drivers/net/mlx5/mlx5_flow.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2016 6WIND S.A.
   3  * Copyright 2016 Mellanox Technologies, Ltd
   4  */
   5
   6 #include <netinet/in.h>
   7 #include <sys/queue.h>
   8 #include <stdalign.h>
   9 #include <stdint.h>
  10 #include <string.h>
  11 #include <stdbool.h>
  12
  13 #include <rte_common.h>
  14 #include <rte_ether.h>
  15 #include <rte_ethdev_driver.h>
  16 #include <rte_eal_paging.h>
  17 #include <rte_flow.h>
  18 #include <rte_cycles.h>
  19 #include <rte_flow_driver.h>
  20 #include <rte_malloc.h>
  21 #include <rte_ip.h>
  22
  23 #include <mlx5_glue.h>
  24 #include <mlx5_devx_cmds.h>
  25 #include <mlx5_prm.h>
  26 #include <mlx5_malloc.h>
  27
  28 #include "mlx5_defs.h"
  29 #include "mlx5.h"
  30 #include "mlx5_flow.h"
  31 #include "mlx5_flow_os.h"
  32 #include "mlx5_rxtx.h"
  33 #include "mlx5_common_os.h"
  34 #include "rte_pmd_mlx5.h"
  35
  36 struct tunnel_default_miss_ctx {
  37         uint16_t *queue;
  38         __extension__
  39         union {
  40                 struct rte_flow_action_rss action_rss;
  41                 struct rte_flow_action_queue miss_queue;
  42                 struct rte_flow_action_jump miss_jump;
  43                 uint8_t raw[0];
  44         };
  45 };
  46
  47 static int
  48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
  49                              struct rte_flow *flow,
  50                              const struct rte_flow_attr *attr,
  51                              const struct rte_flow_action *app_actions,
  52                              uint32_t flow_idx,
  53                              struct tunnel_default_miss_ctx *ctx,
  54                              struct rte_flow_error *error);
  55 static struct mlx5_flow_tunnel *
  56 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
  57 static void
  58 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
  59 static uint32_t
  60 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
  61                                 const struct mlx5_flow_tunnel *tunnel,
  62                                 uint32_t group, uint32_t *table,
  63                                 struct rte_flow_error *error);
  64
  65 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
  66 static void mlx5_flow_pop_thread_workspace(void);
  67
  68
  69 /** Device flow drivers. */
  70 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
  71
  72 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
  73
  74 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
  75         [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
  76 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
  77         [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
  78 #endif
  79         [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
  80         [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
  81 };
  82
  83 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
  84 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
  85         (const int []){ \
  86                 __VA_ARGS__, 0, \
  87         }
  88
  89 /** Node object of input graph for mlx5_flow_expand_rss(). */
  90 struct mlx5_flow_expand_node {
  91         const int *const next;
  92         /**<
  93          * List of next node indexes. Index 0 is interpreted as a terminator.
  94          */
  95         const enum rte_flow_item_type type;
  96         /**< Pattern item type of current node. */
  97         uint64_t rss_types;
  98         /**<
  99          * RSS types bit-field associated with this node
 100          * (see ETH_RSS_* definitions).
 101          */
 102 };
 103
 104 /** Object returned by mlx5_flow_expand_rss(). */
 105 struct mlx5_flow_expand_rss {
 106         uint32_t entries;
 107         /**< Number of entries @p patterns and @p priorities. */
 108         struct {
 109                 struct rte_flow_item *pattern; /**< Expanded pattern array. */
 110                 uint32_t priority; /**< Priority offset for each expansion. */
 111         } entry[];
 112 };
 113
 114 static enum rte_flow_item_type
 115 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
 116 {
 117         enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
 118         uint16_t ether_type = 0;
 119         uint16_t ether_type_m;
 120         uint8_t ip_next_proto = 0;
 121         uint8_t ip_next_proto_m;
 122
 123         if (item == NULL || item->spec == NULL)
 124                 return ret;
 125         switch (item->type) {
 126         case RTE_FLOW_ITEM_TYPE_ETH:
 127                 if (item->mask)
 128                         ether_type_m = ((const struct rte_flow_item_eth *)
 129                                                 (item->mask))->type;
 130                 else
 131                         ether_type_m = rte_flow_item_eth_mask.type;
 132                 if (ether_type_m != RTE_BE16(0xFFFF))
 133                         break;
 134                 ether_type = ((const struct rte_flow_item_eth *)
 135                                 (item->spec))->type;
 136                 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
 137                         ret = RTE_FLOW_ITEM_TYPE_IPV4;
 138                 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
 139                         ret = RTE_FLOW_ITEM_TYPE_IPV6;
 140                 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
 141                         ret = RTE_FLOW_ITEM_TYPE_VLAN;
 142                 else
 143                         ret = RTE_FLOW_ITEM_TYPE_END;
 144                 break;
 145         case RTE_FLOW_ITEM_TYPE_VLAN:
 146                 if (item->mask)
 147                         ether_type_m = ((const struct rte_flow_item_vlan *)
 148                                                 (item->mask))->inner_type;
 149                 else
 150                         ether_type_m = rte_flow_item_vlan_mask.inner_type;
 151                 if (ether_type_m != RTE_BE16(0xFFFF))
 152                         break;
 153                 ether_type = ((const struct rte_flow_item_vlan *)
 154                                 (item->spec))->inner_type;
 155                 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
 156                         ret = RTE_FLOW_ITEM_TYPE_IPV4;
 157                 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
 158                         ret = RTE_FLOW_ITEM_TYPE_IPV6;
 159                 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
 160                         ret = RTE_FLOW_ITEM_TYPE_VLAN;
 161                 else
 162                         ret = RTE_FLOW_ITEM_TYPE_END;
 163                 break;
 164         case RTE_FLOW_ITEM_TYPE_IPV4:
 165                 if (item->mask)
 166                         ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
 167                                         (item->mask))->hdr.next_proto_id;
 168                 else
 169                         ip_next_proto_m =
 170                                 rte_flow_item_ipv4_mask.hdr.next_proto_id;
 171                 if (ip_next_proto_m != 0xFF)
 172                         break;
 173                 ip_next_proto = ((const struct rte_flow_item_ipv4 *)
 174                                 (item->spec))->hdr.next_proto_id;
 175                 if (ip_next_proto == IPPROTO_UDP)
 176                         ret = RTE_FLOW_ITEM_TYPE_UDP;
 177                 else if (ip_next_proto == IPPROTO_TCP)
 178                         ret = RTE_FLOW_ITEM_TYPE_TCP;
 179                 else if (ip_next_proto == IPPROTO_IP)
 180                         ret = RTE_FLOW_ITEM_TYPE_IPV4;
 181                 else if (ip_next_proto == IPPROTO_IPV6)
 182                         ret = RTE_FLOW_ITEM_TYPE_IPV6;
 183                 else
 184                         ret = RTE_FLOW_ITEM_TYPE_END;
 185                 break;
 186         case RTE_FLOW_ITEM_TYPE_IPV6:
 187                 if (item->mask)
 188                         ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
 189                                                 (item->mask))->hdr.proto;
 190                 else
 191                         ip_next_proto_m =
 192                                 rte_flow_item_ipv6_mask.hdr.proto;
 193                 if (ip_next_proto_m != 0xFF)
 194                         break;
 195                 ip_next_proto = ((const struct rte_flow_item_ipv6 *)
 196                                 (item->spec))->hdr.proto;
 197                 if (ip_next_proto == IPPROTO_UDP)
 198                         ret = RTE_FLOW_ITEM_TYPE_UDP;
 199                 else if (ip_next_proto == IPPROTO_TCP)
 200                         ret = RTE_FLOW_ITEM_TYPE_TCP;
 201                 else if (ip_next_proto == IPPROTO_IP)
 202                         ret = RTE_FLOW_ITEM_TYPE_IPV4;
 203                 else if (ip_next_proto == IPPROTO_IPV6)
 204                         ret = RTE_FLOW_ITEM_TYPE_IPV6;
 205                 else
 206                         ret = RTE_FLOW_ITEM_TYPE_END;
 207                 break;
 208         default:
 209                 ret = RTE_FLOW_ITEM_TYPE_VOID;
 210                 break;
 211         }
 212         return ret;
 213 }
 214
 215 #define MLX5_RSS_EXP_ELT_N 8
 216
 217 /**
 218  * Expand RSS flows into several possible flows according to the RSS hash
 219  * fields requested and the driver capabilities.
 220  *
 221  * @param[out] buf
 222  *   Buffer to store the result expansion.
 223  * @param[in] size
 224  *   Buffer size in bytes. If 0, @p buf can be NULL.
 225  * @param[in] pattern
 226  *   User flow pattern.
 227  * @param[in] types
 228  *   RSS types to expand (see ETH_RSS_* definitions).
 229  * @param[in] graph
 230  *   Input graph to expand @p pattern according to @p types.
 231  * @param[in] graph_root_index
 232  *   Index of root node in @p graph, typically 0.
 233  *
 234  * @return
 235  *   A positive value representing the size of @p buf in bytes regardless of
 236  *   @p size on success, a negative errno value otherwise and rte_errno is
 237  *   set, the following errors are defined:
 238  *
 239  *   -E2BIG: graph-depth @p graph is too deep.
 240  */
 241 static int
 242 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
 243                      const struct rte_flow_item *pattern, uint64_t types,
 244                      const struct mlx5_flow_expand_node graph[],
 245                      int graph_root_index)
 246 {
 247         const struct rte_flow_item *item;
 248         const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
 249         const int *next_node;
 250         const int *stack[MLX5_RSS_EXP_ELT_N];
 251         int stack_pos = 0;
 252         struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
 253         unsigned int i;
 254         size_t lsize;
 255         size_t user_pattern_size = 0;
 256         void *addr = NULL;
 257         const struct mlx5_flow_expand_node *next = NULL;
 258         struct rte_flow_item missed_item;
 259         int missed = 0;
 260         int elt = 0;
 261         const struct rte_flow_item *last_item = NULL;
 262
 263         memset(&missed_item, 0, sizeof(missed_item));
 264         lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
 265                 MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
 266         if (lsize <= size) {
 267                 buf->entry[0].priority = 0;
 268                 buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
 269                 buf->entries = 0;
 270                 addr = buf->entry[0].pattern;
 271         }
 272         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
 273                 if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
 274                         last_item = item;
 275                 for (i = 0; node->next && node->next[i]; ++i) {
 276                         next = &graph[node->next[i]];
 277                         if (next->type == item->type)
 278                                 break;
 279                 }
 280                 if (next)
 281                         node = next;
 282                 user_pattern_size += sizeof(*item);
 283         }
 284         user_pattern_size += sizeof(*item); /* Handle END item. */
 285         lsize += user_pattern_size;
 286         /* Copy the user pattern in the first entry of the buffer. */
 287         if (lsize <= size) {
 288                 rte_memcpy(addr, pattern, user_pattern_size);
 289                 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
 290                 buf->entries = 1;
 291         }
 292         /* Start expanding. */
 293         memset(flow_items, 0, sizeof(flow_items));
 294         user_pattern_size -= sizeof(*item);
 295         /*
 296          * Check if the last valid item has spec set, need complete pattern,
 297          * and the pattern can be used for expansion.
 298          */
 299         missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
 300         if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
 301                 /* Item type END indicates expansion is not required. */
 302                 return lsize;
 303         }
 304         if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
 305                 next = NULL;
 306                 missed = 1;
 307                 for (i = 0; node->next && node->next[i]; ++i) {
 308                         next = &graph[node->next[i]];
 309                         if (next->type == missed_item.type) {
 310                                 flow_items[0].type = missed_item.type;
 311                                 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
 312                                 break;
 313                         }
 314                         next = NULL;
 315                 }
 316         }
 317         if (next && missed) {
 318                 elt = 2; /* missed item + item end. */
 319                 node = next;
 320                 lsize += elt * sizeof(*item) + user_pattern_size;
 321                 if ((node->rss_types & types) && lsize <= size) {
 322                         buf->entry[buf->entries].priority = 1;
 323                         buf->entry[buf->entries].pattern = addr;
 324                         buf->entries++;
 325                         rte_memcpy(addr, buf->entry[0].pattern,
 326                                    user_pattern_size);
 327                         addr = (void *)(((uintptr_t)addr) + user_pattern_size);
 328                         rte_memcpy(addr, flow_items, elt * sizeof(*item));
 329                         addr = (void *)(((uintptr_t)addr) +
 330                                         elt * sizeof(*item));
 331                 }
 332         }
 333         memset(flow_items, 0, sizeof(flow_items));
 334         next_node = node->next;
 335         stack[stack_pos] = next_node;
 336         node = next_node ? &graph[*next_node] : NULL;
 337         while (node) {
 338                 flow_items[stack_pos].type = node->type;
 339                 if (node->rss_types & types) {
 340                         /*
 341                          * compute the number of items to copy from the
 342                          * expansion and copy it.
 343                          * When the stack_pos is 0, there are 1 element in it,
 344                          * plus the addition END item.
 345                          */
 346                         elt = stack_pos + 2;
 347                         flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
 348                         lsize += elt * sizeof(*item) + user_pattern_size;
 349                         if (lsize <= size) {
 350                                 size_t n = elt * sizeof(*item);
 351
 352                                 buf->entry[buf->entries].priority =
 353                                         stack_pos + 1 + missed;
 354                                 buf->entry[buf->entries].pattern = addr;
 355                                 buf->entries++;
 356                                 rte_memcpy(addr, buf->entry[0].pattern,
 357                                            user_pattern_size);
 358                                 addr = (void *)(((uintptr_t)addr) +
 359                                                 user_pattern_size);
 360                                 rte_memcpy(addr, &missed_item,
 361                                            missed * sizeof(*item));
 362                                 addr = (void *)(((uintptr_t)addr) +
 363                                         missed * sizeof(*item));
 364                                 rte_memcpy(addr, flow_items, n);
 365                                 addr = (void *)(((uintptr_t)addr) + n);
 366                         }
 367                 }
 368                 /* Go deeper. */
 369                 if (node->next) {
 370                         next_node = node->next;
 371                         if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
 372                                 rte_errno = E2BIG;
 373                                 return -rte_errno;
 374                         }
 375                         stack[stack_pos] = next_node;
 376                 } else if (*(next_node + 1)) {
 377                         /* Follow up with the next possibility. */
 378                         ++next_node;
 379                 } else {
 380                         /* Move to the next path. */
 381                         if (stack_pos)
 382                                 next_node = stack[--stack_pos];
 383                         next_node++;
 384                         stack[stack_pos] = next_node;
 385                 }
 386                 node = *next_node ? &graph[*next_node] : NULL;
 387         };
 388         /* no expanded flows but we have missed item, create one rule for it */
 389         if (buf->entries == 1 && missed != 0) {
 390                 elt = 2;
 391                 lsize += elt * sizeof(*item) + user_pattern_size;
 392                 if (lsize <= size) {
 393                         buf->entry[buf->entries].priority = 1;
 394                         buf->entry[buf->entries].pattern = addr;
 395                         buf->entries++;
 396                         flow_items[0].type = missed_item.type;
 397                         flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
 398                         rte_memcpy(addr, buf->entry[0].pattern,
 399                                    user_pattern_size);
 400                         addr = (void *)(((uintptr_t)addr) + user_pattern_size);
 401                         rte_memcpy(addr, flow_items, elt * sizeof(*item));
 402                 }
 403         }
 404         return lsize;
 405 }
 406
 407 enum mlx5_expansion {
 408         MLX5_EXPANSION_ROOT,
 409         MLX5_EXPANSION_ROOT_OUTER,
 410         MLX5_EXPANSION_ROOT_ETH_VLAN,
 411         MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
 412         MLX5_EXPANSION_OUTER_ETH,
 413         MLX5_EXPANSION_OUTER_ETH_VLAN,
 414         MLX5_EXPANSION_OUTER_VLAN,
 415         MLX5_EXPANSION_OUTER_IPV4,
 416         MLX5_EXPANSION_OUTER_IPV4_UDP,
 417         MLX5_EXPANSION_OUTER_IPV4_TCP,
 418         MLX5_EXPANSION_OUTER_IPV6,
 419         MLX5_EXPANSION_OUTER_IPV6_UDP,
 420         MLX5_EXPANSION_OUTER_IPV6_TCP,
 421         MLX5_EXPANSION_VXLAN,
 422         MLX5_EXPANSION_VXLAN_GPE,
 423         MLX5_EXPANSION_GRE,
 424         MLX5_EXPANSION_MPLS,
 425         MLX5_EXPANSION_ETH,
 426         MLX5_EXPANSION_ETH_VLAN,
 427         MLX5_EXPANSION_VLAN,
 428         MLX5_EXPANSION_IPV4,
 429         MLX5_EXPANSION_IPV4_UDP,
 430         MLX5_EXPANSION_IPV4_TCP,
 431         MLX5_EXPANSION_IPV6,
 432         MLX5_EXPANSION_IPV6_UDP,
 433         MLX5_EXPANSION_IPV6_TCP,
 434 };
 435
 436 /** Supported expansion of items. */
 437 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
 438         [MLX5_EXPANSION_ROOT] = {
 439                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
 440                                                   MLX5_EXPANSION_IPV4,
 441                                                   MLX5_EXPANSION_IPV6),
 442                 .type = RTE_FLOW_ITEM_TYPE_END,
 443         },
 444         [MLX5_EXPANSION_ROOT_OUTER] = {
 445                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
 446                                                   MLX5_EXPANSION_OUTER_IPV4,
 447                                                   MLX5_EXPANSION_OUTER_IPV6),
 448                 .type = RTE_FLOW_ITEM_TYPE_END,
 449         },
 450         [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
 451                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
 452                 .type = RTE_FLOW_ITEM_TYPE_END,
 453         },
 454         [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
 455                 .next = MLX5_FLOW_EXPAND_RSS_NEXT
 456                                                 (MLX5_EXPANSION_OUTER_ETH_VLAN),
 457                 .type = RTE_FLOW_ITEM_TYPE_END,
 458         },
 459         [MLX5_EXPANSION_OUTER_ETH] = {
 460                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
 461                                                   MLX5_EXPANSION_OUTER_IPV6,
 462                                                   MLX5_EXPANSION_MPLS),
 463                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 464                 .rss_types = 0,
 465         },
 466         [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
 467                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
 468                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 469                 .rss_types = 0,
 470         },
 471         [MLX5_EXPANSION_OUTER_VLAN] = {
 472                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
 473                                                   MLX5_EXPANSION_OUTER_IPV6),
 474                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
 475         },
 476         [MLX5_EXPANSION_OUTER_IPV4] = {
 477                 .next = MLX5_FLOW_EXPAND_RSS_NEXT
 478                         (MLX5_EXPANSION_OUTER_IPV4_UDP,
 479                          MLX5_EXPANSION_OUTER_IPV4_TCP,
 480                          MLX5_EXPANSION_GRE,
 481                          MLX5_EXPANSION_IPV4,
 482                          MLX5_EXPANSION_IPV6),
 483                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
 484                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
 485                         ETH_RSS_NONFRAG_IPV4_OTHER,
 486         },
 487         [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
 488                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
 489                                                   MLX5_EXPANSION_VXLAN_GPE),
 490                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 491                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
 492         },
 493         [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
 494                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 495                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
 496         },
 497         [MLX5_EXPANSION_OUTER_IPV6] = {
 498                 .next = MLX5_FLOW_EXPAND_RSS_NEXT
 499                         (MLX5_EXPANSION_OUTER_IPV6_UDP,
 500                          MLX5_EXPANSION_OUTER_IPV6_TCP,
 501                          MLX5_EXPANSION_IPV4,
 502                          MLX5_EXPANSION_IPV6),
 503                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
 504                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
 505                         ETH_RSS_NONFRAG_IPV6_OTHER,
 506         },
 507         [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
 508                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
 509                                                   MLX5_EXPANSION_VXLAN_GPE),
 510                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 511                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
 512         },
 513         [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
 514                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 515                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
 516         },
 517         [MLX5_EXPANSION_VXLAN] = {
 518                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
 519                                                   MLX5_EXPANSION_IPV4,
 520                                                   MLX5_EXPANSION_IPV6),
 521                 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
 522         },
 523         [MLX5_EXPANSION_VXLAN_GPE] = {
 524                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
 525                                                   MLX5_EXPANSION_IPV4,
 526                                                   MLX5_EXPANSION_IPV6),
 527                 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
 528         },
 529         [MLX5_EXPANSION_GRE] = {
 530                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
 531                 .type = RTE_FLOW_ITEM_TYPE_GRE,
 532         },
 533         [MLX5_EXPANSION_MPLS] = {
 534                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 535                                                   MLX5_EXPANSION_IPV6),
 536                 .type = RTE_FLOW_ITEM_TYPE_MPLS,
 537         },
 538         [MLX5_EXPANSION_ETH] = {
 539                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 540                                                   MLX5_EXPANSION_IPV6),
 541                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 542         },
 543         [MLX5_EXPANSION_ETH_VLAN] = {
 544                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
 545                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 546         },
 547         [MLX5_EXPANSION_VLAN] = {
 548                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
 549                                                   MLX5_EXPANSION_IPV6),
 550                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
 551         },
 552         [MLX5_EXPANSION_IPV4] = {
 553                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
 554                                                   MLX5_EXPANSION_IPV4_TCP),
 555                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
 556                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
 557                         ETH_RSS_NONFRAG_IPV4_OTHER,
 558         },
 559         [MLX5_EXPANSION_IPV4_UDP] = {
 560                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 561                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
 562         },
 563         [MLX5_EXPANSION_IPV4_TCP] = {
 564                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 565                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
 566         },
 567         [MLX5_EXPANSION_IPV6] = {
 568                 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
 569                                                   MLX5_EXPANSION_IPV6_TCP),
 570                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
 571                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
 572                         ETH_RSS_NONFRAG_IPV6_OTHER,
 573         },
 574         [MLX5_EXPANSION_IPV6_UDP] = {
 575                 .type = RTE_FLOW_ITEM_TYPE_UDP,
 576                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
 577         },
 578         [MLX5_EXPANSION_IPV6_TCP] = {
 579                 .type = RTE_FLOW_ITEM_TYPE_TCP,
 580                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
 581         },
 582 };
 583
 584 static struct rte_flow_shared_action *
 585 mlx5_shared_action_create(struct rte_eth_dev *dev,
 586                           const struct rte_flow_shared_action_conf *conf,
 587                           const struct rte_flow_action *action,
 588                           struct rte_flow_error *error);
 589 static int mlx5_shared_action_destroy
 590                                 (struct rte_eth_dev *dev,
 591                                  struct rte_flow_shared_action *shared_action,
 592                                  struct rte_flow_error *error);
 593 static int mlx5_shared_action_update
 594                                 (struct rte_eth_dev *dev,
 595                                  struct rte_flow_shared_action *shared_action,
 596                                  const struct rte_flow_action *action,
 597                                  struct rte_flow_error *error);
 598 static int mlx5_shared_action_query
 599                                 (struct rte_eth_dev *dev,
 600                                  const struct rte_flow_shared_action *action,
 601                                  void *data,
 602                                  struct rte_flow_error *error);
 603 static int
 604 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
 605                     struct rte_flow_tunnel *app_tunnel,
 606                     struct rte_flow_action **actions,
 607                     uint32_t *num_of_actions,
 608                     struct rte_flow_error *error);
 609 static int
 610 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
 611                        struct rte_flow_tunnel *app_tunnel,
 612                        struct rte_flow_item **items,
 613                        uint32_t *num_of_items,
 614                        struct rte_flow_error *error);
 615 static int
 616 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
 617                               struct rte_flow_item *pmd_items,
 618                               uint32_t num_items, struct rte_flow_error *err);
 619 static int
 620 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
 621                                 struct rte_flow_action *pmd_actions,
 622                                 uint32_t num_actions,
 623                                 struct rte_flow_error *err);
 624 static int
 625 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
 626                                   struct rte_mbuf *m,
 627                                   struct rte_flow_restore_info *info,
 628                                   struct rte_flow_error *err);
 629
 630 static const struct rte_flow_ops mlx5_flow_ops = {
 631         .validate = mlx5_flow_validate,
 632         .create = mlx5_flow_create,
 633         .destroy = mlx5_flow_destroy,
 634         .flush = mlx5_flow_flush,
 635         .isolate = mlx5_flow_isolate,
 636         .query = mlx5_flow_query,
 637         .dev_dump = mlx5_flow_dev_dump,
 638         .get_aged_flows = mlx5_flow_get_aged_flows,
 639         .shared_action_create = mlx5_shared_action_create,
 640         .shared_action_destroy = mlx5_shared_action_destroy,
 641         .shared_action_update = mlx5_shared_action_update,
 642         .shared_action_query = mlx5_shared_action_query,
 643         .tunnel_decap_set = mlx5_flow_tunnel_decap_set,
 644         .tunnel_match = mlx5_flow_tunnel_match,
 645         .tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
 646         .tunnel_item_release = mlx5_flow_tunnel_item_release,
 647         .get_restore_info = mlx5_flow_tunnel_get_restore_info,
 648 };
 649
 650 /* Tunnel information. */
 651 struct mlx5_flow_tunnel_info {
 652         uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
 653         uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
 654 };
 655
 656 static struct mlx5_flow_tunnel_info tunnels_info[] = {
 657         {
 658                 .tunnel = MLX5_FLOW_LAYER_VXLAN,
 659                 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
 660         },
 661         {
 662                 .tunnel = MLX5_FLOW_LAYER_GENEVE,
 663                 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
 664         },
 665         {
 666                 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
 667                 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
 668         },
 669         {
 670                 .tunnel = MLX5_FLOW_LAYER_GRE,
 671                 .ptype = RTE_PTYPE_TUNNEL_GRE,
 672         },
 673         {
 674                 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
 675                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
 676         },
 677         {
 678                 .tunnel = MLX5_FLOW_LAYER_MPLS,
 679                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
 680         },
 681         {
 682                 .tunnel = MLX5_FLOW_LAYER_NVGRE,
 683                 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
 684         },
 685         {
 686                 .tunnel = MLX5_FLOW_LAYER_IPIP,
 687                 .ptype = RTE_PTYPE_TUNNEL_IP,
 688         },
 689         {
 690                 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
 691                 .ptype = RTE_PTYPE_TUNNEL_IP,
 692         },
 693         {
 694                 .tunnel = MLX5_FLOW_LAYER_GTP,
 695                 .ptype = RTE_PTYPE_TUNNEL_GTPU,
 696         },
 697 };
 698
 699
 700
 701 /**
 702  * Translate tag ID to register.
 703  *
 704  * @param[in] dev
 705  *   Pointer to the Ethernet device structure.
 706  * @param[in] feature
 707  *   The feature that request the register.
 708  * @param[in] id
 709  *   The request register ID.
 710  * @param[out] error
 711  *   Error description in case of any.
 712  *
 713  * @return
 714  *   The request register on success, a negative errno
 715  *   value otherwise and rte_errno is set.
 716  */
 717 int
 718 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
 719                      enum mlx5_feature_name feature,
 720                      uint32_t id,
 721                      struct rte_flow_error *error)
 722 {
 723         struct mlx5_priv *priv = dev->data->dev_private;
 724         struct mlx5_dev_config *config = &priv->config;
 725         enum modify_reg start_reg;
 726         bool skip_mtr_reg = false;
 727
 728         switch (feature) {
 729         case MLX5_HAIRPIN_RX:
 730                 return REG_B;
 731         case MLX5_HAIRPIN_TX:
 732                 return REG_A;
 733         case MLX5_METADATA_RX:
 734                 switch (config->dv_xmeta_en) {
 735                 case MLX5_XMETA_MODE_LEGACY:
 736                         return REG_B;
 737                 case MLX5_XMETA_MODE_META16:
 738                         return REG_C_0;
 739                 case MLX5_XMETA_MODE_META32:
 740                         return REG_C_1;
 741                 }
 742                 break;
 743         case MLX5_METADATA_TX:
 744                 return REG_A;
 745         case MLX5_METADATA_FDB:
 746                 switch (config->dv_xmeta_en) {
 747                 case MLX5_XMETA_MODE_LEGACY:
 748                         return REG_NON;
 749                 case MLX5_XMETA_MODE_META16:
 750                         return REG_C_0;
 751                 case MLX5_XMETA_MODE_META32:
 752                         return REG_C_1;
 753                 }
 754                 break;
 755         case MLX5_FLOW_MARK:
 756                 switch (config->dv_xmeta_en) {
 757                 case MLX5_XMETA_MODE_LEGACY:
 758                         return REG_NON;
 759                 case MLX5_XMETA_MODE_META16:
 760                         return REG_C_1;
 761                 case MLX5_XMETA_MODE_META32:
 762                         return REG_C_0;
 763                 }
 764                 break;
 765         case MLX5_MTR_SFX:
 766                 /*
 767                  * If meter color and flow match share one register, flow match
 768                  * should use the meter color register for match.
 769                  */
 770                 if (priv->mtr_reg_share)
 771                         return priv->mtr_color_reg;
 772                 else
 773                         return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
 774                                REG_C_3;
 775         case MLX5_MTR_COLOR:
 776         case MLX5_ASO_FLOW_HIT: /* Both features use the same REG_C. */
 777                 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
 778                 return priv->mtr_color_reg;
 779         case MLX5_COPY_MARK:
 780                 /*
 781                  * Metadata COPY_MARK register using is in meter suffix sub
 782                  * flow while with meter. It's safe to share the same register.
 783                  */
 784                 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
 785         case MLX5_APP_TAG:
 786                 /*
 787                  * If meter is enable, it will engage the register for color
 788                  * match and flow match. If meter color match is not using the
 789                  * REG_C_2, need to skip the REG_C_x be used by meter color
 790                  * match.
 791                  * If meter is disable, free to use all available registers.
 792                  */
 793                 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
 794                             (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
 795                 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
 796                 if (id > (uint32_t)(REG_C_7 - start_reg))
 797                         return rte_flow_error_set(error, EINVAL,
 798                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 799                                                   NULL, "invalid tag id");
 800                 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
 801                         return rte_flow_error_set(error, ENOTSUP,
 802                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 803                                                   NULL, "unsupported tag id");
 804                 /*
 805                  * This case means meter is using the REG_C_x great than 2.
 806                  * Take care not to conflict with meter color REG_C_x.
 807                  * If the available index REG_C_y >= REG_C_x, skip the
 808                  * color register.
 809                  */
 810                 if (skip_mtr_reg && config->flow_mreg_c
 811                     [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
 812                         if (id >= (uint32_t)(REG_C_7 - start_reg))
 813                                 return rte_flow_error_set(error, EINVAL,
 814                                                        RTE_FLOW_ERROR_TYPE_ITEM,
 815                                                         NULL, "invalid tag id");
 816                         if (config->flow_mreg_c
 817                             [id + 1 + start_reg - REG_C_0] != REG_NON)
 818                                 return config->flow_mreg_c
 819                                                [id + 1 + start_reg - REG_C_0];
 820                         return rte_flow_error_set(error, ENOTSUP,
 821                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 822                                                   NULL, "unsupported tag id");
 823                 }
 824                 return config->flow_mreg_c[id + start_reg - REG_C_0];
 825         }
 826         MLX5_ASSERT(false);
 827         return rte_flow_error_set(error, EINVAL,
 828                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
 829                                   NULL, "invalid feature name");
 830 }
 831
 832 /**
 833  * Check extensive flow metadata register support.
 834  *
 835  * @param dev
 836  *   Pointer to rte_eth_dev structure.
 837  *
 838  * @return
 839  *   True if device supports extensive flow metadata register, otherwise false.
 840  */
 841 bool
 842 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
 843 {
 844         struct mlx5_priv *priv = dev->data->dev_private;
 845         struct mlx5_dev_config *config = &priv->config;
 846
 847         /*
 848          * Having available reg_c can be regarded inclusively as supporting
 849          * extensive flow metadata register, which could mean,
 850          * - metadata register copy action by modify header.
 851          * - 16 modify header actions is supported.
 852          * - reg_c's are preserved across different domain (FDB and NIC) on
 853          *   packet loopback by flow lookup miss.
 854          */
 855         return config->flow_mreg_c[2] != REG_NON;
 856 }
 857
 858 /**
 859  * Verify the @p item specifications (spec, last, mask) are compatible with the
 860  * NIC capabilities.
 861  *
 862  * @param[in] item
 863  *   Item specification.
 864  * @param[in] mask
 865  *   @p item->mask or flow default bit-masks.
 866  * @param[in] nic_mask
 867  *   Bit-masks covering supported fields by the NIC to compare with user mask.
 868  * @param[in] size
 869  *   Bit-masks size in bytes.
 870  * @param[in] range_accepted
 871  *   True if range of values is accepted for specific fields, false otherwise.
 872  * @param[out] error
 873  *   Pointer to error structure.
 874  *
 875  * @return
 876  *   0 on success, a negative errno value otherwise and rte_errno is set.
 877  */
 878 int
 879 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
 880                           const uint8_t *mask,
 881                           const uint8_t *nic_mask,
 882                           unsigned int size,
 883                           bool range_accepted,
 884                           struct rte_flow_error *error)
 885 {
 886         unsigned int i;
 887
 888         MLX5_ASSERT(nic_mask);
 889         for (i = 0; i < size; ++i)
 890                 if ((nic_mask[i] | mask[i]) != nic_mask[i])
 891                         return rte_flow_error_set(error, ENOTSUP,
 892                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 893                                                   item,
 894                                                   "mask enables non supported"
 895                                                   " bits");
 896         if (!item->spec && (item->mask || item->last))
 897                 return rte_flow_error_set(error, EINVAL,
 898                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
 899                                           "mask/last without a spec is not"
 900                                           " supported");
 901         if (item->spec && item->last && !range_accepted) {
 902                 uint8_t spec[size];
 903                 uint8_t last[size];
 904                 unsigned int i;
 905                 int ret;
 906
 907                 for (i = 0; i < size; ++i) {
 908                         spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
 909                         last[i] = ((const uint8_t *)item->last)[i] & mask[i];
 910                 }
 911                 ret = memcmp(spec, last, size);
 912                 if (ret != 0)
 913                         return rte_flow_error_set(error, EINVAL,
 914                                                   RTE_FLOW_ERROR_TYPE_ITEM,
 915                                                   item,
 916                                                   "range is not valid");
 917         }
 918         return 0;
 919 }
 920
 921 /**
 922  * Adjust the hash fields according to the @p flow information.
