struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
2653                                                                0);
2654
2655         if (priv->counter_fallback)
2656                 return flow_dv_counter_alloc_fallback(dev, shared, id);
2657         if (!priv->config.devx) {
2658                 rte_errno = ENOTSUP;
2659                 return NULL;
2660         }
2661         if (shared) {
2662                 cnt_free = flow_dv_counter_shared_search(cont, id);
2663                 if (cnt_free) {
2664                         if (cnt_free->ref_cnt + 1 == 0) {
2665                                 rte_errno = E2BIG;
2666                                 return NULL;
2667                         }
2668                         cnt_free->ref_cnt++;
2669                         return cnt_free;
2670                 }
2671         }
2672         /* Pools which has a free counters are in the start. */
2673         TAILQ_FOREACH(pool, &cont->pool_list, next) {
2674                 /*
2675                  * The free counter reset values must be updated between the
2676                  * counter release to the counter allocation, so, at least one
2677                  * query must be done in this time. ensure it by saving the
2678                  * query generation in the release time.
2679                  * The free list is sorted according to the generation - so if
2680                  * the first one is not updated, all the others are not
2681                  * updated too.
2682                  */
2683                 cnt_free = TAILQ_FIRST(&pool->counters);
2684                 if (cnt_free && cnt_free->query_gen + 1 <
2685                     rte_atomic64_read(&pool->query_gen))
2686                         break;
2687                 cnt_free = NULL;
2688         }
2689         if (!cnt_free) {
2690                 pool = flow_dv_counter_pool_prepare(dev, &cnt_free, batch);
2691                 if (!pool)
2692                         return NULL;
2693         }
2694         cnt_free->batch = batch;
2695         /* Create a DV counter action only in the first time usage. */
2696         if (!cnt_free->action) {
2697                 uint16_t offset;
2698                 struct mlx5_devx_obj *dcs;
2699
2700                 if (batch) {
2701                         offset = cnt_free - &pool->counters_raw[0];
2702                         dcs = pool->min_dcs;
2703                 } else {
2704                         offset = 0;
2705                         dcs = cnt_free->dcs;
2706                 }
2707                 cnt_free->action = mlx5_glue->dv_create_flow_action_counter
2708                                         (dcs->obj, offset);
2709                 if (!cnt_free->action) {
2710                         rte_errno = errno;
2711                         return NULL;
2712                 }
2713         }
2714         /* Update the counter reset values. */
2715         if (_flow_dv_query_count(dev, cnt_free, &cnt_free->hits,
2716                                  &cnt_free->bytes))
2717                 return NULL;
2718         cnt_free->shared = shared;
2719         cnt_free->ref_cnt = 1;
2720         cnt_free->id = id;
2721         if (!priv->sh->cmng.query_thread_on)
2722                 /* Start the asynchronous batch query by the host thread. */
2723                 mlx5_set_query_alarm(priv->sh);
2724         TAILQ_REMOVE(&pool->counters, cnt_free, next);
2725         if (TAILQ_EMPTY(&pool->counters)) {
2726                 /* Move the pool to the end of the container pool list. */
2727                 TAILQ_REMOVE(&cont->pool_list, pool, next);
2728                 TAILQ_INSERT_TAIL(&cont->pool_list, pool, next);
2729         }
2730         return cnt_free;
2731 }
2732
2733 /**
2734  * Release a flow counter.
2735  *
2736  * @param[in] dev
2737  *   Pointer to the Ethernet device structure.
2738  * @param[in] counter
2739  *   Pointer to the counter handler.
2740  */
2741 static void
2742 flow_dv_counter_release(struct rte_eth_dev *dev,
2743                         struct mlx5_flow_counter *counter)
2744 {
2745         struct mlx5_priv *priv = dev->data->dev_private;
2746
2747         if (!counter)
2748                 return;
2749         if (priv->counter_fallback) {
2750                 flow_dv_counter_release_fallback(dev, counter);
2751                 return;
2752         }
2753         if (--counter->ref_cnt == 0) {
2754                 struct mlx5_flow_counter_pool *pool =
2755                                 flow_dv_counter_pool_get(counter);
2756
2757                 /* Put the counter in the end - the last updated one. */
2758                 TAILQ_INSERT_TAIL(&pool->counters, counter, next);
2759                 counter->query_gen = rte_atomic64_read(&pool->query_gen);
2760         }
2761 }
2762
2763 /**
2764  * Verify the @p attributes will be correctly understood by the NIC and store
2765  * them in the @p flow if everything is correct.
2766  *
2767  * @param[in] dev
2768  *   Pointer to dev struct.
2769  * @param[in] attributes
2770  *   Pointer to flow attributes
2771  * @param[out] error
2772  *   Pointer to error structure.
2773  *
2774  * @return
2775  *   0 on success, a negative errno value otherwise and rte_errno is set.
2776  */
2777 static int
2778 flow_dv_validate_attributes(struct rte_eth_dev *dev,
2779                             const struct rte_flow_attr *attributes,
2780                             struct rte_flow_error *error)
2781 {
2782         struct mlx5_priv *priv = dev->data->dev_private;
2783         uint32_t priority_max = priv->config.flow_prio - 1;
2784
2785 #ifndef HAVE_MLX5DV_DR
2786         if (attributes->group)
2787                 return rte_flow_error_set(error, ENOTSUP,
2788                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
2789                                           NULL,
2790                                           "groups is not supported");
2791 #endif
2792         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
2793             attributes->priority >= priority_max)
2794                 return rte_flow_error_set(error, ENOTSUP,
2795                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
2796                                           NULL,
2797                                           "priority out of range");
2798         if (attributes->transfer) {
2799                 if (!priv->config.dv_esw_en)
2800                         return rte_flow_error_set
2801                                 (error, ENOTSUP,
2802                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2803                                  "E-Switch dr is not supported");
2804                 if (!(priv->representor || priv->master))
2805                         return rte_flow_error_set
2806                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2807                                  NULL, "E-Switch configuration can only be"
2808                                  " done by a master or a representor device");
2809                 if (attributes->egress)
2810                         return rte_flow_error_set
2811                                 (error, ENOTSUP,
2812                                  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
2813                                  "egress is not supported");
2814                 if (attributes->group >= MLX5_MAX_TABLES_FDB)
2815                         return rte_flow_error_set
2816                                 (error, EINVAL,
2817                                  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
2818                                  NULL, "group must be smaller than "
2819                                  RTE_STR(MLX5_MAX_TABLES_FDB));
2820         }
2821         if (!(attributes->egress ^ attributes->ingress))
2822                 return rte_flow_error_set(error, ENOTSUP,
2823                                           RTE_FLOW_ERROR_TYPE_ATTR, NULL,
2824                                           "must specify exactly one of "
2825                                           "ingress or egress");
2826         return 0;
2827 }
2828
2829 /**
2830  * Internal validation function. For validating both actions and items.
2831  *
2832  * @param[in] dev
2833  *   Pointer to the rte_eth_dev structure.
2834  * @param[in] attr
2835  *   Pointer to the flow attributes.
2836  * @param[in] items
2837  *   Pointer to the list of items.
2838  * @param[in] actions
2839  *   Pointer to the list of actions.
2840  * @param[out] error
2841  *   Pointer to the error structure.
2842  *
2843  * @return
2844  *   0 on success, a negative errno value otherwise and rte_errno is set.
2845  */
2846 static int
2847 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
2848                  const struct rte_flow_item items[],
2849                  const struct rte_flow_action actions[],
2850                  struct rte_flow_error *error)
2851 {
2852         int ret;
2853         uint64_t action_flags = 0;
2854         uint64_t item_flags = 0;
2855         uint64_t last_item = 0;
2856         uint8_t next_protocol = 0xff;
2857         int actions_n = 0;
2858         const struct rte_flow_item *gre_item = NULL;
2859         struct rte_flow_item_tcp nic_tcp_mask = {
2860                 .hdr = {
2861                         .tcp_flags = 0xFF,
2862                         .src_port = RTE_BE16(UINT16_MAX),
2863                         .dst_port = RTE_BE16(UINT16_MAX),
2864                 }
2865         };
2866
2867         if (items == NULL)
2868                 return -1;
2869         ret = flow_dv_validate_attributes(dev, attr, error);
2870         if (ret < 0)
2871                 return ret;
2872         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2873                 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2874                 switch (items->type) {
2875                 case RTE_FLOW_ITEM_TYPE_VOID:
2876                         break;
2877                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
2878                         ret = flow_dv_validate_item_port_id
2879                                         (dev, items, attr, item_flags, error);
2880                         if (ret < 0)
2881                                 return ret;
2882                         last_item = MLX5_FLOW_ITEM_PORT_ID;
2883                         break;
2884                 case RTE_FLOW_ITEM_TYPE_ETH:
2885                         ret = mlx5_flow_validate_item_eth(items, item_flags,
2886                                                           error);
2887                         if (ret < 0)
2888                                 return ret;
2889                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
2890                                              MLX5_FLOW_LAYER_OUTER_L2;
2891                         break;
2892                 case RTE_FLOW_ITEM_TYPE_VLAN:
2893                         ret = mlx5_flow_validate_item_vlan(items, item_flags,
2894                                                            dev, error);
2895                         if (ret < 0)
2896                                 return ret;
2897                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2898                                              MLX5_FLOW_LAYER_OUTER_VLAN;
2899                         break;
2900                 case RTE_FLOW_ITEM_TYPE_IPV4:
2901                         mlx5_flow_tunnel_ip_check(items, next_protocol,
2902                                                   &item_flags, &tunnel);
2903                         ret = mlx5_flow_validate_item_ipv4(items, item_flags,
2904                                                            NULL, error);
2905                         if (ret < 0)
2906                                 return ret;
2907                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2908                                              MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2909                         if (items->mask != NULL &&
2910                             ((const struct rte_flow_item_ipv4 *)
2911                              items->mask)->hdr.next_proto_id) {
2912                                 next_protocol =
2913                                         ((const struct rte_flow_item_ipv4 *)
2914                                          (items->spec))->hdr.next_proto_id;
2915                                 next_protocol &=
2916                                         ((const struct rte_flow_item_ipv4 *)
2917                                          (items->mask))->hdr.next_proto_id;
2918                         } else {
2919                                 /* Reset for inner layer. */
2920                                 next_protocol = 0xff;
2921                         }
2922                         break;
2923                 case RTE_FLOW_ITEM_TYPE_IPV6:
2924                         mlx5_flow_tunnel_ip_check(items, next_protocol,
2925                                                   &item_flags, &tunnel);
2926                         ret = mlx5_flow_validate_item_ipv6(items, item_flags,
2927                                                            NULL, error);
2928                         if (ret < 0)
2929                                 return ret;
2930                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
2931                                              MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2932                         if (items->mask != NULL &&
2933                             ((const struct rte_flow_item_ipv6 *)
2934                              items->mask)->hdr.proto) {
2935                                 next_protocol =
2936                                         ((const struct rte_flow_item_ipv6 *)
2937                                          items->spec)->hdr.proto;
2938                                 next_protocol &=
2939                                         ((const struct rte_flow_item_ipv6 *)
2940                                          items->mask)->hdr.proto;
2941                         } else {
2942                                 /* Reset for inner layer. */
2943                                 next_protocol = 0xff;
2944                         }
2945                         break;
2946                 case RTE_FLOW_ITEM_TYPE_TCP:
2947                         ret = mlx5_flow_validate_item_tcp
2948                                                 (items, item_flags,
2949                                                  next_protocol,
2950                                                  &nic_tcp_mask,
2951                                                  error);
2952                         if (ret < 0)
2953                                 return ret;
2954                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
2955                                              MLX5_FLOW_LAYER_OUTER_L4_TCP;
2956                         break;
2957                 case RTE_FLOW_ITEM_TYPE_UDP:
2958                         ret = mlx5_flow_validate_item_udp(items, item_flags,
2959                                                           next_protocol,
2960                                                           error);
2961                         if (ret < 0)
2962                                 return ret;
2963                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
2964                                              MLX5_FLOW_LAYER_OUTER_L4_UDP;
2965                         break;
2966                 case RTE_FLOW_ITEM_TYPE_GRE:
2967                         ret = mlx5_flow_validate_item_gre(items, item_flags,
2968                                                           next_protocol, error);
2969                         if (ret < 0)
2970                                 return ret;
2971                         gre_item = items;
2972                         last_item = MLX5_FLOW_LAYER_GRE;
2973                         break;
2974                 case RTE_FLOW_ITEM_TYPE_NVGRE:
2975                         ret = mlx5_flow_validate_item_nvgre(items, item_flags,
2976                                                             next_protocol,
2977                                                             error);
2978                         if (ret < 0)
2979                                 return ret;
2980                         last_item = MLX5_FLOW_LAYER_NVGRE;
2981                         break;
2982                 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
2983                         ret = mlx5_flow_validate_item_gre_key
2984                                 (items, item_flags, gre_item, error);
2985                         if (ret < 0)
2986                                 return ret;
2987                         last_item = MLX5_FLOW_LAYER_GRE_KEY;
2988                         break;
2989                 case RTE_FLOW_ITEM_TYPE_VXLAN:
2990                         ret = mlx5_flow_validate_item_vxlan(items, item_flags,
2991                                                             error);
2992                         if (ret < 0)
2993                                 return ret;
2994                         last_item = MLX5_FLOW_LAYER_VXLAN;
2995                         break;
2996                 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2997                         ret = mlx5_flow_validate_item_vxlan_gpe(items,
2998                                                                 item_flags, dev,
2999                                                                 error);
3000                         if (ret < 0)
3001                                 return ret;
3002                         last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
3003                         break;
3004                 case RTE_FLOW_ITEM_TYPE_MPLS:
3005                         ret = mlx5_flow_validate_item_mpls(dev, items,
3006                                                            item_flags,
3007                                                            last_item, error);
3008                         if (ret < 0)
3009                                 return ret;
3010                         last_item = MLX5_FLOW_LAYER_MPLS;
3011                         break;
3012                 case RTE_FLOW_ITEM_TYPE_META:
3013                         ret = flow_dv_validate_item_meta(dev, items, attr,
3014                                                          error);
3015                         if (ret < 0)
3016                                 return ret;
3017                         last_item = MLX5_FLOW_ITEM_METADATA;
3018                         break;
3019                 case RTE_FLOW_ITEM_TYPE_ICMP:
3020                         ret = mlx5_flow_validate_item_icmp(items, item_flags,
3021                                                            next_protocol,
3022                                                            error);
3023                         if (ret < 0)
3024                                 return ret;
3025                         last_item = MLX5_FLOW_LAYER_ICMP;
3026                         break;
3027                 case RTE_FLOW_ITEM_TYPE_ICMP6:
3028                         ret = mlx5_flow_validate_item_icmp6(items, item_flags,
3029                                                             next_protocol,
3030                                                             error);
3031                         if (ret < 0)
3032                                 return ret;
3033                         last_item = MLX5_FLOW_LAYER_ICMP6;
3034                         break;
3035                 default:
3036                         return rte_flow_error_set(error, ENOTSUP,
3037                                                   RTE_FLOW_ERROR_TYPE_ITEM,
