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 #include <rte_common.h>
  12 #include <rte_ether.h>
  13 #include <rte_ethdev_driver.h>
  14 #include <rte_flow.h>
  15 #include <rte_flow_driver.h>
  16 #include <rte_malloc.h>
  17 #include <rte_cycles.h>
  18 #include <rte_ip.h>
  19 #include <rte_gre.h>
  20 #include <rte_vxlan.h>
  21 #include <rte_gtp.h>
  22 #include <rte_eal_paging.h>
  23 #include <rte_mpls.h>
  24
  25 #include <mlx5_glue.h>
  26 #include <mlx5_devx_cmds.h>
  27 #include <mlx5_prm.h>
  28 #include <mlx5_malloc.h>
  29
  30 #include "mlx5_defs.h"
  31 #include "mlx5.h"
  32 #include "mlx5_common_os.h"
  33 #include "mlx5_flow.h"
  34 #include "mlx5_flow_os.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 /* VLAN header definitions */
  54 #define MLX5DV_FLOW_VLAN_PCP_SHIFT 13
  55 #define MLX5DV_FLOW_VLAN_PCP_MASK (0x7 << MLX5DV_FLOW_VLAN_PCP_SHIFT)
  56 #define MLX5DV_FLOW_VLAN_VID_MASK 0x0fff
  57 #define MLX5DV_FLOW_VLAN_PCP_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK)
  58 #define MLX5DV_FLOW_VLAN_VID_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_VID_MASK)
  59
  60 union flow_dv_attr {
  61         struct {
  62                 uint32_t valid:1;
  63                 uint32_t ipv4:1;
  64                 uint32_t ipv6:1;
  65                 uint32_t tcp:1;
  66                 uint32_t udp:1;
  67                 uint32_t reserved:27;
  68         };
  69         uint32_t attr;
  70 };
  71
  72 static int
  73 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
  74                              struct mlx5_flow_tbl_resource *tbl);
  75
  76 static int
  77 flow_dv_default_miss_resource_release(struct rte_eth_dev *dev);
  78
  79 /**
  80  * Initialize flow attributes structure according to flow items' types.
  81  *
  82  * flow_dv_validate() avoids multiple L3/L4 layers cases other than tunnel
  83  * mode. For tunnel mode, the items to be modified are the outermost ones.
  84  *
  85  * @param[in] item
  86  *   Pointer to item specification.
  87  * @param[out] attr
  88  *   Pointer to flow attributes structure.
  89  * @param[in] dev_flow
  90  *   Pointer to the sub flow.
  91  * @param[in] tunnel_decap
  92  *   Whether action is after tunnel decapsulation.
  93  */
  94 static void
  95 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr,
  96                   struct mlx5_flow *dev_flow, bool tunnel_decap)
  97 {
  98         uint64_t layers = dev_flow->handle->layers;
  99
 100         /*
 101          * If layers is already initialized, it means this dev_flow is the
 102          * suffix flow, the layers flags is set by the prefix flow. Need to
 103          * use the layer flags from prefix flow as the suffix flow may not
 104          * have the user defined items as the flow is split.
 105          */
 106         if (layers) {
 107                 if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
 108                         attr->ipv4 = 1;
 109                 else if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV6)
 110                         attr->ipv6 = 1;
 111                 if (layers & MLX5_FLOW_LAYER_OUTER_L4_TCP)
 112                         attr->tcp = 1;
 113                 else if (layers & MLX5_FLOW_LAYER_OUTER_L4_UDP)
 114                         attr->udp = 1;
 115                 attr->valid = 1;
 116                 return;
 117         }
 118         for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
 119                 uint8_t next_protocol = 0xff;
 120                 switch (item->type) {
 121                 case RTE_FLOW_ITEM_TYPE_GRE:
 122                 case RTE_FLOW_ITEM_TYPE_NVGRE:
 123                 case RTE_FLOW_ITEM_TYPE_VXLAN:
 124                 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
 125                 case RTE_FLOW_ITEM_TYPE_GENEVE:
 126                 case RTE_FLOW_ITEM_TYPE_MPLS:
 127                         if (tunnel_decap)
 128                                 attr->attr = 0;
 129                         break;
 130                 case RTE_FLOW_ITEM_TYPE_IPV4:
 131                         if (!attr->ipv6)
 132                                 attr->ipv4 = 1;
 133                         if (item->mask != NULL &&
 134                             ((const struct rte_flow_item_ipv4 *)
 135                             item->mask)->hdr.next_proto_id)
 136                                 next_protocol =
 137                                     ((const struct rte_flow_item_ipv4 *)
 138                                       (item->spec))->hdr.next_proto_id &
 139                                     ((const struct rte_flow_item_ipv4 *)
 140                                       (item->mask))->hdr.next_proto_id;
 141                         if ((next_protocol == IPPROTO_IPIP ||
 142                             next_protocol == IPPROTO_IPV6) && tunnel_decap)
 143                                 attr->attr = 0;
 144                         break;
 145                 case RTE_FLOW_ITEM_TYPE_IPV6:
 146                         if (!attr->ipv4)
 147                                 attr->ipv6 = 1;
 148                         if (item->mask != NULL &&
 149                             ((const struct rte_flow_item_ipv6 *)
 150                             item->mask)->hdr.proto)
 151                                 next_protocol =
 152                                     ((const struct rte_flow_item_ipv6 *)
 153                                       (item->spec))->hdr.proto &
 154                                     ((const struct rte_flow_item_ipv6 *)
 155                                       (item->mask))->hdr.proto;
 156                         if ((next_protocol == IPPROTO_IPIP ||
 157                             next_protocol == IPPROTO_IPV6) && tunnel_decap)
 158                                 attr->attr = 0;
 159                         break;
 160                 case RTE_FLOW_ITEM_TYPE_UDP:
 161                         if (!attr->tcp)
 162                                 attr->udp = 1;
 163                         break;
 164                 case RTE_FLOW_ITEM_TYPE_TCP:
 165                         if (!attr->udp)
 166                                 attr->tcp = 1;
 167                         break;
 168                 default:
 169                         break;
 170                 }
 171         }
 172         attr->valid = 1;
 173 }
 174
 175 /**
 176  * Convert rte_mtr_color to mlx5 color.
 177  *
 178  * @param[in] rcol
 179  *   rte_mtr_color.
 180  *
 181  * @return
 182  *   mlx5 color.
 183  */
 184 static int
 185 rte_col_2_mlx5_col(enum rte_color rcol)
 186 {
 187         switch (rcol) {
 188         case RTE_COLOR_GREEN:
 189                 return MLX5_FLOW_COLOR_GREEN;
 190         case RTE_COLOR_YELLOW:
 191                 return MLX5_FLOW_COLOR_YELLOW;
 192         case RTE_COLOR_RED:
 193                 return MLX5_FLOW_COLOR_RED;
 194         default:
 195                 break;
 196         }
 197         return MLX5_FLOW_COLOR_UNDEFINED;
 198 }
 199
 200 struct field_modify_info {
 201         uint32_t size; /* Size of field in protocol header, in bytes. */
 202         uint32_t offset; /* Offset of field in protocol header, in bytes. */
 203         enum mlx5_modification_field id;
 204 };
 205
 206 struct field_modify_info modify_eth[] = {
 207         {4,  0, MLX5_MODI_OUT_DMAC_47_16},
 208         {2,  4, MLX5_MODI_OUT_DMAC_15_0},
 209         {4,  6, MLX5_MODI_OUT_SMAC_47_16},
 210         {2, 10, MLX5_MODI_OUT_SMAC_15_0},
 211         {0, 0, 0},
 212 };
 213
 214 struct field_modify_info modify_vlan_out_first_vid[] = {
 215         /* Size in bits !!! */
 216         {12, 0, MLX5_MODI_OUT_FIRST_VID},
 217         {0, 0, 0},
 218 };
 219
 220 struct field_modify_info modify_ipv4[] = {
 221         {1,  1, MLX5_MODI_OUT_IP_DSCP},
 222         {1,  8, MLX5_MODI_OUT_IPV4_TTL},
 223         {4, 12, MLX5_MODI_OUT_SIPV4},
 224         {4, 16, MLX5_MODI_OUT_DIPV4},
 225         {0, 0, 0},
 226 };
 227
 228 struct field_modify_info modify_ipv6[] = {
 229         {1,  0, MLX5_MODI_OUT_IP_DSCP},
 230         {1,  7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
 231         {4,  8, MLX5_MODI_OUT_SIPV6_127_96},
 232         {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
 233         {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
 234         {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
 235         {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
 236         {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
 237         {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
 238         {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
 239         {0, 0, 0},
 240 };
 241
 242 struct field_modify_info modify_udp[] = {
 243         {2, 0, MLX5_MODI_OUT_UDP_SPORT},
 244         {2, 2, MLX5_MODI_OUT_UDP_DPORT},
 245         {0, 0, 0},
 246 };
 247
 248 struct field_modify_info modify_tcp[] = {
 249         {2, 0, MLX5_MODI_OUT_TCP_SPORT},
 250         {2, 2, MLX5_MODI_OUT_TCP_DPORT},
 251         {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
 252         {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
 253         {0, 0, 0},
 254 };
 255
 256 static void
 257 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
 258                           uint8_t next_protocol, uint64_t *item_flags,
 259                           int *tunnel)
 260 {
 261         MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
 262                     item->type == RTE_FLOW_ITEM_TYPE_IPV6);
 263         if (next_protocol == IPPROTO_IPIP) {
 264                 *item_flags |= MLX5_FLOW_LAYER_IPIP;
 265                 *tunnel = 1;
 266         }
 267         if (next_protocol == IPPROTO_IPV6) {
 268                 *item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
 269                 *tunnel = 1;
 270         }
 271 }
 272
 273 /**
 274  * Acquire the synchronizing object to protect multithreaded access
 275  * to shared dv context. Lock occurs only if context is actually
 276  * shared, i.e. we have multiport IB device and representors are
 277  * created.
 278  *
 279  * @param[in] dev
 280  *   Pointer to the rte_eth_dev structure.
 281  */
 282 static void
 283 flow_dv_shared_lock(struct rte_eth_dev *dev)
 284 {
 285         struct mlx5_priv *priv = dev->data->dev_private;
 286         struct mlx5_dev_ctx_shared *sh = priv->sh;
 287
 288         if (sh->dv_refcnt > 1) {
 289                 int ret;
 290
 291                 ret = pthread_mutex_lock(&sh->dv_mutex);
 292                 MLX5_ASSERT(!ret);
 293                 (void)ret;
 294         }
 295 }
 296
 297 static void
 298 flow_dv_shared_unlock(struct rte_eth_dev *dev)
 299 {
 300         struct mlx5_priv *priv = dev->data->dev_private;
 301         struct mlx5_dev_ctx_shared *sh = priv->sh;
 302
 303         if (sh->dv_refcnt > 1) {
 304                 int ret;
 305
 306                 ret = pthread_mutex_unlock(&sh->dv_mutex);
 307                 MLX5_ASSERT(!ret);
 308                 (void)ret;
 309         }
 310 }
 311
 312 /* Update VLAN's VID/PCP based on input rte_flow_action.
 313  *
 314  * @param[in] action
 315  *   Pointer to struct rte_flow_action.
 316  * @param[out] vlan
 317  *   Pointer to struct rte_vlan_hdr.
 318  */
 319 static void
 320 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
 321                          struct rte_vlan_hdr *vlan)
 322 {
 323         uint16_t vlan_tci;
 324         if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
 325                 vlan_tci =
 326                     ((const struct rte_flow_action_of_set_vlan_pcp *)
 327                                                action->conf)->vlan_pcp;
 328                 vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
 329                 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
 330                 vlan->vlan_tci |= vlan_tci;
 331         } else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
 332                 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
 333                 vlan->vlan_tci |= rte_be_to_cpu_16
 334                     (((const struct rte_flow_action_of_set_vlan_vid *)
 335                                              action->conf)->vlan_vid);
 336         }
 337 }
 338
 339 /**
 340  * Fetch 1, 2, 3 or 4 byte field from the byte array
 341  * and return as unsigned integer in host-endian format.
 342  *
 343  * @param[in] data
 344  *   Pointer to data array.
 345  * @param[in] size
 346  *   Size of field to extract.
 347  *
 348  * @return
 349  *   converted field in host endian format.
 350  */
 351 static inline uint32_t
 352 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
 353 {
 354         uint32_t ret;
 355
 356         switch (size) {
 357         case 1:
 358                 ret = *data;
 359                 break;
 360         case 2:
 361                 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
 362                 break;
 363         case 3:
 364                 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
 365                 ret = (ret << 8) | *(data + sizeof(uint16_t));
 366                 break;
 367         case 4:
 368                 ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
 369                 break;
 370         default:
 371                 MLX5_ASSERT(false);
 372                 ret = 0;
 373                 break;
 374         }
 375         return ret;
 376 }
 377
 378 /**
 379  * Convert modify-header action to DV specification.
 380  *
 381  * Data length of each action is determined by provided field description
 382  * and the item mask. Data bit offset and width of each action is determined
 383  * by provided item mask.
 384  *
 385  * @param[in] item
 386  *   Pointer to item specification.
 387  * @param[in] field
 388  *   Pointer to field modification information.
 389  *     For MLX5_MODIFICATION_TYPE_SET specifies destination field.
 390  *     For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
 391  *     For MLX5_MODIFICATION_TYPE_COPY specifies source field.
 392  * @param[in] dcopy
 393  *   Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
 394  *   Negative offset value sets the same offset as source offset.
 395  *   size field is ignored, value is taken from source field.
 396  * @param[in,out] resource
 397  *   Pointer to the modify-header resource.
 398  * @param[in] type
 399  *   Type of modification.
 400  * @param[out] error
 401  *   Pointer to the error structure.
 402  *
 403  * @return
 404  *   0 on success, a negative errno value otherwise and rte_errno is set.
 405  */
 406 static int
 407 flow_dv_convert_modify_action(struct rte_flow_item *item,
 408                               struct field_modify_info *field,
 409                               struct field_modify_info *dcopy,
 410                               struct mlx5_flow_dv_modify_hdr_resource *resource,
 411                               uint32_t type, struct rte_flow_error *error)
 412 {
 413         uint32_t i = resource->actions_num;
 414         struct mlx5_modification_cmd *actions = resource->actions;
 415
 416         /*
 417          * The item and mask are provided in big-endian format.
 418          * The fields should be presented as in big-endian format either.
 419          * Mask must be always present, it defines the actual field width.
 420          */
 421         MLX5_ASSERT(item->mask);
 422         MLX5_ASSERT(field->size);
 423         do {
 424                 unsigned int size_b;
 425                 unsigned int off_b;
 426                 uint32_t mask;
 427                 uint32_t data;
 428
 429                 if (i >= MLX5_MAX_MODIFY_NUM)
 430                         return rte_flow_error_set(error, EINVAL,
 431                                  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 432                                  "too many items to modify");
 433                 /* Fetch variable byte size mask from the array. */
 434                 mask = flow_dv_fetch_field((const uint8_t *)item->mask +
 435                                            field->offset, field->size);
 436                 if (!mask) {
 437                         ++field;
 438                         continue;
 439                 }
 440                 /* Deduce actual data width in bits from mask value. */
 441                 off_b = rte_bsf32(mask);
 442                 size_b = sizeof(uint32_t) * CHAR_BIT -
 443                          off_b - __builtin_clz(mask);
 444                 MLX5_ASSERT(size_b);
 445                 size_b = size_b == sizeof(uint32_t) * CHAR_BIT ? 0 : size_b;
 446                 actions[i] = (struct mlx5_modification_cmd) {
 447                         .action_type = type,
 448                         .field = field->id,
 449                         .offset = off_b,
 450                         .length = size_b,
 451                 };
 452                 /* Convert entire record to expected big-endian format. */
 453                 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
 454                 if (type == MLX5_MODIFICATION_TYPE_COPY) {
 455                         MLX5_ASSERT(dcopy);
 456                         actions[i].dst_field = dcopy->id;
 457                         actions[i].dst_offset =
 458                                 (int)dcopy->offset < 0 ? off_b : dcopy->offset;
 459                         /* Convert entire record to big-endian format. */
 460                         actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
 461                 } else {
 462                         MLX5_ASSERT(item->spec);
 463                         data = flow_dv_fetch_field((const uint8_t *)item->spec +
 464                                                    field->offset, field->size);
 465                         /* Shift out the trailing masked bits from data. */
 466                         data = (data & mask) >> off_b;
 467                         actions[i].data1 = rte_cpu_to_be_32(data);
 468                 }
 469                 ++i;
 470                 ++field;
 471         } while (field->size);
 472         if (resource->actions_num == i)
 473                 return rte_flow_error_set(error, EINVAL,
 474                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 475                                           "invalid modification flow item");
 476         resource->actions_num = i;
 477         return 0;
 478 }
 479
 480 /**
 481  * Convert modify-header set IPv4 address action to DV specification.
 482  *
 483  * @param[in,out] resource
 484  *   Pointer to the modify-header resource.
 485  * @param[in] action
 486  *   Pointer to action specification.
 487  * @param[out] error
 488  *   Pointer to the error structure.
 489  *
 490  * @return
 491  *   0 on success, a negative errno value otherwise and rte_errno is set.
 492  */
 493 static int
 494 flow_dv_convert_action_modify_ipv4
 495                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 496                          const struct rte_flow_action *action,
 497                          struct rte_flow_error *error)
 498 {
 499         const struct rte_flow_action_set_ipv4 *conf =
 500                 (const struct rte_flow_action_set_ipv4 *)(action->conf);
 501         struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
 502         struct rte_flow_item_ipv4 ipv4;
 503         struct rte_flow_item_ipv4 ipv4_mask;
 504
 505         memset(&ipv4, 0, sizeof(ipv4));
 506         memset(&ipv4_mask, 0, sizeof(ipv4_mask));
 507         if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
 508                 ipv4.hdr.src_addr = conf->ipv4_addr;
 509                 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
 510         } else {
 511                 ipv4.hdr.dst_addr = conf->ipv4_addr;
 512                 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
 513         }
 514         item.spec = &ipv4;
 515         item.mask = &ipv4_mask;
 516         return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
 517                                              MLX5_MODIFICATION_TYPE_SET, error);
 518 }
 519
 520 /**
 521  * Convert modify-header set IPv6 address action to DV specification.
 522  *
 523  * @param[in,out] resource
 524  *   Pointer to the modify-header resource.
 525  * @param[in] action
 526  *   Pointer to action specification.
 527  * @param[out] error
 528  *   Pointer to the error structure.
 529  *
 530  * @return
 531  *   0 on success, a negative errno value otherwise and rte_errno is set.
 532  */
 533 static int
 534 flow_dv_convert_action_modify_ipv6
 535                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 536                          const struct rte_flow_action *action,
 537                          struct rte_flow_error *error)
 538 {
 539         const struct rte_flow_action_set_ipv6 *conf =
 540                 (const struct rte_flow_action_set_ipv6 *)(action->conf);
 541         struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
 542         struct rte_flow_item_ipv6 ipv6;
 543         struct rte_flow_item_ipv6 ipv6_mask;
 544
 545         memset(&ipv6, 0, sizeof(ipv6));
 546         memset(&ipv6_mask, 0, sizeof(ipv6_mask));
 547         if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
 548                 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
 549                        sizeof(ipv6.hdr.src_addr));
 550                 memcpy(&ipv6_mask.hdr.src_addr,
 551                        &rte_flow_item_ipv6_mask.hdr.src_addr,
 552                        sizeof(ipv6.hdr.src_addr));
 553         } else {
 554                 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
 555                        sizeof(ipv6.hdr.dst_addr));
 556                 memcpy(&ipv6_mask.hdr.dst_addr,
 557                        &rte_flow_item_ipv6_mask.hdr.dst_addr,
 558                        sizeof(ipv6.hdr.dst_addr));
 559         }
 560         item.spec = &ipv6;
 561         item.mask = &ipv6_mask;
 562         return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
 563                                              MLX5_MODIFICATION_TYPE_SET, error);
 564 }
 565
 566 /**
 567  * Convert modify-header set MAC address action to DV specification.
 568  *
 569  * @param[in,out] resource
 570  *   Pointer to the modify-header resource.
 571  * @param[in] action
 572  *   Pointer to action specification.
 573  * @param[out] error
 574  *   Pointer to the error structure.
 575  *
 576  * @return
 577  *   0 on success, a negative errno value otherwise and rte_errno is set.
 578  */
 579 static int
 580 flow_dv_convert_action_modify_mac
 581                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 582                          const struct rte_flow_action *action,
 583                          struct rte_flow_error *error)
 584 {
 585         const struct rte_flow_action_set_mac *conf =
 586                 (const struct rte_flow_action_set_mac *)(action->conf);
 587         struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
 588         struct rte_flow_item_eth eth;
 589         struct rte_flow_item_eth eth_mask;
 590
 591         memset(&eth, 0, sizeof(eth));
 592         memset(&eth_mask, 0, sizeof(eth_mask));
 593         if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
 594                 memcpy(&eth.src.addr_bytes, &conf->mac_addr,
 595                        sizeof(eth.src.addr_bytes));
 596                 memcpy(&eth_mask.src.addr_bytes,
 597                        &rte_flow_item_eth_mask.src.addr_bytes,
 598                        sizeof(eth_mask.src.addr_bytes));
 599         } else {
 600                 memcpy(&eth.dst.addr_bytes, &conf->mac_addr,
 601                        sizeof(eth.dst.addr_bytes));
 602                 memcpy(&eth_mask.dst.addr_bytes,
 603                        &rte_flow_item_eth_mask.dst.addr_bytes,
 604                        sizeof(eth_mask.dst.addr_bytes));
 605         }
 606         item.spec = &eth;
 607         item.mask = &eth_mask;
 608         return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
 609                                              MLX5_MODIFICATION_TYPE_SET, error);
 610 }
 611
 612 /**
 613  * Convert modify-header set VLAN VID action to DV specification.
 614  *
 615  * @param[in,out] resource
 616  *   Pointer to the modify-header resource.
 617  * @param[in] action
 618  *   Pointer to action specification.
 619  * @param[out] error
 620  *   Pointer to the error structure.
 621  *
 622  * @return
 623  *   0 on success, a negative errno value otherwise and rte_errno is set.
 624  */
 625 static int
 626 flow_dv_convert_action_modify_vlan_vid
 627                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 628                          const struct rte_flow_action *action,
 629                          struct rte_flow_error *error)
 630 {
 631         const struct rte_flow_action_of_set_vlan_vid *conf =
 632                 (const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
 633         int i = resource->actions_num;
 634         struct mlx5_modification_cmd *actions = resource->actions;
 635         struct field_modify_info *field = modify_vlan_out_first_vid;
 636
 637         if (i >= MLX5_MAX_MODIFY_NUM)
 638                 return rte_flow_error_set(error, EINVAL,
 639                          RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 640                          "too many items to modify");
 641         actions[i] = (struct mlx5_modification_cmd) {
 642                 .action_type = MLX5_MODIFICATION_TYPE_SET,
 643                 .field = field->id,
 644                 .length = field->size,
 645                 .offset = field->offset,
 646         };
 647         actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
 648         actions[i].data1 = conf->vlan_vid;
 649         actions[i].data1 = actions[i].data1 << 16;
 650         resource->actions_num = ++i;
 651         return 0;
 652 }
 653
 654 /**
 655  * Convert modify-header set TP action to DV specification.
 656  *
 657  * @param[in,out] resource
 658  *   Pointer to the modify-header resource.
 659  * @param[in] action
 660  *   Pointer to action specification.
 661  * @param[in] items
 662  *   Pointer to rte_flow_item objects list.
 663  * @param[in] attr
 664  *   Pointer to flow attributes structure.
 665  * @param[in] dev_flow
 666  *   Pointer to the sub flow.
 667  * @param[in] tunnel_decap
 668  *   Whether action is after tunnel decapsulation.
 669  * @param[out] error
 670  *   Pointer to the error structure.
 671  *
 672  * @return
 673  *   0 on success, a negative errno value otherwise and rte_errno is set.
 674  */
 675 static int
 676 flow_dv_convert_action_modify_tp
 677                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 678                          const struct rte_flow_action *action,
 679                          const struct rte_flow_item *items,
 680                          union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
 681                          bool tunnel_decap, struct rte_flow_error *error)
 682 {
 683         const struct rte_flow_action_set_tp *conf =
 684                 (const struct rte_flow_action_set_tp *)(action->conf);
 685         struct rte_flow_item item;
 686         struct rte_flow_item_udp udp;
 687         struct rte_flow_item_udp udp_mask;
 688         struct rte_flow_item_tcp tcp;
 689         struct rte_flow_item_tcp tcp_mask;
 690         struct field_modify_info *field;
 691
 692         if (!attr->valid)
 693                 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
 694         if (attr->udp) {
 695                 memset(&udp, 0, sizeof(udp));
 696                 memset(&udp_mask, 0, sizeof(udp_mask));
 697                 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
 698                         udp.hdr.src_port = conf->port;
 699                         udp_mask.hdr.src_port =
 700                                         rte_flow_item_udp_mask.hdr.src_port;
 701                 } else {
 702                         udp.hdr.dst_port = conf->port;
 703                         udp_mask.hdr.dst_port =
 704                                         rte_flow_item_udp_mask.hdr.dst_port;
 705                 }
 706                 item.type = RTE_FLOW_ITEM_TYPE_UDP;
 707                 item.spec = &udp;
 708                 item.mask = &udp_mask;
 709                 field = modify_udp;
 710         } else {
 711                 MLX5_ASSERT(attr->tcp);
 712                 memset(&tcp, 0, sizeof(tcp));
 713                 memset(&tcp_mask, 0, sizeof(tcp_mask));
 714                 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
 715                         tcp.hdr.src_port = conf->port;
 716                         tcp_mask.hdr.src_port =
 717                                         rte_flow_item_tcp_mask.hdr.src_port;
 718                 } else {
 719                         tcp.hdr.dst_port = conf->port;
 720                         tcp_mask.hdr.dst_port =
 721                                         rte_flow_item_tcp_mask.hdr.dst_port;
 722                 }
 723                 item.type = RTE_FLOW_ITEM_TYPE_TCP;
 724                 item.spec = &tcp;
 725                 item.mask = &tcp_mask;
 726                 field = modify_tcp;
 727         }
 728         return flow_dv_convert_modify_action(&item, field, NULL, resource,
 729                                              MLX5_MODIFICATION_TYPE_SET, error);
 730 }
 731
 732 /**
 733  * Convert modify-header set TTL action to DV specification.
 734  *
 735  * @param[in,out] resource
 736  *   Pointer to the modify-header resource.
 737  * @param[in] action
 738  *   Pointer to action specification.
 739  * @param[in] items
 740  *   Pointer to rte_flow_item objects list.
 741  * @param[in] attr
 742  *   Pointer to flow attributes structure.
 743  * @param[in] dev_flow
 744  *   Pointer to the sub flow.
 745  * @param[in] tunnel_decap
 746  *   Whether action is after tunnel decapsulation.
 747  * @param[out] error
 748  *   Pointer to the error structure.
 749  *
 750  * @return
 751  *   0 on success, a negative errno value otherwise and rte_errno is set.
 752  */
 753 static int
 754 flow_dv_convert_action_modify_ttl
 755                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 756                          const struct rte_flow_action *action,
 757                          const struct rte_flow_item *items,
 758                          union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
 759                          bool tunnel_decap, struct rte_flow_error *error)
 760 {
 761         const struct rte_flow_action_set_ttl *conf =
 762                 (const struct rte_flow_action_set_ttl *)(action->conf);
 763         struct rte_flow_item item;
 764         struct rte_flow_item_ipv4 ipv4;
 765         struct rte_flow_item_ipv4 ipv4_mask;
 766         struct rte_flow_item_ipv6 ipv6;
 767         struct rte_flow_item_ipv6 ipv6_mask;
 768         struct field_modify_info *field;
 769
 770         if (!attr->valid)
 771                 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
 772         if (attr->ipv4) {
 773                 memset(&ipv4, 0, sizeof(ipv4));
 774                 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
 775                 ipv4.hdr.time_to_live = conf->ttl_value;
 776                 ipv4_mask.hdr.time_to_live = 0xFF;
 777                 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
 778                 item.spec = &ipv4;
 779                 item.mask = &ipv4_mask;
 780                 field = modify_ipv4;
 781         } else {
 782                 MLX5_ASSERT(attr->ipv6);
 783                 memset(&ipv6, 0, sizeof(ipv6));
 784                 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
 785                 ipv6.hdr.hop_limits = conf->ttl_value;
 786                 ipv6_mask.hdr.hop_limits = 0xFF;
 787                 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
 788                 item.spec = &ipv6;
 789                 item.mask = &ipv6_mask;
 790                 field = modify_ipv6;
 791         }
 792         return flow_dv_convert_modify_action(&item, field, NULL, resource,
 793                                              MLX5_MODIFICATION_TYPE_SET, error);
 794 }
 795
 796 /**
 797  * Convert modify-header decrement TTL action to DV specification.
 798  *
 799  * @param[in,out] resource
 800  *   Pointer to the modify-header resource.
 801  * @param[in] action
 802  *   Pointer to action specification.
 803  * @param[in] items
 804  *   Pointer to rte_flow_item objects list.
 805  * @param[in] attr
 806  *   Pointer to flow attributes structure.
 807  * @param[in] dev_flow
 808  *   Pointer to the sub flow.
 809  * @param[in] tunnel_decap
 810  *   Whether action is after tunnel decapsulation.
 811  * @param[out] error
 812  *   Pointer to the error structure.
 813  *
 814  * @return
 815  *   0 on success, a negative errno value otherwise and rte_errno is set.
 816  */
 817 static int
 818 flow_dv_convert_action_modify_dec_ttl
 819                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 820                          const struct rte_flow_item *items,
 821                          union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
 822                          bool tunnel_decap, struct rte_flow_error *error)
 823 {
 824         struct rte_flow_item item;
 825         struct rte_flow_item_ipv4 ipv4;
 826         struct rte_flow_item_ipv4 ipv4_mask;
 827         struct rte_flow_item_ipv6 ipv6;
 828         struct rte_flow_item_ipv6 ipv6_mask;
 829         struct field_modify_info *field;
 830
 831         if (!attr->valid)
 832                 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
 833         if (attr->ipv4) {
 834                 memset(&ipv4, 0, sizeof(ipv4));
 835                 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
 836                 ipv4.hdr.time_to_live = 0xFF;
 837                 ipv4_mask.hdr.time_to_live = 0xFF;
 838                 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
 839                 item.spec = &ipv4;
 840                 item.mask = &ipv4_mask;
 841                 field = modify_ipv4;
 842         } else {
 843                 MLX5_ASSERT(attr->ipv6);
 844                 memset(&ipv6, 0, sizeof(ipv6));
 845                 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
 846                 ipv6.hdr.hop_limits = 0xFF;
 847                 ipv6_mask.hdr.hop_limits = 0xFF;
 848                 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
 849                 item.spec = &ipv6;
 850                 item.mask = &ipv6_mask;
 851                 field = modify_ipv6;
 852         }
 853         return flow_dv_convert_modify_action(&item, field, NULL, resource,
 854                                              MLX5_MODIFICATION_TYPE_ADD, error);
 855 }
 856
 857 /**
 858  * Convert modify-header increment/decrement TCP Sequence number
 859  * to DV specification.
 860  *
 861  * @param[in,out] resource
 862  *   Pointer to the modify-header resource.
 863  * @param[in] action
 864  *   Pointer to action specification.
 865  * @param[out] error
 866  *   Pointer to the error structure.
 867  *
 868  * @return
 869  *   0 on success, a negative errno value otherwise and rte_errno is set.
 870  */
 871 static int
 872 flow_dv_convert_action_modify_tcp_seq
 873                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 874                          const struct rte_flow_action *action,
 875                          struct rte_flow_error *error)
 876 {
 877         const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
 878         uint64_t value = rte_be_to_cpu_32(*conf);
 879         struct rte_flow_item item;
 880         struct rte_flow_item_tcp tcp;
 881         struct rte_flow_item_tcp tcp_mask;
 882
 883         memset(&tcp, 0, sizeof(tcp));
 884         memset(&tcp_mask, 0, sizeof(tcp_mask));
 885         if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
 886                 /*
 887                  * The HW has no decrement operation, only increment operation.
 888                  * To simulate decrement X from Y using increment operation
 889                  * we need to add UINT32_MAX X times to Y.
 890                  * Each adding of UINT32_MAX decrements Y by 1.
 891                  */
 892                 value *= UINT32_MAX;
 893         tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
 894         tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
 895         item.type = RTE_FLOW_ITEM_TYPE_TCP;
 896         item.spec = &tcp;
 897         item.mask = &tcp_mask;
 898         return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
 899                                              MLX5_MODIFICATION_TYPE_ADD, error);
 900 }
 901
 902 /**
 903  * Convert modify-header increment/decrement TCP Acknowledgment number
 904  * to DV specification.
 905  *
 906  * @param[in,out] resource
 907  *   Pointer to the modify-header resource.
 908  * @param[in] action
 909  *   Pointer to action specification.
 910  * @param[out] error
 911  *   Pointer to the error structure.
 912  *
 913  * @return
 914  *   0 on success, a negative errno value otherwise and rte_errno is set.
 915  */
 916 static int
 917 flow_dv_convert_action_modify_tcp_ack
 918                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 919                          const struct rte_flow_action *action,
 920                          struct rte_flow_error *error)
 921 {
 922         const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
 923         uint64_t value = rte_be_to_cpu_32(*conf);
 924         struct rte_flow_item item;
 925         struct rte_flow_item_tcp tcp;
 926         struct rte_flow_item_tcp tcp_mask;
 927
 928         memset(&tcp, 0, sizeof(tcp));
 929         memset(&tcp_mask, 0, sizeof(tcp_mask));
 930         if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
 931                 /*
 932                  * The HW has no decrement operation, only increment operation.
 933                  * To simulate decrement X from Y using increment operation
 934                  * we need to add UINT32_MAX X times to Y.
 935                  * Each adding of UINT32_MAX decrements Y by 1.
 936                  */
 937                 value *= UINT32_MAX;
 938         tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
 939         tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
 940         item.type = RTE_FLOW_ITEM_TYPE_TCP;
 941         item.spec = &tcp;
 942         item.mask = &tcp_mask;
 943         return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
 944                                              MLX5_MODIFICATION_TYPE_ADD, error);
 945 }
 946
 947 static enum mlx5_modification_field reg_to_field[] = {
 948         [REG_NON] = MLX5_MODI_OUT_NONE,
 949         [REG_A] = MLX5_MODI_META_DATA_REG_A,
 950         [REG_B] = MLX5_MODI_META_DATA_REG_B,
 951         [REG_C_0] = MLX5_MODI_META_REG_C_0,
 952         [REG_C_1] = MLX5_MODI_META_REG_C_1,
 953         [REG_C_2] = MLX5_MODI_META_REG_C_2,
 954         [REG_C_3] = MLX5_MODI_META_REG_C_3,
 955         [REG_C_4] = MLX5_MODI_META_REG_C_4,
 956         [REG_C_5] = MLX5_MODI_META_REG_C_5,
 957         [REG_C_6] = MLX5_MODI_META_REG_C_6,
 958         [REG_C_7] = MLX5_MODI_META_REG_C_7,
 959 };
 960
 961 /**
 962  * Convert register set to DV specification.
 963  *
 964  * @param[in,out] resource
 965  *   Pointer to the modify-header resource.
 966  * @param[in] action
 967  *   Pointer to action specification.
 968  * @param[out] error
 969  *   Pointer to the error structure.
 970  *
 971  * @return
 972  *   0 on success, a negative errno value otherwise and rte_errno is set.
 973  */
 974 static int
 975 flow_dv_convert_action_set_reg
 976                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
 977                          const struct rte_flow_action *action,
 978                          struct rte_flow_error *error)
 979 {
 980         const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
 981         struct mlx5_modification_cmd *actions = resource->actions;
 982         uint32_t i = resource->actions_num;
 983
 984         if (i >= MLX5_MAX_MODIFY_NUM)
 985                 return rte_flow_error_set(error, EINVAL,
 986                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 987                                           "too many items to modify");
 988         MLX5_ASSERT(conf->id != REG_NON);
 989         MLX5_ASSERT(conf->id < RTE_DIM(reg_to_field));
 990         actions[i] = (struct mlx5_modification_cmd) {
 991                 .action_type = MLX5_MODIFICATION_TYPE_SET,
 992                 .field = reg_to_field[conf->id],
 993         };
 994         actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
 995         actions[i].data1 = rte_cpu_to_be_32(conf->data);
 996         ++i;
 997         resource->actions_num = i;
 998         return 0;
 999 }
1000
1001 /**
1002  * Convert SET_TAG action to DV specification.
1003  *
1004  * @param[in] dev
1005  *   Pointer to the rte_eth_dev structure.
1006  * @param[in,out] resource
1007  *   Pointer to the modify-header resource.
1008  * @param[in] conf
1009  *   Pointer to action specification.
1010  * @param[out] error
1011  *   Pointer to the error structure.
1012  *
1013  * @return
1014  *   0 on success, a negative errno value otherwise and rte_errno is set.
1015  */
1016 static int
1017 flow_dv_convert_action_set_tag
1018                         (struct rte_eth_dev *dev,
1019                          struct mlx5_flow_dv_modify_hdr_resource *resource,
1020                          const struct rte_flow_action_set_tag *conf,
1021                          struct rte_flow_error *error)
1022 {
1023         rte_be32_t data = rte_cpu_to_be_32(conf->data);
1024         rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
1025         struct rte_flow_item item = {
1026                 .spec = &data,
1027                 .mask = &mask,
1028         };
1029         struct field_modify_info reg_c_x[] = {
1030                 [1] = {0, 0, 0},
1031         };
1032         enum mlx5_modification_field reg_type;
1033         int ret;
1034
1035         ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1036         if (ret < 0)
1037                 return ret;
1038         MLX5_ASSERT(ret != REG_NON);
1039         MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1040         reg_type = reg_to_field[ret];
1041         MLX5_ASSERT(reg_type > 0);
1042         reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1043         return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1044                                              MLX5_MODIFICATION_TYPE_SET, error);
1045 }
1046
1047 /**
1048  * Convert internal COPY_REG action to DV specification.
1049  *
1050  * @param[in] dev
1051  *   Pointer to the rte_eth_dev structure.
1052  * @param[in,out] res
1053  *   Pointer to the modify-header resource.
1054  * @param[in] action
1055  *   Pointer to action specification.
1056  * @param[out] error
1057  *   Pointer to the error structure.
1058  *
1059  * @return
1060  *   0 on success, a negative errno value otherwise and rte_errno is set.
1061  */
1062 static int
1063 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1064                                  struct mlx5_flow_dv_modify_hdr_resource *res,
1065                                  const struct rte_flow_action *action,
1066                                  struct rte_flow_error *error)
1067 {
1068         const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1069         rte_be32_t mask = RTE_BE32(UINT32_MAX);
1070         struct rte_flow_item item = {
1071                 .spec = NULL,
1072                 .mask = &mask,
1073         };
1074         struct field_modify_info reg_src[] = {
1075                 {4, 0, reg_to_field[conf->src]},
1076                 {0, 0, 0},
1077         };
1078         struct field_modify_info reg_dst = {
1079                 .offset = 0,
1080                 .id = reg_to_field[conf->dst],
1081         };
1082         /* Adjust reg_c[0] usage according to reported mask. */
1083         if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1084                 struct mlx5_priv *priv = dev->data->dev_private;
1085                 uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1086
1087                 MLX5_ASSERT(reg_c0);
1088                 MLX5_ASSERT(priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY);
1089                 if (conf->dst == REG_C_0) {
1090                         /* Copy to reg_c[0], within mask only. */
1091                         reg_dst.offset = rte_bsf32(reg_c0);
1092                         /*
1093                          * Mask is ignoring the enianness, because
1094                          * there is no conversion in datapath.
1095                          */
1096 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1097                         /* Copy from destination lower bits to reg_c[0]. */
1098                         mask = reg_c0 >> reg_dst.offset;
1099 #else
1100                         /* Copy from destination upper bits to reg_c[0]. */
1101                         mask = reg_c0 << (sizeof(reg_c0) * CHAR_BIT -
1102                                           rte_fls_u32(reg_c0));
1103 #endif
1104                 } else {
1105                         mask = rte_cpu_to_be_32(reg_c0);
1106 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1107                         /* Copy from reg_c[0] to destination lower bits. */
1108                         reg_dst.offset = 0;
1109 #else
1110                         /* Copy from reg_c[0] to destination upper bits. */
1111                         reg_dst.offset = sizeof(reg_c0) * CHAR_BIT -
1112                                          (rte_fls_u32(reg_c0) -
1113                                           rte_bsf32(reg_c0));
1114 #endif
1115                 }
1116         }
1117         return flow_dv_convert_modify_action(&item,
1118                                              reg_src, &reg_dst, res,
1119                                              MLX5_MODIFICATION_TYPE_COPY,
1120                                              error);
1121 }
1122
1123 /**
1124  * Convert MARK action to DV specification. This routine is used
1125  * in extensive metadata only and requires metadata register to be
1126  * handled. In legacy mode hardware tag resource is engaged.
1127  *
1128  * @param[in] dev
1129  *   Pointer to the rte_eth_dev structure.
1130  * @param[in] conf
1131  *   Pointer to MARK action specification.
1132  * @param[in,out] resource
1133  *   Pointer to the modify-header resource.
1134  * @param[out] error
1135  *   Pointer to the error structure.
1136  *
1137  * @return
1138  *   0 on success, a negative errno value otherwise and rte_errno is set.
1139  */
1140 static int
1141 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1142                             const struct rte_flow_action_mark *conf,
1143                             struct mlx5_flow_dv_modify_hdr_resource *resource,
1144                             struct rte_flow_error *error)
1145 {
1146         struct mlx5_priv *priv = dev->data->dev_private;
1147         rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1148                                            priv->sh->dv_mark_mask);
1149         rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1150         struct rte_flow_item item = {
1151                 .spec = &data,
1152                 .mask = &mask,
1153         };
1154         struct field_modify_info reg_c_x[] = {
1155                 [1] = {0, 0, 0},
1156         };
1157         int reg;
1158
1159         if (!mask)
1160                 return rte_flow_error_set(error, EINVAL,
1161                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1162                                           NULL, "zero mark action mask");
1163         reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1164         if (reg < 0)
1165                 return reg;
1166         MLX5_ASSERT(reg > 0);
1167         if (reg == REG_C_0) {
1168                 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1169                 uint32_t shl_c0 = rte_bsf32(msk_c0);
1170
1171                 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1172                 mask = rte_cpu_to_be_32(mask) & msk_c0;
1173                 mask = rte_cpu_to_be_32(mask << shl_c0);
1174         }
1175         reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1176         return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1177                                              MLX5_MODIFICATION_TYPE_SET, error);
1178 }
1179
1180 /**
1181  * Get metadata register index for specified steering domain.
1182  *
1183  * @param[in] dev
1184  *   Pointer to the rte_eth_dev structure.
1185  * @param[in] attr
1186  *   Attributes of flow to determine steering domain.
1187  * @param[out] error
1188  *   Pointer to the error structure.
1189  *
1190  * @return
1191  *   positive index on success, a negative errno value otherwise
1192  *   and rte_errno is set.
1193  */
1194 static enum modify_reg
1195 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1196                          const struct rte_flow_attr *attr,
1197                          struct rte_flow_error *error)
1198 {
1199         int reg =
1200                 mlx5_flow_get_reg_id(dev, attr->transfer ?
1201                                           MLX5_METADATA_FDB :
1202                                             attr->egress ?
1203                                             MLX5_METADATA_TX :
1204                                             MLX5_METADATA_RX, 0, error);
1205         if (reg < 0)
1206                 return rte_flow_error_set(error,
1207                                           ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1208                                           NULL, "unavailable "
1209                                           "metadata register");
1210         return reg;
1211 }
1212
1213 /**
1214  * Convert SET_META action to DV specification.
1215  *
1216  * @param[in] dev
1217  *   Pointer to the rte_eth_dev structure.
1218  * @param[in,out] resource
1219  *   Pointer to the modify-header resource.
1220  * @param[in] attr
1221  *   Attributes of flow that includes this item.
1222  * @param[in] conf
1223  *   Pointer to action specification.
1224  * @param[out] error
1225  *   Pointer to the error structure.
1226  *
1227  * @return
1228  *   0 on success, a negative errno value otherwise and rte_errno is set.