 923  *
 924  * @param[in] dev_flow.
 925  *   Pointer to the mlx5_flow.
 926  * @param[in] tunnel
 927  *   1 when the hash field is for a tunnel item.
 928  * @param[in] layer_types
 929  *   ETH_RSS_* types.
 930  * @param[in] hash_fields
 931  *   Item hash fields.
 932  *
 933  * @return
 934  *   The hash fields that should be used.
 935  */
 936 uint64_t
 937 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
 938                             int tunnel __rte_unused, uint64_t layer_types,
 939                             uint64_t hash_fields)
 940 {
 941 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
 942         int rss_request_inner = rss_desc->level >= 2;
 943
 944         /* Check RSS hash level for tunnel. */
 945         if (tunnel && rss_request_inner)
 946                 hash_fields |= IBV_RX_HASH_INNER;
 947         else if (tunnel || rss_request_inner)
 948                 return 0;
 949 #endif
 950         /* Check if requested layer matches RSS hash fields. */
 951         if (!(rss_desc->types & layer_types))
 952                 return 0;
 953         return hash_fields;
 954 }
 955
 956 /**
 957  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
 958  * if several tunnel rules are used on this queue, the tunnel ptype will be
 959  * cleared.
 960  *
 961  * @param rxq_ctrl
 962  *   Rx queue to update.
 963  */
 964 static void
 965 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
 966 {
 967         unsigned int i;
 968         uint32_t tunnel_ptype = 0;
 969
 970         /* Look up for the ptype to use. */
 971         for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
 972                 if (!rxq_ctrl->flow_tunnels_n[i])
 973                         continue;
 974                 if (!tunnel_ptype) {
 975                         tunnel_ptype = tunnels_info[i].ptype;
 976                 } else {
 977                         tunnel_ptype = 0;
 978                         break;
 979                 }
 980         }
 981         rxq_ctrl->rxq.tunnel = tunnel_ptype;
 982 }
 983
 984 /**
 985  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
 986  * flow.
 987  *
 988  * @param[in] dev
 989  *   Pointer to the Ethernet device structure.
 990  * @param[in] dev_handle
 991  *   Pointer to device flow handle structure.
 992  */
 993 static void
 994 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
 995                        struct mlx5_flow_handle *dev_handle)
 996 {
 997         struct mlx5_priv *priv = dev->data->dev_private;
 998         const int mark = dev_handle->mark;
 999         const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1000         struct mlx5_ind_table_obj *ind_tbl = NULL;
1001         unsigned int i;
1002
1003         if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1004                 struct mlx5_hrxq *hrxq;
1005
1006                 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1007                               dev_handle->rix_hrxq);
1008                 if (hrxq)
1009                         ind_tbl = hrxq->ind_table;
1010         } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1011                 struct mlx5_shared_action_rss *shared_rss;
1012
1013                 shared_rss = mlx5_ipool_get
1014                         (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1015                          dev_handle->rix_srss);
1016                 if (shared_rss)
1017                         ind_tbl = shared_rss->ind_tbl;
1018         }
1019         if (!ind_tbl)
1020                 return;
1021         for (i = 0; i != ind_tbl->queues_n; ++i) {
1022                 int idx = ind_tbl->queues[i];
1023                 struct mlx5_rxq_ctrl *rxq_ctrl =
1024                         container_of((*priv->rxqs)[idx],
1025                                      struct mlx5_rxq_ctrl, rxq);
1026
1027                 /*
1028                  * To support metadata register copy on Tx loopback,
1029                  * this must be always enabled (metadata may arive
1030                  * from other port - not from local flows only.
1031                  */
1032                 if (priv->config.dv_flow_en &&
1033                     priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1034                     mlx5_flow_ext_mreg_supported(dev)) {
1035                         rxq_ctrl->rxq.mark = 1;
1036                         rxq_ctrl->flow_mark_n = 1;
1037                 } else if (mark) {
1038                         rxq_ctrl->rxq.mark = 1;
1039                         rxq_ctrl->flow_mark_n++;
1040                 }
1041                 if (tunnel) {
1042                         unsigned int j;
1043
1044                         /* Increase the counter matching the flow. */
1045                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1046                                 if ((tunnels_info[j].tunnel &
1047                                      dev_handle->layers) ==
1048                                     tunnels_info[j].tunnel) {
1049                                         rxq_ctrl->flow_tunnels_n[j]++;
1050                                         break;
1051                                 }
1052                         }
1053                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
1054                 }
1055         }
1056 }
1057
1058 /**
1059  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1060  *
1061  * @param[in] dev
1062  *   Pointer to the Ethernet device structure.
1063  * @param[in] flow
1064  *   Pointer to flow structure.
1065  */
1066 static void
1067 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1068 {
1069         struct mlx5_priv *priv = dev->data->dev_private;
1070         uint32_t handle_idx;
1071         struct mlx5_flow_handle *dev_handle;
1072
1073         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1074                        handle_idx, dev_handle, next)
1075                 flow_drv_rxq_flags_set(dev, dev_handle);
1076 }
1077
1078 /**
1079  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1080  * device flow if no other flow uses it with the same kind of request.
1081  *
1082  * @param dev
1083  *   Pointer to Ethernet device.
1084  * @param[in] dev_handle
1085  *   Pointer to the device flow handle structure.
1086  */
1087 static void
1088 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1089                         struct mlx5_flow_handle *dev_handle)
1090 {
1091         struct mlx5_priv *priv = dev->data->dev_private;
1092         const int mark = dev_handle->mark;
1093         const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1094         struct mlx5_ind_table_obj *ind_tbl = NULL;
1095         unsigned int i;
1096
1097         if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1098                 struct mlx5_hrxq *hrxq;
1099
1100                 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1101                               dev_handle->rix_hrxq);
1102                 if (hrxq)
1103                         ind_tbl = hrxq->ind_table;
1104         } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1105                 struct mlx5_shared_action_rss *shared_rss;
1106
1107                 shared_rss = mlx5_ipool_get
1108                         (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1109                          dev_handle->rix_srss);
1110                 if (shared_rss)
1111                         ind_tbl = shared_rss->ind_tbl;
1112         }
1113         if (!ind_tbl)
1114                 return;
1115         MLX5_ASSERT(dev->data->dev_started);
1116         for (i = 0; i != ind_tbl->queues_n; ++i) {
1117                 int idx = ind_tbl->queues[i];
1118                 struct mlx5_rxq_ctrl *rxq_ctrl =
1119                         container_of((*priv->rxqs)[idx],
1120                                      struct mlx5_rxq_ctrl, rxq);
1121
1122                 if (priv->config.dv_flow_en &&
1123                     priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1124                     mlx5_flow_ext_mreg_supported(dev)) {
1125                         rxq_ctrl->rxq.mark = 1;
1126                         rxq_ctrl->flow_mark_n = 1;
1127                 } else if (mark) {
1128                         rxq_ctrl->flow_mark_n--;
1129                         rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1130                 }
1131                 if (tunnel) {
1132                         unsigned int j;
1133
1134                         /* Decrease the counter matching the flow. */
1135                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1136                                 if ((tunnels_info[j].tunnel &
1137                                      dev_handle->layers) ==
1138                                     tunnels_info[j].tunnel) {
1139                                         rxq_ctrl->flow_tunnels_n[j]--;
1140                                         break;
1141                                 }
1142                         }
1143                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
1144                 }
1145         }
1146 }
1147
1148 /**
1149  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1150  * @p flow if no other flow uses it with the same kind of request.
1151  *
1152  * @param dev
1153  *   Pointer to Ethernet device.
1154  * @param[in] flow
1155  *   Pointer to the flow.
1156  */
1157 static void
1158 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1159 {
1160         struct mlx5_priv *priv = dev->data->dev_private;
1161         uint32_t handle_idx;
1162         struct mlx5_flow_handle *dev_handle;
1163
1164         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1165                        handle_idx, dev_handle, next)
1166                 flow_drv_rxq_flags_trim(dev, dev_handle);
1167 }
1168
1169 /**
1170  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1171  *
1172  * @param dev
1173  *   Pointer to Ethernet device.
1174  */
1175 static void
1176 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1177 {
1178         struct mlx5_priv *priv = dev->data->dev_private;
1179         unsigned int i;
1180
1181         for (i = 0; i != priv->rxqs_n; ++i) {
1182                 struct mlx5_rxq_ctrl *rxq_ctrl;
1183                 unsigned int j;
1184
1185                 if (!(*priv->rxqs)[i])
1186                         continue;
1187                 rxq_ctrl = container_of((*priv->rxqs)[i],
1188                                         struct mlx5_rxq_ctrl, rxq);
1189                 rxq_ctrl->flow_mark_n = 0;
1190                 rxq_ctrl->rxq.mark = 0;
1191                 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1192                         rxq_ctrl->flow_tunnels_n[j] = 0;
1193                 rxq_ctrl->rxq.tunnel = 0;
1194         }
1195 }
1196
1197 /**
1198  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1199  *
1200  * @param[in] dev
1201  *   Pointer to the Ethernet device structure.
1202  */
1203 void
1204 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1205 {
1206         struct mlx5_priv *priv = dev->data->dev_private;
1207         struct mlx5_rxq_data *data;
1208         unsigned int i;
1209
1210         for (i = 0; i != priv->rxqs_n; ++i) {
1211                 if (!(*priv->rxqs)[i])
1212                         continue;
1213                 data = (*priv->rxqs)[i];
1214                 if (!rte_flow_dynf_metadata_avail()) {
1215                         data->dynf_meta = 0;
1216                         data->flow_meta_mask = 0;
1217                         data->flow_meta_offset = -1;
1218                 } else {
1219                         data->dynf_meta = 1;
1220                         data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1221                         data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1222                 }
1223         }
1224 }
1225
1226 /*
1227  * return a pointer to the desired action in the list of actions.
1228  *
1229  * @param[in] actions
1230  *   The list of actions to search the action in.
1231  * @param[in] action
1232  *   The action to find.
1233  *
1234  * @return
1235  *   Pointer to the action in the list, if found. NULL otherwise.
1236  */
1237 const struct rte_flow_action *
1238 mlx5_flow_find_action(const struct rte_flow_action *actions,
1239                       enum rte_flow_action_type action)
1240 {
1241         if (actions == NULL)
1242                 return NULL;
1243         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1244                 if (actions->type == action)
1245                         return actions;
1246         return NULL;
1247 }
1248
1249 /*
1250  * Validate the flag action.
1251  *
1252  * @param[in] action_flags
1253  *   Bit-fields that holds the actions detected until now.
1254  * @param[in] attr
1255  *   Attributes of flow that includes this action.
1256  * @param[out] error
1257  *   Pointer to error structure.
1258  *
1259  * @return
1260  *   0 on success, a negative errno value otherwise and rte_errno is set.
1261  */
1262 int
1263 mlx5_flow_validate_action_flag(uint64_t action_flags,
1264                                const struct rte_flow_attr *attr,
1265                                struct rte_flow_error *error)
1266 {
1267         if (action_flags & MLX5_FLOW_ACTION_MARK)
1268                 return rte_flow_error_set(error, EINVAL,
1269                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1270                                           "can't mark and flag in same flow");
1271         if (action_flags & MLX5_FLOW_ACTION_FLAG)
1272                 return rte_flow_error_set(error, EINVAL,
1273                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1274                                           "can't have 2 flag"
1275                                           " actions in same flow");
1276         if (attr->egress)
1277                 return rte_flow_error_set(error, ENOTSUP,
1278                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1279                                           "flag action not supported for "
1280                                           "egress");
1281         return 0;
1282 }
1283
1284 /*
1285  * Validate the mark action.
1286  *
1287  * @param[in] action
1288  *   Pointer to the queue action.
1289  * @param[in] action_flags
1290  *   Bit-fields that holds the actions detected until now.
1291  * @param[in] attr
1292  *   Attributes of flow that includes this action.
1293  * @param[out] error
1294  *   Pointer to error structure.
1295  *
1296  * @return
1297  *   0 on success, a negative errno value otherwise and rte_errno is set.
1298  */
1299 int
1300 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1301                                uint64_t action_flags,
1302                                const struct rte_flow_attr *attr,
1303                                struct rte_flow_error *error)
1304 {
1305         const struct rte_flow_action_mark *mark = action->conf;
1306
1307         if (!mark)
1308                 return rte_flow_error_set(error, EINVAL,
1309                                           RTE_FLOW_ERROR_TYPE_ACTION,
1310                                           action,
1311                                           "configuration cannot be null");
1312         if (mark->id >= MLX5_FLOW_MARK_MAX)
1313                 return rte_flow_error_set(error, EINVAL,
1314                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1315                                           &mark->id,
1316                                           "mark id must in 0 <= id < "
1317                                           RTE_STR(MLX5_FLOW_MARK_MAX));
1318         if (action_flags & MLX5_FLOW_ACTION_FLAG)
1319                 return rte_flow_error_set(error, EINVAL,
1320                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1321                                           "can't flag and mark in same flow");
1322         if (action_flags & MLX5_FLOW_ACTION_MARK)
1323                 return rte_flow_error_set(error, EINVAL,
1324                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1325                                           "can't have 2 mark actions in same"
1326                                           " flow");
1327         if (attr->egress)
1328                 return rte_flow_error_set(error, ENOTSUP,
1329                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1330                                           "mark action not supported for "
1331                                           "egress");
1332         return 0;
1333 }
1334
1335 /*
1336  * Validate the drop action.
1337  *
1338  * @param[in] action_flags
1339  *   Bit-fields that holds the actions detected until now.
1340  * @param[in] attr
1341  *   Attributes of flow that includes this action.
1342  * @param[out] error
1343  *   Pointer to error structure.
1344  *
1345  * @return
1346  *   0 on success, a negative errno value otherwise and rte_errno is set.
1347  */
1348 int
1349 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1350                                const struct rte_flow_attr *attr,
1351                                struct rte_flow_error *error)
1352 {
1353         if (attr->egress)
1354                 return rte_flow_error_set(error, ENOTSUP,
1355                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1356                                           "drop action not supported for "
1357                                           "egress");
1358         return 0;
1359 }
1360
1361 /*
1362  * Validate the queue action.
1363  *
1364  * @param[in] action
1365  *   Pointer to the queue action.
1366  * @param[in] action_flags
1367  *   Bit-fields that holds the actions detected until now.
1368  * @param[in] dev
1369  *   Pointer to the Ethernet device structure.
1370  * @param[in] attr
1371  *   Attributes of flow that includes this action.
1372  * @param[out] error
1373  *   Pointer to error structure.
1374  *
1375  * @return
1376  *   0 on success, a negative errno value otherwise and rte_errno is set.
1377  */
1378 int
1379 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1380                                 uint64_t action_flags,
1381                                 struct rte_eth_dev *dev,
1382                                 const struct rte_flow_attr *attr,
1383                                 struct rte_flow_error *error)
1384 {
1385         struct mlx5_priv *priv = dev->data->dev_private;
1386         const struct rte_flow_action_queue *queue = action->conf;
1387
1388         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1389                 return rte_flow_error_set(error, EINVAL,
1390                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1391                                           "can't have 2 fate actions in"
1392                                           " same flow");
1393         if (!priv->rxqs_n)
1394                 return rte_flow_error_set(error, EINVAL,
1395                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1396                                           NULL, "No Rx queues configured");
1397         if (queue->index >= priv->rxqs_n)
1398                 return rte_flow_error_set(error, EINVAL,
1399                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1400                                           &queue->index,
1401                                           "queue index out of range");
1402         if (!(*priv->rxqs)[queue->index])
1403                 return rte_flow_error_set(error, EINVAL,
1404                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1405                                           &queue->index,
1406                                           "queue is not configured");
1407         if (attr->egress)
1408                 return rte_flow_error_set(error, ENOTSUP,
1409                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1410                                           "queue action not supported for "
1411                                           "egress");
1412         return 0;
1413 }
1414
1415 /*
1416  * Validate the rss action.
1417  *
1418  * @param[in] dev
1419  *   Pointer to the Ethernet device structure.
1420  * @param[in] action
1421  *   Pointer to the queue action.
1422  * @param[out] error
1423  *   Pointer to error structure.
1424  *
1425  * @return
1426  *   0 on success, a negative errno value otherwise and rte_errno is set.
1427  */
1428 int
1429 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1430                          const struct rte_flow_action *action,
1431                          struct rte_flow_error *error)
1432 {
1433         struct mlx5_priv *priv = dev->data->dev_private;
1434         const struct rte_flow_action_rss *rss = action->conf;
1435         enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1436         unsigned int i;
1437
1438         if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1439             rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1440                 return rte_flow_error_set(error, ENOTSUP,
1441                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1442                                           &rss->func,
1443                                           "RSS hash function not supported");
1444 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1445         if (rss->level > 2)
1446 #else
1447         if (rss->level > 1)
1448 #endif
1449                 return rte_flow_error_set(error, ENOTSUP,
1450                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1451                                           &rss->level,
1452                                           "tunnel RSS is not supported");
1453         /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1454         if (rss->key_len == 0 && rss->key != NULL)
1455                 return rte_flow_error_set(error, ENOTSUP,
1456                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1457                                           &rss->key_len,
1458                                           "RSS hash key length 0");
1459         if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1460                 return rte_flow_error_set(error, ENOTSUP,
1461                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1462                                           &rss->key_len,
1463                                           "RSS hash key too small");
1464         if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1465                 return rte_flow_error_set(error, ENOTSUP,
1466                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1467                                           &rss->key_len,
1468                                           "RSS hash key too large");
1469         if (rss->queue_num > priv->config.ind_table_max_size)
1470                 return rte_flow_error_set(error, ENOTSUP,
1471                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1472                                           &rss->queue_num,
1473                                           "number of queues too large");
1474         if (rss->types & MLX5_RSS_HF_MASK)
1475                 return rte_flow_error_set(error, ENOTSUP,
1476                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1477                                           &rss->types,
1478                                           "some RSS protocols are not"
1479                                           " supported");
1480         if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1481             !(rss->types & ETH_RSS_IP))
1482                 return rte_flow_error_set(error, EINVAL,
1483                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1484                                           "L3 partial RSS requested but L3 RSS"
1485                                           " type not specified");
1486         if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1487             !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1488                 return rte_flow_error_set(error, EINVAL,
1489                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1490                                           "L4 partial RSS requested but L4 RSS"
1491                                           " type not specified");
1492         if (!priv->rxqs_n)
1493                 return rte_flow_error_set(error, EINVAL,
1494                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1495                                           NULL, "No Rx queues configured");
1496         if (!rss->queue_num)
1497                 return rte_flow_error_set(error, EINVAL,
1498                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1499                                           NULL, "No queues configured");
1500         for (i = 0; i != rss->queue_num; ++i) {
1501                 struct mlx5_rxq_ctrl *rxq_ctrl;
1502
1503                 if (rss->queue[i] >= priv->rxqs_n)
1504                         return rte_flow_error_set
1505                                 (error, EINVAL,
1506                                  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1507                                  &rss->queue[i], "queue index out of range");
1508                 if (!(*priv->rxqs)[rss->queue[i]])
1509                         return rte_flow_error_set
1510                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1511                                  &rss->queue[i], "queue is not configured");
1512                 rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1513                                         struct mlx5_rxq_ctrl, rxq);
1514                 if (i == 0)
1515                         rxq_type = rxq_ctrl->type;
1516                 if (rxq_type != rxq_ctrl->type)
1517                         return rte_flow_error_set
1518                                 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1519                                  &rss->queue[i],
1520                                  "combining hairpin and regular RSS queues is not supported");
1521         }
1522         return 0;
1523 }
1524
1525 /*
1526  * Validate the rss action.
1527  *
1528  * @param[in] action
1529  *   Pointer to the queue action.
1530  * @param[in] action_flags
1531  *   Bit-fields that holds the actions detected until now.
1532  * @param[in] dev
1533  *   Pointer to the Ethernet device structure.
1534  * @param[in] attr
1535  *   Attributes of flow that includes this action.
1536  * @param[in] item_flags
1537  *   Items that were detected.
1538  * @param[out] error
1539  *   Pointer to error structure.
1540  *
1541  * @return
1542  *   0 on success, a negative errno value otherwise and rte_errno is set.
1543  */
1544 int
1545 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1546                               uint64_t action_flags,
1547                               struct rte_eth_dev *dev,
1548                               const struct rte_flow_attr *attr,
1549                               uint64_t item_flags,
1550                               struct rte_flow_error *error)
1551 {
1552         const struct rte_flow_action_rss *rss = action->conf;
1553         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1554         int ret;
1555
1556         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1557                 return rte_flow_error_set(error, EINVAL,
1558                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1559                                           "can't have 2 fate actions"
1560                                           " in same flow");
1561         ret = mlx5_validate_action_rss(dev, action, error);
1562         if (ret)
1563                 return ret;
1564         if (attr->egress)
1565                 return rte_flow_error_set(error, ENOTSUP,
1566                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1567                                           "rss action not supported for "
1568                                           "egress");
1569         if (rss->level > 1 && !tunnel)
1570                 return rte_flow_error_set(error, EINVAL,
1571                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1572                                           "inner RSS is not supported for "
1573                                           "non-tunnel flows");
1574         if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1575             !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1576                 return rte_flow_error_set(error, EINVAL,
1577                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1578                                           "RSS on eCPRI is not supported now");
1579         }
1580         return 0;
1581 }
1582
1583 /*
1584  * Validate the default miss action.
1585  *
1586  * @param[in] action_flags
1587  *   Bit-fields that holds the actions detected until now.
1588  * @param[out] error
1589  *   Pointer to error structure.
1590  *
1591  * @return
1592  *   0 on success, a negative errno value otherwise and rte_errno is set.
1593  */
1594 int
1595 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1596                                 const struct rte_flow_attr *attr,
1597                                 struct rte_flow_error *error)
1598 {
1599         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1600                 return rte_flow_error_set(error, EINVAL,
1601                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1602                                           "can't have 2 fate actions in"
1603                                           " same flow");
1604         if (attr->egress)
1605                 return rte_flow_error_set(error, ENOTSUP,
1606                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1607                                           "default miss action not supported "
1608                                           "for egress");
1609         if (attr->group)
1610                 return rte_flow_error_set(error, ENOTSUP,
1611                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1612                                           "only group 0 is supported");
1613         if (attr->transfer)
1614                 return rte_flow_error_set(error, ENOTSUP,
1615                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1616                                           NULL, "transfer is not supported");
1617         return 0;
1618 }
1619
1620 /*
1621  * Validate the count action.
1622  *
1623  * @param[in] dev
1624  *   Pointer to the Ethernet device structure.
1625  * @param[in] attr
1626  *   Attributes of flow that includes this action.
1627  * @param[out] error
1628  *   Pointer to error structure.
1629  *
1630  * @return
1631  *   0 on success, a negative errno value otherwise and rte_errno is set.
1632  */
1633 int
1634 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1635                                 const struct rte_flow_attr *attr,
1636                                 struct rte_flow_error *error)
1637 {
1638         if (attr->egress)
1639                 return rte_flow_error_set(error, ENOTSUP,
1640                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1641                                           "count action not supported for "
1642                                           "egress");
1643         return 0;
1644 }
1645
1646 /**
1647  * Verify the @p attributes will be correctly understood by the NIC and store
1648  * them in the @p flow if everything is correct.
1649  *
1650  * @param[in] dev
1651  *   Pointer to the Ethernet device structure.
1652  * @param[in] attributes
1653  *   Pointer to flow attributes
1654  * @param[out] error
1655  *   Pointer to error structure.
1656  *
1657  * @return
1658  *   0 on success, a negative errno value otherwise and rte_errno is set.
1659  */
1660 int
1661 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1662                               const struct rte_flow_attr *attributes,
1663                               struct rte_flow_error *error)
1664 {
1665         struct mlx5_priv *priv = dev->data->dev_private;
1666         uint32_t priority_max = priv->config.flow_prio - 1;
1667
1668         if (attributes->group)
1669                 return rte_flow_error_set(error, ENOTSUP,
1670                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1671                                           NULL, "groups is not supported");
1672         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1673             attributes->priority >= priority_max)
1674                 return rte_flow_error_set(error, ENOTSUP,
1675                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1676                                           NULL, "priority out of range");
1677         if (attributes->egress)
1678                 return rte_flow_error_set(error, ENOTSUP,
1679                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1680                                           "egress is not supported");
1681         if (attributes->transfer && !priv->config.dv_esw_en)
1682                 return rte_flow_error_set(error, ENOTSUP,
1683                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1684                                           NULL, "transfer is not supported");
1685         if (!attributes->ingress)
1686                 return rte_flow_error_set(error, EINVAL,
1687                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1688                                           NULL,
1689                                           "ingress attribute is mandatory");
1690         return 0;
1691 }
1692
1693 /**
1694  * Validate ICMP6 item.
1695  *
1696  * @param[in] item
1697  *   Item specification.
1698  * @param[in] item_flags
1699  *   Bit-fields that holds the items detected until now.
1700  * @param[in] ext_vlan_sup
1701  *   Whether extended VLAN features are supported or not.
1702  * @param[out] error
1703  *   Pointer to error structure.
1704  *
1705  * @return
1706  *   0 on success, a negative errno value otherwise and rte_errno is set.
1707  */
1708 int
1709 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1710                                uint64_t item_flags,
1711                                uint8_t target_protocol,
1712                                struct rte_flow_error *error)
1713 {
1714         const struct rte_flow_item_icmp6 *mask = item->mask;
1715         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1716         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1717                                       MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1718         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1719                                       MLX5_FLOW_LAYER_OUTER_L4;
1720         int ret;
1721
1722         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1723                 return rte_flow_error_set(error, EINVAL,
1724                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1725                                           "protocol filtering not compatible"
1726                                           " with ICMP6 layer");
1727         if (!(item_flags & l3m))
1728                 return rte_flow_error_set(error, EINVAL,
1729                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1730                                           "IPv6 is mandatory to filter on"
1731                                           " ICMP6");
1732         if (item_flags & l4m)
1733                 return rte_flow_error_set(error, EINVAL,
1734                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1735                                           "multiple L4 layers not supported");
1736         if (!mask)
1737                 mask = &rte_flow_item_icmp6_mask;
1738         ret = mlx5_flow_item_acceptable
1739                 (item, (const uint8_t *)mask,
1740                  (const uint8_t *)&rte_flow_item_icmp6_mask,
1741                  sizeof(struct rte_flow_item_icmp6),
1742                  MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1743         if (ret < 0)
1744                 return ret;
1745         return 0;
1746 }
1747
1748 /**
1749  * Validate ICMP item.
1750  *
1751  * @param[in] item
1752  *   Item specification.
1753  * @param[in] item_flags
1754  *   Bit-fields that holds the items detected until now.
1755  * @param[out] error
1756  *   Pointer to error structure.
1757  *
1758  * @return
1759  *   0 on success, a negative errno value otherwise and rte_errno is set.
1760  */
1761 int
1762 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1763                              uint64_t item_flags,
1764                              uint8_t target_protocol,
1765                              struct rte_flow_error *error)
1766 {
1767         const struct rte_flow_item_icmp *mask = item->mask;
1768         const struct rte_flow_item_icmp nic_mask = {
1769                 .hdr.icmp_type = 0xff,
1770                 .hdr.icmp_code = 0xff,
1771                 .hdr.icmp_ident = RTE_BE16(0xffff),
1772                 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
1773         };
1774         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1775         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1776                                       MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1777         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1778                                       MLX5_FLOW_LAYER_OUTER_L4;
1779         int ret;
1780
1781         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1782                 return rte_flow_error_set(error, EINVAL,
1783                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1784                                           "protocol filtering not compatible"
1785                                           " with ICMP layer");
1786         if (!(item_flags & l3m))
1787                 return rte_flow_error_set(error, EINVAL,
1788                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1789                                           "IPv4 is mandatory to filter"
1790                                           " on ICMP");
1791         if (item_flags & l4m)
1792                 return rte_flow_error_set(error, EINVAL,
1793                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1794                                           "multiple L4 layers not supported");
1795         if (!mask)
1796                 mask = &nic_mask;
1797         ret = mlx5_flow_item_acceptable
1798                 (item, (const uint8_t *)mask,
1799                  (const uint8_t *)&nic_mask,
1800                  sizeof(struct rte_flow_item_icmp),
1801                  MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1802         if (ret < 0)
1803                 return ret;
1804         return 0;
1805 }
1806
1807 /**
1808  * Validate Ethernet item.
1809  *
1810  * @param[in] item
1811  *   Item specification.
1812  * @param[in] item_flags
1813  *   Bit-fields that holds the items detected until now.
1814  * @param[out] error
1815  *   Pointer to error structure.
1816  *
1817  * @return
1818  *   0 on success, a negative errno value otherwise and rte_errno is set.
1819  */
1820 int
1821 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1822                             uint64_t item_flags, bool ext_vlan_sup,
1823                             struct rte_flow_error *error)
1824 {
1825         const struct rte_flow_item_eth *mask = item->mask;
1826         const struct rte_flow_item_eth nic_mask = {
1827                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1828                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1829                 .type = RTE_BE16(0xffff),
1830                 .has_vlan = ext_vlan_sup ? 1 : 0,
1831         };
1832         int ret;
1833         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1834         const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1835                                        MLX5_FLOW_LAYER_OUTER_L2;
1836
1837         if (item_flags & ethm)
1838                 return rte_flow_error_set(error, ENOTSUP,
1839                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1840                                           "multiple L2 layers not supported");
1841         if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1842             (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1843                 return rte_flow_error_set(error, EINVAL,
1844                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1845                                           "L2 layer should not follow "
1846                                           "L3 layers");
1847         if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1848             (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1849                 return rte_flow_error_set(error, EINVAL,
1850                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1851                                           "L2 layer should not follow VLAN");
1852         if (!mask)
1853                 mask = &rte_flow_item_eth_mask;
1854         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1855                                         (const uint8_t *)&nic_mask,
1856                                         sizeof(struct rte_flow_item_eth),
1857                                         MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1858         return ret;
1859 }
1860
1861 /**
1862  * Validate VLAN item.
1863  *
1864  * @param[in] item
1865  *   Item specification.
1866  * @param[in] item_flags
1867  *   Bit-fields that holds the items detected until now.
1868  * @param[in] dev
1869  *   Ethernet device flow is being created on.
1870  * @param[out] error
1871  *   Pointer to error structure.
1872  *
1873  * @return
1874  *   0 on success, a negative errno value otherwise and rte_errno is set.
1875  */
1876 int
1877 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1878                              uint64_t item_flags,
1879                              struct rte_eth_dev *dev,
1880                              struct rte_flow_error *error)
1881 {
1882         const struct rte_flow_item_vlan *spec = item->spec;
1883         const struct rte_flow_item_vlan *mask = item->mask;
1884         const struct rte_flow_item_vlan nic_mask = {
1885                 .tci = RTE_BE16(UINT16_MAX),
1886                 .inner_type = RTE_BE16(UINT16_MAX),
1887         };
1888         uint16_t vlan_tag = 0;
1889         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1890         int ret;
1891         const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1892                                         MLX5_FLOW_LAYER_INNER_L4) :
1893                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1894                                         MLX5_FLOW_LAYER_OUTER_L4);
1895         const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1896                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1897
1898         if (item_flags & vlanm)
1899                 return rte_flow_error_set(error, EINVAL,
1900                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1901                                           "multiple VLAN layers not supported");
1902         else if ((item_flags & l34m) != 0)
1903                 return rte_flow_error_set(error, EINVAL,
1904                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1905                                           "VLAN cannot follow L3/L4 layer");
1906         if (!mask)
1907                 mask = &rte_flow_item_vlan_mask;
1908         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1909                                         (const uint8_t *)&nic_mask,
1910                                         sizeof(struct rte_flow_item_vlan),
1911                                         MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1912         if (ret)
1913                 return ret;
1914         if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1915                 struct mlx5_priv *priv = dev->data->dev_private;
1916
1917                 if (priv->vmwa_context) {
1918                         /*
1919                          * Non-NULL context means we have a virtual machine
1920                          * and SR-IOV enabled, we have to create VLAN interface
1921                          * to make hypervisor to setup E-Switch vport
1922                          * context correctly. We avoid creating the multiple
1923                          * VLAN interfaces, so we cannot support VLAN tag mask.
1924                          */
1925                         return rte_flow_error_set(error, EINVAL,
1926                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1927                                                   item,
1928                                                   "VLAN tag mask is not"
1929                                                   " supported in virtual"
1930                                                   " environment");
1931                 }
1932         }
1933         if (spec) {
1934                 vlan_tag = spec->tci;
1935                 vlan_tag &= mask->tci;
1936         }
1937         /*
1938          * From verbs perspective an empty VLAN is equivalent
1939          * to a packet without VLAN layer.
1940          */
1941         if (!vlan_tag)
1942                 return rte_flow_error_set(error, EINVAL,
1943                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1944                                           item->spec,
1945                                           "VLAN cannot be empty");
1946         return 0;
1947 }
1948
1949 /**
1950  * Validate IPV4 item.
1951  *
1952  * @param[in] item
1953  *   Item specification.
1954  * @param[in] item_flags
1955  *   Bit-fields that holds the items detected until now.
1956  * @param[in] last_item
1957  *   Previous validated item in the pattern items.
1958  * @param[in] ether_type
1959  *   Type in the ethernet layer header (including dot1q).
1960  * @param[in] acc_mask
1961  *   Acceptable mask, if NULL default internal default mask
1962  *   will be used to check whether item fields are supported.
1963  * @param[in] range_accepted
1964  *   True if range of values is accepted for specific fields, false otherwise.