3038                                                   NULL, "item not supported");
3039                 }
3040                 item_flags |= last_item;
3041         }
3042         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3043                 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
3044                         return rte_flow_error_set(error, ENOTSUP,
3045                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3046                                                   actions, "too many actions");
3047                 switch (actions->type) {
3048                 case RTE_FLOW_ACTION_TYPE_VOID:
3049                         break;
3050                 case RTE_FLOW_ACTION_TYPE_PORT_ID:
3051                         ret = flow_dv_validate_action_port_id(dev,
3052                                                               action_flags,
3053                                                               actions,
3054                                                               attr,
3055                                                               error);
3056                         if (ret)
3057                                 return ret;
3058                         action_flags |= MLX5_FLOW_ACTION_PORT_ID;
3059                         ++actions_n;
3060                         break;
3061                 case RTE_FLOW_ACTION_TYPE_FLAG:
3062                         ret = mlx5_flow_validate_action_flag(action_flags,
3063                                                              attr, error);
3064                         if (ret < 0)
3065                                 return ret;
3066                         action_flags |= MLX5_FLOW_ACTION_FLAG;
3067                         ++actions_n;
3068                         break;
3069                 case RTE_FLOW_ACTION_TYPE_MARK:
3070                         ret = mlx5_flow_validate_action_mark(actions,
3071                                                              action_flags,
3072                                                              attr, error);
3073                         if (ret < 0)
3074                                 return ret;
3075                         action_flags |= MLX5_FLOW_ACTION_MARK;
3076                         ++actions_n;
3077                         break;
3078                 case RTE_FLOW_ACTION_TYPE_DROP:
3079                         ret = mlx5_flow_validate_action_drop(action_flags,
3080                                                              attr, error);
3081                         if (ret < 0)
3082                                 return ret;
3083                         action_flags |= MLX5_FLOW_ACTION_DROP;
3084                         ++actions_n;
3085                         break;
3086                 case RTE_FLOW_ACTION_TYPE_QUEUE:
3087                         ret = mlx5_flow_validate_action_queue(actions,
3088                                                               action_flags, dev,
3089                                                               attr, error);
3090                         if (ret < 0)
3091                                 return ret;
3092                         action_flags |= MLX5_FLOW_ACTION_QUEUE;
3093                         ++actions_n;
3094                         break;
3095                 case RTE_FLOW_ACTION_TYPE_RSS:
3096                         ret = mlx5_flow_validate_action_rss(actions,
3097                                                             action_flags, dev,
3098                                                             attr, item_flags,
3099                                                             error);
3100                         if (ret < 0)
3101                                 return ret;
3102                         action_flags |= MLX5_FLOW_ACTION_RSS;
3103                         ++actions_n;
3104                         break;
3105                 case RTE_FLOW_ACTION_TYPE_COUNT:
3106                         ret = flow_dv_validate_action_count(dev, error);
3107                         if (ret < 0)
3108                                 return ret;
3109                         action_flags |= MLX5_FLOW_ACTION_COUNT;
3110                         ++actions_n;
3111                         break;
3112                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3113                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3114                         ret = flow_dv_validate_action_l2_encap(action_flags,
3115                                                                actions, attr,
3116                                                                error);
3117                         if (ret < 0)
3118                                 return ret;
3119                         action_flags |= actions->type ==
3120                                         RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP ?
3121                                         MLX5_FLOW_ACTION_VXLAN_ENCAP :
3122                                         MLX5_FLOW_ACTION_NVGRE_ENCAP;
3123                         ++actions_n;
3124                         break;
3125                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3126                 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3127                         ret = flow_dv_validate_action_l2_decap(action_flags,
3128                                                                attr, error);
3129                         if (ret < 0)
3130                                 return ret;
3131                         action_flags |= actions->type ==
3132                                         RTE_FLOW_ACTION_TYPE_VXLAN_DECAP ?
3133                                         MLX5_FLOW_ACTION_VXLAN_DECAP :
3134                                         MLX5_FLOW_ACTION_NVGRE_DECAP;
3135                         ++actions_n;
3136                         break;
3137                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3138                         ret = flow_dv_validate_action_raw_encap(action_flags,
3139                                                                 actions, attr,
3140                                                                 error);
3141                         if (ret < 0)
3142                                 return ret;
3143                         action_flags |= MLX5_FLOW_ACTION_RAW_ENCAP;
3144                         ++actions_n;
3145                         break;
3146                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3147                         ret = flow_dv_validate_action_raw_decap(action_flags,
3148                                                                 actions, attr,
3149                                                                 error);
3150                         if (ret < 0)
3151                                 return ret;
3152                         action_flags |= MLX5_FLOW_ACTION_RAW_DECAP;
3153                         ++actions_n;
3154                         break;
3155                 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3156                 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3157                         ret = flow_dv_validate_action_modify_mac(action_flags,
3158                                                                  actions,
3159                                                                  item_flags,
3160                                                                  error);
3161                         if (ret < 0)
3162                                 return ret;
3163                         /* Count all modify-header actions as one action. */
3164                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3165                                 ++actions_n;
3166                         action_flags |= actions->type ==
3167                                         RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
3168                                                 MLX5_FLOW_ACTION_SET_MAC_SRC :
3169                                                 MLX5_FLOW_ACTION_SET_MAC_DST;
3170                         break;
3171
3172                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3173                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3174                         ret = flow_dv_validate_action_modify_ipv4(action_flags,
3175                                                                   actions,
3176                                                                   item_flags,
3177                                                                   error);
3178                         if (ret < 0)
3179                                 return ret;
3180                         /* Count all modify-header actions as one action. */
3181                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3182                                 ++actions_n;
3183                         action_flags |= actions->type ==
3184                                         RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
3185                                                 MLX5_FLOW_ACTION_SET_IPV4_SRC :
3186                                                 MLX5_FLOW_ACTION_SET_IPV4_DST;
3187                         break;
3188                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3189                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3190                         ret = flow_dv_validate_action_modify_ipv6(action_flags,
3191                                                                   actions,
3192                                                                   item_flags,
3193                                                                   error);
3194                         if (ret < 0)
3195                                 return ret;
3196                         /* Count all modify-header actions as one action. */
3197                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3198                                 ++actions_n;
3199                         action_flags |= actions->type ==
3200                                         RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
3201                                                 MLX5_FLOW_ACTION_SET_IPV6_SRC :
3202                                                 MLX5_FLOW_ACTION_SET_IPV6_DST;
3203                         break;
3204                 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3205                 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3206                         ret = flow_dv_validate_action_modify_tp(action_flags,
3207                                                                 actions,
3208                                                                 item_flags,
3209                                                                 error);
3210                         if (ret < 0)
3211                                 return ret;
3212                         /* Count all modify-header actions as one action. */
3213                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3214                                 ++actions_n;
3215                         action_flags |= actions->type ==
3216                                         RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
3217                                                 MLX5_FLOW_ACTION_SET_TP_SRC :
3218                                                 MLX5_FLOW_ACTION_SET_TP_DST;
3219                         break;
3220                 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3221                 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3222                         ret = flow_dv_validate_action_modify_ttl(action_flags,
3223                                                                  actions,
3224                                                                  item_flags,
3225                                                                  error);
3226                         if (ret < 0)
3227                                 return ret;
3228                         /* Count all modify-header actions as one action. */
3229                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3230                                 ++actions_n;
3231                         action_flags |= actions->type ==
3232                                         RTE_FLOW_ACTION_TYPE_SET_TTL ?
3233                                                 MLX5_FLOW_ACTION_SET_TTL :
3234                                                 MLX5_FLOW_ACTION_DEC_TTL;
3235                         break;
3236                 case RTE_FLOW_ACTION_TYPE_JUMP:
3237                         ret = flow_dv_validate_action_jump(actions,
3238                                                            action_flags,
3239                                                            attr->group, error);
3240                         if (ret)
3241                                 return ret;
3242                         ++actions_n;
3243                         action_flags |= MLX5_FLOW_ACTION_JUMP;
3244                         break;
3245                 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3246                 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3247                         ret = flow_dv_validate_action_modify_tcp_seq
3248                                                                 (action_flags,
3249                                                                  actions,
3250                                                                  item_flags,
3251                                                                  error);
3252                         if (ret < 0)
3253                                 return ret;
3254                         /* Count all modify-header actions as one action. */
3255                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3256                                 ++actions_n;
3257                         action_flags |= actions->type ==
3258                                         RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
3259                                                 MLX5_FLOW_ACTION_INC_TCP_SEQ :
3260                                                 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
3261                         break;
3262                 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3263                 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3264                         ret = flow_dv_validate_action_modify_tcp_ack
3265                                                                 (action_flags,
3266                                                                  actions,
3267                                                                  item_flags,
3268                                                                  error);
3269                         if (ret < 0)
3270                                 return ret;
3271                         /* Count all modify-header actions as one action. */
3272                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3273                                 ++actions_n;
3274                         action_flags |= actions->type ==
3275                                         RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
3276                                                 MLX5_FLOW_ACTION_INC_TCP_ACK :
3277                                                 MLX5_FLOW_ACTION_DEC_TCP_ACK;
3278                         break;
3279                 default:
3280                         return rte_flow_error_set(error, ENOTSUP,
3281                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3282                                                   actions,
3283                                                   "action not supported");
3284                 }
3285         }
3286         /* Eswitch has few restrictions on using items and actions */
3287         if (attr->transfer) {
3288                 if (action_flags & MLX5_FLOW_ACTION_FLAG)
3289                         return rte_flow_error_set(error, ENOTSUP,
3290                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3291                                                   NULL,
3292                                                   "unsupported action FLAG");
3293                 if (action_flags & MLX5_FLOW_ACTION_MARK)
3294                         return rte_flow_error_set(error, ENOTSUP,
3295                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3296                                                   NULL,
3297                                                   "unsupported action MARK");
3298                 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
3299                         return rte_flow_error_set(error, ENOTSUP,
3300                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3301                                                   NULL,
3302                                                   "unsupported action QUEUE");
3303                 if (action_flags & MLX5_FLOW_ACTION_RSS)
3304                         return rte_flow_error_set(error, ENOTSUP,
3305                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3306                                                   NULL,
3307                                                   "unsupported action RSS");
3308                 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3309                         return rte_flow_error_set(error, EINVAL,
3310                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3311                                                   actions,
3312                                                   "no fate action is found");
3313         } else {
3314                 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
3315                         return rte_flow_error_set(error, EINVAL,
3316                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3317                                                   actions,
3318                                                   "no fate action is found");
3319         }
3320         return 0;
3321 }
3322
3323 /**
3324  * Internal preparation function. Allocates the DV flow size,
3325  * this size is constant.
3326  *
3327  * @param[in] attr
3328  *   Pointer to the flow attributes.
3329  * @param[in] items
3330  *   Pointer to the list of items.
3331  * @param[in] actions
3332  *   Pointer to the list of actions.
3333  * @param[out] error
3334  *   Pointer to the error structure.
3335  *
3336  * @return
3337  *   Pointer to mlx5_flow object on success,
3338  *   otherwise NULL and rte_errno is set.
3339  */
3340 static struct mlx5_flow *
3341 flow_dv_prepare(const struct rte_flow_attr *attr __rte_unused,
3342                 const struct rte_flow_item items[] __rte_unused,
3343                 const struct rte_flow_action actions[] __rte_unused,
3344                 struct rte_flow_error *error)
3345 {
3346         uint32_t size = sizeof(struct mlx5_flow);
3347         struct mlx5_flow *flow;
3348
3349         flow = rte_calloc(__func__, 1, size, 0);
3350         if (!flow) {
3351                 rte_flow_error_set(error, ENOMEM,
3352                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3353                                    "not enough memory to create flow");
3354                 return NULL;
3355         }
3356         flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
3357         return flow;
3358 }
3359
3360 #ifndef NDEBUG
3361 /**
3362  * Sanity check for match mask and value. Similar to check_valid_spec() in
3363  * kernel driver. If unmasked bit is present in value, it returns failure.