1229  */
1230 static int
1231 flow_dv_convert_action_set_meta
1232                         (struct rte_eth_dev *dev,
1233                          struct mlx5_flow_dv_modify_hdr_resource *resource,
1234                          const struct rte_flow_attr *attr,
1235                          const struct rte_flow_action_set_meta *conf,
1236                          struct rte_flow_error *error)
1237 {
1238         uint32_t data = conf->data;
1239         uint32_t mask = conf->mask;
1240         struct rte_flow_item item = {
1241                 .spec = &data,
1242                 .mask = &mask,
1243         };
1244         struct field_modify_info reg_c_x[] = {
1245                 [1] = {0, 0, 0},
1246         };
1247         int reg = flow_dv_get_metadata_reg(dev, attr, error);
1248
1249         if (reg < 0)
1250                 return reg;
1251         /*
1252          * In datapath code there is no endianness
1253          * coversions for perfromance reasons, all
1254          * pattern conversions are done in rte_flow.
1255          */
1256         if (reg == REG_C_0) {
1257                 struct mlx5_priv *priv = dev->data->dev_private;
1258                 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1259                 uint32_t shl_c0;
1260
1261                 MLX5_ASSERT(msk_c0);
1262 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1263                 shl_c0 = rte_bsf32(msk_c0);
1264 #else
1265                 shl_c0 = sizeof(msk_c0) * CHAR_BIT - rte_fls_u32(msk_c0);
1266 #endif
1267                 mask <<= shl_c0;
1268                 data <<= shl_c0;
1269                 MLX5_ASSERT(!(~msk_c0 & rte_cpu_to_be_32(mask)));
1270         }
1271         reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1272         /* The routine expects parameters in memory as big-endian ones. */
1273         return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1274                                              MLX5_MODIFICATION_TYPE_SET, error);
1275 }
1276
1277 /**
1278  * Convert modify-header set IPv4 DSCP action to DV specification.
1279  *
1280  * @param[in,out] resource
1281  *   Pointer to the modify-header resource.
1282  * @param[in] action
1283  *   Pointer to action specification.
1284  * @param[out] error
1285  *   Pointer to the error structure.
1286  *
1287  * @return
1288  *   0 on success, a negative errno value otherwise and rte_errno is set.
1289  */
1290 static int
1291 flow_dv_convert_action_modify_ipv4_dscp
1292                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
1293                          const struct rte_flow_action *action,
1294                          struct rte_flow_error *error)
1295 {
1296         const struct rte_flow_action_set_dscp *conf =
1297                 (const struct rte_flow_action_set_dscp *)(action->conf);
1298         struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1299         struct rte_flow_item_ipv4 ipv4;
1300         struct rte_flow_item_ipv4 ipv4_mask;
1301
1302         memset(&ipv4, 0, sizeof(ipv4));
1303         memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1304         ipv4.hdr.type_of_service = conf->dscp;
1305         ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1306         item.spec = &ipv4;
1307         item.mask = &ipv4_mask;
1308         return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1309                                              MLX5_MODIFICATION_TYPE_SET, error);
1310 }
1311
1312 /**
1313  * Convert modify-header set IPv6 DSCP action to DV specification.
1314  *
1315  * @param[in,out] resource
1316  *   Pointer to the modify-header resource.
1317  * @param[in] action
1318  *   Pointer to action specification.
1319  * @param[out] error
1320  *   Pointer to the error structure.
1321  *
1322  * @return
1323  *   0 on success, a negative errno value otherwise and rte_errno is set.
1324  */
1325 static int
1326 flow_dv_convert_action_modify_ipv6_dscp
1327                         (struct mlx5_flow_dv_modify_hdr_resource *resource,
1328                          const struct rte_flow_action *action,
1329                          struct rte_flow_error *error)
1330 {
1331         const struct rte_flow_action_set_dscp *conf =
1332                 (const struct rte_flow_action_set_dscp *)(action->conf);
1333         struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1334         struct rte_flow_item_ipv6 ipv6;
1335         struct rte_flow_item_ipv6 ipv6_mask;
1336
1337         memset(&ipv6, 0, sizeof(ipv6));
1338         memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1339         /*
1340          * Even though the DSCP bits offset of IPv6 is not byte aligned,
1341          * rdma-core only accept the DSCP bits byte aligned start from
1342          * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1343          * bits in IPv6 case as rdma-core requires byte aligned value.
1344          */
1345         ipv6.hdr.vtc_flow = conf->dscp;
1346         ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1347         item.spec = &ipv6;
1348         item.mask = &ipv6_mask;
1349         return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1350                                              MLX5_MODIFICATION_TYPE_SET, error);
1351 }
1352
1353 /**
1354  * Validate MARK item.
1355  *
1356  * @param[in] dev
1357  *   Pointer to the rte_eth_dev structure.
1358  * @param[in] item
1359  *   Item specification.
1360  * @param[in] attr
1361  *   Attributes of flow that includes this item.
1362  * @param[out] error
1363  *   Pointer to error structure.
1364  *
1365  * @return
1366  *   0 on success, a negative errno value otherwise and rte_errno is set.
1367  */
1368 static int
1369 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1370                            const struct rte_flow_item *item,
1371                            const struct rte_flow_attr *attr __rte_unused,
1372                            struct rte_flow_error *error)
1373 {
1374         struct mlx5_priv *priv = dev->data->dev_private;
1375         struct mlx5_dev_config *config = &priv->config;
1376         const struct rte_flow_item_mark *spec = item->spec;
1377         const struct rte_flow_item_mark *mask = item->mask;
1378         const struct rte_flow_item_mark nic_mask = {
1379                 .id = priv->sh->dv_mark_mask,
1380         };
1381         int ret;
1382
1383         if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1384                 return rte_flow_error_set(error, ENOTSUP,
1385                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1386                                           "extended metadata feature"
1387                                           " isn't enabled");
1388         if (!mlx5_flow_ext_mreg_supported(dev))
1389                 return rte_flow_error_set(error, ENOTSUP,
1390                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1391                                           "extended metadata register"
1392                                           " isn't supported");
1393         if (!nic_mask.id)
1394                 return rte_flow_error_set(error, ENOTSUP,
1395                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1396                                           "extended metadata register"
1397                                           " isn't available");
1398         ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1399         if (ret < 0)
1400                 return ret;
1401         if (!spec)
1402                 return rte_flow_error_set(error, EINVAL,
1403                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1404                                           item->spec,
1405                                           "data cannot be empty");
1406         if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1407                 return rte_flow_error_set(error, EINVAL,
1408                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1409                                           &spec->id,
1410                                           "mark id exceeds the limit");
1411         if (!mask)
1412                 mask = &nic_mask;
1413         if (!mask->id)
1414                 return rte_flow_error_set(error, EINVAL,
1415                                         RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1416                                         "mask cannot be zero");
1417
1418         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1419                                         (const uint8_t *)&nic_mask,
1420                                         sizeof(struct rte_flow_item_mark),
1421                                         error);
1422         if (ret < 0)
1423                 return ret;
1424         return 0;
1425 }
1426
1427 /**
1428  * Validate META item.
1429  *
1430  * @param[in] dev
1431  *   Pointer to the rte_eth_dev structure.
1432  * @param[in] item
1433  *   Item specification.
1434  * @param[in] attr
1435  *   Attributes of flow that includes this item.
1436  * @param[out] error
1437  *   Pointer to error structure.
1438  *
1439  * @return
1440  *   0 on success, a negative errno value otherwise and rte_errno is set.
1441  */
1442 static int
1443 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1444                            const struct rte_flow_item *item,
1445                            const struct rte_flow_attr *attr,
1446                            struct rte_flow_error *error)
1447 {
1448         struct mlx5_priv *priv = dev->data->dev_private;
1449         struct mlx5_dev_config *config = &priv->config;
1450         const struct rte_flow_item_meta *spec = item->spec;
1451         const struct rte_flow_item_meta *mask = item->mask;
1452         struct rte_flow_item_meta nic_mask = {
1453                 .data = UINT32_MAX
1454         };
1455         int reg;
1456         int ret;
1457
1458         if (!spec)
1459                 return rte_flow_error_set(error, EINVAL,
1460                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1461                                           item->spec,
1462                                           "data cannot be empty");
1463         if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
1464                 if (!mlx5_flow_ext_mreg_supported(dev))
1465                         return rte_flow_error_set(error, ENOTSUP,
1466                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1467                                           "extended metadata register"
1468                                           " isn't supported");
1469                 reg = flow_dv_get_metadata_reg(dev, attr, error);
1470                 if (reg < 0)
1471                         return reg;
1472                 if (reg == REG_B)
1473                         return rte_flow_error_set(error, ENOTSUP,
1474                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1475                                           "match on reg_b "
1476                                           "isn't supported");
1477                 if (reg != REG_A)
1478                         nic_mask.data = priv->sh->dv_meta_mask;
1479         } else if (attr->transfer) {
1480                 return rte_flow_error_set(error, ENOTSUP,
1481                                         RTE_FLOW_ERROR_TYPE_ITEM, item,
1482                                         "extended metadata feature "
1483                                         "should be enabled when "
1484                                         "meta item is requested "
1485                                         "with e-switch mode ");
1486         }
1487         if (!mask)
1488                 mask = &rte_flow_item_meta_mask;
1489         if (!mask->data)
1490                 return rte_flow_error_set(error, EINVAL,
1491                                         RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1492                                         "mask cannot be zero");
1493
1494         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1495                                         (const uint8_t *)&nic_mask,
1496                                         sizeof(struct rte_flow_item_meta),
1497                                         error);
1498         return ret;
1499 }
1500
1501 /**
1502  * Validate TAG item.
1503  *
1504  * @param[in] dev
1505  *   Pointer to the rte_eth_dev structure.
1506  * @param[in] item
1507  *   Item specification.
1508  * @param[in] attr
1509  *   Attributes of flow that includes this item.
1510  * @param[out] error
1511  *   Pointer to error structure.
1512  *
1513  * @return
1514  *   0 on success, a negative errno value otherwise and rte_errno is set.
1515  */
1516 static int
1517 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
1518                           const struct rte_flow_item *item,
1519                           const struct rte_flow_attr *attr __rte_unused,
1520                           struct rte_flow_error *error)
1521 {
1522         const struct rte_flow_item_tag *spec = item->spec;
1523         const struct rte_flow_item_tag *mask = item->mask;
1524         const struct rte_flow_item_tag nic_mask = {
1525                 .data = RTE_BE32(UINT32_MAX),
1526                 .index = 0xff,
1527         };
1528         int ret;
1529
1530         if (!mlx5_flow_ext_mreg_supported(dev))
1531                 return rte_flow_error_set(error, ENOTSUP,
1532                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1533                                           "extensive metadata register"
1534                                           " isn't supported");
1535         if (!spec)
1536                 return rte_flow_error_set(error, EINVAL,
1537                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1538                                           item->spec,
1539                                           "data cannot be empty");
1540         if (!mask)
1541                 mask = &rte_flow_item_tag_mask;
1542         if (!mask->data)
1543                 return rte_flow_error_set(error, EINVAL,
1544                                         RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1545                                         "mask cannot be zero");
1546
1547         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1548                                         (const uint8_t *)&nic_mask,
1549                                         sizeof(struct rte_flow_item_tag),
1550                                         error);
1551         if (ret < 0)
1552                 return ret;
1553         if (mask->index != 0xff)
1554                 return rte_flow_error_set(error, EINVAL,
1555                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1556                                           "partial mask for tag index"
1557                                           " is not supported");
1558         ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
1559         if (ret < 0)
1560                 return ret;
1561         MLX5_ASSERT(ret != REG_NON);
1562         return 0;
1563 }
1564
1565 /**
1566  * Validate vport item.
1567  *
1568  * @param[in] dev
1569  *   Pointer to the rte_eth_dev structure.
1570  * @param[in] item
1571  *   Item specification.
1572  * @param[in] attr
1573  *   Attributes of flow that includes this item.
1574  * @param[in] item_flags
1575  *   Bit-fields that holds the items detected until now.
1576  * @param[out] error
1577  *   Pointer to error structure.
1578  *
1579  * @return
1580  *   0 on success, a negative errno value otherwise and rte_errno is set.
1581  */
1582 static int
1583 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
1584                               const struct rte_flow_item *item,
1585                               const struct rte_flow_attr *attr,
1586                               uint64_t item_flags,
1587                               struct rte_flow_error *error)
1588 {
1589         const struct rte_flow_item_port_id *spec = item->spec;
1590         const struct rte_flow_item_port_id *mask = item->mask;
1591         const struct rte_flow_item_port_id switch_mask = {
1592                         .id = 0xffffffff,
1593         };
1594         struct mlx5_priv *esw_priv;
1595         struct mlx5_priv *dev_priv;
1596         int ret;
1597
1598         if (!attr->transfer)
1599                 return rte_flow_error_set(error, EINVAL,
1600                                           RTE_FLOW_ERROR_TYPE_ITEM,
1601                                           NULL,
1602                                           "match on port id is valid only"
1603                                           " when transfer flag is enabled");
1604         if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
1605                 return rte_flow_error_set(error, ENOTSUP,
1606                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1607                                           "multiple source ports are not"
1608                                           " supported");
1609         if (!mask)
1610                 mask = &switch_mask;
1611         if (mask->id != 0xffffffff)
1612                 return rte_flow_error_set(error, ENOTSUP,
1613                                            RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1614                                            mask,
1615                                            "no support for partial mask on"
1616                                            " \"id\" field");
1617         ret = mlx5_flow_item_acceptable
1618                                 (item, (const uint8_t *)mask,
1619                                  (const uint8_t *)&rte_flow_item_port_id_mask,
1620                                  sizeof(struct rte_flow_item_port_id),
1621                                  error);
1622         if (ret)
1623                 return ret;
1624         if (!spec)
1625                 return 0;
1626         esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
1627         if (!esw_priv)
1628                 return rte_flow_error_set(error, rte_errno,
1629                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1630                                           "failed to obtain E-Switch info for"
1631                                           " port");
1632         dev_priv = mlx5_dev_to_eswitch_info(dev);
1633         if (!dev_priv)
1634                 return rte_flow_error_set(error, rte_errno,
1635                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1636                                           NULL,
1637                                           "failed to obtain E-Switch info");
1638         if (esw_priv->domain_id != dev_priv->domain_id)
1639                 return rte_flow_error_set(error, EINVAL,
1640                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1641                                           "cannot match on a port from a"
1642                                           " different E-Switch");
1643         return 0;
1644 }
1645
1646 /**
1647  * Validate VLAN item.
1648  *
1649  * @param[in] item
1650  *   Item specification.
1651  * @param[in] item_flags
1652  *   Bit-fields that holds the items detected until now.
1653  * @param[in] dev
1654  *   Ethernet device flow is being created on.
1655  * @param[out] error
1656  *   Pointer to error structure.
1657  *
1658  * @return
1659  *   0 on success, a negative errno value otherwise and rte_errno is set.
1660  */
1661 static int
1662 flow_dv_validate_item_vlan(const struct rte_flow_item *item,
1663                            uint64_t item_flags,
1664                            struct rte_eth_dev *dev,
1665                            struct rte_flow_error *error)
1666 {
1667         const struct rte_flow_item_vlan *mask = item->mask;
1668         const struct rte_flow_item_vlan nic_mask = {
1669                 .tci = RTE_BE16(UINT16_MAX),
1670                 .inner_type = RTE_BE16(UINT16_MAX),
1671         };
1672         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1673         int ret;
1674         const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1675                                         MLX5_FLOW_LAYER_INNER_L4) :
1676                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1677                                         MLX5_FLOW_LAYER_OUTER_L4);
1678         const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1679                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1680
1681         if (item_flags & vlanm)
1682                 return rte_flow_error_set(error, EINVAL,
1683                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1684                                           "multiple VLAN layers not supported");
1685         else if ((item_flags & l34m) != 0)
1686                 return rte_flow_error_set(error, EINVAL,
1687                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1688                                           "VLAN cannot follow L3/L4 layer");
1689         if (!mask)
1690                 mask = &rte_flow_item_vlan_mask;
1691         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1692                                         (const uint8_t *)&nic_mask,
1693                                         sizeof(struct rte_flow_item_vlan),
1694                                         error);
1695         if (ret)
1696                 return ret;
1697         if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1698                 struct mlx5_priv *priv = dev->data->dev_private;
1699
1700                 if (priv->vmwa_context) {
1701                         /*
1702                          * Non-NULL context means we have a virtual machine
1703                          * and SR-IOV enabled, we have to create VLAN interface
1704                          * to make hypervisor to setup E-Switch vport
1705                          * context correctly. We avoid creating the multiple
1706                          * VLAN interfaces, so we cannot support VLAN tag mask.
1707                          */
1708                         return rte_flow_error_set(error, EINVAL,
1709                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1710                                                   item,
1711                                                   "VLAN tag mask is not"
1712                                                   " supported in virtual"
1713                                                   " environment");
1714                 }
1715         }
1716         return 0;
1717 }
1718
1719 /*
1720  * GTP flags are contained in 1 byte of the format:
1721  * -------------------------------------------
1722  * | bit   | 0 - 2   | 3  | 4   | 5 | 6 | 7  |
1723  * |-----------------------------------------|
1724  * | value | Version | PT | Res | E | S | PN |
1725  * -------------------------------------------
1726  *
1727  * Matching is supported only for GTP flags E, S, PN.
1728  */
1729 #define MLX5_GTP_FLAGS_MASK     0x07
1730
1731 /**
1732  * Validate GTP item.
1733  *
1734  * @param[in] dev
1735  *   Pointer to the rte_eth_dev structure.
1736  * @param[in] item
1737  *   Item specification.
1738  * @param[in] item_flags
1739  *   Bit-fields that holds the items detected until now.
1740  * @param[out] error
1741  *   Pointer to error structure.
1742  *
1743  * @return
1744  *   0 on success, a negative errno value otherwise and rte_errno is set.
1745  */
1746 static int
1747 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
1748                           const struct rte_flow_item *item,
1749                           uint64_t item_flags,
1750                           struct rte_flow_error *error)
1751 {
1752         struct mlx5_priv *priv = dev->data->dev_private;
1753         const struct rte_flow_item_gtp *spec = item->spec;
1754         const struct rte_flow_item_gtp *mask = item->mask;
1755         const struct rte_flow_item_gtp nic_mask = {
1756                 .v_pt_rsv_flags = MLX5_GTP_FLAGS_MASK,
1757                 .msg_type = 0xff,
1758                 .teid = RTE_BE32(0xffffffff),
1759         };
1760
1761         if (!priv->config.hca_attr.tunnel_stateless_gtp)
1762                 return rte_flow_error_set(error, ENOTSUP,
1763                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1764                                           "GTP support is not enabled");
1765         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1766                 return rte_flow_error_set(error, ENOTSUP,
1767                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1768                                           "multiple tunnel layers not"
1769                                           " supported");
1770         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1771                 return rte_flow_error_set(error, EINVAL,
1772                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1773                                           "no outer UDP layer found");
1774         if (!mask)
1775                 mask = &rte_flow_item_gtp_mask;
1776         if (spec && spec->v_pt_rsv_flags & ~MLX5_GTP_FLAGS_MASK)
1777                 return rte_flow_error_set(error, ENOTSUP,
1778                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1779                                           "Match is supported for GTP"
1780                                           " flags only");
1781         return mlx5_flow_item_acceptable
1782                 (item, (const uint8_t *)mask,
1783                  (const uint8_t *)&nic_mask,
1784                  sizeof(struct rte_flow_item_gtp),
1785                  error);
1786 }
1787
1788 /**
1789  * Validate the pop VLAN action.
1790  *
1791  * @param[in] dev
1792  *   Pointer to the rte_eth_dev structure.
1793  * @param[in] action_flags
1794  *   Holds the actions detected until now.
1795  * @param[in] action
1796  *   Pointer to the pop vlan action.
1797  * @param[in] item_flags
1798  *   The items found in this flow rule.
1799  * @param[in] attr
1800  *   Pointer to flow attributes.
1801  * @param[out] error
1802  *   Pointer to error structure.
1803  *
1804  * @return
1805  *   0 on success, a negative errno value otherwise and rte_errno is set.
1806  */
1807 static int
1808 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
1809                                  uint64_t action_flags,
1810                                  const struct rte_flow_action *action,
1811                                  uint64_t item_flags,
1812                                  const struct rte_flow_attr *attr,
1813                                  struct rte_flow_error *error)
1814 {
1815         const struct mlx5_priv *priv = dev->data->dev_private;
1816
1817         (void)action;
1818         (void)attr;
1819         if (!priv->sh->pop_vlan_action)
1820                 return rte_flow_error_set(error, ENOTSUP,
1821                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1822                                           NULL,
1823                                           "pop vlan action is not supported");
1824         if (attr->egress)
1825                 return rte_flow_error_set(error, ENOTSUP,
1826                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
1827                                           NULL,
1828                                           "pop vlan action not supported for "
1829                                           "egress");
1830         if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
1831                 return rte_flow_error_set(error, ENOTSUP,
1832                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1833                                           "no support for multiple VLAN "
1834                                           "actions");
1835         /* Pop VLAN with preceding Decap requires inner header with VLAN. */
1836         if ((action_flags & MLX5_FLOW_ACTION_DECAP) &&
1837             !(item_flags & MLX5_FLOW_LAYER_INNER_VLAN))
1838                 return rte_flow_error_set(error, ENOTSUP,
1839                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1840                                           NULL,
1841                                           "cannot pop vlan after decap without "
1842                                           "match on inner vlan in the flow");
1843         /* Pop VLAN without preceding Decap requires outer header with VLAN. */
1844         if (!(action_flags & MLX5_FLOW_ACTION_DECAP) &&
1845             !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
1846                 return rte_flow_error_set(error, ENOTSUP,
1847                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1848                                           NULL,
1849                                           "cannot pop vlan without a "
1850                                           "match on (outer) vlan in the flow");
1851         if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1852                 return rte_flow_error_set(error, EINVAL,
1853                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1854                                           "wrong action order, port_id should "
1855                                           "be after pop VLAN action");
1856         if (!attr->transfer && priv->representor)
1857                 return rte_flow_error_set(error, ENOTSUP,
1858                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1859                                           "pop vlan action for VF representor "
1860                                           "not supported on NIC table");
1861         return 0;
1862 }
1863
1864 /**
1865  * Get VLAN default info from vlan match info.
1866  *
1867  * @param[in] items
1868  *   the list of item specifications.
1869  * @param[out] vlan
1870  *   pointer VLAN info to fill to.
1871  *
1872  * @return
1873  *   0 on success, a negative errno value otherwise and rte_errno is set.
1874  */
1875 static void
1876 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
1877                                   struct rte_vlan_hdr *vlan)
1878 {
1879         const struct rte_flow_item_vlan nic_mask = {
1880                 .tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
1881                                 MLX5DV_FLOW_VLAN_VID_MASK),
1882                 .inner_type = RTE_BE16(0xffff),
1883         };
1884
1885         if (items == NULL)
1886                 return;
1887         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1888                 int type = items->type;
1889
1890                 if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
1891                     type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
1892                         break;
1893         }
1894         if (items->type != RTE_FLOW_ITEM_TYPE_END) {
1895                 const struct rte_flow_item_vlan *vlan_m = items->mask;
1896                 const struct rte_flow_item_vlan *vlan_v = items->spec;
1897
1898                 /* If VLAN item in pattern doesn't contain data, return here. */
1899                 if (!vlan_v)
1900                         return;
1901                 if (!vlan_m)
1902                         vlan_m = &nic_mask;
1903                 /* Only full match values are accepted */
1904                 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
1905                      MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
1906                         vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
1907                         vlan->vlan_tci |=
1908                                 rte_be_to_cpu_16(vlan_v->tci &
1909                                                  MLX5DV_FLOW_VLAN_PCP_MASK_BE);
1910                 }
1911                 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
1912                      MLX5DV_FLOW_VLAN_VID_MASK_BE) {
1913                         vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
1914                         vlan->vlan_tci |=
1915                                 rte_be_to_cpu_16(vlan_v->tci &
1916                                                  MLX5DV_FLOW_VLAN_VID_MASK_BE);
1917                 }
1918                 if (vlan_m->inner_type == nic_mask.inner_type)
1919                         vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
1920                                                            vlan_m->inner_type);
1921         }
1922 }
1923
1924 /**
1925  * Validate the push VLAN action.
1926  *
1927  * @param[in] dev
1928  *   Pointer to the rte_eth_dev structure.
1929  * @param[in] action_flags
1930  *   Holds the actions detected until now.
1931  * @param[in] item_flags
1932  *   The items found in this flow rule.
1933  * @param[in] action
1934  *   Pointer to the action structure.
1935  * @param[in] attr
1936  *   Pointer to flow attributes
1937  * @param[out] error
1938  *   Pointer to error structure.
1939  *
1940  * @return
1941  *   0 on success, a negative errno value otherwise and rte_errno is set.
1942  */
1943 static int
1944 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
1945                                   uint64_t action_flags,
1946                                   const struct rte_flow_item_vlan *vlan_m,
1947                                   const struct rte_flow_action *action,
1948                                   const struct rte_flow_attr *attr,
1949                                   struct rte_flow_error *error)
1950 {
1951         const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
1952         const struct mlx5_priv *priv = dev->data->dev_private;
1953
1954         if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
1955             push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
1956                 return rte_flow_error_set(error, EINVAL,
1957                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1958                                           "invalid vlan ethertype");
1959         if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1960                 return rte_flow_error_set(error, EINVAL,
1961                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1962                                           "wrong action order, port_id should "
1963                                           "be after push VLAN");
1964         if (!attr->transfer && priv->representor)
1965                 return rte_flow_error_set(error, ENOTSUP,
1966                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1967                                           "push vlan action for VF representor "
1968                                           "not supported on NIC table");
1969         if (vlan_m &&
1970             (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
1971             (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
1972                 MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
1973             !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
1974             !(mlx5_flow_find_action
1975                 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
1976                 return rte_flow_error_set(error, EINVAL,
1977                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1978                                           "not full match mask on VLAN PCP and "
1979                                           "there is no of_set_vlan_pcp action, "
1980                                           "push VLAN action cannot figure out "
1981                                           "PCP value");
1982         if (vlan_m &&
1983             (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
1984             (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
1985                 MLX5DV_FLOW_VLAN_VID_MASK_BE &&
1986             !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
1987             !(mlx5_flow_find_action
1988                 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
1989                 return rte_flow_error_set(error, EINVAL,
1990                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1991                                           "not full match mask on VLAN VID and "
1992                                           "there is no of_set_vlan_vid action, "
1993                                           "push VLAN action cannot figure out "
1994                                           "VID value");
1995         (void)attr;
1996         return 0;
1997 }
1998
1999 /**
2000  * Validate the set VLAN PCP.
2001  *
2002  * @param[in] action_flags
2003  *   Holds the actions detected until now.
2004  * @param[in] actions
2005  *   Pointer to the list of actions remaining in the flow rule.
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_set_vlan_pcp(uint64_t action_flags,
2014                                      const struct rte_flow_action actions[],
2015                                      struct rte_flow_error *error)
2016 {
2017         const struct rte_flow_action *action = actions;
2018         const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
2019
2020         if (conf->vlan_pcp > 7)
2021                 return rte_flow_error_set(error, EINVAL,
2022                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2023                                           "VLAN PCP value is too big");
2024         if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
2025                 return rte_flow_error_set(error, ENOTSUP,
2026                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2027                                           "set VLAN PCP action must follow "
2028                                           "the push VLAN action");
2029         if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
2030                 return rte_flow_error_set(error, ENOTSUP,
2031                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2032                                           "Multiple VLAN PCP modification are "
2033                                           "not supported");
2034         if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2035                 return rte_flow_error_set(error, EINVAL,
2036                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2037                                           "wrong action order, port_id should "
2038                                           "be after set VLAN PCP");
2039         return 0;
2040 }
2041
2042 /**
2043  * Validate the set VLAN VID.
2044  *
2045  * @param[in] item_flags
2046  *   Holds the items detected in this rule.
2047  * @param[in] action_flags
2048  *   Holds the actions detected until now.
2049  * @param[in] actions
2050  *   Pointer to the list of actions remaining in the flow rule.
2051  * @param[out] error
2052  *   Pointer to error structure.
2053  *
2054  * @return
2055  *   0 on success, a negative errno value otherwise and rte_errno is set.
2056  */
2057 static int
2058 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
2059                                      uint64_t action_flags,
2060                                      const struct rte_flow_action actions[],
2061                                      struct rte_flow_error *error)
2062 {
2063         const struct rte_flow_action *action = actions;
2064         const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
2065
2066         if (rte_be_to_cpu_16(conf->vlan_vid) > 0xFFE)
2067                 return rte_flow_error_set(error, EINVAL,
2068                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2069                                           "VLAN VID value is too big");
2070         if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2071             !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2072                 return rte_flow_error_set(error, ENOTSUP,
2073                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2074                                           "set VLAN VID action must follow push"
2075                                           " VLAN action or match on VLAN item");
2076         if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
2077                 return rte_flow_error_set(error, ENOTSUP,
2078                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2079                                           "Multiple VLAN VID modifications are "
2080                                           "not supported");
2081         if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2082                 return rte_flow_error_set(error, EINVAL,
2083                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2084                                           "wrong action order, port_id should "
2085                                           "be after set VLAN VID");
2086         return 0;
2087 }
2088
2089 /*
2090  * Validate the FLAG action.
2091  *
2092  * @param[in] dev
2093  *   Pointer to the rte_eth_dev structure.
2094  * @param[in] action_flags
2095  *   Holds the actions detected until now.
2096  * @param[in] attr
2097  *   Pointer to flow attributes
2098  * @param[out] error
2099  *   Pointer to error structure.
2100  *
2101  * @return
2102  *   0 on success, a negative errno value otherwise and rte_errno is set.
2103  */
2104 static int
2105 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
2106                              uint64_t action_flags,
2107                              const struct rte_flow_attr *attr,
2108                              struct rte_flow_error *error)
2109 {
2110         struct mlx5_priv *priv = dev->data->dev_private;
2111         struct mlx5_dev_config *config = &priv->config;
2112         int ret;
2113
2114         /* Fall back if no extended metadata register support. */
2115         if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2116                 return mlx5_flow_validate_action_flag(action_flags, attr,
2117                                                       error);
2118         /* Extensive metadata mode requires registers. */
2119         if (!mlx5_flow_ext_mreg_supported(dev))
2120                 return rte_flow_error_set(error, ENOTSUP,
2121                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2122                                           "no metadata registers "
2123                                           "to support flag action");
2124         if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
2125                 return rte_flow_error_set(error, ENOTSUP,
2126                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2127                                           "extended metadata register"
2128                                           " isn't available");
2129         ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2130         if (ret < 0)
2131                 return ret;
2132         MLX5_ASSERT(ret > 0);
2133         if (action_flags & MLX5_FLOW_ACTION_MARK)
2134                 return rte_flow_error_set(error, EINVAL,
2135                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2136                                           "can't mark and flag in same flow");
2137         if (action_flags & MLX5_FLOW_ACTION_FLAG)
2138                 return rte_flow_error_set(error, EINVAL,
2139                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2140                                           "can't have 2 flag"
2141                                           " actions in same flow");
2142         return 0;
2143 }
2144
2145 /**
2146  * Validate MARK action.
2147  *
2148  * @param[in] dev
2149  *   Pointer to the rte_eth_dev structure.
2150  * @param[in] action
2151  *   Pointer to action.
2152  * @param[in] action_flags
2153  *   Holds the actions detected until now.
2154  * @param[in] attr
2155  *   Pointer to flow attributes
2156  * @param[out] error
2157  *   Pointer to error structure.
2158  *
2159  * @return
2160  *   0 on success, a negative errno value otherwise and rte_errno is set.
2161  */
2162 static int
2163 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
2164                              const struct rte_flow_action *action,
2165                              uint64_t action_flags,
2166                              const struct rte_flow_attr *attr,
2167                              struct rte_flow_error *error)
2168 {
2169         struct mlx5_priv *priv = dev->data->dev_private;
2170         struct mlx5_dev_config *config = &priv->config;
2171         const struct rte_flow_action_mark *mark = action->conf;
2172         int ret;
2173
2174         /* Fall back if no extended metadata register support. */
2175         if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2176                 return mlx5_flow_validate_action_mark(action, action_flags,
2177                                                       attr, error);
2178         /* Extensive metadata mode requires registers. */
2179         if (!mlx5_flow_ext_mreg_supported(dev))
2180                 return rte_flow_error_set(error, ENOTSUP,
2181                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2182                                           "no metadata registers "
2183                                           "to support mark action");
2184         if (!priv->sh->dv_mark_mask)
2185                 return rte_flow_error_set(error, ENOTSUP,
2186                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2187                                           "extended metadata register"
2188                                           " isn't available");
2189         ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2190         if (ret < 0)
2191                 return ret;
2192         MLX5_ASSERT(ret > 0);
2193         if (!mark)
2194                 return rte_flow_error_set(error, EINVAL,
2195                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2196                                           "configuration cannot be null");
2197         if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
2198                 return rte_flow_error_set(error, EINVAL,
2199                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2200                                           &mark->id,
2201                                           "mark id exceeds the limit");
2202         if (action_flags & MLX5_FLOW_ACTION_FLAG)
2203                 return rte_flow_error_set(error, EINVAL,
2204                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2205                                           "can't flag and mark in same flow");
2206         if (action_flags & MLX5_FLOW_ACTION_MARK)
2207                 return rte_flow_error_set(error, EINVAL,
2208                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2209                                           "can't have 2 mark actions in same"
2210                                           " flow");
2211         return 0;
2212 }
2213
2214 /**
2215  * Validate SET_META action.
2216  *
2217  * @param[in] dev
2218  *   Pointer to the rte_eth_dev structure.
2219  * @param[in] action
2220  *   Pointer to the action structure.
2221  * @param[in] action_flags
2222  *   Holds the actions detected until now.
2223  * @param[in] attr
2224  *   Pointer to flow attributes
2225  * @param[out] error
2226  *   Pointer to error structure.
2227  *
2228  * @return
2229  *   0 on success, a negative errno value otherwise and rte_errno is set.
2230  */
2231 static int
2232 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
2233                                  const struct rte_flow_action *action,
2234                                  uint64_t action_flags __rte_unused,
2235                                  const struct rte_flow_attr *attr,
2236                                  struct rte_flow_error *error)
2237 {
2238         const struct rte_flow_action_set_meta *conf;
2239         uint32_t nic_mask = UINT32_MAX;
2240         int reg;
2241
2242         if (!mlx5_flow_ext_mreg_supported(dev))
2243                 return rte_flow_error_set(error, ENOTSUP,
2244                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2245                                           "extended metadata register"
2246                                           " isn't supported");
2247         reg = flow_dv_get_metadata_reg(dev, attr, error);
2248         if (reg < 0)
2249                 return reg;
2250         if (reg != REG_A && reg != REG_B) {
2251                 struct mlx5_priv *priv = dev->data->dev_private;
2252
2253                 nic_mask = priv->sh->dv_meta_mask;
2254         }
2255         if (!(action->conf))
2256                 return rte_flow_error_set(error, EINVAL,
2257                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2258                                           "configuration cannot be null");
2259         conf = (const struct rte_flow_action_set_meta *)action->conf;
2260         if (!conf->mask)
2261                 return rte_flow_error_set(error, EINVAL,
2262                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2263                                           "zero mask doesn't have any effect");
2264         if (conf->mask & ~nic_mask)
2265                 return rte_flow_error_set(error, EINVAL,
2266                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2267                                           "meta data must be within reg C0");
2268         return 0;
2269 }
2270
2271 /**
2272  * Validate SET_TAG action.
2273  *
2274  * @param[in] dev
2275  *   Pointer to the rte_eth_dev structure.
2276  * @param[in] action
2277  *   Pointer to the action structure.
2278  * @param[in] action_flags
2279  *   Holds the actions detected until now.
2280  * @param[in] attr
2281  *   Pointer to flow attributes
2282  * @param[out] error
2283  *   Pointer to error structure.
2284  *
2285  * @return
2286  *   0 on success, a negative errno value otherwise and rte_errno is set.
2287  */
2288 static int
2289 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
2290                                 const struct rte_flow_action *action,
2291                                 uint64_t action_flags,
2292                                 const struct rte_flow_attr *attr,
2293                                 struct rte_flow_error *error)
2294 {
2295         const struct rte_flow_action_set_tag *conf;
2296         const uint64_t terminal_action_flags =
2297                 MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
2298                 MLX5_FLOW_ACTION_RSS;
2299         int ret;
2300
2301         if (!mlx5_flow_ext_mreg_supported(dev))
2302                 return rte_flow_error_set(error, ENOTSUP,
2303                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2304                                           "extensive metadata register"
2305                                           " isn't supported");
2306         if (!(action->conf))
2307                 return rte_flow_error_set(error, EINVAL,
2308                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2309                                           "configuration cannot be null");
2310         conf = (const struct rte_flow_action_set_tag *)action->conf;
2311         if (!conf->mask)
2312                 return rte_flow_error_set(error, EINVAL,
2313                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2314                                           "zero mask doesn't have any effect");
2315         ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
2316         if (ret < 0)
2317                 return ret;
2318         if (!attr->transfer && attr->ingress &&
2319             (action_flags & terminal_action_flags))
2320                 return rte_flow_error_set(error, EINVAL,
2321                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2322                                           "set_tag has no effect"
2323                                           " with terminal actions");
2324         return 0;
2325 }
2326
2327 /**
2328  * Validate count action.
2329  *
2330  * @param[in] dev
2331  *   Pointer to rte_eth_dev structure.
2332  * @param[out] error
2333  *   Pointer to error structure.
2334  *
2335  * @return
2336  *   0 on success, a negative errno value otherwise and rte_errno is set.
2337  */
2338 static int
2339 flow_dv_validate_action_count(struct rte_eth_dev *dev,
2340                               struct rte_flow_error *error)
2341 {
2342         struct mlx5_priv *priv = dev->data->dev_private;
2343
2344         if (!priv->config.devx)
2345                 goto notsup_err;
2346 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
2347         return 0;
2348 #endif
2349 notsup_err:
2350         return rte_flow_error_set
2351                       (error, ENOTSUP,
2352                        RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2353                        NULL,
2354                        "count action not supported");
2355 }
2356
2357 /**
2358  * Validate the L2 encap action.
2359  *
2360  * @param[in] dev
2361  *   Pointer to the rte_eth_dev structure.
2362  * @param[in] action_flags
2363  *   Holds the actions detected until now.
2364  * @param[in] action
2365  *   Pointer to the action structure.
2366  * @param[in] attr
2367  *   Pointer to flow attributes.
2368  * @param[out] error
2369  *   Pointer to error structure.
2370  *
2371  * @return
2372  *   0 on success, a negative errno value otherwise and rte_errno is set.
2373  */
2374 static int
2375 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
2376                                  uint64_t action_flags,
2377                                  const struct rte_flow_action *action,
2378                                  const struct rte_flow_attr *attr,
2379                                  struct rte_flow_error *error)
2380 {
2381         const struct mlx5_priv *priv = dev->data->dev_private;
2382
2383         if (!(action->conf))
2384                 return rte_flow_error_set(error, EINVAL,
2385                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
2386                                           "configuration cannot be null");
2387         if (action_flags & MLX5_FLOW_ACTION_ENCAP)
2388                 return rte_flow_error_set(error, EINVAL,
2389                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2390                                           "can only have a single encap action "
2391                                           "in a flow");
2392         if (!attr->transfer && priv->representor)
2393                 return rte_flow_error_set(error, ENOTSUP,
2394                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2395                                           "encap action for VF representor "
2396                                           "not supported on NIC table");
2397         return 0;
2398 }
2399
2400 /**
2401  * Validate a decap action.
2402  *
2403  * @param[in] dev
2404  *   Pointer to the rte_eth_dev structure.
2405  * @param[in] action_flags
2406  *   Holds the actions detected until now.
2407  * @param[in] attr
2408  *   Pointer to flow attributes
2409  * @param[out] error
2410  *   Pointer to error structure.
2411  *
2412  * @return
2413  *   0 on success, a negative errno value otherwise and rte_errno is set.
2414  */
2415 static int
2416 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
2417                               uint64_t action_flags,
2418                               const struct rte_flow_attr *attr,
2419                               struct rte_flow_error *error)
2420 {
2421         const struct mlx5_priv *priv = dev->data->dev_private;
2422
2423         if (priv->config.hca_attr.scatter_fcs_w_decap_disable &&
2424             !priv->config.decap_en)
2425                 return rte_flow_error_set(error, ENOTSUP,
2426                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2427                                           "decap is not enabled");
2428         if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
2429                 return rte_flow_error_set(error, ENOTSUP,
2430                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2431                                           action_flags &
2432                                           MLX5_FLOW_ACTION_DECAP ? "can only "
2433                                           "have a single decap action" : "decap "
2434                                           "after encap is not supported");
2435         if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
2436                 return rte_flow_error_set(error, EINVAL,
2437                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2438                                           "can't have decap action after"
2439                                           " modify action");
2440         if (attr->egress)
2441                 return rte_flow_error_set(error, ENOTSUP,
2442                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2443                                           NULL,
2444                                           "decap action not supported for "
2445                                           "egress");
2446         if (!attr->transfer && priv->representor)
2447                 return rte_flow_error_set(error, ENOTSUP,
2448                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2449                                           "decap action for VF representor "
2450                                           "not supported on NIC table");
2451         return 0;
2452 }
2453
2454 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
2455
2456 /**
2457  * Validate the raw encap and decap actions.
2458  *
2459  * @param[in] dev
2460  *   Pointer to the rte_eth_dev structure.
2461  * @param[in] decap
2462  *   Pointer to the decap action.
2463  * @param[in] encap
2464  *   Pointer to the encap action.
2465  * @param[in] attr
2466  *   Pointer to flow attributes
2467  * @param[in/out] action_flags
2468  *   Holds the actions detected until now.
2469  * @param[out] actions_n
2470  *   pointer to the number of actions counter.
2471  * @param[out] error
2472  *   Pointer to error structure.
2473  *
2474  * @return
2475  *   0 on success, a negative errno value otherwise and rte_errno is set.
2476  */
2477 static int
2478 flow_dv_validate_action_raw_encap_decap
2479         (struct rte_eth_dev *dev,
2480          const struct rte_flow_action_raw_decap *decap,
2481          const struct rte_flow_action_raw_encap *encap,
2482          const struct rte_flow_attr *attr, uint64_t *action_flags,
2483          int *actions_n, struct rte_flow_error *error)
2484 {
2485         const struct mlx5_priv *priv = dev->data->dev_private;
2486         int ret;
2487
2488         if (encap && (!encap->size || !encap->data))
2489                 return rte_flow_error_set(error, EINVAL,
2490                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2491                                           "raw encap data cannot be empty");
2492         if (decap && encap) {
2493                 if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
2494                     encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2495                         /* L3 encap. */
2496                         decap = NULL;
2497                 else if (encap->size <=
2498                            MLX5_ENCAPSULATION_DECISION_SIZE &&
2499                            decap->size >
2500                            MLX5_ENCAPSULATION_DECISION_SIZE)
2501                         /* L3 decap. */
2502                         encap = NULL;
2503                 else if (encap->size >
2504                            MLX5_ENCAPSULATION_DECISION_SIZE &&
2505                            decap->size >
2506                            MLX5_ENCAPSULATION_DECISION_SIZE)
2507                         /* 2 L2 actions: encap and decap. */
2508                         ;
2509                 else
2510                         return rte_flow_error_set(error,
2511                                 ENOTSUP,
2512                                 RTE_FLOW_ERROR_TYPE_ACTION,
2513                                 NULL, "unsupported too small "
2514                                 "raw decap and too small raw "
2515                                 "encap combination");
2516         }
2517         if (decap) {
2518                 ret = flow_dv_validate_action_decap(dev, *action_flags, attr,
2519                                                     error);
2520                 if (ret < 0)
2521                         return ret;
2522                 *action_flags |= MLX5_FLOW_ACTION_DECAP;
2523                 ++(*actions_n);
2524         }
2525         if (encap) {
2526                 if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
2527                         return rte_flow_error_set(error, ENOTSUP,
2528                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2529                                                   NULL,
2530                                                   "small raw encap size");
2531                 if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
2532                         return rte_flow_error_set(error, EINVAL,
2533                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2534                                                   NULL,
2535                                                   "more than one encap action");
2536                 if (!attr->transfer && priv->representor)
2537                         return rte_flow_error_set
2538                                         (error, ENOTSUP,
2539                                          RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2540                                          "encap action for VF representor "
2541                                          "not supported on NIC table");
2542                 *action_flags |= MLX5_FLOW_ACTION_ENCAP;
2543                 ++(*actions_n);
2544         }
2545         return 0;
2546 }
2547
2548 /**
2549  * Find existing encap/decap resource or create and register a new one.
