drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c

   1 /*
   2  * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #ifdef CONFIG_RFS_ACCEL
  34
  35 #include <linux/hash.h>
  36 #include <linux/mlx5/fs.h>
  37 #include <linux/ip.h>
  38 #include <linux/ipv6.h>
  39 #include "en.h"
  40
  41 struct arfs_tuple {
  42         __be16 etype;
  43         u8     ip_proto;
  44         union {
  45                 __be32 src_ipv4;
  46                 struct in6_addr src_ipv6;
  47         };
  48         union {
  49                 __be32 dst_ipv4;
  50                 struct in6_addr dst_ipv6;
  51         };
  52         __be16 src_port;
  53         __be16 dst_port;
  54 };
  55
  56 struct arfs_rule {
  57         struct mlx5e_priv       *priv;
  58         struct work_struct      arfs_work;
  59         struct mlx5_flow_rule   *rule;
  60         struct hlist_node       hlist;
  61         int                     rxq;
  62         /* Flow ID passed to ndo_rx_flow_steer */
  63         int                     flow_id;
  64         /* Filter ID returned by ndo_rx_flow_steer */
  65         int                     filter_id;
  66         struct arfs_tuple       tuple;
  67 };
  68
  69 #define mlx5e_for_each_arfs_rule(hn, tmp, arfs_tables, i, j) \
  70         for (i = 0; i < ARFS_NUM_TYPES; i++) \
  71                 mlx5e_for_each_hash_arfs_rule(hn, tmp, arfs_tables[i].rules_hash, j)
  72
  73 #define mlx5e_for_each_hash_arfs_rule(hn, tmp, hash, j) \
  74         for (j = 0; j < ARFS_HASH_SIZE; j++) \
  75                 hlist_for_each_entry_safe(hn, tmp, &hash[j], hlist)
  76
  77 static enum mlx5e_traffic_types arfs_get_tt(enum arfs_type type)
  78 {
  79         switch (type) {
  80         case ARFS_IPV4_TCP:
  81                 return MLX5E_TT_IPV4_TCP;
  82         case ARFS_IPV4_UDP:
  83                 return MLX5E_TT_IPV4_UDP;
  84         case ARFS_IPV6_TCP:
  85                 return MLX5E_TT_IPV6_TCP;
  86         case ARFS_IPV6_UDP:
  87                 return MLX5E_TT_IPV6_UDP;
  88         default:
  89                 return -EINVAL;
  90         }
  91 }
  92
  93 static int arfs_disable(struct mlx5e_priv *priv)
  94 {
  95         struct mlx5_flow_destination dest;
  96         struct mlx5e_tir *tir = priv->indir_tir;
  97         int err = 0;
  98         int tt;
  99         int i;
 100
 101         dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 102         for (i = 0; i < ARFS_NUM_TYPES; i++) {
 103                 dest.tir_num = tir[i].tirn;
 104                 tt = arfs_get_tt(i);
 105                 /* Modify ttc rules destination to bypass the aRFS tables*/
 106                 err = mlx5_modify_rule_destination(priv->fs.ttc.rules[tt],
 107                                                    &dest);
 108                 if (err) {
 109                         netdev_err(priv->netdev,
 110                                    "%s: modify ttc destination failed\n",
 111                                    __func__);
 112                         return err;
 113                 }
 114         }
 115         return 0;
 116 }
 117
 118 static void arfs_del_rules(struct mlx5e_priv *priv);
 119
 120 int mlx5e_arfs_disable(struct mlx5e_priv *priv)
 121 {
 122         arfs_del_rules(priv);
 123
 124         return arfs_disable(priv);
 125 }
 126
 127 int mlx5e_arfs_enable(struct mlx5e_priv *priv)
 128 {
 129         struct mlx5_flow_destination dest;
 130         int err = 0;
 131         int tt;
 132         int i;
 133
 134         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 135         for (i = 0; i < ARFS_NUM_TYPES; i++) {
 136                 dest.ft = priv->fs.arfs.arfs_tables[i].ft.t;
 137                 tt = arfs_get_tt(i);
 138                 /* Modify ttc rules destination to point on the aRFS FTs */
 139                 err = mlx5_modify_rule_destination(priv->fs.ttc.rules[tt],
 140                                                    &dest);