3364  *
3365  * @param match_mask
3366  *   pointer to match mask buffer.
3367  * @param match_value
3368  *   pointer to match value buffer.
3369  *
3370  * @return
3371  *   0 if valid, -EINVAL otherwise.
3372  */
3373 static int
3374 flow_dv_check_valid_spec(void *match_mask, void *match_value)
3375 {
3376         uint8_t *m = match_mask;
3377         uint8_t *v = match_value;
3378         unsigned int i;
3379
3380         for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
3381                 if (v[i] & ~m[i]) {
3382                         DRV_LOG(ERR,
3383                                 "match_value differs from match_criteria"
3384                                 " %p[%u] != %p[%u]",
3385                                 match_value, i, match_mask, i);
3386                         return -EINVAL;
3387                 }
3388         }
3389         return 0;
3390 }
3391 #endif
3392
3393 /**
3394  * Add Ethernet item to matcher and to the value.
3395  *
3396  * @param[in, out] matcher
3397  *   Flow matcher.
3398  * @param[in, out] key
3399  *   Flow matcher value.
3400  * @param[in] item
3401  *   Flow pattern to translate.
3402  * @param[in] inner
3403  *   Item is inner pattern.
3404  */
3405 static void
3406 flow_dv_translate_item_eth(void *matcher, void *key,
3407                            const struct rte_flow_item *item, int inner)
3408 {
3409         const struct rte_flow_item_eth *eth_m = item->mask;
3410         const struct rte_flow_item_eth *eth_v = item->spec;
3411         const struct rte_flow_item_eth nic_mask = {
3412                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
3413                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
3414                 .type = RTE_BE16(0xffff),
3415         };
3416         void *headers_m;
3417         void *headers_v;
3418         char *l24_v;
3419         unsigned int i;
3420
3421         if (!eth_v)
3422                 return;
3423         if (!eth_m)
3424                 eth_m = &nic_mask;
3425         if (inner) {
3426                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3427                                          inner_headers);
3428                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3429         } else {
3430                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3431                                          outer_headers);
3432                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3433         }
3434         memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
3435                &eth_m->dst, sizeof(eth_m->dst));
3436         /* The value must be in the range of the mask. */
3437         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
3438         for (i = 0; i < sizeof(eth_m->dst); ++i)
3439                 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
3440         memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
3441                &eth_m->src, sizeof(eth_m->src));
3442         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
3443         /* The value must be in the range of the mask. */
3444         for (i = 0; i < sizeof(eth_m->dst); ++i)
3445                 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
3446         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
3447                  rte_be_to_cpu_16(eth_m->type));
3448         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, ethertype);
3449         *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
3450 }
3451
3452 /**
3453  * Add VLAN item to matcher and to the value.
3454  *
3455  * @param[in, out] dev_flow
3456  *   Flow descriptor.
3457  * @param[in, out] matcher
3458  *   Flow matcher.
3459  * @param[in, out] key
3460  *   Flow matcher value.
3461  * @param[in] item
3462  *   Flow pattern to translate.
3463  * @param[in] inner
3464  *   Item is inner pattern.
3465  */
3466 static void
3467 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
3468                             void *matcher, void *key,
3469                             const struct rte_flow_item *item,
3470                             int inner)
3471 {
3472         const struct rte_flow_item_vlan *vlan_m = item->mask;
3473         const struct rte_flow_item_vlan *vlan_v = item->spec;
3474         void *headers_m;
3475         void *headers_v;
3476         uint16_t tci_m;
3477         uint16_t tci_v;
3478
3479         if (!vlan_v)
3480                 return;
3481         if (!vlan_m)
3482                 vlan_m = &rte_flow_item_vlan_mask;
3483         if (inner) {
3484                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3485                                          inner_headers);
3486                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3487         } else {
3488                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3489                                          outer_headers);
3490                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3491                 /*
3492                  * This is workaround, masks are not supported,
3493                  * and pre-validated.
3494                  */
3495                 dev_flow->dv.vf_vlan.tag =
3496                         rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
3497         }
3498         tci_m = rte_be_to_cpu_16(vlan_m->tci);
3499         tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
3500         MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
3501         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
3502         MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
3503         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
3504         MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
3505         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
3506         MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
3507         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
3508         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
3509                  rte_be_to_cpu_16(vlan_m->inner_type));
3510         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
3511                  rte_be_to_cpu_16(vlan_m->inner_type & vlan_v->inner_type));
3512 }
3513
3514 /**
3515  * Add IPV4 item to matcher and to the value.
3516  *
3517  * @param[in, out] matcher
3518  *   Flow matcher.
3519  * @param[in, out] key
3520  *   Flow matcher value.
3521  * @param[in] item
3522  *   Flow pattern to translate.
3523  * @param[in] inner
3524  *   Item is inner pattern.
3525  * @param[in] group
3526  *   The group to insert the rule.
3527  */
3528 static void
3529 flow_dv_translate_item_ipv4(void *matcher, void *key,
3530                             const struct rte_flow_item *item,
3531                             int inner, uint32_t group)
3532 {
3533         const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
3534         const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
3535         const struct rte_flow_item_ipv4 nic_mask = {
3536                 .hdr = {
3537                         .src_addr = RTE_BE32(0xffffffff),
3538                         .dst_addr = RTE_BE32(0xffffffff),
3539                         .type_of_service = 0xff,
3540                         .next_proto_id = 0xff,
3541                 },
3542         };
3543         void *headers_m;
3544         void *headers_v;
3545         char *l24_m;
3546         char *l24_v;
3547         uint8_t tos;
3548
3549         if (inner) {
3550                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3551                                          inner_headers);
3552                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3553         } else {
3554                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3555                                          outer_headers);
3556                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3557         }
3558         if (group == 0)
3559                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
3560         else
3561                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x4);
3562         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 4);
3563         if (!ipv4_v)
3564                 return;
3565         if (!ipv4_m)
3566                 ipv4_m = &nic_mask;
3567         l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
3568                              dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
3569         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
3570                              dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
3571         *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
3572         *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
3573         l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
3574                           src_ipv4_src_ipv6.ipv4_layout.ipv4);
3575         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
3576                           src_ipv4_src_ipv6.ipv4_layout.ipv4);
3577         *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
3578         *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
3579         tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
3580         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
3581                  ipv4_m->hdr.type_of_service);
3582         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
3583         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
3584                  ipv4_m->hdr.type_of_service >> 2);
3585         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
3586         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
3587                  ipv4_m->hdr.next_proto_id);
3588         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
3589                  ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
3590 }
3591
3592 /**
3593  * Add IPV6 item to matcher and to the value.
3594  *
3595  * @param[in, out] matcher
3596  *   Flow matcher.
3597  * @param[in, out] key
3598  *   Flow matcher value.
3599  * @param[in] item
3600  *   Flow pattern to translate.
3601  * @param[in] inner
3602  *   Item is inner pattern.
3603  * @param[in] group
3604  *   The group to insert the rule.
3605  */
3606 static void
3607 flow_dv_translate_item_ipv6(void *matcher, void *key,
3608                             const struct rte_flow_item *item,
3609                             int inner, uint32_t group)
3610 {
3611         const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
3612         const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
3613         const struct rte_flow_item_ipv6 nic_mask = {
3614                 .hdr = {
3615                         .src_addr =
3616                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
3617                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
3618                         .dst_addr =
3619                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
3620                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
3621                         .vtc_flow = RTE_BE32(0xffffffff),
3622                         .proto = 0xff,
3623                         .hop_limits = 0xff,
3624                 },
3625         };
3626         void *headers_m;
3627         void *headers_v;
3628         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3629         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3630         char *l24_m;
3631         char *l24_v;
3632         uint32_t vtc_m;
3633         uint32_t vtc_v;
3634         int i;
3635         int size;
3636
3637         if (inner) {
3638                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3639                                          inner_headers);
3640                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3641         } else {
3642                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3643                                          outer_headers);
3644                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3645         }
3646         if (group == 0)
3647                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
3648         else
3649                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x6);
3650         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 6);
3651         if (!ipv6_v)
3652                 return;
3653         if (!ipv6_m)
3654                 ipv6_m = &nic_mask;
3655         size = sizeof(ipv6_m->hdr.dst_addr);
3656         l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
3657                              dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
3658         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
3659                              dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
3660         memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
3661         for (i = 0; i < size; ++i)
3662                 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
3663         l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
3664                              src_ipv4_src_ipv6.ipv6_layout.ipv6);
3665         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
3666                              src_ipv4_src_ipv6.ipv6_layout.ipv6);
3667         memcpy(l24_m, ipv6_m->hdr.src_addr, size);
3668         for (i = 0; i < size; ++i)
3669                 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
3670         /* TOS. */
3671         vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
3672         vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
3673         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
3674         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
3675         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
3676         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
3677         /* Label. */
3678         if (inner) {
3679                 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
3680                          vtc_m);
3681                 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
3682                          vtc_v);
3683         } else {
3684                 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
3685                          vtc_m);
3686                 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
3687                          vtc_v);
3688         }
3689         /* Protocol. */
3690         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
3691                  ipv6_m->hdr.proto);
3692         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
3693                  ipv6_v->hdr.proto & ipv6_m->hdr.proto);
3694 }
3695
3696 /**
3697  * Add TCP item to matcher and to the value.
3698  *
3699  * @param[in, out] matcher
3700  *   Flow matcher.
3701  * @param[in, out] key
3702  *   Flow matcher value.
3703  * @param[in] item
3704  *   Flow pattern to translate.
3705  * @param[in] inner
3706  *   Item is inner pattern.
3707  */
3708 static void
3709 flow_dv_translate_item_tcp(void *matcher, void *key,
3710                            const struct rte_flow_item *item,
3711                            int inner)
3712 {
3713         const struct rte_flow_item_tcp *tcp_m = item->mask;
3714         const struct rte_flow_item_tcp *tcp_v = item->spec;
3715         void *headers_m;
3716         void *headers_v;
3717
3718         if (inner) {
3719                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3720                                          inner_headers);
3721                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3722         } else {
3723                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3724                                          outer_headers);
3725                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3726         }
3727         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
3728         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
3729         if (!tcp_v)
3730                 return;
3731         if (!tcp_m)
3732                 tcp_m = &rte_flow_item_tcp_mask;
3733         MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
3734                  rte_be_to_cpu_16(tcp_m->hdr.src_port));
3735         MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
3736                  rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
3737         MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
3738                  rte_be_to_cpu_16(tcp_m->hdr.dst_port));
3739         MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
3740                  rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
3741         MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
3742                  tcp_m->hdr.tcp_flags);
3743         MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
3744                  (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
3745 }
3746
3747 /**
3748  * Add UDP item to matcher and to the value.
3749  *
3750  * @param[in, out] matcher
3751  *   Flow matcher.
3752  * @param[in, out] key
3753  *   Flow matcher value.
3754  * @param[in] item
3755  *   Flow pattern to translate.
3756  * @param[in] inner
3757  *   Item is inner pattern.
3758  */
3759 static void
3760 flow_dv_translate_item_udp(void *matcher, void *key,
3761                            const struct rte_flow_item *item,
3762                            int inner)
3763 {
3764         const struct rte_flow_item_udp *udp_m = item->mask;
3765         const struct rte_flow_item_udp *udp_v = item->spec;
3766         void *headers_m;
3767         void *headers_v;
3768
3769         if (inner) {
3770                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3771                                          inner_headers);
3772                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3773         } else {
3774                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3775                                          outer_headers);
3776                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3777         }
3778         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
3779         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
3780         if (!udp_v)
3781                 return;
3782         if (!udp_m)
3783                 udp_m = &rte_flow_item_udp_mask;
3784         MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
3785                  rte_be_to_cpu_16(udp_m->hdr.src_port));
3786         MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
3787                  rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
3788         MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
3789                  rte_be_to_cpu_16(udp_m->hdr.dst_port));
3790         MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
3791                  rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
3792 }
3793
3794 /**
3795  * Add GRE optional Key item to matcher and to the value.
3796  *
3797  * @param[in, out] matcher
3798  *   Flow matcher.
3799  * @param[in, out] key
3800  *   Flow matcher value.
3801  * @param[in] item
3802  *   Flow pattern to translate.
3803  * @param[in] inner
3804  *   Item is inner pattern.
3805  */
3806 static void
3807 flow_dv_translate_item_gre_key(void *matcher, void *key,
3808                                    const struct rte_flow_item *item)
3809 {
3810         const rte_be32_t *key_m = item->mask;
3811         const rte_be32_t *key_v = item->spec;
3812         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3813         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3814         rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
3815
3816         if (!key_v)
3817                 return;
3818         if (!key_m)
3819                 key_m = &gre_key_default_mask;
3820         /* GRE K bit must be on and should already be validated */
3821         MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
3822         MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
3823         MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
3824                  rte_be_to_cpu_32(*key_m) >> 8);
3825         MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
3826                  rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
3827         MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
3828                  rte_be_to_cpu_32(*key_m) & 0xFF);
3829         MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
3830                  rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
3831 }
3832
3833 /**
3834  * Add GRE item to matcher and to the value.
3835  *
3836  * @param[in, out] matcher
3837  *   Flow matcher.
3838  * @param[in, out] key
3839  *   Flow matcher value.
3840  * @param[in] item
3841  *   Flow pattern to translate.
3842  * @param[in] inner
3843  *   Item is inner pattern.