2550  *
2551  * @param[in, out] dev
2552  *   Pointer to rte_eth_dev structure.
2553  * @param[in, out] resource
2554  *   Pointer to encap/decap resource.
2555  * @parm[in, out] dev_flow
2556  *   Pointer to the dev_flow.
2557  * @param[out] error
2558  *   pointer to error structure.
2559  *
2560  * @return
2561  *   0 on success otherwise -errno and errno is set.
2562  */
2563 static int
2564 flow_dv_encap_decap_resource_register
2565                         (struct rte_eth_dev *dev,
2566                          struct mlx5_flow_dv_encap_decap_resource *resource,
2567                          struct mlx5_flow *dev_flow,
2568                          struct rte_flow_error *error)
2569 {
2570         struct mlx5_priv *priv = dev->data->dev_private;
2571         struct mlx5_dev_ctx_shared *sh = priv->sh;
2572         struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2573         struct mlx5dv_dr_domain *domain;
2574         uint32_t idx = 0;
2575         int ret;
2576
2577         resource->flags = dev_flow->dv.group ? 0 : 1;
2578         if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2579                 domain = sh->fdb_domain;
2580         else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2581                 domain = sh->rx_domain;
2582         else
2583                 domain = sh->tx_domain;
2584         /* Lookup a matching resource from cache. */
2585         ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], sh->encaps_decaps, idx,
2586                       cache_resource, next) {
2587                 if (resource->reformat_type == cache_resource->reformat_type &&
2588                     resource->ft_type == cache_resource->ft_type &&
2589                     resource->flags == cache_resource->flags &&
2590                     resource->size == cache_resource->size &&
2591                     !memcmp((const void *)resource->buf,
2592                             (const void *)cache_resource->buf,
2593                             resource->size)) {
2594                         DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
2595                                 (void *)cache_resource,
2596                                 rte_atomic32_read(&cache_resource->refcnt));
2597                         rte_atomic32_inc(&cache_resource->refcnt);
2598                         dev_flow->handle->dvh.rix_encap_decap = idx;
2599                         dev_flow->dv.encap_decap = cache_resource;
2600                         return 0;
2601                 }
2602         }
2603         /* Register new encap/decap resource. */
2604         cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2605                                        &dev_flow->handle->dvh.rix_encap_decap);
2606         if (!cache_resource)
2607                 return rte_flow_error_set(error, ENOMEM,
2608                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2609                                           "cannot allocate resource memory");
2610         *cache_resource = *resource;
2611         ret = mlx5_flow_os_create_flow_action_packet_reformat
2612                                         (sh->ctx, domain, cache_resource,
2613                                          &cache_resource->action);
2614         if (ret) {
2615                 mlx5_free(cache_resource);
2616                 return rte_flow_error_set(error, ENOMEM,
2617                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2618                                           NULL, "cannot create action");
2619         }
2620         rte_atomic32_init(&cache_resource->refcnt);
2621         rte_atomic32_inc(&cache_resource->refcnt);
2622         ILIST_INSERT(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], &sh->encaps_decaps,
2623                      dev_flow->handle->dvh.rix_encap_decap, cache_resource,
2624                      next);
2625         dev_flow->dv.encap_decap = cache_resource;
2626         DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
2627                 (void *)cache_resource,
2628                 rte_atomic32_read(&cache_resource->refcnt));
2629         return 0;
2630 }
2631
2632 /**
2633  * Find existing table jump resource or create and register a new one.
2634  *
2635  * @param[in, out] dev
2636  *   Pointer to rte_eth_dev structure.
2637  * @param[in, out] tbl
2638  *   Pointer to flow table resource.
2639  * @parm[in, out] dev_flow
2640  *   Pointer to the dev_flow.
2641  * @param[out] error
2642  *   pointer to error structure.
2643  *
2644  * @return
2645  *   0 on success otherwise -errno and errno is set.
2646  */
2647 static int
2648 flow_dv_jump_tbl_resource_register
2649                         (struct rte_eth_dev *dev __rte_unused,
2650                          struct mlx5_flow_tbl_resource *tbl,
2651                          struct mlx5_flow *dev_flow,
2652                          struct rte_flow_error *error)
2653 {
2654         struct mlx5_flow_tbl_data_entry *tbl_data =
2655                 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
2656         int cnt, ret;
2657
2658         MLX5_ASSERT(tbl);
2659         cnt = rte_atomic32_read(&tbl_data->jump.refcnt);
2660         if (!cnt) {
2661                 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
2662                                 (tbl->obj, &tbl_data->jump.action);
2663                 if (ret)
2664                         return rte_flow_error_set(error, ENOMEM,
2665                                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2666                                         NULL, "cannot create jump action");
2667                 DRV_LOG(DEBUG, "new jump table resource %p: refcnt %d++",
2668                         (void *)&tbl_data->jump, cnt);
2669         } else {
2670                 /* old jump should not make the table ref++. */
2671                 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
2672                 MLX5_ASSERT(tbl_data->jump.action);
2673                 DRV_LOG(DEBUG, "existed jump table resource %p: refcnt %d++",
2674                         (void *)&tbl_data->jump, cnt);
2675         }
2676         rte_atomic32_inc(&tbl_data->jump.refcnt);
2677         dev_flow->handle->rix_jump = tbl_data->idx;
2678         dev_flow->dv.jump = &tbl_data->jump;
2679         return 0;
2680 }
2681
2682 /**
2683  * Find existing default miss resource or create and register a new one.
2684  *
2685  * @param[in, out] dev
2686  *   Pointer to rte_eth_dev structure.
2687  * @param[out] error
2688  *   pointer to error structure.
2689  *
2690  * @return
2691  *   0 on success otherwise -errno and errno is set.
2692  */
2693 static int
2694 flow_dv_default_miss_resource_register(struct rte_eth_dev *dev,
2695                 struct rte_flow_error *error)
2696 {
2697         struct mlx5_priv *priv = dev->data->dev_private;
2698         struct mlx5_dev_ctx_shared *sh = priv->sh;
2699         struct mlx5_flow_default_miss_resource *cache_resource =
2700                         &sh->default_miss;
2701         int cnt = rte_atomic32_read(&cache_resource->refcnt);
2702
2703         if (!cnt) {
2704                 MLX5_ASSERT(cache_resource->action);
2705                 cache_resource->action =
2706                 mlx5_glue->dr_create_flow_action_default_miss();
2707                 if (!cache_resource->action)
2708                         return rte_flow_error_set(error, ENOMEM,
2709                                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2710                                         "cannot create default miss action");
2711                 DRV_LOG(DEBUG, "new default miss resource %p: refcnt %d++",
2712                                 (void *)cache_resource->action, cnt);
2713         }
2714         rte_atomic32_inc(&cache_resource->refcnt);
2715         return 0;
2716 }
2717
2718 /**
2719  * Find existing table port ID resource or create and register a new one.
2720  *
2721  * @param[in, out] dev
2722  *   Pointer to rte_eth_dev structure.
2723  * @param[in, out] resource
2724  *   Pointer to port ID action resource.
2725  * @parm[in, out] dev_flow
2726  *   Pointer to the dev_flow.
2727  * @param[out] error
2728  *   pointer to error structure.
2729  *
2730  * @return
2731  *   0 on success otherwise -errno and errno is set.
2732  */
2733 static int
2734 flow_dv_port_id_action_resource_register
2735                         (struct rte_eth_dev *dev,
2736                          struct mlx5_flow_dv_port_id_action_resource *resource,
2737                          struct mlx5_flow *dev_flow,
2738                          struct rte_flow_error *error)
2739 {
2740         struct mlx5_priv *priv = dev->data->dev_private;
2741         struct mlx5_dev_ctx_shared *sh = priv->sh;
2742         struct mlx5_flow_dv_port_id_action_resource *cache_resource;
2743         uint32_t idx = 0;
2744         int ret;
2745
2746         /* Lookup a matching resource from cache. */
2747         ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PORT_ID], sh->port_id_action_list,
2748                       idx, cache_resource, next) {
2749                 if (resource->port_id == cache_resource->port_id) {
2750                         DRV_LOG(DEBUG, "port id action resource resource %p: "
2751                                 "refcnt %d++",
2752                                 (void *)cache_resource,
2753                                 rte_atomic32_read(&cache_resource->refcnt));
2754                         rte_atomic32_inc(&cache_resource->refcnt);
2755                         dev_flow->handle->rix_port_id_action = idx;
2756                         dev_flow->dv.port_id_action = cache_resource;
2757                         return 0;
2758                 }
2759         }
2760         /* Register new port id action resource. */
2761         cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID],
2762                                        &dev_flow->handle->rix_port_id_action);
2763         if (!cache_resource)
2764                 return rte_flow_error_set(error, ENOMEM,
2765                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2766                                           "cannot allocate resource memory");
2767         *cache_resource = *resource;
2768         ret = mlx5_flow_os_create_flow_action_dest_port
2769                                 (priv->sh->fdb_domain, resource->port_id,
2770                                  &cache_resource->action);
2771         if (ret) {
2772                 mlx5_free(cache_resource);
2773                 return rte_flow_error_set(error, ENOMEM,
2774                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2775                                           NULL, "cannot create action");
2776         }
2777         rte_atomic32_init(&cache_resource->refcnt);
2778         rte_atomic32_inc(&cache_resource->refcnt);
2779         ILIST_INSERT(sh->ipool[MLX5_IPOOL_PORT_ID], &sh->port_id_action_list,
2780                      dev_flow->handle->rix_port_id_action, cache_resource,
2781                      next);
2782         dev_flow->dv.port_id_action = cache_resource;
2783         DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
2784                 (void *)cache_resource,
2785                 rte_atomic32_read(&cache_resource->refcnt));
2786         return 0;
2787 }
2788
2789 /**
2790  * Find existing push vlan resource or create and register a new one.
2791  *
2792  * @param [in, out] dev
2793  *   Pointer to rte_eth_dev structure.
2794  * @param[in, out] resource
2795  *   Pointer to port ID action resource.
2796  * @parm[in, out] dev_flow
2797  *   Pointer to the dev_flow.
2798  * @param[out] error
2799  *   pointer to error structure.
2800  *
2801  * @return
2802  *   0 on success otherwise -errno and errno is set.
2803  */
2804 static int
2805 flow_dv_push_vlan_action_resource_register
2806                        (struct rte_eth_dev *dev,
2807                         struct mlx5_flow_dv_push_vlan_action_resource *resource,
2808                         struct mlx5_flow *dev_flow,
2809                         struct rte_flow_error *error)
2810 {
2811         struct mlx5_priv *priv = dev->data->dev_private;
2812         struct mlx5_dev_ctx_shared *sh = priv->sh;
2813         struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
2814         struct mlx5dv_dr_domain *domain;
2815         uint32_t idx = 0;
2816         int ret;
2817
2818         /* Lookup a matching resource from cache. */
2819         ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2820                       sh->push_vlan_action_list, idx, cache_resource, next) {
2821                 if (resource->vlan_tag == cache_resource->vlan_tag &&
2822                     resource->ft_type == cache_resource->ft_type) {
2823                         DRV_LOG(DEBUG, "push-VLAN action resource resource %p: "
2824                                 "refcnt %d++",
2825                                 (void *)cache_resource,
2826                                 rte_atomic32_read(&cache_resource->refcnt));
2827                         rte_atomic32_inc(&cache_resource->refcnt);
2828                         dev_flow->handle->dvh.rix_push_vlan = idx;
2829                         dev_flow->dv.push_vlan_res = cache_resource;
2830                         return 0;
2831                 }
2832         }
2833         /* Register new push_vlan action resource. */
2834         cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2835                                        &dev_flow->handle->dvh.rix_push_vlan);
2836         if (!cache_resource)
2837                 return rte_flow_error_set(error, ENOMEM,
2838                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2839                                           "cannot allocate resource memory");
2840         *cache_resource = *resource;
2841         if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2842                 domain = sh->fdb_domain;
2843         else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2844                 domain = sh->rx_domain;
2845         else
2846                 domain = sh->tx_domain;
2847         ret = mlx5_flow_os_create_flow_action_push_vlan
2848                                         (domain, resource->vlan_tag,
2849                                          &cache_resource->action);
2850         if (ret) {
2851                 mlx5_free(cache_resource);
2852                 return rte_flow_error_set(error, ENOMEM,
2853                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2854                                           NULL, "cannot create action");
2855         }
2856         rte_atomic32_init(&cache_resource->refcnt);
2857         rte_atomic32_inc(&cache_resource->refcnt);
2858         ILIST_INSERT(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2859                      &sh->push_vlan_action_list,
2860                      dev_flow->handle->dvh.rix_push_vlan,
2861                      cache_resource, next);
2862         dev_flow->dv.push_vlan_res = cache_resource;
2863         DRV_LOG(DEBUG, "new push vlan action resource %p: refcnt %d++",
2864                 (void *)cache_resource,
2865                 rte_atomic32_read(&cache_resource->refcnt));
2866         return 0;
2867 }
2868 /**
2869  * Get the size of specific rte_flow_item_type hdr size
2870  *
2871  * @param[in] item_type
2872  *   Tested rte_flow_item_type.
2873  *
2874  * @return
2875  *   sizeof struct item_type, 0 if void or irrelevant.
2876  */
2877 static size_t
2878 flow_dv_get_item_hdr_len(const enum rte_flow_item_type item_type)
2879 {
2880         size_t retval;
2881
2882         switch (item_type) {
2883         case RTE_FLOW_ITEM_TYPE_ETH:
2884                 retval = sizeof(struct rte_ether_hdr);
2885                 break;
2886         case RTE_FLOW_ITEM_TYPE_VLAN:
2887                 retval = sizeof(struct rte_vlan_hdr);
2888                 break;
2889         case RTE_FLOW_ITEM_TYPE_IPV4:
2890                 retval = sizeof(struct rte_ipv4_hdr);
2891                 break;
2892         case RTE_FLOW_ITEM_TYPE_IPV6:
2893                 retval = sizeof(struct rte_ipv6_hdr);
2894                 break;
2895         case RTE_FLOW_ITEM_TYPE_UDP:
2896                 retval = sizeof(struct rte_udp_hdr);
2897                 break;
2898         case RTE_FLOW_ITEM_TYPE_TCP:
2899                 retval = sizeof(struct rte_tcp_hdr);
2900                 break;
2901         case RTE_FLOW_ITEM_TYPE_VXLAN:
2902         case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2903                 retval = sizeof(struct rte_vxlan_hdr);
2904                 break;
2905         case RTE_FLOW_ITEM_TYPE_GRE:
2906         case RTE_FLOW_ITEM_TYPE_NVGRE:
2907                 retval = sizeof(struct rte_gre_hdr);
2908                 break;
2909         case RTE_FLOW_ITEM_TYPE_MPLS:
2910                 retval = sizeof(struct rte_mpls_hdr);
2911                 break;
2912         case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
2913         default:
2914                 retval = 0;
2915                 break;
2916         }
2917         return retval;
2918 }
2919
2920 #define MLX5_ENCAP_IPV4_VERSION         0x40
2921 #define MLX5_ENCAP_IPV4_IHL_MIN         0x05
2922 #define MLX5_ENCAP_IPV4_TTL_DEF         0x40
2923 #define MLX5_ENCAP_IPV6_VTC_FLOW        0x60000000
2924 #define MLX5_ENCAP_IPV6_HOP_LIMIT       0xff
2925 #define MLX5_ENCAP_VXLAN_FLAGS          0x08000000
2926 #define MLX5_ENCAP_VXLAN_GPE_FLAGS      0x04
2927
2928 /**
2929  * Convert the encap action data from list of rte_flow_item to raw buffer
2930  *
2931  * @param[in] items
2932  *   Pointer to rte_flow_item objects list.
2933  * @param[out] buf
2934  *   Pointer to the output buffer.
2935  * @param[out] size
2936  *   Pointer to the output buffer size.
2937  * @param[out] error
2938  *   Pointer to the error structure.
2939  *
2940  * @return
2941  *   0 on success, a negative errno value otherwise and rte_errno is set.
2942  */
2943 static int
2944 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
2945                            size_t *size, struct rte_flow_error *error)
2946 {
2947         struct rte_ether_hdr *eth = NULL;
2948         struct rte_vlan_hdr *vlan = NULL;
2949         struct rte_ipv4_hdr *ipv4 = NULL;
2950         struct rte_ipv6_hdr *ipv6 = NULL;
2951         struct rte_udp_hdr *udp = NULL;
2952         struct rte_vxlan_hdr *vxlan = NULL;
2953         struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
2954         struct rte_gre_hdr *gre = NULL;
2955         size_t len;
2956         size_t temp_size = 0;
2957
2958         if (!items)
2959                 return rte_flow_error_set(error, EINVAL,
2960                                           RTE_FLOW_ERROR_TYPE_ACTION,
2961                                           NULL, "invalid empty data");
2962         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2963                 len = flow_dv_get_item_hdr_len(items->type);
2964                 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
2965                         return rte_flow_error_set(error, EINVAL,
2966                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2967                                                   (void *)items->type,
2968                                                   "items total size is too big"
2969                                                   " for encap action");
2970                 rte_memcpy((void *)&buf[temp_size], items->spec, len);
2971                 switch (items->type) {
2972                 case RTE_FLOW_ITEM_TYPE_ETH:
2973                         eth = (struct rte_ether_hdr *)&buf[temp_size];
2974                         break;
2975                 case RTE_FLOW_ITEM_TYPE_VLAN:
2976                         vlan = (struct rte_vlan_hdr *)&buf[temp_size];
2977                         if (!eth)
2978                                 return rte_flow_error_set(error, EINVAL,
2979                                                 RTE_FLOW_ERROR_TYPE_ACTION,
2980                                                 (void *)items->type,
2981                                                 "eth header not found");
2982                         if (!eth->ether_type)
2983                                 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
2984                         break;
2985                 case RTE_FLOW_ITEM_TYPE_IPV4:
2986                         ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
2987                         if (!vlan && !eth)
2988                                 return rte_flow_error_set(error, EINVAL,
2989                                                 RTE_FLOW_ERROR_TYPE_ACTION,
2990                                                 (void *)items->type,
2991                                                 "neither eth nor vlan"
2992                                                 " header found");
2993                         if (vlan && !vlan->eth_proto)
2994                                 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2995                         else if (eth && !eth->ether_type)
2996                                 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2997                         if (!ipv4->version_ihl)
2998                                 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
2999                                                     MLX5_ENCAP_IPV4_IHL_MIN;
3000                         if (!ipv4->time_to_live)
3001                                 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
3002                         break;
3003                 case RTE_FLOW_ITEM_TYPE_IPV6:
3004                         ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
3005                         if (!vlan && !eth)
3006                                 return rte_flow_error_set(error, EINVAL,
3007                                                 RTE_FLOW_ERROR_TYPE_ACTION,
3008                                                 (void *)items->type,
3009                                                 "neither eth nor vlan"
3010                                                 " header found");
3011                         if (vlan && !vlan->eth_proto)
3012                                 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3013                         else if (eth && !eth->ether_type)
3014                                 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3015                         if (!ipv6->vtc_flow)
3016                                 ipv6->vtc_flow =
3017                                         RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
3018                         if (!ipv6->hop_limits)
3019                                 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
3020                         break;
3021                 case RTE_FLOW_ITEM_TYPE_UDP:
3022                         udp = (struct rte_udp_hdr *)&buf[temp_size];
3023                         if (!ipv4 && !ipv6)
3024                                 return rte_flow_error_set(error, EINVAL,
3025                                                 RTE_FLOW_ERROR_TYPE_ACTION,
3026                                                 (void *)items->type,
3027                                                 "ip header not found");
3028                         if (ipv4 && !ipv4->next_proto_id)
3029                                 ipv4->next_proto_id = IPPROTO_UDP;
3030                         else if (ipv6 && !ipv6->proto)
3031                                 ipv6->proto = IPPROTO_UDP;
3032                         break;
3033                 case RTE_FLOW_ITEM_TYPE_VXLAN:
3034                         vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
3035                         if (!udp)
3036                                 return rte_flow_error_set(error, EINVAL,
3037                                                 RTE_FLOW_ERROR_TYPE_ACTION,
3038                                                 (void *)items->type,
3039                                                 "udp header not found");
3040                         if (!udp->dst_port)
3041                                 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
3042                         if (!vxlan->vx_flags)
3043                                 vxlan->vx_flags =
3044                                         RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
3045                         break;
3046                 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3047                         vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
3048                         if (!udp)
3049                                 return rte_flow_error_set(error, EINVAL,
3050                                                 RTE_FLOW_ERROR_TYPE_ACTION,
3051                                                 (void *)items->type,
3052                                                 "udp header not found");
3053                         if (!vxlan_gpe->proto)
3054                                 return rte_flow_error_set(error, EINVAL,
3055                                                 RTE_FLOW_ERROR_TYPE_ACTION,
3056                                                 (void *)items->type,
3057                                                 "next protocol not found");
3058                         if (!udp->dst_port)
3059                                 udp->dst_port =
3060                                         RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
3061                         if (!vxlan_gpe->vx_flags)
3062                                 vxlan_gpe->vx_flags =
3063                                                 MLX5_ENCAP_VXLAN_GPE_FLAGS;
3064                         break;
3065                 case RTE_FLOW_ITEM_TYPE_GRE:
3066                 case RTE_FLOW_ITEM_TYPE_NVGRE:
3067                         gre = (struct rte_gre_hdr *)&buf[temp_size];
3068                         if (!gre->proto)
3069                                 return rte_flow_error_set(error, EINVAL,
3070                                                 RTE_FLOW_ERROR_TYPE_ACTION,
3071                                                 (void *)items->type,
3072                                                 "next protocol not found");
3073                         if (!ipv4 && !ipv6)
3074                                 return rte_flow_error_set(error, EINVAL,
3075                                                 RTE_FLOW_ERROR_TYPE_ACTION,
3076                                                 (void *)items->type,
3077                                                 "ip header not found");
3078                         if (ipv4 && !ipv4->next_proto_id)
3079                                 ipv4->next_proto_id = IPPROTO_GRE;
3080                         else if (ipv6 && !ipv6->proto)
3081                                 ipv6->proto = IPPROTO_GRE;
3082                         break;
3083                 case RTE_FLOW_ITEM_TYPE_VOID:
3084                         break;
3085                 default:
3086                         return rte_flow_error_set(error, EINVAL,
3087                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3088                                                   (void *)items->type,
3089                                                   "unsupported item type");
3090                         break;
3091                 }
3092                 temp_size += len;
3093         }
3094         *size = temp_size;
3095         return 0;
3096 }
3097
3098 static int
3099 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
3100 {
3101         struct rte_ether_hdr *eth = NULL;
3102         struct rte_vlan_hdr *vlan = NULL;
3103         struct rte_ipv6_hdr *ipv6 = NULL;
3104         struct rte_udp_hdr *udp = NULL;
3105         char *next_hdr;
3106         uint16_t proto;
3107
3108         eth = (struct rte_ether_hdr *)data;
3109         next_hdr = (char *)(eth + 1);
3110         proto = RTE_BE16(eth->ether_type);
3111
3112         /* VLAN skipping */
3113         while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
3114                 vlan = (struct rte_vlan_hdr *)next_hdr;
3115                 proto = RTE_BE16(vlan->eth_proto);
3116                 next_hdr += sizeof(struct rte_vlan_hdr);
3117         }
3118
3119         /* HW calculates IPv4 csum. no need to proceed */
3120         if (proto == RTE_ETHER_TYPE_IPV4)
3121                 return 0;
3122
3123         /* non IPv4/IPv6 header. not supported */
3124         if (proto != RTE_ETHER_TYPE_IPV6) {
3125                 return rte_flow_error_set(error, ENOTSUP,
3126                                           RTE_FLOW_ERROR_TYPE_ACTION,
3127                                           NULL, "Cannot offload non IPv4/IPv6");
3128         }
3129
3130         ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3131
3132         /* ignore non UDP */
3133         if (ipv6->proto != IPPROTO_UDP)
3134                 return 0;
3135
3136         udp = (struct rte_udp_hdr *)(ipv6 + 1);
3137         udp->dgram_cksum = 0;
3138
3139         return 0;
3140 }
3141
3142 /**
3143  * Convert L2 encap action to DV specification.
3144  *
3145  * @param[in] dev
3146  *   Pointer to rte_eth_dev structure.
3147  * @param[in] action
3148  *   Pointer to action structure.
3149  * @param[in, out] dev_flow
3150  *   Pointer to the mlx5_flow.
3151  * @param[in] transfer
3152  *   Mark if the flow is E-Switch flow.
3153  * @param[out] error
3154  *   Pointer to the error structure.
3155  *
3156  * @return
3157  *   0 on success, a negative errno value otherwise and rte_errno is set.
3158  */
3159 static int
3160 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3161                                const struct rte_flow_action *action,
3162                                struct mlx5_flow *dev_flow,
3163                                uint8_t transfer,
3164                                struct rte_flow_error *error)
3165 {
3166         const struct rte_flow_item *encap_data;
3167         const struct rte_flow_action_raw_encap *raw_encap_data;
3168         struct mlx5_flow_dv_encap_decap_resource res = {
3169                 .reformat_type =
3170                         MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3171                 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3172                                       MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3173         };
3174
3175         if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3176                 raw_encap_data =
3177                         (const struct rte_flow_action_raw_encap *)action->conf;
3178                 res.size = raw_encap_data->size;
3179                 memcpy(res.buf, raw_encap_data->data, res.size);
3180         } else {
3181                 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3182                         encap_data =
3183                                 ((const struct rte_flow_action_vxlan_encap *)
3184                                                 action->conf)->definition;
3185                 else
3186                         encap_data =
3187                                 ((const struct rte_flow_action_nvgre_encap *)
3188                                                 action->conf)->definition;
3189                 if (flow_dv_convert_encap_data(encap_data, res.buf,
3190                                                &res.size, error))
3191                         return -rte_errno;
3192         }
3193         if (flow_dv_zero_encap_udp_csum(res.buf, error))
3194                 return -rte_errno;
3195         if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3196                 return rte_flow_error_set(error, EINVAL,
3197                                           RTE_FLOW_ERROR_TYPE_ACTION,
3198                                           NULL, "can't create L2 encap action");
3199         return 0;
3200 }
3201
3202 /**
3203  * Convert L2 decap action to DV specification.
3204  *
3205  * @param[in] dev
3206  *   Pointer to rte_eth_dev structure.
3207  * @param[in, out] dev_flow
3208  *   Pointer to the mlx5_flow.
3209  * @param[in] transfer
3210  *   Mark if the flow is E-Switch flow.
3211  * @param[out] error
3212  *   Pointer to the error structure.
3213  *
3214  * @return
3215  *   0 on success, a negative errno value otherwise and rte_errno is set.
3216  */
3217 static int
3218 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
3219                                struct mlx5_flow *dev_flow,
3220                                uint8_t transfer,
3221                                struct rte_flow_error *error)
3222 {
3223         struct mlx5_flow_dv_encap_decap_resource res = {
3224                 .size = 0,
3225                 .reformat_type =
3226                         MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3227                 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3228                                       MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3229         };
3230
3231         if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3232                 return rte_flow_error_set(error, EINVAL,
3233                                           RTE_FLOW_ERROR_TYPE_ACTION,
3234                                           NULL, "can't create L2 decap action");
3235         return 0;
3236 }
3237
3238 /**
3239  * Convert raw decap/encap (L3 tunnel) action to DV specification.
3240  *
3241  * @param[in] dev
3242  *   Pointer to rte_eth_dev structure.
3243  * @param[in] action
3244  *   Pointer to action structure.
3245  * @param[in, out] dev_flow
3246  *   Pointer to the mlx5_flow.
3247  * @param[in] attr
3248  *   Pointer to the flow attributes.
3249  * @param[out] error
3250  *   Pointer to the error structure.
3251  *
3252  * @return
3253  *   0 on success, a negative errno value otherwise and rte_errno is set.
3254  */
3255 static int
3256 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3257                                 const struct rte_flow_action *action,
3258                                 struct mlx5_flow *dev_flow,
3259                                 const struct rte_flow_attr *attr,
3260                                 struct rte_flow_error *error)
3261 {
3262         const struct rte_flow_action_raw_encap *encap_data;
3263         struct mlx5_flow_dv_encap_decap_resource res;
3264
3265         memset(&res, 0, sizeof(res));
3266         encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3267         res.size = encap_data->size;
3268         memcpy(res.buf, encap_data->data, res.size);
3269         res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3270                 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3271                 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3272         if (attr->transfer)
3273                 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3274         else
3275                 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3276                                              MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3277         if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3278                 return rte_flow_error_set(error, EINVAL,
3279                                           RTE_FLOW_ERROR_TYPE_ACTION,
3280                                           NULL, "can't create encap action");
3281         return 0;
3282 }
3283
3284 /**
3285  * Create action push VLAN.
3286  *
3287  * @param[in] dev
3288  *   Pointer to rte_eth_dev structure.
3289  * @param[in] attr
3290  *   Pointer to the flow attributes.
3291  * @param[in] vlan
3292  *   Pointer to the vlan to push to the Ethernet header.
3293  * @param[in, out] dev_flow
3294  *   Pointer to the mlx5_flow.
3295  * @param[out] error
3296  *   Pointer to the error structure.
3297  *
3298  * @return
3299  *   0 on success, a negative errno value otherwise and rte_errno is set.
3300  */
3301 static int
3302 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3303                                 const struct rte_flow_attr *attr,
3304                                 const struct rte_vlan_hdr *vlan,
3305                                 struct mlx5_flow *dev_flow,
3306                                 struct rte_flow_error *error)
3307 {
3308         struct mlx5_flow_dv_push_vlan_action_resource res;
3309
3310         memset(&res, 0, sizeof(res));
3311         res.vlan_tag =
3312                 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3313                                  vlan->vlan_tci);
3314         if (attr->transfer)
3315                 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3316         else
3317                 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3318                                              MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3319         return flow_dv_push_vlan_action_resource_register
3320                                             (dev, &res, dev_flow, error);
3321 }
3322
3323 /**
3324  * Validate the modify-header actions.
3325  *
3326  * @param[in] action_flags
3327  *   Holds the actions detected until now.
3328  * @param[in] action
3329  *   Pointer to the modify action.
3330  * @param[out] error
3331  *   Pointer to error structure.
3332  *
3333  * @return
3334  *   0 on success, a negative errno value otherwise and rte_errno is set.
3335  */
3336 static int
3337 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3338                                    const struct rte_flow_action *action,
3339                                    struct rte_flow_error *error)
3340 {
3341         if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3342                 return rte_flow_error_set(error, EINVAL,
3343                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3344                                           NULL, "action configuration not set");
3345         if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3346                 return rte_flow_error_set(error, EINVAL,
3347                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3348                                           "can't have encap action before"
3349                                           " modify action");
3350         return 0;
3351 }
3352
3353 /**
3354  * Validate the modify-header MAC address actions.
3355  *
3356  * @param[in] action_flags
3357  *   Holds the actions detected until now.
3358  * @param[in] action
3359  *   Pointer to the modify action.
3360  * @param[in] item_flags
3361  *   Holds the items detected.
3362  * @param[out] error
3363  *   Pointer to error structure.
3364  *
3365  * @return
3366  *   0 on success, a negative errno value otherwise and rte_errno is set.
3367  */
3368 static int
3369 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3370                                    const struct rte_flow_action *action,
3371                                    const uint64_t item_flags,
3372                                    struct rte_flow_error *error)
3373 {
3374         int ret = 0;
3375
3376         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3377         if (!ret) {
3378                 if (!(item_flags & MLX5_FLOW_LAYER_L2))
3379                         return rte_flow_error_set(error, EINVAL,
3380                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3381                                                   NULL,
3382                                                   "no L2 item in pattern");
3383         }
3384         return ret;
3385 }
3386
3387 /**
3388  * Validate the modify-header IPv4 address actions.
3389  *
3390  * @param[in] action_flags
3391  *   Holds the actions detected until now.
3392  * @param[in] action
3393  *   Pointer to the modify action.
3394  * @param[in] item_flags
3395  *   Holds the items detected.
3396  * @param[out] error
3397  *   Pointer to error structure.
3398  *
3399  * @return
3400  *   0 on success, a negative errno value otherwise and rte_errno is set.
3401  */
3402 static int
3403 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3404                                     const struct rte_flow_action *action,
3405                                     const uint64_t item_flags,
3406                                     struct rte_flow_error *error)
3407 {
3408         int ret = 0;
3409         uint64_t layer;
3410
3411         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3412         if (!ret) {
3413                 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3414                                  MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3415                                  MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3416                 if (!(item_flags & layer))
3417                         return rte_flow_error_set(error, EINVAL,
3418                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3419                                                   NULL,
3420                                                   "no ipv4 item in pattern");
3421         }
3422         return ret;
3423 }
3424
3425 /**
3426  * Validate the modify-header IPv6 address actions.
3427  *
3428  * @param[in] action_flags
3429  *   Holds the actions detected until now.
3430  * @param[in] action
3431  *   Pointer to the modify action.
3432  * @param[in] item_flags
3433  *   Holds the items detected.
3434  * @param[out] error
3435  *   Pointer to error structure.
3436  *
3437  * @return
3438  *   0 on success, a negative errno value otherwise and rte_errno is set.
3439  */
3440 static int
3441 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3442                                     const struct rte_flow_action *action,
3443                                     const uint64_t item_flags,
3444                                     struct rte_flow_error *error)
3445 {
3446         int ret = 0;
3447         uint64_t layer;
3448
3449         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3450         if (!ret) {
3451                 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3452                                  MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3453                                  MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3454                 if (!(item_flags & layer))
3455                         return rte_flow_error_set(error, EINVAL,
3456                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3457                                                   NULL,
3458                                                   "no ipv6 item in pattern");
3459         }
3460         return ret;
3461 }
3462
3463 /**
3464  * Validate the modify-header TP actions.
3465  *
3466  * @param[in] action_flags
3467  *   Holds the actions detected until now.
3468  * @param[in] action
3469  *   Pointer to the modify action.
3470  * @param[in] item_flags
3471  *   Holds the items detected.
3472  * @param[out] error
3473  *   Pointer to error structure.
3474  *
3475  * @return
3476  *   0 on success, a negative errno value otherwise and rte_errno is set.
3477  */
3478 static int
3479 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3480                                   const struct rte_flow_action *action,
3481                                   const uint64_t item_flags,
3482                                   struct rte_flow_error *error)
3483 {
3484         int ret = 0;
3485         uint64_t layer;
3486
3487         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3488         if (!ret) {
3489                 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3490                                  MLX5_FLOW_LAYER_INNER_L4 :
3491                                  MLX5_FLOW_LAYER_OUTER_L4;
3492                 if (!(item_flags & layer))
3493                         return rte_flow_error_set(error, EINVAL,
3494                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3495                                                   NULL, "no transport layer "
3496                                                   "in pattern");
3497         }
3498         return ret;
3499 }
3500
3501 /**
3502  * Validate the modify-header actions of increment/decrement
3503  * TCP Sequence-number.
3504  *
3505  * @param[in] action_flags
3506  *   Holds the actions detected until now.
3507  * @param[in] action
3508  *   Pointer to the modify action.
3509  * @param[in] item_flags
3510  *   Holds the items detected.
3511  * @param[out] error
3512  *   Pointer to error structure.
3513  *
3514  * @return
3515  *   0 on success, a negative errno value otherwise and rte_errno is set.
3516  */
3517 static int
3518 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3519                                        const struct rte_flow_action *action,
3520                                        const uint64_t item_flags,
3521                                        struct rte_flow_error *error)
3522 {
3523         int ret = 0;
3524         uint64_t layer;
3525
3526         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3527         if (!ret) {
3528                 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3529                                  MLX5_FLOW_LAYER_INNER_L4_TCP :
3530                                  MLX5_FLOW_LAYER_OUTER_L4_TCP;
3531                 if (!(item_flags & layer))
3532                         return rte_flow_error_set(error, EINVAL,
3533                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3534                                                   NULL, "no TCP item in"
3535                                                   " pattern");
3536                 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3537                         (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3538                     (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3539                         (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3540                         return rte_flow_error_set(error, EINVAL,
3541                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3542                                                   NULL,
3543                                                   "cannot decrease and increase"
3544                                                   " TCP sequence number"
3545                                                   " at the same time");
3546         }
3547         return ret;
3548 }
3549
3550 /**
3551  * Validate the modify-header actions of increment/decrement
3552  * TCP Acknowledgment number.
3553  *
3554  * @param[in] action_flags
3555  *   Holds the actions detected until now.
3556  * @param[in] action
3557  *   Pointer to the modify action.
3558  * @param[in] item_flags
3559  *   Holds the items detected.
3560  * @param[out] error
3561  *   Pointer to error structure.
3562  *
3563  * @return
3564  *   0 on success, a negative errno value otherwise and rte_errno is set.
3565  */
3566 static int
3567 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3568                                        const struct rte_flow_action *action,
3569                                        const uint64_t item_flags,
3570                                        struct rte_flow_error *error)
3571 {
3572         int ret = 0;
3573         uint64_t layer;
3574
3575         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3576         if (!ret) {
3577                 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3578                                  MLX5_FLOW_LAYER_INNER_L4_TCP :
3579                                  MLX5_FLOW_LAYER_OUTER_L4_TCP;
3580                 if (!(item_flags & layer))
3581                         return rte_flow_error_set(error, EINVAL,
3582                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3583                                                   NULL, "no TCP item in"
3584                                                   " pattern");
3585                 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3586                         (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3587                     (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3588                         (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3589                         return rte_flow_error_set(error, EINVAL,
3590                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3591                                                   NULL,
3592                                                   "cannot decrease and increase"
3593                                                   " TCP acknowledgment number"
3594                                                   " at the same time");
3595         }
3596         return ret;
3597 }
3598
3599 /**
3600  * Validate the modify-header TTL actions.
3601  *
3602  * @param[in] action_flags
3603  *   Holds the actions detected until now.
3604  * @param[in] action
3605  *   Pointer to the modify action.
3606  * @param[in] item_flags
3607  *   Holds the items detected.
3608  * @param[out] error
3609  *   Pointer to error structure.
3610  *
3611  * @return
3612  *   0 on success, a negative errno value otherwise and rte_errno is set.
3613  */
3614 static int
3615 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3616                                    const struct rte_flow_action *action,
3617                                    const uint64_t item_flags,
3618                                    struct rte_flow_error *error)
3619 {
3620         int ret = 0;
3621         uint64_t layer;
3622
3623         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3624         if (!ret) {
3625                 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3626                                  MLX5_FLOW_LAYER_INNER_L3 :
3627                                  MLX5_FLOW_LAYER_OUTER_L3;
3628                 if (!(item_flags & layer))
3629                         return rte_flow_error_set(error, EINVAL,
3630                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3631                                                   NULL,
3632                                                   "no IP protocol in pattern");
3633         }
3634         return ret;
3635 }
3636
3637 /**
3638  * Validate jump action.
3639  *
3640  * @param[in] action
3641  *   Pointer to the jump action.
3642  * @param[in] action_flags
3643  *   Holds the actions detected until now.
3644  * @param[in] attributes
3645  *   Pointer to flow attributes
3646  * @param[in] external
3647  *   Action belongs to flow rule created by request external to PMD.
3648  * @param[out] error
3649  *   Pointer to error structure.
3650  *
3651  * @return
3652  *   0 on success, a negative errno value otherwise and rte_errno is set.
3653  */
3654 static int
3655 flow_dv_validate_action_jump(const struct rte_flow_action *action,
3656                              uint64_t action_flags,
3657                              const struct rte_flow_attr *attributes,
3658                              bool external, struct rte_flow_error *error)
3659 {
3660         uint32_t target_group, table;
3661         int ret = 0;
3662
3663         if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3664                             MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3665                 return rte_flow_error_set(error, EINVAL,
3666                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3667                                           "can't have 2 fate actions in"
3668                                           " same flow");
3669         if (action_flags & MLX5_FLOW_ACTION_METER)
3670                 return rte_flow_error_set(error, ENOTSUP,
3671                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3672                                           "jump with meter not support");
3673         if (!action->conf)
3674                 return rte_flow_error_set(error, EINVAL,
3675                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3676                                           NULL, "action configuration not set");
3677         target_group =
3678                 ((const struct rte_flow_action_jump *)action->conf)->group;
3679         ret = mlx5_flow_group_to_table(attributes, external, target_group,
3680                                        true, &table, error);
3681         if (ret)
3682                 return ret;
3683         if (attributes->group == target_group)
3684                 return rte_flow_error_set(error, EINVAL,
3685                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3686                                           "target group must be other than"
3687                                           " the current flow group");
3688         return 0;
3689 }
3690
3691 /*
3692  * Validate the port_id action.
3693  *
3694  * @param[in] dev
3695  *   Pointer to rte_eth_dev structure.
3696  * @param[in] action_flags
3697  *   Bit-fields that holds the actions detected until now.
3698  * @param[in] action
3699  *   Port_id RTE action structure.
3700  * @param[in] attr
3701  *   Attributes of flow that includes this action.
3702  * @param[out] error
3703  *   Pointer to error structure.
3704  *
3705  * @return
3706  *   0 on success, a negative errno value otherwise and rte_errno is set.
3707  */
3708 static int
3709 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
3710                                 uint64_t action_flags,
3711                                 const struct rte_flow_action *action,
3712                                 const struct rte_flow_attr *attr,
3713                                 struct rte_flow_error *error)
3714 {
3715         const struct rte_flow_action_port_id *port_id;
3716         struct mlx5_priv *act_priv;
3717         struct mlx5_priv *dev_priv;
3718         uint16_t port;
3719
3720         if (!attr->transfer)
3721                 return rte_flow_error_set(error, ENOTSUP,
3722                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3723                                           NULL,
3724                                           "port id action is valid in transfer"
3725                                           " mode only");
3726         if (!action || !action->conf)
3727                 return rte_flow_error_set(error, ENOTSUP,
3728                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3729                                           NULL,
3730                                           "port id action parameters must be"
3731                                           " specified");
3732         if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3733                             MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3734                 return rte_flow_error_set(error, EINVAL,
3735                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3736                                           "can have only one fate actions in"
3737                                           " a flow");
3738         dev_priv = mlx5_dev_to_eswitch_info(dev);
3739         if (!dev_priv)
3740                 return rte_flow_error_set(error, rte_errno,
3741                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3742                                           NULL,
3743                                           "failed to obtain E-Switch info");
3744         port_id = action->conf;
3745         port = port_id->original ? dev->data->port_id : port_id->id;
3746         act_priv = mlx5_port_to_eswitch_info(port, false);
3747         if (!act_priv)
3748                 return rte_flow_error_set
3749                                 (error, rte_errno,
3750                                  RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
3751                                  "failed to obtain E-Switch port id for port");
3752         if (act_priv->domain_id != dev_priv->domain_id)
3753                 return rte_flow_error_set
3754                                 (error, EINVAL,
3755                                  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3756                                  "port does not belong to"
3757                                  " E-Switch being configured");
3758         return 0;
3759 }
3760
3761 /**
3762  * Get the maximum number of modify header actions.
3763  *
3764  * @param dev
3765  *   Pointer to rte_eth_dev structure.
3766  * @param flags
3767  *   Flags bits to check if root level.
3768  *
3769  * @return
3770  *   Max number of modify header actions device can support.
3771  */
3772 static inline unsigned int
3773 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
3774                               uint64_t flags)
3775 {
3776         /*
3777          * There's no way to directly query the max capacity from FW.
3778          * The maximal value on root table should be assumed to be supported.
3779          */
3780         if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
3781                 return MLX5_MAX_MODIFY_NUM;
3782         else
3783                 return MLX5_ROOT_TBL_MODIFY_NUM;
3784 }
3785
3786 /**
3787  * Validate the meter action.
3788  *
3789  * @param[in] dev
3790  *   Pointer to rte_eth_dev structure.
3791  * @param[in] action_flags
3792  *   Bit-fields that holds the actions detected until now.
3793  * @param[in] action
3794  *   Pointer to the meter action.
3795  * @param[in] attr
3796  *   Attributes of flow that includes this action.
3797  * @param[out] error
3798  *   Pointer to error structure.
3799  *
3800  * @return
3801  *   0 on success, a negative errno value otherwise and rte_ernno is set.