 141                 if (err) {
 142                         netdev_err(priv->netdev,
 143                                    "%s: modify ttc destination failed err=%d\n",
 144                                    __func__, err);
 145                         arfs_disable(priv);
 146                         return err;
 147                 }
 148         }
 149         return 0;
 150 }
 151
 152 static void arfs_destroy_table(struct arfs_table *arfs_t)
 153 {
 154         mlx5_del_flow_rule(arfs_t->default_rule);
 155         mlx5e_destroy_flow_table(&arfs_t->ft);
 156 }
 157
 158 void mlx5e_arfs_destroy_tables(struct mlx5e_priv *priv)
 159 {
 160         int i;
 161
 162         if (!(priv->netdev->hw_features & NETIF_F_NTUPLE))
 163                 return;
 164
 165         arfs_del_rules(priv);
 166         destroy_workqueue(priv->fs.arfs.wq);
 167         for (i = 0; i < ARFS_NUM_TYPES; i++) {
 168                 if (!IS_ERR_OR_NULL(priv->fs.arfs.arfs_tables[i].ft.t))
 169                         arfs_destroy_table(&priv->fs.arfs.arfs_tables[i]);
 170         }
 171 }
 172
 173 static int arfs_add_default_rule(struct mlx5e_priv *priv,
 174                                  enum arfs_type type)
 175 {
 176         struct arfs_table *arfs_t = &priv->fs.arfs.arfs_tables[type];
 177         struct mlx5_flow_destination dest;
 178         u8 match_criteria_enable = 0;
 179         struct mlx5e_tir *tir = priv->indir_tir;
 180         u32 *match_criteria;
 181         u32 *match_value;
 182         int err = 0;
 183
 184         match_value     = mlx5_vzalloc(MLX5_ST_SZ_BYTES(fte_match_param));
 185         match_criteria  = mlx5_vzalloc(MLX5_ST_SZ_BYTES(fte_match_param));
 186         if (!match_value || !match_criteria) {
 187                 netdev_err(priv->netdev, "%s: alloc failed\n", __func__);
 188                 err = -ENOMEM;
 189                 goto out;
 190         }
 191
 192         dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 193         switch (type) {
 194         case ARFS_IPV4_TCP:
 195                 dest.tir_num = tir[MLX5E_TT_IPV4_TCP].tirn;
 196                 break;
 197         case ARFS_IPV4_UDP:
 198                 dest.tir_num = tir[MLX5E_TT_IPV4_UDP].tirn;
 199                 break;
 200         case ARFS_IPV6_TCP:
 201                 dest.tir_num = tir[MLX5E_TT_IPV6_TCP].tirn;
 202                 break;
 203         case ARFS_IPV6_UDP:
 204                 dest.tir_num = tir[MLX5E_TT_IPV6_UDP].tirn;
 205                 break;
 206         default:
 207                 err = -EINVAL;
 208                 goto out;
 209         }
 210
 211         arfs_t->default_rule = mlx5_add_flow_rule(arfs_t->ft.t, match_criteria_enable,
 212                                                   match_criteria, match_value,
 213                                                   MLX5_FLOW_CONTEXT_ACTION_FWD_DEST,
 214                                                   MLX5_FS_DEFAULT_FLOW_TAG,
 215                                                   &dest);
 216         if (IS_ERR(arfs_t->default_rule)) {
 217                 err = PTR_ERR(arfs_t->default_rule);
 218                 arfs_t->default_rule = NULL;
 219                 netdev_err(priv->netdev, "%s: add rule failed, arfs type=%d\n",
 220                            __func__, type);
 221         }
 222 out:
 223         kvfree(match_criteria);
 224         kvfree(match_value);
 225         return err;
 226 }
 227
 228 #define MLX5E_ARFS_NUM_GROUPS   2
 229 #define MLX5E_ARFS_GROUP1_SIZE  BIT(12)
 230 #define MLX5E_ARFS_GROUP2_SIZE  BIT(0)
 231 #define MLX5E_ARFS_TABLE_SIZE   (MLX5E_ARFS_GROUP1_SIZE +\
 232                                  MLX5E_ARFS_GROUP2_SIZE)
 233 static int arfs_create_groups(struct mlx5e_flow_table *ft,
 234                               enum  arfs_type type)
 235 {
 236         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 237         void *outer_headers_c;
 238         int ix = 0;