3844  */
3845 static void
3846 flow_dv_translate_item_gre(void *matcher, void *key,
3847                            const struct rte_flow_item *item,
3848                            int inner)
3849 {
3850         const struct rte_flow_item_gre *gre_m = item->mask;
3851         const struct rte_flow_item_gre *gre_v = item->spec;
3852         void *headers_m;
3853         void *headers_v;
3854         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3855         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3856         struct {
3857                 union {
3858                         __extension__
3859                         struct {
3860                                 uint16_t version:3;
3861                                 uint16_t rsvd0:9;
3862                                 uint16_t s_present:1;
3863                                 uint16_t k_present:1;
3864                                 uint16_t rsvd_bit1:1;
3865                                 uint16_t c_present:1;
3866                         };
3867                         uint16_t value;
3868                 };
3869         } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
3870
3871         if (inner) {
3872                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3873                                          inner_headers);
3874                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3875         } else {
3876                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3877                                          outer_headers);
3878                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3879         }
3880         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
3881         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
3882         if (!gre_v)
3883                 return;
3884         if (!gre_m)
3885                 gre_m = &rte_flow_item_gre_mask;
3886         MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
3887                  rte_be_to_cpu_16(gre_m->protocol));
3888         MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
3889                  rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
3890         gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
3891         gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
3892         MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
3893                  gre_crks_rsvd0_ver_m.c_present);
3894         MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
3895                  gre_crks_rsvd0_ver_v.c_present &
3896                  gre_crks_rsvd0_ver_m.c_present);
3897         MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
3898                  gre_crks_rsvd0_ver_m.k_present);
3899         MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
3900                  gre_crks_rsvd0_ver_v.k_present &
3901                  gre_crks_rsvd0_ver_m.k_present);
3902         MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
3903                  gre_crks_rsvd0_ver_m.s_present);
3904         MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
3905                  gre_crks_rsvd0_ver_v.s_present &
3906                  gre_crks_rsvd0_ver_m.s_present);
3907 }
3908
3909 /**
3910  * Add NVGRE item to matcher and to the value.
3911  *
3912  * @param[in, out] matcher
3913  *   Flow matcher.
3914  * @param[in, out] key
3915  *   Flow matcher value.
3916  * @param[in] item
3917  *   Flow pattern to translate.
3918  * @param[in] inner
3919  *   Item is inner pattern.
3920  */
3921 static void
3922 flow_dv_translate_item_nvgre(void *matcher, void *key,
3923                              const struct rte_flow_item *item,
3924                              int inner)
3925 {
3926         const struct rte_flow_item_nvgre *nvgre_m = item->mask;
3927         const struct rte_flow_item_nvgre *nvgre_v = item->spec;
3928         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3929         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3930         const char *tni_flow_id_m = (const char *)nvgre_m->tni;
3931         const char *tni_flow_id_v = (const char *)nvgre_v->tni;
3932         char *gre_key_m;
3933         char *gre_key_v;
3934         int size;
3935         int i;
3936
3937         /* For NVGRE, GRE header fields must be set with defined values. */
3938         const struct rte_flow_item_gre gre_spec = {
3939                 .c_rsvd0_ver = RTE_BE16(0x2000),
3940                 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
3941         };
3942         const struct rte_flow_item_gre gre_mask = {
3943                 .c_rsvd0_ver = RTE_BE16(0xB000),
3944                 .protocol = RTE_BE16(UINT16_MAX),
3945         };
3946         const struct rte_flow_item gre_item = {
3947                 .spec = &gre_spec,
3948                 .mask = &gre_mask,
3949                 .last = NULL,
3950         };
3951         flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
3952         if (!nvgre_v)
3953                 return;
3954         if (!nvgre_m)
3955                 nvgre_m = &rte_flow_item_nvgre_mask;
3956         size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
3957         gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
3958         gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
3959         memcpy(gre_key_m, tni_flow_id_m, size);
3960         for (i = 0; i < size; ++i)
3961                 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
3962 }
3963
3964 /**
3965  * Add VXLAN item to matcher and to the value.
3966  *
3967  * @param[in, out] matcher
3968  *   Flow matcher.
3969  * @param[in, out] key
3970  *   Flow matcher value.
3971  * @param[in] item
3972  *   Flow pattern to translate.
3973  * @param[in] inner
3974  *   Item is inner pattern.
3975  */
3976 static void
3977 flow_dv_translate_item_vxlan(void *matcher, void *key,
3978                              const struct rte_flow_item *item,
3979                              int inner)
3980 {
3981         const struct rte_flow_item_vxlan *vxlan_m = item->mask;
3982         const struct rte_flow_item_vxlan *vxlan_v = item->spec;
3983         void *headers_m;
3984         void *headers_v;
3985         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3986         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3987         char *vni_m;
3988         char *vni_v;
3989         uint16_t dport;
3990         int size;
3991         int i;
3992
3993         if (inner) {
3994                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3995                                          inner_headers);
3996                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3997         } else {
3998                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3999                                          outer_headers);
4000                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
4001         }
4002         dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
4003                 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
4004         if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
4005                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
4006                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
4007         }
4008         if (!vxlan_v)
4009                 return;
4010         if (!vxlan_m)
4011                 vxlan_m = &rte_flow_item_vxlan_mask;
4012         size = sizeof(vxlan_m->vni);
4013         vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
4014         vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
4015         memcpy(vni_m, vxlan_m->vni, size);
4016         for (i = 0; i < size; ++i)
4017                 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
4018 }
4019
4020 /**
4021  * Add MPLS item to matcher and to the value.
4022  *
4023  * @param[in, out] matcher
4024  *   Flow matcher.
4025  * @param[in, out] key
4026  *   Flow matcher value.
4027  * @param[in] item
4028  *   Flow pattern to translate.
4029  * @param[in] prev_layer
4030  *   The protocol layer indicated in previous item.
4031  * @param[in] inner
4032  *   Item is inner pattern.
4033  */
4034 static void
4035 flow_dv_translate_item_mpls(void *matcher, void *key,
4036                             const struct rte_flow_item *item,
4037                             uint64_t prev_layer,
4038                             int inner)
4039 {
4040         const uint32_t *in_mpls_m = item->mask;
4041         const uint32_t *in_mpls_v = item->spec;
4042         uint32_t *out_mpls_m = 0;
4043         uint32_t *out_mpls_v = 0;
4044         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
4045         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
4046         void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
4047                                      misc_parameters_2);
4048         void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
4049         void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
4050         void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
4051
4052         switch (prev_layer) {
4053         case MLX5_FLOW_LAYER_OUTER_L4_UDP:
4054                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
4055                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
4056                          MLX5_UDP_PORT_MPLS);
4057                 break;
4058         case MLX5_FLOW_LAYER_GRE:
4059                 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
4060                 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
4061                          RTE_ETHER_TYPE_MPLS);
4062                 break;
4063         default:
4064                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
4065                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
4066                          IPPROTO_MPLS);
4067                 break;
4068         }
4069         if (!in_mpls_v)
4070                 return;
4071         if (!in_mpls_m)
4072                 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
4073         switch (prev_layer) {
4074         case MLX5_FLOW_LAYER_OUTER_L4_UDP:
4075                 out_mpls_m =
4076                         (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
4077                                                  outer_first_mpls_over_udp);
4078                 out_mpls_v =
4079                         (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
4080                                                  outer_first_mpls_over_udp);
4081                 break;
4082         case MLX5_FLOW_LAYER_GRE:
4083                 out_mpls_m =
4084                         (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
4085                                                  outer_first_mpls_over_gre);
4086                 out_mpls_v =
4087                         (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
4088                                                  outer_first_mpls_over_gre);
4089                 break;
4090         default:
4091                 /* Inner MPLS not over GRE is not supported. */
4092                 if (!inner) {
4093                         out_mpls_m =
4094                                 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
4095                                                          misc2_m,
4096                                                          outer_first_mpls);
4097                         out_mpls_v =
4098                                 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
4099                                                          misc2_v,
4100                                                          outer_first_mpls);
4101                 }
4102                 break;
4103         }
4104         if (out_mpls_m && out_mpls_v) {
4105                 *out_mpls_m = *in_mpls_m;
4106                 *out_mpls_v = *in_mpls_v & *in_mpls_m;
4107         }
4108 }
4109
4110 /**
4111  * Add META item to matcher
4112  *
4113  * @param[in, out] matcher
4114  *   Flow matcher.
4115  * @param[in, out] key
4116  *   Flow matcher value.
4117  * @param[in] item
4118  *   Flow pattern to translate.
4119  * @param[in] inner
4120  *   Item is inner pattern.
4121  */
4122 static void
4123 flow_dv_translate_item_meta(void *matcher, void *key,
4124                             const struct rte_flow_item *item)
4125 {
4126         const struct rte_flow_item_meta *meta_m;
4127         const struct rte_flow_item_meta *meta_v;
4128         void *misc2_m =
4129                 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
4130         void *misc2_v =
4131                 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
4132
4133         meta_m = (const void *)item->mask;
4134         if (!meta_m)
4135                 meta_m = &rte_flow_item_meta_mask;
4136         meta_v = (const void *)item->spec;
4137         if (meta_v) {
4138                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a,
4139                          rte_be_to_cpu_32(meta_m->data));
4140                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a,
4141                          rte_be_to_cpu_32(meta_v->data & meta_m->data));
4142         }
4143 }
4144
4145 /**
4146  * Add source vport match to the specified matcher.
4147  *
4148  * @param[in, out] matcher
4149  *   Flow matcher.
4150  * @param[in, out] key
4151  *   Flow matcher value.
4152  * @param[in] port
4153  *   Source vport value to match
4154  * @param[in] mask
4155  *   Mask
4156  */
4157 static void
4158 flow_dv_translate_item_source_vport(void *matcher, void *key,
4159                                     int16_t port, uint16_t mask)
4160 {
4161         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
4162         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
4163
4164         MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
4165         MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
4166 }
4167
4168 /**
4169  * Translate port-id item to eswitch match on  port-id.
4170  *
4171  * @param[in] dev
4172  *   The devich to configure through.
4173  * @param[in, out] matcher
4174  *   Flow matcher.
4175  * @param[in, out] key
4176  *   Flow matcher value.
4177  * @param[in] item
4178  *   Flow pattern to translate.
4179  *
4180  * @return
4181  *   0 on success, a negative errno value otherwise.
4182  */
4183 static int
4184 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
4185                                void *key, const struct rte_flow_item *item)
4186 {
4187         const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
4188         const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
4189         uint16_t mask, val, id;
4190         int ret;
4191
4192         mask = pid_m ? pid_m->id : 0xffff;
4193         id = pid_v ? pid_v->id : dev->data->port_id;
4194         ret = mlx5_port_to_eswitch_info(id, NULL, &val);
4195         if (ret)
4196                 return ret;
4197         flow_dv_translate_item_source_vport(matcher, key, val, mask);
4198         return 0;
4199 }
4200
4201 /**
4202  * Add ICMP6 item to matcher and to the value.
4203  *
4204  * @param[in, out] matcher
4205  *   Flow matcher.
4206  * @param[in, out] key
4207  *   Flow matcher value.
4208  * @param[in] item
4209  *   Flow pattern to translate.
4210  * @param[in] inner
4211  *   Item is inner pattern.
4212  */
4213 static void
4214 flow_dv_translate_item_icmp6(void *matcher, void *key,
4215                               const struct rte_flow_item *item,
4216                               int inner)
4217 {
4218         const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
4219         const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
4220         void *headers_m;
4221         void *headers_v;
4222         void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
4223                                      misc_parameters_3);
4224         void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
4225         if (inner) {
4226                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
4227                                          inner_headers);
4228                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
4229         } else {
4230                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
4231                                          outer_headers);
4232                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
4233         }
4234         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
4235         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
4236         if (!icmp6_v)
4237                 return;
4238         if (!icmp6_m)
4239                 icmp6_m = &rte_flow_item_icmp6_mask;
4240         MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
4241         MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
4242                  icmp6_v->type & icmp6_m->type);
4243         MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
4244         MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
4245                  icmp6_v->code & icmp6_m->code);
4246 }
4247
4248 /**
4249  * Add ICMP item to matcher and to the value.
4250  *
4251  * @param[in, out] matcher
4252  *   Flow matcher.
4253  * @param[in, out] key
4254  *   Flow matcher value.
4255  * @param[in] item
4256  *   Flow pattern to translate.
4257  * @param[in] inner
4258  *   Item is inner pattern.
4259  */
4260 static void
4261 flow_dv_translate_item_icmp(void *matcher, void *key,
4262                             const struct rte_flow_item *item,
4263                             int inner)
4264 {
4265         const struct rte_flow_item_icmp *icmp_m = item->mask;
4266         const struct rte_flow_item_icmp *icmp_v = item->spec;
4267         void *headers_m;
4268         void *headers_v;
4269         void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
4270                                      misc_parameters_3);
4271         void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
4272         if (inner) {
4273                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
4274                                          inner_headers);
4275                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
4276         } else {
4277                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
4278                                          outer_headers);
4279                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
4280         }
4281         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
4282         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
4283         if (!icmp_v)
4284                 return;
4285         if (!icmp_m)
4286                 icmp_m = &rte_flow_item_icmp_mask;
4287         MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
4288                  icmp_m->hdr.icmp_type);
4289         MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
4290                  icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
4291         MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
4292                  icmp_m->hdr.icmp_code);
4293         MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
4294                  icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
4295 }
4296
4297 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
4298
4299 #define HEADER_IS_ZERO(match_criteria, headers)                              \
4300         !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers),     \
4301                  matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
4302
4303 /**
4304  * Calculate flow matcher enable bitmap.