1965  * @param[out] error
1966  *   Pointer to error structure.
1967  *
1968  * @return
1969  *   0 on success, a negative errno value otherwise and rte_errno is set.
1970  */
1971 int
1972 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1973                              uint64_t item_flags,
1974                              uint64_t last_item,
1975                              uint16_t ether_type,
1976                              const struct rte_flow_item_ipv4 *acc_mask,
1977                              bool range_accepted,
1978                              struct rte_flow_error *error)
1979 {
1980         const struct rte_flow_item_ipv4 *mask = item->mask;
1981         const struct rte_flow_item_ipv4 *spec = item->spec;
1982         const struct rte_flow_item_ipv4 nic_mask = {
1983                 .hdr = {
1984                         .src_addr = RTE_BE32(0xffffffff),
1985                         .dst_addr = RTE_BE32(0xffffffff),
1986                         .type_of_service = 0xff,
1987                         .next_proto_id = 0xff,
1988                 },
1989         };
1990         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1991         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1992                                       MLX5_FLOW_LAYER_OUTER_L3;
1993         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1994                                       MLX5_FLOW_LAYER_OUTER_L4;
1995         int ret;
1996         uint8_t next_proto = 0xFF;
1997         const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1998                                   MLX5_FLOW_LAYER_OUTER_VLAN |
1999                                   MLX5_FLOW_LAYER_INNER_VLAN);
2000
2001         if ((last_item & l2_vlan) && ether_type &&
2002             ether_type != RTE_ETHER_TYPE_IPV4)
2003                 return rte_flow_error_set(error, EINVAL,
2004                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2005                                           "IPv4 cannot follow L2/VLAN layer "
2006                                           "which ether type is not IPv4");
2007         if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2008                 if (mask && spec)
2009                         next_proto = mask->hdr.next_proto_id &
2010                                      spec->hdr.next_proto_id;
2011                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2012                         return rte_flow_error_set(error, EINVAL,
2013                                                   RTE_FLOW_ERROR_TYPE_ITEM,
2014                                                   item,
2015                                                   "multiple tunnel "
2016                                                   "not supported");
2017         }
2018         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2019                 return rte_flow_error_set(error, EINVAL,
2020                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2021                                           "wrong tunnel type - IPv6 specified "
2022                                           "but IPv4 item provided");
2023         if (item_flags & l3m)
2024                 return rte_flow_error_set(error, ENOTSUP,
2025                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2026                                           "multiple L3 layers not supported");
2027         else if (item_flags & l4m)
2028                 return rte_flow_error_set(error, EINVAL,
2029                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2030                                           "L3 cannot follow an L4 layer.");
2031         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2032                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2033                 return rte_flow_error_set(error, EINVAL,
2034                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2035                                           "L3 cannot follow an NVGRE layer.");
2036         if (!mask)
2037                 mask = &rte_flow_item_ipv4_mask;
2038         else if (mask->hdr.next_proto_id != 0 &&
2039                  mask->hdr.next_proto_id != 0xff)
2040                 return rte_flow_error_set(error, EINVAL,
2041                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2042                                           "partial mask is not supported"
2043                                           " for protocol");
2044         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2045                                         acc_mask ? (const uint8_t *)acc_mask
2046                                                  : (const uint8_t *)&nic_mask,
2047                                         sizeof(struct rte_flow_item_ipv4),
2048                                         range_accepted, error);
2049         if (ret < 0)
2050                 return ret;
2051         return 0;
2052 }
2053
2054 /**
2055  * Validate IPV6 item.
2056  *
2057  * @param[in] item
2058  *   Item specification.
2059  * @param[in] item_flags
2060  *   Bit-fields that holds the items detected until now.
2061  * @param[in] last_item
2062  *   Previous validated item in the pattern items.
2063  * @param[in] ether_type
2064  *   Type in the ethernet layer header (including dot1q).
2065  * @param[in] acc_mask
2066  *   Acceptable mask, if NULL default internal default mask
2067  *   will be used to check whether item fields are supported.
2068  * @param[out] error
2069  *   Pointer to error structure.
2070  *
2071  * @return
2072  *   0 on success, a negative errno value otherwise and rte_errno is set.
2073  */
2074 int
2075 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2076                              uint64_t item_flags,
2077                              uint64_t last_item,
2078                              uint16_t ether_type,
2079                              const struct rte_flow_item_ipv6 *acc_mask,
2080                              struct rte_flow_error *error)
2081 {
2082         const struct rte_flow_item_ipv6 *mask = item->mask;
2083         const struct rte_flow_item_ipv6 *spec = item->spec;
2084         const struct rte_flow_item_ipv6 nic_mask = {
2085                 .hdr = {
2086                         .src_addr =
2087                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
2088                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
2089                         .dst_addr =
2090                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
2091                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
2092                         .vtc_flow = RTE_BE32(0xffffffff),
2093                         .proto = 0xff,
2094                 },
2095         };
2096         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2097         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2098                                       MLX5_FLOW_LAYER_OUTER_L3;
2099         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2100                                       MLX5_FLOW_LAYER_OUTER_L4;
2101         int ret;
2102         uint8_t next_proto = 0xFF;
2103         const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2104                                   MLX5_FLOW_LAYER_OUTER_VLAN |
2105                                   MLX5_FLOW_LAYER_INNER_VLAN);
2106
2107         if ((last_item & l2_vlan) && ether_type &&
2108             ether_type != RTE_ETHER_TYPE_IPV6)
2109                 return rte_flow_error_set(error, EINVAL,
2110                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2111                                           "IPv6 cannot follow L2/VLAN layer "
2112                                           "which ether type is not IPv6");
2113         if (mask && mask->hdr.proto == UINT8_MAX && spec)
2114                 next_proto = spec->hdr.proto;
2115         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2116                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2117                         return rte_flow_error_set(error, EINVAL,
2118                                                   RTE_FLOW_ERROR_TYPE_ITEM,
2119                                                   item,
2120                                                   "multiple tunnel "
2121                                                   "not supported");
2122         }
2123         if (next_proto == IPPROTO_HOPOPTS  ||
2124             next_proto == IPPROTO_ROUTING  ||
2125             next_proto == IPPROTO_FRAGMENT ||
2126             next_proto == IPPROTO_ESP      ||
2127             next_proto == IPPROTO_AH       ||
2128             next_proto == IPPROTO_DSTOPTS)
2129                 return rte_flow_error_set(error, EINVAL,
2130                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2131                                           "IPv6 proto (next header) should "
2132                                           "not be set as extension header");
2133         if (item_flags & MLX5_FLOW_LAYER_IPIP)
2134                 return rte_flow_error_set(error, EINVAL,
2135                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2136                                           "wrong tunnel type - IPv4 specified "
2137                                           "but IPv6 item provided");
2138         if (item_flags & l3m)
2139                 return rte_flow_error_set(error, ENOTSUP,
2140                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2141                                           "multiple L3 layers not supported");
2142         else if (item_flags & l4m)
2143                 return rte_flow_error_set(error, EINVAL,
2144                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2145                                           "L3 cannot follow an L4 layer.");
2146         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2147                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2148                 return rte_flow_error_set(error, EINVAL,
2149                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2150                                           "L3 cannot follow an NVGRE layer.");
2151         if (!mask)
2152                 mask = &rte_flow_item_ipv6_mask;
2153         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2154                                         acc_mask ? (const uint8_t *)acc_mask
2155                                                  : (const uint8_t *)&nic_mask,
2156                                         sizeof(struct rte_flow_item_ipv6),
2157                                         MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2158         if (ret < 0)
2159                 return ret;
2160         return 0;
2161 }
2162
2163 /**
2164  * Validate UDP item.
2165  *
2166  * @param[in] item
2167  *   Item specification.
2168  * @param[in] item_flags
2169  *   Bit-fields that holds the items detected until now.
2170  * @param[in] target_protocol
2171  *   The next protocol in the previous item.
2172  * @param[in] flow_mask
2173  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2174  * @param[out] error
2175  *   Pointer to error structure.
2176  *
2177  * @return
2178  *   0 on success, a negative errno value otherwise and rte_errno is set.
2179  */
2180 int
2181 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2182                             uint64_t item_flags,
2183                             uint8_t target_protocol,
2184                             struct rte_flow_error *error)
2185 {
2186         const struct rte_flow_item_udp *mask = item->mask;
2187         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2188         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2189                                       MLX5_FLOW_LAYER_OUTER_L3;
2190         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2191                                       MLX5_FLOW_LAYER_OUTER_L4;
2192         int ret;
2193
2194         if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2195                 return rte_flow_error_set(error, EINVAL,
2196                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2197                                           "protocol filtering not compatible"
2198                                           " with UDP layer");
2199         if (!(item_flags & l3m))
2200                 return rte_flow_error_set(error, EINVAL,
2201                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2202                                           "L3 is mandatory to filter on L4");
2203         if (item_flags & l4m)
2204                 return rte_flow_error_set(error, EINVAL,
2205                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2206                                           "multiple L4 layers not supported");
2207         if (!mask)
2208                 mask = &rte_flow_item_udp_mask;
2209         ret = mlx5_flow_item_acceptable
2210                 (item, (const uint8_t *)mask,
2211                  (const uint8_t *)&rte_flow_item_udp_mask,
2212                  sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2213                  error);
2214         if (ret < 0)
2215                 return ret;
2216         return 0;
2217 }
2218
2219 /**
2220  * Validate TCP item.
2221  *
2222  * @param[in] item
2223  *   Item specification.
2224  * @param[in] item_flags
2225  *   Bit-fields that holds the items detected until now.
2226  * @param[in] target_protocol
2227  *   The next protocol in the previous item.
2228  * @param[out] error
2229  *   Pointer to error structure.
2230  *
2231  * @return
2232  *   0 on success, a negative errno value otherwise and rte_errno is set.
2233  */
2234 int
2235 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2236                             uint64_t item_flags,
2237                             uint8_t target_protocol,
2238                             const struct rte_flow_item_tcp *flow_mask,
2239                             struct rte_flow_error *error)
2240 {
2241         const struct rte_flow_item_tcp *mask = item->mask;
2242         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2243         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2244                                       MLX5_FLOW_LAYER_OUTER_L3;
2245         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2246                                       MLX5_FLOW_LAYER_OUTER_L4;
2247         int ret;
2248
2249         MLX5_ASSERT(flow_mask);
2250         if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2251                 return rte_flow_error_set(error, EINVAL,
2252                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2253                                           "protocol filtering not compatible"
2254                                           " with TCP layer");
2255         if (!(item_flags & l3m))
2256                 return rte_flow_error_set(error, EINVAL,
2257                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2258                                           "L3 is mandatory to filter on L4");
2259         if (item_flags & l4m)
2260                 return rte_flow_error_set(error, EINVAL,
2261                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2262                                           "multiple L4 layers not supported");
2263         if (!mask)
2264                 mask = &rte_flow_item_tcp_mask;
2265         ret = mlx5_flow_item_acceptable
2266                 (item, (const uint8_t *)mask,
2267                  (const uint8_t *)flow_mask,
2268                  sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2269                  error);
2270         if (ret < 0)
2271                 return ret;
2272         return 0;
2273 }
2274
2275 /**
2276  * Validate VXLAN item.
2277  *
2278  * @param[in] item
2279  *   Item specification.
2280  * @param[in] item_flags
2281  *   Bit-fields that holds the items detected until now.
2282  * @param[in] target_protocol
2283  *   The next protocol in the previous item.
2284  * @param[out] error
2285  *   Pointer to error structure.
2286  *
2287  * @return
2288  *   0 on success, a negative errno value otherwise and rte_errno is set.
2289  */
2290 int
2291 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2292                               uint64_t item_flags,
2293                               struct rte_flow_error *error)
2294 {
2295         const struct rte_flow_item_vxlan *spec = item->spec;
2296         const struct rte_flow_item_vxlan *mask = item->mask;
2297         int ret;
2298         union vni {
2299                 uint32_t vlan_id;
2300                 uint8_t vni[4];
2301         } id = { .vlan_id = 0, };
2302
2303
2304         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2305                 return rte_flow_error_set(error, ENOTSUP,
2306                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2307                                           "multiple tunnel layers not"
2308                                           " supported");
2309         /*
2310          * Verify only UDPv4 is present as defined in
2311          * https://tools.ietf.org/html/rfc7348
2312          */
2313         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2314                 return rte_flow_error_set(error, EINVAL,
2315                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2316                                           "no outer UDP layer found");
2317         if (!mask)
2318                 mask = &rte_flow_item_vxlan_mask;
2319         ret = mlx5_flow_item_acceptable
2320                 (item, (const uint8_t *)mask,
2321                  (const uint8_t *)&rte_flow_item_vxlan_mask,
2322                  sizeof(struct rte_flow_item_vxlan),
2323                  MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2324         if (ret < 0)
2325                 return ret;
2326         if (spec) {
2327                 memcpy(&id.vni[1], spec->vni, 3);
2328                 memcpy(&id.vni[1], mask->vni, 3);
2329         }
2330         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2331                 return rte_flow_error_set(error, ENOTSUP,
2332                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2333                                           "VXLAN tunnel must be fully defined");
2334         return 0;
2335 }
2336
2337 /**
2338  * Validate VXLAN_GPE item.
2339  *
2340  * @param[in] item
2341  *   Item specification.
2342  * @param[in] item_flags
2343  *   Bit-fields that holds the items detected until now.
2344  * @param[in] priv
2345  *   Pointer to the private data structure.
2346  * @param[in] target_protocol
2347  *   The next protocol in the previous item.
2348  * @param[out] error
2349  *   Pointer to error structure.
2350  *
2351  * @return
2352  *   0 on success, a negative errno value otherwise and rte_errno is set.
2353  */
2354 int
2355 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2356                                   uint64_t item_flags,
2357                                   struct rte_eth_dev *dev,
2358                                   struct rte_flow_error *error)
2359 {
2360         struct mlx5_priv *priv = dev->data->dev_private;
2361         const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2362         const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2363         int ret;
2364         union vni {
2365                 uint32_t vlan_id;
2366                 uint8_t vni[4];
2367         } id = { .vlan_id = 0, };
2368
2369         if (!priv->config.l3_vxlan_en)
2370                 return rte_flow_error_set(error, ENOTSUP,
2371                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2372                                           "L3 VXLAN is not enabled by device"
2373                                           " parameter and/or not configured in"
2374                                           " firmware");
2375         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2376                 return rte_flow_error_set(error, ENOTSUP,
2377                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2378                                           "multiple tunnel layers not"
2379                                           " supported");
2380         /*
2381          * Verify only UDPv4 is present as defined in
2382          * https://tools.ietf.org/html/rfc7348
2383          */
2384         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2385                 return rte_flow_error_set(error, EINVAL,
2386                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2387                                           "no outer UDP layer found");
2388         if (!mask)
2389                 mask = &rte_flow_item_vxlan_gpe_mask;
2390         ret = mlx5_flow_item_acceptable
2391                 (item, (const uint8_t *)mask,
2392                  (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2393                  sizeof(struct rte_flow_item_vxlan_gpe),
2394                  MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2395         if (ret < 0)
2396                 return ret;
2397         if (spec) {
2398                 if (spec->protocol)
2399                         return rte_flow_error_set(error, ENOTSUP,
2400                                                   RTE_FLOW_ERROR_TYPE_ITEM,
2401                                                   item,
2402                                                   "VxLAN-GPE protocol"
2403                                                   " not supported");
2404                 memcpy(&id.vni[1], spec->vni, 3);
2405                 memcpy(&id.vni[1], mask->vni, 3);
2406         }
2407         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2408                 return rte_flow_error_set(error, ENOTSUP,
2409                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2410                                           "VXLAN-GPE tunnel must be fully"
2411                                           " defined");
2412         return 0;
2413 }
2414 /**
2415  * Validate GRE Key item.
2416  *
2417  * @param[in] item
2418  *   Item specification.
2419  * @param[in] item_flags
2420  *   Bit flags to mark detected items.
2421  * @param[in] gre_item
2422  *   Pointer to gre_item
2423  * @param[out] error
2424  *   Pointer to error structure.
2425  *
2426  * @return
2427  *   0 on success, a negative errno value otherwise and rte_errno is set.
2428  */
2429 int
2430 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2431                                 uint64_t item_flags,
2432                                 const struct rte_flow_item *gre_item,
2433                                 struct rte_flow_error *error)
2434 {
2435         const rte_be32_t *mask = item->mask;
2436         int ret = 0;
2437         rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2438         const struct rte_flow_item_gre *gre_spec;
2439         const struct rte_flow_item_gre *gre_mask;
2440
2441         if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2442                 return rte_flow_error_set(error, ENOTSUP,
2443                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2444                                           "Multiple GRE key not support");
2445         if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2446                 return rte_flow_error_set(error, ENOTSUP,
2447                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2448                                           "No preceding GRE header");
2449         if (item_flags & MLX5_FLOW_LAYER_INNER)
2450                 return rte_flow_error_set(error, ENOTSUP,
2451                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2452                                           "GRE key following a wrong item");
2453         gre_mask = gre_item->mask;
2454         if (!gre_mask)
2455                 gre_mask = &rte_flow_item_gre_mask;
2456         gre_spec = gre_item->spec;
2457         if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2458                          !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2459                 return rte_flow_error_set(error, EINVAL,
2460                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2461                                           "Key bit must be on");
2462
2463         if (!mask)
2464                 mask = &gre_key_default_mask;
2465         ret = mlx5_flow_item_acceptable
2466                 (item, (const uint8_t *)mask,
2467                  (const uint8_t *)&gre_key_default_mask,
2468                  sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2469         return ret;
2470 }
2471
2472 /**
2473  * Validate GRE item.
2474  *
2475  * @param[in] item
2476  *   Item specification.
2477  * @param[in] item_flags
2478  *   Bit flags to mark detected items.
2479  * @param[in] target_protocol
2480  *   The next protocol in the previous item.
2481  * @param[out] error
2482  *   Pointer to error structure.
2483  *
2484  * @return
2485  *   0 on success, a negative errno value otherwise and rte_errno is set.
2486  */
2487 int
2488 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2489                             uint64_t item_flags,
2490                             uint8_t target_protocol,
2491                             struct rte_flow_error *error)
2492 {
2493         const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2494         const struct rte_flow_item_gre *mask = item->mask;
2495         int ret;
2496         const struct rte_flow_item_gre nic_mask = {
2497                 .c_rsvd0_ver = RTE_BE16(0xB000),
2498                 .protocol = RTE_BE16(UINT16_MAX),
2499         };
2500
2501         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2502                 return rte_flow_error_set(error, EINVAL,
2503                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2504                                           "protocol filtering not compatible"
2505                                           " with this GRE layer");
2506         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2507                 return rte_flow_error_set(error, ENOTSUP,
2508                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2509                                           "multiple tunnel layers not"
2510                                           " supported");
2511         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2512                 return rte_flow_error_set(error, ENOTSUP,
2513                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2514                                           "L3 Layer is missing");
2515         if (!mask)
2516                 mask = &rte_flow_item_gre_mask;
2517         ret = mlx5_flow_item_acceptable
2518                 (item, (const uint8_t *)mask,
2519                  (const uint8_t *)&nic_mask,
2520                  sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2521                  error);
2522         if (ret < 0)
2523                 return ret;
2524 #ifndef HAVE_MLX5DV_DR
2525 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2526         if (spec && (spec->protocol & mask->protocol))
2527                 return rte_flow_error_set(error, ENOTSUP,
2528                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2529                                           "without MPLS support the"
2530                                           " specification cannot be used for"
2531                                           " filtering");
2532 #endif
2533 #endif
2534         return 0;
2535 }
2536
2537 /**
2538  * Validate Geneve item.
2539  *
2540  * @param[in] item
2541  *   Item specification.
2542  * @param[in] itemFlags
2543  *   Bit-fields that holds the items detected until now.
2544  * @param[in] enPriv
2545  *   Pointer to the private data structure.
2546  * @param[out] error
2547  *   Pointer to error structure.
2548  *
2549  * @return
2550  *   0 on success, a negative errno value otherwise and rte_errno is set.
2551  */
2552
2553 int
2554 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2555                                uint64_t item_flags,
2556                                struct rte_eth_dev *dev,
2557                                struct rte_flow_error *error)
2558 {
2559         struct mlx5_priv *priv = dev->data->dev_private;
2560         const struct rte_flow_item_geneve *spec = item->spec;
2561         const struct rte_flow_item_geneve *mask = item->mask;
2562         int ret;
2563         uint16_t gbhdr;
2564         uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2565                           MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2566         const struct rte_flow_item_geneve nic_mask = {
2567                 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2568                 .vni = "\xff\xff\xff",
2569                 .protocol = RTE_BE16(UINT16_MAX),
2570         };
2571
2572         if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2573                 return rte_flow_error_set(error, ENOTSUP,
2574                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2575                                           "L3 Geneve is not enabled by device"
2576                                           " parameter and/or not configured in"
2577                                           " firmware");
2578         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2579                 return rte_flow_error_set(error, ENOTSUP,
2580                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2581                                           "multiple tunnel layers not"
2582                                           " supported");
2583         /*
2584          * Verify only UDPv4 is present as defined in
2585          * https://tools.ietf.org/html/rfc7348
2586          */
2587         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2588                 return rte_flow_error_set(error, EINVAL,
2589                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2590                                           "no outer UDP layer found");
2591         if (!mask)
2592                 mask = &rte_flow_item_geneve_mask;
2593         ret = mlx5_flow_item_acceptable
2594                                   (item, (const uint8_t *)mask,
2595                                    (const uint8_t *)&nic_mask,
2596                                    sizeof(struct rte_flow_item_geneve),
2597                                    MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2598         if (ret)
2599                 return ret;
2600         if (spec) {
2601                 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2602                 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2603                      MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2604                      MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2605                         return rte_flow_error_set(error, ENOTSUP,
2606                                                   RTE_FLOW_ERROR_TYPE_ITEM,
2607                                                   item,
2608                                                   "Geneve protocol unsupported"
2609                                                   " fields are being used");
2610                 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2611                         return rte_flow_error_set
2612                                         (error, ENOTSUP,
2613                                          RTE_FLOW_ERROR_TYPE_ITEM,
2614                                          item,
2615                                          "Unsupported Geneve options length");
2616         }
2617         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2618                 return rte_flow_error_set
2619                                     (error, ENOTSUP,
2620                                      RTE_FLOW_ERROR_TYPE_ITEM, item,
2621                                      "Geneve tunnel must be fully defined");
2622         return 0;
2623 }
2624
2625 /**
2626  * Validate Geneve TLV option item.
2627  *
2628  * @param[in] item
2629  *   Item specification.
2630  * @param[in] last_item
2631  *   Previous validated item in the pattern items.
2632  * @param[in] geneve_item
2633  *   Previous GENEVE item specification.
2634  * @param[in] dev
2635  *   Pointer to the rte_eth_dev structure.
2636  * @param[out] error
2637  *   Pointer to error structure.
2638  *
2639  * @return
2640  *   0 on success, a negative errno value otherwise and rte_errno is set.
2641  */
2642 int
2643 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2644                                    uint64_t last_item,
2645                                    const struct rte_flow_item *geneve_item,
2646                                    struct rte_eth_dev *dev,
2647                                    struct rte_flow_error *error)
2648 {
2649         struct mlx5_priv *priv = dev->data->dev_private;
2650         struct mlx5_dev_ctx_shared *sh = priv->sh;
2651         struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2652         struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2653         uint8_t data_max_supported =
2654                         hca_attr->max_geneve_tlv_option_data_len * 4;
2655         struct mlx5_dev_config *config = &priv->config;
2656         const struct rte_flow_item_geneve *geneve_spec;
2657         const struct rte_flow_item_geneve *geneve_mask;
2658         const struct rte_flow_item_geneve_opt *spec = item->spec;
2659         const struct rte_flow_item_geneve_opt *mask = item->mask;
2660         unsigned int i;
2661         unsigned int data_len;
2662         uint8_t tlv_option_len;
2663         uint16_t optlen_m, optlen_v;
2664         const struct rte_flow_item_geneve_opt full_mask = {
2665                 .option_class = RTE_BE16(0xffff),
2666                 .option_type = 0xff,
2667                 .option_len = 0x1f,
2668         };
2669
2670         if (!mask)
2671                 mask = &rte_flow_item_geneve_opt_mask;
2672         if (!spec)
2673                 return rte_flow_error_set
2674                         (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2675                         "Geneve TLV opt class/type/length must be specified");
2676         if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2677                 return rte_flow_error_set
2678                         (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2679                         "Geneve TLV opt length exceeeds the limit (31)");
2680         /* Check if class type and length masks are full. */
2681         if (full_mask.option_class != mask->option_class ||
2682             full_mask.option_type != mask->option_type ||
2683             full_mask.option_len != (mask->option_len & full_mask.option_len))
2684                 return rte_flow_error_set
2685                         (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2686                         "Geneve TLV opt class/type/length masks must be full");
2687         /* Check if length is supported */
2688         if ((uint32_t)spec->option_len >
2689                         config->hca_attr.max_geneve_tlv_option_data_len)
2690                 return rte_flow_error_set
2691                         (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2692                         "Geneve TLV opt length not supported");
2693         if (config->hca_attr.max_geneve_tlv_options > 1)
2694                 DRV_LOG(DEBUG,
2695                         "max_geneve_tlv_options supports more than 1 option");
2696         /* Check GENEVE item preceding. */
2697         if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2698                 return rte_flow_error_set
2699                         (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2700                         "Geneve opt item must be preceded with Geneve item");
2701         geneve_spec = geneve_item->spec;
2702         geneve_mask = geneve_item->mask ? geneve_item->mask :
2703                                           &rte_flow_item_geneve_mask;
2704         /* Check if GENEVE TLV option size doesn't exceed option length */
2705         if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2706                             geneve_spec->ver_opt_len_o_c_rsvd0)) {
2707                 tlv_option_len = spec->option_len & mask->option_len;
2708                 optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2709                 optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2710                 optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2711                 optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2712                 if ((optlen_v & optlen_m) <= tlv_option_len)
2713                         return rte_flow_error_set
2714                                 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2715                                  "GENEVE TLV option length exceeds optlen");
2716         }
2717         /* Check if length is 0 or data is 0. */
2718         if (spec->data == NULL || spec->option_len == 0)
2719                 return rte_flow_error_set
2720                         (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2721                         "Geneve TLV opt with zero data/length not supported");
2722         /* Check not all data & mask are 0. */
2723         data_len = spec->option_len * 4;
2724         if (mask->data == NULL) {
2725                 for (i = 0; i < data_len; i++)
2726                         if (spec->data[i])
2727                                 break;
2728                 if (i == data_len)
2729                         return rte_flow_error_set(error, ENOTSUP,
2730                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
2731                                 "Can't match on Geneve option data 0");
2732         } else {
2733                 for (i = 0; i < data_len; i++)
2734                         if (spec->data[i] & mask->data[i])
2735                                 break;
2736                 if (i == data_len)
2737                         return rte_flow_error_set(error, ENOTSUP,
2738                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
2739                                 "Can't match on Geneve option data and mask 0");
2740                 /* Check data mask supported. */
2741                 for (i = data_max_supported; i < data_len ; i++)
2742                         if (mask->data[i])
2743                                 return rte_flow_error_set(error, ENOTSUP,
2744                                         RTE_FLOW_ERROR_TYPE_ITEM, item,
2745                                         "Data mask is of unsupported size");
2746         }
2747         /* Check GENEVE option is supported in NIC. */
2748         if (!config->hca_attr.geneve_tlv_opt)
2749                 return rte_flow_error_set
2750                         (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2751                         "Geneve TLV opt not supported");
2752         /* Check if we already have geneve option with different type/class. */
2753         rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2754         geneve_opt_resource = sh->geneve_tlv_option_resource;
2755         if (geneve_opt_resource != NULL)
2756                 if (geneve_opt_resource->option_class != spec->option_class ||
2757                     geneve_opt_resource->option_type != spec->option_type ||
2758                     geneve_opt_resource->length != spec->option_len) {
2759                         rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2760                         return rte_flow_error_set(error, ENOTSUP,
2761                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
2762                                 "Only one Geneve TLV option supported");
2763                 }
2764         rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2765         return 0;
2766 }
2767
2768 /**
2769  * Validate MPLS item.
2770  *
2771  * @param[in] dev
2772  *   Pointer to the rte_eth_dev structure.
2773  * @param[in] item
2774  *   Item specification.
2775  * @param[in] item_flags
2776  *   Bit-fields that holds the items detected until now.
2777  * @param[in] prev_layer
2778  *   The protocol layer indicated in previous item.
2779  * @param[out] error
2780  *   Pointer to error structure.
2781  *
2782  * @return
2783  *   0 on success, a negative errno value otherwise and rte_errno is set.
2784  */
2785 int
2786 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2787                              const struct rte_flow_item *item __rte_unused,
2788                              uint64_t item_flags __rte_unused,
2789                              uint64_t prev_layer __rte_unused,
2790                              struct rte_flow_error *error)
2791 {
2792 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2793         const struct rte_flow_item_mpls *mask = item->mask;
2794         struct mlx5_priv *priv = dev->data->dev_private;
2795         int ret;
2796
2797         if (!priv->config.mpls_en)
2798                 return rte_flow_error_set(error, ENOTSUP,
2799                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2800                                           "MPLS not supported or"
2801                                           " disabled in firmware"
2802                                           " configuration.");
2803         /* MPLS over IP, UDP, GRE is allowed */
2804         if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2805                             MLX5_FLOW_LAYER_OUTER_L4_UDP |
2806                             MLX5_FLOW_LAYER_GRE |
2807                             MLX5_FLOW_LAYER_GRE_KEY)))
2808                 return rte_flow_error_set(error, EINVAL,
2809                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2810                                           "protocol filtering not compatible"
2811                                           " with MPLS layer");
2812         /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2813         if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2814             !(item_flags & MLX5_FLOW_LAYER_GRE))
2815                 return rte_flow_error_set(error, ENOTSUP,
2816                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2817                                           "multiple tunnel layers not"
2818                                           " supported");
2819         if (!mask)
2820                 mask = &rte_flow_item_mpls_mask;
2821         ret = mlx5_flow_item_acceptable
2822                 (item, (const uint8_t *)mask,
2823                  (const uint8_t *)&rte_flow_item_mpls_mask,
2824                  sizeof(struct rte_flow_item_mpls),
2825                  MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2826         if (ret < 0)
2827                 return ret;
2828         return 0;
2829 #else
2830         return rte_flow_error_set(error, ENOTSUP,
2831                                   RTE_FLOW_ERROR_TYPE_ITEM, item,
2832                                   "MPLS is not supported by Verbs, please"
2833                                   " update.");
2834 #endif
2835 }
2836
2837 /**
2838  * Validate NVGRE item.
2839  *
2840  * @param[in] item
2841  *   Item specification.
2842  * @param[in] item_flags
2843  *   Bit flags to mark detected items.
2844  * @param[in] target_protocol
2845  *   The next protocol in the previous item.
2846  * @param[out] error
2847  *   Pointer to error structure.
2848  *
2849  * @return
2850  *   0 on success, a negative errno value otherwise and rte_errno is set.
2851  */
2852 int
2853 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2854                               uint64_t item_flags,
2855                               uint8_t target_protocol,
2856                               struct rte_flow_error *error)
2857 {
2858         const struct rte_flow_item_nvgre *mask = item->mask;
2859         int ret;
2860
2861         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2862                 return rte_flow_error_set(error, EINVAL,
2863                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2864                                           "protocol filtering not compatible"
2865                                           " with this GRE layer");
2866         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2867                 return rte_flow_error_set(error, ENOTSUP,
2868                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2869                                           "multiple tunnel layers not"
2870                                           " supported");
2871         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2872                 return rte_flow_error_set(error, ENOTSUP,
2873                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2874                                           "L3 Layer is missing");
2875         if (!mask)
2876                 mask = &rte_flow_item_nvgre_mask;
2877         ret = mlx5_flow_item_acceptable
2878                 (item, (const uint8_t *)mask,
2879                  (const uint8_t *)&rte_flow_item_nvgre_mask,
2880                  sizeof(struct rte_flow_item_nvgre),
2881                  MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2882         if (ret < 0)
2883                 return ret;
2884         return 0;
2885 }
2886
2887 /**
2888  * Validate eCPRI item.
2889  *
2890  * @param[in] item
2891  *   Item specification.
2892  * @param[in] item_flags
2893  *   Bit-fields that holds the items detected until now.
2894  * @param[in] last_item
2895  *   Previous validated item in the pattern items.
2896  * @param[in] ether_type
2897  *   Type in the ethernet layer header (including dot1q).
2898  * @param[in] acc_mask
2899  *   Acceptable mask, if NULL default internal default mask
2900  *   will be used to check whether item fields are supported.
2901  * @param[out] error
2902  *   Pointer to error structure.
2903  *
2904  * @return
2905  *   0 on success, a negative errno value otherwise and rte_errno is set.