 239         u32 *in;
 240         int err;
 241         u8 *mc;
 242
 243         ft->g = kcalloc(MLX5E_ARFS_NUM_GROUPS,
 244                         sizeof(*ft->g), GFP_KERNEL);
 245         in = mlx5_vzalloc(inlen);
 246         if  (!in || !ft->g) {
 247                 kvfree(ft->g);
 248                 kvfree(in);
 249                 return -ENOMEM;
 250         }
 251
 252         mc = MLX5_ADDR_OF(create_flow_group_in, in, match_criteria);
 253         outer_headers_c = MLX5_ADDR_OF(fte_match_param, mc,
 254                                        outer_headers);
 255         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, ethertype);
 256         switch (type) {
 257         case ARFS_IPV4_TCP:
 258         case ARFS_IPV6_TCP:
 259                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_dport);
 260                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_sport);
 261                 break;
 262         case ARFS_IPV4_UDP:
 263         case ARFS_IPV6_UDP:
 264                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_dport);
 265                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_sport);
 266                 break;
 267         default:
 268                 err = -EINVAL;
 269                 goto out;
 270         }
 271
 272         switch (type) {
 273         case ARFS_IPV4_TCP:
 274         case ARFS_IPV4_UDP:
 275                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
 276                                  src_ipv4_src_ipv6.ipv4_layout.ipv4);
 277                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
 278                                  dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
 279                 break;
 280         case ARFS_IPV6_TCP:
 281         case ARFS_IPV6_UDP:
 282                 memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
 283                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 284                        0xff, 16);
 285                 memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
 286                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 287                        0xff, 16);
 288                 break;
 289         default:
 290                 err = -EINVAL;
 291                 goto out;
 292         }
 293
 294         MLX5_SET_CFG(in, match_criteria_enable, MLX5_MATCH_OUTER_HEADERS);
 295         MLX5_SET_CFG(in, start_flow_index, ix);
 296         ix += MLX5E_ARFS_GROUP1_SIZE;
 297         MLX5_SET_CFG(in, end_flow_index, ix - 1);
 298         ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
 299         if (IS_ERR(ft->g[ft->num_groups]))
 300                 goto err;
 301         ft->num_groups++;
 302
 303         memset(in, 0, inlen);
 304         MLX5_SET_CFG(in, start_flow_index, ix);
 305         ix += MLX5E_ARFS_GROUP2_SIZE;
 306         MLX5_SET_CFG(in, end_flow_index, ix - 1);
 307         ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
 308         if (IS_ERR(ft->g[ft->num_groups]))
 309                 goto err;
 310         ft->num_groups++;
 311
 312         kvfree(in);
 313         return 0;
 314
 315 err:
 316         err = PTR_ERR(ft->g[ft->num_groups]);
 317         ft->g[ft->num_groups] = NULL;
 318 out:
 319         kvfree(in);
 320
 321         return err;
 322 }
 323
 324 static int arfs_create_table(struct mlx5e_priv *priv,
 325                              enum arfs_type type)
 326 {
 327         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 328         struct mlx5e_flow_table *ft = &arfs->arfs_tables[type].ft;
 329         int err;
 330
 331         ft->t = mlx5_create_flow_table(priv->fs.ns, MLX5E_NIC_PRIO,
 332                                        MLX5E_ARFS_TABLE_SIZE, MLX5E_ARFS_FT_LEVEL);
 333         if (IS_ERR(ft->t)) {
 334                 err = PTR_ERR(ft->t);
 335                 ft->t = NULL;
 336                 return err;
 337         }