4305  *
4306  * @param match_criteria
4307  *   Pointer to flow matcher criteria.
4308  *
4309  * @return
4310  *   Bitmap of enabled fields.
4311  */
4312 static uint8_t
4313 flow_dv_matcher_enable(uint32_t *match_criteria)
4314 {
4315         uint8_t match_criteria_enable;
4316
4317         match_criteria_enable =
4318                 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
4319                 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
4320         match_criteria_enable |=
4321                 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
4322                 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
4323         match_criteria_enable |=
4324                 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
4325                 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
4326         match_criteria_enable |=
4327                 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
4328                 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
4329         match_criteria_enable |=
4330                 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
4331                 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
4332         return match_criteria_enable;
4333 }
4334
4335
4336 /**
4337  * Get a flow table.
4338  *
4339  * @param dev[in, out]
4340  *   Pointer to rte_eth_dev structure.
4341  * @param[in] table_id
4342  *   Table id to use.
4343  * @param[in] egress
4344  *   Direction of the table.
4345  * @param[in] transfer
4346  *   E-Switch or NIC flow.
4347  * @param[out] error
4348  *   pointer to error structure.
4349  *
4350  * @return
4351  *   Returns tables resource based on the index, NULL in case of failed.
4352  */
4353 static struct mlx5_flow_tbl_resource *
4354 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
4355                          uint32_t table_id, uint8_t egress,
4356                          uint8_t transfer,
4357                          struct rte_flow_error *error)
4358 {
4359         struct mlx5_priv *priv = dev->data->dev_private;
4360         struct mlx5_ibv_shared *sh = priv->sh;
4361         struct mlx5_flow_tbl_resource *tbl;
4362
4363 #ifdef HAVE_MLX5DV_DR
4364         if (transfer) {
4365                 tbl = &sh->fdb_tbl[table_id];
4366                 if (!tbl->obj)
4367                         tbl->obj = mlx5_glue->dr_create_flow_tbl
4368                                 (sh->fdb_domain, table_id);
4369         } else if (egress) {
4370                 tbl = &sh->tx_tbl[table_id];
4371                 if (!tbl->obj)
4372                         tbl->obj = mlx5_glue->dr_create_flow_tbl
4373                                 (sh->tx_domain, table_id);
4374         } else {
4375                 tbl = &sh->rx_tbl[table_id];
4376                 if (!tbl->obj)
4377                         tbl->obj = mlx5_glue->dr_create_flow_tbl
4378                                 (sh->rx_domain, table_id);
4379         }
4380         if (!tbl->obj) {
4381                 rte_flow_error_set(error, ENOMEM,
4382                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4383                                    NULL, "cannot create table");
4384                 return NULL;
4385         }
4386         rte_atomic32_inc(&tbl->refcnt);
4387         return tbl;
4388 #else
4389         (void)error;
4390         (void)tbl;
4391         if (transfer)
4392                 return &sh->fdb_tbl[table_id];
4393         else if (egress)
4394                 return &sh->tx_tbl[table_id];
4395         else
4396                 return &sh->rx_tbl[table_id];
4397 #endif
4398 }
4399
4400 /**
4401  * Release a flow table.
4402  *
4403  * @param[in] tbl
4404  *   Table resource to be released.
4405  *
4406  * @return
4407  *   Returns 0 if table was released, else return 1;
4408  */
4409 static int
4410 flow_dv_tbl_resource_release(struct mlx5_flow_tbl_resource *tbl)
4411 {
4412         if (!tbl)
4413                 return 0;
4414         if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
4415                 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
4416                 tbl->obj = NULL;
4417                 return 0;
4418         }
4419         return 1;
4420 }
4421
4422 /**
4423  * Register the flow matcher.
4424  *
4425  * @param dev[in, out]
4426  *   Pointer to rte_eth_dev structure.
4427  * @param[in, out] matcher
4428  *   Pointer to flow matcher.
4429  * @parm[in, out] dev_flow
4430  *   Pointer to the dev_flow.
4431  * @param[out] error
4432  *   pointer to error structure.
4433  *
4434  * @return
4435  *   0 on success otherwise -errno and errno is set.
4436  */
4437 static int
4438 flow_dv_matcher_register(struct rte_eth_dev *dev,
4439                          struct mlx5_flow_dv_matcher *matcher,
4440                          struct mlx5_flow *dev_flow,
4441                          struct rte_flow_error *error)
4442 {
4443         struct mlx5_priv *priv = dev->data->dev_private;
4444         struct mlx5_ibv_shared *sh = priv->sh;
4445         struct mlx5_flow_dv_matcher *cache_matcher;
4446         struct mlx5dv_flow_matcher_attr dv_attr = {
4447                 .type = IBV_FLOW_ATTR_NORMAL,
4448                 .match_mask = (void *)&matcher->mask,
4449         };
4450         struct mlx5_flow_tbl_resource *tbl = NULL;
4451
4452         /* Lookup from cache. */
4453         LIST_FOREACH(cache_matcher, &sh->matchers, next) {
4454                 if (matcher->crc == cache_matcher->crc &&
4455                     matcher->priority == cache_matcher->priority &&
4456                     matcher->egress == cache_matcher->egress &&
4457                     matcher->group == cache_matcher->group &&
4458                     matcher->transfer == cache_matcher->transfer &&
4459                     !memcmp((const void *)matcher->mask.buf,
4460                             (const void *)cache_matcher->mask.buf,
4461                             cache_matcher->mask.size)) {
4462                         DRV_LOG(DEBUG,
4463                                 "priority %hd use %s matcher %p: refcnt %d++",
4464                                 cache_matcher->priority,
4465                                 cache_matcher->egress ? "tx" : "rx",
4466                                 (void *)cache_matcher,
4467                                 rte_atomic32_read(&cache_matcher->refcnt));
4468                         rte_atomic32_inc(&cache_matcher->refcnt);
4469                         dev_flow->dv.matcher = cache_matcher;
4470                         return 0;
4471                 }
4472         }
4473         /* Register new matcher. */
4474         cache_matcher = rte_calloc(__func__, 1, sizeof(*cache_matcher), 0);
4475         if (!cache_matcher)
4476                 return rte_flow_error_set(error, ENOMEM,
4477                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4478                                           "cannot allocate matcher memory");
4479         tbl = flow_dv_tbl_resource_get(dev, matcher->group * MLX5_GROUP_FACTOR,
4480                                        matcher->egress, matcher->transfer,
4481                                        error);
4482         if (!tbl) {
4483                 rte_free(cache_matcher);
4484                 return rte_flow_error_set(error, ENOMEM,
4485                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4486                                           NULL, "cannot create table");
4487         }
4488         *cache_matcher = *matcher;
4489         dv_attr.match_criteria_enable =
4490                 flow_dv_matcher_enable(cache_matcher->mask.buf);
4491         dv_attr.priority = matcher->priority;
4492         if (matcher->egress)
4493                 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
4494         cache_matcher->matcher_object =
4495                 mlx5_glue->dv_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj);
4496         if (!cache_matcher->matcher_object) {
4497                 rte_free(cache_matcher);
4498 #ifdef HAVE_MLX5DV_DR
4499                 flow_dv_tbl_resource_release(tbl);
4500 #endif
4501                 return rte_flow_error_set(error, ENOMEM,
4502                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4503                                           NULL, "cannot create matcher");
4504         }
4505         rte_atomic32_inc(&cache_matcher->refcnt);
4506         LIST_INSERT_HEAD(&sh->matchers, cache_matcher, next);
4507         dev_flow->dv.matcher = cache_matcher;
4508         DRV_LOG(DEBUG, "priority %hd new %s matcher %p: refcnt %d",
4509                 cache_matcher->priority,
4510                 cache_matcher->egress ? "tx" : "rx", (void *)cache_matcher,
4511                 rte_atomic32_read(&cache_matcher->refcnt));
4512         rte_atomic32_inc(&tbl->refcnt);
4513         return 0;
4514 }
4515
4516 /**
4517  * Find existing tag resource or create and register a new one.
4518  *
4519  * @param dev[in, out]
4520  *   Pointer to rte_eth_dev structure.
4521  * @param[in, out] resource
4522  *   Pointer to tag resource.
4523  * @parm[in, out] dev_flow
4524  *   Pointer to the dev_flow.
4525  * @param[out] error
4526  *   pointer to error structure.
4527  *
4528  * @return
4529  *   0 on success otherwise -errno and errno is set.
4530  */
4531 static int
4532 flow_dv_tag_resource_register
4533                         (struct rte_eth_dev *dev,
4534                          struct mlx5_flow_dv_tag_resource *resource,
4535                          struct mlx5_flow *dev_flow,
4536                          struct rte_flow_error *error)
4537 {
4538         struct mlx5_priv *priv = dev->data->dev_private;
4539         struct mlx5_ibv_shared *sh = priv->sh;
4540         struct mlx5_flow_dv_tag_resource *cache_resource;
4541
4542         /* Lookup a matching resource from cache. */
4543         LIST_FOREACH(cache_resource, &sh->tags, next) {
4544                 if (resource->tag == cache_resource->tag) {
4545                         DRV_LOG(DEBUG, "tag resource %p: refcnt %d++",
4546                                 (void *)cache_resource,
4547                                 rte_atomic32_read(&cache_resource->refcnt));
4548                         rte_atomic32_inc(&cache_resource->refcnt);
4549                         dev_flow->flow->tag_resource = cache_resource;
4550                         return 0;
4551                 }
4552         }
4553         /* Register new  resource. */
4554         cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
4555         if (!cache_resource)
4556                 return rte_flow_error_set(error, ENOMEM,
4557                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4558                                           "cannot allocate resource memory");
4559         *cache_resource = *resource;
4560         cache_resource->action = mlx5_glue->dv_create_flow_action_tag
4561                 (resource->tag);
4562         if (!cache_resource->action) {
4563                 rte_free(cache_resource);
4564                 return rte_flow_error_set(error, ENOMEM,
4565                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4566                                           NULL, "cannot create action");
4567         }
4568         rte_atomic32_init(&cache_resource->refcnt);
4569         rte_atomic32_inc(&cache_resource->refcnt);
4570         LIST_INSERT_HEAD(&sh->tags, cache_resource, next);
4571         dev_flow->flow->tag_resource = cache_resource;
4572         DRV_LOG(DEBUG, "new tag resource %p: refcnt %d++",
4573                 (void *)cache_resource,
4574                 rte_atomic32_read(&cache_resource->refcnt));
4575         return 0;
4576 }
4577
4578 /**
4579  * Release the tag.
4580  *
4581  * @param dev
4582  *   Pointer to Ethernet device.
4583  * @param flow
4584  *   Pointer to mlx5_flow.
4585  *
4586  * @return
4587  *   1 while a reference on it exists, 0 when freed.
4588  */
4589 static int
4590 flow_dv_tag_release(struct rte_eth_dev *dev,
4591                     struct mlx5_flow_dv_tag_resource *tag)
4592 {
4593         assert(tag);
4594         DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
4595                 dev->data->port_id, (void *)tag,
4596                 rte_atomic32_read(&tag->refcnt));
4597         if (rte_atomic32_dec_and_test(&tag->refcnt)) {
4598                 claim_zero(mlx5_glue->destroy_flow_action(tag->action));
4599                 LIST_REMOVE(tag, next);
4600                 DRV_LOG(DEBUG, "port %u tag %p: removed",
4601                         dev->data->port_id, (void *)tag);
4602                 rte_free(tag);
4603                 return 0;
4604         }
4605         return 1;
4606 }
4607
4608 /**
4609  * Translate port ID action to vport.
4610  *
4611  * @param[in] dev
4612  *   Pointer to rte_eth_dev structure.
4613  * @param[in] action
4614  *   Pointer to the port ID action.
4615  * @param[out] dst_port_id
4616  *   The target port ID.
4617  * @param[out] error
4618  *   Pointer to the error structure.
4619  *
4620  * @return
4621  *   0 on success, a negative errno value otherwise and rte_errno is set.
4622  */
4623 static int
4624 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
4625                                  const struct rte_flow_action *action,
4626                                  uint32_t *dst_port_id,
4627                                  struct rte_flow_error *error)
4628 {
4629         uint32_t port;
4630         uint16_t port_id;
4631         int ret;
4632         const struct rte_flow_action_port_id *conf =
4633                         (const struct rte_flow_action_port_id *)action->conf;
4634
4635         port = conf->original ? dev->data->port_id : conf->id;
4636         ret = mlx5_port_to_eswitch_info(port, NULL, &port_id);
4637         if (ret)
4638                 return rte_flow_error_set(error, -ret,
4639                                           RTE_FLOW_ERROR_TYPE_ACTION,
4640                                           NULL,
4641                                           "No eswitch info was found for port");
4642         *dst_port_id = port_id;
4643         return 0;
4644 }
4645
4646 /**
4647  * Fill the flow with DV spec.
4648  *
4649  * @param[in] dev
4650  *   Pointer to rte_eth_dev structure.
4651  * @param[in, out] dev_flow
4652  *   Pointer to the sub flow.
4653  * @param[in] attr
4654  *   Pointer to the flow attributes.
4655  * @param[in] items
4656  *   Pointer to the list of items.
4657  * @param[in] actions
4658  *   Pointer to the list of actions.
4659  * @param[out] error
4660  *   Pointer to the error structure.
4661  *
4662  * @return
4663  *   0 on success, a negative errno value otherwise and rte_errno is set.