2906  */
2907 int
2908 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2909                               uint64_t item_flags,
2910                               uint64_t last_item,
2911                               uint16_t ether_type,
2912                               const struct rte_flow_item_ecpri *acc_mask,
2913                               struct rte_flow_error *error)
2914 {
2915         const struct rte_flow_item_ecpri *mask = item->mask;
2916         const struct rte_flow_item_ecpri nic_mask = {
2917                 .hdr = {
2918                         .common = {
2919                                 .u32 =
2920                                 RTE_BE32(((const struct rte_ecpri_common_hdr) {
2921                                         .type = 0xFF,
2922                                         }).u32),
2923                         },
2924                         .dummy[0] = 0xFFFFFFFF,
2925                 },
2926         };
2927         const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2928                                         MLX5_FLOW_LAYER_OUTER_VLAN);
2929         struct rte_flow_item_ecpri mask_lo;
2930
2931         if (!(last_item & outer_l2_vlan) &&
2932             last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2933                 return rte_flow_error_set(error, EINVAL,
2934                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2935                                           "eCPRI can only follow L2/VLAN layer or UDP layer");
2936         if ((last_item & outer_l2_vlan) && ether_type &&
2937             ether_type != RTE_ETHER_TYPE_ECPRI)
2938                 return rte_flow_error_set(error, EINVAL,
2939                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2940                                           "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2941         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2942                 return rte_flow_error_set(error, EINVAL,
2943                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2944                                           "eCPRI with tunnel is not supported right now");
2945         if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2946                 return rte_flow_error_set(error, ENOTSUP,
2947                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2948                                           "multiple L3 layers not supported");
2949         else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2950                 return rte_flow_error_set(error, EINVAL,
2951                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2952                                           "eCPRI cannot coexist with a TCP layer");
2953         /* In specification, eCPRI could be over UDP layer. */
2954         else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2955                 return rte_flow_error_set(error, EINVAL,
2956                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
2957                                           "eCPRI over UDP layer is not yet supported right now");
2958         /* Mask for type field in common header could be zero. */
2959         if (!mask)
2960                 mask = &rte_flow_item_ecpri_mask;
2961         mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2962         /* Input mask is in big-endian format. */
2963         if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2964                 return rte_flow_error_set(error, EINVAL,
2965                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2966                                           "partial mask is not supported for protocol");
2967         else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2968                 return rte_flow_error_set(error, EINVAL,
2969                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2970                                           "message header mask must be after a type mask");
2971         return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2972                                          acc_mask ? (const uint8_t *)acc_mask
2973                                                   : (const uint8_t *)&nic_mask,
2974                                          sizeof(struct rte_flow_item_ecpri),
2975                                          MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2976 }
2977
2978 /**
2979  * Release resource related QUEUE/RSS action split.
2980  *
2981  * @param dev
2982  *   Pointer to Ethernet device.
2983  * @param flow
2984  *   Flow to release id's from.
2985  */
2986 static void
2987 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2988                              struct rte_flow *flow)
2989 {
2990         struct mlx5_priv *priv = dev->data->dev_private;
2991         uint32_t handle_idx;
2992         struct mlx5_flow_handle *dev_handle;
2993
2994         SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2995                        handle_idx, dev_handle, next)
2996                 if (dev_handle->split_flow_id)
2997                         mlx5_ipool_free(priv->sh->ipool
2998                                         [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
2999                                         dev_handle->split_flow_id);
3000 }
3001
3002 static int
3003 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3004                    const struct rte_flow_attr *attr __rte_unused,
3005                    const struct rte_flow_item items[] __rte_unused,
3006                    const struct rte_flow_action actions[] __rte_unused,
3007                    bool external __rte_unused,
3008                    int hairpin __rte_unused,
3009                    struct rte_flow_error *error)
3010 {
3011         return rte_flow_error_set(error, ENOTSUP,
3012                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3013 }
3014
3015 static struct mlx5_flow *
3016 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3017                   const struct rte_flow_attr *attr __rte_unused,
3018                   const struct rte_flow_item items[] __rte_unused,
3019                   const struct rte_flow_action actions[] __rte_unused,
3020                   struct rte_flow_error *error)
3021 {
3022         rte_flow_error_set(error, ENOTSUP,
3023                            RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3024         return NULL;
3025 }
3026
3027 static int
3028 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3029                     struct mlx5_flow *dev_flow __rte_unused,
3030                     const struct rte_flow_attr *attr __rte_unused,
3031                     const struct rte_flow_item items[] __rte_unused,
3032                     const struct rte_flow_action actions[] __rte_unused,
3033                     struct rte_flow_error *error)
3034 {
3035         return rte_flow_error_set(error, ENOTSUP,
3036                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3037 }
3038
3039 static int
3040 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3041                 struct rte_flow *flow __rte_unused,
3042                 struct rte_flow_error *error)
3043 {
3044         return rte_flow_error_set(error, ENOTSUP,
3045                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3046 }
3047
3048 static void
3049 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3050                  struct rte_flow *flow __rte_unused)
3051 {
3052 }
3053
3054 static void
3055 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3056                   struct rte_flow *flow __rte_unused)
3057 {
3058 }
3059
3060 static int
3061 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3062                 struct rte_flow *flow __rte_unused,
3063                 const struct rte_flow_action *actions __rte_unused,
3064                 void *data __rte_unused,
3065                 struct rte_flow_error *error)
3066 {
3067         return rte_flow_error_set(error, ENOTSUP,
3068                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3069 }
3070
3071 static int
3072 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3073                       uint32_t domains __rte_unused,
3074                       uint32_t flags __rte_unused)
3075 {
3076         return 0;
3077 }
3078
3079 /* Void driver to protect from null pointer reference. */
3080 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3081         .validate = flow_null_validate,
3082         .prepare = flow_null_prepare,
3083         .translate = flow_null_translate,
3084         .apply = flow_null_apply,
3085         .remove = flow_null_remove,
3086         .destroy = flow_null_destroy,
3087         .query = flow_null_query,
3088         .sync_domain = flow_null_sync_domain,
3089 };
3090
3091 /**
3092  * Select flow driver type according to flow attributes and device
3093  * configuration.
3094  *
3095  * @param[in] dev
3096  *   Pointer to the dev structure.
3097  * @param[in] attr
3098  *   Pointer to the flow attributes.
3099  *
3100  * @return
3101  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3102  */
3103 static enum mlx5_flow_drv_type
3104 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3105 {
3106         struct mlx5_priv *priv = dev->data->dev_private;
3107         /* The OS can determine first a specific flow type (DV, VERBS) */
3108         enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3109
3110         if (type != MLX5_FLOW_TYPE_MAX)
3111                 return type;
3112         /* If no OS specific type - continue with DV/VERBS selection */
3113         if (attr->transfer && priv->config.dv_esw_en)
3114                 type = MLX5_FLOW_TYPE_DV;
3115         if (!attr->transfer)
3116                 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3117                                                  MLX5_FLOW_TYPE_VERBS;
3118         return type;
3119 }
3120
3121 #define flow_get_drv_ops(type) flow_drv_ops[type]
3122
3123 /**
3124  * Flow driver validation API. This abstracts calling driver specific functions.
3125  * The type of flow driver is determined according to flow attributes.
3126  *
3127  * @param[in] dev
3128  *   Pointer to the dev structure.
3129  * @param[in] attr
3130  *   Pointer to the flow attributes.
3131  * @param[in] items
3132  *   Pointer to the list of items.
3133  * @param[in] actions
3134  *   Pointer to the list of actions.
3135  * @param[in] external
3136  *   This flow rule is created by request external to PMD.
3137  * @param[in] hairpin
3138  *   Number of hairpin TX actions, 0 means classic flow.
3139  * @param[out] error
3140  *   Pointer to the error structure.
3141  *
3142  * @return
3143  *   0 on success, a negative errno value otherwise and rte_errno is set.
3144  */
3145 static inline int
3146 flow_drv_validate(struct rte_eth_dev *dev,
3147                   const struct rte_flow_attr *attr,
3148                   const struct rte_flow_item items[],
3149                   const struct rte_flow_action actions[],
3150                   bool external, int hairpin, struct rte_flow_error *error)
3151 {
3152         const struct mlx5_flow_driver_ops *fops;
3153         enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3154
3155         fops = flow_get_drv_ops(type);
3156         return fops->validate(dev, attr, items, actions, external,
3157                               hairpin, error);
3158 }
3159
3160 /**
3161  * Flow driver preparation API. This abstracts calling driver specific
3162  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3163  * calculates the size of memory required for device flow, allocates the memory,
3164  * initializes the device flow and returns the pointer.
3165  *
3166  * @note
3167  *   This function initializes device flow structure such as dv or verbs in
3168  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3169  *   rest. For example, adding returning device flow to flow->dev_flow list and
3170  *   setting backward reference to the flow should be done out of this function.
3171  *   layers field is not filled either.
3172  *
3173  * @param[in] dev
3174  *   Pointer to the dev structure.
3175  * @param[in] attr
3176  *   Pointer to the flow attributes.
3177  * @param[in] items
3178  *   Pointer to the list of items.
3179  * @param[in] actions
3180  *   Pointer to the list of actions.
3181  * @param[in] flow_idx
3182  *   This memory pool index to the flow.
3183  * @param[out] error
3184  *   Pointer to the error structure.
3185  *
3186  * @return
3187  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3188  */
3189 static inline struct mlx5_flow *
3190 flow_drv_prepare(struct rte_eth_dev *dev,
3191                  const struct rte_flow *flow,
3192                  const struct rte_flow_attr *attr,
3193                  const struct rte_flow_item items[],
3194                  const struct rte_flow_action actions[],
3195                  uint32_t flow_idx,
3196                  struct rte_flow_error *error)
3197 {
3198         const struct mlx5_flow_driver_ops *fops;
3199         enum mlx5_flow_drv_type type = flow->drv_type;
3200         struct mlx5_flow *mlx5_flow = NULL;
3201
3202         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3203         fops = flow_get_drv_ops(type);
3204         mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3205         if (mlx5_flow)
3206                 mlx5_flow->flow_idx = flow_idx;
3207         return mlx5_flow;
3208 }
3209
3210 /**
3211  * Flow driver translation API. This abstracts calling driver specific
3212  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3213  * translates a generic flow into a driver flow. flow_drv_prepare() must
3214  * precede.
3215  *
3216  * @note
3217  *   dev_flow->layers could be filled as a result of parsing during translation
3218  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3219  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3220  *   flow->actions could be overwritten even though all the expanded dev_flows
3221  *   have the same actions.
3222  *
3223  * @param[in] dev
3224  *   Pointer to the rte dev structure.
3225  * @param[in, out] dev_flow
3226  *   Pointer to the mlx5 flow.
3227  * @param[in] attr
3228  *   Pointer to the flow attributes.
3229  * @param[in] items
3230  *   Pointer to the list of items.
3231  * @param[in] actions
3232  *   Pointer to the list of actions.
3233  * @param[out] error
3234  *   Pointer to the error structure.
3235  *
3236  * @return
3237  *   0 on success, a negative errno value otherwise and rte_errno is set.
3238  */
3239 static inline int
3240 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3241                    const struct rte_flow_attr *attr,
3242                    const struct rte_flow_item items[],
3243                    const struct rte_flow_action actions[],
3244                    struct rte_flow_error *error)
3245 {
3246         const struct mlx5_flow_driver_ops *fops;
3247         enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3248
3249         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3250         fops = flow_get_drv_ops(type);
3251         return fops->translate(dev, dev_flow, attr, items, actions, error);
3252 }
3253
3254 /**
3255  * Flow driver apply API. This abstracts calling driver specific functions.
3256  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3257  * translated driver flows on to device. flow_drv_translate() must precede.
3258  *
3259  * @param[in] dev
3260  *   Pointer to Ethernet device structure.
3261  * @param[in, out] flow
3262  *   Pointer to flow structure.
3263  * @param[out] error
3264  *   Pointer to error structure.
3265  *
3266  * @return
3267  *   0 on success, a negative errno value otherwise and rte_errno is set.
3268  */
3269 static inline int
3270 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3271                struct rte_flow_error *error)
3272 {
3273         const struct mlx5_flow_driver_ops *fops;
3274         enum mlx5_flow_drv_type type = flow->drv_type;
3275
3276         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3277         fops = flow_get_drv_ops(type);
3278         return fops->apply(dev, flow, error);
3279 }
3280
3281 /**
3282  * Flow driver destroy API. This abstracts calling driver specific functions.
3283  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3284  * on device and releases resources of the flow.
3285  *
3286  * @param[in] dev
3287  *   Pointer to Ethernet device.
3288  * @param[in, out] flow
3289  *   Pointer to flow structure.
3290  */
3291 static inline void
3292 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3293 {
3294         const struct mlx5_flow_driver_ops *fops;
3295         enum mlx5_flow_drv_type type = flow->drv_type;
3296
3297         flow_mreg_split_qrss_release(dev, flow);
3298         MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3299         fops = flow_get_drv_ops(type);
3300         fops->destroy(dev, flow);
3301 }
3302
3303 /**
3304  * Get RSS action from the action list.
3305  *
3306  * @param[in] actions
3307  *   Pointer to the list of actions.
3308  *
3309  * @return
3310  *   Pointer to the RSS action if exist, else return NULL.
3311  */
3312 static const struct rte_flow_action_rss*
3313 flow_get_rss_action(const struct rte_flow_action actions[])
3314 {
3315         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3316                 switch (actions->type) {
3317                 case RTE_FLOW_ACTION_TYPE_RSS:
3318                         return (const struct rte_flow_action_rss *)
3319                                actions->conf;
3320                 default:
3321                         break;
3322                 }
3323         }
3324         return NULL;
3325 }
3326
3327 /**
3328  * Get ASO age action by index.
3329  *
3330  * @param[in] dev
3331  *   Pointer to the Ethernet device structure.
3332  * @param[in] age_idx
3333  *   Index to the ASO age action.
3334  *
3335  * @return
3336  *   The specified ASO age action.
3337  */
3338 struct mlx5_aso_age_action*
3339 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3340 {
3341         uint16_t pool_idx = age_idx & UINT16_MAX;
3342         uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3343         struct mlx5_priv *priv = dev->data->dev_private;
3344         struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3345         struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3346
3347         return &pool->actions[offset - 1];
3348 }
3349
3350 /* maps shared action to translated non shared in some actions array */
3351 struct mlx5_translated_shared_action {
3352         struct rte_flow_shared_action *action; /**< Shared action */
3353         int index; /**< Index in related array of rte_flow_action */
3354 };
3355
3356 /**
3357  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3358  * non shared action if translation possible.
3359  * This functionality used to run same execution path for both shared & non
3360  * shared actions on flow create. All necessary preparations for shared
3361  * action handling should be preformed on *shared* actions list returned
3362  * from this call.
3363  *
3364  * @param[in] dev
3365  *   Pointer to Ethernet device.
3366  * @param[in] actions
3367  *   List of actions to translate.
3368  * @param[out] shared
3369  *   List to store translated shared actions.
3370  * @param[in, out] shared_n
3371  *   Size of *shared* array. On return should be updated with number of shared
3372  *   actions retrieved from the *actions* list.
3373  * @param[out] translated_actions
3374  *   List of actions where all shared actions were translated to non shared
3375  *   if possible. NULL if no translation took place.
3376  * @param[out] error
3377  *   Pointer to the error structure.
3378  *
3379  * @return
3380  *   0 on success, a negative errno value otherwise and rte_errno is set.
3381  */
3382 static int
3383 flow_shared_actions_translate(struct rte_eth_dev *dev,
3384                               const struct rte_flow_action actions[],
3385                               struct mlx5_translated_shared_action *shared,
3386                               int *shared_n,
3387                               struct rte_flow_action **translated_actions,
3388                               struct rte_flow_error *error)
3389 {
3390         struct mlx5_priv *priv = dev->data->dev_private;
3391         struct rte_flow_action *translated = NULL;
3392         size_t actions_size;
3393         int n;
3394         int copied_n = 0;
3395         struct mlx5_translated_shared_action *shared_end = NULL;
3396
3397         for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3398                 if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3399                         continue;
3400                 if (copied_n == *shared_n) {
3401                         return rte_flow_error_set
3402                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3403                                  NULL, "too many shared actions");
3404                 }
3405                 rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3406                            sizeof(actions[n].conf));
3407                 shared[copied_n].index = n;
3408                 copied_n++;
3409         }
3410         n++;
3411         *shared_n = copied_n;
3412         if (!copied_n)
3413                 return 0;
3414         actions_size = sizeof(struct rte_flow_action) * n;
3415         translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3416         if (!translated) {
3417                 rte_errno = ENOMEM;
3418                 return -ENOMEM;
3419         }
3420         memcpy(translated, actions, actions_size);
3421         for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3422                 struct mlx5_shared_action_rss *shared_rss;
3423                 uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3424                 uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3425                 uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET)
3426                                                                            - 1);
3427
3428                 switch (type) {
3429                 case MLX5_SHARED_ACTION_TYPE_RSS:
3430                         shared_rss = mlx5_ipool_get
3431                           (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3432                         translated[shared->index].type =
3433                                 RTE_FLOW_ACTION_TYPE_RSS;
3434                         translated[shared->index].conf =
3435                                 &shared_rss->origin;
3436                         break;
3437                 case MLX5_SHARED_ACTION_TYPE_AGE:
3438                         if (priv->sh->flow_hit_aso_en) {
3439                                 translated[shared->index].type =
3440                                         (enum rte_flow_action_type)
3441                                         MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3442                                 translated[shared->index].conf =
3443                                                          (void *)(uintptr_t)idx;
3444                                 break;
3445                         }
3446                         /* Fall-through */
3447                 default:
3448                         mlx5_free(translated);
3449                         return rte_flow_error_set
3450                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3451                                  NULL, "invalid shared action type");
3452                 }
3453         }
3454         *translated_actions = translated;
3455         return 0;
3456 }
3457
3458 /**
3459  * Get Shared RSS action from the action list.
3460  *
3461  * @param[in] dev
3462  *   Pointer to Ethernet device.
3463  * @param[in] shared
3464  *   Pointer to the list of actions.
3465  * @param[in] shared_n
3466  *   Actions list length.
3467  *
3468  * @return
3469  *   The MLX5 RSS action ID if exists, otherwise return 0.
3470  */
3471 static uint32_t
3472 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3473                            struct mlx5_translated_shared_action *shared,
3474                            int shared_n)
3475 {
3476         struct mlx5_translated_shared_action *shared_end;
3477         struct mlx5_priv *priv = dev->data->dev_private;
3478         struct mlx5_shared_action_rss *shared_rss;
3479
3480
3481         for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3482                 uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3483                 uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3484                 uint32_t idx = act_idx &
3485                                    ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
3486                 switch (type) {
3487                 case MLX5_SHARED_ACTION_TYPE_RSS:
3488                         shared_rss = mlx5_ipool_get
3489                                 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3490                                                                            idx);
3491                         __atomic_add_fetch(&shared_rss->refcnt, 1,
3492                                            __ATOMIC_RELAXED);
3493                         return idx;
3494                 default:
3495                         break;
3496                 }
3497         }
3498         return 0;
3499 }
3500
3501 static unsigned int
3502 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3503 {
3504         const struct rte_flow_item *item;
3505         unsigned int has_vlan = 0;
3506
3507         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3508                 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3509                         has_vlan = 1;
3510                         break;
3511                 }
3512         }
3513         if (has_vlan)
3514                 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3515                                        MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3516         return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3517                                MLX5_EXPANSION_ROOT_OUTER;
3518 }
3519
3520 /**
3521  *  Get layer flags from the prefix flow.
3522  *
3523  *  Some flows may be split to several subflows, the prefix subflow gets the
3524  *  match items and the suffix sub flow gets the actions.
3525  *  Some actions need the user defined match item flags to get the detail for
3526  *  the action.
3527  *  This function helps the suffix flow to get the item layer flags from prefix
3528  *  subflow.
3529  *
3530  * @param[in] dev_flow
3531  *   Pointer the created preifx subflow.
3532  *
3533  * @return
3534  *   The layers get from prefix subflow.
3535  */
3536 static inline uint64_t
3537 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3538 {
3539         uint64_t layers = 0;
3540
3541         /*
3542          * Layers bits could be localization, but usually the compiler will
3543          * help to do the optimization work for source code.
3544          * If no decap actions, use the layers directly.
3545          */
3546         if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3547                 return dev_flow->handle->layers;
3548         /* Convert L3 layers with decap action. */
3549         if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3550                 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3551         else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3552                 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3553         /* Convert L4 layers with decap action.  */
3554         if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3555                 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3556         else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3557                 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3558         return layers;
3559 }
3560
3561 /**
3562  * Get metadata split action information.
3563  *
3564  * @param[in] actions
3565  *   Pointer to the list of actions.
3566  * @param[out] qrss
3567  *   Pointer to the return pointer.
3568  * @param[out] qrss_type
3569  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3570  *   if no QUEUE/RSS is found.
3571  * @param[out] encap_idx
3572  *   Pointer to the index of the encap action if exists, otherwise the last
3573  *   action index.
3574  *
3575  * @return
3576  *   Total number of actions.
3577  */
3578 static int
3579 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3580                                        const struct rte_flow_action **qrss,
3581                                        int *encap_idx)
3582 {
3583         const struct rte_flow_action_raw_encap *raw_encap;
3584         int actions_n = 0;
3585         int raw_decap_idx = -1;
3586
3587         *encap_idx = -1;
3588         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3589                 switch (actions->type) {
3590                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3591                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3592                         *encap_idx = actions_n;
3593                         break;
3594                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3595                         raw_decap_idx = actions_n;
3596                         break;
3597                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3598                         raw_encap = actions->conf;
3599                         if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3600                                 *encap_idx = raw_decap_idx != -1 ?
3601                                                       raw_decap_idx : actions_n;
3602                         break;
3603                 case RTE_FLOW_ACTION_TYPE_QUEUE:
3604                 case RTE_FLOW_ACTION_TYPE_RSS:
3605                         *qrss = actions;
3606                         break;
3607                 default:
3608                         break;
3609                 }
3610                 actions_n++;
3611         }
3612         if (*encap_idx == -1)
3613                 *encap_idx = actions_n;
3614         /* Count RTE_FLOW_ACTION_TYPE_END. */
3615         return actions_n + 1;
3616 }
3617
3618 /**
3619  * Check meter action from the action list.
3620  *
3621  * @param[in] actions
3622  *   Pointer to the list of actions.
3623  * @param[out] mtr
3624  *   Pointer to the meter exist flag.
3625  *
3626  * @return
3627  *   Total number of actions.
3628  */
3629 static int
3630 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3631 {
3632         int actions_n = 0;
3633
3634         MLX5_ASSERT(mtr);
3635         *mtr = 0;
3636         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3637                 switch (actions->type) {
3638                 case RTE_FLOW_ACTION_TYPE_METER:
3639                         *mtr = 1;
3640                         break;
3641                 default:
3642                         break;
3643                 }
3644                 actions_n++;
3645         }
3646         /* Count RTE_FLOW_ACTION_TYPE_END. */
3647         return actions_n + 1;
3648 }
3649
3650 /**
3651  * Check if the flow should be split due to hairpin.
3652  * The reason for the split is that in current HW we can't
3653  * support encap and push-vlan on Rx, so if a flow contains
3654  * these actions we move it to Tx.
3655  *
3656  * @param dev
3657  *   Pointer to Ethernet device.
3658  * @param[in] attr
3659  *   Flow rule attributes.
3660  * @param[in] actions
3661  *   Associated actions (list terminated by the END action).
3662  *
3663  * @return
3664  *   > 0 the number of actions and the flow should be split,
3665  *   0 when no split required.
3666  */
3667 static int
3668 flow_check_hairpin_split(struct rte_eth_dev *dev,
3669                          const struct rte_flow_attr *attr,
3670                          const struct rte_flow_action actions[])
3671 {
3672         int queue_action = 0;
3673         int action_n = 0;
3674         int split = 0;
3675         const struct rte_flow_action_queue *queue;
3676         const struct rte_flow_action_rss *rss;
3677         const struct rte_flow_action_raw_encap *raw_encap;
3678         const struct rte_eth_hairpin_conf *conf;
3679
3680         if (!attr->ingress)
3681                 return 0;
3682         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3683                 switch (actions->type) {
3684                 case RTE_FLOW_ACTION_TYPE_QUEUE:
3685                         queue = actions->conf;
3686                         if (queue == NULL)
3687                                 return 0;
3688                         conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3689                         if (conf == NULL || conf->tx_explicit != 0)
3690                                 return 0;
3691                         queue_action = 1;
3692                         action_n++;
3693                         break;
3694                 case RTE_FLOW_ACTION_TYPE_RSS:
3695                         rss = actions->conf;
3696                         if (rss == NULL || rss->queue_num == 0)
3697                                 return 0;
3698                         conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3699                         if (conf == NULL || conf->tx_explicit != 0)
3700                                 return 0;
3701                         queue_action = 1;
3702                         action_n++;
3703                         break;
3704                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3705                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3706                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3707                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3708                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3709                         split++;
3710                         action_n++;
3711                         break;
3712                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3713                         raw_encap = actions->conf;
3714                         if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3715                                 split++;
3716                         action_n++;
3717                         break;
3718                 default:
3719                         action_n++;
3720                         break;
3721                 }
3722         }
3723         if (split && queue_action)
3724                 return action_n;
3725         return 0;
3726 }
3727
3728 /* Declare flow create/destroy prototype in advance. */
3729 static uint32_t
3730 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3731                  const struct rte_flow_attr *attr,
3732                  const struct rte_flow_item items[],
3733                  const struct rte_flow_action actions[],
3734                  bool external, struct rte_flow_error *error);
3735
3736 static void
3737 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3738                   uint32_t flow_idx);
3739
3740 int
3741 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3742                       struct mlx5_hlist_entry *entry,
3743                       uint64_t key, void *cb_ctx __rte_unused)
3744 {
3745         struct mlx5_flow_mreg_copy_resource *mcp_res =
3746                 container_of(entry, typeof(*mcp_res), hlist_ent);
3747
3748         return mcp_res->mark_id != key;
3749 }
3750
3751 struct mlx5_hlist_entry *
3752 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3753                        void *cb_ctx)
3754 {
3755         struct rte_eth_dev *dev = list->ctx;
3756         struct mlx5_priv *priv = dev->data->dev_private;
3757         struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3758         struct mlx5_flow_mreg_copy_resource *mcp_res;
3759         struct rte_flow_error *error = ctx->error;
3760         uint32_t idx = 0;
3761         int ret;
3762         uint32_t mark_id = key;
3763         struct rte_flow_attr attr = {
3764                 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3765                 .ingress = 1,
3766         };
3767         struct mlx5_rte_flow_item_tag tag_spec = {
3768                 .data = mark_id,
3769         };
3770         struct rte_flow_item items[] = {
3771                 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3772         };
3773         struct rte_flow_action_mark ftag = {
3774                 .id = mark_id,
3775         };
3776         struct mlx5_flow_action_copy_mreg cp_mreg = {
3777                 .dst = REG_B,
3778                 .src = REG_NON,
3779         };
3780         struct rte_flow_action_jump jump = {
3781                 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3782         };
3783         struct rte_flow_action actions[] = {
3784                 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3785         };
3786
3787         /* Fill the register fileds in the flow. */
3788         ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3789         if (ret < 0)
3790                 return NULL;
3791         tag_spec.id = ret;
3792         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3793         if (ret < 0)
3794                 return NULL;
3795         cp_mreg.src = ret;
3796         /* Provide the full width of FLAG specific value. */
3797         if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3798                 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3799         /* Build a new flow. */
3800         if (mark_id != MLX5_DEFAULT_COPY_ID) {
3801                 items[0] = (struct rte_flow_item){
3802                         .type = (enum rte_flow_item_type)
3803                                 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3804                         .spec = &tag_spec,
3805                 };
3806                 items[1] = (struct rte_flow_item){
3807                         .type = RTE_FLOW_ITEM_TYPE_END,
3808                 };
3809                 actions[0] = (struct rte_flow_action){
3810                         .type = (enum rte_flow_action_type)
3811                                 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3812                         .conf = &ftag,
3813                 };
3814                 actions[1] = (struct rte_flow_action){
3815                         .type = (enum rte_flow_action_type)
3816                                 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3817                         .conf = &cp_mreg,
3818                 };
3819                 actions[2] = (struct rte_flow_action){
3820                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
3821                         .conf = &jump,
3822                 };
3823                 actions[3] = (struct rte_flow_action){
3824                         .type = RTE_FLOW_ACTION_TYPE_END,
3825                 };
3826         } else {
3827                 /* Default rule, wildcard match. */
3828                 attr.priority = MLX5_FLOW_PRIO_RSVD;
3829                 items[0] = (struct rte_flow_item){
3830                         .type = RTE_FLOW_ITEM_TYPE_END,
3831                 };
3832                 actions[0] = (struct rte_flow_action){
3833                         .type = (enum rte_flow_action_type)
3834                                 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3835                         .conf = &cp_mreg,
3836                 };
3837                 actions[1] = (struct rte_flow_action){
3838                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
3839                         .conf = &jump,
3840                 };
3841                 actions[2] = (struct rte_flow_action){
3842                         .type = RTE_FLOW_ACTION_TYPE_END,
3843                 };
3844         }
3845         /* Build a new entry. */
3846         mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3847         if (!mcp_res) {
3848                 rte_errno = ENOMEM;
3849                 return NULL;
3850         }
3851         mcp_res->idx = idx;
3852         mcp_res->mark_id = mark_id;
3853         /*
3854          * The copy Flows are not included in any list. There
3855          * ones are referenced from other Flows and can not
3856          * be applied, removed, deleted in ardbitrary order
3857          * by list traversing.
3858          */
3859         mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3860                                          actions, false, error);
3861         if (!mcp_res->rix_flow) {
3862                 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3863                 return NULL;
3864         }
3865         return &mcp_res->hlist_ent;
3866 }
3867
3868 /**
3869  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3870  *
3871  * As mark_id is unique, if there's already a registered flow for the mark_id,
3872  * return by increasing the reference counter of the resource. Otherwise, create
3873  * the resource (mcp_res) and flow.
3874  *
3875  * Flow looks like,
3876  *   - If ingress port is ANY and reg_c[1] is mark_id,
3877  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3878  *
3879  * For default flow (zero mark_id), flow is like,
3880  *   - If ingress port is ANY,
3881  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3882  *
3883  * @param dev
3884  *   Pointer to Ethernet device.
3885  * @param mark_id
3886  *   ID of MARK action, zero means default flow for META.
3887  * @param[out] error
3888  *   Perform verbose error reporting if not NULL.
3889  *
3890  * @return
3891  *   Associated resource on success, NULL otherwise and rte_errno is set.
3892  */
3893 static struct mlx5_flow_mreg_copy_resource *
3894 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3895                           struct rte_flow_error *error)
3896 {
3897         struct mlx5_priv *priv = dev->data->dev_private;
3898         struct mlx5_hlist_entry *entry;
3899         struct mlx5_flow_cb_ctx ctx = {
3900                 .dev = dev,
3901                 .error = error,
3902         };
3903
3904         /* Check if already registered. */
3905         MLX5_ASSERT(priv->mreg_cp_tbl);
3906         entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3907         if (!entry)
3908                 return NULL;
3909         return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3910                             hlist_ent);
3911 }
3912
3913 void
3914 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3915 {
3916         struct mlx5_flow_mreg_copy_resource *mcp_res =
3917                 container_of(entry, typeof(*mcp_res), hlist_ent);
3918         struct rte_eth_dev *dev = list->ctx;
3919         struct mlx5_priv *priv = dev->data->dev_private;
3920
3921         MLX5_ASSERT(mcp_res->rix_flow);
3922         flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3923         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3924 }
3925
3926 /**
3927  * Release flow in RX_CP_TBL.
3928  *
3929  * @param dev
3930  *   Pointer to Ethernet device.
3931  * @flow
3932  *   Parent flow for wich copying is provided.
3933  */
3934 static void
3935 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3936                           struct rte_flow *flow)
3937 {
3938         struct mlx5_flow_mreg_copy_resource *mcp_res;
3939         struct mlx5_priv *priv = dev->data->dev_private;
3940
3941         if (!flow->rix_mreg_copy)
3942                 return;
3943         mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3944                                  flow->rix_mreg_copy);
3945         if (!mcp_res || !priv->mreg_cp_tbl)
3946                 return;
3947         MLX5_ASSERT(mcp_res->rix_flow);
3948         mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3949         flow->rix_mreg_copy = 0;
3950 }
3951
3952 /**
3953  * Remove the default copy action from RX_CP_TBL.
3954  *
3955  * This functions is called in the mlx5_dev_start(). No thread safe
3956  * is guaranteed.
3957  *
3958  * @param dev
3959  *   Pointer to Ethernet device.
3960  */
3961 static void
3962 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3963 {
3964         struct mlx5_hlist_entry *entry;
3965         struct mlx5_priv *priv = dev->data->dev_private;
3966
3967         /* Check if default flow is registered. */
3968         if (!priv->mreg_cp_tbl)
3969                 return;
3970         entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
3971                                   MLX5_DEFAULT_COPY_ID, NULL);
3972         if (!entry)
3973                 return;
3974         mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
3975 }
3976
3977 /**
3978  * Add the default copy action in in RX_CP_TBL.
3979  *
3980  * This functions is called in the mlx5_dev_start(). No thread safe
3981  * is guaranteed.
3982  *
3983  * @param dev
3984  *   Pointer to Ethernet device.
3985  * @param[out] error
3986  *   Perform verbose error reporting if not NULL.
3987  *
3988  * @return
3989  *   0 for success, negative value otherwise and rte_errno is set.
3990  */
3991 static int
3992 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3993                                   struct rte_flow_error *error)
3994 {
3995         struct mlx5_priv *priv = dev->data->dev_private;
3996         struct mlx5_flow_mreg_copy_resource *mcp_res;
3997
3998         /* Check whether extensive metadata feature is engaged. */
3999         if (!priv->config.dv_flow_en ||
4000             priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4001             !mlx5_flow_ext_mreg_supported(dev) ||
4002             !priv->sh->dv_regc0_mask)
4003                 return 0;
4004         /*
4005          * Add default mreg copy flow may be called multiple time, but
4006          * only be called once in stop. Avoid register it twice.
4007          */
4008         if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4009                 return 0;
4010         mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4011         if (!mcp_res)
4012                 return -rte_errno;
4013         return 0;
4014 }
4015
4016 /**
4017  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4018  *
4019  * All the flow having Q/RSS action should be split by
4020  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4021  * performs the following,
4022  *   - CQE->flow_tag := reg_c[1] (MARK)
4023  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4024  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4025  * but there should be a flow per each MARK ID set by MARK action.
4026  *
4027  * For the aforementioned reason, if there's a MARK action in flow's action
4028  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4029  * the MARK ID to CQE's flow_tag like,
4030  *   - If reg_c[1] is mark_id,
4031  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4032  *
4033  * For SET_META action which stores value in reg_c[0], as the destination is
4034  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4035  * MARK ID means the default flow. The default flow looks like,
4036  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4037  *
4038  * @param dev
4039  *   Pointer to Ethernet device.