 338
 339         err = arfs_create_groups(ft, type);
 340         if (err)
 341                 goto err;
 342
 343         err = arfs_add_default_rule(priv, type);
 344         if (err)
 345                 goto err;
 346
 347         return 0;
 348 err:
 349         mlx5e_destroy_flow_table(ft);
 350         return err;
 351 }
 352
 353 int mlx5e_arfs_create_tables(struct mlx5e_priv *priv)
 354 {
 355         int err = 0;
 356         int i;
 357
 358         if (!(priv->netdev->hw_features & NETIF_F_NTUPLE))
 359                 return 0;
 360
 361         spin_lock_init(&priv->fs.arfs.arfs_lock);
 362         INIT_LIST_HEAD(&priv->fs.arfs.rules);
 363         priv->fs.arfs.wq = create_singlethread_workqueue("mlx5e_arfs");
 364         if (!priv->fs.arfs.wq)
 365                 return -ENOMEM;
 366
 367         for (i = 0; i < ARFS_NUM_TYPES; i++) {
 368                 err = arfs_create_table(priv, i);
 369                 if (err)
 370                         goto err;
 371         }
 372         return 0;
 373 err:
 374         mlx5e_arfs_destroy_tables(priv);
 375         return err;
 376 }
 377
 378 #define MLX5E_ARFS_EXPIRY_QUOTA 60
 379
 380 static void arfs_may_expire_flow(struct mlx5e_priv *priv)
 381 {
 382         struct arfs_rule *arfs_rule;
 383         struct hlist_node *htmp;
 384         int quota = 0;
 385         int i;
 386         int j;
 387
 388         HLIST_HEAD(del_list);
 389         spin_lock_bh(&priv->fs.arfs.arfs_lock);
 390         mlx5e_for_each_arfs_rule(arfs_rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
 391                 if (quota++ > MLX5E_ARFS_EXPIRY_QUOTA)
 392                         break;
 393                 if (!work_pending(&arfs_rule->arfs_work) &&
 394                     rps_may_expire_flow(priv->netdev,
 395                                         arfs_rule->rxq, arfs_rule->flow_id,
 396                                         arfs_rule->filter_id)) {
 397                         hlist_del_init(&arfs_rule->hlist);
 398                         hlist_add_head(&arfs_rule->hlist, &del_list);
 399                 }
 400         }
 401         spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 402         hlist_for_each_entry_safe(arfs_rule, htmp, &del_list, hlist) {
 403                 if (arfs_rule->rule)
 404                         mlx5_del_flow_rule(arfs_rule->rule);
 405                 hlist_del(&arfs_rule->hlist);
 406                 kfree(arfs_rule);
 407         }
 408 }
 409
 410 static void arfs_del_rules(struct mlx5e_priv *priv)
 411 {
 412         struct hlist_node *htmp;
 413         struct arfs_rule *rule;
 414         int i;
 415         int j;
 416
 417         HLIST_HEAD(del_list);
 418         spin_lock_bh(&priv->fs.arfs.arfs_lock);
 419         mlx5e_for_each_arfs_rule(rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
 420                 hlist_del_init(&rule->hlist);
 421                 hlist_add_head(&rule->hlist, &del_list);
 422         }
 423         spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 424
 425         hlist_for_each_entry_safe(rule, htmp, &del_list, hlist) {
 426                 cancel_work_sync(&rule->arfs_work);
 427                 if (rule->rule)
 428                         mlx5_del_flow_rule(rule->rule);
 429                 hlist_del(&rule->hlist);
 430                 kfree(rule);
 431         }
 432 }
 433
 434 static struct hlist_head *
 435 arfs_hash_bucket(struct arfs_table *arfs_t, __be16 src_port,
 436                  __be16 dst_port)
 437 {
 438         unsigned long l;
 439         int bucket_idx;
 440
 441         l = (__force unsigned long)src_port |
 442             ((__force unsigned long)dst_port << 2);
 443
 444         bucket_idx = hash_long(l, ARFS_HASH_SHIFT);
 445
 446         return &arfs_t->rules_hash[bucket_idx];
 447 }
 448
 449 static u8 arfs_get_ip_proto(const struct sk_buff *skb)
 450 {
 451         return (skb->protocol == htons(ETH_P_IP)) ?
 452                 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
 453 }
 454
 455 static struct arfs_table *arfs_get_table(struct mlx5e_arfs_tables *arfs,
 456                                          u8 ip_proto, __be16 etype)
 457 {
 458         if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_TCP)
 459                 return &arfs->arfs_tables[ARFS_IPV4_TCP];
 460         if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_UDP)
 461                 return &arfs->arfs_tables[ARFS_IPV4_UDP];
 462         if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_TCP)
 463                 return &arfs->arfs_tables[ARFS_IPV6_TCP];
 464         if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_UDP)
 465                 return &arfs->arfs_tables[ARFS_IPV6_UDP];
 466
 467         return NULL;
 468 }
 469
 470 static struct mlx5_flow_rule *arfs_add_rule(struct mlx5e_priv *priv,
 471                                             struct arfs_rule *arfs_rule)
 472 {
 473         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 474         struct arfs_tuple *tuple = &arfs_rule->tuple;
 475         struct mlx5_flow_rule *rule = NULL;
 476         struct mlx5_flow_destination dest;
 477         struct arfs_table *arfs_table;
 478         u8 match_criteria_enable = 0;
 479         struct mlx5_flow_table *ft;
 480         u32 *match_criteria;
 481         u32 *match_value;
 482         int err = 0;
 483
 484         match_value     = mlx5_vzalloc(MLX5_ST_SZ_BYTES(fte_match_param));
 485         match_criteria  = mlx5_vzalloc(MLX5_ST_SZ_BYTES(fte_match_param));
 486         if (!match_value || !match_criteria) {
 487                 netdev_err(priv->netdev, "%s: alloc failed\n", __func__);
 488                 err = -ENOMEM;
 489                 goto out;
 490         }
 491         match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
 492         MLX5_SET_TO_ONES(fte_match_param, match_criteria,
 493                          outer_headers.ethertype);
 494         MLX5_SET(fte_match_param, match_value, outer_headers.ethertype,
 495                  ntohs(tuple->etype));
 496         arfs_table = arfs_get_table(arfs, tuple->ip_proto, tuple->etype);
 497         if (!arfs_table) {
 498                 err = -EINVAL;
 499                 goto out;
 500         }
 501
 502         ft = arfs_table->ft.t;
 503         if (tuple->ip_proto == IPPROTO_TCP) {
 504                 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
 505                                  outer_headers.tcp_dport);
 506                 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
 507                                  outer_headers.tcp_sport);
 508                 MLX5_SET(fte_match_param, match_value, outer_headers.tcp_dport,
 509                          ntohs(tuple->dst_port));
 510                 MLX5_SET(fte_match_param, match_value, outer_headers.tcp_sport,
 511                          ntohs(tuple->src_port));
 512         } else {
 513                 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
 514                                  outer_headers.udp_dport);
 515                 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
 516                                  outer_headers.udp_sport);
 517                 MLX5_SET(fte_match_param, match_value, outer_headers.udp_dport,
 518                          ntohs(tuple->dst_port));
 519                 MLX5_SET(fte_match_param, match_value, outer_headers.udp_sport,
 520                          ntohs(tuple->src_port));
 521         }
 522         if (tuple->etype == htons(ETH_P_IP)) {
 523                 memcpy(MLX5_ADDR_OF(fte_match_param, match_value,
 524                                     outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4),
 525                        &tuple->src_ipv4,
 526                        4);
 527                 memcpy(MLX5_ADDR_OF(fte_match_param, match_value,
 528                                     outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 529                        &tuple->dst_ipv4,
 530                        4);
 531                 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
 532                                  outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4);
 533                 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
 534                                  outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
 535         } else {
 536                 memcpy(MLX5_ADDR_OF(fte_match_param, match_value,
 537                                     outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
 538                        &tuple->src_ipv6,
 539                        16);
 540                 memcpy(MLX5_ADDR_OF(fte_match_param, match_value,
 541                                     outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 542                        &tuple->dst_ipv6,
 543                        16);
 544                 memset(MLX5_ADDR_OF(fte_match_param, match_criteria,
 545                                     outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
 546                        0xff,
 547                        16);
 548                 memset(MLX5_ADDR_OF(fte_match_param, match_criteria,
 549                                     outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 550                        0xff,
 551                        16);
 552         }
 553         dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 554         dest.tir_num = priv->direct_tir[arfs_rule->rxq].tirn;
 555         rule = mlx5_add_flow_rule(ft, match_criteria_enable, match_criteria,
 556                                   match_value, MLX5_FLOW_CONTEXT_ACTION_FWD_DEST,
 557                                   MLX5_FS_DEFAULT_FLOW_TAG,
 558                                   &dest);
 559         if (IS_ERR(rule)) {
 560                 err = PTR_ERR(rule);
 561                 netdev_err(priv->netdev, "%s: add rule(filter id=%d, rq idx=%d) failed, err=%d\n",
 562                            __func__, arfs_rule->filter_id, arfs_rule->rxq, err);
 563         }
 564
 565 out:
 566         kvfree(match_criteria);
 567         kvfree(match_value);
 568         return err ? ERR_PTR(err) : rule;
 569 }
 570
 571 static void arfs_modify_rule_rq(struct mlx5e_priv *priv,
 572                                 struct mlx5_flow_rule *rule, u16 rxq)
 573 {
 574         struct mlx5_flow_destination dst;
 575         int err = 0;
 576
 577         dst.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 578         dst.tir_num = priv->direct_tir[rxq].tirn;
 579         err =  mlx5_modify_rule_destination(rule, &dst);
 580         if (err)
 581                 netdev_warn(priv->netdev,
 582                             "Failed to modfiy aRFS rule destination to rq=%d\n", rxq);
 583 }
 584
 585 static void arfs_handle_work(struct work_struct *work)
 586 {
 587         struct arfs_rule *arfs_rule = container_of(work,
 588                                                    struct arfs_rule,
 589                                                    arfs_work);
 590         struct mlx5e_priv *priv = arfs_rule->priv;
 591         struct mlx5_flow_rule *rule;
 592
 593         mutex_lock(&priv->state_lock);
 594         if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
 595                 spin_lock_bh(&priv->fs.arfs.arfs_lock);
 596                 hlist_del(&arfs_rule->hlist);
 597                 spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 598
 599                 mutex_unlock(&priv->state_lock);
 600                 kfree(arfs_rule);
 601                 goto out;
 602         }
 603         mutex_unlock(&priv->state_lock);
 604
 605         if (!arfs_rule->rule) {
 606                 rule = arfs_add_rule(priv, arfs_rule);
 607                 if (IS_ERR(rule))
 608                         goto out;
 609                 arfs_rule->rule = rule;
 610         } else {
 611                 arfs_modify_rule_rq(priv, arfs_rule->rule,
 612                                     arfs_rule->rxq);
 613         }
 614 out:
 615         arfs_may_expire_flow(priv);
 616 }
 617
 618 /* return L4 destination port from ip4/6 packets */
 619 static __be16 arfs_get_dst_port(const struct sk_buff *skb)
 620 {
 621         char *transport_header;
 622
 623         transport_header = skb_transport_header(skb);
 624         if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
 625                 return ((struct tcphdr *)transport_header)->dest;
 626         return ((struct udphdr *)transport_header)->dest;
 627 }
 628
 629 /* return L4 source port from ip4/6 packets */
 630 static __be16 arfs_get_src_port(const struct sk_buff *skb)
 631 {
 632         char *transport_header;
 633
 634         transport_header = skb_transport_header(skb);
 635         if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
 636                 return ((struct tcphdr *)transport_header)->source;
 637         return ((struct udphdr *)transport_header)->source;
 638 }
 639
 640 static struct arfs_rule *arfs_alloc_rule(struct mlx5e_priv *priv,
 641                                          struct arfs_table *arfs_t,
 642                                          const struct sk_buff *skb,
 643                                          u16 rxq, u32 flow_id)
 644 {
 645         struct arfs_rule *rule;
 646         struct arfs_tuple *tuple;
 647
 648         rule = kzalloc(sizeof(*rule), GFP_ATOMIC);
 649         if (!rule)
 650                 return NULL;
 651
 652         rule->priv = priv;
 653         rule->rxq = rxq;
 654         INIT_WORK(&rule->arfs_work, arfs_handle_work);
 655
 656         tuple = &rule->tuple;
 657         tuple->etype = skb->protocol;
 658         if (tuple->etype == htons(ETH_P_IP)) {
 659                 tuple->src_ipv4 = ip_hdr(skb)->saddr;
 660                 tuple->dst_ipv4 = ip_hdr(skb)->daddr;
 661         } else {
 662                 memcpy(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
 663                        sizeof(struct in6_addr));
 664                 memcpy(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
 665                        sizeof(struct in6_addr));
 666         }
 667         tuple->ip_proto = arfs_get_ip_proto(skb);
 668         tuple->src_port = arfs_get_src_port(skb);
 669         tuple->dst_port = arfs_get_dst_port(skb);
 670
 671         rule->flow_id = flow_id;
 672         rule->filter_id = priv->fs.arfs.last_filter_id++ % RPS_NO_FILTER;
 673
 674         hlist_add_head(&rule->hlist,
 675                        arfs_hash_bucket(arfs_t, tuple->src_port,
 676                                         tuple->dst_port));
 677         return rule;
 678 }
 679
 680 static bool arfs_cmp_ips(struct arfs_tuple *tuple,
 681                          const struct sk_buff *skb)
 682 {
 683         if (tuple->etype == htons(ETH_P_IP) &&
 684             tuple->src_ipv4 == ip_hdr(skb)->saddr &&
 685             tuple->dst_ipv4 == ip_hdr(skb)->daddr)
 686                 return true;
 687         if (tuple->etype == htons(ETH_P_IPV6) &&
 688             (!memcmp(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
 689                      sizeof(struct in6_addr))) &&
 690             (!memcmp(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
 691                      sizeof(struct in6_addr))))
 692                 return true;
 693         return false;
 694 }
 695
 696 static struct arfs_rule *arfs_find_rule(struct arfs_table *arfs_t,
 697                                         const struct sk_buff *skb)
 698 {
 699         struct arfs_rule *arfs_rule;
 700         struct hlist_head *head;
 701         __be16 src_port = arfs_get_src_port(skb);
 702         __be16 dst_port = arfs_get_dst_port(skb);
 703
 704         head = arfs_hash_bucket(arfs_t, src_port, dst_port);
 705         hlist_for_each_entry(arfs_rule, head, hlist) {
 706                 if (arfs_rule->tuple.src_port == src_port &&
 707                     arfs_rule->tuple.dst_port == dst_port &&
 708                     arfs_cmp_ips(&arfs_rule->tuple, skb)) {
 709                         return arfs_rule;
 710                 }
 711         }
 712
 713         return NULL;
 714 }
 715
 716 int mlx5e_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
 717                         u16 rxq_index, u32 flow_id)
 718 {
 719         struct mlx5e_priv *priv = netdev_priv(dev);
 720         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 721         struct arfs_table *arfs_t;
 722         struct arfs_rule *arfs_rule;
 723
 724         if (skb->protocol != htons(ETH_P_IP) &&
 725             skb->protocol != htons(ETH_P_IPV6))
 726                 return -EPROTONOSUPPORT;
 727
 728         arfs_t = arfs_get_table(arfs, arfs_get_ip_proto(skb), skb->protocol);
 729         if (!arfs_t)
 730                 return -EPROTONOSUPPORT;
 731
 732         spin_lock_bh(&arfs->arfs_lock);
 733         arfs_rule = arfs_find_rule(arfs_t, skb);
 734         if (arfs_rule) {
 735                 if (arfs_rule->rxq == rxq_index) {
 736                         spin_unlock_bh(&arfs->arfs_lock);
 737                         return arfs_rule->filter_id;
 738                 }
 739                 arfs_rule->rxq = rxq_index;
 740         } else {
 741                 arfs_rule = arfs_alloc_rule(priv, arfs_t, skb,
 742                                             rxq_index, flow_id);
 743                 if (!arfs_rule) {
 744                         spin_unlock_bh(&arfs->arfs_lock);
 745                         return -ENOMEM;
 746                 }
 747         }
 748         queue_work(priv->fs.arfs.wq, &arfs_rule->arfs_work);
 749         spin_unlock_bh(&arfs->arfs_lock);
 750         return arfs_rule->filter_id;
 751 }
 752 #endif