4664  */
4665 static int
4666 flow_dv_translate(struct rte_eth_dev *dev,
4667                   struct mlx5_flow *dev_flow,
4668                   const struct rte_flow_attr *attr,
4669                   const struct rte_flow_item items[],
4670                   const struct rte_flow_action actions[],
4671                   struct rte_flow_error *error)
4672 {
4673         struct mlx5_priv *priv = dev->data->dev_private;
4674         struct rte_flow *flow = dev_flow->flow;
4675         uint64_t item_flags = 0;
4676         uint64_t last_item = 0;
4677         uint64_t action_flags = 0;
4678         uint64_t priority = attr->priority;
4679         struct mlx5_flow_dv_matcher matcher = {
4680                 .mask = {
4681                         .size = sizeof(matcher.mask.buf),
4682                 },
4683         };
4684         int actions_n = 0;
4685         bool actions_end = false;
4686         struct mlx5_flow_dv_modify_hdr_resource res = {
4687                 .ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
4688                                           MLX5DV_FLOW_TABLE_TYPE_NIC_RX
4689         };
4690         union flow_dv_attr flow_attr = { .attr = 0 };
4691         struct mlx5_flow_dv_tag_resource tag_resource;
4692         uint32_t modify_action_position = UINT32_MAX;
4693         void *match_mask = matcher.mask.buf;
4694         void *match_value = dev_flow->dv.value.buf;
4695         uint8_t next_protocol = 0xff;
4696
4697         flow->group = attr->group;
4698         if (attr->transfer)
4699                 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
4700         if (priority == MLX5_FLOW_PRIO_RSVD)
4701                 priority = priv->config.flow_prio - 1;
4702         for (; !actions_end ; actions++) {
4703                 const struct rte_flow_action_queue *queue;
4704                 const struct rte_flow_action_rss *rss;
4705                 const struct rte_flow_action *action = actions;
4706                 const struct rte_flow_action_count *count = action->conf;
4707                 const uint8_t *rss_key;
4708                 const struct rte_flow_action_jump *jump_data;
4709                 struct mlx5_flow_dv_jump_tbl_resource jump_tbl_resource;
4710                 struct mlx5_flow_tbl_resource *tbl;
4711                 uint32_t port_id = 0;
4712                 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
4713
4714                 switch (actions->type) {
4715                 case RTE_FLOW_ACTION_TYPE_VOID:
4716                         break;
4717                 case RTE_FLOW_ACTION_TYPE_PORT_ID:
4718                         if (flow_dv_translate_action_port_id(dev, action,
4719                                                              &port_id, error))
4720                                 return -rte_errno;
4721                         port_id_resource.port_id = port_id;
4722                         if (flow_dv_port_id_action_resource_register
4723                             (dev, &port_id_resource, dev_flow, error))
4724                                 return -rte_errno;
4725                         dev_flow->dv.actions[actions_n++] =
4726                                 dev_flow->dv.port_id_action->action;
4727                         action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4728                         break;
4729                 case RTE_FLOW_ACTION_TYPE_FLAG:
4730                         tag_resource.tag =
4731                                 mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
4732                         if (!flow->tag_resource)
4733                                 if (flow_dv_tag_resource_register
4734                                     (dev, &tag_resource, dev_flow, error))
4735                                         return errno;
4736                         dev_flow->dv.actions[actions_n++] =
4737                                 flow->tag_resource->action;
4738                         action_flags |= MLX5_FLOW_ACTION_FLAG;
4739                         break;
4740                 case RTE_FLOW_ACTION_TYPE_MARK:
4741                         tag_resource.tag = mlx5_flow_mark_set
4742                               (((const struct rte_flow_action_mark *)
4743                                (actions->conf))->id);
4744                         if (!flow->tag_resource)
4745                                 if (flow_dv_tag_resource_register
4746                                     (dev, &tag_resource, dev_flow, error))
4747                                         return errno;
4748                         dev_flow->dv.actions[actions_n++] =
4749                                 flow->tag_resource->action;
4750                         action_flags |= MLX5_FLOW_ACTION_MARK;
4751                         break;
4752                 case RTE_FLOW_ACTION_TYPE_DROP:
4753                         action_flags |= MLX5_FLOW_ACTION_DROP;
4754                         break;
4755                 case RTE_FLOW_ACTION_TYPE_QUEUE:
4756                         queue = actions->conf;
4757                         flow->rss.queue_num = 1;
4758                         (*flow->queue)[0] = queue->index;
4759                         action_flags |= MLX5_FLOW_ACTION_QUEUE;
4760                         break;
4761                 case RTE_FLOW_ACTION_TYPE_RSS:
4762                         rss = actions->conf;
4763                         if (flow->queue)
4764                                 memcpy((*flow->queue), rss->queue,
4765                                        rss->queue_num * sizeof(uint16_t));
4766                         flow->rss.queue_num = rss->queue_num;
4767                         /* NULL RSS key indicates default RSS key. */
4768                         rss_key = !rss->key ? rss_hash_default_key : rss->key;
4769                         memcpy(flow->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
4770                         /* RSS type 0 indicates default RSS type ETH_RSS_IP. */
4771                         flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4772                         flow->rss.level = rss->level;
4773                         action_flags |= MLX5_FLOW_ACTION_RSS;
4774                         break;
4775                 case RTE_FLOW_ACTION_TYPE_COUNT:
4776                         if (!priv->config.devx) {
4777                                 rte_errno = ENOTSUP;
4778                                 goto cnt_err;
4779                         }
4780                         flow->counter = flow_dv_counter_alloc(dev,
4781                                                               count->shared,
4782                                                               count->id,
4783                                                               attr->group);
4784                         if (flow->counter == NULL)
4785                                 goto cnt_err;
4786                         dev_flow->dv.actions[actions_n++] =
4787                                 flow->counter->action;
4788                         action_flags |= MLX5_FLOW_ACTION_COUNT;
4789                         break;
4790 cnt_err:
4791                         if (rte_errno == ENOTSUP)
4792                                 return rte_flow_error_set
4793                                               (error, ENOTSUP,
4794                                                RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4795                                                NULL,
4796                                                "count action not supported");
4797                         else
4798                                 return rte_flow_error_set
4799                                                 (error, rte_errno,
4800                                                  RTE_FLOW_ERROR_TYPE_ACTION,
4801                                                  action,
4802                                                  "cannot create counter"
4803                                                   " object.");
4804                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4805                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4806                         if (flow_dv_create_action_l2_encap(dev, actions,
4807                                                            dev_flow,
4808                                                            attr->transfer,
4809                                                            error))
4810                                 return -rte_errno;
4811                         dev_flow->dv.actions[actions_n++] =
4812                                 dev_flow->dv.encap_decap->verbs_action;
4813                         action_flags |= actions->type ==
4814                                         RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP ?
4815                                         MLX5_FLOW_ACTION_VXLAN_ENCAP :
4816                                         MLX5_FLOW_ACTION_NVGRE_ENCAP;
4817                         break;
4818                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4819                 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4820                         if (flow_dv_create_action_l2_decap(dev, dev_flow,
4821                                                            attr->transfer,
4822                                                            error))
4823                                 return -rte_errno;
4824                         dev_flow->dv.actions[actions_n++] =
4825                                 dev_flow->dv.encap_decap->verbs_action;
4826                         action_flags |= actions->type ==
4827                                         RTE_FLOW_ACTION_TYPE_VXLAN_DECAP ?
4828                                         MLX5_FLOW_ACTION_VXLAN_DECAP :
4829                                         MLX5_FLOW_ACTION_NVGRE_DECAP;
4830                         break;
4831                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4832                         /* Handle encap with preceding decap. */
4833                         if (action_flags & MLX5_FLOW_ACTION_RAW_DECAP) {
4834                                 if (flow_dv_create_action_raw_encap
4835                                         (dev, actions, dev_flow, attr, error))
4836                                         return -rte_errno;
4837                                 dev_flow->dv.actions[actions_n++] =
4838                                         dev_flow->dv.encap_decap->verbs_action;
4839                         } else {
4840                                 /* Handle encap without preceding decap. */
4841                                 if (flow_dv_create_action_l2_encap
4842                                     (dev, actions, dev_flow, attr->transfer,
4843                                      error))
4844                                         return -rte_errno;
4845                                 dev_flow->dv.actions[actions_n++] =
4846                                         dev_flow->dv.encap_decap->verbs_action;
4847                         }
4848                         action_flags |= MLX5_FLOW_ACTION_RAW_ENCAP;
4849                         break;
4850                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4851                         /* Check if this decap is followed by encap. */
4852                         for (; action->type != RTE_FLOW_ACTION_TYPE_END &&
4853                                action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP;
4854                                action++) {
4855                         }
4856                         /* Handle decap only if it isn't followed by encap. */
4857                         if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
4858                                 if (flow_dv_create_action_l2_decap
4859                                     (dev, dev_flow, attr->transfer, error))
4860                                         return -rte_errno;
4861                                 dev_flow->dv.actions[actions_n++] =
4862                                         dev_flow->dv.encap_decap->verbs_action;
4863                         }
4864                         /* If decap is followed by encap, handle it at encap. */
4865                         action_flags |= MLX5_FLOW_ACTION_RAW_DECAP;
4866                         break;
4867                 case RTE_FLOW_ACTION_TYPE_JUMP:
4868                         jump_data = action->conf;
4869                         tbl = flow_dv_tbl_resource_get(dev, jump_data->group *
4870                                                        MLX5_GROUP_FACTOR,
4871                                                        attr->egress,
4872                                                        attr->transfer, error);
4873                         if (!tbl)
4874                                 return rte_flow_error_set
4875                                                 (error, errno,
4876                                                  RTE_FLOW_ERROR_TYPE_ACTION,
4877                                                  NULL,
4878                                                  "cannot create jump action.");
4879                         jump_tbl_resource.tbl = tbl;
4880                         if (flow_dv_jump_tbl_resource_register
4881                             (dev, &jump_tbl_resource, dev_flow, error)) {
4882                                 flow_dv_tbl_resource_release(tbl);
4883                                 return rte_flow_error_set
4884                                                 (error, errno,
4885                                                  RTE_FLOW_ERROR_TYPE_ACTION,
4886                                                  NULL,
4887                                                  "cannot create jump action.");
4888                         }
4889                         dev_flow->dv.actions[actions_n++] =
4890                                 dev_flow->dv.jump->action;
4891                         action_flags |= MLX5_FLOW_ACTION_JUMP;
4892                         break;
4893                 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4894                 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4895                         if (flow_dv_convert_action_modify_mac(&res, actions,
4896                                                               error))
4897                                 return -rte_errno;
4898                         action_flags |= actions->type ==
4899                                         RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
4900                                         MLX5_FLOW_ACTION_SET_MAC_SRC :
4901                                         MLX5_FLOW_ACTION_SET_MAC_DST;
4902                         break;
4903                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4904                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4905                         if (flow_dv_convert_action_modify_ipv4(&res, actions,
4906                                                                error))
4907                                 return -rte_errno;
4908                         action_flags |= actions->type ==
4909                                         RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
4910                                         MLX5_FLOW_ACTION_SET_IPV4_SRC :
4911                                         MLX5_FLOW_ACTION_SET_IPV4_DST;
4912                         break;
4913                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4914                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4915                         if (flow_dv_convert_action_modify_ipv6(&res, actions,
4916                                                                error))
4917                                 return -rte_errno;
4918                         action_flags |= actions->type ==
4919                                         RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
4920                                         MLX5_FLOW_ACTION_SET_IPV6_SRC :
4921                                         MLX5_FLOW_ACTION_SET_IPV6_DST;
4922                         break;
4923                 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4924                 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4925                         if (flow_dv_convert_action_modify_tp(&res, actions,
4926                                                              items, &flow_attr,
4927                                                              error))
4928                                 return -rte_errno;
4929                         action_flags |= actions->type ==
4930                                         RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
4931                                         MLX5_FLOW_ACTION_SET_TP_SRC :
4932                                         MLX5_FLOW_ACTION_SET_TP_DST;
4933                         break;
4934                 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4935                         if (flow_dv_convert_action_modify_dec_ttl(&res, items,
4936                                                                   &flow_attr,
4937                                                                   error))
4938                                 return -rte_errno;
4939                         action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
4940                         break;
4941                 case RTE_FLOW_ACTION_TYPE_SET_TTL:
4942                         if (flow_dv_convert_action_modify_ttl(&res, actions,
4943                                                              items, &flow_attr,
4944                                                              error))
4945                                 return -rte_errno;
4946                         action_flags |= MLX5_FLOW_ACTION_SET_TTL;
4947                         break;
4948                 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4949                 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4950                         if (flow_dv_convert_action_modify_tcp_seq(&res, actions,
4951                                                                   error))
4952                                 return -rte_errno;
4953                         action_flags |= actions->type ==
4954                                         RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
4955                                         MLX5_FLOW_ACTION_INC_TCP_SEQ :
4956                                         MLX5_FLOW_ACTION_DEC_TCP_SEQ;
4957                         break;
4958