4040  * @param flow
4041  *   Pointer to flow structure.
4042  * @param[in] actions
4043  *   Pointer to the list of actions.
4044  * @param[out] error
4045  *   Perform verbose error reporting if not NULL.
4046  *
4047  * @return
4048  *   0 on success, negative value otherwise and rte_errno is set.
4049  */
4050 static int
4051 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4052                             struct rte_flow *flow,
4053                             const struct rte_flow_action *actions,
4054                             struct rte_flow_error *error)
4055 {
4056         struct mlx5_priv *priv = dev->data->dev_private;
4057         struct mlx5_dev_config *config = &priv->config;
4058         struct mlx5_flow_mreg_copy_resource *mcp_res;
4059         const struct rte_flow_action_mark *mark;
4060
4061         /* Check whether extensive metadata feature is engaged. */
4062         if (!config->dv_flow_en ||
4063             config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4064             !mlx5_flow_ext_mreg_supported(dev) ||
4065             !priv->sh->dv_regc0_mask)
4066                 return 0;
4067         /* Find MARK action. */
4068         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4069                 switch (actions->type) {
4070                 case RTE_FLOW_ACTION_TYPE_FLAG:
4071                         mcp_res = flow_mreg_add_copy_action
4072                                 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4073                         if (!mcp_res)
4074                                 return -rte_errno;
4075                         flow->rix_mreg_copy = mcp_res->idx;
4076                         return 0;
4077                 case RTE_FLOW_ACTION_TYPE_MARK:
4078                         mark = (const struct rte_flow_action_mark *)
4079                                 actions->conf;
4080                         mcp_res =
4081                                 flow_mreg_add_copy_action(dev, mark->id, error);
4082                         if (!mcp_res)
4083                                 return -rte_errno;
4084                         flow->rix_mreg_copy = mcp_res->idx;
4085                         return 0;
4086                 default:
4087                         break;
4088                 }
4089         }
4090         return 0;
4091 }
4092
4093 #define MLX5_MAX_SPLIT_ACTIONS 24
4094 #define MLX5_MAX_SPLIT_ITEMS 24
4095
4096 /**
4097  * Split the hairpin flow.
4098  * Since HW can't support encap and push-vlan on Rx, we move these
4099  * actions to Tx.
4100  * If the count action is after the encap then we also
4101  * move the count action. in this case the count will also measure
4102  * the outer bytes.
4103  *
4104  * @param dev
4105  *   Pointer to Ethernet device.
4106  * @param[in] actions
4107  *   Associated actions (list terminated by the END action).
4108  * @param[out] actions_rx
4109  *   Rx flow actions.
4110  * @param[out] actions_tx
4111  *   Tx flow actions..
4112  * @param[out] pattern_tx
4113  *   The pattern items for the Tx flow.
4114  * @param[out] flow_id
4115  *   The flow ID connected to this flow.
4116  *
4117  * @return
4118  *   0 on success.
4119  */
4120 static int
4121 flow_hairpin_split(struct rte_eth_dev *dev,
4122                    const struct rte_flow_action actions[],
4123                    struct rte_flow_action actions_rx[],
4124                    struct rte_flow_action actions_tx[],
4125                    struct rte_flow_item pattern_tx[],
4126                    uint32_t flow_id)
4127 {
4128         const struct rte_flow_action_raw_encap *raw_encap;
4129         const struct rte_flow_action_raw_decap *raw_decap;
4130         struct mlx5_rte_flow_action_set_tag *set_tag;
4131         struct rte_flow_action *tag_action;
4132         struct mlx5_rte_flow_item_tag *tag_item;
4133         struct rte_flow_item *item;
4134         char *addr;
4135         int encap = 0;
4136
4137         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4138                 switch (actions->type) {
4139                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4140                 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4141                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4142                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4143                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4144                         rte_memcpy(actions_tx, actions,
4145                                sizeof(struct rte_flow_action));
4146                         actions_tx++;
4147                         break;
4148                 case RTE_FLOW_ACTION_TYPE_COUNT:
4149                         if (encap) {
4150                                 rte_memcpy(actions_tx, actions,
4151                                            sizeof(struct rte_flow_action));
4152                                 actions_tx++;
4153                         } else {
4154                                 rte_memcpy(actions_rx, actions,
4155                                            sizeof(struct rte_flow_action));
4156                                 actions_rx++;
4157                         }
4158                         break;
4159                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4160                         raw_encap = actions->conf;
4161                         if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4162                                 memcpy(actions_tx, actions,
4163                                        sizeof(struct rte_flow_action));
4164                                 actions_tx++;
4165                                 encap = 1;
4166                         } else {
4167                                 rte_memcpy(actions_rx, actions,
4168                                            sizeof(struct rte_flow_action));
4169                                 actions_rx++;
4170                         }
4171                         break;
4172                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4173                         raw_decap = actions->conf;
4174                         if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4175                                 memcpy(actions_tx, actions,
4176                                        sizeof(struct rte_flow_action));
4177                                 actions_tx++;
4178                         } else {
4179                                 rte_memcpy(actions_rx, actions,
4180                                            sizeof(struct rte_flow_action));
4181                                 actions_rx++;
4182                         }
4183                         break;
4184                 default:
4185                         rte_memcpy(actions_rx, actions,
4186                                    sizeof(struct rte_flow_action));
4187                         actions_rx++;
4188                         break;
4189                 }
4190         }
4191         /* Add set meta action and end action for the Rx flow. */
4192         tag_action = actions_rx;
4193         tag_action->type = (enum rte_flow_action_type)
4194                            MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4195         actions_rx++;
4196         rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4197         actions_rx++;
4198         set_tag = (void *)actions_rx;
4199         set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4200         MLX5_ASSERT(set_tag->id > REG_NON);
4201         set_tag->data = flow_id;
4202         tag_action->conf = set_tag;
4203         /* Create Tx item list. */
4204         rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4205         addr = (void *)&pattern_tx[2];
4206         item = pattern_tx;
4207         item->type = (enum rte_flow_item_type)
4208                      MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4209         tag_item = (void *)addr;
4210         tag_item->data = flow_id;
4211         tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4212         MLX5_ASSERT(set_tag->id > REG_NON);
4213         item->spec = tag_item;
4214         addr += sizeof(struct mlx5_rte_flow_item_tag);
4215         tag_item = (void *)addr;
4216         tag_item->data = UINT32_MAX;
4217         tag_item->id = UINT16_MAX;
4218         item->mask = tag_item;
4219         item->last = NULL;
4220         item++;
4221         item->type = RTE_FLOW_ITEM_TYPE_END;
4222         return 0;
4223 }
4224
4225 /**
4226  * The last stage of splitting chain, just creates the subflow
4227  * without any modification.
4228  *
4229  * @param[in] dev
4230  *   Pointer to Ethernet device.
4231  * @param[in] flow
4232  *   Parent flow structure pointer.
4233  * @param[in, out] sub_flow
4234  *   Pointer to return the created subflow, may be NULL.
4235  * @param[in] attr
4236  *   Flow rule attributes.
4237  * @param[in] items
4238  *   Pattern specification (list terminated by the END pattern item).
4239  * @param[in] actions
4240  *   Associated actions (list terminated by the END action).
4241  * @param[in] flow_split_info
4242  *   Pointer to flow split info structure.
4243  * @param[out] error
4244  *   Perform verbose error reporting if not NULL.
4245  * @return
4246  *   0 on success, negative value otherwise
4247  */
4248 static int
4249 flow_create_split_inner(struct rte_eth_dev *dev,
4250                         struct rte_flow *flow,
4251                         struct mlx5_flow **sub_flow,
4252                         const struct rte_flow_attr *attr,
4253                         const struct rte_flow_item items[],
4254                         const struct rte_flow_action actions[],
4255                         struct mlx5_flow_split_info *flow_split_info,
4256                         struct rte_flow_error *error)
4257 {
4258         struct mlx5_flow *dev_flow;
4259
4260         dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4261                                     flow_split_info->flow_idx, error);
4262         if (!dev_flow)
4263                 return -rte_errno;
4264         dev_flow->flow = flow;
4265         dev_flow->external = flow_split_info->external;
4266         dev_flow->skip_scale = flow_split_info->skip_scale;
4267         /* Subflow object was created, we must include one in the list. */
4268         SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4269                       dev_flow->handle, next);
4270         /*
4271          * If dev_flow is as one of the suffix flow, some actions in suffix
4272          * flow may need some user defined item layer flags, and pass the
4273          * Metadate rxq mark flag to suffix flow as well.
4274          */
4275         if (flow_split_info->prefix_layers)
4276                 dev_flow->handle->layers = flow_split_info->prefix_layers;
4277         if (flow_split_info->prefix_mark)
4278                 dev_flow->handle->mark = 1;
4279         if (sub_flow)
4280                 *sub_flow = dev_flow;
4281         return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4282 }
4283
4284 /**
4285  * Split the meter flow.
4286  *
4287  * As meter flow will split to three sub flow, other than meter
4288  * action, the other actions make sense to only meter accepts
4289  * the packet. If it need to be dropped, no other additional
4290  * actions should be take.
4291  *
4292  * One kind of special action which decapsulates the L3 tunnel
4293  * header will be in the prefix sub flow, as not to take the
4294  * L3 tunnel header into account.
4295  *
4296  * @param dev
4297  *   Pointer to Ethernet device.
4298  * @param[in] items
4299  *   Pattern specification (list terminated by the END pattern item).
4300  * @param[out] sfx_items
4301  *   Suffix flow match items (list terminated by the END pattern item).
4302  * @param[in] actions
4303  *   Associated actions (list terminated by the END action).
4304  * @param[out] actions_sfx
4305  *   Suffix flow actions.
4306  * @param[out] actions_pre
4307  *   Prefix flow actions.
4308  * @param[out] pattern_sfx
4309  *   The pattern items for the suffix flow.
4310  * @param[out] tag_sfx
4311  *   Pointer to suffix flow tag.
4312  *
4313  * @return
4314  *   0 on success.
4315  */
4316 static int
4317 flow_meter_split_prep(struct rte_eth_dev *dev,
4318                  const struct rte_flow_item items[],
4319                  struct rte_flow_item sfx_items[],
4320                  const struct rte_flow_action actions[],
4321                  struct rte_flow_action actions_sfx[],
4322                  struct rte_flow_action actions_pre[])
4323 {
4324         struct mlx5_priv *priv = dev->data->dev_private;
4325         struct rte_flow_action *tag_action = NULL;
4326         struct rte_flow_item *tag_item;
4327         struct mlx5_rte_flow_action_set_tag *set_tag;
4328         struct rte_flow_error error;
4329         const struct rte_flow_action_raw_encap *raw_encap;
4330         const struct rte_flow_action_raw_decap *raw_decap;
4331         struct mlx5_rte_flow_item_tag *tag_spec;
4332         struct mlx5_rte_flow_item_tag *tag_mask;
4333         uint32_t tag_id = 0;
4334         bool copy_vlan = false;
4335
4336         /* Prepare the actions for prefix and suffix flow. */
4337         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4338                 struct rte_flow_action **action_cur = NULL;
4339
4340                 switch (actions->type) {
4341                 case RTE_FLOW_ACTION_TYPE_METER:
4342                         /* Add the extra tag action first. */
4343                         tag_action = actions_pre;
4344                         tag_action->type = (enum rte_flow_action_type)
4345                                            MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4346                         actions_pre++;
4347                         action_cur = &actions_pre;
4348                         break;
4349                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4350                 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4351                         action_cur = &actions_pre;
4352                         break;
4353                 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4354                         raw_encap = actions->conf;
4355                         if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4356                                 action_cur = &actions_pre;
4357                         break;
4358                 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4359                         raw_decap = actions->conf;
4360                         if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4361                                 action_cur = &actions_pre;
4362                         break;
4363                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4364                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4365                         copy_vlan = true;
4366                         break;
4367                 default:
4368                         break;
4369                 }
4370                 if (!action_cur)
4371                         action_cur = &actions_sfx;
4372                 memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4373                 (*action_cur)++;
4374         }
4375         /* Add end action to the actions. */
4376         actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4377         actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4378         actions_pre++;
4379         /* Set the tag. */
4380         set_tag = (void *)actions_pre;
4381         set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4382         mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4383                           &tag_id);
4384         if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4385                 DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4386                         dev->data->port_id);
4387                 mlx5_ipool_free(priv->sh->ipool
4388                                 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4389                 return 0;
4390         } else if (!tag_id) {
4391                 return 0;
4392         }
4393         set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4394         assert(tag_action);
4395         tag_action->conf = set_tag;
4396         /* Prepare the suffix subflow items. */
4397         tag_item = sfx_items++;
4398         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4399                 int item_type = items->type;
4400
4401                 switch (item_type) {
4402                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4403                         memcpy(sfx_items, items, sizeof(*sfx_items));
4404                         sfx_items++;
4405                         break;
4406                 case RTE_FLOW_ITEM_TYPE_VLAN:
4407                         if (copy_vlan) {
4408                                 memcpy(sfx_items, items, sizeof(*sfx_items));
4409                                 /*
4410                                  * Convert to internal match item, it is used
4411                                  * for vlan push and set vid.
4412                                  */
4413                                 sfx_items->type = (enum rte_flow_item_type)
4414                                                   MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4415                                 sfx_items++;
4416                         }
4417                         break;
4418                 default:
4419                         break;
4420                 }
4421         }
4422         sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4423         sfx_items++;
4424         tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4425         tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4426         tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4427         tag_mask = tag_spec + 1;
4428         tag_mask->data = 0xffffff00;
4429         tag_item->type = (enum rte_flow_item_type)
4430                          MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4431         tag_item->spec = tag_spec;
4432         tag_item->last = NULL;
4433         tag_item->mask = tag_mask;
4434         return tag_id;
4435 }
4436
4437 /**
4438  * Split action list having QUEUE/RSS for metadata register copy.
4439  *
4440  * Once Q/RSS action is detected in user's action list, the flow action
4441  * should be split in order to copy metadata registers, which will happen in
4442  * RX_CP_TBL like,
4443  *   - CQE->flow_tag := reg_c[1] (MARK)
4444  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4445  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4446  * This is because the last action of each flow must be a terminal action
4447  * (QUEUE, RSS or DROP).
4448  *
4449  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4450  * stored and kept in the mlx5_flow structure per each sub_flow.
4451  *
4452  * The Q/RSS action is replaced with,
4453  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4454  * And the following JUMP action is added at the end,
4455  *   - JUMP, to RX_CP_TBL.
4456  *
4457  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4458  * flow_create_split_metadata() routine. The flow will look like,
4459  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4460  *
4461  * @param dev
4462  *   Pointer to Ethernet device.
4463  * @param[out] split_actions
4464  *   Pointer to store split actions to jump to CP_TBL.
4465  * @param[in] actions
4466  *   Pointer to the list of original flow actions.
4467  * @param[in] qrss
4468  *   Pointer to the Q/RSS action.
4469  * @param[in] actions_n
4470  *   Number of original actions.
4471  * @param[out] error
4472  *   Perform verbose error reporting if not NULL.
4473  *
4474  * @return
4475  *   non-zero unique flow_id on success, otherwise 0 and
4476  *   error/rte_error are set.
4477  */
4478 static uint32_t
4479 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4480                           struct rte_flow_action *split_actions,
4481                           const struct rte_flow_action *actions,
4482                           const struct rte_flow_action *qrss,
4483                           int actions_n, struct rte_flow_error *error)
4484 {
4485         struct mlx5_priv *priv = dev->data->dev_private;
4486         struct mlx5_rte_flow_action_set_tag *set_tag;
4487         struct rte_flow_action_jump *jump;
4488         const int qrss_idx = qrss - actions;
4489         uint32_t flow_id = 0;
4490         int ret = 0;
4491
4492         /*
4493          * Given actions will be split
4494          * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4495          * - Add jump to mreg CP_TBL.
4496          * As a result, there will be one more action.
4497          */
4498         ++actions_n;
4499         memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4500         set_tag = (void *)(split_actions + actions_n);
4501         /*
4502          * If tag action is not set to void(it means we are not the meter
4503          * suffix flow), add the tag action. Since meter suffix flow already
4504          * has the tag added.
4505          */
4506         if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4507                 /*
4508                  * Allocate the new subflow ID. This one is unique within
4509                  * device and not shared with representors. Otherwise,
4510                  * we would have to resolve multi-thread access synch
4511                  * issue. Each flow on the shared device is appended
4512                  * with source vport identifier, so the resulting
4513                  * flows will be unique in the shared (by master and
4514                  * representors) domain even if they have coinciding
4515                  * IDs.
4516                  */
4517                 mlx5_ipool_malloc(priv->sh->ipool
4518                                   [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4519                 if (!flow_id)
4520                         return rte_flow_error_set(error, ENOMEM,
4521                                                   RTE_FLOW_ERROR_TYPE_ACTION,
4522                                                   NULL, "can't allocate id "
4523                                                   "for split Q/RSS subflow");
4524                 /* Internal SET_TAG action to set flow ID. */
4525                 *set_tag = (struct mlx5_rte_flow_action_set_tag){
4526                         .data = flow_id,
4527                 };
4528                 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4529                 if (ret < 0)
4530                         return ret;
4531                 set_tag->id = ret;
4532                 /* Construct new actions array. */
4533                 /* Replace QUEUE/RSS action. */
4534                 split_actions[qrss_idx] = (struct rte_flow_action){
4535                         .type = (enum rte_flow_action_type)
4536                                 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4537                         .conf = set_tag,
4538                 };
4539         }
4540         /* JUMP action to jump to mreg copy table (CP_TBL). */
4541         jump = (void *)(set_tag + 1);
4542         *jump = (struct rte_flow_action_jump){
4543                 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4544         };
4545         split_actions[actions_n - 2] = (struct rte_flow_action){
4546                 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4547                 .conf = jump,
4548         };
4549         split_actions[actions_n - 1] = (struct rte_flow_action){
4550                 .type = RTE_FLOW_ACTION_TYPE_END,
4551         };
4552         return flow_id;
4553 }
4554
4555 /**
4556  * Extend the given action list for Tx metadata copy.
4557  *
4558  * Copy the given action list to the ext_actions and add flow metadata register
4559  * copy action in order to copy reg_a set by WQE to reg_c[0].
4560  *
4561  * @param[out] ext_actions
4562  *   Pointer to the extended action list.
4563  * @param[in] actions
4564  *   Pointer to the list of actions.
4565  * @param[in] actions_n
4566  *   Number of actions in the list.
4567  * @param[out] error
4568  *   Perform verbose error reporting if not NULL.
4569  * @param[in] encap_idx
4570  *   The encap action inndex.
4571  *
4572  * @return
4573  *   0 on success, negative value otherwise
4574  */
4575 static int
4576 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4577                        struct rte_flow_action *ext_actions,
4578                        const struct rte_flow_action *actions,
4579                        int actions_n, struct rte_flow_error *error,
4580                        int encap_idx)
4581 {
4582         struct mlx5_flow_action_copy_mreg *cp_mreg =
4583                 (struct mlx5_flow_action_copy_mreg *)
4584                         (ext_actions + actions_n + 1);
4585         int ret;
4586
4587         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4588         if (ret < 0)
4589                 return ret;
4590         cp_mreg->dst = ret;
4591         ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4592         if (ret < 0)
4593                 return ret;
4594         cp_mreg->src = ret;
4595         if (encap_idx != 0)
4596                 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4597         if (encap_idx == actions_n - 1) {
4598                 ext_actions[actions_n - 1] = (struct rte_flow_action){
4599                         .type = (enum rte_flow_action_type)
4600                                 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4601                         .conf = cp_mreg,
4602                 };
4603                 ext_actions[actions_n] = (struct rte_flow_action){
4604                         .type = RTE_FLOW_ACTION_TYPE_END,
4605                 };
4606         } else {
4607                 ext_actions[encap_idx] = (struct rte_flow_action){
4608                         .type = (enum rte_flow_action_type)
4609                                 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4610                         .conf = cp_mreg,
4611                 };
4612                 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4613                                 sizeof(*ext_actions) * (actions_n - encap_idx));
4614         }
4615         return 0;
4616 }
4617
4618 /**
4619  * Check the match action from the action list.
4620  *
4621  * @param[in] actions
4622  *   Pointer to the list of actions.
4623  * @param[in] attr
4624  *   Flow rule attributes.
4625  * @param[in] action
4626  *   The action to be check if exist.
4627  * @param[out] match_action_pos
4628  *   Pointer to the position of the matched action if exists, otherwise is -1.
4629  * @param[out] qrss_action_pos
4630  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4631  *
4632  * @return
4633  *   > 0 the total number of actions.
4634  *   0 if not found match action in action list.
4635  */
4636 static int
4637 flow_check_match_action(const struct rte_flow_action actions[],
4638                         const struct rte_flow_attr *attr,
4639                         enum rte_flow_action_type action,
4640                         int *match_action_pos, int *qrss_action_pos)
4641 {
4642         const struct rte_flow_action_sample *sample;
4643         int actions_n = 0;
4644         int jump_flag = 0;
4645         uint32_t ratio = 0;
4646         int sub_type = 0;
4647         int flag = 0;
4648
4649         *match_action_pos = -1;
4650         *qrss_action_pos = -1;
4651         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4652                 if (actions->type == action) {
4653                         flag = 1;
4654                         *match_action_pos = actions_n;
4655                 }
4656                 if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4657                     actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4658                         *qrss_action_pos = actions_n;
4659                 if (actions->type == RTE_FLOW_ACTION_TYPE_JUMP)
4660                         jump_flag = 1;
4661                 if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4662                         sample = actions->conf;
4663                         ratio = sample->ratio;
4664                         sub_type = ((const struct rte_flow_action *)
4665                                         (sample->actions))->type;
4666                 }
4667                 actions_n++;
4668         }
4669         if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4670                 if (ratio == 1) {
4671                         /* JUMP Action not support for Mirroring;
4672                          * Mirroring support multi-destination;
4673                          */
4674                         if (!jump_flag && sub_type != RTE_FLOW_ACTION_TYPE_END)
4675                                 flag = 0;
4676                 }
4677         }
4678         /* Count RTE_FLOW_ACTION_TYPE_END. */
4679         return flag ? actions_n + 1 : 0;
4680 }
4681
4682 #define SAMPLE_SUFFIX_ITEM 2
4683
4684 /**
4685  * Split the sample flow.
4686  *
4687  * As sample flow will split to two sub flow, sample flow with
4688  * sample action, the other actions will move to new suffix flow.
4689  *
4690  * Also add unique tag id with tag action in the sample flow,
4691  * the same tag id will be as match in the suffix flow.
4692  *
4693  * @param dev
4694  *   Pointer to Ethernet device.
4695  * @param[in] fdb_tx
4696  *   FDB egress flow flag.
4697  * @param[out] sfx_items
4698  *   Suffix flow match items (list terminated by the END pattern item).
4699  * @param[in] actions
4700  *   Associated actions (list terminated by the END action).
4701  * @param[out] actions_sfx
4702  *   Suffix flow actions.
4703  * @param[out] actions_pre
4704  *   Prefix flow actions.
4705  * @param[in] actions_n
4706  *  The total number of actions.
4707  * @param[in] sample_action_pos
4708  *   The sample action position.
4709  * @param[in] qrss_action_pos
4710  *   The Queue/RSS action position.
4711  * @param[out] error
4712  *   Perform verbose error reporting if not NULL.
4713  *
4714  * @return
4715  *   0 on success, or unique flow_id, a negative errno value
4716  *   otherwise and rte_errno is set.
4717  */
4718 static int
4719 flow_sample_split_prep(struct rte_eth_dev *dev,
4720                        uint32_t fdb_tx,
4721                        struct rte_flow_item sfx_items[],
4722                        const struct rte_flow_action actions[],
4723                        struct rte_flow_action actions_sfx[],
4724                        struct rte_flow_action actions_pre[],
4725                        int actions_n,
4726                        int sample_action_pos,
4727                        int qrss_action_pos,
4728                        struct rte_flow_error *error)
4729 {
4730         struct mlx5_priv *priv = dev->data->dev_private;
4731         struct mlx5_rte_flow_action_set_tag *set_tag;
4732         struct mlx5_rte_flow_item_tag *tag_spec;
4733         struct mlx5_rte_flow_item_tag *tag_mask;
4734         uint32_t tag_id = 0;
4735         int index;
4736         int ret;
4737
4738         if (sample_action_pos < 0)
4739                 return rte_flow_error_set(error, EINVAL,
4740                                           RTE_FLOW_ERROR_TYPE_ACTION,
4741                                           NULL, "invalid position of sample "
4742                                           "action in list");
4743         if (!fdb_tx) {
4744                 /* Prepare the prefix tag action. */
4745                 set_tag = (void *)(actions_pre + actions_n + 1);
4746                 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4747                 if (ret < 0)
4748                         return ret;
4749                 set_tag->id = ret;
4750                 mlx5_ipool_malloc(priv->sh->ipool
4751                                   [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4752                 set_tag->data = tag_id;
4753                 /* Prepare the suffix subflow items. */
4754                 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4755                 tag_spec->data = tag_id;
4756                 tag_spec->id = set_tag->id;
4757                 tag_mask = tag_spec + 1;
4758                 tag_mask->data = UINT32_MAX;
4759                 sfx_items[0] = (struct rte_flow_item){
4760                         .type = (enum rte_flow_item_type)
4761                                 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4762                         .spec = tag_spec,
4763                         .last = NULL,
4764                         .mask = tag_mask,
4765                 };
4766                 sfx_items[1] = (struct rte_flow_item){
4767                         .type = (enum rte_flow_item_type)
4768                                 RTE_FLOW_ITEM_TYPE_END,
4769                 };
4770         }
4771         /* Prepare the actions for prefix and suffix flow. */
4772         if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4773                 index = qrss_action_pos;
4774                 /* Put the preceding the Queue/RSS action into prefix flow. */
4775                 if (index != 0)
4776                         memcpy(actions_pre, actions,
4777                                sizeof(struct rte_flow_action) * index);
4778                 /* Put others preceding the sample action into prefix flow. */
4779                 if (sample_action_pos > index + 1)
4780                         memcpy(actions_pre + index, actions + index + 1,
4781                                sizeof(struct rte_flow_action) *
4782                                (sample_action_pos - index - 1));
4783                 index = sample_action_pos - 1;
4784                 /* Put Queue/RSS action into Suffix flow. */
4785                 memcpy(actions_sfx, actions + qrss_action_pos,
4786                        sizeof(struct rte_flow_action));
4787                 actions_sfx++;
4788         } else {
4789                 index = sample_action_pos;
4790                 if (index != 0)
4791                         memcpy(actions_pre, actions,
4792                                sizeof(struct rte_flow_action) * index);
4793         }
4794         /* Add the extra tag action for NIC-RX and E-Switch ingress. */
4795         if (!fdb_tx) {
4796                 actions_pre[index++] =
4797                         (struct rte_flow_action){
4798                         .type = (enum rte_flow_action_type)
4799                                 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4800                         .conf = set_tag,
4801                 };
4802         }
4803         memcpy(actions_pre + index, actions + sample_action_pos,
4804                sizeof(struct rte_flow_action));
4805         index += 1;
4806         actions_pre[index] = (struct rte_flow_action){
4807                 .type = (enum rte_flow_action_type)
4808                         RTE_FLOW_ACTION_TYPE_END,
4809         };
4810         /* Put the actions after sample into Suffix flow. */
4811         memcpy(actions_sfx, actions + sample_action_pos + 1,
4812                sizeof(struct rte_flow_action) *
4813                (actions_n - sample_action_pos - 1));
4814         return tag_id;
4815 }
4816
4817 /**
4818  * The splitting for metadata feature.
4819  *
4820  * - Q/RSS action on NIC Rx should be split in order to pass by
4821  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4822  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4823  *
4824  * - All the actions on NIC Tx should have a mreg copy action to
4825  *   copy reg_a from WQE to reg_c[0].
4826  *
4827  * @param dev
4828  *   Pointer to Ethernet device.
4829  * @param[in] flow
4830  *   Parent flow structure pointer.
4831  * @param[in] attr
4832  *   Flow rule attributes.
4833  * @param[in] items
4834  *   Pattern specification (list terminated by the END pattern item).
4835  * @param[in] actions
4836  *   Associated actions (list terminated by the END action).
4837  * @param[in] flow_split_info
4838  *   Pointer to flow split info structure.
4839  * @param[out] error
4840  *   Perform verbose error reporting if not NULL.
4841  * @return
4842  *   0 on success, negative value otherwise
4843  */
4844 static int
4845 flow_create_split_metadata(struct rte_eth_dev *dev,
4846                            struct rte_flow *flow,
4847                            const struct rte_flow_attr *attr,
4848                            const struct rte_flow_item items[],
4849                            const struct rte_flow_action actions[],
4850                            struct mlx5_flow_split_info *flow_split_info,
4851                            struct rte_flow_error *error)
4852 {
4853         struct mlx5_priv *priv = dev->data->dev_private;
4854         struct mlx5_dev_config *config = &priv->config;
4855         const struct rte_flow_action *qrss = NULL;
4856         struct rte_flow_action *ext_actions = NULL;
4857         struct mlx5_flow *dev_flow = NULL;
4858         uint32_t qrss_id = 0;
4859         int mtr_sfx = 0;
4860         size_t act_size;
4861         int actions_n;
4862         int encap_idx;
4863         int ret;
4864
4865         /* Check whether extensive metadata feature is engaged. */
4866         if (!config->dv_flow_en ||
4867             config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4868             !mlx5_flow_ext_mreg_supported(dev))
4869                 return flow_create_split_inner(dev, flow, NULL, attr, items,
4870                                                actions, flow_split_info, error);
4871         actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4872                                                            &encap_idx);
4873         if (qrss) {
4874                 /* Exclude hairpin flows from splitting. */
4875                 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4876                         const struct rte_flow_action_queue *queue;
4877
4878                         queue = qrss->conf;
4879                         if (mlx5_rxq_get_type(dev, queue->index) ==
4880                             MLX5_RXQ_TYPE_HAIRPIN)
4881                                 qrss = NULL;
4882                 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4883                         const struct rte_flow_action_rss *rss;
4884
4885                         rss = qrss->conf;
4886                         if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4887                             MLX5_RXQ_TYPE_HAIRPIN)
4888                                 qrss = NULL;
4889                 }
4890         }
4891         if (qrss) {
4892                 /* Check if it is in meter suffix table. */
4893                 mtr_sfx = attr->group == (attr->transfer ?
4894                           (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4895                           MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4896                 /*
4897                  * Q/RSS action on NIC Rx should be split in order to pass by
4898                  * the mreg copy table (RX_CP_TBL) and then it jumps to the
4899                  * action table (RX_ACT_TBL) which has the split Q/RSS action.
4900                  */
4901                 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4902                            sizeof(struct rte_flow_action_set_tag) +
4903                            sizeof(struct rte_flow_action_jump);
4904                 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4905                                           SOCKET_ID_ANY);
4906                 if (!ext_actions)
4907                         return rte_flow_error_set(error, ENOMEM,
4908                                                   RTE_FLOW_ERROR_TYPE_ACTION,
4909                                                   NULL, "no memory to split "
4910                                                   "metadata flow");
4911                 /*
4912                  * If we are the suffix flow of meter, tag already exist.
4913                  * Set the tag action to void.
4914                  */
4915                 if (mtr_sfx)
4916                         ext_actions[qrss - actions].type =
4917                                                 RTE_FLOW_ACTION_TYPE_VOID;
4918                 else
4919                         ext_actions[qrss - actions].type =
4920                                                 (enum rte_flow_action_type)
4921                                                 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4922                 /*
4923                  * Create the new actions list with removed Q/RSS action
4924                  * and appended set tag and jump to register copy table
4925                  * (RX_CP_TBL). We should preallocate unique tag ID here
4926                  * in advance, because it is needed for set tag action.
4927                  */
4928                 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4929                                                     qrss, actions_n, error);
4930                 if (!mtr_sfx && !qrss_id) {
4931                         ret = -rte_errno;
4932                         goto exit;
4933                 }
4934         } else if (attr->egress && !attr->transfer) {
4935                 /*
4936                  * All the actions on NIC Tx should have a metadata register
4937                  * copy action to copy reg_a from WQE to reg_c[meta]
4938                  */
4939                 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4940                            sizeof(struct mlx5_flow_action_copy_mreg);
4941                 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4942                                           SOCKET_ID_ANY);
4943                 if (!ext_actions)
4944                         return rte_flow_error_set(error, ENOMEM,
4945                                                   RTE_FLOW_ERROR_TYPE_ACTION,
4946                                                   NULL, "no memory to split "
4947                                                   "metadata flow");
4948                 /* Create the action list appended with copy register. */
4949                 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4950                                              actions_n, error, encap_idx);
4951                 if (ret < 0)
4952                         goto exit;
4953         }
4954         /* Add the unmodified original or prefix subflow. */
4955         ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
4956                                       items, ext_actions ? ext_actions :
4957                                       actions, flow_split_info, error);
4958         if (ret < 0)
4959                 goto exit;
4960         MLX5_ASSERT(dev_flow);
4961         if (qrss) {
4962                 const struct rte_flow_attr q_attr = {
4963                         .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4964                         .ingress = 1,
4965                 };
4966                 /* Internal PMD action to set register. */
4967                 struct mlx5_rte_flow_item_tag q_tag_spec = {
4968                         .data = qrss_id,
4969                         .id = REG_NON,
4970                 };
4971                 struct rte_flow_item q_items[] = {
4972                         {
4973                                 .type = (enum rte_flow_item_type)
4974                                         MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4975                                 .spec = &q_tag_spec,
4976                                 .last = NULL,
4977                                 .mask = NULL,
4978                         },
4979                         {
4980                                 .type = RTE_FLOW_ITEM_TYPE_END,
4981                         },
4982                 };
4983                 struct rte_flow_action q_actions[] = {
4984                         {
4985                                 .type = qrss->type,
4986                                 .conf = qrss->conf,
4987                         },
4988                         {
4989                                 .type = RTE_FLOW_ACTION_TYPE_END,
4990                         },
4991                 };
4992                 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4993
4994                 /*
4995                  * Configure the tag item only if there is no meter subflow.
4996                  * Since tag is already marked in the meter suffix subflow
4997                  * we can just use the meter suffix items as is.
4998                  */
4999                 if (qrss_id) {
5000                         /* Not meter subflow. */
5001                         MLX5_ASSERT(!mtr_sfx);
5002                         /*
5003                          * Put unique id in prefix flow due to it is destroyed
5004                          * after suffix flow and id will be freed after there
5005                          * is no actual flows with this id and identifier
5006                          * reallocation becomes possible (for example, for
5007                          * other flows in other threads).
5008                          */
5009                         dev_flow->handle->split_flow_id = qrss_id;
5010                         ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5011                                                    error);
5012                         if (ret < 0)
5013                                 goto exit;
5014                         q_tag_spec.id = ret;
5015                 }
5016                 dev_flow = NULL;
5017                 /* Add suffix subflow to execute Q/RSS. */
5018                 flow_split_info->prefix_layers = layers;
5019                 flow_split_info->prefix_mark = 0;
5020                 ret = flow_create_split_inner(dev, flow, &dev_flow,
5021                                               &q_attr, mtr_sfx ? items :
5022                                               q_items, q_actions,
5023                                               flow_split_info, error);
5024                 if (ret < 0)
5025                         goto exit;
5026                 /* qrss ID should be freed if failed. */
5027                 qrss_id = 0;
5028                 MLX5_ASSERT(dev_flow);
5029         }
5030
5031 exit:
5032         /*
5033          * We do not destroy the partially created sub_flows in case of error.
5034          * These ones are included into parent flow list and will be destroyed
5035          * by flow_drv_destroy.
5036          */
5037         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5038                         qrss_id);
5039         mlx5_free(ext_actions);
5040         return ret;
5041 }
5042
5043 /**
5044  * The splitting for meter feature.
5045  *
5046  * - The meter flow will be split to two flows as prefix and
5047  *   suffix flow. The packets make sense only it pass the prefix
5048  *   meter action.
5049  *
5050  * - Reg_C_5 is used for the packet to match betweend prefix and
5051  *   suffix flow.
5052  *
5053  * @param dev
5054  *   Pointer to Ethernet device.
5055  * @param[in] flow
5056  *   Parent flow structure pointer.
5057  * @param[in] attr
5058  *   Flow rule attributes.
5059  * @param[in] items
5060  *   Pattern specification (list terminated by the END pattern item).
5061  * @param[in] actions
5062  *   Associated actions (list terminated by the END action).
5063  * @param[in] flow_split_info
5064  *   Pointer to flow split info structure.
5065  * @param[out] error
5066  *   Perform verbose error reporting if not NULL.
5067  * @return
5068  *   0 on success, negative value otherwise
5069  */
5070 static int
5071 flow_create_split_meter(struct rte_eth_dev *dev,
5072                         struct rte_flow *flow,
5073                         const struct rte_flow_attr *attr,
5074                         const struct rte_flow_item items[],
5075                         const struct rte_flow_action actions[],
5076                         struct mlx5_flow_split_info *flow_split_info,
5077                         struct rte_flow_error *error)
5078 {
5079         struct mlx5_priv *priv = dev->data->dev_private;
5080         struct rte_flow_action *sfx_actions = NULL;
5081         struct rte_flow_action *pre_actions = NULL;
5082         struct rte_flow_item *sfx_items = NULL;
5083         struct mlx5_flow *dev_flow = NULL;
5084         struct rte_flow_attr sfx_attr = *attr;
5085         uint32_t mtr = 0;
5086         uint32_t mtr_tag_id = 0;
5087         size_t act_size;
5088         size_t item_size;
5089         int actions_n = 0;
5090         int ret;
5091
5092         if (priv->mtr_en)
5093                 actions_n = flow_check_meter_action(actions, &mtr);
5094         if (mtr) {
5095                 /* The five prefix actions: meter, decap, encap, tag, end. */
5096                 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5097                            sizeof(struct mlx5_rte_flow_action_set_tag);
5098                 /* tag, vlan, port id, end. */
5099 #define METER_SUFFIX_ITEM 4
5100                 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5101                             sizeof(struct mlx5_rte_flow_item_tag) * 2;
5102                 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5103                                           0, SOCKET_ID_ANY);
5104                 if (!sfx_actions)
5105                         return rte_flow_error_set(error, ENOMEM,
5106                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5107                                                   NULL, "no memory to split "
5108                                                   "meter flow");
5109                 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5110                              act_size);
5111                 pre_actions = sfx_actions + actions_n;
5112                 mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5113                                                    actions, sfx_actions,
5114                                                    pre_actions);
5115                 if (!mtr_tag_id) {
5116                         ret = -rte_errno;
5117                         goto exit;
5118                 }
5119                 /* Add the prefix subflow. */
5120                 flow_split_info->prefix_mark = 0;
5121                 ret = flow_create_split_inner(dev, flow, &dev_flow,
5122                                               attr, items, pre_actions,
5123                                               flow_split_info, error);
5124                 if (ret) {
5125                         ret = -rte_errno;
5126                         goto exit;
5127                 }
5128                 dev_flow->handle->split_flow_id = mtr_tag_id;
5129                 /* Setting the sfx group atrr. */
5130                 sfx_attr.group = sfx_attr.transfer ?
5131                                 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5132                                  MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5133                 flow_split_info->prefix_layers =
5134                                 flow_get_prefix_layer_flags(dev_flow);
5135                 flow_split_info->prefix_mark = dev_flow->handle->mark;
5136         }
5137         /* Add the prefix subflow. */
5138         ret = flow_create_split_metadata(dev, flow,
5139                                          &sfx_attr, sfx_items ?
5140                                          sfx_items : items,
5141                                          sfx_actions ? sfx_actions : actions,
5142                                          flow_split_info, error);
5143 exit:
5144         if (sfx_actions)
5145                 mlx5_free(sfx_actions);
5146         return ret;
5147 }
5148
5149 /**
5150  * The splitting for sample feature.
5151  *
5152  * Once Sample action is detected in the action list, the flow actions should
5153  * be split into prefix sub flow and suffix sub flow.
5154  *
5155  * The original items remain in the prefix sub flow, all actions preceding the
5156  * sample action and the sample action itself will be copied to the prefix
5157  * sub flow, the actions following the sample action will be copied to the
5158  * suffix sub flow, Queue action always be located in the suffix sub flow.
5159  *
5160  * In order to make the packet from prefix sub flow matches with suffix sub
5161  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5162  * flow uses tag item with the unique flow id.
5163  *
5164  * @param dev
5165  *   Pointer to Ethernet device.
5166  * @param[in] flow
5167  *   Parent flow structure pointer.
5168  * @param[in] attr
5169  *   Flow rule attributes.
5170  * @param[in] items
5171  *   Pattern specification (list terminated by the END pattern item).
5172  * @param[in] actions
5173  *   Associated actions (list terminated by the END action).
5174  * @param[in] flow_split_info
5175  *   Pointer to flow split info structure.
5176  * @param[out] error
5177  *   Perform verbose error reporting if not NULL.
5178  * @return
5179  *   0 on success, negative value otherwise
5180  */
5181 static int
5182 flow_create_split_sample(struct rte_eth_dev *dev,
5183                          struct rte_flow *flow,
5184                          const struct rte_flow_attr *attr,
5185                          const struct rte_flow_item items[],
5186                          const struct rte_flow_action actions[],
5187                          struct mlx5_flow_split_info *flow_split_info,
5188                          struct rte_flow_error *error)
5189 {
5190         struct mlx5_priv *priv = dev->data->dev_private;
5191         struct rte_flow_action *sfx_actions = NULL;
5192         struct rte_flow_action *pre_actions = NULL;
5193         struct rte_flow_item *sfx_items = NULL;
5194         struct mlx5_flow *dev_flow = NULL;
5195         struct rte_flow_attr sfx_attr = *attr;
5196 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5197         struct mlx5_flow_dv_sample_resource *sample_res;
5198         struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5199         struct mlx5_flow_tbl_resource *sfx_tbl;
5200 #endif
5201         size_t act_size;
5202         size_t item_size;
5203         uint32_t fdb_tx = 0;
5204         int32_t tag_id = 0;
5205         int actions_n = 0;
5206         int sample_action_pos;
5207         int qrss_action_pos;
5208         int ret = 0;
5209
5210         if (priv->sampler_en)
5211                 actions_n = flow_check_match_action(actions, attr,
5212                                         RTE_FLOW_ACTION_TYPE_SAMPLE,
5213                                         &sample_action_pos, &qrss_action_pos);
5214         if (actions_n) {
5215                 /* The prefix actions must includes sample, tag, end. */
5216                 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5217                            + sizeof(struct mlx5_rte_flow_action_set_tag);
5218                 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5219                             sizeof(struct mlx5_rte_flow_item_tag) * 2;
5220                 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5221                                           item_size), 0, SOCKET_ID_ANY);
5222                 if (!sfx_actions)
5223                         return rte_flow_error_set(error, ENOMEM,
5224                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5225                                                   NULL, "no memory to split "
5226                                                   "sample flow");
5227                 /* The representor_id is -1 for uplink. */
5228                 fdb_tx = (attr->transfer && priv->representor_id != -1);
5229                 if (!fdb_tx)
5230                         sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5231                                         + act_size);
5232                 pre_actions = sfx_actions + actions_n;
5233                 tag_id = flow_sample_split_prep(dev, fdb_tx, sfx_items,
5234                                                 actions, sfx_actions,
5235                                                 pre_actions, actions_n,
5236                                                 sample_action_pos,
5237                                                 qrss_action_pos, error);
5238                 if (tag_id < 0 || (!fdb_tx && !tag_id)) {
5239                         ret = -rte_errno;
5240                         goto exit;
5241                 }
5242                 /* Add the prefix subflow. */
5243                 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5244                                               items, pre_actions,
5245                                               flow_split_info, error);
5246                 if (ret) {
5247                         ret = -rte_errno;
5248                         goto exit;
5249                 }
5250                 dev_flow->handle->split_flow_id = tag_id;
5251 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5252                 /* Set the sfx group attr. */
5253                 sample_res = (struct mlx5_flow_dv_sample_resource *)
5254                                         dev_flow->dv.sample_res;
5255                 sfx_tbl = (struct mlx5_flow_tbl_resource *)
5256                                         sample_res->normal_path_tbl;
5257                 sfx_tbl_data = container_of(sfx_tbl,
5258                                         struct mlx5_flow_tbl_data_entry, tbl);
5259                 sfx_attr.group = sfx_attr.transfer ?
5260                                         (sfx_tbl_data->table_id - 1) :
5261                                          sfx_tbl_data->table_id;
5262                 flow_split_info->prefix_layers =
5263                                 flow_get_prefix_layer_flags(dev_flow);
5264                 flow_split_info->prefix_mark = dev_flow->handle->mark;
5265                 /* Suffix group level already be scaled with factor, set
5266                  * skip_scale to 1 to avoid scale again in translation.
5267                  */
5268                 flow_split_info->skip_scale = 1;
5269 #endif
5270         }
5271         /* Add the suffix subflow. */
5272         ret = flow_create_split_meter(dev, flow, &sfx_attr,
5273                                       sfx_items ? sfx_items : items,
5274                                       sfx_actions ? sfx_actions : actions,
5275                                       flow_split_info, error);
5276 exit:
5277         if (sfx_actions)
5278                 mlx5_free(sfx_actions);
5279         return ret;
5280 }
5281
5282 /**
5283  * Split the flow to subflow set. The splitters might be linked
5284  * in the chain, like this:
5285  * flow_create_split_outer() calls:
5286  *   flow_create_split_meter() calls:
5287  *     flow_create_split_metadata(meter_subflow_0) calls:
5288  *       flow_create_split_inner(metadata_subflow_0)
5289  *       flow_create_split_inner(metadata_subflow_1)
5290  *       flow_create_split_inner(metadata_subflow_2)
5291  *     flow_create_split_metadata(meter_subflow_1) calls:
5292  *       flow_create_split_inner(metadata_subflow_0)
5293  *       flow_create_split_inner(metadata_subflow_1)
5294  *       flow_create_split_inner(metadata_subflow_2)
5295  *
5296  * This provide flexible way to add new levels of flow splitting.
5297  * The all of successfully created subflows are included to the
5298  * parent flow dev_flow list.
5299  *
5300  * @param dev
5301  *   Pointer to Ethernet device.
5302  * @param[in] flow
5303  *   Parent flow structure pointer.
5304  * @param[in] attr
5305  *   Flow rule attributes.
5306  * @param[in] items
5307  *   Pattern specification (list terminated by the END pattern item).
5308  * @param[in] actions
5309  *   Associated actions (list terminated by the END action).
5310  * @param[in] flow_split_info
5311  *   Pointer to flow split info structure.
5312  * @param[out] error
5313  *   Perform verbose error reporting if not NULL.
5314  * @return
5315  *   0 on success, negative value otherwise
5316  */
5317 static int
5318 flow_create_split_outer(struct rte_eth_dev *dev,
5319                         struct rte_flow *flow,
5320                         const struct rte_flow_attr *attr,
5321                         const struct rte_flow_item items[],
5322                         const struct rte_flow_action actions[],
5323                         struct mlx5_flow_split_info *flow_split_info,
5324                         struct rte_flow_error *error)
5325 {
5326         int ret;
5327
5328         ret = flow_create_split_sample(dev, flow, attr, items,
5329                                        actions, flow_split_info, error);
5330         MLX5_ASSERT(ret <= 0);
5331         return ret;
5332 }
5333
5334 static struct mlx5_flow_tunnel *
5335 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5336                       const struct rte_flow_attr *attr,
5337                       const struct rte_flow_item items[],
5338                       const struct rte_flow_action actions[])
5339 {
5340         struct mlx5_flow_tunnel *tunnel;
5341
5342 #pragma GCC diagnostic push
5343 #pragma GCC diagnostic ignored "-Wcast-qual"
5344         if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5345                 tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5346         else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5347                 tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5348         else
5349                 tunnel = NULL;
5350 #pragma GCC diagnostic pop
5351
5352         return tunnel;
5353 }
5354
5355 /**
5356  * Adjust flow RSS workspace if needed.
5357  *
5358  * @param wks
5359  *   Pointer to thread flow work space.
5360  * @param rss_desc
5361  *   Pointer to RSS descriptor.
5362  * @param[in] nrssq_num
5363  *   New RSS queue number.
5364  *
5365  * @return
5366  *   0 on success, -1 otherwise and rte_errno is set.
5367  */
5368 static int
5369 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5370                           struct mlx5_flow_rss_desc *rss_desc,
5371                           uint32_t nrssq_num)
5372 {
5373         if (likely(nrssq_num <= wks->rssq_num))
5374                 return 0;
5375         rss_desc->queue = realloc(rss_desc->queue,
5376                           sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5377         if (!rss_desc->queue) {
5378                 rte_errno = ENOMEM;
5379                 return -1;
5380         }
5381         wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5382         return 0;
5383 }
5384
5385 /**
5386  * Create a flow and add it to @p list.
5387  *
5388  * @param dev
5389  *   Pointer to Ethernet device.
5390  * @param list
5391  *   Pointer to a TAILQ flow list. If this parameter NULL,
5392  *   no list insertion occurred, flow is just created,
5393  *   this is caller's responsibility to track the
5394  *   created flow.
5395  * @param[in] attr
5396  *   Flow rule attributes.
5397  * @param[in] items
5398  *   Pattern specification (list terminated by the END pattern item).
5399  * @param[in] actions
5400  *   Associated actions (list terminated by the END action).
5401  * @param[in] external
5402  *   This flow rule is created by request external to PMD.
5403  * @param[out] error
5404  *   Perform verbose error reporting if not NULL.
5405  *
5406  * @return
5407  *   A flow index on success, 0 otherwise and rte_errno is set.
5408  */
5409 static uint32_t
5410 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5411                  const struct rte_flow_attr *attr,
5412                  const struct rte_flow_item items[],
5413                  const struct rte_flow_action original_actions[],
5414                  bool external, struct rte_flow_error *error)
5415 {
5416         struct mlx5_priv *priv = dev->data->dev_private;
5417         struct rte_flow *flow = NULL;
5418         struct mlx5_flow *dev_flow;
5419         const struct rte_flow_action_rss *rss = NULL;
5420         struct mlx5_translated_shared_action
5421                 shared_actions[MLX5_MAX_SHARED_ACTIONS];
5422         int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5423         union {
5424                 struct mlx5_flow_expand_rss buf;
5425                 uint8_t buffer[2048];
5426         } expand_buffer;
5427         union {
5428                 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5429                 uint8_t buffer[2048];
5430         } actions_rx;
5431         union {
5432                 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5433                 uint8_t buffer[2048];
5434         } actions_hairpin_tx;
5435         union {
5436                 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5437                 uint8_t buffer[2048];
5438         } items_tx;
5439         struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5440         struct mlx5_flow_rss_desc *rss_desc;
5441         const struct rte_flow_action *p_actions_rx;
5442         uint32_t i;
5443         uint32_t idx = 0;
5444         int hairpin_flow;
5445         struct rte_flow_attr attr_tx = { .priority = 0 };
5446         const struct rte_flow_action *actions;
5447         struct rte_flow_action *translated_actions = NULL;
5448         struct mlx5_flow_tunnel *tunnel;
5449         struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5450         struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5451         struct mlx5_flow_split_info flow_split_info = {
5452                 .external = !!external,
5453                 .skip_scale = 0,
5454                 .flow_idx = 0,
5455                 .prefix_mark = 0,
5456                 .prefix_layers = 0
5457         };
5458         int ret;
5459
5460         MLX5_ASSERT(wks);
5461         rss_desc = &wks->rss_desc;
5462         ret = flow_shared_actions_translate(dev, original_actions,
5463                                             shared_actions,
5464                                             &shared_actions_n,
5465                                             &translated_actions, error);
5466         if (ret < 0) {
5467                 MLX5_ASSERT(translated_actions == NULL);
5468                 return 0;
5469         }
5470         actions = translated_actions ? translated_actions : original_actions;
5471         p_actions_rx = actions;
5472         hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5473         ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5474                                 external, hairpin_flow, error);
5475         if (ret < 0)
5476                 goto error_before_hairpin_split;
5477         flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5478         if (!flow) {
5479                 rte_errno = ENOMEM;
5480                 goto error_before_hairpin_split;
5481         }
5482         if (hairpin_flow > 0) {
5483                 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5484                         rte_errno = EINVAL;
5485                         goto error_before_hairpin_split;
5486                 }
5487                 flow_hairpin_split(dev, actions, actions_rx.actions,
5488                                    actions_hairpin_tx.actions, items_tx.items,
5489                                    idx);
5490                 p_actions_rx = actions_rx.actions;
5491         }
5492         flow_split_info.flow_idx = idx;
5493         flow->drv_type = flow_get_drv_type(dev, attr);
5494         MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5495                     flow->drv_type < MLX5_FLOW_TYPE_MAX);
5496         memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5497         /* RSS Action only works on NIC RX domain */
5498         if (attr->ingress && !attr->transfer)
5499                 rss = flow_get_rss_action(p_actions_rx);
5500         if (rss) {
5501                 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5502                         return 0;
5503                 /*
5504                  * The following information is required by
5505                  * mlx5_flow_hashfields_adjust() in advance.
5506                  */
5507                 rss_desc->level = rss->level;
5508                 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5509                 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5510         }
5511         flow->dev_handles = 0;
5512         if (rss && rss->types) {
5513                 unsigned int graph_root;
5514
5515                 graph_root = find_graph_root(items, rss->level);
5516                 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5517                                            items, rss->types,
5518                                            mlx5_support_expansion, graph_root);
5519                 MLX5_ASSERT(ret > 0 &&
5520                        (unsigned int)ret < sizeof(expand_buffer.buffer));
5521         } else {
5522                 buf->entries = 1;
5523                 buf->entry[0].pattern = (void *)(uintptr_t)items;
5524         }
5525         rss_desc->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
5526                                                       shared_actions_n);
5527         for (i = 0; i < buf->entries; ++i) {
5528                 /* Initialize flow split data. */
5529                 flow_split_info.prefix_layers = 0;
5530                 flow_split_info.prefix_mark = 0;
5531                 flow_split_info.skip_scale = 0;
5532                 /*
5533                  * The splitter may create multiple dev_flows,
5534                  * depending on configuration. In the simplest
5535                  * case it just creates unmodified original flow.
5536                  */
5537                 ret = flow_create_split_outer(dev, flow, attr,
5538                                               buf->entry[i].pattern,
5539                                               p_actions_rx, &flow_split_info,
5540                                               error);
5541                 if (ret < 0)
5542                         goto error;
5543                 if (is_flow_tunnel_steer_rule(dev, attr,
5544                                               buf->entry[i].pattern,
5545                                               p_actions_rx)) {
5546                         ret = flow_tunnel_add_default_miss(dev, flow, attr,
5547                                                            p_actions_rx,
5548                                                            idx,
5549                                                            &default_miss_ctx,
5550                                                            error);
5551                         if (ret < 0) {
5552                                 mlx5_free(default_miss_ctx.queue);
5553                                 goto error;
5554                         }
5555                 }
5556         }
5557         /* Create the tx flow. */
5558         if (hairpin_flow) {
5559                 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5560                 attr_tx.ingress = 0;
5561                 attr_tx.egress = 1;
5562                 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5563                                          actions_hairpin_tx.actions,
5564                                          idx, error);
5565                 if (!dev_flow)
5566                         goto error;
5567                 dev_flow->flow = flow;
5568                 dev_flow->external = 0;
5569                 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5570                               dev_flow->handle, next);
5571                 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5572                                          items_tx.items,
5573                                          actions_hairpin_tx.actions, error);
5574                 if (ret < 0)
5575                         goto error;
5576         }
5577         /*
5578          * Update the metadata register copy table. If extensive
5579          * metadata feature is enabled and registers are supported
5580          * we might create the extra rte_flow for each unique
5581          * MARK/FLAG action ID.
5582          *
5583          * The table is updated for ingress Flows only, because
5584          * the egress Flows belong to the different device and
5585          * copy table should be updated in peer NIC Rx domain.
5586          */
5587         if (attr->ingress &&
5588             (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5589                 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5590                 if (ret)
5591                         goto error;
5592         }
5593         /*
5594          * If the flow is external (from application) OR device is started,
5595          * OR mreg discover, then apply immediately.
5596          */
5597         if (external || dev->data->dev_started ||
5598             (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5599              attr->priority == MLX5_FLOW_PRIO_RSVD)) {
5600                 ret = flow_drv_apply(dev, flow, error);
5601                 if (ret < 0)
5602                         goto error;
5603         }
5604         if (list) {
5605                 rte_spinlock_lock(&priv->flow_list_lock);
5606                 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5607                              flow, next);
5608                 rte_spinlock_unlock(&priv->flow_list_lock);
5609         }
5610         flow_rxq_flags_set(dev, flow);
5611         rte_free(translated_actions);
5612         tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5613         if (tunnel) {
5614                 flow->tunnel = 1;
5615                 flow->tunnel_id = tunnel->tunnel_id;
5616                 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5617                 mlx5_free(default_miss_ctx.queue);
5618         }
5619         mlx5_flow_pop_thread_workspace();
5620         return idx;
5621 error:
5622         MLX5_ASSERT(flow);
5623         ret = rte_errno; /* Save rte_errno before cleanup. */
5624         flow_mreg_del_copy_action(dev, flow);
5625         flow_drv_destroy(dev, flow);
5626         if (rss_desc->shared_rss)
5627                 __atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5628                         mlx5_ipool_get
5629                         (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5630                         rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5631         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5632         rte_errno = ret; /* Restore rte_errno. */
5633         ret = rte_errno;
5634         rte_errno = ret;
5635         mlx5_flow_pop_thread_workspace();
5636 error_before_hairpin_split:
5637         rte_free(translated_actions);
5638         return 0;
5639 }
5640
5641 /**
5642  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5643  * incoming packets to table 1.
5644  *
5645  * Other flow rules, requested for group n, will be created in
5646  * e-switch table n+1.
5647  * Jump action to e-switch group n will be created to group n+1.
5648  *
5649  * Used when working in switchdev mode, to utilise advantages of table 1
5650  * and above.
5651  *
5652  * @param dev
5653  *   Pointer to Ethernet device.
5654  *
5655  * @return
5656  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5657  */
5658 struct rte_flow *
5659 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5660 {
5661         const struct rte_flow_attr attr = {
5662                 .group = 0,
5663                 .priority = 0,
5664                 .ingress = 1,
5665                 .egress = 0,
5666                 .transfer = 1,
5667         };
5668         const struct rte_flow_item pattern = {
5669                 .type = RTE_FLOW_ITEM_TYPE_END,
5670         };
5671         struct rte_flow_action_jump jump = {
5672                 .group = 1,
5673         };
5674         const struct rte_flow_action actions[] = {
5675                 {
5676                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
5677                         .conf = &jump,
5678                 },
5679                 {
5680                         .type = RTE_FLOW_ACTION_TYPE_END,
5681                 },
5682         };
5683         struct mlx5_priv *priv = dev->data->dev_private;
5684         struct rte_flow_error error;
5685
5686         return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5687                                                    &attr, &pattern,
5688                                                    actions, false, &error);
5689 }
5690
5691 /**
5692  * Validate a flow supported by the NIC.
5693  *
5694  * @see rte_flow_validate()
5695  * @see rte_flow_ops
5696  */
5697 int
5698 mlx5_flow_validate(struct rte_eth_dev *dev,
5699                    const struct rte_flow_attr *attr,
5700                    const struct rte_flow_item items[],
5701                    const struct rte_flow_action original_actions[],
5702                    struct rte_flow_error *error)
5703 {
5704         int hairpin_flow;
5705         struct mlx5_translated_shared_action
5706                 shared_actions[MLX5_MAX_SHARED_ACTIONS];
5707         int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5708         const struct rte_flow_action *actions;
5709         struct rte_flow_action *translated_actions = NULL;
5710         int ret = flow_shared_actions_translate(dev, original_actions,
5711                                                 shared_actions,
5712                                                 &shared_actions_n,
5713                                                 &translated_actions, error);
5714
5715         if (ret)
5716                 return ret;
5717         actions = translated_actions ? translated_actions : original_actions;
5718         hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5719         ret = flow_drv_validate(dev, attr, items, actions,
5720                                 true, hairpin_flow, error);
5721         rte_free(translated_actions);
5722         return ret;
5723 }
5724
5725 /**
5726  * Create a flow.
5727  *
5728  * @see rte_flow_create()
5729  * @see rte_flow_ops
5730  */
5731 struct rte_flow *
5732 mlx5_flow_create(struct rte_eth_dev *dev,
5733                  const struct rte_flow_attr *attr,
5734                  const struct rte_flow_item items[],
5735                  const struct rte_flow_action actions[],
5736                  struct rte_flow_error *error)
5737 {
5738         struct mlx5_priv *priv = dev->data->dev_private;
5739
5740         /*
5741          * If the device is not started yet, it is not allowed to created a
5742          * flow from application. PMD default flows and traffic control flows
5743          * are not affected.
5744          */
5745         if (unlikely(!dev->data->dev_started)) {
5746                 DRV_LOG(DEBUG, "port %u is not started when "
5747                         "inserting a flow", dev->data->port_id);
5748                 rte_flow_error_set(error, ENODEV,
5749                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5750                                    NULL,
5751                                    "port not started");
5752                 return NULL;
5753         }
5754
5755         return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5756                                   attr, items, actions, true, error);
5757 }
5758
5759 /**
5760  * Destroy a flow in a list.
5761  *
5762  * @param dev
5763  *   Pointer to Ethernet device.
5764  * @param list
5765  *   Pointer to the Indexed flow list. If this parameter NULL,
5766  *   there is no flow removal from the list. Be noted that as
5767  *   flow is add to the indexed list, memory of the indexed
5768  *   list points to maybe changed as flow destroyed.
5769  * @param[in] flow_idx
5770  *   Index of flow to destroy.
5771  */
5772 static void
5773 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5774                   uint32_t flow_idx)
5775 {
5776         struct mlx5_priv *priv = dev->data->dev_private;
5777         struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5778                                                [MLX5_IPOOL_RTE_FLOW], flow_idx);
5779
5780         if (!flow)
5781                 return;
5782         /*
5783          * Update RX queue flags only if port is started, otherwise it is
5784          * already clean.
5785          */
5786         if (dev->data->dev_started)
5787                 flow_rxq_flags_trim(dev, flow);
5788         flow_drv_destroy(dev, flow);
5789         if (list) {
5790                 rte_spinlock_lock(&priv->flow_list_lock);
5791                 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5792                              flow_idx, flow, next);
5793                 rte_spinlock_unlock(&priv->flow_list_lock);
5794         }
5795         if (flow->tunnel) {
5796                 struct mlx5_flow_tunnel *tunnel;
5797
5798                 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5799                 RTE_VERIFY(tunnel);
5800                 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5801                         mlx5_flow_tunnel_free(dev, tunnel);
5802         }
5803         flow_mreg_del_copy_action(dev, flow);
5804         mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5805 }
5806
5807 /**
5808  * Destroy all flows.
5809  *
5810  * @param dev
5811  *   Pointer to Ethernet device.
5812  * @param list
5813  *   Pointer to the Indexed flow list.
5814  * @param active
5815  *   If flushing is called avtively.
5816  */
5817 void
5818 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5819 {
5820         uint32_t num_flushed = 0;
5821
5822         while (*list) {
5823                 flow_list_destroy(dev, list, *list);
5824                 num_flushed++;
5825         }
5826         if (active) {
5827                 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5828                         dev->data->port_id, num_flushed);
5829         }
5830 }
5831
5832 /**
5833  * Stop all default actions for flows.
5834  *
5835  * @param dev
5836  *   Pointer to Ethernet device.
5837  */
5838 void
5839 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5840 {
5841         flow_mreg_del_default_copy_action(dev);
5842         flow_rxq_flags_clear(dev);
5843 }
5844
5845 /**
5846  * Start all default actions for flows.
5847  *
5848  * @param dev
5849  *   Pointer to Ethernet device.
5850  * @return
5851  *   0 on success, a negative errno value otherwise and rte_errno is set.
5852  */
5853 int
5854 mlx5_flow_start_default(struct rte_eth_dev *dev)
5855 {
5856         struct rte_flow_error error;
5857
5858         /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5859         return flow_mreg_add_default_copy_action(dev, &error);
5860 }
5861
5862 /**
5863  * Release key of thread specific flow workspace data.
5864  */
5865 void
5866 flow_release_workspace(void *data)
5867 {
5868         struct mlx5_flow_workspace *wks = data;
5869         struct mlx5_flow_workspace *next;
5870
5871         while (wks) {
5872                 next = wks->next;
5873                 free(wks->rss_desc.queue);
5874                 free(wks);
5875                 wks = next;
5876         }
5877 }
5878
5879 /**
5880  * Get thread specific current flow workspace.
5881  *
5882  * @return pointer to thread specific flow workspace data, NULL on error.
5883  */
5884 struct mlx5_flow_workspace*
5885 mlx5_flow_get_thread_workspace(void)
5886 {
5887         struct mlx5_flow_workspace *data;
5888
5889         data = mlx5_flow_os_get_specific_workspace();
5890         MLX5_ASSERT(data && data->inuse);
5891         if (!data || !data->inuse)
5892                 DRV_LOG(ERR, "flow workspace not initialized.");
5893         return data;
5894 }
5895
5896 /**
5897  * Allocate and init new flow workspace.
5898  *
5899  * @return pointer to flow workspace data, NULL on error.
5900  */
5901 static struct mlx5_flow_workspace*
5902 flow_alloc_thread_workspace(void)
5903 {
5904         struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
5905
5906         if (!data) {
5907                 DRV_LOG(ERR, "Failed to allocate flow workspace "
5908                         "memory.");
5909                 return NULL;
5910         }
5911         data->rss_desc.queue = calloc(1,
5912                         sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
5913         if (!data->rss_desc.queue)
5914                 goto err;
5915         data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
5916         return data;
5917 err:
5918         if (data->rss_desc.queue)
5919                 free(data->rss_desc.queue);
5920         free(data);
5921         return NULL;
5922 }
5923
5924 /**
5925  * Get new thread specific flow workspace.
5926  *
5927  * If current workspace inuse, create new one and set as current.
5928  *
5929  * @return pointer to thread specific flow workspace data, NULL on error.
5930  */
5931 static struct mlx5_flow_workspace*
5932 mlx5_flow_push_thread_workspace(void)
5933 {
5934         struct mlx5_flow_workspace *curr;
5935         struct mlx5_flow_workspace *data;
5936
5937         curr = mlx5_flow_os_get_specific_workspace();
5938         if (!curr) {
5939                 data = flow_alloc_thread_workspace();
5940                 if (!data)
5941                         return NULL;
5942         } else if (!curr->inuse) {
5943                 data = curr;
5944         } else if (curr->next) {
5945                 data = curr->next;
5946         } else {
5947                 data = flow_alloc_thread_workspace();
5948                 if (!data)
5949                         return NULL;
5950                 curr->next = data;
5951                 data->prev = curr;
5952         }
5953         data->inuse = 1;
5954         data->flow_idx = 0;
5955         /* Set as current workspace */
5956         if (mlx5_flow_os_set_specific_workspace(data))
5957                 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
5958         return data;
5959 }
5960
5961 /**
5962  * Close current thread specific flow workspace.
5963  *
5964  * If previous workspace available, set it as current.
5965  *
5966  * @return pointer to thread specific flow workspace data, NULL on error.
5967  */
5968 static void
5969 mlx5_flow_pop_thread_workspace(void)
5970 {
5971         struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
5972
5973         if (!data)
5974                 return;
5975         if (!data->inuse) {
5976                 DRV_LOG(ERR, "Failed to close unused flow workspace.");
5977                 return;
5978         }
5979         data->inuse = 0;
5980         if (!data->prev)
5981                 return;
5982         if (mlx5_flow_os_set_specific_workspace(data->prev))
5983                 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
5984 }
5985
5986 /**
5987  * Verify the flow list is empty
5988  *
5989  * @param dev
5990  *  Pointer to Ethernet device.
5991  *
5992  * @return the number of flows not released.
5993  */
5994 int
5995 mlx5_flow_verify(struct rte_eth_dev *dev)
5996 {
5997         struct mlx5_priv *priv = dev->data->dev_private;
5998         struct rte_flow *flow;
5999         uint32_t idx;
6000         int ret = 0;
6001
6002         ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6003                       flow, next) {
6004                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
6005                         dev->data->port_id, (void *)flow);
6006                 ++ret;
6007         }
6008         return ret;
6009 }
6010
6011 /**
6012  * Enable default hairpin egress flow.
6013  *
6014  * @param dev
6015  *   Pointer to Ethernet device.
6016  * @param queue
6017  *   The queue index.
6018  *
6019  * @return
6020  *   0 on success, a negative errno value otherwise and rte_errno is set.
6021  */
6022 int
6023 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6024                             uint32_t queue)
6025 {
6026         struct mlx5_priv *priv = dev->data->dev_private;
6027         const struct rte_flow_attr attr = {
6028                 .egress = 1,
6029                 .priority = 0,
6030         };
6031         struct mlx5_rte_flow_item_tx_queue queue_spec = {
6032                 .queue = queue,
6033         };
6034         struct mlx5_rte_flow_item_tx_queue queue_mask = {
6035                 .queue = UINT32_MAX,
6036         };
6037         struct rte_flow_item items[] = {
6038                 {
6039                         .type = (enum rte_flow_item_type)
6040                                 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6041                         .spec = &queue_spec,
6042                         .last = NULL,
6043                         .mask = &queue_mask,
6044                 },
6045                 {
6046                         .type = RTE_FLOW_ITEM_TYPE_END,
6047                 },
6048         };
6049         struct rte_flow_action_jump jump = {
6050                 .group = MLX5_HAIRPIN_TX_TABLE,
6051         };
6052         struct rte_flow_action actions[2];
6053         uint32_t flow_idx;
6054         struct rte_flow_error error;
6055
6056         actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6057         actions[0].conf = &jump;
6058         actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6059         flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6060                                 &attr, items, actions, false, &error);
6061         if (!flow_idx) {
6062                 DRV_LOG(DEBUG,
6063                         "Failed to create ctrl flow: rte_errno(%d),"
6064                         " type(%d), message(%s)",
6065                         rte_errno, error.type,
6066                         error.message ? error.message : " (no stated reason)");
6067                 return -rte_errno;
6068         }
6069         return 0;
6070 }
6071
6072 /**
6073  * Enable a control flow configured from the control plane.
6074  *
6075  * @param dev
6076  *   Pointer to Ethernet device.
6077  * @param eth_spec
6078  *   An Ethernet flow spec to apply.
6079  * @param eth_mask
6080  *   An Ethernet flow mask to apply.
6081  * @param vlan_spec
6082  *   A VLAN flow spec to apply.
6083  * @param vlan_mask
6084  *   A VLAN flow mask to apply.
6085  *
6086  * @return
6087  *   0 on success, a negative errno value otherwise and rte_errno is set.
6088  */
6089 int
6090 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6091                     struct rte_flow_item_eth *eth_spec,
6092                     struct rte_flow_item_eth *eth_mask,
6093                     struct rte_flow_item_vlan *vlan_spec,
6094                     struct rte_flow_item_vlan *vlan_mask)
6095 {
6096         struct mlx5_priv *priv = dev->data->dev_private;
6097         const struct rte_flow_attr attr = {
6098                 .ingress = 1,
6099                 .priority = MLX5_FLOW_PRIO_RSVD,
6100         };
6101         struct rte_flow_item items[] = {
6102                 {
6103                         .type = RTE_FLOW_ITEM_TYPE_ETH,
6104                         .spec = eth_spec,
6105                         .last = NULL,
6106                         .mask = eth_mask,
6107                 },
6108                 {
6109                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6110                                               RTE_FLOW_ITEM_TYPE_END,
6111                         .spec = vlan_spec,
6112                         .last = NULL,
6113                         .mask = vlan_mask,
6114                 },
6115                 {
6116                         .type = RTE_FLOW_ITEM_TYPE_END,
6117                 },
6118         };
6119         uint16_t queue[priv->reta_idx_n];
6120         struct rte_flow_action_rss action_rss = {
6121                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6122                 .level = 0,
6123                 .types = priv->rss_conf.rss_hf,
6124                 .key_len = priv->rss_conf.rss_key_len,
6125                 .queue_num = priv->reta_idx_n,
6126                 .key = priv->rss_conf.rss_key,
6127                 .queue = queue,
6128         };
6129         struct rte_flow_action actions[] = {
6130                 {
6131                         .type = RTE_FLOW_ACTION_TYPE_RSS,
6132                         .conf = &action_rss,
6133                 },
6134                 {
6135                         .type = RTE_FLOW_ACTION_TYPE_END,
6136                 },
6137         };
6138         uint32_t flow_idx;
6139         struct rte_flow_error error;
6140         unsigned int i;
6141
6142         if (!priv->reta_idx_n || !priv->rxqs_n) {
6143                 return 0;
6144         }
6145         if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6146                 action_rss.types = 0;
6147         for (i = 0; i != priv->reta_idx_n; ++i)
6148                 queue[i] = (*priv->reta_idx)[i];
6149         flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6150                                 &attr, items, actions, false, &error);
6151         if (!flow_idx)
6152                 return -rte_errno;
6153         return 0;
6154 }
6155
6156 /**
6157  * Enable a flow control configured from the control plane.
6158  *
6159  * @param dev
6160  *   Pointer to Ethernet device.
6161  * @param eth_spec
6162  *   An Ethernet flow spec to apply.
6163  * @param eth_mask
6164  *   An Ethernet flow mask to apply.
6165  *
6166  * @return
6167  *   0 on success, a negative errno value otherwise and rte_errno is set.
6168  */
6169 int
6170 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6171                struct rte_flow_item_eth *eth_spec,
6172                struct rte_flow_item_eth *eth_mask)
6173 {
6174         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6175 }
6176
6177 /**
6178  * Create default miss flow rule matching lacp traffic
6179  *
6180  * @param dev
6181  *   Pointer to Ethernet device.
6182  * @param eth_spec
6183  *   An Ethernet flow spec to apply.
6184  *
6185  * @return
6186  *   0 on success, a negative errno value otherwise and rte_errno is set.
6187  */
6188 int
6189 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6190 {
6191         struct mlx5_priv *priv = dev->data->dev_private;
6192         /*
6193          * The LACP matching is done by only using ether type since using
6194          * a multicast dst mac causes kernel to give low priority to this flow.
6195          */
6196         static const struct rte_flow_item_eth lacp_spec = {
6197                 .type = RTE_BE16(0x8809),
6198         };
6199         static const struct rte_flow_item_eth lacp_mask = {
6200                 .type = 0xffff,
6201         };
6202         const struct rte_flow_attr attr = {
6203                 .ingress = 1,
6204         };
6205         struct rte_flow_item items[] = {
6206                 {
6207                         .type = RTE_FLOW_ITEM_TYPE_ETH,
6208                         .spec = &lacp_spec,
6209                         .mask = &lacp_mask,
6210                 },
6211                 {
6212                         .type = RTE_FLOW_ITEM_TYPE_END,
6213                 },
6214         };
6215         struct rte_flow_action actions[] = {
6216                 {
6217                         .type = (enum rte_flow_action_type)
6218                                 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6219                 },
6220                 {
6221                         .type = RTE_FLOW_ACTION_TYPE_END,
6222                 },
6223         };
6224         struct rte_flow_error error;
6225         uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6226                                 &attr, items, actions, false, &error);
6227
6228         if (!flow_idx)
6229                 return -rte_errno;
6230         return 0;
6231 }
6232
6233 /**
6234  * Destroy a flow.
6235  *
6236  * @see rte_flow_destroy()
6237  * @see rte_flow_ops
6238  */
6239 int
6240 mlx5_flow_destroy(struct rte_eth_dev *dev,
6241                   struct rte_flow *flow,
6242                   struct rte_flow_error *error __rte_unused)
6243 {
6244         struct mlx5_priv *priv = dev->data->dev_private;
6245
6246         flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6247         return 0;
6248 }
6249
6250 /**
6251  * Destroy all flows.
6252  *
6253  * @see rte_flow_flush()
6254  * @see rte_flow_ops
6255  */
6256 int
6257 mlx5_flow_flush(struct rte_eth_dev *dev,
6258                 struct rte_flow_error *error __rte_unused)
6259 {
6260         struct mlx5_priv *priv = dev->data->dev_private;
6261
6262         mlx5_flow_list_flush(dev, &priv->flows, false);
6263         return 0;
6264 }
6265
6266 /**
6267  * Isolated mode.
6268  *
6269  * @see rte_flow_isolate()
6270  * @see rte_flow_ops
6271  */
6272 int
6273 mlx5_flow_isolate(struct rte_eth_dev *dev,
6274                   int enable,
6275                   struct rte_flow_error *error)
6276 {
6277         struct mlx5_priv *priv = dev->data->dev_private;
6278
6279         if (dev->data->dev_started) {
6280                 rte_flow_error_set(error, EBUSY,
6281                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6282                                    NULL,
6283                                    "port must be stopped first");
6284                 return -rte_errno;
6285         }
6286         priv->isolated = !!enable;
6287         if (enable)
6288                 dev->dev_ops = &mlx5_dev_ops_isolate;
6289         else
6290                 dev->dev_ops = &mlx5_dev_ops;
6291
6292         dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6293         dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6294
6295         return 0;
6296 }
6297
6298 /**
6299  * Query a flow.
6300  *
6301  * @see rte_flow_query()
6302  * @see rte_flow_ops
6303  */
6304 static int
6305 flow_drv_query(struct rte_eth_dev *dev,
6306                uint32_t flow_idx,
6307                const struct rte_flow_action *actions,
6308                void *data,
6309                struct rte_flow_error *error)
6310 {
6311         struct mlx5_priv *priv = dev->data->dev_private;
6312         const struct mlx5_flow_driver_ops *fops;
6313         struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6314                                                [MLX5_IPOOL_RTE_FLOW],
6315                                                flow_idx);
6316         enum mlx5_flow_drv_type ftype;
6317
6318         if (!flow) {
6319                 return rte_flow_error_set(error, ENOENT,
6320                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6321                           NULL,
6322                           "invalid flow handle");
6323         }
6324         ftype = flow->drv_type;
6325         MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6326         fops = flow_get_drv_ops(ftype);
6327
6328         return fops->query(dev, flow, actions, data, error);
6329 }
6330
6331 /**
6332  * Query a flow.
6333  *
6334  * @see rte_flow_query()
6335  * @see rte_flow_ops
6336  */
6337 int
6338 mlx5_flow_query(struct rte_eth_dev *dev,
6339                 struct rte_flow *flow,
6340                 const struct rte_flow_action *actions,
6341                 void *data,
6342                 struct rte_flow_error *error)
6343 {
6344         int ret;
6345
6346         ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6347                              error);
6348         if (ret < 0)
6349                 return ret;
6350         return 0;
6351 }
6352
6353 /**
6354  * Manage filter operations.
6355  *
6356  * @param dev
6357  *   Pointer to Ethernet device structure.
6358  * @param filter_type
6359  *   Filter type.
6360  * @param filter_op
6361  *   Operation to perform.
6362  * @param arg
6363  *   Pointer to operation-specific structure.
6364  *
6365  * @return
6366  *   0 on success, a negative errno value otherwise and rte_errno is set.
6367  */
6368 int
6369 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6370                      enum rte_filter_type filter_type,
6371                      enum rte_filter_op filter_op,
6372                      void *arg)
6373 {
6374         switch (filter_type) {
6375         case RTE_ETH_FILTER_GENERIC:
6376                 if (filter_op != RTE_ETH_FILTER_GET) {
6377                         rte_errno = EINVAL;
6378                         return -rte_errno;
6379                 }
6380                 *(const void **)arg = &mlx5_flow_ops;
6381                 return 0;
6382         default:
6383                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
6384                         dev->data->port_id, filter_type);
6385                 rte_errno = ENOTSUP;
6386                 return -rte_errno;
6387         }
6388         return 0;
6389 }
6390
6391 /**
6392  * Create the needed meter and suffix tables.
6393  *
6394  * @param[in] dev
6395  *   Pointer to Ethernet device.
6396  * @param[in] fm
6397  *   Pointer to the flow meter.
6398  *
6399  * @return
6400  *   Pointer to table set on success, NULL otherwise.
6401  */
6402 struct mlx5_meter_domains_infos *
6403 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6404                           const struct mlx5_flow_meter *fm)
6405 {
6406         const struct mlx5_flow_driver_ops *fops;
6407
6408         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6409         return fops->create_mtr_tbls(dev, fm);
6410 }
6411
6412 /**
6413  * Destroy the meter table set.
6414  *
6415  * @param[in] dev
6416  *   Pointer to Ethernet device.
6417  * @param[in] tbl
6418  *   Pointer to the meter table set.
6419  *
6420  * @return
6421  *   0 on success.
6422  */
6423 int
6424 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6425                            struct mlx5_meter_domains_infos *tbls)
6426 {
6427         const struct mlx5_flow_driver_ops *fops;
6428
6429         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6430         return fops->destroy_mtr_tbls(dev, tbls);
6431 }
6432
6433 /**
6434  * Create policer rules.
6435  *
6436  * @param[in] dev
6437  *   Pointer to Ethernet device.
6438  * @param[in] fm
6439  *   Pointer to flow meter structure.
6440  * @param[in] attr
6441  *   Pointer to flow attributes.
6442  *
6443  * @return
6444  *   0 on success, -1 otherwise.
6445  */
6446 int
6447 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6448                                struct mlx5_flow_meter *fm,
6449                                const struct rte_flow_attr *attr)
6450 {
6451         const struct mlx5_flow_driver_ops *fops;
6452
6453         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6454         return fops->create_policer_rules(dev, fm, attr);
6455 }
6456
6457 /**
6458  * Destroy policer rules.
6459  *
6460  * @param[in] fm
6461  *   Pointer to flow meter structure.
6462  * @param[in] attr
6463  *   Pointer to flow attributes.
6464  *
6465  * @return
6466  *   0 on success, -1 otherwise.
6467  */
6468 int
6469 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6470                                 struct mlx5_flow_meter *fm,
6471                                 const struct rte_flow_attr *attr)
6472 {
6473         const struct mlx5_flow_driver_ops *fops;
6474
6475         fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6476         return fops->destroy_policer_rules(dev, fm, attr);
6477 }
6478
6479 /**
6480  * Allocate a counter.
6481  *
6482  * @param[in] dev
6483  *   Pointer to Ethernet device structure.
6484  *
6485  * @return
6486  *   Index to allocated counter  on success, 0 otherwise.
6487  */
6488 uint32_t
6489 mlx5_counter_alloc(struct rte_eth_dev *dev)
6490 {
6491         const struct mlx5_flow_driver_ops *fops;
6492         struct rte_flow_attr attr = { .transfer = 0 };
6493
6494         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6495                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6496                 return fops->counter_alloc(dev);
6497         }
6498         DRV_LOG(ERR,
6499                 "port %u counter allocate is not supported.",
6500                  dev->data->port_id);
6501         return 0;
6502 }
6503
6504 /**
6505  * Free a counter.
6506  *
6507  * @param[in] dev
6508  *   Pointer to Ethernet device structure.
6509  * @param[in] cnt
6510  *   Index to counter to be free.
6511  */
6512 void
6513 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6514 {
6515         const struct mlx5_flow_driver_ops *fops;
6516         struct rte_flow_attr attr = { .transfer = 0 };
6517
6518         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6519                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6520                 fops->counter_free(dev, cnt);
6521                 return;
6522         }
6523         DRV_LOG(ERR,
6524                 "port %u counter free is not supported.",
6525                  dev->data->port_id);
6526 }
6527
6528 /**
6529  * Query counter statistics.
6530  *
6531  * @param[in] dev
6532  *   Pointer to Ethernet device structure.
6533  * @param[in] cnt
6534  *   Index to counter to query.
6535  * @param[in] clear
6536  *   Set to clear counter statistics.
6537  * @param[out] pkts
6538  *   The counter hits packets number to save.
6539  * @param[out] bytes
6540  *   The counter hits bytes number to save.
6541  *
6542  * @return
6543  *   0 on success, a negative errno value otherwise.
6544  */
6545 int
6546 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6547                    bool clear, uint64_t *pkts, uint64_t *bytes)
6548 {
6549         const struct mlx5_flow_driver_ops *fops;
6550         struct rte_flow_attr attr = { .transfer = 0 };
6551
6552         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6553                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6554                 return fops->counter_query(dev, cnt, clear, pkts, bytes);
6555         }
6556         DRV_LOG(ERR,
6557                 "port %u counter query is not supported.",
6558                  dev->data->port_id);
6559         return -ENOTSUP;
6560 }
6561
6562 /**
6563  * Allocate a new memory for the counter values wrapped by all the needed
6564  * management.
6565  *
6566  * @param[in] sh
6567  *   Pointer to mlx5_dev_ctx_shared object.
6568  *
6569  * @return
6570  *   0 on success, a negative errno value otherwise.
6571  */
6572 static int
6573 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6574 {
6575         struct mlx5_devx_mkey_attr mkey_attr;
6576         struct mlx5_counter_stats_mem_mng *mem_mng;
6577         volatile struct flow_counter_stats *raw_data;
6578         int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6579         int size = (sizeof(struct flow_counter_stats) *
6580                         MLX5_COUNTERS_PER_POOL +
6581                         sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6582                         sizeof(struct mlx5_counter_stats_mem_mng);
6583         size_t pgsize = rte_mem_page_size();
6584         uint8_t *mem;
6585         int i;
6586
6587         if (pgsize == (size_t)-1) {
6588                 DRV_LOG(ERR, "Failed to get mem page size");
6589                 rte_errno = ENOMEM;
6590                 return -ENOMEM;
6591         }
6592         mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6593         if (!mem) {
6594                 rte_errno = ENOMEM;
6595                 return -ENOMEM;
6596         }
6597         mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6598         size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6599         mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
6600                                                  IBV_ACCESS_LOCAL_WRITE);
6601         if (!mem_mng->umem) {
6602                 rte_errno = errno;
6603                 mlx5_free(mem);
6604                 return -rte_errno;
6605         }
6606         mkey_attr.addr = (uintptr_t)mem;
6607         mkey_attr.size = size;
6608         mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6609         mkey_attr.pd = sh->pdn;
6610         mkey_attr.log_entity_size = 0;
6611         mkey_attr.pg_access = 0;
6612         mkey_attr.klm_array = NULL;
6613         mkey_attr.klm_num = 0;
6614         mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6615         mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6616         mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6617         if (!mem_mng->dm) {
6618                 mlx5_os_umem_dereg(mem_mng->umem);
6619                 rte_errno = errno;
6620                 mlx5_free(mem);
6621                 return -rte_errno;
6622         }
6623         mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6624         raw_data = (volatile struct flow_counter_stats *)mem;
6625         for (i = 0; i < raws_n; ++i) {
6626                 mem_mng->raws[i].mem_mng = mem_mng;
6627                 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6628         }
6629         for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6630                 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6631                                  mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6632                                  next);
6633         LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6634         sh->cmng.mem_mng = mem_mng;
6635         return 0;
6636 }
6637
6638 /**
6639  * Set the statistic memory to the new counter pool.
6640  *
6641  * @param[in] sh
6642  *   Pointer to mlx5_dev_ctx_shared object.
6643  * @param[in] pool
6644  *   Pointer to the pool to set the statistic memory.
6645  *
6646  * @return
6647  *   0 on success, a negative errno value otherwise.
6648  */
6649 static int
6650 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6651                                struct mlx5_flow_counter_pool *pool)
6652 {
6653         struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6654         /* Resize statistic memory once used out. */
6655         if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6656             mlx5_flow_create_counter_stat_mem_mng(sh)) {
6657                 DRV_LOG(ERR, "Cannot resize counter stat mem.");
6658                 return -1;
6659         }
6660         rte_spinlock_lock(&pool->sl);
6661         pool->raw = cmng->mem_mng->raws + pool->index %
6662                     MLX5_CNT_CONTAINER_RESIZE;
6663         rte_spinlock_unlock(&pool->sl);
6664         pool->raw_hw = NULL;
6665         return 0;
6666 }
6667
6668 #define MLX5_POOL_QUERY_FREQ_US 1000000
6669
6670 /**
6671  * Set the periodic procedure for triggering asynchronous batch queries for all
6672  * the counter pools.
6673  *
6674  * @param[in] sh
6675  *   Pointer to mlx5_dev_ctx_shared object.
6676  */
6677 void
6678 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6679 {
6680         uint32_t pools_n, us;
6681
6682         pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6683         us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6684         DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6685         if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6686                 sh->cmng.query_thread_on = 0;
6687                 DRV_LOG(ERR, "Cannot reinitialize query alarm");
6688         } else {
6689                 sh->cmng.query_thread_on = 1;
6690         }
6691 }
6692
6693 /**
6694  * The periodic procedure for triggering asynchronous batch queries for all the
6695  * counter pools. This function is probably called by the host thread.
6696  *
6697  * @param[in] arg
6698  *   The parameter for the alarm process.
6699  */
6700 void
6701 mlx5_flow_query_alarm(void *arg)
6702 {
6703         struct mlx5_dev_ctx_shared *sh = arg;
6704         int ret;
6705         uint16_t pool_index = sh->cmng.pool_index;
6706         struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6707         struct mlx5_flow_counter_pool *pool;
6708         uint16_t n_valid;
6709
6710         if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6711                 goto set_alarm;
6712         rte_spinlock_lock(&cmng->pool_update_sl);
6713         pool = cmng->pools[pool_index];
6714         n_valid = cmng->n_valid;
6715         rte_spinlock_unlock(&cmng->pool_update_sl);
6716         /* Set the statistic memory to the new created pool. */
6717         if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6718                 goto set_alarm;
6719         if (pool->raw_hw)
6720                 /* There is a pool query in progress. */
6721                 goto set_alarm;
6722         pool->raw_hw =
6723                 LIST_FIRST(&sh->cmng.free_stat_raws);
6724         if (!pool->raw_hw)
6725                 /* No free counter statistics raw memory. */
6726                 goto set_alarm;
6727         /*
6728          * Identify the counters released between query trigger and query
6729          * handle more efficiently. The counter released in this gap period
6730          * should wait for a new round of query as the new arrived packets
6731          * will not be taken into account.
6732          */
6733         pool->query_gen++;
6734         ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6735                                                MLX5_COUNTERS_PER_POOL,
6736                                                NULL, NULL,
6737                                                pool->raw_hw->mem_mng->dm->id,
6738                                                (void *)(uintptr_t)
6739                                                pool->raw_hw->data,
6740                                                sh->devx_comp,
6741                                                (uint64_t)(uintptr_t)pool);
6742         if (ret) {
6743                 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6744                         " %d", pool->min_dcs->id);
6745                 pool->raw_hw = NULL;
6746                 goto set_alarm;
6747         }
6748         LIST_REMOVE(pool->raw_hw, next);
6749         sh->cmng.pending_queries++;
6750         pool_index++;
6751         if (pool_index >= n_valid)
6752                 pool_index = 0;
6753 set_alarm:
6754         sh->cmng.pool_index = pool_index;
6755         mlx5_set_query_alarm(sh);
6756 }
6757
6758 /**
6759  * Check and callback event for new aged flow in the counter pool
6760  *
6761  * @param[in] sh
6762  *   Pointer to mlx5_dev_ctx_shared object.
6763  * @param[in] pool
6764  *   Pointer to Current counter pool.
6765  */
6766 static void
6767 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6768                    struct mlx5_flow_counter_pool *pool)
6769 {
6770         struct mlx5_priv *priv;
6771         struct mlx5_flow_counter *cnt;
6772         struct mlx5_age_info *age_info;
6773         struct mlx5_age_param *age_param;
6774         struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6775         struct mlx5_counter_stats_raw *prev = pool->raw;
6776         const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6777         const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6778         uint16_t expected = AGE_CANDIDATE;
6779         uint32_t i;
6780
6781         pool->time_of_last_age_check = curr_time;
6782         for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6783                 cnt = MLX5_POOL_GET_CNT(pool, i);
6784                 age_param = MLX5_CNT_TO_AGE(cnt);
6785                 if (__atomic_load_n(&age_param->state,
6786                                     __ATOMIC_RELAXED) != AGE_CANDIDATE)
6787                         continue;
6788                 if (cur->data[i].hits != prev->data[i].hits) {
6789                         __atomic_store_n(&age_param->sec_since_last_hit, 0,
6790                                          __ATOMIC_RELAXED);
6791                         continue;
6792                 }
6793                 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6794                                        time_delta,
6795                                        __ATOMIC_RELAXED) <= age_param->timeout)
6796                         continue;
6797                 /**
6798                  * Hold the lock first, or if between the
6799                  * state AGE_TMOUT and tailq operation the
6800                  * release happened, the release procedure
6801                  * may delete a non-existent tailq node.
6802                  */
6803                 priv = rte_eth_devices[age_param->port_id].data->dev_private;
6804                 age_info = GET_PORT_AGE_INFO(priv);
6805                 rte_spinlock_lock(&age_info->aged_sl);
6806                 if (__atomic_compare_exchange_n(&age_param->state, &expected,
6807                                                 AGE_TMOUT, false,
6808                                                 __ATOMIC_RELAXED,
6809                                                 __ATOMIC_RELAXED)) {
6810                         TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6811                         MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6812                 }
6813                 rte_spinlock_unlock(&age_info->aged_sl);
6814         }
6815         mlx5_age_event_prepare(sh);
6816 }
6817
6818 /**
6819  * Handler for the HW respond about ready values from an asynchronous batch
6820  * query. This function is probably called by the host thread.
6821  *
6822  * @param[in] sh
6823  *   The pointer to the shared device context.
6824  * @param[in] async_id
6825  *   The Devx async ID.
6826  * @param[in] status
6827  *   The status of the completion.
6828  */
6829 void
6830 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6831                                   uint64_t async_id, int status)
6832 {
6833         struct mlx5_flow_counter_pool *pool =
6834                 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6835         struct mlx5_counter_stats_raw *raw_to_free;
6836         uint8_t query_gen = pool->query_gen ^ 1;
6837         struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6838         enum mlx5_counter_type cnt_type =
6839                 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6840                                 MLX5_COUNTER_TYPE_ORIGIN;
6841
6842         if (unlikely(status)) {
6843                 raw_to_free = pool->raw_hw;
6844         } else {
6845                 raw_to_free = pool->raw;
6846                 if (pool->is_aged)
6847                         mlx5_flow_aging_check(sh, pool);
6848                 rte_spinlock_lock(&pool->sl);
6849                 pool->raw = pool->raw_hw;
6850                 rte_spinlock_unlock(&pool->sl);
6851                 /* Be sure the new raw counters data is updated in memory. */
6852                 rte_io_wmb();
6853                 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6854                         rte_spinlock_lock(&cmng->csl[cnt_type]);
6855                         TAILQ_CONCAT(&cmng->counters[cnt_type],
6856                                      &pool->counters[query_gen], next);
6857                         rte_spinlock_unlock(&cmng->csl[cnt_type]);
6858                 }
6859         }
6860         LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6861         pool->raw_hw = NULL;
6862         sh->cmng.pending_queries--;
6863 }
6864
6865 static int
6866 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
6867                     const struct flow_grp_info *grp_info,
6868                     struct rte_flow_error *error)
6869 {
6870         if (grp_info->transfer && grp_info->external &&
6871             grp_info->fdb_def_rule) {
6872                 if (group == UINT32_MAX)
6873                         return rte_flow_error_set
6874                                                 (error, EINVAL,
6875                                                  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6876                                                  NULL,
6877                                                  "group index not supported");
6878                 *table = group + 1;
6879         } else {
6880                 *table = group;
6881         }
6882         DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
6883         return 0;
6884 }
6885
6886 /**
6887  * Translate the rte_flow group index to HW table value.
6888  *
6889  * If tunnel offload is disabled, all group ids converted to flow table
6890  * id using the standard method.
6891  * If tunnel offload is enabled, group id can be converted using the
6892  * standard or tunnel conversion method. Group conversion method
6893  * selection depends on flags in `grp_info` parameter:
6894  * - Internal (grp_info.external == 0) groups conversion uses the
6895  *   standard method.
6896  * - Group ids in JUMP action converted with the tunnel conversion.
6897  * - Group id in rule attribute conversion depends on a rule type and
6898  *   group id value:
6899  *   ** non zero group attributes converted with the tunnel method
6900  *   ** zero group attribute in non-tunnel rule is converted using the
6901  *      standard method - there's only one root table
6902  *   ** zero group attribute in steer tunnel rule is converted with the
6903  *      standard method - single root table
6904  *   ** zero group attribute in match tunnel rule is a special OvS
6905  *      case: that value is used for portability reasons. That group
6906  *      id is converted with the tunnel conversion method.
6907  *
6908  * @param[in] dev
6909  *   Port device
6910  * @param[in] tunnel
6911  *   PMD tunnel offload object
6912  * @param[in] group
6913  *   rte_flow group index value.
6914  * @param[out] table
6915  *   HW table value.
6916  * @param[in] grp_info
6917  *   flags used for conversion
6918  * @param[out] error
6919  *   Pointer to error structure.
6920  *
6921  * @return
6922  *   0 on success, a negative errno value otherwise and rte_errno is set.
6923  */
6924 int
6925 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
6926                          const struct mlx5_flow_tunnel *tunnel,
6927                          uint32_t group, uint32_t *table,
6928                          const struct flow_grp_info *grp_info,
6929                          struct rte_flow_error *error)
6930 {
6931         int ret;
6932         bool standard_translation;
6933
6934         if (!grp_info->skip_scale && grp_info->external &&
6935             group < MLX5_MAX_TABLES_EXTERNAL)
6936                 group *= MLX5_FLOW_TABLE_FACTOR;
6937         if (is_tunnel_offload_active(dev)) {
6938                 standard_translation = !grp_info->external ||
6939                                         grp_info->std_tbl_fix;
6940         } else {
6941                 standard_translation = true;
6942         }
6943         DRV_LOG(DEBUG,
6944                 "port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
6945                 dev->data->port_id, group, grp_info->transfer,
6946                 grp_info->external, grp_info->fdb_def_rule,
6947                 standard_translation ? "STANDARD" : "TUNNEL");
6948         if (standard_translation)
6949                 ret = flow_group_to_table(dev->data->port_id, group, table,
6950                                           grp_info, error);
6951         else
6952                 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
6953                                                       table, error);
6954
6955         return ret;
6956 }
6957
6958 /**
6959  * Discover availability of metadata reg_c's.
6960  *
6961  * Iteratively use test flows to check availability.
6962  *
6963  * @param[in] dev
6964  *   Pointer to the Ethernet device structure.
6965  *
6966  * @return
6967  *   0 on success, a negative errno value otherwise and rte_errno is set.
6968  */
6969 int
6970 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6971 {
6972         struct mlx5_priv *priv = dev->data->dev_private;
6973         struct mlx5_dev_config *config = &priv->config;
6974         enum modify_reg idx;
6975         int n = 0;
6976
6977         /* reg_c[0] and reg_c[1] are reserved. */
6978         config->flow_mreg_c[n++] = REG_C_0;
6979         config->flow_mreg_c[n++] = REG_C_1;
6980         /* Discover availability of other reg_c's. */
6981         for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6982                 struct rte_flow_attr attr = {
6983                         .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6984                         .priority = MLX5_FLOW_PRIO_RSVD,
6985                         .ingress = 1,
6986                 };
6987                 struct rte_flow_item items[] = {
6988                         [0] = {
6989                                 .type = RTE_FLOW_ITEM_TYPE_END,
6990                         },
6991                 };
6992                 struct rte_flow_action actions[] = {
6993                         [0] = {
6994                                 .type = (enum rte_flow_action_type)
6995                                         MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6996                                 .conf = &(struct mlx5_flow_action_copy_mreg){
6997                                         .src = REG_C_1,
6998                                         .dst = idx,
6999                                 },
7000                         },
7001                         [1] = {
7002                                 .type = RTE_FLOW_ACTION_TYPE_JUMP,
7003                                 .conf = &(struct rte_flow_action_jump){
7004                                         .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7005                                 },
7006                         },
7007                         [2] = {
7008                                 .type = RTE_FLOW_ACTION_TYPE_END,
7009                         },
7010                 };
7011                 uint32_t flow_idx;
7012                 struct rte_flow *flow;
7013                 struct rte_flow_error error;
7014
7015                 if (!config->dv_flow_en)
7016                         break;
7017                 /* Create internal flow, validation skips copy action. */
7018                 flow_idx = flow_list_create(dev, NULL, &attr, items,
7019                                             actions, false, &error);
7020                 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7021                                       flow_idx);
7022                 if (!flow)
7023                         continue;
7024                 config->flow_mreg_c[n++] = idx;
7025                 flow_list_destroy(dev, NULL, flow_idx);
7026         }
7027         for (; n < MLX5_MREG_C_NUM; ++n)
7028                 config->flow_mreg_c[n] = REG_NON;
7029         return 0;
7030 }
7031
7032 /**
7033  * Dump flow raw hw data to file
7034  *
7035  * @param[in] dev
7036  *    The pointer to Ethernet device.
7037  * @param[in] file
7038  *   A pointer to a file for output.