3802  */
3803 static int
3804 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
3805                                 uint64_t action_flags,
3806                                 const struct rte_flow_action *action,
3807                                 const struct rte_flow_attr *attr,
3808                                 struct rte_flow_error *error)
3809 {
3810         struct mlx5_priv *priv = dev->data->dev_private;
3811         const struct rte_flow_action_meter *am = action->conf;
3812         struct mlx5_flow_meter *fm;
3813
3814         if (!am)
3815                 return rte_flow_error_set(error, EINVAL,
3816                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3817                                           "meter action conf is NULL");
3818
3819         if (action_flags & MLX5_FLOW_ACTION_METER)
3820                 return rte_flow_error_set(error, ENOTSUP,
3821                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3822                                           "meter chaining not support");
3823         if (action_flags & MLX5_FLOW_ACTION_JUMP)
3824                 return rte_flow_error_set(error, ENOTSUP,
3825                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3826                                           "meter with jump not support");
3827         if (!priv->mtr_en)
3828                 return rte_flow_error_set(error, ENOTSUP,
3829                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3830                                           NULL,
3831                                           "meter action not supported");
3832         fm = mlx5_flow_meter_find(priv, am->mtr_id);
3833         if (!fm)
3834                 return rte_flow_error_set(error, EINVAL,
3835                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3836                                           "Meter not found");
3837         if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
3838               (!fm->ingress && !attr->ingress && attr->egress) ||
3839               (!fm->egress && !attr->egress && attr->ingress))))
3840                 return rte_flow_error_set(error, EINVAL,
3841                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3842                                           "Flow attributes are either invalid "
3843                                           "or have a conflict with current "
3844                                           "meter attributes");
3845         return 0;
3846 }
3847
3848 /**
3849  * Validate the age action.
3850  *
3851  * @param[in] action_flags
3852  *   Holds the actions detected until now.
3853  * @param[in] action
3854  *   Pointer to the age action.
3855  * @param[in] dev
3856  *   Pointer to the Ethernet device structure.
3857  * @param[out] error
3858  *   Pointer to error structure.
3859  *
3860  * @return
3861  *   0 on success, a negative errno value otherwise and rte_errno is set.
3862  */
3863 static int
3864 flow_dv_validate_action_age(uint64_t action_flags,
3865                             const struct rte_flow_action *action,
3866                             struct rte_eth_dev *dev,
3867                             struct rte_flow_error *error)
3868 {
3869         struct mlx5_priv *priv = dev->data->dev_private;
3870         const struct rte_flow_action_age *age = action->conf;
3871
3872         if (!priv->config.devx || priv->counter_fallback)
3873                 return rte_flow_error_set(error, ENOTSUP,
3874                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3875                                           NULL,
3876                                           "age action not supported");
3877         if (!(action->conf))
3878                 return rte_flow_error_set(error, EINVAL,
3879                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
3880                                           "configuration cannot be null");
3881         if (age->timeout >= UINT16_MAX / 2 / 10)
3882                 return rte_flow_error_set(error, ENOTSUP,
3883                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
3884                                           "Max age time: 3275 seconds");
3885         if (action_flags & MLX5_FLOW_ACTION_AGE)
3886                 return rte_flow_error_set(error, EINVAL,
3887                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3888                                           "Duplicate age ctions set");
3889         return 0;
3890 }
3891
3892 /**
3893  * Validate the modify-header IPv4 DSCP actions.
3894  *
3895  * @param[in] action_flags
3896  *   Holds the actions detected until now.
3897  * @param[in] action
3898  *   Pointer to the modify action.
3899  * @param[in] item_flags
3900  *   Holds the items detected.
3901  * @param[out] error
3902  *   Pointer to error structure.
3903  *
3904  * @return
3905  *   0 on success, a negative errno value otherwise and rte_errno is set.
3906  */
3907 static int
3908 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
3909                                          const struct rte_flow_action *action,
3910                                          const uint64_t item_flags,
3911                                          struct rte_flow_error *error)
3912 {
3913         int ret = 0;
3914
3915         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3916         if (!ret) {
3917                 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
3918                         return rte_flow_error_set(error, EINVAL,
3919                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3920                                                   NULL,
3921                                                   "no ipv4 item in pattern");
3922         }
3923         return ret;
3924 }
3925
3926 /**
3927  * Validate the modify-header IPv6 DSCP actions.
3928  *
3929  * @param[in] action_flags
3930  *   Holds the actions detected until now.
3931  * @param[in] action
3932  *   Pointer to the modify action.
3933  * @param[in] item_flags
3934  *   Holds the items detected.
3935  * @param[out] error
3936  *   Pointer to error structure.
3937  *
3938  * @return
3939  *   0 on success, a negative errno value otherwise and rte_errno is set.
3940  */
3941 static int
3942 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
3943                                          const struct rte_flow_action *action,
3944                                          const uint64_t item_flags,
3945                                          struct rte_flow_error *error)
3946 {
3947         int ret = 0;
3948
3949         ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3950         if (!ret) {
3951                 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
3952                         return rte_flow_error_set(error, EINVAL,
3953                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3954                                                   NULL,
3955                                                   "no ipv6 item in pattern");
3956         }
3957         return ret;
3958 }
3959
3960 /**
3961  * Match modify-header resource.
3962  *
3963  * @param entry
3964  *   Pointer to exist resource entry object.
3965  * @param ctx
3966  *   Pointer to new modify-header resource.
3967  *
3968  * @return
3969  *   0 on matching, -1 otherwise.
3970  */
3971 static int
3972 flow_dv_modify_hdr_resource_match(struct mlx5_hlist_entry *entry, void *ctx)
3973 {
3974         struct mlx5_flow_dv_modify_hdr_resource *resource;
3975         struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
3976         uint32_t actions_len;
3977
3978         resource = (struct mlx5_flow_dv_modify_hdr_resource *)ctx;
3979         cache_resource = container_of(entry,
3980                                       struct mlx5_flow_dv_modify_hdr_resource,
3981                                       entry);
3982         actions_len = resource->actions_num * sizeof(resource->actions[0]);
3983         if (resource->entry.key == cache_resource->entry.key &&
3984             resource->ft_type == cache_resource->ft_type &&
3985             resource->actions_num == cache_resource->actions_num &&
3986             resource->flags == cache_resource->flags &&
3987             !memcmp((const void *)resource->actions,
3988                     (const void *)cache_resource->actions,
3989                     actions_len))
3990                 return 0;
3991         return -1;
3992 }
3993
3994 /**
3995  * Find existing modify-header resource or create and register a new one.
3996  *
3997  * @param dev[in, out]
3998  *   Pointer to rte_eth_dev structure.
3999  * @param[in, out] resource
4000  *   Pointer to modify-header resource.
4001  * @parm[in, out] dev_flow
4002  *   Pointer to the dev_flow.
4003  * @param[out] error
4004  *   pointer to error structure.
4005  *
4006  * @return
4007  *   0 on success otherwise -errno and errno is set.
4008  */
4009 static int
4010 flow_dv_modify_hdr_resource_register
4011                         (struct rte_eth_dev *dev,
4012                          struct mlx5_flow_dv_modify_hdr_resource *resource,
4013                          struct mlx5_flow *dev_flow,
4014                          struct rte_flow_error *error)
4015 {
4016         struct mlx5_priv *priv = dev->data->dev_private;
4017         struct mlx5_dev_ctx_shared *sh = priv->sh;
4018         struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
4019         struct mlx5dv_dr_domain *ns;
4020         uint32_t actions_len;
4021         struct mlx5_hlist_entry *entry;
4022         union mlx5_flow_modify_hdr_key hdr_mod_key = {
4023                 {
4024                         .ft_type = resource->ft_type,
4025                         .actions_num = resource->actions_num,
4026                         .group = dev_flow->dv.group,
4027                         .cksum = 0,
4028                 }
4029         };
4030         int ret;
4031
4032         resource->flags = dev_flow->dv.group ? 0 :
4033                           MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4034         if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
4035                                     resource->flags))
4036                 return rte_flow_error_set(error, EOVERFLOW,
4037                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4038                                           "too many modify header items");
4039         if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
4040                 ns = sh->fdb_domain;
4041         else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
4042                 ns = sh->tx_domain;
4043         else
4044                 ns = sh->rx_domain;
4045         /* Lookup a matching resource from cache. */
4046         actions_len = resource->actions_num * sizeof(resource->actions[0]);
4047         hdr_mod_key.cksum = __rte_raw_cksum(resource->actions, actions_len, 0);
4048         resource->entry.key = hdr_mod_key.v64;
4049         entry = mlx5_hlist_lookup_ex(sh->modify_cmds, resource->entry.key,
4050                                      flow_dv_modify_hdr_resource_match,
4051                                      (void *)resource);
4052         if (entry) {
4053                 cache_resource = container_of(entry,
4054                                         struct mlx5_flow_dv_modify_hdr_resource,
4055                                         entry);
4056                 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
4057                         (void *)cache_resource,
4058                         rte_atomic32_read(&cache_resource->refcnt));
4059                 rte_atomic32_inc(&cache_resource->refcnt);
4060                 dev_flow->handle->dvh.modify_hdr = cache_resource;
4061                 return 0;
4062
4063         }
4064         /* Register new modify-header resource. */
4065         cache_resource = mlx5_malloc(MLX5_MEM_ZERO,
4066                                     sizeof(*cache_resource) + actions_len, 0,
4067                                     SOCKET_ID_ANY);
4068         if (!cache_resource)
4069                 return rte_flow_error_set(error, ENOMEM,
4070                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4071                                           "cannot allocate resource memory");
4072         *cache_resource = *resource;
4073         rte_memcpy(cache_resource->actions, resource->actions, actions_len);
4074         ret = mlx5_flow_os_create_flow_action_modify_header
4075                                         (sh->ctx, ns, cache_resource,
4076                                          actions_len, &cache_resource->action);
4077         if (ret) {
4078                 mlx5_free(cache_resource);
4079                 return rte_flow_error_set(error, ENOMEM,
4080                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4081                                           NULL, "cannot create action");
4082         }
4083         rte_atomic32_init(&cache_resource->refcnt);
4084         rte_atomic32_inc(&cache_resource->refcnt);
4085         if (mlx5_hlist_insert_ex(sh->modify_cmds, &cache_resource->entry,
4086                                  flow_dv_modify_hdr_resource_match,
4087                                  (void *)cache_resource)) {
4088                 claim_zero(mlx5_flow_os_destroy_flow_action
4089                                                 (cache_resource->action));
4090                 mlx5_free(cache_resource);
4091                 return rte_flow_error_set(error, EEXIST,
4092                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4093                                           NULL, "action exist");
4094         }
4095         dev_flow->handle->dvh.modify_hdr = cache_resource;
4096         DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
4097                 (void *)cache_resource,
4098                 rte_atomic32_read(&cache_resource->refcnt));
4099         return 0;
4100 }
4101
4102 /**
4103  * Get DV flow counter by index.
4104  *
4105  * @param[in] dev
4106  *   Pointer to the Ethernet device structure.
4107  * @param[in] idx
4108  *   mlx5 flow counter index in the container.
4109  * @param[out] ppool
4110  *   mlx5 flow counter pool in the container,
4111  *
4112  * @return
4113  *   Pointer to the counter, NULL otherwise.
4114  */
4115 static struct mlx5_flow_counter *
4116 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
4117                            uint32_t idx,
4118                            struct mlx5_flow_counter_pool **ppool)
4119 {
4120         struct mlx5_priv *priv = dev->data->dev_private;
4121         struct mlx5_pools_container *cont;
4122         struct mlx5_flow_counter_pool *pool;
4123         uint32_t batch = 0, age = 0;
4124
4125         idx--;
4126         age = MLX_CNT_IS_AGE(idx);
4127         idx = age ? idx - MLX5_CNT_AGE_OFFSET : idx;
4128         if (idx >= MLX5_CNT_BATCH_OFFSET) {
4129                 idx -= MLX5_CNT_BATCH_OFFSET;
4130                 batch = 1;
4131         }
4132         cont = MLX5_CNT_CONTAINER(priv->sh, batch, age);
4133         MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cont->n);
4134         pool = cont->pools[idx / MLX5_COUNTERS_PER_POOL];
4135         MLX5_ASSERT(pool);
4136         if (ppool)
4137                 *ppool = pool;
4138         return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
4139 }
4140
4141 /**
4142  * Check the devx counter belongs to the pool.
4143  *
4144  * @param[in] pool
4145  *   Pointer to the counter pool.
4146  * @param[in] id
4147  *   The counter devx ID.
4148  *
4149  * @return
4150  *   True if counter belongs to the pool, false otherwise.
4151  */
4152 static bool
4153 flow_dv_is_counter_in_pool(struct mlx5_flow_counter_pool *pool, int id)
4154 {
4155         int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
4156                    MLX5_COUNTERS_PER_POOL;
4157
4158         if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
4159                 return true;
4160         return false;
4161 }
4162
4163 /**
4164  * Get a pool by devx counter ID.
4165  *
4166  * @param[in] cont
4167  *   Pointer to the counter container.
4168  * @param[in] id
4169  *   The counter devx ID.
4170  *
4171  * @return
4172  *   The counter pool pointer if exists, NULL otherwise,
4173  */
4174 static struct mlx5_flow_counter_pool *
4175 flow_dv_find_pool_by_id(struct mlx5_pools_container *cont, int id)
4176 {
4177         uint32_t i;
4178
4179         /* Check last used pool. */
4180         if (cont->last_pool_idx != POOL_IDX_INVALID &&
4181             flow_dv_is_counter_in_pool(cont->pools[cont->last_pool_idx], id))
4182                 return cont->pools[cont->last_pool_idx];
4183         /* ID out of range means no suitable pool in the container. */
4184         if (id > cont->max_id || id < cont->min_id)
4185                 return NULL;
4186         /*
4187          * Find the pool from the end of the container, since mostly counter
4188          * ID is sequence increasing, and the last pool should be the needed
4189          * one.
4190          */
4191         i = rte_atomic16_read(&cont->n_valid);
4192         while (i--) {
4193                 struct mlx5_flow_counter_pool *pool = cont->pools[i];
4194
4195                 if (flow_dv_is_counter_in_pool(pool, id))
4196                         return pool;
4197         }
4198         return NULL;
4199 }
4200
4201 /**
4202  * Allocate a new memory for the counter values wrapped by all the needed
4203  * management.
4204  *
4205  * @param[in] dev
4206  *   Pointer to the Ethernet device structure.
4207  * @param[in] raws_n
4208  *   The raw memory areas - each one for MLX5_COUNTERS_PER_POOL counters.
4209  *
4210  * @return
4211  *   The new memory management pointer on success, otherwise NULL and rte_errno
4212  *   is set.
4213  */
4214 static struct mlx5_counter_stats_mem_mng *
4215 flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
4216 {
4217         struct mlx5_priv *priv = dev->data->dev_private;
4218         struct mlx5_dev_ctx_shared *sh = priv->sh;
4219         struct mlx5_devx_mkey_attr mkey_attr;
4220         struct mlx5_counter_stats_mem_mng *mem_mng;
4221         volatile struct flow_counter_stats *raw_data;
4222         int size = (sizeof(struct flow_counter_stats) *
4223                         MLX5_COUNTERS_PER_POOL +
4224                         sizeof(struct mlx5_counter_stats_raw)) * raws_n +
4225                         sizeof(struct mlx5_counter_stats_mem_mng);
4226         size_t pgsize = rte_mem_page_size();
4227         if (pgsize == (size_t)-1) {
4228                 DRV_LOG(ERR, "Failed to get mem page size");
4229                 rte_errno = ENOMEM;
4230                 return NULL;
4231         }
4232         uint8_t *mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize,
4233                                   SOCKET_ID_ANY);
4234         int i;
4235
4236         if (!mem) {
4237                 rte_errno = ENOMEM;
4238                 return NULL;
4239         }
4240         mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
4241         size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
4242         mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
4243                                                  IBV_ACCESS_LOCAL_WRITE);
4244         if (!mem_mng->umem) {
4245                 rte_errno = errno;
4246                 mlx5_free(mem);
4247                 return NULL;
4248         }
4249         mkey_attr.addr = (uintptr_t)mem;
4250         mkey_attr.size = size;
4251         mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
4252         mkey_attr.pd = sh->pdn;
4253         mkey_attr.log_entity_size = 0;
4254         mkey_attr.pg_access = 0;
4255         mkey_attr.klm_array = NULL;
4256         mkey_attr.klm_num = 0;
4257         if (priv->config.hca_attr.relaxed_ordering_write &&
4258                 priv->config.hca_attr.relaxed_ordering_read  &&
4259                 !haswell_broadwell_cpu)
4260                 mkey_attr.relaxed_ordering = 1;
4261         mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
4262         if (!mem_mng->dm) {
4263                 mlx5_glue->devx_umem_dereg(mem_mng->umem);
4264                 rte_errno = errno;
4265                 mlx5_free(mem);
4266                 return NULL;
4267         }
4268         mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
4269         raw_data = (volatile struct flow_counter_stats *)mem;
4270         for (i = 0; i < raws_n; ++i) {
4271                 mem_mng->raws[i].mem_mng = mem_mng;
4272                 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
4273         }
4274         LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
4275         return mem_mng;
4276 }
4277
4278 /**
4279  * Resize a counter container.
4280  *
4281  * @param[in] dev
4282  *   Pointer to the Ethernet device structure.
4283  * @param[in] batch
4284  *   Whether the pool is for counter that was allocated by batch command.
4285  * @param[in] age
4286  *   Whether the pool is for Aging counter.
4287  *
4288  * @return
4289  *   0 on success, otherwise negative errno value and rte_errno is set.
4290  */
4291 static int
4292 flow_dv_container_resize(struct rte_eth_dev *dev,
4293                                 uint32_t batch, uint32_t age)
4294 {
4295         struct mlx5_priv *priv = dev->data->dev_private;
4296         struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4297                                                                age);
4298         struct mlx5_counter_stats_mem_mng *mem_mng = NULL;
4299         void *old_pools = cont->pools;
4300         uint32_t resize = cont->n + MLX5_CNT_CONTAINER_RESIZE;
4301         uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4302         void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
4303
4304         if (!pools) {
4305                 rte_errno = ENOMEM;
4306                 return -ENOMEM;
4307         }
4308         if (old_pools)
4309                 memcpy(pools, old_pools, cont->n *
4310                                        sizeof(struct mlx5_flow_counter_pool *));
4311         /*
4312          * Fallback mode query the counter directly, no background query
4313          * resources are needed.
4314          */
4315         if (!priv->counter_fallback) {
4316                 int i;
4317
4318                 mem_mng = flow_dv_create_counter_stat_mem_mng(dev,
4319                           MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES);
4320                 if (!mem_mng) {
4321                         mlx5_free(pools);
4322                         return -ENOMEM;
4323                 }
4324                 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
4325                         LIST_INSERT_HEAD(&priv->sh->cmng.free_stat_raws,
4326                                          mem_mng->raws +
4327                                          MLX5_CNT_CONTAINER_RESIZE +
4328                                          i, next);
4329         }
4330         rte_spinlock_lock(&cont->resize_sl);
4331         cont->n = resize;
4332         cont->mem_mng = mem_mng;
4333         cont->pools = pools;
4334         rte_spinlock_unlock(&cont->resize_sl);
4335         if (old_pools)
4336                 mlx5_free(old_pools);
4337         return 0;
4338 }
4339
4340 /**
4341  * Query a devx flow counter.
4342  *
4343  * @param[in] dev
4344  *   Pointer to the Ethernet device structure.
4345  * @param[in] cnt
4346  *   Index to the flow counter.
4347  * @param[out] pkts
4348  *   The statistics value of packets.
4349  * @param[out] bytes
4350  *   The statistics value of bytes.
4351  *
4352  * @return
4353  *   0 on success, otherwise a negative errno value and rte_errno is set.
4354  */
4355 static inline int
4356 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4357                      uint64_t *bytes)
4358 {
4359         struct mlx5_priv *priv = dev->data->dev_private;
4360         struct mlx5_flow_counter_pool *pool = NULL;
4361         struct mlx5_flow_counter *cnt;
4362         struct mlx5_flow_counter_ext *cnt_ext = NULL;
4363         int offset;
4364
4365         cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4366         MLX5_ASSERT(pool);
4367         if (counter < MLX5_CNT_BATCH_OFFSET) {
4368                 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4369                 if (priv->counter_fallback)
4370                         return mlx5_devx_cmd_flow_counter_query(cnt_ext->dcs, 0,
4371                                         0, pkts, bytes, 0, NULL, NULL, 0);
4372         }
4373
4374         rte_spinlock_lock(&pool->sl);
4375         /*
4376          * The single counters allocation may allocate smaller ID than the
4377          * current allocated in parallel to the host reading.
4378          * In this case the new counter values must be reported as 0.
4379          */
4380         if (unlikely(cnt_ext && cnt_ext->dcs->id < pool->raw->min_dcs_id)) {
4381                 *pkts = 0;
4382                 *bytes = 0;
4383         } else {
4384                 offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
4385                 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4386                 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4387         }
4388         rte_spinlock_unlock(&pool->sl);
4389         return 0;
4390 }
4391
4392 /**
4393  * Create and initialize a new counter pool.
4394  *
4395  * @param[in] dev
4396  *   Pointer to the Ethernet device structure.
4397  * @param[out] dcs
4398  *   The devX counter handle.
4399  * @param[in] batch
4400  *   Whether the pool is for counter that was allocated by batch command.
4401  * @param[in] age
4402  *   Whether the pool is for counter that was allocated for aging.
4403  * @param[in/out] cont_cur
4404  *   Pointer to the container pointer, it will be update in pool resize.
4405  *
4406  * @return
4407  *   The pool container pointer on success, NULL otherwise and rte_errno is set.
4408  */
4409 static struct mlx5_flow_counter_pool *
4410 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4411                     uint32_t batch, uint32_t age)
4412 {
4413         struct mlx5_priv *priv = dev->data->dev_private;
4414         struct mlx5_flow_counter_pool *pool;
4415         struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4416                                                                age);
4417         int16_t n_valid = rte_atomic16_read(&cont->n_valid);
4418         uint32_t size = sizeof(*pool);
4419
4420         if (cont->n == n_valid && flow_dv_container_resize(dev, batch, age))
4421                 return NULL;
4422         size += MLX5_COUNTERS_PER_POOL * CNT_SIZE;
4423         size += (batch ? 0 : MLX5_COUNTERS_PER_POOL * CNTEXT_SIZE);
4424         size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * AGE_SIZE);
4425         pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
4426         if (!pool) {
4427                 rte_errno = ENOMEM;
4428                 return NULL;
4429         }
4430         pool->min_dcs = dcs;
4431         if (!priv->counter_fallback)
4432                 pool->raw = cont->mem_mng->raws + n_valid %
4433                                                       MLX5_CNT_CONTAINER_RESIZE;
4434         pool->raw_hw = NULL;
4435         pool->type = 0;
4436         pool->type |= (batch ? 0 :  CNT_POOL_TYPE_EXT);
4437         pool->type |= (!age ? 0 :  CNT_POOL_TYPE_AGE);
4438         pool->query_gen = 0;
4439         rte_spinlock_init(&pool->sl);
4440         TAILQ_INIT(&pool->counters[0]);
4441         TAILQ_INIT(&pool->counters[1]);
4442         TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
4443         pool->index = n_valid;
4444         cont->pools[n_valid] = pool;
4445         if (!batch) {
4446                 int base = RTE_ALIGN_FLOOR(dcs->id, MLX5_COUNTERS_PER_POOL);
4447
4448                 if (base < cont->min_id)
4449                         cont->min_id = base;
4450                 if (base > cont->max_id)
4451                         cont->max_id = base + MLX5_COUNTERS_PER_POOL - 1;
4452                 cont->last_pool_idx = pool->index;
4453         }
4454         /* Pool initialization must be updated before host thread access. */
4455         rte_cio_wmb();
4456         rte_atomic16_add(&cont->n_valid, 1);
4457         return pool;
4458 }
4459
4460 /**
4461  * Restore skipped counters in the pool.
4462  *
4463  * As counter pool query requires the first counter dcs
4464  * ID start with 4 alinged, if the pool counters with
4465  * min_dcs ID are not aligned with 4, the counters will
4466  * be skipped.
4467  * Once other min_dcs ID less than these skipped counter
4468  * dcs ID appears, the skipped counters will be safe to
4469  * use.
4470  * Should be called when min_dcs is updated.
4471  *
4472  * @param[in] pool
4473  *   Current counter pool.
4474  * @param[in] last_min_dcs
4475  *   Last min_dcs.
4476  */
4477 static void
4478 flow_dv_counter_restore(struct mlx5_flow_counter_pool *pool,
4479                         struct mlx5_devx_obj *last_min_dcs)
4480 {
4481         struct mlx5_flow_counter_ext *cnt_ext;
4482         uint32_t offset, new_offset;
4483         uint32_t skip_cnt = 0;
4484         uint32_t i;
4485
4486         if (!pool->skip_cnt)
4487                 return;
4488         /*
4489          * If last min_dcs is not valid. The skipped counter may even after
4490          * last min_dcs, set the offset to the whole pool.
4491          */
4492         if (last_min_dcs->id & (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1))
4493                 offset = MLX5_COUNTERS_PER_POOL;
4494         else
4495                 offset = last_min_dcs->id % MLX5_COUNTERS_PER_POOL;
4496         new_offset = pool->min_dcs->id % MLX5_COUNTERS_PER_POOL;
4497         /*
4498          * Check the counters from 1 to the last_min_dcs range. Counters
4499          * before new min_dcs indicates pool still has skipped counters.
4500          * Counters be skipped after new min_dcs will be ready to use.
4501          * Offset 0 counter must be empty or min_dcs, start from 1.
4502          */
4503         for (i = 1; i < offset; i++) {
4504                 cnt_ext = MLX5_GET_POOL_CNT_EXT(pool, i);
4505                 if (cnt_ext->skipped) {
4506                         if (i > new_offset) {
4507                                 cnt_ext->skipped = 0;
4508                                 TAILQ_INSERT_TAIL
4509                                         (&pool->counters[pool->query_gen],
4510                                          MLX5_POOL_GET_CNT(pool, i), next);
4511                         } else {
4512                                 skip_cnt++;
4513                         }
4514                 }
4515         }
4516         if (!skip_cnt)
4517                 pool->skip_cnt = 0;
4518 }
4519
4520 /**
4521  * Prepare a new counter and/or a new counter pool.
4522  *
4523  * @param[in] dev
4524  *   Pointer to the Ethernet device structure.
4525  * @param[out] cnt_free
4526  *   Where to put the pointer of a new counter.
4527  * @param[in] batch
4528  *   Whether the pool is for counter that was allocated by batch command.
4529  * @param[in] age
4530  *   Whether the pool is for counter that was allocated for aging.
4531  *
4532  * @return
4533  *   The counter pool pointer and @p cnt_free is set on success,
4534  *   NULL otherwise and rte_errno is set.
4535  */
4536 static struct mlx5_flow_counter_pool *
4537 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4538                              struct mlx5_flow_counter **cnt_free,
4539                              uint32_t batch, uint32_t age)
4540 {
4541         struct mlx5_priv *priv = dev->data->dev_private;
4542         struct mlx5_pools_container *cont;
4543         struct mlx5_flow_counter_pool *pool;
4544         struct mlx5_counters tmp_tq;
4545         struct mlx5_devx_obj *last_min_dcs;
4546         struct mlx5_devx_obj *dcs = NULL;
4547         struct mlx5_flow_counter *cnt;
4548         uint32_t add2other;
4549         uint32_t i;
4550
4551         cont = MLX5_CNT_CONTAINER(priv->sh, batch, age);
4552         if (!batch) {
4553 retry:
4554                 add2other = 0;
4555                 /* bulk_bitmap must be 0 for single counter allocation. */
4556                 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4557                 if (!dcs)
4558                         return NULL;
4559                 pool = flow_dv_find_pool_by_id(cont, dcs->id);
4560                 /* Check if counter belongs to exist pool ID range. */
4561                 if (!pool) {
4562                         pool = flow_dv_find_pool_by_id
4563                                (MLX5_CNT_CONTAINER
4564                                (priv->sh, batch, (age ^ 0x1)), dcs->id);
4565                         /*
4566                          * Pool eixsts, counter will be added to the other
4567                          * container, need to reallocate it later.
4568                          */
4569                         if (pool) {
4570                                 add2other = 1;
4571                         } else {
4572                                 pool = flow_dv_pool_create(dev, dcs, batch,
4573                                                            age);
4574                                 if (!pool) {
4575                                         mlx5_devx_cmd_destroy(dcs);
4576                                         return NULL;
4577                                 }
4578                         }
4579                 }
4580                 if ((dcs->id < pool->min_dcs->id ||
4581                     pool->min_dcs->id &
4582                     (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1)) &&
4583                     !(dcs->id & (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1))) {
4584                         /*
4585                          * Update the pool min_dcs only if current dcs is
4586                          * valid and exist min_dcs is not valid or greater
4587                          * than new dcs.
4588                          */
4589                         last_min_dcs = pool->min_dcs;
4590                         rte_atomic64_set(&pool->a64_dcs,
4591                                          (int64_t)(uintptr_t)dcs);
4592                         /*
4593                          * Restore any skipped counters if the new min_dcs
4594                          * ID is smaller or min_dcs is not valid.
4595                          */
4596                         if (dcs->id < last_min_dcs->id ||
4597                             last_min_dcs->id &
4598                             (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1))
4599                                 flow_dv_counter_restore(pool, last_min_dcs);
4600                 }
4601                 i = dcs->id % MLX5_COUNTERS_PER_POOL;
4602                 cnt = MLX5_POOL_GET_CNT(pool, i);
4603                 cnt->pool = pool;
4604                 MLX5_GET_POOL_CNT_EXT(pool, i)->dcs = dcs;
4605                 /*
4606                  * If min_dcs is not valid, it means the new allocated dcs
4607                  * also fail to become the valid min_dcs, just skip it.
4608                  * Or if min_dcs is valid, and new dcs ID is smaller than
4609                  * min_dcs, but not become the min_dcs, also skip it.
4610                  */
4611                 if (pool->min_dcs->id &
4612                     (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1) ||
4613                     dcs->id < pool->min_dcs->id) {
4614                         MLX5_GET_POOL_CNT_EXT(pool, i)->skipped = 1;
4615                         pool->skip_cnt = 1;
4616                         goto retry;
4617                 }
4618                 if (add2other) {
4619                         TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen],
4620                                           cnt, next);
4621                         goto retry;
4622                 }
4623                 *cnt_free = cnt;
4624                 return pool;
4625         }
4626         /* bulk_bitmap is in 128 counters units. */
4627         if (priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4)
4628                 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4629         if (!dcs) {
4630                 rte_errno = ENODATA;
4631                 return NULL;
4632         }
4633         pool = flow_dv_pool_create(dev, dcs, batch, age);
4634         if (!pool) {
4635                 mlx5_devx_cmd_destroy(dcs);
4636                 return NULL;
4637         }
4638         TAILQ_INIT(&tmp_tq);
4639         for (i = 1; i < MLX5_COUNTERS_PER_POOL; ++i) {
4640                 cnt = MLX5_POOL_GET_CNT(pool, i);
4641                 cnt->pool = pool;
4642                 TAILQ_INSERT_HEAD(&tmp_tq, cnt, next);
4643         }
4644         rte_spinlock_lock(&cont->csl);
4645         TAILQ_CONCAT(&cont->counters, &tmp_tq, next);
4646         rte_spinlock_unlock(&cont->csl);
4647         *cnt_free = MLX5_POOL_GET_CNT(pool, 0);
4648         (*cnt_free)->pool = pool;
4649         return pool;
4650 }
4651
4652 /**
4653  * Search for existed shared counter.
4654  *
4655  * @param[in] dev
4656  *   Pointer to the Ethernet device structure.
4657  * @param[in] id
4658  *   The shared counter ID to search.
4659  * @param[out] ppool
4660  *   mlx5 flow counter pool in the container,
4661  *
4662  * @return
4663  *   NULL if not existed, otherwise pointer to the shared extend counter.
4664  */
4665 static struct mlx5_flow_counter_ext *
4666 flow_dv_counter_shared_search(struct rte_eth_dev *dev, uint32_t id,
4667                               struct mlx5_flow_counter_pool **ppool)
4668 {
4669         struct mlx5_priv *priv = dev->data->dev_private;
4670         union mlx5_l3t_data data;
4671         uint32_t cnt_idx;
4672
4673         if (mlx5_l3t_get_entry(priv->sh->cnt_id_tbl, id, &data) || !data.dword)
4674                 return NULL;
4675         cnt_idx = data.dword;
4676         /*
4677          * Shared counters don't have age info. The counter extend is after
4678          * the counter datat structure.
4679          */
4680         return (struct mlx5_flow_counter_ext *)
4681                ((flow_dv_counter_get_by_idx(dev, cnt_idx, ppool)) + 1);
4682 }
4683
4684 /**
4685  * Allocate a flow counter.
4686  *
4687  * @param[in] dev
4688  *   Pointer to the Ethernet device structure.
4689  * @param[in] shared
4690  *   Indicate if this counter is shared with other flows.
4691  * @param[in] id
4692  *   Counter identifier.
4693  * @param[in] group
4694  *   Counter flow group.
4695  * @param[in] age
4696  *   Whether the counter was allocated for aging.
4697  *
4698  * @return
4699  *   Index to flow counter on success, 0 otherwise and rte_errno is set.
4700  */
4701 static uint32_t
4702 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t shared, uint32_t id,
4703                       uint16_t group, uint32_t age)
4704 {
4705         struct mlx5_priv *priv = dev->data->dev_private;
4706         struct mlx5_flow_counter_pool *pool = NULL;
4707         struct mlx5_flow_counter *cnt_free = NULL;
4708         struct mlx5_flow_counter_ext *cnt_ext = NULL;
4709         /*
4710          * Currently group 0 flow counter cannot be assigned to a flow if it is
4711          * not the first one in the batch counter allocation, so it is better
4712          * to allocate counters one by one for these flows in a separate
4713          * container.
4714          * A counter can be shared between different groups so need to take
4715          * shared counters from the single container.
4716          */
4717         uint32_t batch = (group && !shared && !priv->counter_fallback) ? 1 : 0;
4718         struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4719                                                                age);
4720         uint32_t cnt_idx;
4721
4722         if (!priv->config.devx) {
4723                 rte_errno = ENOTSUP;
4724                 return 0;
4725         }
4726         if (shared) {
4727                 cnt_ext = flow_dv_counter_shared_search(dev, id, &pool);
4728                 if (cnt_ext) {
4729                         if (cnt_ext->ref_cnt + 1 == 0) {
4730                                 rte_errno = E2BIG;
4731                                 return 0;
4732                         }
4733                         cnt_ext->ref_cnt++;
4734                         cnt_idx = pool->index * MLX5_COUNTERS_PER_POOL +
4735                                   (cnt_ext->dcs->id % MLX5_COUNTERS_PER_POOL)
4736                                   + 1;
4737                         return cnt_idx;
4738                 }
4739         }
4740         /* Get free counters from container. */
4741         rte_spinlock_lock(&cont->csl);
4742         cnt_free = TAILQ_FIRST(&cont->counters);
4743         if (cnt_free)
4744                 TAILQ_REMOVE(&cont->counters, cnt_free, next);
4745         rte_spinlock_unlock(&cont->csl);
4746         if (!cnt_free && !flow_dv_counter_pool_prepare(dev, &cnt_free,
4747                                                        batch, age))
4748                 goto err;
4749         pool = cnt_free->pool;
4750         if (!batch)
4751                 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt_free);
4752         /* Create a DV counter action only in the first time usage. */
4753         if (!cnt_free->action) {
4754                 uint16_t offset;
4755                 struct mlx5_devx_obj *dcs;
4756                 int ret;
4757
4758                 if (batch) {
4759                         offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
4760                         dcs = pool->min_dcs;
4761                 } else {
4762                         offset = 0;
4763                         dcs = cnt_ext->dcs;
4764                 }
4765                 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, offset,
4766                                                             &cnt_free->action);
4767                 if (ret) {
4768                         rte_errno = errno;
4769                         goto err;
4770                 }
4771         }
4772         cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
4773                                 MLX5_CNT_ARRAY_IDX(pool, cnt_free));
4774         cnt_idx += batch * MLX5_CNT_BATCH_OFFSET;
4775         cnt_idx += age * MLX5_CNT_AGE_OFFSET;
4776         /* Update the counter reset values. */
4777         if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
4778                                  &cnt_free->bytes))
4779                 goto err;
4780         if (cnt_ext) {
4781                 cnt_ext->shared = shared;
4782                 cnt_ext->ref_cnt = 1;
4783                 cnt_ext->id = id;
4784                 if (shared) {
4785                         union mlx5_l3t_data data;
4786
4787                         data.dword = cnt_idx;
4788                         if (mlx5_l3t_set_entry(priv->sh->cnt_id_tbl, id, &data))
4789                                 return 0;
4790                 }
4791         }
4792         if (!priv->counter_fallback && !priv->sh->cmng.query_thread_on)
4793                 /* Start the asynchronous batch query by the host thread. */
4794                 mlx5_set_query_alarm(priv->sh);
4795         return cnt_idx;
4796 err:
4797         if (cnt_free) {
4798                 cnt_free->pool = pool;
4799                 rte_spinlock_lock(&cont->csl);
4800                 TAILQ_INSERT_TAIL(&cont->counters, cnt_free, next);
4801                 rte_spinlock_unlock(&cont->csl);
4802         }
4803         return 0;
4804 }
4805
4806 /**
4807  * Get age param from counter index.
4808  *
4809  * @param[in] dev
4810  *   Pointer to the Ethernet device structure.
4811  * @param[in] counter
4812  *   Index to the counter handler.
4813  *
4814  * @return
4815  *   The aging parameter specified for the counter index.
4816  */
4817 static struct mlx5_age_param*
4818 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
4819                                 uint32_t counter)
4820 {
4821         struct mlx5_flow_counter *cnt;
4822         struct mlx5_flow_counter_pool *pool = NULL;
4823
4824         flow_dv_counter_get_by_idx(dev, counter, &pool);
4825         counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
4826         cnt = MLX5_POOL_GET_CNT(pool, counter);
4827         return MLX5_CNT_TO_AGE(cnt);
4828 }
4829
4830 /**
4831  * Remove a flow counter from aged counter list.
4832  *
4833  * @param[in] dev
4834  *   Pointer to the Ethernet device structure.
4835  * @param[in] counter
4836  *   Index to the counter handler.
4837  * @param[in] cnt
4838  *   Pointer to the counter handler.
4839  */
4840 static void
4841 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
4842                                 uint32_t counter, struct mlx5_flow_counter *cnt)
4843 {
4844         struct mlx5_age_info *age_info;
4845         struct mlx5_age_param *age_param;
4846         struct mlx5_priv *priv = dev->data->dev_private;
4847
4848         age_info = GET_PORT_AGE_INFO(priv);
4849         age_param = flow_dv_counter_idx_get_age(dev, counter);
4850         if (rte_atomic16_cmpset((volatile uint16_t *)
4851                         &age_param->state,
4852                         AGE_CANDIDATE, AGE_FREE)
4853                         != AGE_CANDIDATE) {
4854                 /**
4855                  * We need the lock even it is age timeout,
4856                  * since counter may still in process.
4857                  */
4858                 rte_spinlock_lock(&age_info->aged_sl);
4859                 TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
4860                 rte_spinlock_unlock(&age_info->aged_sl);
4861         }
4862         rte_atomic16_set(&age_param->state, AGE_FREE);
4863 }
4864 /**
4865  * Release a flow counter.
4866  *
4867  * @param[in] dev
4868  *   Pointer to the Ethernet device structure.
4869  * @param[in] counter
4870  *   Index to the counter handler.
4871  */
4872 static void
4873 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
4874 {
4875         struct mlx5_priv *priv = dev->data->dev_private;
4876         struct mlx5_flow_counter_pool *pool = NULL;
4877         struct mlx5_flow_counter *cnt;
4878         struct mlx5_flow_counter_ext *cnt_ext = NULL;
4879
4880         if (!counter)
4881                 return;
4882         cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4883         MLX5_ASSERT(pool);
4884         if (counter < MLX5_CNT_BATCH_OFFSET) {
4885                 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4886                 if (cnt_ext) {
4887                         if (--cnt_ext->ref_cnt)
4888                                 return;
4889                         if (cnt_ext->shared)
4890                                 mlx5_l3t_clear_entry(priv->sh->cnt_id_tbl,
4891                                                      cnt_ext->id);
4892                 }
4893         }
4894         if (IS_AGE_POOL(pool))
4895                 flow_dv_counter_remove_from_age(dev, counter, cnt);
4896         cnt->pool = pool;
4897         /*
4898          * Put the counter back to list to be updated in none fallback mode.
4899          * Currently, we are using two list alternately, while one is in query,
4900          * add the freed counter to the other list based on the pool query_gen
4901          * value. After query finishes, add counter the list to the global
4902          * container counter list. The list changes while query starts. In
4903          * this case, lock will not be needed as query callback and release
4904          * function both operate with the different list.
4905          *
4906          */
4907         if (!priv->counter_fallback)
4908                 TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen], cnt, next);
4909         else
4910                 TAILQ_INSERT_TAIL(&((MLX5_CNT_CONTAINER
4911                                   (priv->sh, 0, 0))->counters),
4912                                   cnt, next);
4913 }
4914
4915 /**
4916  * Verify the @p attributes will be correctly understood by the NIC and store
4917  * them in the @p flow if everything is correct.
4918  *
4919  * @param[in] dev
4920  *   Pointer to dev struct.
4921  * @param[in] attributes
4922  *   Pointer to flow attributes
4923  * @param[in] external
4924  *   This flow rule is created by request external to PMD.
4925  * @param[out] error
4926  *   Pointer to error structure.
4927  *
4928  * @return
4929  *   - 0 on success and non root table.
4930  *   - 1 on success and root table.
4931  *   - a negative errno value otherwise and rte_errno is set.
4932  */
4933 static int
4934 flow_dv_validate_attributes(struct rte_eth_dev *dev,
4935                             const struct rte_flow_attr *attributes,
4936                             bool external __rte_unused,
4937                             struct rte_flow_error *error)
4938 {
4939         struct mlx5_priv *priv = dev->data->dev_private;
4940         uint32_t priority_max = priv->config.flow_prio - 1;
4941         int ret = 0;
4942
4943 #ifndef HAVE_MLX5DV_DR
4944         if (attributes->group)
4945                 return rte_flow_error_set(error, ENOTSUP,
4946                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
4947                                           NULL,
4948                                           "groups are not supported");
4949 #else
4950         uint32_t table = 0;
4951
4952         ret = mlx5_flow_group_to_table(attributes, external,
4953                                        attributes->group, !!priv->fdb_def_rule,
4954                                        &table, error);
4955         if (ret)
4956                 return ret;
4957         if (!table)
4958                 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4959 #endif
4960         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
4961             attributes->priority >= priority_max)
4962                 return rte_flow_error_set(error, ENOTSUP,
4963                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
4964                                           NULL,
4965                                           "priority out of range");
4966         if (attributes->transfer) {
4967                 if (!priv->config.dv_esw_en)
4968                         return rte_flow_error_set
4969                                 (error, ENOTSUP,
4970                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4971                                  "E-Switch dr is not supported");
4972                 if (!(priv->representor || priv->master))
4973                         return rte_flow_error_set
4974                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4975                                  NULL, "E-Switch configuration can only be"
4976                                  " done by a master or a representor device");
4977                 if (attributes->egress)
4978                         return rte_flow_error_set
4979                                 (error, ENOTSUP,
4980                                  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
4981                                  "egress is not supported");
4982         }
4983         if (!(attributes->egress ^ attributes->ingress))
4984                 return rte_flow_error_set(error, ENOTSUP,
4985                                           RTE_FLOW_ERROR_TYPE_ATTR, NULL,
4986                                           "must specify exactly one of "
4987                                           "ingress or egress");
4988         return ret;
4989 }
4990
4991 /**
4992  * Internal validation function. For validating both actions and items.
4993  *
4994  * @param[in] dev
4995  *   Pointer to the rte_eth_dev structure.
4996  * @param[in] attr
4997  *   Pointer to the flow attributes.
4998  * @param[in] items
4999  *   Pointer to the list of items.
5000  * @param[in] actions
5001  *   Pointer to the list of actions.
5002  * @param[in] external
5003  *   This flow rule is created by request external to PMD.
5004  * @param[in] hairpin
5005  *   Number of hairpin TX actions, 0 means classic flow.
5006  * @param[out] error
5007  *   Pointer to the error structure.
5008  *
5009  * @return
5010  *   0 on success, a negative errno value otherwise and rte_errno is set.