4959                 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4960                 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4961                         if (flow_dv_convert_action_modify_tcp_ack(&res, actions,
4962                                                                   error))
4963                                 return -rte_errno;
4964                         action_flags |= actions->type ==
4965                                         RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
4966                                         MLX5_FLOW_ACTION_INC_TCP_ACK :
4967                                         MLX5_FLOW_ACTION_DEC_TCP_ACK;
4968                         break;
4969                 case RTE_FLOW_ACTION_TYPE_END:
4970                         actions_end = true;
4971                         if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS) {
4972                                 /* create modify action if needed. */
4973                                 if (flow_dv_modify_hdr_resource_register
4974                                                                 (dev, &res,
4975                                                                  dev_flow,
4976                                                                  error))
4977                                         return -rte_errno;
4978                                 dev_flow->dv.actions[modify_action_position] =
4979                                         dev_flow->dv.modify_hdr->verbs_action;
4980                         }
4981                         break;
4982                 default:
4983                         break;
4984                 }
4985                 if ((action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS) &&
4986                     modify_action_position == UINT32_MAX)
4987                         modify_action_position = actions_n++;
4988         }
4989         dev_flow->dv.actions_n = actions_n;
4990         flow->actions = action_flags;
4991         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4992                 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
4993
4994                 switch (items->type) {
4995                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4996                         flow_dv_translate_item_port_id(dev, match_mask,
4997                                                        match_value, items);
4998                         last_item = MLX5_FLOW_ITEM_PORT_ID;
4999                         break;
5000                 case RTE_FLOW_ITEM_TYPE_ETH:
5001                         flow_dv_translate_item_eth(match_mask, match_value,
5002                                                    items, tunnel);
5003                         matcher.priority = MLX5_PRIORITY_MAP_L2;
5004                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
5005                                              MLX5_FLOW_LAYER_OUTER_L2;
5006                         break;
5007                 case RTE_FLOW_ITEM_TYPE_VLAN:
5008                         flow_dv_translate_item_vlan(dev_flow,
5009                                                     match_mask, match_value,
5010                                                     items, tunnel);
5011                         matcher.priority = MLX5_PRIORITY_MAP_L2;
5012                         last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
5013                                               MLX5_FLOW_LAYER_INNER_VLAN) :
5014                                              (MLX5_FLOW_LAYER_OUTER_L2 |
5015                                               MLX5_FLOW_LAYER_OUTER_VLAN);
5016                         break;
5017                 case RTE_FLOW_ITEM_TYPE_IPV4:
5018                         mlx5_flow_tunnel_ip_check(items, next_protocol,
5019                                                   &item_flags, &tunnel);
5020                         flow_dv_translate_item_ipv4(match_mask, match_value,
5021                                                     items, tunnel, attr->group);
5022                         matcher.priority = MLX5_PRIORITY_MAP_L3;
5023                         dev_flow->dv.hash_fields |=
5024                                 mlx5_flow_hashfields_adjust
5025                                         (dev_flow, tunnel,
5026                                          MLX5_IPV4_LAYER_TYPES,
5027                                          MLX5_IPV4_IBV_RX_HASH);
5028                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
5029                                              MLX5_FLOW_LAYER_OUTER_L3_IPV4;
5030                         if (items->mask != NULL &&
5031                             ((const struct rte_flow_item_ipv4 *)
5032                              items->mask)->hdr.next_proto_id) {
5033                                 next_protocol =
5034                                         ((const struct rte_flow_item_ipv4 *)
5035                                          (items->spec))->hdr.next_proto_id;
5036                                 next_protocol &=
5037                                         ((const struct rte_flow_item_ipv4 *)
5038                                          (items->mask))->hdr.next_proto_id;
5039                         } else {
5040                                 /* Reset for inner layer. */
5041                                 next_protocol = 0xff;
5042                         }
5043                         break;
5044                 case RTE_FLOW_ITEM_TYPE_IPV6:
5045                         mlx5_flow_tunnel_ip_check(items, next_protocol,
5046                                                   &item_flags, &tunnel);
5047                         flow_dv_translate_item_ipv6(match_mask, match_value,
5048                                                     items, tunnel, attr->group);
5049                         matcher.priority = MLX5_PRIORITY_MAP_L3;
5050                         dev_flow->dv.hash_fields |=
5051                                 mlx5_flow_hashfields_adjust
5052                                         (dev_flow, tunnel,
5053                                          MLX5_IPV6_LAYER_TYPES,
5054                                          MLX5_IPV6_IBV_RX_HASH);
5055                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
5056                                              MLX5_FLOW_LAYER_OUTER_L3_IPV6;
5057                         if (items->mask != NULL &&
5058                             ((const struct rte_flow_item_ipv6 *)
5059                              items->mask)->hdr.proto) {
5060                                 next_protocol =
5061                                         ((const struct rte_flow_item_ipv6 *)
5062                                          items->spec)->hdr.proto;
5063                                 next_protocol &=
5064                                         ((const struct rte_flow_item_ipv6 *)
5065                                          items->mask)->hdr.proto;
5066                         } else {
5067                                 /* Reset for inner layer. */
5068                                 next_protocol = 0xff;
5069                         }
5070                         break;
5071                 case RTE_FLOW_ITEM_TYPE_TCP:
5072                         flow_dv_translate_item_tcp(match_mask, match_value,
5073                                                    items, tunnel);
5074                         matcher.priority = MLX5_PRIORITY_MAP_L4;
5075                         dev_flow->dv.hash_fields |=
5076                                 mlx5_flow_hashfields_adjust
5077                                         (dev_flow, tunnel, ETH_RSS_TCP,
5078                                          IBV_RX_HASH_SRC_PORT_TCP |
5079                                          IBV_RX_HASH_DST_PORT_TCP);
5080                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
5081                                              MLX5_FLOW_LAYER_OUTER_L4_TCP;
5082                         break;
5083                 case RTE_FLOW_ITEM_TYPE_UDP:
5084                         flow_dv_translate_item_udp(match_mask, match_value,
5085                                                    items, tunnel);
5086                         matcher.priority = MLX5_PRIORITY_MAP_L4;
5087                         dev_flow->dv.hash_fields |=
5088                                 mlx5_flow_hashfields_adjust
5089                                         (dev_flow, tunnel, ETH_RSS_UDP,
5090                                          IBV_RX_HASH_SRC_PORT_UDP |
5091                                          IBV_RX_HASH_DST_PORT_UDP);
5092                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
5093                                              MLX5_FLOW_LAYER_OUTER_L4_UDP;
5094                         break;
5095                 case RTE_FLOW_ITEM_TYPE_GRE:
5096                         flow_dv_translate_item_gre(match_mask, match_value,
5097                                                    items, tunnel);
5098                         last_item = MLX5_FLOW_LAYER_GRE;
5099                         break;
5100                 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
5101                         flow_dv_translate_item_gre_key(match_mask,
5102                                                        match_value, items);
5103                         last_item = MLX5_FLOW_LAYER_GRE_KEY;
5104                         break;
5105                 case RTE_FLOW_ITEM_TYPE_NVGRE:
5106                         flow_dv_translate_item_nvgre(match_mask, match_value,
5107                                                      items, tunnel);
5108                         last_item = MLX5_FLOW_LAYER_GRE;
5109                         break;
5110                 case RTE_FLOW_ITEM_TYPE_VXLAN:
5111                         flow_dv_translate_item_vxlan(match_mask, match_value,
5112                                                      items, tunnel);
5113                         last_item = MLX5_FLOW_LAYER_VXLAN;
5114                         break;
5115                 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
5116                         flow_dv_translate_item_vxlan(match_mask, match_value,
5117                                                      items, tunnel);
5118                         last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
5119                         break;
5120                 case RTE_FLOW_ITEM_TYPE_MPLS:
5121                         flow_dv_translate_item_mpls(match_mask, match_value,
5122                                                     items, last_item, tunnel);
5123                         last_item = MLX5_FLOW_LAYER_MPLS;
5124                         break;
5125                 case RTE_FLOW_ITEM_TYPE_META:
5126                         flow_dv_translate_item_meta(match_mask, match_value,
5127                                                     items);
5128                         last_item = MLX5_FLOW_ITEM_METADATA;
5129                         break;
5130                 case RTE_FLOW_ITEM_TYPE_ICMP:
5131                         flow_dv_translate_item_icmp(match_mask, match_value,
5132                                                     items, tunnel);
5133                         last_item = MLX5_FLOW_LAYER_ICMP;
5134                         break;
5135                 case RTE_FLOW_ITEM_TYPE_ICMP6:
5136                         flow_dv_translate_item_icmp6(match_mask, match_value,
5137                                                       items, tunnel);
5138                         last_item = MLX5_FLOW_LAYER_ICMP6;
5139                         break;
5140                 default:
5141                         break;
5142                 }
5143                 item_flags |= last_item;
5144         }
5145         /*
5146          * In case of ingress traffic when E-Switch mode is enabled,
5147          * we have two cases where we need to set the source port manually.
5148          * The first one, is in case of Nic steering rule, and the second is
5149          * E-Switch rule where no port_id item was found. In both cases
5150          * the source port is set according the current port in use.
5151          */
5152         if ((attr->ingress && !(item_flags & MLX5_FLOW_ITEM_PORT_ID)) &&
5153             (priv->representor || priv->master)) {
5154                 if (flow_dv_translate_item_port_id(dev, match_mask,
5155                                                    match_value, NULL))
5156                         return -rte_errno;
5157         }
5158         assert(!flow_dv_check_valid_spec(matcher.mask.buf,
5159                                          dev_flow->dv.value.buf));
5160         dev_flow->layers = item_flags;
5161         /* Register matcher. */
5162         matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
5163                                     matcher.mask.size);
5164         matcher.priority = mlx5_flow_adjust_priority(dev, priority,
5165                                                      matcher.priority);
5166         matcher.egress = attr->egress;
5167         matcher.group = attr->group;
5168         matcher.transfer = attr->transfer;
5169         if (flow_dv_matcher_register(dev, &matcher, dev_flow, error))
5170                 return -rte_errno;
5171         return 0;
5172 }
5173
5174 /**
5175  * Apply the flow to the NIC.
5176  *
5177  * @param[in] dev
5178  *   Pointer to the Ethernet device structure.
5179  * @param[in, out] flow
5180  *   Pointer to flow structure.
5181  * @param[out] error
5182  *   Pointer to error structure.
5183  *
5184  * @return
5185  *   0 on success, a negative errno value otherwise and rte_errno is set.
5186  */
5187 static int
5188 flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
5189               struct rte_flow_error *error)
5190 {
5191         struct mlx5_flow_dv *dv;
5192         struct mlx5_flow *dev_flow;
5193         struct mlx5_priv *priv = dev->data->dev_private;
5194         int n;
5195         int err;
5196
5197         LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
5198                 dv = &dev_flow->dv;
5199                 n = dv->actions_n;
5200                 if (flow->actions & MLX5_FLOW_ACTION_DROP) {
5201                         if (flow->transfer) {
5202                                 dv->actions[n++] = priv->sh->esw_drop_action;
5203                         } else {
5204                                 dv->hrxq = mlx5_hrxq_drop_new(dev);
5205                                 if (!dv->hrxq) {
5206                                         rte_flow_error_set
5207                                                 (error, errno,
5208                                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5209                                                  NULL,
5210                                                  "cannot get drop hash queue");
5211                                         goto error;
5212                                 }
5213                                 dv->actions[n++] = dv->hrxq->action;
5214                         }
5215                 } else if (flow->actions &
5216                            (MLX5_FLOW_ACTION_QUEUE | MLX5_FLOW_ACTION_RSS)) {
5217                         struct mlx5_hrxq *hrxq;
5218
5219                         hrxq = mlx5_hrxq_get(dev, flow->key,
5220                                              MLX5_RSS_HASH_KEY_LEN,
5221                                              dv->hash_fields,
5222                                              (*flow->queue),
5223                                              flow->rss.queue_num);
5224                         if (!hrxq) {
5225                                 hrxq = mlx5_hrxq_new
5226                                         (dev, flow->key, MLX5_RSS_HASH_KEY_LEN,
5227                                          dv->hash_fields, (*flow->queue),
5228                                          flow->rss.queue_num,
5229                                          !!(dev_flow->layers &
5230                                             MLX5_FLOW_LAYER_TUNNEL));
5231                         }
5232                         if (!hrxq) {
5233                                 rte_flow_error_set
5234                                         (error, rte_errno,
5235                                          RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5236                                          "cannot get hash queue");
5237                                 goto error;
5238                         }
5239                         dv->hrxq = hrxq;
5240                         dv->actions[n++] = dv->hrxq->action;
5241                 }
5242                 dv->flow =
5243                         mlx5_glue->dv_create_flow(dv->matcher->matcher_object,
5244                                                   (void *)&dv->value, n,
5245                                                   dv->actions);
5246                 if (!dv->flow) {
5247                         rte_flow_error_set(error, errno,
5248                                            RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5249                                            NULL,
5250                                            "hardware refuses to create flow");
5251                         goto error;
5252                 }
5253                 if (priv->vmwa_context &&
5254                     dev_flow->dv.vf_vlan.tag &&
5255                     !dev_flow->dv.vf_vlan.created) {
5256                         /*
5257                          * The rule contains the VLAN pattern.
5258                          * For VF we are going to create VLAN
5259                          * interface to make hypervisor set correct
5260                          * e-Switch vport context.
5261                          */
5262                         mlx5_vlan_vmwa_acquire(dev, &dev_flow->dv.vf_vlan);
5263                 }
5264         }
5265         return 0;
5266 error:
5267         err = rte_errno; /* Save rte_errno before cleanup. */
5268         LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
5269                 struct mlx5_flow_dv *dv = &dev_flow->dv;
5270                 if (dv->hrxq) {
5271                         if (flow->actions & MLX5_FLOW_ACTION_DROP)
5272                                 mlx5_hrxq_drop_release(dev);
5273                         else
5274                                 mlx5_hrxq_release(dev, dv->hrxq);
5275                         dv->hrxq = NULL;
5276                 }
5277                 if (dev_flow->dv.vf_vlan.tag &&
5278                     dev_flow->dv.vf_vlan.created)
5279                         mlx5_vlan_vmwa_release(dev, &dev_flow->dv.vf_vlan);
5280         }
5281         rte_errno = err; /* Restore rte_errno. */
5282         return -rte_errno;
5283 }
5284
5285 /**
5286  * Release the flow matcher.