7039  * @param[out] error
7040  *   Perform verbose error reporting if not NULL. PMDs initialize this
7041  *   structure in case of error only.
7042  * @return
7043  *   0 on success, a nagative value otherwise.
7044  */
7045 int
7046 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7047                    FILE *file,
7048                    struct rte_flow_error *error __rte_unused)
7049 {
7050         struct mlx5_priv *priv = dev->data->dev_private;
7051         struct mlx5_dev_ctx_shared *sh = priv->sh;
7052
7053         if (!priv->config.dv_flow_en) {
7054                 if (fputs("device dv flow disabled\n", file) <= 0)
7055                         return -errno;
7056                 return -ENOTSUP;
7057         }
7058         return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7059                                        sh->tx_domain, file);
7060 }
7061
7062 /**
7063  * Get aged-out flows.
7064  *
7065  * @param[in] dev
7066  *   Pointer to the Ethernet device structure.
7067  * @param[in] context
7068  *   The address of an array of pointers to the aged-out flows contexts.
7069  * @param[in] nb_countexts
7070  *   The length of context array pointers.
7071  * @param[out] error
7072  *   Perform verbose error reporting if not NULL. Initialized in case of
7073  *   error only.
7074  *
7075  * @return
7076  *   how many contexts get in success, otherwise negative errno value.
7077  *   if nb_contexts is 0, return the amount of all aged contexts.
7078  *   if nb_contexts is not 0 , return the amount of aged flows reported
7079  *   in the context array.
7080  */
7081 int
7082 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7083                         uint32_t nb_contexts, struct rte_flow_error *error)
7084 {
7085         const struct mlx5_flow_driver_ops *fops;
7086         struct rte_flow_attr attr = { .transfer = 0 };
7087
7088         if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7089                 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7090                 return fops->get_aged_flows(dev, contexts, nb_contexts,
7091                                                     error);
7092         }
7093         DRV_LOG(ERR,
7094                 "port %u get aged flows is not supported.",
7095                  dev->data->port_id);
7096         return -ENOTSUP;
7097 }
7098
7099 /* Wrapper for driver action_validate op callback */
7100 static int
7101 flow_drv_action_validate(struct rte_eth_dev *dev,
7102                          const struct rte_flow_shared_action_conf *conf,
7103                          const struct rte_flow_action *action,
7104                          const struct mlx5_flow_driver_ops *fops,
7105                          struct rte_flow_error *error)
7106 {
7107         static const char err_msg[] = "shared action validation unsupported";
7108
7109         if (!fops->action_validate) {
7110                 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7111                 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7112                                    NULL, err_msg);
7113                 return -rte_errno;
7114         }
7115         return fops->action_validate(dev, conf, action, error);
7116 }
7117
7118 /**
7119  * Destroys the shared action by handle.
7120  *
7121  * @param dev
7122  *   Pointer to Ethernet device structure.
7123  * @param[in] action
7124  *   Handle for the shared action to be destroyed.
7125  * @param[out] error
7126  *   Perform verbose error reporting if not NULL. PMDs initialize this
7127  *   structure in case of error only.
7128  *
7129  * @return
7130  *   0 on success, a negative errno value otherwise and rte_errno is set.
7131  *
7132  * @note: wrapper for driver action_create op callback.
7133  */
7134 static int
7135 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7136                            struct rte_flow_shared_action *action,
7137                            struct rte_flow_error *error)
7138 {
7139         static const char err_msg[] = "shared action destruction unsupported";
7140         struct rte_flow_attr attr = { .transfer = 0 };
7141         const struct mlx5_flow_driver_ops *fops =
7142                         flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7143
7144         if (!fops->action_destroy) {
7145                 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7146                 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7147                                    NULL, err_msg);
7148                 return -rte_errno;
7149         }
7150         return fops->action_destroy(dev, action, error);
7151 }
7152
7153 /* Wrapper for driver action_destroy op callback */
7154 static int
7155 flow_drv_action_update(struct rte_eth_dev *dev,
7156                        struct rte_flow_shared_action *action,
7157                        const void *action_conf,
7158                        const struct mlx5_flow_driver_ops *fops,
7159                        struct rte_flow_error *error)
7160 {
7161         static const char err_msg[] = "shared action update unsupported";
7162
7163         if (!fops->action_update) {
7164                 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7165                 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7166                                    NULL, err_msg);
7167                 return -rte_errno;
7168         }
7169         return fops->action_update(dev, action, action_conf, error);
7170 }
7171
7172 /* Wrapper for driver action_destroy op callback */
7173 static int
7174 flow_drv_action_query(struct rte_eth_dev *dev,
7175                       const struct rte_flow_shared_action *action,
7176                       void *data,
7177                       const struct mlx5_flow_driver_ops *fops,
7178                       struct rte_flow_error *error)
7179 {
7180         static const char err_msg[] = "shared action query unsupported";
7181
7182         if (!fops->action_query) {
7183                 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7184                 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7185                                    NULL, err_msg);
7186                 return -rte_errno;
7187         }
7188         return fops->action_query(dev, action, data, error);
7189 }
7190
7191 /**
7192  * Create shared action for reuse in multiple flow rules.
7193  *
7194  * @param dev
7195  *   Pointer to Ethernet device structure.
7196  * @param[in] action
7197  *   Action configuration for shared action creation.
7198  * @param[out] error
7199  *   Perform verbose error reporting if not NULL. PMDs initialize this
7200  *   structure in case of error only.
7201  * @return
7202  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7203  */
7204 static struct rte_flow_shared_action *
7205 mlx5_shared_action_create(struct rte_eth_dev *dev,
7206                           const struct rte_flow_shared_action_conf *conf,
7207                           const struct rte_flow_action *action,
7208                           struct rte_flow_error *error)
7209 {
7210         static const char err_msg[] = "shared action creation unsupported";
7211         struct rte_flow_attr attr = { .transfer = 0 };
7212         const struct mlx5_flow_driver_ops *fops =
7213                         flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7214
7215         if (flow_drv_action_validate(dev, conf, action, fops, error))
7216                 return NULL;
7217         if (!fops->action_create) {
7218                 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7219                 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7220                                    NULL, err_msg);
7221                 return NULL;
7222         }
7223         return fops->action_create(dev, conf, action, error);
7224 }
7225
7226 /**
7227  * Updates inplace the shared action configuration pointed by *action* handle
7228  * with the configuration provided as *action* argument.
7229  * The update of the shared action configuration effects all flow rules reusing
7230  * the action via handle.
7231  *
7232  * @param dev
7233  *   Pointer to Ethernet device structure.
7234  * @param[in] shared_action
7235  *   Handle for the shared action to be updated.
7236  * @param[in] action
7237  *   Action specification used to modify the action pointed by handle.
7238  *   *action* should be of same type with the action pointed by the *action*
7239  *   handle argument, otherwise considered as invalid.
7240  * @param[out] error
7241  *   Perform verbose error reporting if not NULL. PMDs initialize this
7242  *   structure in case of error only.
7243  *
7244  * @return
7245  *   0 on success, a negative errno value otherwise and rte_errno is set.
7246  */
7247 static int
7248 mlx5_shared_action_update(struct rte_eth_dev *dev,
7249                 struct rte_flow_shared_action *shared_action,
7250                 const struct rte_flow_action *action,
7251                 struct rte_flow_error *error)
7252 {
7253         struct rte_flow_attr attr = { .transfer = 0 };
7254         const struct mlx5_flow_driver_ops *fops =
7255                         flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7256         int ret;
7257
7258         ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7259         if (ret)
7260                 return ret;
7261         return flow_drv_action_update(dev, shared_action, action->conf, fops,
7262                                       error);
7263 }
7264
7265 /**
7266  * Query the shared action by handle.
7267  *
7268  * This function allows retrieving action-specific data such as counters.
7269  * Data is gathered by special action which may be present/referenced in
7270  * more than one flow rule definition.
7271  *
7272  * \see RTE_FLOW_ACTION_TYPE_COUNT
7273  *
7274  * @param dev
7275  *   Pointer to Ethernet device structure.
7276  * @param[in] action
7277  *   Handle for the shared action to query.
7278  * @param[in, out] data
7279  *   Pointer to storage for the associated query data type.
7280  * @param[out] error
7281  *   Perform verbose error reporting if not NULL. PMDs initialize this
7282  *   structure in case of error only.
7283  *
7284  * @return
7285  *   0 on success, a negative errno value otherwise and rte_errno is set.
7286  */
7287 static int
7288 mlx5_shared_action_query(struct rte_eth_dev *dev,
7289                          const struct rte_flow_shared_action *action,
7290                          void *data,
7291                          struct rte_flow_error *error)
7292 {
7293         struct rte_flow_attr attr = { .transfer = 0 };
7294         const struct mlx5_flow_driver_ops *fops =
7295                         flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7296
7297         return flow_drv_action_query(dev, action, data, fops, error);
7298 }
7299
7300 /**
7301  * Destroy all shared actions.
7302  *
7303  * @param dev
7304  *   Pointer to Ethernet device.
7305  *
7306  * @return
7307  *   0 on success, a negative errno value otherwise and rte_errno is set.
7308  */
7309 int
7310 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7311 {
7312         struct rte_flow_error error;
7313         struct mlx5_priv *priv = dev->data->dev_private;
7314         struct mlx5_shared_action_rss *action;
7315         int ret = 0;
7316         uint32_t idx;
7317
7318         ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7319                       priv->rss_shared_actions, idx, action, next) {
7320                 ret |= mlx5_shared_action_destroy(dev,
7321                        (struct rte_flow_shared_action *)(uintptr_t)idx, &error);
7322         }
7323         return ret;
7324 }
7325
7326 #ifndef HAVE_MLX5DV_DR
7327 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7328 #else
7329 #define MLX5_DOMAIN_SYNC_FLOW \
7330         (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7331 #endif
7332
7333 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7334 {
7335         struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7336         const struct mlx5_flow_driver_ops *fops;
7337         int ret;
7338         struct rte_flow_attr attr = { .transfer = 0 };
7339
7340         fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7341         ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7342         if (ret > 0)
7343                 ret = -ret;
7344         return ret;
7345 }
7346
7347 /**
7348  * tunnel offload functionalilty is defined for DV environment only
7349  */
7350 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7351 __extension__
7352 union tunnel_offload_mark {
7353         uint32_t val;
7354         struct {
7355                 uint32_t app_reserve:8;
7356                 uint32_t table_id:15;
7357                 uint32_t transfer:1;
7358                 uint32_t _unused_:8;
7359         };
7360 };
7361
7362 static bool
7363 mlx5_access_tunnel_offload_db
7364         (struct rte_eth_dev *dev,
7365          bool (*match)(struct rte_eth_dev *,
7366                        struct mlx5_flow_tunnel *, const void *),
7367          void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7368          void (*miss)(struct rte_eth_dev *, void *),
7369          void *ctx, bool lock_op);
7370
7371 static int
7372 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
7373                              struct rte_flow *flow,
7374                              const struct rte_flow_attr *attr,
7375                              const struct rte_flow_action *app_actions,
7376                              uint32_t flow_idx,
7377                              struct tunnel_default_miss_ctx *ctx,
7378                              struct rte_flow_error *error)
7379 {
7380         struct mlx5_priv *priv = dev->data->dev_private;
7381         struct mlx5_flow *dev_flow;
7382         struct rte_flow_attr miss_attr = *attr;
7383         const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
7384         const struct rte_flow_item miss_items[2] = {
7385                 {
7386                         .type = RTE_FLOW_ITEM_TYPE_ETH,
7387                         .spec = NULL,
7388                         .last = NULL,
7389                         .mask = NULL
7390                 },
7391                 {
7392                         .type = RTE_FLOW_ITEM_TYPE_END,
7393                         .spec = NULL,
7394                         .last = NULL,
7395                         .mask = NULL
7396                 }
7397         };
7398         union tunnel_offload_mark mark_id;
7399         struct rte_flow_action_mark miss_mark;
7400         struct rte_flow_action miss_actions[3] = {
7401                 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
7402                 [2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
7403         };
7404         const struct rte_flow_action_jump *jump_data;
7405         uint32_t i, flow_table = 0; /* prevent compilation warning */
7406         struct flow_grp_info grp_info = {
7407                 .external = 1,
7408                 .transfer = attr->transfer,
7409                 .fdb_def_rule = !!priv->fdb_def_rule,
7410                 .std_tbl_fix = 0,
7411         };
7412         int ret;
7413
7414         if (!attr->transfer) {
7415                 uint32_t q_size;
7416
7417                 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
7418                 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
7419                 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
7420                                          0, SOCKET_ID_ANY);
7421                 if (!ctx->queue)
7422                         return rte_flow_error_set
7423                                 (error, ENOMEM,
7424                                 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7425                                 NULL, "invalid default miss RSS");
7426                 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7427                 ctx->action_rss.level = 0,
7428                 ctx->action_rss.types = priv->rss_conf.rss_hf,
7429                 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
7430                 ctx->action_rss.queue_num = priv->reta_idx_n,
7431                 ctx->action_rss.key = priv->rss_conf.rss_key,
7432                 ctx->action_rss.queue = ctx->queue;
7433                 if (!priv->reta_idx_n || !priv->rxqs_n)
7434                         return rte_flow_error_set
7435                                 (error, EINVAL,
7436                                 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7437                                 NULL, "invalid port configuration");
7438                 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
7439                         ctx->action_rss.types = 0;
7440                 for (i = 0; i != priv->reta_idx_n; ++i)
7441                         ctx->queue[i] = (*priv->reta_idx)[i];
7442         } else {
7443                 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
7444                 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
7445         }
7446         miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
7447         for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
7448         jump_data = app_actions->conf;
7449         miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
7450         miss_attr.group = jump_data->group;
7451         ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
7452                                        &flow_table, &grp_info, error);
7453         if (ret)
7454                 return rte_flow_error_set(error, EINVAL,
7455                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7456                                           NULL, "invalid tunnel id");
7457         mark_id.app_reserve = 0;
7458         mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
7459         mark_id.transfer = !!attr->transfer;
7460         mark_id._unused_ = 0;
7461         miss_mark.id = mark_id.val;
7462         dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
7463                                     miss_items, miss_actions, flow_idx, error);
7464         if (!dev_flow)
7465                 return -rte_errno;
7466         dev_flow->flow = flow;
7467         dev_flow->external = true;
7468         dev_flow->tunnel = tunnel;
7469         /* Subflow object was created, we must include one in the list. */
7470         SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
7471                       dev_flow->handle, next);
7472         DRV_LOG(DEBUG,
7473                 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
7474                 dev->data->port_id, tunnel->app_tunnel.type,
7475                 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
7476         ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
7477                                   miss_actions, error);
7478         if (!ret)
7479                 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
7480                                                   error);
7481
7482         return ret;
7483 }
7484
7485 static const struct mlx5_flow_tbl_data_entry  *
7486 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7487 {
7488         struct mlx5_priv *priv = dev->data->dev_private;
7489         struct mlx5_dev_ctx_shared *sh = priv->sh;
7490         struct mlx5_hlist_entry *he;
7491         union tunnel_offload_mark mbits = { .val = mark };
7492         union mlx5_flow_tbl_key table_key = {
7493                 {
7494                         .table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7495                         .dummy = 0,
7496                         .domain = !!mbits.transfer,
7497                         .direction = 0,
7498                 }
7499         };
7500         he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
7501         return he ?
7502                container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7503 }
7504
7505 static void
7506 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
7507                                    struct mlx5_hlist_entry *entry)
7508 {
7509         struct mlx5_dev_ctx_shared *sh = list->ctx;
7510         struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7511
7512         mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7513                         tunnel_flow_tbl_to_id(tte->flow_table));
7514         mlx5_free(tte);
7515 }
7516
7517 static int
7518 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
7519                                   struct mlx5_hlist_entry *entry,
7520                                   uint64_t key, void *cb_ctx __rte_unused)
7521 {
7522         union tunnel_tbl_key tbl = {
7523                 .val = key,
7524         };
7525         struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7526
7527         return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
7528 }
7529
7530 static struct mlx5_hlist_entry *
7531 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
7532                                    void *ctx __rte_unused)
7533 {
7534         struct mlx5_dev_ctx_shared *sh = list->ctx;
7535         struct tunnel_tbl_entry *tte;
7536         union tunnel_tbl_key tbl = {
7537                 .val = key,
7538         };
7539
7540         tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7541                           sizeof(*tte), 0,
7542                           SOCKET_ID_ANY);
7543         if (!tte)
7544                 goto err;
7545         mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7546                           &tte->flow_table);
7547         if (tte->flow_table >= MLX5_MAX_TABLES) {
7548                 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7549                         tte->flow_table);
7550                 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7551                                 tte->flow_table);
7552                 goto err;
7553         } else if (!tte->flow_table) {
7554                 goto err;
7555         }
7556         tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7557         tte->tunnel_id = tbl.tunnel_id;
7558         tte->group = tbl.group;
7559         return &tte->hash;
7560 err:
7561         if (tte)
7562                 mlx5_free(tte);
7563         return NULL;
7564 }
7565
7566 static uint32_t
7567 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7568                                 const struct mlx5_flow_tunnel *tunnel,
7569                                 uint32_t group, uint32_t *table,
7570                                 struct rte_flow_error *error)
7571 {
7572         struct mlx5_hlist_entry *he;
7573         struct tunnel_tbl_entry *tte;
7574         union tunnel_tbl_key key = {
7575                 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7576                 .group = group
7577         };
7578         struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7579         struct mlx5_hlist *group_hash;
7580
7581         group_hash = tunnel ? tunnel->groups : thub->groups;
7582         he = mlx5_hlist_register(group_hash, key.val, NULL);
7583         if (!he)
7584                 return rte_flow_error_set(error, EINVAL,
7585                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7586                                           NULL,
7587                                           "tunnel group index not supported");
7588         tte = container_of(he, typeof(*tte), hash);
7589         *table = tte->flow_table;
7590         DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7591                 dev->data->port_id, key.tunnel_id, group, *table);
7592         return 0;
7593 }
7594
7595 static void
7596 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7597                       struct mlx5_flow_tunnel *tunnel)
7598 {
7599         struct mlx5_priv *priv = dev->data->dev_private;
7600         struct mlx5_indexed_pool *ipool;
7601
7602         DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7603                 dev->data->port_id, tunnel->tunnel_id);
7604         LIST_REMOVE(tunnel, chain);
7605         mlx5_hlist_destroy(tunnel->groups);
7606         ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7607         mlx5_ipool_free(ipool, tunnel->tunnel_id);
7608 }
7609
7610 static bool
7611 mlx5_access_tunnel_offload_db
7612         (struct rte_eth_dev *dev,
7613          bool (*match)(struct rte_eth_dev *,
7614                        struct mlx5_flow_tunnel *, const void *),
7615          void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7616          void (*miss)(struct rte_eth_dev *, void *),
7617          void *ctx, bool lock_op)
7618 {
7619         bool verdict = false;
7620         struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7621         struct mlx5_flow_tunnel *tunnel;
7622
7623         rte_spinlock_lock(&thub->sl);
7624         LIST_FOREACH(tunnel, &thub->tunnels, chain) {
7625                 verdict = match(dev, tunnel, (const void *)ctx);
7626                 if (verdict)
7627                         break;
7628         }
7629         if (!lock_op)
7630                 rte_spinlock_unlock(&thub->sl);
7631         if (verdict && hit)
7632                 hit(dev, tunnel, ctx);
7633         if (!verdict && miss)
7634                 miss(dev, ctx);
7635         if (lock_op)
7636                 rte_spinlock_unlock(&thub->sl);
7637
7638         return verdict;
7639 }
7640
7641 struct tunnel_db_find_tunnel_id_ctx {
7642         uint32_t tunnel_id;
7643         struct mlx5_flow_tunnel *tunnel;
7644 };
7645
7646 static bool
7647 find_tunnel_id_match(struct rte_eth_dev *dev,
7648                      struct mlx5_flow_tunnel *tunnel, const void *x)
7649 {
7650         const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7651
7652         RTE_SET_USED(dev);
7653         return tunnel->tunnel_id == ctx->tunnel_id;
7654 }
7655
7656 static void
7657 find_tunnel_id_hit(struct rte_eth_dev *dev,
7658                    struct mlx5_flow_tunnel *tunnel, void *x)
7659 {
7660         struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7661         RTE_SET_USED(dev);
7662         ctx->tunnel = tunnel;
7663 }
7664
7665 static struct mlx5_flow_tunnel *
7666 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7667 {
7668         struct tunnel_db_find_tunnel_id_ctx ctx = {
7669                 .tunnel_id = id,
7670         };
7671
7672         mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
7673                                       find_tunnel_id_hit, NULL, &ctx, true);
7674
7675         return ctx.tunnel;
7676 }
7677
7678 static struct mlx5_flow_tunnel *
7679 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7680                           const struct rte_flow_tunnel *app_tunnel)
7681 {
7682         struct mlx5_priv *priv = dev->data->dev_private;
7683         struct mlx5_indexed_pool *ipool;
7684         struct mlx5_flow_tunnel *tunnel;
7685         uint32_t id;
7686
7687         ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7688         tunnel = mlx5_ipool_zmalloc(ipool, &id);
7689         if (!tunnel)
7690                 return NULL;
7691         if (id >= MLX5_MAX_TUNNELS) {
7692                 mlx5_ipool_free(ipool, id);
7693                 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7694                 return NULL;
7695         }
7696         tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7697                                            mlx5_flow_tunnel_grp2tbl_create_cb,
7698                                            mlx5_flow_tunnel_grp2tbl_match_cb,
7699                                            mlx5_flow_tunnel_grp2tbl_remove_cb);
7700         if (!tunnel->groups) {
7701                 mlx5_ipool_free(ipool, id);
7702                 return NULL;
7703         }
7704         tunnel->groups->ctx = priv->sh;
7705         /* initiate new PMD tunnel */
7706         memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7707         tunnel->tunnel_id = id;
7708         tunnel->action.type = (typeof(tunnel->action.type))
7709                               MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7710         tunnel->action.conf = tunnel;
7711         tunnel->item.type = (typeof(tunnel->item.type))
7712                             MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7713         tunnel->item.spec = tunnel;
7714         tunnel->item.last = NULL;
7715         tunnel->item.mask = NULL;
7716
7717         DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7718                 dev->data->port_id, tunnel->tunnel_id);
7719
7720         return tunnel;
7721 }
7722
7723 struct tunnel_db_get_tunnel_ctx {
7724         const struct rte_flow_tunnel *app_tunnel;
7725         struct mlx5_flow_tunnel *tunnel;
7726 };
7727
7728 static bool get_tunnel_match(struct rte_eth_dev *dev,
7729                              struct mlx5_flow_tunnel *tunnel, const void *x)
7730 {
7731         const struct tunnel_db_get_tunnel_ctx *ctx = x;
7732
7733         RTE_SET_USED(dev);
7734         return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
7735                        sizeof(*ctx->app_tunnel));
7736 }
7737
7738 static void get_tunnel_hit(struct rte_eth_dev *dev,
7739                            struct mlx5_flow_tunnel *tunnel, void *x)
7740 {
7741         /* called under tunnel spinlock protection */
7742         struct tunnel_db_get_tunnel_ctx *ctx = x;
7743
7744         RTE_SET_USED(dev);
7745         tunnel->refctn++;
7746         ctx->tunnel = tunnel;
7747 }
7748
7749 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
7750 {
7751         /* called under tunnel spinlock protection */
7752         struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7753         struct tunnel_db_get_tunnel_ctx *ctx = x;
7754
7755         rte_spinlock_unlock(&thub->sl);
7756         ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
7757         ctx->tunnel->refctn = 1;
7758         rte_spinlock_lock(&thub->sl);
7759         if (ctx->tunnel)
7760                 LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
7761 }
7762
7763
7764 static int
7765 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7766                      const struct rte_flow_tunnel *app_tunnel,
7767                      struct mlx5_flow_tunnel **tunnel)
7768 {
7769         struct tunnel_db_get_tunnel_ctx ctx = {
7770                 .app_tunnel = app_tunnel,
7771         };
7772
7773         mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
7774                                       get_tunnel_miss, &ctx, true);
7775         *tunnel = ctx.tunnel;
7776         return ctx.tunnel ? 0 : -ENOMEM;
7777 }
7778
7779 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7780 {
7781         struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7782
7783         if (!thub)
7784                 return;
7785         if (!LIST_EMPTY(&thub->tunnels))
7786                 DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
7787         mlx5_hlist_destroy(thub->groups);
7788         mlx5_free(thub);
7789 }
7790
7791 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7792 {
7793         int err;
7794         struct mlx5_flow_tunnel_hub *thub;
7795
7796         thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7797                            0, SOCKET_ID_ANY);
7798         if (!thub)
7799                 return -ENOMEM;
7800         LIST_INIT(&thub->tunnels);
7801         rte_spinlock_init(&thub->sl);
7802         thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES, 0,
7803                                          0, mlx5_flow_tunnel_grp2tbl_create_cb,
7804                                          mlx5_flow_tunnel_grp2tbl_match_cb,
7805                                          mlx5_flow_tunnel_grp2tbl_remove_cb);
7806         if (!thub->groups) {
7807                 err = -rte_errno;
7808                 goto err;
7809         }
7810         thub->groups->ctx = sh;
7811         sh->tunnel_hub = thub;
7812
7813         return 0;
7814
7815 err:
7816         if (thub->groups)
7817                 mlx5_hlist_destroy(thub->groups);
7818         if (thub)
7819                 mlx5_free(thub);
7820         return err;
7821 }
7822
7823 static inline bool
7824 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
7825                           struct rte_flow_tunnel *tunnel,
7826                           const char *err_msg)
7827 {
7828         err_msg = NULL;
7829         if (!is_tunnel_offload_active(dev)) {
7830                 err_msg = "tunnel offload was not activated";
7831                 goto out;
7832         } else if (!tunnel) {
7833                 err_msg = "no application tunnel";
7834                 goto out;
7835         }
7836
7837         switch (tunnel->type) {
7838         default:
7839                 err_msg = "unsupported tunnel type";
7840                 goto out;
7841         case RTE_FLOW_ITEM_TYPE_VXLAN:
7842                 break;
7843         }
7844
7845 out:
7846         return !err_msg;
7847 }
7848
7849 static int
7850 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
7851                     struct rte_flow_tunnel *app_tunnel,
7852                     struct rte_flow_action **actions,
7853                     uint32_t *num_of_actions,
7854                     struct rte_flow_error *error)
7855 {
7856         int ret;
7857         struct mlx5_flow_tunnel *tunnel;
7858         const char *err_msg = NULL;
7859         bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7860
7861         if (!verdict)
7862                 return rte_flow_error_set(error, EINVAL,
7863                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7864                                           err_msg);
7865         ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7866         if (ret < 0) {
7867                 return rte_flow_error_set(error, ret,
7868                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7869                                           "failed to initialize pmd tunnel");
7870         }
7871         *actions = &tunnel->action;
7872         *num_of_actions = 1;
7873         return 0;
7874 }
7875
7876 static int
7877 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
7878                        struct rte_flow_tunnel *app_tunnel,
7879                        struct rte_flow_item **items,
7880                        uint32_t *num_of_items,
7881                        struct rte_flow_error *error)
7882 {
7883         int ret;
7884         struct mlx5_flow_tunnel *tunnel;
7885         const char *err_msg = NULL;
7886         bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7887
7888         if (!verdict)
7889                 return rte_flow_error_set(error, EINVAL,
7890                                           RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7891                                           err_msg);
7892         ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7893         if (ret < 0) {
7894                 return rte_flow_error_set(error, ret,
7895                                           RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7896                                           "failed to initialize pmd tunnel");
7897         }
7898         *items = &tunnel->item;
7899         *num_of_items = 1;
7900         return 0;
7901 }
7902
7903 struct tunnel_db_element_release_ctx {
7904         struct rte_flow_item *items;
7905         struct rte_flow_action *actions;
7906         uint32_t num_elements;
7907         struct rte_flow_error *error;
7908         int ret;
7909 };
7910
7911 static bool
7912 tunnel_element_release_match(struct rte_eth_dev *dev,
7913                              struct mlx5_flow_tunnel *tunnel, const void *x)
7914 {
7915         const struct tunnel_db_element_release_ctx *ctx = x;
7916
7917         RTE_SET_USED(dev);
7918         if (ctx->num_elements != 1)
7919                 return false;
7920         else if (ctx->items)
7921                 return ctx->items == &tunnel->item;
7922         else if (ctx->actions)
7923                 return ctx->actions == &tunnel->action;
7924
7925         return false;
7926 }
7927
7928 static void
7929 tunnel_element_release_hit(struct rte_eth_dev *dev,
7930                            struct mlx5_flow_tunnel *tunnel, void *x)
7931 {
7932         struct tunnel_db_element_release_ctx *ctx = x;
7933         ctx->ret = 0;
7934         if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
7935                 mlx5_flow_tunnel_free(dev, tunnel);
7936 }
7937
7938 static void
7939 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
7940 {
7941         struct tunnel_db_element_release_ctx *ctx = x;
7942         RTE_SET_USED(dev);
7943         ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
7944                                       RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7945                                       "invalid argument");
7946 }
7947
7948 static int
7949 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
7950                        struct rte_flow_item *pmd_items,
7951                        uint32_t num_items, struct rte_flow_error *err)
7952 {
7953         struct tunnel_db_element_release_ctx ctx = {
7954                 .items = pmd_items,
7955                 .actions = NULL,
7956                 .num_elements = num_items,
7957                 .error = err,
7958         };
7959
7960         mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
7961                                       tunnel_element_release_hit,
7962                                       tunnel_element_release_miss, &ctx, false);
7963
7964         return ctx.ret;
7965 }
7966
7967 static int
7968 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
7969                          struct rte_flow_action *pmd_actions,
7970                          uint32_t num_actions, struct rte_flow_error *err)
7971 {
7972         struct tunnel_db_element_release_ctx ctx = {
7973                 .items = NULL,
7974                 .actions = pmd_actions,
7975                 .num_elements = num_actions,
7976                 .error = err,
7977         };
7978
7979         mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
7980                                       tunnel_element_release_hit,
7981                                       tunnel_element_release_miss, &ctx, false);
7982
7983         return ctx.ret;
7984 }
7985
7986 static int
7987 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
7988                                   struct rte_mbuf *m,
7989                                   struct rte_flow_restore_info *info,
7990                                   struct rte_flow_error *err)
7991 {
7992         uint64_t ol_flags = m->ol_flags;
7993         const struct mlx5_flow_tbl_data_entry *tble;
7994         const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
7995
7996         if (!is_tunnel_offload_active(dev)) {
7997                 info->flags = 0;
7998                 return 0;
7999         }
8000
8001         if ((ol_flags & mask) != mask)
8002                 goto err;
8003         tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8004         if (!tble) {
8005                 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8006                         dev->data->port_id, m->hash.fdir.hi);
8007                 goto err;
8008         }
8009         MLX5_ASSERT(tble->tunnel);
8010         memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8011         info->group_id = tble->group_id;
8012         info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8013                       RTE_FLOW_RESTORE_INFO_GROUP_ID |
8014                       RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8015
8016         return 0;
8017
8018 err:
8019         return rte_flow_error_set(err, EINVAL,
8020                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8021                                   "failed to get restore info");
8022 }
8023
8024 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8025 static int
8026 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8027                            __rte_unused struct rte_flow_tunnel *app_tunnel,
8028                            __rte_unused struct rte_flow_action **actions,
8029                            __rte_unused uint32_t *num_of_actions,
8030                            __rte_unused struct rte_flow_error *error)
8031 {
8032         return -ENOTSUP;
8033 }
8034
8035 static int
8036 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8037                        __rte_unused struct rte_flow_tunnel *app_tunnel,
8038                        __rte_unused struct rte_flow_item **items,
8039                        __rte_unused uint32_t *num_of_items,
8040                        __rte_unused struct rte_flow_error *error)
8041 {
8042         return -ENOTSUP;
8043 }
8044
8045 static int
8046 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8047                               __rte_unused struct rte_flow_item *pmd_items,
8048                               __rte_unused uint32_t num_items,
8049                               __rte_unused struct rte_flow_error *err)
8050 {
8051         return -ENOTSUP;
8052 }
8053
8054 static int
8055 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8056                                 __rte_unused struct rte_flow_action *pmd_action,
8057                                 __rte_unused uint32_t num_actions,
8058                                 __rte_unused struct rte_flow_error *err)
8059 {
8060         return -ENOTSUP;
8061 }
8062
8063 static int
8064 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8065                                   __rte_unused struct rte_mbuf *m,
8066                                   __rte_unused struct rte_flow_restore_info *i,
8067                                   __rte_unused struct rte_flow_error *err)
8068 {
8069         return -ENOTSUP;
8070 }
8071
8072 static int
8073 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8074                              __rte_unused struct rte_flow *flow,
8075                              __rte_unused const struct rte_flow_attr *attr,
8076                              __rte_unused const struct rte_flow_action *actions,
8077                              __rte_unused uint32_t flow_idx,
8078                              __rte_unused struct tunnel_default_miss_ctx *ctx,
8079                              __rte_unused struct rte_flow_error *error)
8080 {
8081         return -ENOTSUP;
8082 }
8083
8084 static struct mlx5_flow_tunnel *
8085 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8086                     __rte_unused uint32_t id)
8087 {
8088         return NULL;
8089 }
8090
8091 static void
8092 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8093                       __rte_unused struct mlx5_flow_tunnel *tunnel)
8094 {
8095 }
8096
8097 static uint32_t
8098 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8099                                 __rte_unused const struct mlx5_flow_tunnel *t,
8100                                 __rte_unused uint32_t group,
8101                                 __rte_unused uint32_t *table,
8102                                 struct rte_flow_error *error)
8103 {
8104         return rte_flow_error_set(error, ENOTSUP,
8105                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8106                                   "tunnel offload requires DV support");
8107 }
8108
8109 void
8110 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8111                         __rte_unused  uint16_t port_id)
8112 {
8113 }
8114 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
8115