5011  */
5012 static int
5013 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
5014                  const struct rte_flow_item items[],
5015                  const struct rte_flow_action actions[],
5016                  bool external, int hairpin, struct rte_flow_error *error)
5017 {
5018         int ret;
5019         uint64_t action_flags = 0;
5020         uint64_t item_flags = 0;
5021         uint64_t last_item = 0;
5022         uint8_t next_protocol = 0xff;
5023         uint16_t ether_type = 0;
5024         int actions_n = 0;
5025         uint8_t item_ipv6_proto = 0;
5026         const struct rte_flow_item *gre_item = NULL;
5027         const struct rte_flow_action_raw_decap *decap;
5028         const struct rte_flow_action_raw_encap *encap;
5029         const struct rte_flow_action_rss *rss;
5030         const struct rte_flow_item_tcp nic_tcp_mask = {
5031                 .hdr = {
5032                         .tcp_flags = 0xFF,
5033                         .src_port = RTE_BE16(UINT16_MAX),
5034                         .dst_port = RTE_BE16(UINT16_MAX),
5035                 }
5036         };
5037         const struct rte_flow_item_ipv4 nic_ipv4_mask = {
5038                 .hdr = {
5039                         .src_addr = RTE_BE32(0xffffffff),
5040                         .dst_addr = RTE_BE32(0xffffffff),
5041                         .type_of_service = 0xff,
5042                         .next_proto_id = 0xff,
5043                         .time_to_live = 0xff,
5044                 },
5045         };
5046         const struct rte_flow_item_ipv6 nic_ipv6_mask = {
5047                 .hdr = {
5048                         .src_addr =
5049                         "\xff\xff\xff\xff\xff\xff\xff\xff"
5050                         "\xff\xff\xff\xff\xff\xff\xff\xff",
5051                         .dst_addr =
5052                         "\xff\xff\xff\xff\xff\xff\xff\xff"
5053                         "\xff\xff\xff\xff\xff\xff\xff\xff",
5054                         .vtc_flow = RTE_BE32(0xffffffff),
5055                         .proto = 0xff,
5056                         .hop_limits = 0xff,
5057                 },
5058         };
5059         const struct rte_flow_item_ecpri nic_ecpri_mask = {
5060                 .hdr = {
5061                         .common = {
5062                                 .u32 =
5063                                 RTE_BE32(((const struct rte_ecpri_common_hdr) {
5064                                         .type = 0xFF,
5065                                         }).u32),
5066                         },
5067                         .dummy[0] = 0xffffffff,
5068                 },
5069         };
5070         struct mlx5_priv *priv = dev->data->dev_private;
5071         struct mlx5_dev_config *dev_conf = &priv->config;
5072         uint16_t queue_index = 0xFFFF;
5073         const struct rte_flow_item_vlan *vlan_m = NULL;
5074         int16_t rw_act_num = 0;
5075         uint64_t is_root;
5076
5077         if (items == NULL)
5078                 return -1;
5079         ret = flow_dv_validate_attributes(dev, attr, external, error);
5080         if (ret < 0)
5081                 return ret;
5082         is_root = (uint64_t)ret;
5083         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5084                 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
5085                 int type = items->type;
5086
5087                 if (!mlx5_flow_os_item_supported(type))
5088                         return rte_flow_error_set(error, ENOTSUP,
5089                                                   RTE_FLOW_ERROR_TYPE_ITEM,
5090                                                   NULL, "item not supported");
5091                 switch (type) {
5092                 case RTE_FLOW_ITEM_TYPE_VOID:
5093                         break;
5094                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
5095                         ret = flow_dv_validate_item_port_id
5096                                         (dev, items, attr, item_flags, error);
5097                         if (ret < 0)
5098                                 return ret;
5099                         last_item = MLX5_FLOW_ITEM_PORT_ID;
5100                         break;
5101                 case RTE_FLOW_ITEM_TYPE_ETH:
5102                         ret = mlx5_flow_validate_item_eth(items, item_flags,
5103                                                           error);
5104                         if (ret < 0)
5105                                 return ret;
5106                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
5107                                              MLX5_FLOW_LAYER_OUTER_L2;
5108                         if (items->mask != NULL && items->spec != NULL) {
5109                                 ether_type =
5110                                         ((const struct rte_flow_item_eth *)
5111                                          items->spec)->type;
5112                                 ether_type &=
5113                                         ((const struct rte_flow_item_eth *)
5114                                          items->mask)->type;
5115                                 ether_type = rte_be_to_cpu_16(ether_type);
5116                         } else {
5117                                 ether_type = 0;
5118                         }
5119                         break;
5120                 case RTE_FLOW_ITEM_TYPE_VLAN:
5121                         ret = flow_dv_validate_item_vlan(items, item_flags,
5122                                                          dev, error);
5123                         if (ret < 0)
5124                                 return ret;
5125                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
5126                                              MLX5_FLOW_LAYER_OUTER_VLAN;
5127                         if (items->mask != NULL && items->spec != NULL) {
5128                                 ether_type =
5129                                         ((const struct rte_flow_item_vlan *)
5130                                          items->spec)->inner_type;
5131                                 ether_type &=
5132                                         ((const struct rte_flow_item_vlan *)
5133                                          items->mask)->inner_type;
5134                                 ether_type = rte_be_to_cpu_16(ether_type);
5135                         } else {
5136                                 ether_type = 0;
5137                         }
5138                         /* Store outer VLAN mask for of_push_vlan action. */
5139                         if (!tunnel)
5140                                 vlan_m = items->mask;
5141                         break;
5142                 case RTE_FLOW_ITEM_TYPE_IPV4:
5143                         mlx5_flow_tunnel_ip_check(items, next_protocol,
5144                                                   &item_flags, &tunnel);
5145                         ret = mlx5_flow_validate_item_ipv4(items, item_flags,
5146                                                            last_item,
5147                                                            ether_type,
5148                                                            &nic_ipv4_mask,
5149                                                            error);
5150                         if (ret < 0)
5151                                 return ret;
5152                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
5153                                              MLX5_FLOW_LAYER_OUTER_L3_IPV4;
5154                         if (items->mask != NULL &&
5155                             ((const struct rte_flow_item_ipv4 *)
5156                              items->mask)->hdr.next_proto_id) {
5157                                 next_protocol =
5158                                         ((const struct rte_flow_item_ipv4 *)
5159                                          (items->spec))->hdr.next_proto_id;
5160                                 next_protocol &=
5161                                         ((const struct rte_flow_item_ipv4 *)
5162                                          (items->mask))->hdr.next_proto_id;
5163                         } else {
5164                                 /* Reset for inner layer. */
5165                                 next_protocol = 0xff;
5166                         }
5167                         break;
5168                 case RTE_FLOW_ITEM_TYPE_IPV6:
5169                         mlx5_flow_tunnel_ip_check(items, next_protocol,
5170                                                   &item_flags, &tunnel);
5171                         ret = mlx5_flow_validate_item_ipv6(items, item_flags,
5172                                                            last_item,
5173                                                            ether_type,
5174                                                            &nic_ipv6_mask,
5175                                                            error);
5176                         if (ret < 0)
5177                                 return ret;
5178                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
5179                                              MLX5_FLOW_LAYER_OUTER_L3_IPV6;
5180                         if (items->mask != NULL &&
5181                             ((const struct rte_flow_item_ipv6 *)
5182                              items->mask)->hdr.proto) {
5183                                 item_ipv6_proto =
5184                                         ((const struct rte_flow_item_ipv6 *)
5185                                          items->spec)->hdr.proto;
5186                                 next_protocol =
5187                                         ((const struct rte_flow_item_ipv6 *)
5188                                          items->spec)->hdr.proto;
5189                                 next_protocol &=
5190                                         ((const struct rte_flow_item_ipv6 *)
5191                                          items->mask)->hdr.proto;
5192                         } else {
5193                                 /* Reset for inner layer. */
5194                                 next_protocol = 0xff;
5195                         }
5196                         break;
5197                 case RTE_FLOW_ITEM_TYPE_TCP:
5198                         ret = mlx5_flow_validate_item_tcp
5199                                                 (items, item_flags,
5200                                                  next_protocol,
5201                                                  &nic_tcp_mask,
5202                                                  error);
5203                         if (ret < 0)
5204                                 return ret;
5205                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
5206                                              MLX5_FLOW_LAYER_OUTER_L4_TCP;
5207                         break;
5208                 case RTE_FLOW_ITEM_TYPE_UDP:
5209                         ret = mlx5_flow_validate_item_udp(items, item_flags,
5210                                                           next_protocol,
5211                                                           error);
5212                         if (ret < 0)
5213                                 return ret;
5214                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
5215                                              MLX5_FLOW_LAYER_OUTER_L4_UDP;
5216                         break;
5217                 case RTE_FLOW_ITEM_TYPE_GRE:
5218                         ret = mlx5_flow_validate_item_gre(items, item_flags,
5219                                                           next_protocol, error);
5220                         if (ret < 0)
5221                                 return ret;
5222                         gre_item = items;
5223                         last_item = MLX5_FLOW_LAYER_GRE;
5224                         break;
5225                 case RTE_FLOW_ITEM_TYPE_NVGRE:
5226                         ret = mlx5_flow_validate_item_nvgre(items, item_flags,
5227                                                             next_protocol,
5228                                                             error);
5229                         if (ret < 0)
5230                                 return ret;
5231                         last_item = MLX5_FLOW_LAYER_NVGRE;
5232                         break;
5233                 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
5234                         ret = mlx5_flow_validate_item_gre_key
5235                                 (items, item_flags, gre_item, error);
5236                         if (ret < 0)
5237                                 return ret;
5238                         last_item = MLX5_FLOW_LAYER_GRE_KEY;
5239                         break;
5240                 case RTE_FLOW_ITEM_TYPE_VXLAN:
5241                         ret = mlx5_flow_validate_item_vxlan(items, item_flags,
5242                                                             error);
5243                         if (ret < 0)
5244                                 return ret;
5245                         last_item = MLX5_FLOW_LAYER_VXLAN;
5246                         break;
5247                 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
5248                         ret = mlx5_flow_validate_item_vxlan_gpe(items,
5249                                                                 item_flags, dev,
5250                                                                 error);
5251                         if (ret < 0)
5252                                 return ret;
5253                         last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
5254                         break;
5255                 case RTE_FLOW_ITEM_TYPE_GENEVE:
5256                         ret = mlx5_flow_validate_item_geneve(items,
5257                                                              item_flags, dev,
5258                                                              error);
5259                         if (ret < 0)
5260                                 return ret;
5261                         last_item = MLX5_FLOW_LAYER_GENEVE;
5262                         break;
5263                 case RTE_FLOW_ITEM_TYPE_MPLS:
5264                         ret = mlx5_flow_validate_item_mpls(dev, items,
5265                                                            item_flags,
5266                                                            last_item, error);
5267                         if (ret < 0)
5268                                 return ret;
5269                         last_item = MLX5_FLOW_LAYER_MPLS;
5270                         break;
5271
5272                 case RTE_FLOW_ITEM_TYPE_MARK:
5273                         ret = flow_dv_validate_item_mark(dev, items, attr,
5274                                                          error);
5275                         if (ret < 0)
5276                                 return ret;
5277                         last_item = MLX5_FLOW_ITEM_MARK;
5278                         break;
5279                 case RTE_FLOW_ITEM_TYPE_META:
5280                         ret = flow_dv_validate_item_meta(dev, items, attr,
5281                                                          error);
5282                         if (ret < 0)
5283                                 return ret;
5284                         last_item = MLX5_FLOW_ITEM_METADATA;
5285                         break;
5286                 case RTE_FLOW_ITEM_TYPE_ICMP:
5287                         ret = mlx5_flow_validate_item_icmp(items, item_flags,
5288                                                            next_protocol,
5289                                                            error);
5290                         if (ret < 0)
5291                                 return ret;
5292                         last_item = MLX5_FLOW_LAYER_ICMP;
5293                         break;
5294                 case RTE_FLOW_ITEM_TYPE_ICMP6:
5295                         ret = mlx5_flow_validate_item_icmp6(items, item_flags,
5296                                                             next_protocol,
5297                                                             error);
5298                         if (ret < 0)
5299                                 return ret;
5300                         item_ipv6_proto = IPPROTO_ICMPV6;
5301                         last_item = MLX5_FLOW_LAYER_ICMP6;
5302                         break;
5303                 case RTE_FLOW_ITEM_TYPE_TAG:
5304                         ret = flow_dv_validate_item_tag(dev, items,
5305                                                         attr, error);
5306                         if (ret < 0)
5307                                 return ret;
5308                         last_item = MLX5_FLOW_ITEM_TAG;
5309                         break;
5310                 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
5311                 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
5312                         break;
5313                 case RTE_FLOW_ITEM_TYPE_GTP:
5314                         ret = flow_dv_validate_item_gtp(dev, items, item_flags,
5315                                                         error);
5316                         if (ret < 0)
5317                                 return ret;
5318                         last_item = MLX5_FLOW_LAYER_GTP;
5319                         break;
5320                 case RTE_FLOW_ITEM_TYPE_ECPRI:
5321                         /* Capacity will be checked in the translate stage. */
5322                         ret = mlx5_flow_validate_item_ecpri(items, item_flags,
5323                                                             last_item,
5324                                                             ether_type,
5325                                                             &nic_ecpri_mask,
5326                                                             error);
5327                         if (ret < 0)
5328                                 return ret;
5329                         last_item = MLX5_FLOW_LAYER_ECPRI;
5330                         break;
5331                 default:
5332                         return rte_flow_error_set(error, ENOTSUP,
5333                                                   RTE_FLOW_ERROR_TYPE_ITEM,
5334                                                   NULL, "item not supported");
5335                 }
5336                 item_flags |= last_item;
5337         }
5338         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5339                 int type = actions->type;
5340
5341                 if (!mlx5_flow_os_action_supported(type))
5342                         return rte_flow_error_set(error, ENOTSUP,
5343                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5344                                                   actions,
5345                                                   "action not supported");
5346                 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5347                         return rte_flow_error_set(error, ENOTSUP,
5348                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5349                                                   actions, "too many actions");
5350                 switch (type) {
5351                 case RTE_FLOW_ACTION_TYPE_VOID:
5352                         break;
5353                 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5354                         ret = flow_dv_validate_action_port_id(dev,
5355                                                               action_flags,
5356                                                               actions,
5357                                                               attr,
5358                                                               error);
5359                         if (ret)
5360                                 return ret;
5361                         action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5362                         ++actions_n;
5363                         break;
5364                 case RTE_FLOW_ACTION_TYPE_FLAG:
5365                         ret = flow_dv_validate_action_flag(dev, action_flags,
5366                                                            attr, error);
5367                         if (ret < 0)
5368                                 return ret;
5369                         if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5370                                 /* Count all modify-header actions as one. */
5371                                 if (!(action_flags &
5372                                       MLX5_FLOW_MODIFY_HDR_ACTIONS))
5373                                         ++actions_n;
5374                                 action_flags |= MLX5_FLOW_ACTION_FLAG |
5375                                                 MLX5_FLOW_ACTION_MARK_EXT;
5376                         } else {
5377                                 action_flags |= MLX5_FLOW_ACTION_FLAG;
5378                                 ++actions_n;
5379                         }
5380                         rw_act_num += MLX5_ACT_NUM_SET_MARK;
5381                         break;
5382                 case RTE_FLOW_ACTION_TYPE_MARK:
5383                         ret = flow_dv_validate_action_mark(dev, actions,
5384                                                            action_flags,
5385                                                            attr, error);
5386                         if (ret < 0)
5387                                 return ret;
5388                         if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5389                                 /* Count all modify-header actions as one. */
5390                                 if (!(action_flags &
5391                                       MLX5_FLOW_MODIFY_HDR_ACTIONS))
5392                                         ++actions_n;
5393                                 action_flags |= MLX5_FLOW_ACTION_MARK |
5394                                                 MLX5_FLOW_ACTION_MARK_EXT;
5395                         } else {
5396                                 action_flags |= MLX5_FLOW_ACTION_MARK;
5397                                 ++actions_n;
5398                         }
5399                         rw_act_num += MLX5_ACT_NUM_SET_MARK;
5400                         break;
5401                 case RTE_FLOW_ACTION_TYPE_SET_META:
5402                         ret = flow_dv_validate_action_set_meta(dev, actions,
5403                                                                action_flags,
5404                                                                attr, error);
5405                         if (ret < 0)
5406                                 return ret;
5407                         /* Count all modify-header actions as one action. */
5408                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5409                                 ++actions_n;
5410                         action_flags |= MLX5_FLOW_ACTION_SET_META;
5411                         rw_act_num += MLX5_ACT_NUM_SET_META;
5412                         break;
5413                 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5414                         ret = flow_dv_validate_action_set_tag(dev, actions,
5415                                                               action_flags,
5416                                                               attr, error);
5417                         if (ret < 0)
5418                                 return ret;
5419                         /* Count all modify-header actions as one action. */
5420                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5421                                 ++actions_n;
5422                         action_flags |= MLX5_FLOW_ACTION_SET_TAG;
5423                         rw_act_num += MLX5_ACT_NUM_SET_TAG;
5424                         break;
5425                 case RTE_FLOW_ACTION_TYPE_DROP:
5426                         ret = mlx5_flow_validate_action_drop(action_flags,
5427                                                              attr, error);
5428                         if (ret < 0)
5429                                 return ret;
5430                         action_flags |= MLX5_FLOW_ACTION_DROP;
5431                         ++actions_n;
5432                         break;
5433                 case RTE_FLOW_ACTION_TYPE_QUEUE:
5434                         ret = mlx5_flow_validate_action_queue(actions,
5435                                                               action_flags, dev,
5436                                                               attr, error);
5437                         if (ret < 0)
5438                                 return ret;
5439                         queue_index = ((const struct rte_flow_action_queue *)
5440                                                         (actions->conf))->index;
5441                         action_flags |= MLX5_FLOW_ACTION_QUEUE;
5442                         ++actions_n;
5443                         break;
5444                 case RTE_FLOW_ACTION_TYPE_RSS:
5445                         rss = actions->conf;
5446                         ret = mlx5_flow_validate_action_rss(actions,
5447                                                             action_flags, dev,
5448                                                             attr, item_flags,
5449                                                             error);
5450                         if (ret < 0)
5451                                 return ret;
5452                         if (rss != NULL && rss->queue_num)
5453                                 queue_index = rss->queue[0];
5454                         action_flags |= MLX5_FLOW_ACTION_RSS;
5455                         ++actions_n;
5456                         break;
5457                 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
5458                         ret =
5459                         mlx5_flow_validate_action_default_miss(action_flags,
5460                                         attr, error);
5461                         if (ret < 0)
5462                                 return ret;
5463                         action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
5464                         ++actions_n;
5465                         break;
5466                 case RTE_FLOW_ACTION_TYPE_COUNT:
5467                         ret = flow_dv_validate_action_count(dev, error);
5468                         if (ret < 0)
5469                                 return ret;
5470                         action_flags |= MLX5_FLOW_ACTION_COUNT;
5471                         ++actions_n;
5472                         break;
5473                 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5474                         if (flow_dv_validate_action_pop_vlan(dev,
5475                                                              action_flags,
5476                                                              actions,
5477                                                              item_flags, attr,
5478                                                              error))
5479                                 return -rte_errno;
5480                         action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5481                         ++actions_n;
5482                         break;
5483                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5484                         ret = flow_dv_validate_action_push_vlan(dev,
5485                                                                 action_flags,
5486                                                                 vlan_m,
5487                                                                 actions, attr,
5488                                                                 error);
5489                         if (ret < 0)
5490                                 return ret;
5491                         action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5492                         ++actions_n;
5493                         break;
5494                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5495                         ret = flow_dv_validate_action_set_vlan_pcp
5496                                                 (action_flags, actions, error);
5497                         if (ret < 0)
5498                                 return ret;
5499                         /* Count PCP with push_vlan command. */
5500                         action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5501                         break;
5502                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5503                         ret = flow_dv_validate_action_set_vlan_vid
5504                                                 (item_flags, action_flags,
5505                                                  actions, error);
5506                         if (ret < 0)
5507                                 return ret;
5508                         /* Count VID with push_vlan command. */
5509                         action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5510                         rw_act_num += MLX5_ACT_NUM_MDF_VID;
5511                         break;
5512                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5513                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5514                         ret = flow_dv_validate_action_l2_encap(dev,
5515                                                                action_flags,
5516                                                                actions, attr,
5517                                                                error);
5518                         if (ret < 0)
5519                                 return ret;
5520                         action_flags |= MLX5_FLOW_ACTION_ENCAP;
5521                         ++actions_n;
5522                         break;
5523                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5524                 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5525                         ret = flow_dv_validate_action_decap(dev, action_flags,
5526                                                             attr, error);
5527                         if (ret < 0)
5528                                 return ret;
5529                         action_flags |= MLX5_FLOW_ACTION_DECAP;
5530                         ++actions_n;
5531                         break;
5532                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5533                         ret = flow_dv_validate_action_raw_encap_decap
5534                                 (dev, NULL, actions->conf, attr, &action_flags,
5535                                  &actions_n, error);
5536                         if (ret < 0)
5537                                 return ret;
5538                         break;
5539                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5540                         decap = actions->conf;
5541                         while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5542                                 ;
5543                         if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5544                                 encap = NULL;
5545                                 actions--;
5546                         } else {
5547                                 encap = actions->conf;
5548                         }
5549                         ret = flow_dv_validate_action_raw_encap_decap
5550                                            (dev,
5551                                             decap ? decap : &empty_decap, encap,
5552                                             attr, &action_flags, &actions_n,
5553                                             error);
5554                         if (ret < 0)
5555                                 return ret;
5556                         break;
5557                 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5558                 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5559                         ret = flow_dv_validate_action_modify_mac(action_flags,
5560                                                                  actions,
5561                                                                  item_flags,
5562                                                                  error);
5563                         if (ret < 0)
5564                                 return ret;
5565                         /* Count all modify-header actions as one action. */
5566                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5567                                 ++actions_n;
5568                         action_flags |= actions->type ==
5569                                         RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5570                                                 MLX5_FLOW_ACTION_SET_MAC_SRC :
5571                                                 MLX5_FLOW_ACTION_SET_MAC_DST;
5572                         /*
5573                          * Even if the source and destination MAC addresses have
5574                          * overlap in the header with 4B alignment, the convert
5575                          * function will handle them separately and 4 SW actions
5576                          * will be created. And 2 actions will be added each
5577                          * time no matter how many bytes of address will be set.
5578                          */
5579                         rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5580                         break;
5581                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5582                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5583                         ret = flow_dv_validate_action_modify_ipv4(action_flags,
5584                                                                   actions,
5585                                                                   item_flags,
5586                                                                   error);
5587                         if (ret < 0)
5588                                 return ret;
5589                         /* Count all modify-header actions as one action. */
5590                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5591                                 ++actions_n;
5592                         action_flags |= actions->type ==
5593                                         RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5594                                                 MLX5_FLOW_ACTION_SET_IPV4_SRC :
5595                                                 MLX5_FLOW_ACTION_SET_IPV4_DST;
5596                         rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5597                         break;
5598                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5599                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5600                         ret = flow_dv_validate_action_modify_ipv6(action_flags,
5601                                                                   actions,
5602                                                                   item_flags,
5603                                                                   error);
5604                         if (ret < 0)
5605                                 return ret;
5606                         if (item_ipv6_proto == IPPROTO_ICMPV6)
5607                                 return rte_flow_error_set(error, ENOTSUP,
5608                                         RTE_FLOW_ERROR_TYPE_ACTION,
5609                                         actions,
5610                                         "Can't change header "
5611                                         "with ICMPv6 proto");
5612                         /* Count all modify-header actions as one action. */
5613                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5614                                 ++actions_n;
5615                         action_flags |= actions->type ==
5616                                         RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5617                                                 MLX5_FLOW_ACTION_SET_IPV6_SRC :
5618                                                 MLX5_FLOW_ACTION_SET_IPV6_DST;
5619                         rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5620                         break;
5621                 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5622                 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5623                         ret = flow_dv_validate_action_modify_tp(action_flags,
5624                                                                 actions,
5625                                                                 item_flags,
5626                                                                 error);
5627                         if (ret < 0)
5628                                 return ret;
5629                         /* Count all modify-header actions as one action. */
5630                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5631                                 ++actions_n;
5632                         action_flags |= actions->type ==
5633                                         RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5634                                                 MLX5_FLOW_ACTION_SET_TP_SRC :
5635                                                 MLX5_FLOW_ACTION_SET_TP_DST;
5636                         rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5637                         break;
5638                 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5639                 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5640                         ret = flow_dv_validate_action_modify_ttl(action_flags,
5641                                                                  actions,
5642                                                                  item_flags,
5643                                                                  error);
5644                         if (ret < 0)
5645                                 return ret;
5646                         /* Count all modify-header actions as one action. */
5647                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5648                                 ++actions_n;
5649                         action_flags |= actions->type ==
5650                                         RTE_FLOW_ACTION_TYPE_SET_TTL ?
5651                                                 MLX5_FLOW_ACTION_SET_TTL :
5652                                                 MLX5_FLOW_ACTION_DEC_TTL;
5653                         rw_act_num += MLX5_ACT_NUM_MDF_TTL;
5654                         break;
5655                 case RTE_FLOW_ACTION_TYPE_JUMP:
5656                         ret = flow_dv_validate_action_jump(actions,
5657                                                            action_flags,
5658                                                            attr, external,
5659                                                            error);
5660                         if (ret)
5661                                 return ret;
5662                         ++actions_n;
5663                         action_flags |= MLX5_FLOW_ACTION_JUMP;
5664                         break;
5665                 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5666                 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5667                         ret = flow_dv_validate_action_modify_tcp_seq
5668                                                                 (action_flags,
5669                                                                  actions,
5670                                                                  item_flags,
5671                                                                  error);
5672                         if (ret < 0)
5673                                 return ret;
5674                         /* Count all modify-header actions as one action. */
5675                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5676                                 ++actions_n;
5677                         action_flags |= actions->type ==
5678                                         RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5679                                                 MLX5_FLOW_ACTION_INC_TCP_SEQ :
5680                                                 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5681                         rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
5682                         break;
5683                 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5684                 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5685                         ret = flow_dv_validate_action_modify_tcp_ack
5686                                                                 (action_flags,
5687                                                                  actions,
5688                                                                  item_flags,
5689                                                                  error);
5690                         if (ret < 0)
5691                                 return ret;
5692                         /* Count all modify-header actions as one action. */
5693                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5694                                 ++actions_n;
5695                         action_flags |= actions->type ==
5696                                         RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5697                                                 MLX5_FLOW_ACTION_INC_TCP_ACK :
5698                                                 MLX5_FLOW_ACTION_DEC_TCP_ACK;
5699                         rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
5700                         break;
5701                 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5702                         break;
5703                 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5704                 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5705                         rw_act_num += MLX5_ACT_NUM_SET_TAG;
5706                         break;
5707                 case RTE_FLOW_ACTION_TYPE_METER:
5708                         ret = mlx5_flow_validate_action_meter(dev,
5709                                                               action_flags,
5710                                                               actions, attr,
5711                                                               error);
5712                         if (ret < 0)
5713                                 return ret;
5714                         action_flags |= MLX5_FLOW_ACTION_METER;
5715                         ++actions_n;
5716                         /* Meter action will add one more TAG action. */
5717                         rw_act_num += MLX5_ACT_NUM_SET_TAG;
5718                         break;
5719                 case RTE_FLOW_ACTION_TYPE_AGE:
5720                         ret = flow_dv_validate_action_age(action_flags,
5721                                                           actions, dev,
5722                                                           error);
5723                         if (ret < 0)
5724                                 return ret;
5725                         action_flags |= MLX5_FLOW_ACTION_AGE;
5726                         ++actions_n;
5727                         break;
5728                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5729                         ret = flow_dv_validate_action_modify_ipv4_dscp
5730                                                          (action_flags,
5731                                                           actions,
5732                                                           item_flags,
5733                                                           error);
5734                         if (ret < 0)
5735                                 return ret;
5736                         /* Count all modify-header actions as one action. */
5737                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5738                                 ++actions_n;
5739                         action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
5740                         rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5741                         break;
5742                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5743                         ret = flow_dv_validate_action_modify_ipv6_dscp
5744                                                                 (action_flags,
5745                                                                  actions,
5746                                                                  item_flags,
5747                                                                  error);
5748                         if (ret < 0)
5749                                 return ret;
5750                         /* Count all modify-header actions as one action. */
5751                         if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5752                                 ++actions_n;
5753                         action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
5754                         rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5755                         break;
5756                 default:
5757                         return rte_flow_error_set(error, ENOTSUP,
5758                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5759                                                   actions,
5760                                                   "action not supported");
5761                 }
5762         }
5763         /*
5764          * Validate the drop action mutual exclusion with other actions.
5765          * Drop action is mutually-exclusive with any other action, except for
5766          * Count action.
5767          */
5768         if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
5769             (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
5770                 return rte_flow_error_set(error, EINVAL,
5771                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5772                                           "Drop action is mutually-exclusive "
5773                                           "with any other action, except for "
5774                                           "Count action");
5775         /* Eswitch has few restrictions on using items and actions */
5776         if (attr->transfer) {
5777                 if (!mlx5_flow_ext_mreg_supported(dev) &&
5778                     action_flags & MLX5_FLOW_ACTION_FLAG)
5779                         return rte_flow_error_set(error, ENOTSUP,
5780                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5781                                                   NULL,
5782                                                   "unsupported action FLAG");
5783                 if (!mlx5_flow_ext_mreg_supported(dev) &&
5784                     action_flags & MLX5_FLOW_ACTION_MARK)
5785                         return rte_flow_error_set(error, ENOTSUP,
5786                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5787                                                   NULL,
5788                                                   "unsupported action MARK");
5789                 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
5790                         return rte_flow_error_set(error, ENOTSUP,
5791                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5792                                                   NULL,
5793                                                   "unsupported action QUEUE");
5794                 if (action_flags & MLX5_FLOW_ACTION_RSS)
5795                         return rte_flow_error_set(error, ENOTSUP,
5796                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5797                                                   NULL,
5798                                                   "unsupported action RSS");
5799                 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
5800                         return rte_flow_error_set(error, EINVAL,
5801                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5802                                                   actions,
5803                                                   "no fate action is found");
5804         } else {
5805                 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
5806                         return rte_flow_error_set(error, EINVAL,
5807                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5808                                                   actions,
5809                                                   "no fate action is found");
5810         }
5811         /* Continue validation for Xcap and VLAN actions.*/
5812         if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
5813                              MLX5_FLOW_VLAN_ACTIONS)) &&
5814             (queue_index == 0xFFFF ||
5815              mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
5816                 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
5817                     MLX5_FLOW_XCAP_ACTIONS)
5818                         return rte_flow_error_set(error, ENOTSUP,
5819                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5820                                                   NULL, "encap and decap "
5821                                                   "combination aren't supported");
5822                 if (!attr->transfer && attr->ingress) {
5823                         if (action_flags & MLX5_FLOW_ACTION_ENCAP)
5824                                 return rte_flow_error_set
5825                                                 (error, ENOTSUP,
5826                                                  RTE_FLOW_ERROR_TYPE_ACTION,
5827                                                  NULL, "encap is not supported"
5828                                                  " for ingress traffic");
5829                         else if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
5830                                 return rte_flow_error_set
5831                                                 (error, ENOTSUP,
5832                                                  RTE_FLOW_ERROR_TYPE_ACTION,
5833                                                  NULL, "push VLAN action not "
5834                                                  "supported for ingress");
5835                         else if ((action_flags & MLX5_FLOW_VLAN_ACTIONS) ==
5836                                         MLX5_FLOW_VLAN_ACTIONS)
5837                                 return rte_flow_error_set
5838                                                 (error, ENOTSUP,
5839                                                  RTE_FLOW_ERROR_TYPE_ACTION,
5840                                                  NULL, "no support for "
5841                                                  "multiple VLAN actions");
5842                 }
5843         }
5844         /* Hairpin flow will add one more TAG action. */
5845         if (hairpin > 0)
5846                 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5847         /* extra metadata enabled: one more TAG action will be add. */
5848         if (dev_conf->dv_flow_en &&
5849             dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
5850             mlx5_flow_ext_mreg_supported(dev))
5851                 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5852         if ((uint32_t)rw_act_num >
5853                         flow_dv_modify_hdr_action_max(dev, is_root)) {
5854                 return rte_flow_error_set(error, ENOTSUP,
5855                                           RTE_FLOW_ERROR_TYPE_ACTION,
5856                                           NULL, "too many header modify"
5857                                           " actions to support");
5858         }
5859         return 0;
5860 }
5861
5862 /**
5863  * Internal preparation function. Allocates the DV flow size,
5864  * this size is constant.
5865  *
5866  * @param[in] dev
5867  *   Pointer to the rte_eth_dev structure.
5868  * @param[in] attr
5869  *   Pointer to the flow attributes.
5870  * @param[in] items
5871  *   Pointer to the list of items.
5872  * @param[in] actions
5873  *   Pointer to the list of actions.
5874  * @param[out] error
5875  *   Pointer to the error structure.
5876  *
5877  * @return
5878  *   Pointer to mlx5_flow object on success,
5879  *   otherwise NULL and rte_errno is set.
5880  */
5881 static struct mlx5_flow *
5882 flow_dv_prepare(struct rte_eth_dev *dev,
5883                 const struct rte_flow_attr *attr __rte_unused,
5884                 const struct rte_flow_item items[] __rte_unused,
5885                 const struct rte_flow_action actions[] __rte_unused,
5886                 struct rte_flow_error *error)
5887 {
5888         uint32_t handle_idx = 0;
5889         struct mlx5_flow *dev_flow;
5890         struct mlx5_flow_handle *dev_handle;
5891         struct mlx5_priv *priv = dev->data->dev_private;
5892
5893         /* In case of corrupting the memory. */
5894         if (priv->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
5895                 rte_flow_error_set(error, ENOSPC,
5896                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5897                                    "not free temporary device flow");
5898                 return NULL;
5899         }
5900         dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
5901                                    &handle_idx);
5902         if (!dev_handle) {
5903                 rte_flow_error_set(error, ENOMEM,
5904                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5905                                    "not enough memory to create flow handle");
5906                 return NULL;
5907         }
5908         /* No multi-thread supporting. */
5909         dev_flow = &((struct mlx5_flow *)priv->inter_flows)[priv->flow_idx++];
5910         dev_flow->handle = dev_handle;
5911         dev_flow->handle_idx = handle_idx;
5912         /*
5913          * In some old rdma-core releases, before continuing, a check of the
5914          * length of matching parameter will be done at first. It needs to use
5915          * the length without misc4 param. If the flow has misc4 support, then
5916          * the length needs to be adjusted accordingly. Each param member is
5917          * aligned with a 64B boundary naturally.
5918          */
5919         dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param) -
5920                                   MLX5_ST_SZ_BYTES(fte_match_set_misc4);
5921         /*
5922          * The matching value needs to be cleared to 0 before using. In the
5923          * past, it will be automatically cleared when using rte_*alloc
5924          * API. The time consumption will be almost the same as before.
5925          */
5926         memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
5927         dev_flow->ingress = attr->ingress;
5928         dev_flow->dv.transfer = attr->transfer;
5929         return dev_flow;
5930 }
5931
5932 #ifdef RTE_LIBRTE_MLX5_DEBUG
5933 /**
5934  * Sanity check for match mask and value. Similar to check_valid_spec() in
5935  * kernel driver. If unmasked bit is present in value, it returns failure.
5936  *
5937  * @param match_mask
5938  *   pointer to match mask buffer.
5939  * @param match_value
5940  *   pointer to match value buffer.
5941  *
5942  * @return
5943  *   0 if valid, -EINVAL otherwise.
5944  */
5945 static int
5946 flow_dv_check_valid_spec(void *match_mask, void *match_value)
5947 {
5948         uint8_t *m = match_mask;
5949         uint8_t *v = match_value;
5950         unsigned int i;
5951
5952         for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
5953                 if (v[i] & ~m[i]) {
5954                         DRV_LOG(ERR,
5955                                 "match_value differs from match_criteria"
5956                                 " %p[%u] != %p[%u]",
5957                                 match_value, i, match_mask, i);
5958                         return -EINVAL;
5959                 }
5960         }
5961         return 0;
5962 }
5963 #endif
5964
5965 /**
5966  * Add match of ip_version.
5967  *
5968  * @param[in] group
5969  *   Flow group.
5970  * @param[in] headers_v
5971  *   Values header pointer.
5972  * @param[in] headers_m
5973  *   Masks header pointer.
5974  * @param[in] ip_version
5975  *   The IP version to set.
5976  */
5977 static inline void
5978 flow_dv_set_match_ip_version(uint32_t group,
5979                              void *headers_v,
5980                              void *headers_m,
5981                              uint8_t ip_version)
5982 {
5983         if (group == 0)
5984                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
5985         else
5986                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
5987                          ip_version);
5988         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
5989         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
5990         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
5991 }
5992
5993 /**
5994  * Add Ethernet item to matcher and to the value.
5995  *
5996  * @param[in, out] matcher
5997  *   Flow matcher.
5998  * @param[in, out] key
5999  *   Flow matcher value.
6000  * @param[in] item
6001  *   Flow pattern to translate.
6002  * @param[in] inner
6003  *   Item is inner pattern.
6004  */
6005 static void
6006 flow_dv_translate_item_eth(void *matcher, void *key,
6007                            const struct rte_flow_item *item, int inner,
6008                            uint32_t group)
6009 {
6010         const struct rte_flow_item_eth *eth_m = item->mask;
6011         const struct rte_flow_item_eth *eth_v = item->spec;
6012         const struct rte_flow_item_eth nic_mask = {
6013                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6014                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6015                 .type = RTE_BE16(0xffff),
6016         };
6017         void *headers_m;
6018         void *headers_v;
6019         char *l24_v;
6020         unsigned int i;
6021
6022         if (!eth_v)
6023                 return;
6024         if (!eth_m)
6025                 eth_m = &nic_mask;
6026         if (inner) {
6027                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6028                                          inner_headers);
6029                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6030         } else {
6031                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6032                                          outer_headers);
6033                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6034         }
6035         memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
6036                &eth_m->dst, sizeof(eth_m->dst));
6037         /* The value must be in the range of the mask. */
6038         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
6039         for (i = 0; i < sizeof(eth_m->dst); ++i)
6040                 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
6041         memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
6042                &eth_m->src, sizeof(eth_m->src));
6043         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
6044         /* The value must be in the range of the mask. */
6045         for (i = 0; i < sizeof(eth_m->dst); ++i)
6046                 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
6047         if (eth_v->type) {
6048                 /* When ethertype is present set mask for tagged VLAN. */
6049                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6050                 /* Set value for tagged VLAN if ethertype is 802.1Q. */
6051                 if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_VLAN) ||
6052                     eth_v->type == RTE_BE16(RTE_ETHER_TYPE_QINQ)) {
6053                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag,
6054                                  1);
6055                         /* Return here to avoid setting match on ethertype. */
6056                         return;
6057                 }
6058         }
6059         /*
6060          * HW supports match on one Ethertype, the Ethertype following the last
6061          * VLAN tag of the packet (see PRM).
6062          * Set match on ethertype only if ETH header is not followed by VLAN.
6063          * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6064          * ethertype, and use ip_version field instead.
6065          * eCPRI over Ether layer will use type value 0xAEFE.
6066          */
6067         if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
6068             eth_m->type == 0xFFFF) {
6069                 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6070         } else if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
6071                    eth_m->type == 0xFFFF) {
6072                 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6073         } else {
6074                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
6075                          rte_be_to_cpu_16(eth_m->type));
6076                 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6077                                      ethertype);
6078                 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
6079         }
6080 }
6081
6082 /**
6083  * Add VLAN item to matcher and to the value.
6084  *
6085  * @param[in, out] dev_flow
6086  *   Flow descriptor.
6087  * @param[in, out] matcher
6088  *   Flow matcher.
6089  * @param[in, out] key
6090  *   Flow matcher value.
6091  * @param[in] item
6092  *   Flow pattern to translate.
6093  * @param[in] inner
6094  *   Item is inner pattern.
6095  */
6096 static void
6097 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
6098                             void *matcher, void *key,
6099                             const struct rte_flow_item *item,
6100                             int inner, uint32_t group)
6101 {
6102         const struct rte_flow_item_vlan *vlan_m = item->mask;
6103         const struct rte_flow_item_vlan *vlan_v = item->spec;
6104         void *headers_m;
6105         void *headers_v;
6106         uint16_t tci_m;
6107         uint16_t tci_v;
6108
6109         if (inner) {
6110                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6111                                          inner_headers);
6112                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6113         } else {
6114                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6115                                          outer_headers);
6116                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6117                 /*
6118                  * This is workaround, masks are not supported,
6119                  * and pre-validated.
6120                  */
6121                 if (vlan_v)
6122                         dev_flow->handle->vf_vlan.tag =
6123                                         rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
6124         }
6125         /*
6126          * When VLAN item exists in flow, mark packet as tagged,
6127          * even if TCI is not specified.
6128          */
6129         MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6130         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
6131         if (!vlan_v)
6132                 return;
6133         if (!vlan_m)
6134                 vlan_m = &rte_flow_item_vlan_mask;
6135         tci_m = rte_be_to_cpu_16(vlan_m->tci);
6136         tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
6137         MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
6138         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
6139         MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
6140         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
6141         MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
6142         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
6143         /*
6144          * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6145          * ethertype, and use ip_version field instead.
6146          */
6147         if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
6148             vlan_m->inner_type == 0xFFFF) {
6149                 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6150         } else if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
6151                    vlan_m->inner_type == 0xFFFF) {
6152                 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6153         } else {
6154                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
6155                          rte_be_to_cpu_16(vlan_m->inner_type));
6156                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
6157                          rte_be_to_cpu_16(vlan_m->inner_type &
6158                                           vlan_v->inner_type));
6159         }
6160 }
6161
6162 /**
6163  * Add IPV4 item to matcher and to the value.
6164  *
6165  * @param[in, out] matcher
6166  *   Flow matcher.
6167  * @param[in, out] key
6168  *   Flow matcher value.
6169  * @param[in] item
6170  *   Flow pattern to translate.
6171  * @param[in] item_flags
6172  *   Bit-fields that holds the items detected until now.
6173  * @param[in] inner
6174  *   Item is inner pattern.
6175  * @param[in] group
6176  *   The group to insert the rule.
6177  */
6178 static void
6179 flow_dv_translate_item_ipv4(void *matcher, void *key,
6180                             const struct rte_flow_item *item,
6181                             const uint64_t item_flags,
6182                             int inner, uint32_t group)
6183 {
6184         const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
6185         const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
6186         const struct rte_flow_item_ipv4 nic_mask = {
6187                 .hdr = {
6188                         .src_addr = RTE_BE32(0xffffffff),
6189                         .dst_addr = RTE_BE32(0xffffffff),
6190                         .type_of_service = 0xff,
6191                         .next_proto_id = 0xff,
6192                         .time_to_live = 0xff,
6193                 },
6194         };
6195         void *headers_m;
6196         void *headers_v;
6197         char *l24_m;
6198         char *l24_v;
6199         uint8_t tos;
6200
6201         if (inner) {
6202                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6203                                          inner_headers);
6204                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6205         } else {
6206                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6207                                          outer_headers);
6208                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6209         }
6210         flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6211         /*
6212          * On outer header (which must contains L2), or inner header with L2,
6213          * set cvlan_tag mask bit to mark this packet as untagged.
6214          * This should be done even if item->spec is empty.
6215          */
6216         if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
6217                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6218         if (!ipv4_v)
6219                 return;
6220         if (!ipv4_m)
6221                 ipv4_m = &nic_mask;
6222         l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6223                              dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6224         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6225                              dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6226         *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
6227         *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
6228         l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6229                           src_ipv4_src_ipv6.ipv4_layout.ipv4);
6230         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6231                           src_ipv4_src_ipv6.ipv4_layout.ipv4);
6232         *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
6233         *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
6234         tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
6235         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
6236                  ipv4_m->hdr.type_of_service);
6237         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
6238         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
6239                  ipv4_m->hdr.type_of_service >> 2);
6240         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
6241         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6242                  ipv4_m->hdr.next_proto_id);
6243         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6244                  ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
6245         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6246                  ipv4_m->hdr.time_to_live);
6247         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6248                  ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
6249 }
6250
6251 /**
6252  * Add IPV6 item to matcher and to the value.
6253  *
6254  * @param[in, out] matcher
6255  *   Flow matcher.
6256  * @param[in, out] key
6257  *   Flow matcher value.
6258  * @param[in] item
6259  *   Flow pattern to translate.
6260  * @param[in] item_flags
6261  *   Bit-fields that holds the items detected until now.
6262  * @param[in] inner
6263  *   Item is inner pattern.
6264  * @param[in] group
6265  *   The group to insert the rule.
6266  */
6267 static void
6268 flow_dv_translate_item_ipv6(void *matcher, void *key,
6269                             const struct rte_flow_item *item,
6270                             const uint64_t item_flags,
6271                             int inner, uint32_t group)
6272 {
6273         const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
6274         const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
6275         const struct rte_flow_item_ipv6 nic_mask = {
6276                 .hdr = {
6277                         .src_addr =
6278                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
6279                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
6280                         .dst_addr =
6281                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
6282                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
6283                         .vtc_flow = RTE_BE32(0xffffffff),
6284                         .proto = 0xff,
6285                         .hop_limits = 0xff,
6286                 },
6287         };
6288         void *headers_m;
6289         void *headers_v;
6290         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6291         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6292         char *l24_m;
6293         char *l24_v;
6294         uint32_t vtc_m;
6295         uint32_t vtc_v;
6296         int i;
6297         int size;
6298
6299         if (inner) {
6300                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6301                                          inner_headers);
6302                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6303         } else {
6304                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6305                                          outer_headers);
6306                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6307         }
6308         flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6309         /*
6310          * On outer header (which must contains L2), or inner header with L2,
6311          * set cvlan_tag mask bit to mark this packet as untagged.
6312          * This should be done even if item->spec is empty.
6313          */
6314         if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
6315                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6316         if (!ipv6_v)
6317                 return;
6318         if (!ipv6_m)
6319                 ipv6_m = &nic_mask;
6320         size = sizeof(ipv6_m->hdr.dst_addr);
6321         l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6322                              dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6323         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6324                              dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6325         memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
6326         for (i = 0; i < size; ++i)
6327                 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
6328         l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6329                              src_ipv4_src_ipv6.ipv6_layout.ipv6);
6330         l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6331                              src_ipv4_src_ipv6.ipv6_layout.ipv6);
6332         memcpy(l24_m, ipv6_m->hdr.src_addr, size);
6333         for (i = 0; i < size; ++i)
6334                 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
6335         /* TOS. */
6336         vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
6337         vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
6338         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
6339         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
6340         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
6341         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
6342         /* Label. */
6343         if (inner) {
6344                 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
6345                          vtc_m);
6346                 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
6347                          vtc_v);
6348         } else {
6349                 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
6350                          vtc_m);
6351                 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
6352                          vtc_v);
6353         }
6354         /* Protocol. */
6355         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6356                  ipv6_m->hdr.proto);
6357         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6358                  ipv6_v->hdr.proto & ipv6_m->hdr.proto);
6359         /* Hop limit. */
6360         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6361                  ipv6_m->hdr.hop_limits);
6362         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6363                  ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
6364 }
6365
6366 /**
6367  * Add TCP item to matcher and to the value.
6368  *
6369  * @param[in, out] matcher
6370  *   Flow matcher.
6371  * @param[in, out] key
6372  *   Flow matcher value.
6373  * @param[in] item
6374  *   Flow pattern to translate.
6375  * @param[in] inner
6376  *   Item is inner pattern.
6377  */
6378 static void
6379 flow_dv_translate_item_tcp(void *matcher, void *key,
6380                            const struct rte_flow_item *item,
6381                            int inner)
6382 {
6383         const struct rte_flow_item_tcp *tcp_m = item->mask;
6384         const struct rte_flow_item_tcp *tcp_v = item->spec;
6385         void *headers_m;
6386         void *headers_v;
6387
6388         if (inner) {
6389                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6390                                          inner_headers);
6391                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6392         } else {
6393                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6394                                          outer_headers);
6395                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6396         }
6397         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6398         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
6399         if (!tcp_v)
6400                 return;
6401         if (!tcp_m)
6402                 tcp_m = &rte_flow_item_tcp_mask;
6403         MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
6404                  rte_be_to_cpu_16(tcp_m->hdr.src_port));
6405         MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
6406                  rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
6407         MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
6408                  rte_be_to_cpu_16(tcp_m->hdr.dst_port));
6409         MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
6410                  rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
6411         MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
6412                  tcp_m->hdr.tcp_flags);
6413         MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
6414                  (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
6415 }
6416
6417 /**
6418  * Add UDP item to matcher and to the value.
6419  *
6420  * @param[in, out] matcher
6421  *   Flow matcher.
6422  * @param[in, out] key
6423  *   Flow matcher value.
6424  * @param[in] item
6425  *   Flow pattern to translate.
6426  * @param[in] inner
6427  *   Item is inner pattern.
6428  */
6429 static void
6430 flow_dv_translate_item_udp(void *matcher, void *key,
6431                            const struct rte_flow_item *item,
6432                            int inner)
6433 {
6434         const struct rte_flow_item_udp *udp_m = item->mask;
6435         const struct rte_flow_item_udp *udp_v = item->spec;
6436         void *headers_m;
6437         void *headers_v;
6438
6439         if (inner) {
6440                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6441                                          inner_headers);
6442                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6443         } else {
6444                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6445                                          outer_headers);
6446                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6447         }
6448         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6449         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
6450         if (!udp_v)
6451                 return;
6452         if (!udp_m)
6453                 udp_m = &rte_flow_item_udp_mask;
6454         MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
6455                  rte_be_to_cpu_16(udp_m->hdr.src_port));
6456         MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
6457                  rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
6458         MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
6459                  rte_be_to_cpu_16(udp_m->hdr.dst_port));
6460         MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6461                  rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
6462 }
6463
6464 /**
6465  * Add GRE optional Key item to matcher and to the value.
6466  *
6467  * @param[in, out] matcher
6468  *   Flow matcher.
6469  * @param[in, out] key
6470  *   Flow matcher value.
6471  * @param[in] item
6472  *   Flow pattern to translate.
6473  * @param[in] inner
6474  *   Item is inner pattern.
6475  */
6476 static void
6477 flow_dv_translate_item_gre_key(void *matcher, void *key,
6478                                    const struct rte_flow_item *item)
6479 {
6480         const rte_be32_t *key_m = item->mask;
6481         const rte_be32_t *key_v = item->spec;
6482         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6483         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6484         rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
6485
6486         /* GRE K bit must be on and should already be validated */
6487         MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
6488         MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
6489         if (!key_v)
6490                 return;
6491         if (!key_m)
6492                 key_m = &gre_key_default_mask;
6493         MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
6494                  rte_be_to_cpu_32(*key_m) >> 8);
6495         MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
6496                  rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
6497         MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
6498                  rte_be_to_cpu_32(*key_m) & 0xFF);
6499         MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
6500                  rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
6501 }
6502
6503 /**
6504  * Add GRE item to matcher and to the value.
6505  *
6506  * @param[in, out] matcher
6507  *   Flow matcher.
6508  * @param[in, out] key
6509  *   Flow matcher value.
6510  * @param[in] item
6511  *   Flow pattern to translate.
6512  * @param[in] inner
6513  *   Item is inner pattern.
6514  */
6515 static void
6516 flow_dv_translate_item_gre(void *matcher, void *key,
6517                            const struct rte_flow_item *item,
6518                            int inner)
6519 {
6520         const struct rte_flow_item_gre *gre_m = item->mask;
6521         const struct rte_flow_item_gre *gre_v = item->spec;
6522         void *headers_m;
6523         void *headers_v;
6524         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6525         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6526         struct {
6527                 union {
6528                         __extension__
6529                         struct {
6530                                 uint16_t version:3;
6531                                 uint16_t rsvd0:9;
6532                                 uint16_t s_present:1;
6533                                 uint16_t k_present:1;
6534                                 uint16_t rsvd_bit1:1;
6535                                 uint16_t c_present:1;
6536                         };
6537                         uint16_t value;
6538                 };
6539         } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
6540
6541         if (inner) {
6542                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6543                                          inner_headers);
6544                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6545         } else {
6546                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6547                                          outer_headers);
6548                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6549         }
6550         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6551         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
6552         if (!gre_v)
6553                 return;
6554         if (!gre_m)
6555                 gre_m = &rte_flow_item_gre_mask;
6556         MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
6557                  rte_be_to_cpu_16(gre_m->protocol));
6558         MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6559                  rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
6560         gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
6561         gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
6562         MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
6563                  gre_crks_rsvd0_ver_m.c_present);
6564         MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
6565                  gre_crks_rsvd0_ver_v.c_present &
6566                  gre_crks_rsvd0_ver_m.c_present);
6567         MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
6568                  gre_crks_rsvd0_ver_m.k_present);
6569         MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
6570                  gre_crks_rsvd0_ver_v.k_present &
6571                  gre_crks_rsvd0_ver_m.k_present);
6572         MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
6573                  gre_crks_rsvd0_ver_m.s_present);
6574         MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
6575                  gre_crks_rsvd0_ver_v.s_present &
6576                  gre_crks_rsvd0_ver_m.s_present);
6577 }
6578
6579 /**
6580  * Add NVGRE item to matcher and to the value.
6581  *
6582  * @param[in, out] matcher
6583  *   Flow matcher.
6584  * @param[in, out] key
6585  *   Flow matcher value.
6586  * @param[in] item
6587  *   Flow pattern to translate.
6588  * @param[in] inner
6589  *   Item is inner pattern.
6590  */
6591 static void
6592 flow_dv_translate_item_nvgre(void *matcher, void *key,
6593                              const struct rte_flow_item *item,
6594                              int inner)
6595 {
6596         const struct rte_flow_item_nvgre *nvgre_m = item->mask;
6597         const struct rte_flow_item_nvgre *nvgre_v = item->spec;
6598         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6599         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6600         const char *tni_flow_id_m;
6601         const char *tni_flow_id_v;
6602         char *gre_key_m;
6603         char *gre_key_v;
6604         int size;
6605         int i;
6606
6607         /* For NVGRE, GRE header fields must be set with defined values. */
6608         const struct rte_flow_item_gre gre_spec = {
6609                 .c_rsvd0_ver = RTE_BE16(0x2000),
6610                 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6611         };
6612         const struct rte_flow_item_gre gre_mask = {
6613                 .c_rsvd0_ver = RTE_BE16(0xB000),
6614                 .protocol = RTE_BE16(UINT16_MAX),
6615         };
6616         const struct rte_flow_item gre_item = {
6617                 .spec = &gre_spec,
6618                 .mask = &gre_mask,
6619                 .last = NULL,
6620         };
6621         flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6622         if (!nvgre_v)
6623                 return;
6624         if (!nvgre_m)
6625                 nvgre_m = &rte_flow_item_nvgre_mask;
6626         tni_flow_id_m = (const char *)nvgre_m->tni;
6627         tni_flow_id_v = (const char *)nvgre_v->tni;
6628         size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
6629         gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
6630         gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
6631         memcpy(gre_key_m, tni_flow_id_m, size);
6632         for (i = 0; i < size; ++i)
6633                 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
6634 }
6635
6636 /**
6637  * Add VXLAN item to matcher and to the value.
6638  *
6639  * @param[in, out] matcher
6640  *   Flow matcher.
6641  * @param[in, out] key
6642  *   Flow matcher value.
6643  * @param[in] item
6644  *   Flow pattern to translate.
6645  * @param[in] inner
6646  *   Item is inner pattern.
6647  */
6648 static void
6649 flow_dv_translate_item_vxlan(void *matcher, void *key,
6650                              const struct rte_flow_item *item,
6651                              int inner)
6652 {
6653         const struct rte_flow_item_vxlan *vxlan_m = item->mask;
6654         const struct rte_flow_item_vxlan *vxlan_v = item->spec;
6655         void *headers_m;
6656         void *headers_v;
6657         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6658         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6659         char *vni_m;
6660         char *vni_v;
6661         uint16_t dport;
6662         int size;
6663         int i;
6664
6665         if (inner) {
6666                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6667                                          inner_headers);
6668                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6669         } else {
6670                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6671                                          outer_headers);
6672                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6673         }
6674         dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6675                 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6676         if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6677                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6678                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6679         }
6680         if (!vxlan_v)
6681                 return;
6682         if (!vxlan_m)
6683                 vxlan_m = &rte_flow_item_vxlan_mask;
6684         size = sizeof(vxlan_m->vni);
6685         vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
6686         vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
6687         memcpy(vni_m, vxlan_m->vni, size);
6688         for (i = 0; i < size; ++i)
6689                 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6690 }
6691
6692 /**
6693  * Add VXLAN-GPE item to matcher and to the value.
6694  *
6695  * @param[in, out] matcher
6696  *   Flow matcher.
6697  * @param[in, out] key
6698  *   Flow matcher value.
6699  * @param[in] item
6700  *   Flow pattern to translate.
6701  * @param[in] inner
6702  *   Item is inner pattern.
6703  */
6704
6705 static void
6706 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
6707                                  const struct rte_flow_item *item, int inner)
6708 {
6709         const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
6710         const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
6711         void *headers_m;
6712         void *headers_v;
6713         void *misc_m =
6714                 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
6715         void *misc_v =
6716                 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6717         char *vni_m;
6718         char *vni_v;
6719         uint16_t dport;
6720         int size;
6721         int i;
6722         uint8_t flags_m = 0xff;
6723         uint8_t flags_v = 0xc;
6724
6725         if (inner) {
6726                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6727                                          inner_headers);
6728                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6729         } else {
6730                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6731                                          outer_headers);
6732                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6733         }
6734         dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6735                 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6736         if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6737                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6738                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6739         }
6740         if (!vxlan_v)
6741                 return;
6742         if (!vxlan_m)
6743                 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
6744         size = sizeof(vxlan_m->vni);
6745         vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
6746         vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
6747         memcpy(vni_m, vxlan_m->vni, size);
6748         for (i = 0; i < size; ++i)
6749                 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6750         if (vxlan_m->flags) {
6751                 flags_m = vxlan_m->flags;
6752                 flags_v = vxlan_v->flags;
6753         }
6754         MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
6755         MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
6756         MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
6757                  vxlan_m->protocol);
6758         MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
6759                  vxlan_v->protocol);
6760 }
6761
6762 /**
6763  * Add Geneve item to matcher and to the value.
6764  *
6765  * @param[in, out] matcher
6766  *   Flow matcher.
6767  * @param[in, out] key
6768  *   Flow matcher value.
6769  * @param[in] item
6770  *   Flow pattern to translate.
6771  * @param[in] inner
6772  *   Item is inner pattern.
6773  */
6774
6775 static void
6776 flow_dv_translate_item_geneve(void *matcher, void *key,
6777                               const struct rte_flow_item *item, int inner)
6778 {
6779         const struct rte_flow_item_geneve *geneve_m = item->mask;
6780         const struct rte_flow_item_geneve *geneve_v = item->spec;
6781         void *headers_m;
6782         void *headers_v;
6783         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6784         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6785         uint16_t dport;
6786         uint16_t gbhdr_m;
6787         uint16_t gbhdr_v;
6788         char *vni_m;
6789         char *vni_v;
6790         size_t size, i;
6791
6792         if (inner) {
6793                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6794                                          inner_headers);
6795                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6796         } else {
6797                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6798                                          outer_headers);
6799                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6800         }
6801         dport = MLX5_UDP_PORT_GENEVE;
6802         if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6803                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6804                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6805         }
6806         if (!geneve_v)
6807                 return;
6808         if (!geneve_m)
6809                 geneve_m = &rte_flow_item_geneve_mask;
6810         size = sizeof(geneve_m->vni);
6811         vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
6812         vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
6813         memcpy(vni_m, geneve_m->vni, size);
6814         for (i = 0; i < size; ++i)
6815                 vni_v[i] = vni_m[i] & geneve_v->vni[i];
6816         MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
6817                  rte_be_to_cpu_16(geneve_m->protocol));
6818         MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
6819                  rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
6820         gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
6821         gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
6822         MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
6823                  MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6824         MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
6825                  MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6826         MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
6827                  MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6828         MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
6829                  MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
6830                  MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6831 }
6832
6833 /**
6834  * Add MPLS item to matcher and to the value.
6835  *
6836  * @param[in, out] matcher
6837  *   Flow matcher.
6838  * @param[in, out] key
6839  *   Flow matcher value.
6840  * @param[in] item
6841  *   Flow pattern to translate.
6842  * @param[in] prev_layer
6843  *   The protocol layer indicated in previous item.
6844  * @param[in] inner
6845  *   Item is inner pattern.
6846  */
6847 static void
6848 flow_dv_translate_item_mpls(void *matcher, void *key,
6849                             const struct rte_flow_item *item,
6850                             uint64_t prev_layer,
6851                             int inner)
6852 {
6853         const uint32_t *in_mpls_m = item->mask;
6854         const uint32_t *in_mpls_v = item->spec;
6855         uint32_t *out_mpls_m = 0;
6856         uint32_t *out_mpls_v = 0;
6857         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6858         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6859         void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
6860                                      misc_parameters_2);
6861         void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6862         void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
6863         void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6864
6865         switch (prev_layer) {
6866         case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6867                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
6868                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6869                          MLX5_UDP_PORT_MPLS);
6870                 break;
6871         case MLX5_FLOW_LAYER_GRE:
6872                 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
6873                 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6874                          RTE_ETHER_TYPE_MPLS);
6875                 break;
6876         default:
6877                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6878                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6879                          IPPROTO_MPLS);
6880                 break;
6881         }
6882         if (!in_mpls_v)
6883                 return;
6884         if (!in_mpls_m)
6885                 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
6886         switch (prev_layer) {
6887         case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6888                 out_mpls_m =
6889                         (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6890                                                  outer_first_mpls_over_udp);
6891                 out_mpls_v =
6892                         (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6893                                                  outer_first_mpls_over_udp);
6894                 break;
6895         case MLX5_FLOW_LAYER_GRE:
6896                 out_mpls_m =
6897                         (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6898                                                  outer_first_mpls_over_gre);
6899                 out_mpls_v =
6900                         (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6901                                                  outer_first_mpls_over_gre);
6902                 break;
6903         default:
6904                 /* Inner MPLS not over GRE is not supported. */
6905                 if (!inner) {
6906                         out_mpls_m =
6907                                 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6908                                                          misc2_m,
6909                                                          outer_first_mpls);
6910                         out_mpls_v =
6911                                 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6912                                                          misc2_v,
6913                                                          outer_first_mpls);
6914                 }
6915                 break;
6916         }
6917         if (out_mpls_m && out_mpls_v) {
6918                 *out_mpls_m = *in_mpls_m;
6919                 *out_mpls_v = *in_mpls_v & *in_mpls_m;
6920         }
6921 }
6922
6923 /**
6924  * Add metadata register item to matcher
6925  *
6926  * @param[in, out] matcher
6927  *   Flow matcher.
6928  * @param[in, out] key
6929  *   Flow matcher value.
6930  * @param[in] reg_type
6931  *   Type of device metadata register
6932  * @param[in] value
6933  *   Register value
6934  * @param[in] mask
6935  *   Register mask
6936  */
6937 static void
6938 flow_dv_match_meta_reg(void *matcher, void *key,
6939                        enum modify_reg reg_type,
6940                        uint32_t data, uint32_t mask)
6941 {
6942         void *misc2_m =
6943                 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
6944         void *misc2_v =
6945                 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6946         uint32_t temp;
6947
6948         data &= mask;
6949         switch (reg_type) {
6950         case REG_A:
6951                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
6952                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
6953                 break;
6954         case REG_B:
6955                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
6956                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
6957                 break;
6958         case REG_C_0:
6959                 /*
6960                  * The metadata register C0 field might be divided into
6961                  * source vport index and META item value, we should set
6962                  * this field according to specified mask, not as whole one.
6963                  */
6964                 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
6965                 temp |= mask;
6966                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
6967                 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
6968                 temp &= ~mask;
6969                 temp |= data;
6970                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
6971                 break;
6972         case REG_C_1:
6973                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
6974                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
6975                 break;
6976         case REG_C_2:
6977                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
6978                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
6979                 break;
6980         case REG_C_3:
6981                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
6982                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
6983                 break;
6984         case REG_C_4:
6985                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
6986                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
6987                 break;
6988         case REG_C_5:
6989                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
6990                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
6991                 break;
6992         case REG_C_6:
6993                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
6994                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
6995                 break;
6996         case REG_C_7:
6997                 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
6998                 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
6999                 break;
7000         default:
7001                 MLX5_ASSERT(false);
7002                 break;
7003         }
7004 }
7005
7006 /**
7007  * Add MARK item to matcher
7008  *
7009  * @param[in] dev
7010  *   The device to configure through.
7011  * @param[in, out] matcher
7012  *   Flow matcher.
7013  * @param[in, out] key
7014  *   Flow matcher value.
7015  * @param[in] item
7016  *   Flow pattern to translate.
7017  */
7018 static void
7019 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
7020                             void *matcher, void *key,
7021                             const struct rte_flow_item *item)
7022 {
7023         struct mlx5_priv *priv = dev->data->dev_private;
7024         const struct rte_flow_item_mark *mark;
7025         uint32_t value;
7026         uint32_t mask;
7027
7028         mark = item->mask ? (const void *)item->mask :
7029                             &rte_flow_item_mark_mask;
7030         mask = mark->id & priv->sh->dv_mark_mask;
7031         mark = (const void *)item->spec;
7032         MLX5_ASSERT(mark);
7033         value = mark->id & priv->sh->dv_mark_mask & mask;
7034         if (mask) {
7035                 enum modify_reg reg;
7036
7037                 /* Get the metadata register index for the mark. */
7038                 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
7039                 MLX5_ASSERT(reg > 0);
7040                 if (reg == REG_C_0) {
7041                         struct mlx5_priv *priv = dev->data->dev_private;
7042                         uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7043                         uint32_t shl_c0 = rte_bsf32(msk_c0);
7044
7045                         mask &= msk_c0;
7046                         mask <<= shl_c0;
7047                         value <<= shl_c0;
7048                 }
7049                 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7050         }
7051 }
7052
7053 /**
7054  * Add META item to matcher
7055  *
7056  * @param[in] dev
7057  *   The devich to configure through.
7058  * @param[in, out] matcher
7059  *   Flow matcher.
7060  * @param[in, out] key
7061  *   Flow matcher value.
7062  * @param[in] attr
7063  *   Attributes of flow that includes this item.
7064  * @param[in] item
7065  *   Flow pattern to translate.
7066  */
7067 static void
7068 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
7069                             void *matcher, void *key,
7070                             const struct rte_flow_attr *attr,
7071                             const struct rte_flow_item *item)
7072 {
7073         const struct rte_flow_item_meta *meta_m;
7074         const struct rte_flow_item_meta *meta_v;
7075
7076         meta_m = (const void *)item->mask;
7077         if (!meta_m)
7078                 meta_m = &rte_flow_item_meta_mask;
7079         meta_v = (const void *)item->spec;
7080         if (meta_v) {
7081                 int reg;
7082                 uint32_t value = meta_v->data;
7083                 uint32_t mask = meta_m->data;
7084
7085                 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
7086                 if (reg < 0)
7087                         return;
7088                 /*
7089                  * In datapath code there is no endianness
7090                  * coversions for perfromance reasons, all
7091                  * pattern conversions are done in rte_flow.
7092                  */
7093                 value = rte_cpu_to_be_32(value);
7094                 mask = rte_cpu_to_be_32(mask);
7095                 if (reg == REG_C_0) {
7096                         struct mlx5_priv *priv = dev->data->dev_private;
7097                         uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7098                         uint32_t shl_c0 = rte_bsf32(msk_c0);
7099 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
7100                         uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
7101
7102                         value >>= shr_c0;
7103                         mask >>= shr_c0;
7104 #endif
7105                         value <<= shl_c0;
7106                         mask <<= shl_c0;
7107                         MLX5_ASSERT(msk_c0);
7108                         MLX5_ASSERT(!(~msk_c0 & mask));
7109                 }
7110                 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7111         }
7112 }
7113
7114 /**
7115  * Add vport metadata Reg C0 item to matcher
7116  *
7117  * @param[in, out] matcher
7118  *   Flow matcher.
7119  * @param[in, out] key
7120  *   Flow matcher value.
7121  * @param[in] reg
7122  *   Flow pattern to translate.
7123  */
7124 static void
7125 flow_dv_translate_item_meta_vport(void *matcher, void *key,
7126                                   uint32_t value, uint32_t mask)
7127 {
7128         flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
7129 }
7130
7131 /**
7132  * Add tag item to matcher
7133  *
7134  * @param[in] dev
7135  *   The devich to configure through.
7136  * @param[in, out] matcher
7137  *   Flow matcher.
7138  * @param[in, out] key
7139  *   Flow matcher value.
7140  * @param[in] item
7141  *   Flow pattern to translate.
7142  */
7143 static void
7144 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
7145                                 void *matcher, void *key,
7146                                 const struct rte_flow_item *item)
7147 {
7148         const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
7149         const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
7150         uint32_t mask, value;
7151
7152         MLX5_ASSERT(tag_v);
7153         value = tag_v->data;
7154         mask = tag_m ? tag_m->data : UINT32_MAX;
7155         if (tag_v->id == REG_C_0) {
7156                 struct mlx5_priv *priv = dev->data->dev_private;
7157                 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7158                 uint32_t shl_c0 = rte_bsf32(msk_c0);
7159
7160                 mask &= msk_c0;
7161                 mask <<= shl_c0;
7162                 value <<= shl_c0;
7163         }
7164         flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
7165 }
7166
7167 /**
7168  * Add TAG item to matcher
7169  *
7170  * @param[in] dev
7171  *   The devich to configure through.
7172  * @param[in, out] matcher
7173  *   Flow matcher.
7174  * @param[in, out] key
7175  *   Flow matcher value.
7176  * @param[in] item
7177  *   Flow pattern to translate.
7178  */
7179 static void
7180 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
7181                            void *matcher, void *key,
7182                            const struct rte_flow_item *item)
7183 {
7184         const struct rte_flow_item_tag *tag_v = item->spec;
7185         const struct rte_flow_item_tag *tag_m = item->mask;
7186         enum modify_reg reg;
7187
7188         MLX5_ASSERT(tag_v);
7189         tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
7190         /* Get the metadata register index for the tag. */
7191         reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
7192         MLX5_ASSERT(reg > 0);
7193         flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
7194 }
7195
7196 /**
7197  * Add source vport match to the specified matcher.
7198  *
7199  * @param[in, out] matcher
7200  *   Flow matcher.
7201  * @param[in, out] key
7202  *   Flow matcher value.
7203  * @param[in] port
7204  *   Source vport value to match
7205  * @param[in] mask
7206  *   Mask
7207  */
7208 static void
7209 flow_dv_translate_item_source_vport(void *matcher, void *key,
7210                                     int16_t port, uint16_t mask)
7211 {
7212         void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7213         void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7214
7215         MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
7216         MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
7217 }
7218
7219 /**
7220  * Translate port-id item to eswitch match on  port-id.
7221  *
7222  * @param[in] dev
7223  *   The devich to configure through.
7224  * @param[in, out] matcher
7225  *   Flow matcher.
7226  * @param[in, out] key
7227  *   Flow matcher value.
7228  * @param[in] item
7229  *   Flow pattern to translate.
7230  *
7231  * @return
7232  *   0 on success, a negative errno value otherwise.
7233  */
7234 static int
7235 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
7236                                void *key, const struct rte_flow_item *item)
7237 {
7238         const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
7239         const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
7240         struct mlx5_priv *priv;
7241         uint16_t mask, id;
7242
7243         mask = pid_m ? pid_m->id : 0xffff;
7244         id = pid_v ? pid_v->id : dev->data->port_id;
7245         priv = mlx5_port_to_eswitch_info(id, item == NULL);
7246         if (!priv)
7247                 return -rte_errno;
7248         /* Translate to vport field or to metadata, depending on mode. */
7249         if (priv->vport_meta_mask)
7250                 flow_dv_translate_item_meta_vport(matcher, key,
7251                                                   priv->vport_meta_tag,
7252                                                   priv->vport_meta_mask);
7253         else
7254                 flow_dv_translate_item_source_vport(matcher, key,
7255                                                     priv->vport_id, mask);
7256         return 0;
7257 }
7258
7259 /**
7260  * Add ICMP6 item to matcher and to the value.
7261  *
7262  * @param[in, out] matcher
7263  *   Flow matcher.
7264  * @param[in, out] key
7265  *   Flow matcher value.
7266  * @param[in] item
7267  *   Flow pattern to translate.
7268  * @param[in] inner
7269  *   Item is inner pattern.
7270  */
7271 static void
7272 flow_dv_translate_item_icmp6(void *matcher, void *key,
7273                               const struct rte_flow_item *item,
7274                               int inner)
7275 {
7276         const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
7277         const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
7278         void *headers_m;
7279         void *headers_v;
7280         void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7281                                      misc_parameters_3);
7282         void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7283         if (inner) {
7284                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7285                                          inner_headers);
7286                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7287         } else {
7288                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7289                                          outer_headers);
7290                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7291         }
7292         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7293         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
7294         if (!icmp6_v)
7295                 return;
7296         if (!icmp6_m)
7297                 icmp6_m = &rte_flow_item_icmp6_mask;
7298         /*
7299          * Force flow only to match the non-fragmented IPv6 ICMPv6 packets.
7300          * If only the protocol is specified, no need to match the frag.
7301          */
7302         MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
7303         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
7304         MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
7305         MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
7306                  icmp6_v->type & icmp6_m->type);
7307         MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
7308         MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
7309                  icmp6_v->code & icmp6_m->code);
7310 }
7311
7312 /**
7313  * Add ICMP item to matcher and to the value.
7314  *
7315  * @param[in, out] matcher
7316  *   Flow matcher.
7317  * @param[in, out] key
7318  *   Flow matcher value.
7319  * @param[in] item
7320  *   Flow pattern to translate.
7321  * @param[in] inner
7322  *   Item is inner pattern.
7323  */
7324 static void
7325 flow_dv_translate_item_icmp(void *matcher, void *key,
7326                             const struct rte_flow_item *item,
7327                             int inner)
7328 {
7329         const struct rte_flow_item_icmp *icmp_m = item->mask;
7330         const struct rte_flow_item_icmp *icmp_v = item->spec;
7331         void *headers_m;
7332         void *headers_v;
7333         void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7334                                      misc_parameters_3);
7335         void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7336         if (inner) {
7337                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7338                                          inner_headers);
7339                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7340         } else {
7341                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7342                                          outer_headers);
7343                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7344         }
7345         MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7346         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
7347         if (!icmp_v)
7348                 return;
7349         if (!icmp_m)
7350                 icmp_m = &rte_flow_item_icmp_mask;
7351         /*
7352          * Force flow only to match the non-fragmented IPv4 ICMP packets.
7353          * If only the protocol is specified, no need to match the frag.
7354          */
7355         MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
7356         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
7357         MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
7358                  icmp_m->hdr.icmp_type);
7359         MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
7360                  icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
7361         MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
7362                  icmp_m->hdr.icmp_code);
7363         MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
7364                  icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
7365 }
7366
7367 /**
7368  * Add GTP item to matcher and to the value.
7369  *
7370  * @param[in, out] matcher
7371  *   Flow matcher.
7372  * @param[in, out] key
7373  *   Flow matcher value.
7374  * @param[in] item
7375  *   Flow pattern to translate.
7376  * @param[in] inner
7377  *   Item is inner pattern.
7378  */
7379 static void
7380 flow_dv_translate_item_gtp(void *matcher, void *key,
7381                            const struct rte_flow_item *item, int inner)
7382 {
7383         const struct rte_flow_item_gtp *gtp_m = item->mask;
7384         const struct rte_flow_item_gtp *gtp_v = item->spec;
7385         void *headers_m;
7386         void *headers_v;
7387         void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7388                                      misc_parameters_3);
7389         void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7390         uint16_t dport = RTE_GTPU_UDP_PORT;
7391
7392         if (inner) {
7393                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7394                                          inner_headers);
7395                 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7396         } else {
7397                 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7398                                          outer_headers);
7399                 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7400         }
7401         if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7402                 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7403                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7404         }
7405         if (!gtp_v)
7406                 return;
7407         if (!gtp_m)
7408                 gtp_m = &rte_flow_item_gtp_mask;
7409         MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
7410                  gtp_m->v_pt_rsv_flags);
7411         MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
7412                  gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
7413         MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
7414         MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
7415                  gtp_v->msg_type & gtp_m->msg_type);
7416         MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
7417                  rte_be_to_cpu_32(gtp_m->teid));
7418         MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
7419                  rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
7420 }
7421
7422 /**
7423  * Add eCPRI item to matcher and to the value.
7424  *
7425  * @param[in] dev
7426  *   The devich to configure through.
7427  * @param[in, out] matcher
7428  *   Flow matcher.
7429  * @param[in, out] key
7430  *   Flow matcher value.
7431  * @param[in] item
7432  *   Flow pattern to translate.
7433  * @param[in] samples
7434  *   Sample IDs to be used in the matching.
7435  */
7436 static void
7437 flow_dv_translate_item_ecpri(struct rte_eth_dev *dev, void *matcher,
7438                              void *key, const struct rte_flow_item *item)
7439 {
7440         struct mlx5_priv *priv = dev->data->dev_private;
7441         const struct rte_flow_item_ecpri *ecpri_m = item->mask;
7442         const struct rte_flow_item_ecpri *ecpri_v = item->spec;
7443         void *misc4_m = MLX5_ADDR_OF(fte_match_param, matcher,
7444                                      misc_parameters_4);
7445         void *misc4_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_4);
7446         uint32_t *samples;
7447         void *dw_m;
7448         void *dw_v;
7449
7450         if (!ecpri_v)
7451                 return;
7452         if (!ecpri_m)
7453                 ecpri_m = &rte_flow_item_ecpri_mask;
7454         /*
7455          * Maximal four DW samples are supported in a single matching now.
7456          * Two are used now for a eCPRI matching:
7457          * 1. Type: one byte, mask should be 0x00ff0000 in network order
7458          * 2. ID of a message: one or two bytes, mask 0xffff0000 or 0xff000000
7459          *    if any.
7460          */
7461         if (!ecpri_m->hdr.common.u32)
7462                 return;
7463         samples = priv->sh->fp[MLX5_FLEX_PARSER_ECPRI_0].ids;
7464         /* Need to take the whole DW as the mask to fill the entry. */
7465         dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7466                             prog_sample_field_value_0);
7467         dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7468                             prog_sample_field_value_0);
7469         /* Already big endian (network order) in the header. */
7470         *(uint32_t *)dw_m = ecpri_m->hdr.common.u32;
7471         *(uint32_t *)dw_v = ecpri_v->hdr.common.u32;
7472         /* Sample#0, used for matching type, offset 0. */
7473         MLX5_SET(fte_match_set_misc4, misc4_m,
7474                  prog_sample_field_id_0, samples[0]);
7475         /* It makes no sense to set the sample ID in the mask field. */
7476         MLX5_SET(fte_match_set_misc4, misc4_v,
7477                  prog_sample_field_id_0, samples[0]);
7478         /*
7479          * Checking if message body part needs to be matched.
7480          * Some wildcard rules only matching type field should be supported.
7481          */
7482         if (ecpri_m->hdr.dummy[0]) {
7483                 switch (ecpri_v->hdr.common.type) {
7484                 case RTE_ECPRI_MSG_TYPE_IQ_DATA:
7485                 case RTE_ECPRI_MSG_TYPE_RTC_CTRL:
7486                 case RTE_ECPRI_MSG_TYPE_DLY_MSR:
7487                         dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7488                                             prog_sample_field_value_1);
7489                         dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7490                                             prog_sample_field_value_1);
7491                         *(uint32_t *)dw_m = ecpri_m->hdr.dummy[0];
7492                         *(uint32_t *)dw_v = ecpri_v->hdr.dummy[0];
7493                         /* Sample#1, to match message body, offset 4. */
7494                         MLX5_SET(fte_match_set_misc4, misc4_m,
7495                                  prog_sample_field_id_1, samples[1]);
7496                         MLX5_SET(fte_match_set_misc4, misc4_v,
7497                                  prog_sample_field_id_1, samples[1]);
7498                         break;
7499                 default:
7500                         /* Others, do not match any sample ID. */
7501                         break;
7502                 }
7503         }
7504 }
7505
7506 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
7507
7508 #define HEADER_IS_ZERO(match_criteria, headers)                              \
7509         !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers),     \
7510                  matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
7511
7512 /**
7513  * Calculate flow matcher enable bitmap.
7514  *
7515  * @param match_criteria
7516  *   Pointer to flow matcher criteria.
7517  *
7518  * @return
7519  *   Bitmap of enabled fields.
7520  */
7521 static uint8_t
7522 flow_dv_matcher_enable(uint32_t *match_criteria)
7523 {
7524         uint8_t match_criteria_enable;
7525
7526         match_criteria_enable =
7527                 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
7528                 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
7529         match_criteria_enable |=
7530                 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
7531                 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
7532         match_criteria_enable |=
7533                 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
7534                 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
7535         match_criteria_enable |=
7536                 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
7537                 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
7538         match_criteria_enable |=
7539                 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
7540                 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
7541         match_criteria_enable |=
7542                 (!HEADER_IS_ZERO(match_criteria, misc_parameters_4)) <<
7543                 MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT;
7544         return match_criteria_enable;
7545 }
7546
7547
7548 /**
7549  * Get a flow table.
7550  *
7551  * @param[in, out] dev
7552  *   Pointer to rte_eth_dev structure.
7553  * @param[in] table_id
7554  *   Table id to use.
7555  * @param[in] egress
7556  *   Direction of the table.
7557  * @param[in] transfer
7558  *   E-Switch or NIC flow.
7559  * @param[out] error
7560  *   pointer to error structure.
7561  *
7562  * @return
7563  *   Returns tables resource based on the index, NULL in case of failed.
7564  */
7565 static struct mlx5_flow_tbl_resource *
7566 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
7567                          uint32_t table_id, uint8_t egress,
7568                          uint8_t transfer,
7569                          struct rte_flow_error *error)
7570 {
7571         struct mlx5_priv *priv = dev->data->dev_private;
7572         struct mlx5_dev_ctx_shared *sh = priv->sh;
7573         struct mlx5_flow_tbl_resource *tbl;
7574         union mlx5_flow_tbl_key table_key = {
7575                 {
7576                         .table_id = table_id,
7577                         .reserved = 0,
7578                         .domain = !!transfer,
7579                         .direction = !!egress,
7580                 }
7581         };
7582         struct mlx5_hlist_entry *pos = mlx5_hlist_lookup(sh->flow_tbls,
7583                                                          table_key.v64);
7584         struct mlx5_flow_tbl_data_entry *tbl_data;
7585         uint32_t idx = 0;
7586         int ret;
7587         void *domain;
7588
7589         if (pos) {
7590                 tbl_data = container_of(pos, struct mlx5_flow_tbl_data_entry,
7591                                         entry);
7592                 tbl = &tbl_data->tbl;
7593                 rte_atomic32_inc(&tbl->refcnt);
7594                 return tbl;
7595         }
7596         tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
7597         if (!tbl_data) {
7598                 rte_flow_error_set(error, ENOMEM,
7599                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7600                                    NULL,
7601                                    "cannot allocate flow table data entry");
7602                 return NULL;
7603         }
7604         tbl_data->idx = idx;
7605         tbl = &tbl_data->tbl;
7606         pos = &tbl_data->entry;
7607         if (transfer)
7608                 domain = sh->fdb_domain;
7609         else if (egress)
7610                 domain = sh->tx_domain;
7611         else
7612                 domain = sh->rx_domain;
7613         ret = mlx5_flow_os_create_flow_tbl(domain, table_id, &tbl->obj);
7614         if (ret) {
7615                 rte_flow_error_set(error, ENOMEM,
7616                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7617                                    NULL, "cannot create flow table object");
7618                 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7619                 return NULL;
7620         }
7621         /*
7622          * No multi-threads now, but still better to initialize the reference
7623          * count before insert it into the hash list.
7624          */
7625         rte_atomic32_init(&tbl->refcnt);
7626         /* Jump action reference count is initialized here. */
7627         rte_atomic32_init(&tbl_data->jump.refcnt);
7628         pos->key = table_key.v64;
7629         ret = mlx5_hlist_insert(sh->flow_tbls, pos);
7630         if (ret < 0) {
7631                 rte_flow_error_set(error, -ret,
7632                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7633                                    "cannot insert flow table data entry");
7634                 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7635                 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7636         }
7637         rte_atomic32_inc(&tbl->refcnt);
7638         return tbl;
7639 }
7640
7641 /**
7642  * Release a flow table.
7643  *
7644  * @param[in] dev
7645  *   Pointer to rte_eth_dev structure.
7646  * @param[in] tbl
7647  *   Table resource to be released.
7648  *
7649  * @return
7650  *   Returns 0 if table was released, else return 1;
7651  */
7652 static int
7653 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
7654                              struct mlx5_flow_tbl_resource *tbl)
7655 {
7656         struct mlx5_priv *priv = dev->data->dev_private;
7657         struct mlx5_dev_ctx_shared *sh = priv->sh;
7658         struct mlx5_flow_tbl_data_entry *tbl_data =
7659                 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7660
7661         if (!tbl)
7662                 return 0;
7663         if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
7664                 struct mlx5_hlist_entry *pos = &tbl_data->entry;
7665
7666                 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7667                 tbl->obj = NULL;
7668                 /* remove the entry from the hash list and free memory. */
7669                 mlx5_hlist_remove(sh->flow_tbls, pos);
7670                 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_JUMP],
7671                                 tbl_data->idx);
7672                 return 0;
7673         }
7674         return 1;
7675 }
7676
7677 /**
7678  * Register the flow matcher.
7679  *
7680  * @param[in, out] dev
7681  *   Pointer to rte_eth_dev structure.
7682  * @param[in, out] matcher
7683  *   Pointer to flow matcher.
7684  * @param[in, out] key
7685  *   Pointer to flow table key.
7686  * @parm[in, out] dev_flow
7687  *   Pointer to the dev_flow.
7688  * @param[out] error
7689  *   pointer to error structure.
7690  *
7691  * @return
7692  *   0 on success otherwise -errno and errno is set.
7693  */
7694 static int
7695 flow_dv_matcher_register(struct rte_eth_dev *dev,
7696                          struct mlx5_flow_dv_matcher *matcher,
7697                          union mlx5_flow_tbl_key *key,
7698                          struct mlx5_flow *dev_flow,
7699                          struct rte_flow_error *error)
7700 {
7701         struct mlx5_priv *priv = dev->data->dev_private;
7702         struct mlx5_dev_ctx_shared *sh = priv->sh;
7703         struct mlx5_flow_dv_matcher *cache_matcher;
7704         struct mlx5dv_flow_matcher_attr dv_attr = {
7705                 .type = IBV_FLOW_ATTR_NORMAL,
7706                 .match_mask = (void *)&matcher->mask,
7707         };
7708         struct mlx5_flow_tbl_resource *tbl;
7709         struct mlx5_flow_tbl_data_entry *tbl_data;
7710         int ret;
7711
7712         tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
7713                                        key->domain, error);
7714         if (!tbl)
7715                 return -rte_errno;      /* No need to refill the error info */
7716         tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7717         /* Lookup from cache. */
7718         LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
7719                 if (matcher->crc == cache_matcher->crc &&
7720                     matcher->priority == cache_matcher->priority &&
7721                     !memcmp((const void *)matcher->mask.buf,
7722                             (const void *)cache_matcher->mask.buf,
7723                             cache_matcher->mask.size)) {
7724                         DRV_LOG(DEBUG,
7725                                 "%s group %u priority %hd use %s "
7726                                 "matcher %p: refcnt %d++",
7727                                 key->domain ? "FDB" : "NIC", key->table_id,
7728                                 cache_matcher->priority,
7729                                 key->direction ? "tx" : "rx",
7730                                 (void *)cache_matcher,
7731                                 rte_atomic32_read(&cache_matcher->refcnt));
7732                         rte_atomic32_inc(&cache_matcher->refcnt);
7733                         dev_flow->handle->dvh.matcher = cache_matcher;
7734                         /* old matcher should not make the table ref++. */
7735                         flow_dv_tbl_resource_release(dev, tbl);
7736                         return 0;
7737                 }
7738         }
7739         /* Register new matcher. */
7740         cache_matcher = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*cache_matcher), 0,
7741                                     SOCKET_ID_ANY);
7742         if (!cache_matcher) {
7743                 flow_dv_tbl_resource_release(dev, tbl);
7744                 return rte_flow_error_set(error, ENOMEM,
7745                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7746                                           "cannot allocate matcher memory");
7747         }
7748         *cache_matcher = *matcher;
7749         dv_attr.match_criteria_enable =
7750                 flow_dv_matcher_enable(cache_matcher->mask.buf);
7751         dv_attr.priority = matcher->priority;
7752         if (key->direction)
7753                 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
7754         ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj,
7755                                                &cache_matcher->matcher_object);
7756         if (ret) {
7757                 mlx5_free(cache_matcher);
7758 #ifdef HAVE_MLX5DV_DR
7759                 flow_dv_tbl_resource_release(dev, tbl);
7760 #endif
7761                 return rte_flow_error_set(error, ENOMEM,
7762                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7763                                           NULL, "cannot create matcher");
7764         }
7765         /* Save the table information */
7766         cache_matcher->tbl = tbl;
7767         rte_atomic32_init(&cache_matcher->refcnt);
7768         /* only matcher ref++, table ref++ already done above in get API. */
7769         rte_atomic32_inc(&cache_matcher->refcnt);
7770         LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
7771         dev_flow->handle->dvh.matcher = cache_matcher;
7772         DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
7773                 key->domain ? "FDB" : "NIC", key->table_id,
7774                 cache_matcher->priority,
7775                 key->direction ? "tx" : "rx", (void *)cache_matcher,
7776                 rte_atomic32_read(&cache_matcher->refcnt));
7777         return 0;
7778 }
7779
7780 /**
7781  * Find existing tag resource or create and register a new one.
7782  *
7783  * @param dev[in, out]
7784  *   Pointer to rte_eth_dev structure.
7785  * @param[in, out] tag_be24
7786  *   Tag value in big endian then R-shift 8.
7787  * @parm[in, out] dev_flow
7788  *   Pointer to the dev_flow.
7789  * @param[out] error
7790  *   pointer to error structure.
7791  *
7792  * @return
7793  *   0 on success otherwise -errno and errno is set.
7794  */
7795 static int
7796 flow_dv_tag_resource_register
7797                         (struct rte_eth_dev *dev,
7798                          uint32_t tag_be24,
7799                          struct mlx5_flow *dev_flow,
7800                          struct rte_flow_error *error)
7801 {
7802         struct mlx5_priv *priv = dev->data->dev_private;
7803         struct mlx5_dev_ctx_shared *sh = priv->sh;
7804         struct mlx5_flow_dv_tag_resource *cache_resource;
7805         struct mlx5_hlist_entry *entry;
7806         int ret;
7807
7808         /* Lookup a matching resource from cache. */
7809         entry = mlx5_hlist_lookup(sh->tag_table, (uint64_t)tag_be24);
7810         if (entry) {
7811                 cache_resource = container_of
7812                         (entry, struct mlx5_flow_dv_tag_resource, entry);
7813                 rte_atomic32_inc(&cache_resource->refcnt);
7814                 dev_flow->handle->dvh.rix_tag = cache_resource->idx;
7815                 dev_flow->dv.tag_resource = cache_resource;
7816                 DRV_LOG(DEBUG, "cached tag resource %p: refcnt now %d++",
7817                         (void *)cache_resource,
7818                         rte_atomic32_read(&cache_resource->refcnt));
7819                 return 0;
7820         }
7821         /* Register new resource. */
7822         cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG],
7823                                        &dev_flow->handle->dvh.rix_tag);
7824         if (!cache_resource)
7825                 return rte_flow_error_set(error, ENOMEM,
7826                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7827                                           "cannot allocate resource memory");
7828         cache_resource->entry.key = (uint64_t)tag_be24;
7829         ret = mlx5_flow_os_create_flow_action_tag(tag_be24,
7830                                                   &cache_resource->action);
7831         if (ret) {
7832                 mlx5_free(cache_resource);
7833                 return rte_flow_error_set(error, ENOMEM,
7834                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7835                                           NULL, "cannot create action");
7836         }
7837         rte_atomic32_init(&cache_resource->refcnt);
7838         rte_atomic32_inc(&cache_resource->refcnt);
7839         if (mlx5_hlist_insert(sh->tag_table, &cache_resource->entry)) {
7840                 mlx5_flow_os_destroy_flow_action(cache_resource->action);
7841                 mlx5_free(cache_resource);
7842                 return rte_flow_error_set(error, EEXIST,
7843                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7844                                           NULL, "cannot insert tag");
7845         }
7846         dev_flow->dv.tag_resource = cache_resource;
7847         DRV_LOG(DEBUG, "new tag resource %p: refcnt now %d++",
7848                 (void *)cache_resource,
7849                 rte_atomic32_read(&cache_resource->refcnt));
7850         return 0;
7851 }
7852
7853 /**
7854  * Release the tag.
7855  *
7856  * @param dev
7857  *   Pointer to Ethernet device.
7858  * @param tag_idx
7859  *   Tag index.
7860  *
7861  * @return
7862  *   1 while a reference on it exists, 0 when freed.
7863  */
7864 static int
7865 flow_dv_tag_release(struct rte_eth_dev *dev,
7866                     uint32_t tag_idx)
7867 {
7868         struct mlx5_priv *priv = dev->data->dev_private;
7869         struct mlx5_dev_ctx_shared *sh = priv->sh;
7870         struct mlx5_flow_dv_tag_resource *tag;
7871
7872         tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7873         if (!tag)
7874                 return 0;
7875         DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
7876                 dev->data->port_id, (void *)tag,
7877                 rte_atomic32_read(&tag->refcnt));
7878         if (rte_atomic32_dec_and_test(&tag->refcnt)) {
7879                 claim_zero(mlx5_flow_os_destroy_flow_action(tag->action));
7880                 mlx5_hlist_remove(sh->tag_table, &tag->entry);
7881                 DRV_LOG(DEBUG, "port %u tag %p: removed",
7882                         dev->data->port_id, (void *)tag);
7883                 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7884                 return 0;
7885         }
7886         return 1;
7887 }
7888
7889 /**
7890  * Translate port ID action to vport.
7891  *
7892  * @param[in] dev
7893  *   Pointer to rte_eth_dev structure.
7894  * @param[in] action
7895  *   Pointer to the port ID action.
7896  * @param[out] dst_port_id
7897  *   The target port ID.
7898  * @param[out] error
7899  *   Pointer to the error structure.
7900  *
7901  * @return
7902  *   0 on success, a negative errno value otherwise and rte_errno is set.
7903  */
7904 static int
7905 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
7906                                  const struct rte_flow_action *action,
7907                                  uint32_t *dst_port_id,
7908                                  struct rte_flow_error *error)
7909 {
7910         uint32_t port;
7911         struct mlx5_priv *priv;
7912         const struct rte_flow_action_port_id *conf =
7913                         (const struct rte_flow_action_port_id *)action->conf;
7914
7915         port = conf->original ? dev->data->port_id : conf->id;
7916         priv = mlx5_port_to_eswitch_info(port, false);
7917         if (!priv)
7918                 return rte_flow_error_set(error, -rte_errno,
7919                                           RTE_FLOW_ERROR_TYPE_ACTION,
7920                                           NULL,
7921                                           "No eswitch info was found for port");
7922 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
7923         /*
7924          * This parameter is transferred to
7925          * mlx5dv_dr_action_create_dest_ib_port().
7926          */
7927         *dst_port_id = priv->dev_port;
7928 #else
7929         /*
7930          * Legacy mode, no LAG configurations is supported.
7931          * This parameter is transferred to
7932          * mlx5dv_dr_action_create_dest_vport().
7933          */
7934         *dst_port_id = priv->vport_id;
7935 #endif
7936         return 0;
7937 }
7938
7939 /**
7940  * Create a counter with aging configuration.
7941  *
7942  * @param[in] dev
7943  *   Pointer to rte_eth_dev structure.
7944  * @param[out] count
7945  *   Pointer to the counter action configuration.
7946  * @param[in] age
7947  *   Pointer to the aging action configuration.
7948  *
7949  * @return
7950  *   Index to flow counter on success, 0 otherwise.
7951  */
7952 static uint32_t
7953 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
7954                                 struct mlx5_flow *dev_flow,
7955                                 const struct rte_flow_action_count *count,
7956                                 const struct rte_flow_action_age *age)
7957 {
7958         uint32_t counter;
7959         struct mlx5_age_param *age_param;
7960
7961         counter = flow_dv_counter_alloc(dev,
7962                                 count ? count->shared : 0,
7963                                 count ? count->id : 0,
7964                                 dev_flow->dv.group, !!age);
7965         if (!counter || age == NULL)
7966                 return counter;
7967         age_param  = flow_dv_counter_idx_get_age(dev, counter);
7968         /*
7969          * The counter age accuracy may have a bit delay. Have 3/4
7970          * second bias on the timeount in order to let it age in time.
7971          */
7972         age_param->context = age->context ? age->context :
7973                 (void *)(uintptr_t)(dev_flow->flow_idx);
7974         /*
7975          * The counter age accuracy may have a bit delay. Have 3/4
7976          * second bias on the timeount in order to let it age in time.
7977          */
7978         age_param->timeout = age->timeout * 10 - MLX5_AGING_TIME_DELAY;
7979         /* Set expire time in unit of 0.1 sec. */
7980         age_param->port_id = dev->data->port_id;
7981         age_param->expire = age_param->timeout +
7982                         rte_rdtsc() / (rte_get_tsc_hz() / 10);
7983         rte_atomic16_set(&age_param->state, AGE_CANDIDATE);
7984         return counter;
7985 }
7986 /**
7987  * Add Tx queue matcher
7988  *
7989  * @param[in] dev
7990  *   Pointer to the dev struct.
7991  * @param[in, out] matcher
7992  *   Flow matcher.
7993  * @param[in, out] key
7994  *   Flow matcher value.
7995  * @param[in] item
7996  *   Flow pattern to translate.
7997  * @param[in] inner
7998  *   Item is inner pattern.
7999  */
8000 static void
8001 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
8002                                 void *matcher, void *key,
8003                                 const struct rte_flow_item *item)
8004 {
8005         const struct mlx5_rte_flow_item_tx_queue *queue_m;
8006         const struct mlx5_rte_flow_item_tx_queue *queue_v;
8007         void *misc_m =
8008                 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8009         void *misc_v =
8010                 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8011         struct mlx5_txq_ctrl *txq;
8012         uint32_t queue;
8013
8014
8015         queue_m = (const void *)item->mask;
8016         if (!queue_m)
8017                 return;
8018         queue_v = (const void *)item->spec;
8019         if (!queue_v)
8020                 return;
8021         txq = mlx5_txq_get(dev, queue_v->queue);
8022         if (!txq)
8023                 return;
8024         queue = txq->obj->sq->id;
8025         MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
8026         MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
8027                  queue & queue_m->queue);
8028         mlx5_txq_release(dev, queue_v->queue);
8029 }
8030
8031 /**
8032  * Set the hash fields according to the @p flow information.
8033  *
8034  * @param[in] dev_flow
8035  *   Pointer to the mlx5_flow.
8036  * @param[in] rss_desc
8037  *   Pointer to the mlx5_flow_rss_desc.
8038  */
8039 static void
8040 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
8041                        struct mlx5_flow_rss_desc *rss_desc)
8042 {
8043         uint64_t items = dev_flow->handle->layers;
8044         int rss_inner = 0;
8045         uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
8046
8047         dev_flow->hash_fields = 0;
8048 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
8049         if (rss_desc->level >= 2) {
8050                 dev_flow->hash_fields |= IBV_RX_HASH_INNER;
8051                 rss_inner = 1;
8052         }
8053 #endif
8054         if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
8055             (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
8056                 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
8057                         if (rss_types & ETH_RSS_L3_SRC_ONLY)
8058                                 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
8059                         else if (rss_types & ETH_RSS_L3_DST_ONLY)
8060                                 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
8061                         else
8062                                 dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
8063                 }
8064         } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
8065                    (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
8066                 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
8067                         if (rss_types & ETH_RSS_L3_SRC_ONLY)
8068                                 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
8069                         else if (rss_types & ETH_RSS_L3_DST_ONLY)
8070                                 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
8071                         else
8072                                 dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
8073                 }
8074         }
8075         if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
8076             (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
8077                 if (rss_types & ETH_RSS_UDP) {
8078                         if (rss_types & ETH_RSS_L4_SRC_ONLY)
8079                                 dev_flow->hash_fields |=
8080                                                 IBV_RX_HASH_SRC_PORT_UDP;
8081                         else if (rss_types & ETH_RSS_L4_DST_ONLY)
8082                                 dev_flow->hash_fields |=
8083                                                 IBV_RX_HASH_DST_PORT_UDP;
8084                         else
8085                                 dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
8086                 }
8087         } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
8088                    (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
8089                 if (rss_types & ETH_RSS_TCP) {
8090                         if (rss_types & ETH_RSS_L4_SRC_ONLY)
8091                                 dev_flow->hash_fields |=
8092                                                 IBV_RX_HASH_SRC_PORT_TCP;
8093                         else if (rss_types & ETH_RSS_L4_DST_ONLY)
8094                                 dev_flow->hash_fields |=
8095                                                 IBV_RX_HASH_DST_PORT_TCP;
8096                         else
8097                                 dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
8098                 }
8099         }
8100 }
8101
8102 /**
8103  * Fill the flow with DV spec, lock free
8104  * (mutex should be acquired by caller).
8105  *
8106  * @param[in] dev
8107  *   Pointer to rte_eth_dev structure.
8108  * @param[in, out] dev_flow
8109  *   Pointer to the sub flow.
8110  * @param[in] attr
8111  *   Pointer to the flow attributes.
8112  * @param[in] items
8113  *   Pointer to the list of items.
8114  * @param[in] actions
8115  *   Pointer to the list of actions.
8116  * @param[out] error
8117  *   Pointer to the error structure.
8118  *
8119  * @return
8120  *   0 on success, a negative errno value otherwise and rte_errno is set.
8121  */
8122 static int
8123 __flow_dv_translate(struct rte_eth_dev *dev,
8124                     struct mlx5_flow *dev_flow,
8125                     const struct rte_flow_attr *attr,
8126                     const struct rte_flow_item items[],
8127                     const struct rte_flow_action actions[],
8128                     struct rte_flow_error *error)
8129 {
8130         struct mlx5_priv *priv = dev->data->dev_private;
8131         struct mlx5_dev_config *dev_conf = &priv->config;
8132         struct rte_flow *flow = dev_flow->flow;
8133         struct mlx5_flow_handle *handle = dev_flow->handle;
8134         struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
8135                                               priv->rss_desc)
8136                                               [!!priv->flow_nested_idx];
8137         uint64_t item_flags = 0;
8138         uint64_t last_item = 0;
8139         uint64_t action_flags = 0;
8140         uint64_t priority = attr->priority;
8141         struct mlx5_flow_dv_matcher matcher = {
8142                 .mask = {
8143                         .size = sizeof(matcher.mask.buf) -
8144                                 MLX5_ST_SZ_BYTES(fte_match_set_misc4),
8145                 },
8146         };
8147         int actions_n = 0;
8148         bool actions_end = false;
8149         union {
8150                 struct mlx5_flow_dv_modify_hdr_resource res;
8151                 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
8152                             sizeof(struct mlx5_modification_cmd) *
8153                             (MLX5_MAX_MODIFY_NUM + 1)];
8154         } mhdr_dummy;
8155         struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
8156         const struct rte_flow_action_count *count = NULL;
8157         const struct rte_flow_action_age *age = NULL;
8158         union flow_dv_attr flow_attr = { .attr = 0 };
8159         uint32_t tag_be;
8160         union mlx5_flow_tbl_key tbl_key;
8161         uint32_t modify_action_position = UINT32_MAX;
8162         void *match_mask = matcher.mask.buf;
8163         void *match_value = dev_flow->dv.value.buf;
8164         uint8_t next_protocol = 0xff;
8165         struct rte_vlan_hdr vlan = { 0 };
8166         uint32_t table;
8167         int ret = 0;
8168
8169         mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
8170                                            MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
8171         ret = mlx5_flow_group_to_table(attr, dev_flow->external, attr->group,
8172                                        !!priv->fdb_def_rule, &table, error);
8173         if (ret)
8174                 return ret;
8175         dev_flow->dv.group = table;
8176         if (attr->transfer)
8177                 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
8178         if (priority == MLX5_FLOW_PRIO_RSVD)
8179                 priority = dev_conf->flow_prio - 1;
8180         /* number of actions must be set to 0 in case of dirty stack. */
8181         mhdr_res->actions_num = 0;
8182         for (; !actions_end ; actions++) {
8183                 const struct rte_flow_action_queue *queue;
8184                 const struct rte_flow_action_rss *rss;
8185                 const struct rte_flow_action *action = actions;
8186                 const uint8_t *rss_key;
8187                 const struct rte_flow_action_jump *jump_data;
8188                 const struct rte_flow_action_meter *mtr;
8189                 struct mlx5_flow_tbl_resource *tbl;
8190                 uint32_t port_id = 0;
8191                 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
8192                 int action_type = actions->type;
8193                 const struct rte_flow_action *found_action = NULL;
8194                 struct mlx5_flow_meter *fm = NULL;
8195
8196                 if (!mlx5_flow_os_action_supported(action_type))
8197                         return rte_flow_error_set(error, ENOTSUP,
8198                                                   RTE_FLOW_ERROR_TYPE_ACTION,
8199                                                   actions,
8200                                                   "action not supported");
8201                 switch (action_type) {
8202                 case RTE_FLOW_ACTION_TYPE_VOID:
8203                         break;
8204                 case RTE_FLOW_ACTION_TYPE_PORT_ID:
8205                         if (flow_dv_translate_action_port_id(dev, action,
8206                                                              &port_id, error))
8207                                 return -rte_errno;
8208                         port_id_resource.port_id = port_id;
8209                         MLX5_ASSERT(!handle->rix_port_id_action);
8210                         if (flow_dv_port_id_action_resource_register
8211                             (dev, &port_id_resource, dev_flow, error))
8212                                 return -rte_errno;
8213                         dev_flow->dv.actions[actions_n++] =
8214                                         dev_flow->dv.port_id_action->action;
8215                         action_flags |= MLX5_FLOW_ACTION_PORT_ID;
8216                         dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
8217                         break;
8218                 case RTE_FLOW_ACTION_TYPE_FLAG:
8219                         action_flags |= MLX5_FLOW_ACTION_FLAG;
8220                         dev_flow->handle->mark = 1;
8221                         if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
8222                                 struct rte_flow_action_mark mark = {
8223                                         .id = MLX5_FLOW_MARK_DEFAULT,
8224                                 };
8225
8226                                 if (flow_dv_convert_action_mark(dev, &mark,
8227                                                                 mhdr_res,
8228                                                                 error))
8229                                         return -rte_errno;
8230                                 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
8231                                 break;
8232                         }
8233                         tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
8234                         /*
8235                          * Only one FLAG or MARK is supported per device flow
8236                          * right now. So the pointer to the tag resource must be
8237                          * zero before the register process.
8238                          */
8239                         MLX5_ASSERT(!handle->dvh.rix_tag);
8240                         if (flow_dv_tag_resource_register(dev, tag_be,
8241                                                           dev_flow, error))
8242                                 return -rte_errno;
8243                         MLX5_ASSERT(dev_flow->dv.tag_resource);
8244                         dev_flow->dv.actions[actions_n++] =
8245                                         dev_flow->dv.tag_resource->action;
8246                         break;
8247                 case RTE_FLOW_ACTION_TYPE_MARK:
8248                         action_flags |= MLX5_FLOW_ACTION_MARK;
8249                         dev_flow->handle->mark = 1;
8250                         if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
8251                                 const struct rte_flow_action_mark *mark =
8252                                         (const struct rte_flow_action_mark *)
8253                                                 actions->conf;
8254
8255                                 if (flow_dv_convert_action_mark(dev, mark,
8256                                                                 mhdr_res,
8257                                                                 error))
8258                                         return -rte_errno;
8259                                 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
8260                                 break;
8261                         }
8262                         /* Fall-through */
8263                 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
8264                         /* Legacy (non-extensive) MARK action. */
8265                         tag_be = mlx5_flow_mark_set
8266                               (((const struct rte_flow_action_mark *)
8267                                (actions->conf))->id);
8268                         MLX5_ASSERT(!handle->dvh.rix_tag);
8269                         if (flow_dv_tag_resource_register(dev, tag_be,
8270                                                           dev_flow, error))
8271                                 return -rte_errno;
8272                         MLX5_ASSERT(dev_flow->dv.tag_resource);
8273                         dev_flow->dv.actions[actions_n++] =
8274                                         dev_flow->dv.tag_resource->action;
8275                         break;
8276                 case RTE_FLOW_ACTION_TYPE_SET_META:
8277                         if (flow_dv_convert_action_set_meta
8278                                 (dev, mhdr_res, attr,
8279                                  (const struct rte_flow_action_set_meta *)
8280                                   actions->conf, error))
8281                                 return -rte_errno;
8282                         action_flags |= MLX5_FLOW_ACTION_SET_META;
8283                         break;
8284                 case RTE_FLOW_ACTION_TYPE_SET_TAG:
8285                         if (flow_dv_convert_action_set_tag
8286                                 (dev, mhdr_res,
8287                                  (const struct rte_flow_action_set_tag *)
8288                                   actions->conf, error))
8289                                 return -rte_errno;
8290                         action_flags |= MLX5_FLOW_ACTION_SET_TAG;
8291                         break;
8292                 case RTE_FLOW_ACTION_TYPE_DROP:
8293                         action_flags |= MLX5_FLOW_ACTION_DROP;
8294                         dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
8295                         break;
8296                 case RTE_FLOW_ACTION_TYPE_QUEUE:
8297                         queue = actions->conf;
8298                         rss_desc->queue_num = 1;
8299                         rss_desc->queue[0] = queue->index;
8300                         action_flags |= MLX5_FLOW_ACTION_QUEUE;
8301                         dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
8302                         break;
8303                 case RTE_FLOW_ACTION_TYPE_RSS:
8304                         rss = actions->conf;
8305                         memcpy(rss_desc->queue, rss->queue,
8306                                rss->queue_num * sizeof(uint16_t));
8307                         rss_desc->queue_num = rss->queue_num;
8308                         /* NULL RSS key indicates default RSS key. */
8309                         rss_key = !rss->key ? rss_hash_default_key : rss->key;
8310                         memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
8311                         /*
8312                          * rss->level and rss.types should be set in advance
8313                          * when expanding items for RSS.
8314                          */
8315                         action_flags |= MLX5_FLOW_ACTION_RSS;
8316                         dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
8317                         break;
8318                 case RTE_FLOW_ACTION_TYPE_AGE:
8319                 case RTE_FLOW_ACTION_TYPE_COUNT:
8320                         if (!dev_conf->devx) {
8321                                 return rte_flow_error_set
8322                                               (error, ENOTSUP,
8323                                                RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8324                                                NULL,
8325                                                "count action not supported");
8326                         }
8327                         /* Save information first, will apply later. */
8328                         if (actions->type == RTE_FLOW_ACTION_TYPE_COUNT)
8329                                 count = action->conf;
8330                         else
8331                                 age = action->conf;
8332                         action_flags |= MLX5_FLOW_ACTION_COUNT;
8333                         break;
8334                 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
8335                         dev_flow->dv.actions[actions_n++] =
8336                                                 priv->sh->pop_vlan_action;
8337                         action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
8338                         break;
8339                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
8340                         if (!(action_flags &
8341                               MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
8342                                 flow_dev_get_vlan_info_from_items(items, &vlan);
8343                         vlan.eth_proto = rte_be_to_cpu_16
8344                              ((((const struct rte_flow_action_of_push_vlan *)
8345                                                    actions->conf)->ethertype));
8346                         found_action = mlx5_flow_find_action
8347                                         (actions + 1,
8348                                          RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
8349                         if (found_action)
8350                                 mlx5_update_vlan_vid_pcp(found_action, &vlan);
8351                         found_action = mlx5_flow_find_action
8352                                         (actions + 1,
8353                                          RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
8354                         if (found_action)
8355                                 mlx5_update_vlan_vid_pcp(found_action, &vlan);
8356                         if (flow_dv_create_action_push_vlan
8357                                             (dev, attr, &vlan, dev_flow, error))
8358                                 return -rte_errno;
8359                         dev_flow->dv.actions[actions_n++] =
8360                                         dev_flow->dv.push_vlan_res->action;
8361                         action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
8362                         break;
8363                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
8364                         /* of_vlan_push action handled this action */
8365                         MLX5_ASSERT(action_flags &
8366                                     MLX5_FLOW_ACTION_OF_PUSH_VLAN);
8367                         break;
8368                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
8369                         if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
8370                                 break;
8371                         flow_dev_get_vlan_info_from_items(items, &vlan);
8372                         mlx5_update_vlan_vid_pcp(actions, &vlan);
8373                         /* If no VLAN push - this is a modify header action */
8374                         if (flow_dv_convert_action_modify_vlan_vid
8375                                                 (mhdr_res, actions, error))
8376                                 return -rte_errno;
8377                         action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
8378                         break;
8379                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
8380                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
8381                         if (flow_dv_create_action_l2_encap(dev, actions,
8382                                                            dev_flow,
8383                                                            attr->transfer,
8384                                                            error))
8385                                 return -rte_errno;
8386                         dev_flow->dv.actions[actions_n++] =
8387                                         dev_flow->dv.encap_decap->action;
8388                         action_flags |= MLX5_FLOW_ACTION_ENCAP;
8389                         break;
8390                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
8391                 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
8392                         if (flow_dv_create_action_l2_decap(dev, dev_flow,
8393                                                            attr->transfer,
8394                                                            error))
8395                                 return -rte_errno;
8396                         dev_flow->dv.actions[actions_n++] =
8397                                         dev_flow->dv.encap_decap->action;
8398                         action_flags |= MLX5_FLOW_ACTION_DECAP;
8399                         break;
8400                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
8401                         /* Handle encap with preceding decap. */
8402                         if (action_flags & MLX5_FLOW_ACTION_DECAP) {
8403                                 if (flow_dv_create_action_raw_encap
8404                                         (dev, actions, dev_flow, attr, error))
8405                                         return -rte_errno;
8406                                 dev_flow->dv.actions[actions_n++] =
8407                                         dev_flow->dv.encap_decap->action;
8408                         } else {
8409                                 /* Handle encap without preceding decap. */
8410                                 if (flow_dv_create_action_l2_encap
8411                                     (dev, actions, dev_flow, attr->transfer,
8412                                      error))
8413                                         return -rte_errno;
8414                                 dev_flow->dv.actions[actions_n++] =
8415                                         dev_flow->dv.encap_decap->action;
8416                         }
8417                         action_flags |= MLX5_FLOW_ACTION_ENCAP;
8418                         break;
8419                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
8420                         while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
8421                                 ;
8422                         if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
8423                                 if (flow_dv_create_action_l2_decap
8424                                     (dev, dev_flow, attr->transfer, error))
8425                                         return -rte_errno;
8426                                 dev_flow->dv.actions[actions_n++] =
8427                                         dev_flow->dv.encap_decap->action;
8428                         }
8429                         /* If decap is followed by encap, handle it at encap. */
8430                         action_flags |= MLX5_FLOW_ACTION_DECAP;
8431                         break;
8432                 case RTE_FLOW_ACTION_TYPE_JUMP:
8433                         jump_data = action->conf;
8434                         ret = mlx5_flow_group_to_table(attr, dev_flow->external,
8435                                                        jump_data->group,
8436                                                        !!priv->fdb_def_rule,
8437                                                        &table, error);
8438                         if (ret)
8439                                 return ret;
8440                         tbl = flow_dv_tbl_resource_get(dev, table,
8441                                                        attr->egress,
8442                                                        attr->transfer, error);
8443                         if (!tbl)
8444                                 return rte_flow_error_set
8445                                                 (error, errno,
8446                                                  RTE_FLOW_ERROR_TYPE_ACTION,
8447                                                  NULL,
8448                                                  "cannot create jump action.");
8449                         if (flow_dv_jump_tbl_resource_register
8450                             (dev, tbl, dev_flow, error)) {
8451                                 flow_dv_tbl_resource_release(dev, tbl);
8452                                 return rte_flow_error_set
8453                                                 (error, errno,
8454                                                  RTE_FLOW_ERROR_TYPE_ACTION,
8455                                                  NULL,
8456                                                  "cannot create jump action.");
8457                         }
8458                         dev_flow->dv.actions[actions_n++] =
8459                                         dev_flow->dv.jump->action;
8460                         action_flags |= MLX5_FLOW_ACTION_JUMP;
8461                         dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
8462                         break;
8463                 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
8464                 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
8465                         if (flow_dv_convert_action_modify_mac
8466                                         (mhdr_res, actions, error))
8467                                 return -rte_errno;
8468                         action_flags |= actions->type ==
8469                                         RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
8470                                         MLX5_FLOW_ACTION_SET_MAC_SRC :
8471                                         MLX5_FLOW_ACTION_SET_MAC_DST;
8472                         break;
8473                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
8474                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
8475                         if (flow_dv_convert_action_modify_ipv4
8476                                         (mhdr_res, actions, error))
8477                                 return -rte_errno;
8478                         action_flags |= actions->type ==
8479                                         RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
8480                                         MLX5_FLOW_ACTION_SET_IPV4_SRC :
8481                                         MLX5_FLOW_ACTION_SET_IPV4_DST;
8482                         break;
8483                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
8484                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
8485                         if (flow_dv_convert_action_modify_ipv6
8486                                         (mhdr_res, actions, error))
8487                                 return -rte_errno;
8488                         action_flags |= actions->type ==
8489                                         RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
8490                                         MLX5_FLOW_ACTION_SET_IPV6_SRC :
8491                                         MLX5_FLOW_ACTION_SET_IPV6_DST;
8492                         break;
8493                 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
8494                 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
8495                         if (flow_dv_convert_action_modify_tp
8496                                         (mhdr_res, actions, items,
8497                                          &flow_attr, dev_flow, !!(action_flags &
8498                                          MLX5_FLOW_ACTION_DECAP), error))
8499                                 return -rte_errno;
8500                         action_flags |= actions->type ==
8501                                         RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
8502                                         MLX5_FLOW_ACTION_SET_TP_SRC :
8503                                         MLX5_FLOW_ACTION_SET_TP_DST;
8504                         break;
8505                 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
8506                         if (flow_dv_convert_action_modify_dec_ttl
8507                                         (mhdr_res, items, &flow_attr, dev_flow,
8508                                          !!(action_flags &
8509                                          MLX5_FLOW_ACTION_DECAP), error))
8510                                 return -rte_errno;
8511                         action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
8512                         break;
8513                 case RTE_FLOW_ACTION_TYPE_SET_TTL:
8514                         if (flow_dv_convert_action_modify_ttl
8515                                         (mhdr_res, actions, items, &flow_attr,
8516                                          dev_flow, !!(action_flags &
8517                                          MLX5_FLOW_ACTION_DECAP), error))
8518                                 return -rte_errno;
8519                         action_flags |= MLX5_FLOW_ACTION_SET_TTL;
8520                         break;
8521                 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
8522                 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
8523                         if (flow_dv_convert_action_modify_tcp_seq
8524                                         (mhdr_res, actions, error))
8525                                 return -rte_errno;
8526                         action_flags |= actions->type ==
8527                                         RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
8528                                         MLX5_FLOW_ACTION_INC_TCP_SEQ :
8529                                         MLX5_FLOW_ACTION_DEC_TCP_SEQ;
8530                         break;
8531
8532                 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
8533                 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
8534                         if (flow_dv_convert_action_modify_tcp_ack
8535                                         (mhdr_res, actions, error))
8536                                 return -rte_errno;
8537                         action_flags |= actions->type ==
8538                                         RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
8539                                         MLX5_FLOW_ACTION_INC_TCP_ACK :
8540                                         MLX5_FLOW_ACTION_DEC_TCP_ACK;
8541                         break;
8542                 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
8543                         if (flow_dv_convert_action_set_reg
8544                                         (mhdr_res, actions, error))
8545                                 return -rte_errno;
8546                         action_flags |= MLX5_FLOW_ACTION_SET_TAG;
8547                         break;
8548                 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
8549                         if (flow_dv_convert_action_copy_mreg
8550                                         (dev, mhdr_res, actions, error))
8551                                 return -rte_errno;
8552                         action_flags |= MLX5_FLOW_ACTION_SET_TAG;
8553                         break;
8554                 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
8555                         action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
8556                         dev_flow->handle->fate_action =
8557                                         MLX5_FLOW_FATE_DEFAULT_MISS;
8558                         break;
8559                 case RTE_FLOW_ACTION_TYPE_METER:
8560                         mtr = actions->conf;
8561                         if (!flow->meter) {
8562                                 fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
8563                                                             attr, error);
8564                                 if (!fm)
8565                                         return rte_flow_error_set(error,
8566                                                 rte_errno,
8567                                                 RTE_FLOW_ERROR_TYPE_ACTION,
8568                                                 NULL,
8569                                                 "meter not found "
8570                                                 "or invalid parameters");
8571                                 flow->meter = fm->idx;
8572                         }
8573                         /* Set the meter action. */
8574                         if (!fm) {
8575                                 fm = mlx5_ipool_get(priv->sh->ipool
8576                                                 [MLX5_IPOOL_MTR], flow->meter);
8577                                 if (!fm)
8578                                         return rte_flow_error_set(error,
8579                                                 rte_errno,
8580                                                 RTE_FLOW_ERROR_TYPE_ACTION,
8581                                                 NULL,
8582                                                 "meter not found "
8583                                                 "or invalid parameters");
8584                         }
8585                         dev_flow->dv.actions[actions_n++] =
8586                                 fm->mfts->meter_action;
8587                         action_flags |= MLX5_FLOW_ACTION_METER;
8588                         break;
8589                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
8590                         if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
8591                                                               actions, error))
8592                                 return -rte_errno;
8593                         action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
8594                         break;
8595                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
8596                         if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
8597                                                               actions, error))
8598                                 return -rte_errno;
8599                         action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
8600                         break;
8601                 case RTE_FLOW_ACTION_TYPE_END:
8602                         actions_end = true;
8603                         if (mhdr_res->actions_num) {
8604                                 /* create modify action if needed. */
8605                                 if (flow_dv_modify_hdr_resource_register
8606                                         (dev, mhdr_res, dev_flow, error))
8607                                         return -rte_errno;
8608                                 dev_flow->dv.actions[modify_action_position] =
8609                                         handle->dvh.modify_hdr->action;
8610                         }
8611                         if (action_flags & MLX5_FLOW_ACTION_COUNT) {
8612                                 flow->counter =
8613                                         flow_dv_translate_create_counter(dev,
8614                                                 dev_flow, count, age);
8615
8616                                 if (!flow->counter)
8617                                         return rte_flow_error_set
8618                                                 (error, rte_errno,
8619                                                 RTE_FLOW_ERROR_TYPE_ACTION,
8620                                                 NULL,
8621                                                 "cannot create counter"
8622                                                 " object.");
8623                                 dev_flow->dv.actions[actions_n++] =
8624                                           (flow_dv_counter_get_by_idx(dev,
8625                                           flow->counter, NULL))->action;
8626                         }
8627                         break;
8628                 default:
8629                         break;
8630                 }
8631                 if (mhdr_res->actions_num &&
8632                     modify_action_position == UINT32_MAX)
8633                         modify_action_position = actions_n++;
8634         }
8635         dev_flow->dv.actions_n = actions_n;
8636         dev_flow->act_flags = action_flags;
8637         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
8638                 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
8639                 int item_type = items->type;
8640
8641                 if (!mlx5_flow_os_item_supported(item_type))
8642                         return rte_flow_error_set(error, ENOTSUP,
8643                                                   RTE_FLOW_ERROR_TYPE_ITEM,
8644                                                   NULL, "item not supported");
8645                 switch (item_type) {
8646                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
8647                         flow_dv_translate_item_port_id(dev, match_mask,
8648                                                        match_value, items);
8649                         last_item = MLX5_FLOW_ITEM_PORT_ID;
8650                         break;
8651                 case RTE_FLOW_ITEM_TYPE_ETH:
8652                         flow_dv_translate_item_eth(match_mask, match_value,
8653                                                    items, tunnel,
8654                                                    dev_flow->dv.group);
8655                         matcher.priority = action_flags &
8656                                         MLX5_FLOW_ACTION_DEFAULT_MISS &&
8657                                         !dev_flow->external ?
8658                                         MLX5_PRIORITY_MAP_L3 :
8659                                         MLX5_PRIORITY_MAP_L2;
8660                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
8661                                              MLX5_FLOW_LAYER_OUTER_L2;
8662                         break;
8663                 case RTE_FLOW_ITEM_TYPE_VLAN:
8664                         flow_dv_translate_item_vlan(dev_flow,
8665                                                     match_mask, match_value,
8666                                                     items, tunnel,
8667                                                     dev_flow->dv.group);
8668                         matcher.priority = MLX5_PRIORITY_MAP_L2;
8669                         last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
8670                                               MLX5_FLOW_LAYER_INNER_VLAN) :
8671                                              (MLX5_FLOW_LAYER_OUTER_L2 |
8672                                               MLX5_FLOW_LAYER_OUTER_VLAN);
8673                         break;
8674                 case RTE_FLOW_ITEM_TYPE_IPV4:
8675                         mlx5_flow_tunnel_ip_check(items, next_protocol,
8676                                                   &item_flags, &tunnel);
8677                         flow_dv_translate_item_ipv4(match_mask, match_value,
8678                                                     items, item_flags, tunnel,
8679                                                     dev_flow->dv.group);
8680                         matcher.priority = MLX5_PRIORITY_MAP_L3;
8681                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
8682                                              MLX5_FLOW_LAYER_OUTER_L3_IPV4;
8683                         if (items->mask != NULL &&
8684                             ((const struct rte_flow_item_ipv4 *)
8685                              items->mask)->hdr.next_proto_id) {
8686                                 next_protocol =
8687                                         ((const struct rte_flow_item_ipv4 *)
8688                                          (items->spec))->hdr.next_proto_id;
8689                                 next_protocol &=
8690                                         ((const struct rte_flow_item_ipv4 *)
8691                                          (items->mask))->hdr.next_proto_id;
8692                         } else {
8693                                 /* Reset for inner layer. */
8694                                 next_protocol = 0xff;
8695                         }
8696                         break;
8697                 case RTE_FLOW_ITEM_TYPE_IPV6:
8698                         mlx5_flow_tunnel_ip_check(items, next_protocol,
8699                                                   &item_flags, &tunnel);
8700                         flow_dv_translate_item_ipv6(match_mask, match_value,
8701                                                     items, item_flags, tunnel,
8702                                                     dev_flow->dv.group);
8703                         matcher.priority = MLX5_PRIORITY_MAP_L3;
8704                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
8705                                              MLX5_FLOW_LAYER_OUTER_L3_IPV6;
8706                         if (items->mask != NULL &&
8707                             ((const struct rte_flow_item_ipv6 *)
8708                              items->mask)->hdr.proto) {
8709                                 next_protocol =
8710                                         ((const struct rte_flow_item_ipv6 *)
8711                                          items->spec)->hdr.proto;
8712                                 next_protocol &=
8713                                         ((const struct rte_flow_item_ipv6 *)
8714                                          items->mask)->hdr.proto;
8715                         } else {
8716                                 /* Reset for inner layer. */
8717                                 next_protocol = 0xff;
8718                         }
8719                         break;
8720                 case RTE_FLOW_ITEM_TYPE_TCP:
8721                         flow_dv_translate_item_tcp(match_mask, match_value,
8722                                                    items, tunnel);
8723                         matcher.priority = MLX5_PRIORITY_MAP_L4;
8724                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
8725                                              MLX5_FLOW_LAYER_OUTER_L4_TCP;
8726                         break;
8727                 case RTE_FLOW_ITEM_TYPE_UDP:
8728                         flow_dv_translate_item_udp(match_mask, match_value,
8729                                                    items, tunnel);
8730                         matcher.priority = MLX5_PRIORITY_MAP_L4;
8731                         last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
8732                                              MLX5_FLOW_LAYER_OUTER_L4_UDP;
8733                         break;
8734                 case RTE_FLOW_ITEM_TYPE_GRE:
8735                         flow_dv_translate_item_gre(match_mask, match_value,
8736                                                    items, tunnel);
8737                         matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8738                         last_item = MLX5_FLOW_LAYER_GRE;
8739                         break;
8740                 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
8741                         flow_dv_translate_item_gre_key(match_mask,
8742                                                        match_value, items);
8743                         last_item = MLX5_FLOW_LAYER_GRE_KEY;
8744                         break;
8745                 case RTE_FLOW_ITEM_TYPE_NVGRE:
8746                         flow_dv_translate_item_nvgre(match_mask, match_value,
8747                                                      items, tunnel);
8748                         matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8749                         last_item = MLX5_FLOW_LAYER_GRE;
8750                         break;
8751                 case RTE_FLOW_ITEM_TYPE_VXLAN:
8752                         flow_dv_translate_item_vxlan(match_mask, match_value,
8753                                                      items, tunnel);
8754                         matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8755                         last_item = MLX5_FLOW_LAYER_VXLAN;
8756                         break;
8757                 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
8758                         flow_dv_translate_item_vxlan_gpe(match_mask,
8759                                                          match_value, items,
8760                                                          tunnel);
8761                         matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8762                         last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
8763                         break;
8764                 case RTE_FLOW_ITEM_TYPE_GENEVE:
8765                         flow_dv_translate_item_geneve(match_mask, match_value,
8766                                                       items, tunnel);
8767                         matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8768                         last_item = MLX5_FLOW_LAYER_GENEVE;
8769                         break;
8770                 case RTE_FLOW_ITEM_TYPE_MPLS:
8771                         flow_dv_translate_item_mpls(match_mask, match_value,
8772                                                     items, last_item, tunnel);
8773                         matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8774                         last_item = MLX5_FLOW_LAYER_MPLS;
8775                         break;
8776                 case RTE_FLOW_ITEM_TYPE_MARK:
8777                         flow_dv_translate_item_mark(dev, match_mask,
8778                                                     match_value, items);
8779                         last_item = MLX5_FLOW_ITEM_MARK;
8780                         break;
8781                 case RTE_FLOW_ITEM_TYPE_META:
8782                         flow_dv_translate_item_meta(dev, match_mask,
8783                                                     match_value, attr, items);
8784                         last_item = MLX5_FLOW_ITEM_METADATA;
8785                         break;
8786                 case RTE_FLOW_ITEM_TYPE_ICMP:
8787                         flow_dv_translate_item_icmp(match_mask, match_value,
8788                                                     items, tunnel);
8789                         last_item = MLX5_FLOW_LAYER_ICMP;
8790                         break;
8791                 case RTE_FLOW_ITEM_TYPE_ICMP6:
8792                         flow_dv_translate_item_icmp6(match_mask, match_value,
8793                                                       items, tunnel);
8794                         last_item = MLX5_FLOW_LAYER_ICMP6;
8795                         break;
8796                 case RTE_FLOW_ITEM_TYPE_TAG:
8797                         flow_dv_translate_item_tag(dev, match_mask,
8798                                                    match_value, items);
8799                         last_item = MLX5_FLOW_ITEM_TAG;
8800                         break;
8801                 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
8802                         flow_dv_translate_mlx5_item_tag(dev, match_mask,
8803                                                         match_value, items);
8804                         last_item = MLX5_FLOW_ITEM_TAG;
8805                         break;
8806                 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
8807                         flow_dv_translate_item_tx_queue(dev, match_mask,
8808                                                         match_value,
8809                                                         items);
8810                         last_item = MLX5_FLOW_ITEM_TX_QUEUE;
8811                         break;
8812                 case RTE_FLOW_ITEM_TYPE_GTP:
8813                         flow_dv_translate_item_gtp(match_mask, match_value,
8814                                                    items, tunnel);
8815                         matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8816                         last_item = MLX5_FLOW_LAYER_GTP;
8817                         break;
8818                 case RTE_FLOW_ITEM_TYPE_ECPRI:
8819                         if (!mlx5_flex_parser_ecpri_exist(dev)) {
8820                                 /* Create it only the first time to be used. */
8821                                 ret = mlx5_flex_parser_ecpri_alloc(dev);
8822                                 if (ret)
8823                                         return rte_flow_error_set
8824                                                 (error, -ret,
8825                                                 RTE_FLOW_ERROR_TYPE_ITEM,
8826                                                 NULL,
8827                                                 "cannot create eCPRI parser");
8828                         }
8829                         /* Adjust the length matcher and device flow value. */
8830                         matcher.mask.size = MLX5_ST_SZ_BYTES(fte_match_param);
8831                         dev_flow->dv.value.size =
8832                                         MLX5_ST_SZ_BYTES(fte_match_param);
8833                         flow_dv_translate_item_ecpri(dev, match_mask,
8834                                                      match_value, items);
8835                         /* No other protocol should follow eCPRI layer. */
8836                         last_item = MLX5_FLOW_LAYER_ECPRI;
8837                         break;
8838                 default:
8839                         break;
8840                 }
8841                 item_flags |= last_item;
8842         }
8843         /*
8844          * When E-Switch mode is enabled, we have two cases where we need to
8845          * set the source port manually.
8846          * The first one, is in case of Nic steering rule, and the second is
8847          * E-Switch rule where no port_id item was found. In both cases
8848          * the source port is set according the current port in use.
8849          */
8850         if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
8851             (priv->representor || priv->master)) {
8852                 if (flow_dv_translate_item_port_id(dev, match_mask,
8853                                                    match_value, NULL))
8854                         return -rte_errno;
8855         }
8856 #ifdef RTE_LIBRTE_MLX5_DEBUG
8857         MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
8858                                               dev_flow->dv.value.buf));
8859 #endif
8860         /*
8861          * Layers may be already initialized from prefix flow if this dev_flow
8862          * is the suffix flow.
8863          */
8864         handle->layers |= item_flags;
8865         if (action_flags & MLX5_FLOW_ACTION_RSS)
8866                 flow_dv_hashfields_set(dev_flow, rss_desc);
8867         /* Register matcher. */
8868         matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
8869                                     matcher.mask.size);
8870         matcher.priority = mlx5_flow_adjust_priority(dev, priority,
8871                                                      matcher.priority);
8872         /* reserved field no needs to be set to 0 here. */
8873         tbl_key.domain = attr->transfer;
8874         tbl_key.direction = attr->egress;
8875         tbl_key.table_id = dev_flow->dv.group;
8876         if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
8877                 return -rte_errno;
8878         return 0;
8879 }
8880
8881 /**
8882  * Apply the flow to the NIC, lock free,
8883  * (mutex should be acquired by caller).
8884  *
8885  * @param[in] dev
8886  *   Pointer to the Ethernet device structure.
8887  * @param[in, out] flow
8888  *   Pointer to flow structure.
8889  * @param[out] error
8890  *   Pointer to error structure.
8891  *
8892  * @return
8893  *   0 on success, a negative errno value otherwise and rte_errno is set.
8894  */
8895 static int
8896 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
8897                 struct rte_flow_error *error)
8898 {
8899         struct mlx5_flow_dv_workspace *dv;
8900         struct mlx5_flow_handle *dh;
8901         struct mlx5_flow_handle_dv *dv_h;
8902         struct mlx5_flow *dev_flow;
8903         struct mlx5_priv *priv = dev->data->dev_private;
8904         uint32_t handle_idx;
8905         int n;
8906         int err;
8907         int idx;
8908
8909         for (idx = priv->flow_idx - 1; idx >= priv->flow_nested_idx; idx--) {
8910                 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[idx];
8911                 dv = &dev_flow->dv;
8912                 dh = dev_flow->handle;
8913                 dv_h = &dh->dvh;
8914                 n = dv->actions_n;
8915                 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8916                         if (dv->transfer) {
8917                                 dv->actions[n++] = priv->sh->esw_drop_action;
8918                         } else {
8919                                 struct mlx5_hrxq *drop_hrxq;
8920                                 drop_hrxq = mlx5_hrxq_drop_new(dev);
8921                                 if (!drop_hrxq) {
8922                                         rte_flow_error_set
8923                                                 (error, errno,
8924                                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8925                                                  NULL,
8926                                                  "cannot get drop hash queue");
8927                                         goto error;
8928                                 }
8929                                 /*
8930                                  * Drop queues will be released by the specify
8931                                  * mlx5_hrxq_drop_release() function. Assign
8932                                  * the special index to hrxq to mark the queue
8933                                  * has been allocated.
8934                                  */
8935                                 dh->rix_hrxq = UINT32_MAX;
8936                                 dv->actions[n++] = drop_hrxq->action;
8937                         }
8938                 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8939                         struct mlx5_hrxq *hrxq;
8940                         uint32_t hrxq_idx;
8941                         struct mlx5_flow_rss_desc *rss_desc =
8942                                 &((struct mlx5_flow_rss_desc *)priv->rss_desc)
8943                                 [!!priv->flow_nested_idx];
8944
8945                         MLX5_ASSERT(rss_desc->queue_num);
8946                         hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
8947                                                  MLX5_RSS_HASH_KEY_LEN,
8948                                                  dev_flow->hash_fields,
8949                                                  rss_desc->queue,
8950                                                  rss_desc->queue_num);
8951                         if (!hrxq_idx) {
8952                                 hrxq_idx = mlx5_hrxq_new
8953                                                 (dev, rss_desc->key,
8954                                                 MLX5_RSS_HASH_KEY_LEN,
8955                                                 dev_flow->hash_fields,
8956                                                 rss_desc->queue,
8957                                                 rss_desc->queue_num,
8958                                                 !!(dh->layers &
8959                                                 MLX5_FLOW_LAYER_TUNNEL));
8960                         }
8961                         hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8962                                               hrxq_idx);
8963                         if (!hrxq) {
8964                                 rte_flow_error_set
8965                                         (error, rte_errno,
8966                                          RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8967                                          "cannot get hash queue");
8968                                 goto error;
8969                         }
8970                         dh->rix_hrxq = hrxq_idx;
8971                         dv->actions[n++] = hrxq->action;
8972                 } else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
8973                         if (flow_dv_default_miss_resource_register
8974                                         (dev, error)) {
8975                                 rte_flow_error_set
8976                                         (error, rte_errno,
8977                                          RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8978                                          "cannot create default miss resource");
8979                                 goto error_default_miss;
8980                         }
8981                         dh->rix_default_fate =  MLX5_FLOW_FATE_DEFAULT_MISS;
8982                         dv->actions[n++] = priv->sh->default_miss.action;
8983                 }
8984                 err = mlx5_flow_os_create_flow(dv_h->matcher->matcher_object,
8985                                                (void *)&dv->value, n,
8986                                                dv->actions, &dh->drv_flow);
8987                 if (err) {
8988                         rte_flow_error_set(error, errno,
8989                                            RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8990                                            NULL,
8991                                            "hardware refuses to create flow");
8992                         goto error;
8993                 }
8994                 if (priv->vmwa_context &&
8995                     dh->vf_vlan.tag && !dh->vf_vlan.created) {
8996                         /*
8997                          * The rule contains the VLAN pattern.
8998                          * For VF we are going to create VLAN
8999                          * interface to make hypervisor set correct
9000                          * e-Switch vport context.
9001                          */
9002                         mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
9003                 }
9004         }
9005         return 0;
9006 error:
9007         if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS)
9008                 flow_dv_default_miss_resource_release(dev);
9009 error_default_miss:
9010         err = rte_errno; /* Save rte_errno before cleanup. */
9011         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
9012                        handle_idx, dh, next) {
9013                 /* hrxq is union, don't clear it if the flag is not set. */
9014                 if (dh->rix_hrxq) {
9015                         if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
9016                                 mlx5_hrxq_drop_release(dev);
9017                                 dh->rix_hrxq = 0;
9018                         } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
9019                                 mlx5_hrxq_release(dev, dh->rix_hrxq);
9020                                 dh->rix_hrxq = 0;
9021                         }
9022                 }
9023                 if (dh->vf_vlan.tag && dh->vf_vlan.created)
9024                         mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
9025         }
9026         rte_errno = err; /* Restore rte_errno. */
9027         return -rte_errno;
9028 }
9029
9030 /**
9031  * Release the flow matcher.
9032  *
9033  * @param dev
9034  *   Pointer to Ethernet device.
9035  * @param handle
9036  *   Pointer to mlx5_flow_handle.
9037  *
9038  * @return
9039  *   1 while a reference on it exists, 0 when freed.
9040  */
9041 static int
9042 flow_dv_matcher_release(struct rte_eth_dev *dev,
9043                         struct mlx5_flow_handle *handle)
9044 {
9045         struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
9046
9047         MLX5_ASSERT(matcher->matcher_object);
9048         DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
9049                 dev->data->port_id, (void *)matcher,
9050                 rte_atomic32_read(&matcher->refcnt));
9051         if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
9052                 claim_zero(mlx5_flow_os_destroy_flow_matcher
9053                            (matcher->matcher_object));
9054                 LIST_REMOVE(matcher, next);
9055                 /* table ref-- in release interface. */
9056                 flow_dv_tbl_resource_release(dev, matcher->tbl);
9057                 mlx5_free(matcher);
9058                 DRV_LOG(DEBUG, "port %u matcher %p: removed",
9059                         dev->data->port_id, (void *)matcher);
9060                 return 0;
9061         }
9062         return 1;
9063 }
9064
9065 /**
9066  * Release an encap/decap resource.
9067  *
9068  * @param dev
9069  *   Pointer to Ethernet device.
9070  * @param handle
9071  *   Pointer to mlx5_flow_handle.
9072  *
9073  * @return
9074  *   1 while a reference on it exists, 0 when freed.
9075  */
9076 static int
9077 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
9078                                      struct mlx5_flow_handle *handle)
9079 {
9080         struct mlx5_priv *priv = dev->data->dev_private;
9081         uint32_t idx = handle->dvh.rix_encap_decap;
9082         struct mlx5_flow_dv_encap_decap_resource *cache_resource;
9083
9084         cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
9085                          idx);
9086         if (!cache_resource)
9087                 return 0;
9088         MLX5_ASSERT(cache_resource->action);
9089         DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
9090                 (void *)cache_resource,
9091                 rte_atomic32_read(&cache_resource->refcnt));
9092         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9093                 claim_zero(mlx5_flow_os_destroy_flow_action
9094                                                 (cache_resource->action));
9095                 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
9096                              &priv->sh->encaps_decaps, idx,
9097                              cache_resource, next);
9098                 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
9099                 DRV_LOG(DEBUG, "encap/decap resource %p: removed",
9100                         (void *)cache_resource);
9101                 return 0;
9102         }
9103         return 1;
9104 }
9105
9106 /**
9107  * Release an jump to table action resource.
9108  *
9109  * @param dev
9110  *   Pointer to Ethernet device.
9111  * @param handle
9112  *   Pointer to mlx5_flow_handle.
9113  *
9114  * @return
9115  *   1 while a reference on it exists, 0 when freed.
9116  */
9117 static int
9118 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
9119                                   struct mlx5_flow_handle *handle)
9120 {
9121         struct mlx5_priv *priv = dev->data->dev_private;
9122         struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
9123         struct mlx5_flow_tbl_data_entry *tbl_data;
9124
9125         tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
9126                              handle->rix_jump);
9127         if (!tbl_data)
9128                 return 0;
9129         cache_resource = &tbl_data->jump;
9130         MLX5_ASSERT(cache_resource->action);
9131         DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
9132                 (void *)cache_resource,
9133                 rte_atomic32_read(&cache_resource->refcnt));
9134         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9135                 claim_zero(mlx5_flow_os_destroy_flow_action
9136                                                 (cache_resource->action));
9137                 /* jump action memory free is inside the table release. */
9138                 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
9139                 DRV_LOG(DEBUG, "jump table resource %p: removed",
9140                         (void *)cache_resource);
9141                 return 0;
9142         }
9143         return 1;
9144 }
9145
9146 /**
9147  * Release a default miss resource.
9148  *
9149  * @param dev
9150  *   Pointer to Ethernet device.
9151  * @return
9152  *   1 while a reference on it exists, 0 when freed.
9153  */
9154 static int
9155 flow_dv_default_miss_resource_release(struct rte_eth_dev *dev)
9156 {
9157         struct mlx5_priv *priv = dev->data->dev_private;
9158         struct mlx5_dev_ctx_shared *sh = priv->sh;
9159         struct mlx5_flow_default_miss_resource *cache_resource =
9160                         &sh->default_miss;
9161
9162         MLX5_ASSERT(cache_resource->action);
9163         DRV_LOG(DEBUG, "default miss resource %p: refcnt %d--",
9164                         (void *)cache_resource->action,
9165                         rte_atomic32_read(&cache_resource->refcnt));
9166         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9167                 claim_zero(mlx5_glue->destroy_flow_action
9168                                 (cache_resource->action));
9169                 DRV_LOG(DEBUG, "default miss resource %p: removed",
9170                                 (void *)cache_resource->action);
9171                 return 0;
9172         }
9173         return 1;
9174 }
9175
9176 /**
9177  * Release a modify-header resource.
9178  *
9179  * @param dev
9180  *   Pointer to Ethernet device.
9181  * @param handle
9182  *   Pointer to mlx5_flow_handle.
9183  *
9184  * @return
9185  *   1 while a reference on it exists, 0 when freed.
9186  */
9187 static int
9188 flow_dv_modify_hdr_resource_release(struct rte_eth_dev *dev,
9189                                     struct mlx5_flow_handle *handle)
9190 {
9191         struct mlx5_priv *priv = dev->data->dev_private;
9192         struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
9193                                                         handle->dvh.modify_hdr;
9194
9195         MLX5_ASSERT(cache_resource->action);
9196         DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
9197                 (void *)cache_resource,
9198                 rte_atomic32_read(&cache_resource->refcnt));
9199         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9200                 claim_zero(mlx5_flow_os_destroy_flow_action
9201                                                 (cache_resource->action));
9202                 mlx5_hlist_remove(priv->sh->modify_cmds,
9203                                   &cache_resource->entry);
9204                 mlx5_free(cache_resource);
9205                 DRV_LOG(DEBUG, "modify-header resource %p: removed",
9206                         (void *)cache_resource);
9207                 return 0;
9208         }
9209         return 1;
9210 }
9211
9212 /**
9213  * Release port ID action resource.
9214  *
9215  * @param dev
9216  *   Pointer to Ethernet device.
9217  * @param handle
9218  *   Pointer to mlx5_flow_handle.
9219  *
9220  * @return
9221  *   1 while a reference on it exists, 0 when freed.
9222  */
9223 static int
9224 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
9225                                         struct mlx5_flow_handle *handle)
9226 {
9227         struct mlx5_priv *priv = dev->data->dev_private;
9228         struct mlx5_flow_dv_port_id_action_resource *cache_resource;
9229         uint32_t idx = handle->rix_port_id_action;
9230
9231         cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
9232                                         idx);
9233         if (!cache_resource)
9234                 return 0;
9235         MLX5_ASSERT(cache_resource->action);
9236         DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
9237                 (void *)cache_resource,
9238                 rte_atomic32_read(&cache_resource->refcnt));
9239         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9240                 claim_zero(mlx5_flow_os_destroy_flow_action
9241                                                 (cache_resource->action));
9242                 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
9243                              &priv->sh->port_id_action_list, idx,
9244                              cache_resource, next);
9245                 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
9246                 DRV_LOG(DEBUG, "port id action resource %p: removed",
9247                         (void *)cache_resource);
9248                 return 0;
9249         }
9250         return 1;
9251 }
9252
9253 /**
9254  * Release push vlan action resource.
9255  *
9256  * @param dev
9257  *   Pointer to Ethernet device.
9258  * @param handle
9259  *   Pointer to mlx5_flow_handle.
9260  *
9261  * @return
9262  *   1 while a reference on it exists, 0 when freed.
9263  */
9264 static int
9265 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
9266                                           struct mlx5_flow_handle *handle)
9267 {
9268         struct mlx5_priv *priv = dev->data->dev_private;
9269         uint32_t idx = handle->dvh.rix_push_vlan;
9270         struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
9271
9272         cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
9273                                         idx);
9274         if (!cache_resource)
9275                 return 0;
9276         MLX5_ASSERT(cache_resource->action);
9277         DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
9278                 (void *)cache_resource,
9279                 rte_atomic32_read(&cache_resource->refcnt));
9280         if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9281                 claim_zero(mlx5_flow_os_destroy_flow_action
9282                                                 (cache_resource->action));
9283                 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
9284                              &priv->sh->push_vlan_action_list, idx,
9285                              cache_resource, next);
9286                 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
9287                 DRV_LOG(DEBUG, "push vlan action resource %p: removed",
9288                         (void *)cache_resource);
9289                 return 0;
9290         }
9291         return 1;
9292 }
9293
9294 /**
9295  * Release the fate resource.
9296  *
9297  * @param dev
9298  *   Pointer to Ethernet device.
9299  * @param handle
9300  *   Pointer to mlx5_flow_handle.
9301  */
9302 static void
9303 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
9304                                struct mlx5_flow_handle *handle)
9305 {
9306         if (!handle->rix_fate)
9307                 return;
9308         switch (handle->fate_action) {
9309         case MLX5_FLOW_FATE_DROP:
9310                 mlx5_hrxq_drop_release(dev);
9311                 break;
9312         case MLX5_FLOW_FATE_QUEUE:
9313                 mlx5_hrxq_release(dev, handle->rix_hrxq);
9314                 break;
9315         case MLX5_FLOW_FATE_JUMP:
9316                 flow_dv_jump_tbl_resource_release(dev, handle);
9317                 break;
9318         case MLX5_FLOW_FATE_PORT_ID:
9319                 flow_dv_port_id_action_resource_release(dev, handle);
9320                 break;
9321         case MLX5_FLOW_FATE_DEFAULT_MISS:
9322                 flow_dv_default_miss_resource_release(dev);
9323                 break;
9324         default:
9325                 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
9326                 break;
9327         }
9328         handle->rix_fate = 0;
9329 }
9330
9331 /**
9332  * Remove the flow from the NIC but keeps it in memory.
9333  * Lock free, (mutex should be acquired by caller).
9334  *
9335  * @param[in] dev
9336  *   Pointer to Ethernet device.
9337  * @param[in, out] flow
9338  *   Pointer to flow structure.
9339  */
9340 static void
9341 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
9342 {
9343         struct mlx5_flow_handle *dh;
9344         uint32_t handle_idx;
9345         struct mlx5_priv *priv = dev->data->dev_private;
9346
9347         if (!flow)
9348                 return;
9349         handle_idx = flow->dev_handles;
9350         while (handle_idx) {
9351                 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
9352                                     handle_idx);
9353                 if (!dh)
9354                         return;
9355                 if (dh->drv_flow) {
9356                         claim_zero(mlx5_flow_os_destroy_flow(dh->drv_flow));
9357                         dh->drv_flow = NULL;
9358                 }
9359                 if (dh->fate_action == MLX5_FLOW_FATE_DROP ||
9360                     dh->fate_action == MLX5_FLOW_FATE_QUEUE ||
9361                     dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS)
9362                         flow_dv_fate_resource_release(dev, dh);
9363                 if (dh->vf_vlan.tag && dh->vf_vlan.created)
9364                         mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
9365                 handle_idx = dh->next.next;
9366         }
9367 }
9368
9369 /**
9370  * Remove the flow from the NIC and the memory.
9371  * Lock free, (mutex should be acquired by caller).
9372  *
9373  * @param[in] dev
9374  *   Pointer to the Ethernet device structure.
9375  * @param[in, out] flow
9376  *   Pointer to flow structure.
9377  */
9378 static void
9379 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
9380 {
9381         struct mlx5_flow_handle *dev_handle;
9382         struct mlx5_priv *priv = dev->data->dev_private;
9383
9384         if (!flow)
9385                 return;
9386         __flow_dv_remove(dev, flow);
9387         if (flow->counter) {
9388                 flow_dv_counter_release(dev, flow->counter);
9389                 flow->counter = 0;
9390         }
9391         if (flow->meter) {
9392                 struct mlx5_flow_meter *fm;
9393
9394                 fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
9395                                     flow->meter);
9396                 if (fm)
9397                         mlx5_flow_meter_detach(fm);
9398                 flow->meter = 0;
9399         }
9400         while (flow->dev_handles) {
9401                 uint32_t tmp_idx = flow->dev_handles;
9402
9403                 dev_handle = mlx5_ipool_get(priv->sh->ipool
9404                                             [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
9405                 if (!dev_handle)
9406                         return;
9407                 flow->dev_handles = dev_handle->next.next;
9408                 if (dev_handle->dvh.matcher)
9409                         flow_dv_matcher_release(dev, dev_handle);
9410                 if (dev_handle->dvh.rix_encap_decap)
9411                         flow_dv_encap_decap_resource_release(dev, dev_handle);
9412                 if (dev_handle->dvh.modify_hdr)
9413                         flow_dv_modify_hdr_resource_release(dev, dev_handle);
9414                 if (dev_handle->dvh.rix_push_vlan)
9415                         flow_dv_push_vlan_action_resource_release(dev,
9416                                                                   dev_handle);
9417                 if (dev_handle->dvh.rix_tag)
9418                         flow_dv_tag_release(dev,
9419                                             dev_handle->dvh.rix_tag);
9420                 flow_dv_fate_resource_release(dev, dev_handle);
9421                 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
9422                            tmp_idx);
9423         }
9424 }
9425
9426 /**
9427  * Query a dv flow  rule for its statistics via devx.
9428  *
9429  * @param[in] dev
9430  *   Pointer to Ethernet device.
9431  * @param[in] flow
9432  *   Pointer to the sub flow.
9433  * @param[out] data
9434  *   data retrieved by the query.
9435  * @param[out] error
9436  *   Perform verbose error reporting if not NULL.
9437  *
9438  * @return
9439  *   0 on success, a negative errno value otherwise and rte_errno is set.
9440  */
9441 static int
9442 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
9443                     void *data, struct rte_flow_error *error)
9444 {
9445         struct mlx5_priv *priv = dev->data->dev_private;
9446         struct rte_flow_query_count *qc = data;
9447
9448         if (!priv->config.devx)
9449                 return rte_flow_error_set(error, ENOTSUP,
9450                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9451                                           NULL,
9452                                           "counters are not supported");
9453         if (flow->counter) {
9454                 uint64_t pkts, bytes;
9455                 struct mlx5_flow_counter *cnt;
9456
9457                 cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
9458                                                  NULL);
9459                 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
9460                                                &bytes);
9461
9462                 if (err)
9463                         return rte_flow_error_set(error, -err,
9464                                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9465                                         NULL, "cannot read counters");
9466                 qc->hits_set = 1;
9467                 qc->bytes_set = 1;
9468                 qc->hits = pkts - cnt->hits;
9469                 qc->bytes = bytes - cnt->bytes;
9470                 if (qc->reset) {
9471                         cnt->hits = pkts;
9472                         cnt->bytes = bytes;
9473                 }
9474                 return 0;
9475         }
9476         return rte_flow_error_set(error, EINVAL,
9477                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9478                                   NULL,
9479                                   "counters are not available");
9480 }
9481
9482 /**
9483  * Query a flow.
9484  *
9485  * @see rte_flow_query()
9486  * @see rte_flow_ops
9487  */
9488 static int
9489 flow_dv_query(struct rte_eth_dev *dev,
9490               struct rte_flow *flow __rte_unused,
9491               const struct rte_flow_action *actions __rte_unused,
9492               void *data __rte_unused,
9493               struct rte_flow_error *error __rte_unused)
9494 {
9495         int ret = -EINVAL;
9496
9497         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
9498                 switch (actions->type) {
9499                 case RTE_FLOW_ACTION_TYPE_VOID:
9500                         break;
9501                 case RTE_FLOW_ACTION_TYPE_COUNT:
9502                         ret = flow_dv_query_count(dev, flow, data, error);
9503                         break;
9504                 default:
9505                         return rte_flow_error_set(error, ENOTSUP,
9506                                                   RTE_FLOW_ERROR_TYPE_ACTION,
9507                                                   actions,
9508                                                   "action not supported");
9509                 }
9510         }
9511         return ret;
9512 }
9513
9514 /**
9515  * Destroy the meter table set.
9516  * Lock free, (mutex should be acquired by caller).
9517  *
9518  * @param[in] dev
9519  *   Pointer to Ethernet device.
9520  * @param[in] tbl
9521  *   Pointer to the meter table set.
9522  *
9523  * @return
9524  *   Always 0.
9525  */
9526 static int
9527 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
9528                         struct mlx5_meter_domains_infos *tbl)
9529 {
9530         struct mlx5_priv *priv = dev->data->dev_private;
9531         struct mlx5_meter_domains_infos *mtd =
9532                                 (struct mlx5_meter_domains_infos *)tbl;
9533
9534         if (!mtd || !priv->config.dv_flow_en)
9535                 return 0;
9536         if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
9537                 claim_zero(mlx5_flow_os_destroy_flow
9538                            (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
9539         if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
9540                 claim_zero(mlx5_flow_os_destroy_flow
9541                            (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
9542         if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
9543                 claim_zero(mlx5_flow_os_destroy_flow
9544                            (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
9545         if (mtd->egress.color_matcher)
9546                 claim_zero(mlx5_flow_os_destroy_flow_matcher
9547                            (mtd->egress.color_matcher));
9548         if (mtd->egress.any_matcher)
9549                 claim_zero(mlx5_flow_os_destroy_flow_matcher
9550                            (mtd->egress.any_matcher));
9551         if (mtd->egress.tbl)
9552                 flow_dv_tbl_resource_release(dev, mtd->egress.tbl);
9553         if (mtd->egress.sfx_tbl)
9554                 flow_dv_tbl_resource_release(dev, mtd->egress.sfx_tbl);
9555         if (mtd->ingress.color_matcher)
9556                 claim_zero(mlx5_flow_os_destroy_flow_matcher
9557                            (mtd->ingress.color_matcher));
9558         if (mtd->ingress.any_matcher)
9559                 claim_zero(mlx5_flow_os_destroy_flow_matcher
9560                            (mtd->ingress.any_matcher));
9561         if (mtd->ingress.tbl)
9562                 flow_dv_tbl_resource_release(dev, mtd->ingress.tbl);
9563         if (mtd->ingress.sfx_tbl)
9564                 flow_dv_tbl_resource_release(dev, mtd->ingress.sfx_tbl);
9565         if (mtd->transfer.color_matcher)
9566                 claim_zero(mlx5_flow_os_destroy_flow_matcher
9567                            (mtd->transfer.color_matcher));
9568         if (mtd->transfer.any_matcher)
9569                 claim_zero(mlx5_flow_os_destroy_flow_matcher
9570                            (mtd->transfer.any_matcher));
9571         if (mtd->transfer.tbl)
9572                 flow_dv_tbl_resource_release(dev, mtd->transfer.tbl);
9573         if (mtd->transfer.sfx_tbl)
9574                 flow_dv_tbl_resource_release(dev, mtd->transfer.sfx_tbl);
9575         if (mtd->drop_actn)
9576                 claim_zero(mlx5_flow_os_destroy_flow_action(mtd->drop_actn));
9577         mlx5_free(mtd);
9578         return 0;
9579 }
9580
9581 /* Number of meter flow actions, count and jump or count and drop. */
9582 #define METER_ACTIONS 2
9583
9584 /**
9585  * Create specify domain meter table and suffix table.
9586  *
9587  * @param[in] dev
9588  *   Pointer to Ethernet device.
9589  * @param[in,out] mtb
9590  *   Pointer to DV meter table set.
9591  * @param[in] egress
9592  *   Table attribute.
9593  * @param[in] transfer
9594  *   Table attribute.
9595  * @param[in] color_reg_c_idx
9596  *   Reg C index for color match.
9597  *
9598  * @return
9599  *   0 on success, -1 otherwise and rte_errno is set.
9600  */
9601 static int
9602 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
9603                            struct mlx5_meter_domains_infos *mtb,
9604                            uint8_t egress, uint8_t transfer,
9605                            uint32_t color_reg_c_idx)
9606 {
9607         struct mlx5_priv *priv = dev->data->dev_private;
9608         struct mlx5_dev_ctx_shared *sh = priv->sh;
9609         struct mlx5_flow_dv_match_params mask = {
9610                 .size = sizeof(mask.buf),
9611         };
9612         struct mlx5_flow_dv_match_params value = {
9613                 .size = sizeof(value.buf),
9614         };
9615         struct mlx5dv_flow_matcher_attr dv_attr = {
9616                 .type = IBV_FLOW_ATTR_NORMAL,
9617                 .priority = 0,
9618                 .match_criteria_enable = 0,
9619                 .match_mask = (void *)&mask,
9620         };
9621         void *actions[METER_ACTIONS];
9622         struct mlx5_meter_domain_info *dtb;
9623         struct rte_flow_error error;
9624         int i = 0;
9625         int ret;
9626
9627         if (transfer)
9628                 dtb = &mtb->transfer;
9629         else if (egress)
9630                 dtb = &mtb->egress;
9631         else
9632                 dtb = &mtb->ingress;
9633         /* Create the meter table with METER level. */
9634         dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
9635                                             egress, transfer, &error);
9636         if (!dtb->tbl) {
9637                 DRV_LOG(ERR, "Failed to create meter policer table.");
9638                 return -1;
9639         }
9640         /* Create the meter suffix table with SUFFIX level. */
9641         dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
9642                                             MLX5_FLOW_TABLE_LEVEL_SUFFIX,
9643                                             egress, transfer, &error);
9644         if (!dtb->sfx_tbl) {
9645                 DRV_LOG(ERR, "Failed to create meter suffix table.");
9646                 return -1;
9647         }
9648         /* Create matchers, Any and Color. */
9649         dv_attr.priority = 3;
9650         dv_attr.match_criteria_enable = 0;
9651         ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
9652                                                &dtb->any_matcher);
9653         if (ret) {
9654                 DRV_LOG(ERR, "Failed to create meter"
9655                              " policer default matcher.");
9656                 goto error_exit;
9657         }
9658         dv_attr.priority = 0;
9659         dv_attr.match_criteria_enable =
9660                                 1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
9661         flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
9662                                rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
9663         ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
9664                                                &dtb->color_matcher);
9665         if (ret) {
9666                 DRV_LOG(ERR, "Failed to create meter policer color matcher.");
9667                 goto error_exit;
9668         }
9669         if (mtb->count_actns[RTE_MTR_DROPPED])
9670                 actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
9671         actions[i++] = mtb->drop_actn;
9672         /* Default rule: lowest priority, match any, actions: drop. */
9673         ret = mlx5_flow_os_create_flow(dtb->any_matcher, (void *)&value, i,
9674                                        actions,
9675                                        &dtb->policer_rules[RTE_MTR_DROPPED]);
9676         if (ret) {
9677                 DRV_LOG(ERR, "Failed to create meter policer drop rule.");
9678                 goto error_exit;
9679         }
9680         return 0;
9681 error_exit:
9682         return -1;
9683 }
9684
9685 /**
9686  * Create the needed meter and suffix tables.
9687  * Lock free, (mutex should be acquired by caller).
9688  *
9689  * @param[in] dev
9690  *   Pointer to Ethernet device.
9691  * @param[in] fm
9692  *   Pointer to the flow meter.
9693  *
9694  * @return
9695  *   Pointer to table set on success, NULL otherwise and rte_errno is set.
9696  */
9697 static struct mlx5_meter_domains_infos *
9698 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
9699                        const struct mlx5_flow_meter *fm)
9700 {
9701         struct mlx5_priv *priv = dev->data->dev_private;
9702         struct mlx5_meter_domains_infos *mtb;
9703         int ret;
9704         int i;
9705
9706         if (!priv->mtr_en) {
9707                 rte_errno = ENOTSUP;
9708                 return NULL;
9709         }
9710         mtb = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*mtb), 0, SOCKET_ID_ANY);
9711         if (!mtb) {
9712                 DRV_LOG(ERR, "Failed to allocate memory for meter.");
9713                 return NULL;
9714         }
9715         /* Create meter count actions */
9716         for (i = 0; i <= RTE_MTR_DROPPED; i++) {
9717                 struct mlx5_flow_counter *cnt;
9718                 if (!fm->policer_stats.cnt[i])
9719                         continue;
9720                 cnt = flow_dv_counter_get_by_idx(dev,
9721                       fm->policer_stats.cnt[i], NULL);
9722                 mtb->count_actns[i] = cnt->action;
9723         }
9724         /* Create drop action. */
9725         ret = mlx5_flow_os_create_flow_action_drop(&mtb->drop_actn);
9726         if (ret) {
9727                 DRV_LOG(ERR, "Failed to create drop action.");
9728                 goto error_exit;
9729         }
9730         /* Egress meter table. */
9731         ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
9732         if (ret) {
9733                 DRV_LOG(ERR, "Failed to prepare egress meter table.");
9734                 goto error_exit;
9735         }
9736         /* Ingress meter table. */
9737         ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
9738         if (ret) {
9739                 DRV_LOG(ERR, "Failed to prepare ingress meter table.");
9740                 goto error_exit;
9741         }
9742         /* FDB meter table. */
9743         if (priv->config.dv_esw_en) {
9744                 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
9745                                                  priv->mtr_color_reg);
9746                 if (ret) {
9747                         DRV_LOG(ERR, "Failed to prepare fdb meter table.");
9748                         goto error_exit;
9749                 }
9750         }
9751         return mtb;
9752 error_exit:
9753         flow_dv_destroy_mtr_tbl(dev, mtb);
9754         return NULL;
9755 }
9756
9757 /**
9758  * Destroy domain policer rule.
9759  *
9760  * @param[in] dt
9761  *   Pointer to domain table.
9762  */
9763 static void
9764 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
9765 {
9766         int i;
9767
9768         for (i = 0; i < RTE_MTR_DROPPED; i++) {
9769                 if (dt->policer_rules[i]) {
9770                         claim_zero(mlx5_flow_os_destroy_flow
9771                                    (dt->policer_rules[i]));
9772                         dt->policer_rules[i] = NULL;
9773                 }
9774         }
9775         if (dt->jump_actn) {
9776                 claim_zero(mlx5_flow_os_destroy_flow_action(dt->jump_actn));
9777                 dt->jump_actn = NULL;
9778         }
9779 }
9780
9781 /**
9782  * Destroy policer rules.
9783  *
9784  * @param[in] dev
9785  *   Pointer to Ethernet device.
9786  * @param[in] fm
9787  *   Pointer to flow meter structure.
9788  * @param[in] attr
9789  *   Pointer to flow attributes.
9790  *
9791  * @return
9792  *   Always 0.
9793  */
9794 static int
9795 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
9796                               const struct mlx5_flow_meter *fm,
9797                               const struct rte_flow_attr *attr)
9798 {
9799         struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
9800
9801         if (!mtb)
9802                 return 0;
9803         if (attr->egress)
9804                 flow_dv_destroy_domain_policer_rule(&mtb->egress);
9805         if (attr->ingress)
9806                 flow_dv_destroy_domain_policer_rule(&mtb->ingress);
9807         if (attr->transfer)
9808                 flow_dv_destroy_domain_policer_rule(&mtb->transfer);
9809         return 0;
9810 }
9811
9812 /**
9813  * Create specify domain meter policer rule.
9814  *
9815  * @param[in] fm
9816  *   Pointer to flow meter structure.
9817  * @param[in] mtb
9818  *   Pointer to DV meter table set.
9819  * @param[in] mtr_reg_c
9820  *   Color match REG_C.
9821  *
9822  * @return
9823  *   0 on success, -1 otherwise.
9824  */
9825 static int
9826 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
9827                                     struct mlx5_meter_domain_info *dtb,
9828                                     uint8_t mtr_reg_c)
9829 {
9830         struct mlx5_flow_dv_match_params matcher = {
9831                 .size = sizeof(matcher.buf),
9832         };
9833         struct mlx5_flow_dv_match_params value = {
9834                 .size = sizeof(value.buf),
9835         };
9836         struct mlx5_meter_domains_infos *mtb = fm->mfts;
9837         void *actions[METER_ACTIONS];
9838         int i;
9839         int ret = 0;
9840
9841         /* Create jump action. */
9842         if (!dtb->jump_actn)
9843                 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
9844                                 (dtb->sfx_tbl->obj, &dtb->jump_actn);
9845         if (ret) {
9846                 DRV_LOG(ERR, "Failed to create policer jump action.");
9847                 goto error;
9848         }
9849         for (i = 0; i < RTE_MTR_DROPPED; i++) {
9850                 int j = 0;
9851
9852                 flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
9853                                        rte_col_2_mlx5_col(i), UINT8_MAX);
9854                 if (mtb->count_actns[i])
9855                         actions[j++] = mtb->count_actns[i];
9856                 if (fm->action[i] == MTR_POLICER_ACTION_DROP)
9857                         actions[j++] = mtb->drop_actn;
9858                 else
9859                         actions[j++] = dtb->jump_actn;
9860                 ret = mlx5_flow_os_create_flow(dtb->color_matcher,
9861                                                (void *)&value, j, actions,
9862                                                &dtb->policer_rules[i]);
9863                 if (ret) {
9864                         DRV_LOG(ERR, "Failed to create policer rule.");
9865                         goto error;
9866                 }
9867         }
9868         return 0;
9869 error:
9870         rte_errno = errno;
9871         return -1;
9872 }
9873
9874 /**
9875  * Create policer rules.
9876  *
9877  * @param[in] dev
9878  *   Pointer to Ethernet device.
9879  * @param[in] fm
9880  *   Pointer to flow meter structure.
9881  * @param[in] attr
9882  *   Pointer to flow attributes.
9883  *
9884  * @return
9885  *   0 on success, -1 otherwise.
9886  */
9887 static int
9888 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
9889                              struct mlx5_flow_meter *fm,
9890                              const struct rte_flow_attr *attr)
9891 {
9892         struct mlx5_priv *priv = dev->data->dev_private;
9893         struct mlx5_meter_domains_infos *mtb = fm->mfts;
9894         int ret;
9895
9896         if (attr->egress) {
9897                 ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
9898                                                 priv->mtr_color_reg);
9899                 if (ret) {
9900                         DRV_LOG(ERR, "Failed to create egress policer.");
9901                         goto error;
9902                 }
9903         }
9904         if (attr->ingress) {
9905                 ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
9906                                                 priv->mtr_color_reg);
9907                 if (ret) {
9908                         DRV_LOG(ERR, "Failed to create ingress policer.");
9909                         goto error;
9910                 }
9911         }
9912         if (attr->transfer) {
9913                 ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
9914                                                 priv->mtr_color_reg);
9915                 if (ret) {
9916                         DRV_LOG(ERR, "Failed to create transfer policer.");
9917                         goto error;
9918                 }
9919         }
9920         return 0;
9921 error:
9922         flow_dv_destroy_policer_rules(dev, fm, attr);
9923         return -1;
9924 }
9925
9926 /**
9927  * Query a devx counter.
9928  *
9929  * @param[in] dev
9930  *   Pointer to the Ethernet device structure.
9931  * @param[in] cnt
9932  *   Index to the flow counter.
9933  * @param[in] clear
9934  *   Set to clear the counter statistics.
9935  * @param[out] pkts
9936  *   The statistics value of packets.
9937  * @param[out] bytes
9938  *   The statistics value of bytes.
9939  *
9940  * @return
9941  *   0 on success, otherwise return -1.
9942  */
9943 static int
9944 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
9945                       uint64_t *pkts, uint64_t *bytes)
9946 {
9947         struct mlx5_priv *priv = dev->data->dev_private;
9948         struct mlx5_flow_counter *cnt;
9949         uint64_t inn_pkts, inn_bytes;
9950         int ret;
9951
9952         if (!priv->config.devx)
9953                 return -1;
9954
9955         ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
9956         if (ret)
9957                 return -1;
9958         cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
9959         *pkts = inn_pkts - cnt->hits;
9960         *bytes = inn_bytes - cnt->bytes;
9961         if (clear) {
9962                 cnt->hits = inn_pkts;
9963                 cnt->bytes = inn_bytes;
9964         }
9965         return 0;
9966 }
9967
9968 /**
9969  * Get aged-out flows.
9970  *
9971  * @param[in] dev
9972  *   Pointer to the Ethernet device structure.
9973  * @param[in] context
9974  *   The address of an array of pointers to the aged-out flows contexts.
9975  * @param[in] nb_contexts
9976  *   The length of context array pointers.
9977  * @param[out] error
9978  *   Perform verbose error reporting if not NULL. Initialized in case of
9979  *   error only.
9980  *
9981  * @return
9982  *   how many contexts get in success, otherwise negative errno value.
9983  *   if nb_contexts is 0, return the amount of all aged contexts.
9984  *   if nb_contexts is not 0 , return the amount of aged flows reported
9985  *   in the context array.
9986  * @note: only stub for now
9987  */
9988 static int
9989 flow_get_aged_flows(struct rte_eth_dev *dev,
9990                     void **context,
9991                     uint32_t nb_contexts,
9992                     struct rte_flow_error *error)
9993 {
9994         struct mlx5_priv *priv = dev->data->dev_private;
9995         struct mlx5_age_info *age_info;
9996         struct mlx5_age_param *age_param;
9997         struct mlx5_flow_counter *counter;
9998         int nb_flows = 0;
9999
10000         if (nb_contexts && !context)
10001                 return rte_flow_error_set(error, EINVAL,
10002                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10003                                           NULL,
10004                                           "Should assign at least one flow or"
10005                                           " context to get if nb_contexts != 0");
10006         age_info = GET_PORT_AGE_INFO(priv);
10007         rte_spinlock_lock(&age_info->aged_sl);
10008         TAILQ_FOREACH(counter, &age_info->aged_counters, next) {
10009                 nb_flows++;
10010                 if (nb_contexts) {
10011                         age_param = MLX5_CNT_TO_AGE(counter);
10012                         context[nb_flows - 1] = age_param->context;
10013                         if (!(--nb_contexts))
10014                                 break;
10015                 }
10016         }
10017         rte_spinlock_unlock(&age_info->aged_sl);
10018         MLX5_AGE_SET(age_info, MLX5_AGE_TRIGGER);
10019         return nb_flows;
10020 }
10021
10022 /*
10023  * Mutex-protected thunk to lock-free  __flow_dv_translate().
10024  */
10025 static int
10026 flow_dv_translate(struct rte_eth_dev *dev,
10027                   struct mlx5_flow *dev_flow,
10028                   const struct rte_flow_attr *attr,
10029                   const struct rte_flow_item items[],
10030                   const struct rte_flow_action actions[],
10031                   struct rte_flow_error *error)
10032 {
10033         int ret;
10034
10035         flow_dv_shared_lock(dev);
10036         ret = __flow_dv_translate(dev, dev_flow, attr, items, actions, error);
10037         flow_dv_shared_unlock(dev);
10038         return ret;
10039 }
10040
10041 /*
10042  * Mutex-protected thunk to lock-free  __flow_dv_apply().
10043  */
10044 static int
10045 flow_dv_apply(struct rte_eth_dev *dev,
10046               struct rte_flow *flow,
10047               struct rte_flow_error *error)
10048 {
10049         int ret;
10050
10051         flow_dv_shared_lock(dev);
10052         ret = __flow_dv_apply(dev, flow, error);
10053         flow_dv_shared_unlock(dev);
10054         return ret;
10055 }
10056
10057 /*
10058  * Mutex-protected thunk to lock-free __flow_dv_remove().
10059  */
10060 static void
10061 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
10062 {
10063         flow_dv_shared_lock(dev);
10064         __flow_dv_remove(dev, flow);
10065         flow_dv_shared_unlock(dev);
10066 }
10067
10068 /*
10069  * Mutex-protected thunk to lock-free __flow_dv_destroy().
10070  */
10071 static void
10072 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
10073 {
10074         flow_dv_shared_lock(dev);
10075         __flow_dv_destroy(dev, flow);
10076         flow_dv_shared_unlock(dev);
10077 }
10078
10079 /*
10080  * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
10081  */
10082 static uint32_t
10083 flow_dv_counter_allocate(struct rte_eth_dev *dev)
10084 {
10085         uint32_t cnt;
10086
10087         flow_dv_shared_lock(dev);
10088         cnt = flow_dv_counter_alloc(dev, 0, 0, 1, 0);
10089         flow_dv_shared_unlock(dev);
10090         return cnt;
10091 }
10092
10093 /*
10094  * Mutex-protected thunk to lock-free flow_dv_counter_release().
10095  */
10096 static void
10097 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
10098 {
10099         flow_dv_shared_lock(dev);
10100         flow_dv_counter_release(dev, cnt);
10101         flow_dv_shared_unlock(dev);
10102 }
10103
10104 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
10105         .validate = flow_dv_validate,
10106         .prepare = flow_dv_prepare,
10107         .translate = flow_dv_translate,
10108         .apply = flow_dv_apply,
10109         .remove = flow_dv_remove,
10110         .destroy = flow_dv_destroy,
10111         .query = flow_dv_query,
10112         .create_mtr_tbls = flow_dv_create_mtr_tbl,
10113         .destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
10114         .create_policer_rules = flow_dv_create_policer_rules,
10115         .destroy_policer_rules = flow_dv_destroy_policer_rules,
10116         .counter_alloc = flow_dv_counter_allocate,
10117         .counter_free = flow_dv_counter_free,
10118         .counter_query = flow_dv_counter_query,
10119         .get_aged_flows = flow_get_aged_flows,
10120 };
10121
10122 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
10123