5287  *
5288  * @param dev
5289  *   Pointer to Ethernet device.
5290  * @param flow
5291  *   Pointer to mlx5_flow.
5292  *
5293  * @return
5294  *   1 while a reference on it exists, 0 when freed.
5295  */
5296 static int
5297 flow_dv_matcher_release(struct rte_eth_dev *dev,
5298                         struct mlx5_flow *flow)
5299 {
5300         struct mlx5_flow_dv_matcher *matcher = flow->dv.matcher;
5301         struct mlx5_priv *priv = dev->data->dev_private;
5302         struct mlx5_ibv_shared *sh = priv->sh;
5303         struct mlx5_flow_tbl_resource *tbl;
5304
5305         assert(matcher->matcher_object);
5306         DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
5307                 dev->data->port_id, (void *)matcher,
5308                 rte_atomic32_read(&matcher->refcnt));
5309         if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
5310                 claim_zero(mlx5_glue->dv_destroy_flow_matcher
5311                            (matcher->matcher_object));
5312                 LIST_REMOVE(matcher, next);
5313                 if (matcher->egress)
5314                         tbl = &sh->tx_tbl[matcher->group];
5315                 else
5316                         tbl = &sh->rx_tbl[matcher->group];
5317                 flow_dv_tbl_resource_release(tbl);
5318                 rte_free(matcher);
5319                 DRV_LOG(DEBUG, "port %u matcher %p: removed",
5320                         dev->data->port_id, (void *)matcher);
5321                 return 0;
5322         }
5323         return 1;
5324 }
5325
5326 /**
5327  * Release an encap/decap resource.
5328  *
5329  * @param flow
5330  *   Pointer to mlx5_flow.
5331  *
5332  * @return
5333  *   1 while a reference on it exists, 0 when freed.
5334  */
5335 static int
5336 flow_dv_encap_decap_resource_release(struct mlx5_flow *flow)
5337 {
5338         struct mlx5_flow_dv_encap_decap_resource *cache_resource =
5339                                                 flow->dv.encap_decap;
5340
5341         assert(cache_resource->verbs_action);
5342         DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
5343                 (void *)cache_resource,
5344                 rte_atomic32_read(&cache_resource->refcnt));
5345         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
5346                 claim_zero(mlx5_glue->destroy_flow_action
5347                                 (cache_resource->verbs_action));
5348                 LIST_REMOVE(cache_resource, next);
5349                 rte_free(cache_resource);
5350                 DRV_LOG(DEBUG, "encap/decap resource %p: removed",
5351                         (void *)cache_resource);
5352                 return 0;
5353         }
5354         return 1;
5355 }
5356
5357 /**
5358  * Release an jump to table action resource.
5359  *
5360  * @param flow
5361  *   Pointer to mlx5_flow.
5362  *
5363  * @return
5364  *   1 while a reference on it exists, 0 when freed.
5365  */
5366 static int
5367 flow_dv_jump_tbl_resource_release(struct mlx5_flow *flow)
5368 {
5369         struct mlx5_flow_dv_jump_tbl_resource *cache_resource =
5370                                                 flow->dv.jump;
5371
5372         assert(cache_resource->action);
5373         DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
5374                 (void *)cache_resource,
5375                 rte_atomic32_read(&cache_resource->refcnt));
5376         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
5377                 claim_zero(mlx5_glue->destroy_flow_action
5378                                 (cache_resource->action));
5379                 LIST_REMOVE(cache_resource, next);
5380                 flow_dv_tbl_resource_release(cache_resource->tbl);
5381                 rte_free(cache_resource);
5382                 DRV_LOG(DEBUG, "jump table resource %p: removed",
5383                         (void *)cache_resource);
5384                 return 0;
5385         }
5386         return 1;
5387 }
5388
5389 /**
5390  * Release a modify-header resource.
5391  *
5392  * @param flow
5393  *   Pointer to mlx5_flow.
5394  *
5395  * @return
5396  *   1 while a reference on it exists, 0 when freed.
5397  */
5398 static int
5399 flow_dv_modify_hdr_resource_release(struct mlx5_flow *flow)
5400 {
5401         struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
5402                                                 flow->dv.modify_hdr;
5403
5404         assert(cache_resource->verbs_action);
5405         DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
5406                 (void *)cache_resource,
5407                 rte_atomic32_read(&cache_resource->refcnt));
5408         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
5409                 claim_zero(mlx5_glue->destroy_flow_action
5410                                 (cache_resource->verbs_action));
5411                 LIST_REMOVE(cache_resource, next);
5412                 rte_free(cache_resource);
5413                 DRV_LOG(DEBUG, "modify-header resource %p: removed",
5414                         (void *)cache_resource);
5415                 return 0;
5416         }
5417         return 1;
5418 }
5419
5420 /**
5421  * Release port ID action resource.
5422  *
5423  * @param flow
5424  *   Pointer to mlx5_flow.
5425  *
5426  * @return
5427  *   1 while a reference on it exists, 0 when freed.
5428  */
5429 static int
5430 flow_dv_port_id_action_resource_release(struct mlx5_flow *flow)
5431 {
5432         struct mlx5_flow_dv_port_id_action_resource *cache_resource =
5433                 flow->dv.port_id_action;
5434
5435         assert(cache_resource->action);
5436         DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
5437                 (void *)cache_resource,
5438                 rte_atomic32_read(&cache_resource->refcnt));
5439         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
5440                 claim_zero(mlx5_glue->destroy_flow_action
5441                                 (cache_resource->action));
5442                 LIST_REMOVE(cache_resource, next);
5443                 rte_free(cache_resource);
5444                 DRV_LOG(DEBUG, "port id action resource %p: removed",
5445                         (void *)cache_resource);
5446                 return 0;
5447         }
5448         return 1;
5449 }
5450
5451 /**
5452  * Remove the flow from the NIC but keeps it in memory.
5453  *
5454  * @param[in] dev
5455  *   Pointer to Ethernet device.
5456  * @param[in, out] flow
5457  *   Pointer to flow structure.
5458  */
5459 static void
5460 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
5461 {
5462         struct mlx5_flow_dv *dv;
5463         struct mlx5_flow *dev_flow;
5464
5465         if (!flow)
5466                 return;
5467         LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
5468                 dv = &dev_flow->dv;
5469                 if (dv->flow) {
5470                         claim_zero(mlx5_glue->dv_destroy_flow(dv->flow));
5471                         dv->flow = NULL;
5472                 }
5473                 if (dv->hrxq) {
5474                         if (flow->actions & MLX5_FLOW_ACTION_DROP)
5475                                 mlx5_hrxq_drop_release(dev);
5476                         else
5477                                 mlx5_hrxq_release(dev, dv->hrxq);
5478                         dv->hrxq = NULL;
5479                 }
5480                 if (dev_flow->dv.vf_vlan.tag &&
5481                     dev_flow->dv.vf_vlan.created)
5482                         mlx5_vlan_vmwa_release(dev, &dev_flow->dv.vf_vlan);
5483         }
5484 }
5485
5486 /**
5487  * Remove the flow from the NIC and the memory.
5488  *
5489  * @param[in] dev
5490  *   Pointer to the Ethernet device structure.
5491  * @param[in, out] flow
5492  *   Pointer to flow structure.
5493  */
5494 static void
5495 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
5496 {
5497         struct mlx5_flow *dev_flow;
5498
5499         if (!flow)
5500                 return;
5501         flow_dv_remove(dev, flow);
5502         if (flow->counter) {
5503                 flow_dv_counter_release(dev, flow->counter);
5504                 flow->counter = NULL;
5505         }
5506         if (flow->tag_resource) {
5507                 flow_dv_tag_release(dev, flow->tag_resource);
5508                 flow->tag_resource = NULL;
5509         }
5510         while (!LIST_EMPTY(&flow->dev_flows)) {
5511                 dev_flow = LIST_FIRST(&flow->dev_flows);
5512                 LIST_REMOVE(dev_flow, next);
5513                 if (dev_flow->dv.matcher)
5514                         flow_dv_matcher_release(dev, dev_flow);
5515                 if (dev_flow->dv.encap_decap)
5516                         flow_dv_encap_decap_resource_release(dev_flow);
5517                 if (dev_flow->dv.modify_hdr)
5518                         flow_dv_modify_hdr_resource_release(dev_flow);
5519                 if (dev_flow->dv.jump)
5520                         flow_dv_jump_tbl_resource_release(dev_flow);
5521                 if (dev_flow->dv.port_id_action)
5522                         flow_dv_port_id_action_resource_release(dev_flow);
5523                 rte_free(dev_flow);
5524         }
5525 }
5526
5527 /**
5528  * Query a dv flow  rule for its statistics via devx.
5529  *
5530  * @param[in] dev
5531  *   Pointer to Ethernet device.
5532  * @param[in] flow
5533  *   Pointer to the sub flow.
5534  * @param[out] data
5535  *   data retrieved by the query.
5536  * @param[out] error
5537  *   Perform verbose error reporting if not NULL.
5538  *
5539  * @return
5540  *   0 on success, a negative errno value otherwise and rte_errno is set.
5541  */
5542 static int
5543 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
5544                     void *data, struct rte_flow_error *error)
5545 {
5546         struct mlx5_priv *priv = dev->data->dev_private;
5547         struct rte_flow_query_count *qc = data;
5548
5549         if (!priv->config.devx)
5550                 return rte_flow_error_set(error, ENOTSUP,
5551                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5552                                           NULL,
5553                                           "counters are not supported");
5554         if (flow->counter) {
5555                 uint64_t pkts, bytes;
5556                 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
5557                                                &bytes);
5558
5559                 if (err)
5560                         return rte_flow_error_set(error, -err,
5561                                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5562                                         NULL, "cannot read counters");
5563                 qc->hits_set = 1;
5564                 qc->bytes_set = 1;
5565                 qc->hits = pkts - flow->counter->hits;
5566                 qc->bytes = bytes - flow->counter->bytes;
5567                 if (qc->reset) {
5568                         flow->counter->hits = pkts;
5569                         flow->counter->bytes = bytes;
5570                 }
5571                 return 0;
5572         }
5573         return rte_flow_error_set(error, EINVAL,
5574                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5575                                   NULL,
5576                                   "counters are not available");
5577 }
5578
5579 /**
5580  * Query a flow.
5581  *
5582  * @see rte_flow_query()
5583  * @see rte_flow_ops
5584  */
5585 static int
5586 flow_dv_query(struct rte_eth_dev *dev,
5587               struct rte_flow *flow __rte_unused,
5588               const struct rte_flow_action *actions __rte_unused,
5589               void *data __rte_unused,
5590               struct rte_flow_error *error __rte_unused)
5591 {
5592         int ret = -EINVAL;
5593
5594         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5595                 switch (actions->type) {
5596                 case RTE_FLOW_ACTION_TYPE_VOID:
5597                         break;
5598                 case RTE_FLOW_ACTION_TYPE_COUNT:
5599                         ret = flow_dv_query_count(dev, flow, data, error);
5600                         break;
5601                 default:
5602                         return rte_flow_error_set(error, ENOTSUP,
5603                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5604                                                   actions,
5605                                                   "action not supported");
5606                 }
5607         }
5608         return ret;
5609 }
5610
5611 /*
5612  * Mutex-protected thunk to flow_dv_translate().
5613  */
5614 static int
5615 flow_d_translate(struct rte_eth_dev *dev,
5616                  struct mlx5_flow *dev_flow,
5617                  const struct rte_flow_attr *attr,
5618                  const struct rte_flow_item items[],
5619                  const struct rte_flow_action actions[],
5620                  struct rte_flow_error *error)
5621 {
5622         int ret;
5623
5624         flow_d_shared_lock(dev);
5625         ret = flow_dv_translate(dev, dev_flow, attr, items, actions, error);
5626         flow_d_shared_unlock(dev);
5627         return ret;
5628 }
5629
5630 /*
5631  * Mutex-protected thunk to flow_dv_apply().
5632  */
5633 static int
5634 flow_d_apply(struct rte_eth_dev *dev,
5635              struct rte_flow *flow,
5636              struct rte_flow_error *error)
5637 {
5638         int ret;
5639
5640         flow_d_shared_lock(dev);
5641         ret = flow_dv_apply(dev, flow, error);
5642         flow_d_shared_unlock(dev);
5643         return ret;
5644 }
5645
5646 /*
5647  * Mutex-protected thunk to flow_dv_remove().
5648  */
5649 static void
5650 flow_d_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
5651 {
5652         flow_d_shared_lock(dev);
5653         flow_dv_remove(dev, flow);
5654         flow_d_shared_unlock(dev);
5655 }
5656
5657 /*
5658  * Mutex-protected thunk to flow_dv_destroy().
5659  */
5660 static void
5661 flow_d_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
5662 {
5663         flow_d_shared_lock(dev);
5664         flow_dv_destroy(dev, flow);
5665         flow_d_shared_unlock(dev);
5666 }
5667
5668 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
5669         .validate = flow_dv_validate,
5670         .prepare = flow_dv_prepare,
5671         .translate = flow_d_translate,
5672         .apply = flow_d_apply,
5673         .remove = flow_d_remove,
5674         .destroy = flow_d_destroy,
5675         .query = flow_dv_query,
5676 };
5677
5678 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */