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         struct mlx5e_tir *tir = priv->indir_tir;
 179         struct mlx5_flow_spec *spec;
 180         int err = 0;
 181
 182         spec = mlx5_vzalloc(sizeof(*spec));
 183         if (!spec) {
 184                 netdev_err(priv->netdev, "%s: alloc failed\n", __func__);
 185                 err = -ENOMEM;
 186                 goto out;
 187         }
 188
 189         dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 190         switch (type) {
 191         case ARFS_IPV4_TCP:
 192                 dest.tir_num = tir[MLX5E_TT_IPV4_TCP].tirn;
 193                 break;
 194         case ARFS_IPV4_UDP:
 195                 dest.tir_num = tir[MLX5E_TT_IPV4_UDP].tirn;
 196                 break;
 197         case ARFS_IPV6_TCP:
 198                 dest.tir_num = tir[MLX5E_TT_IPV6_TCP].tirn;
 199                 break;
 200         case ARFS_IPV6_UDP:
 201                 dest.tir_num = tir[MLX5E_TT_IPV6_UDP].tirn;
 202                 break;
 203         default:
 204                 err = -EINVAL;
 205                 goto out;
 206         }
 207
 208         arfs_t->default_rule = mlx5_add_flow_rule(arfs_t->ft.t, spec,
 209                                                   MLX5_FLOW_CONTEXT_ACTION_FWD_DEST,
 210                                                   MLX5_FS_DEFAULT_FLOW_TAG,
 211                                                   &dest);
 212         if (IS_ERR(arfs_t->default_rule)) {
 213                 err = PTR_ERR(arfs_t->default_rule);
 214                 arfs_t->default_rule = NULL;
 215                 netdev_err(priv->netdev, "%s: add rule failed, arfs type=%d\n",
 216                            __func__, type);
 217         }
 218 out:
 219         kvfree(spec);
 220         return err;
 221 }
 222
 223 #define MLX5E_ARFS_NUM_GROUPS   2
 224 #define MLX5E_ARFS_GROUP1_SIZE  BIT(12)
 225 #define MLX5E_ARFS_GROUP2_SIZE  BIT(0)
 226 #define MLX5E_ARFS_TABLE_SIZE   (MLX5E_ARFS_GROUP1_SIZE +\
 227                                  MLX5E_ARFS_GROUP2_SIZE)
 228 static int arfs_create_groups(struct mlx5e_flow_table *ft,
 229                               enum  arfs_type type)
 230 {
 231         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 232         void *outer_headers_c;
 233         int ix = 0;
 234         u32 *in;
 235         int err;
 236         u8 *mc;
 237
 238         ft->g = kcalloc(MLX5E_ARFS_NUM_GROUPS,
 239                         sizeof(*ft->g), GFP_KERNEL);
 240         in = mlx5_vzalloc(inlen);
 241         if  (!in || !ft->g) {
 242                 kvfree(ft->g);
 243                 kvfree(in);
 244                 return -ENOMEM;
 245         }
 246
 247         mc = MLX5_ADDR_OF(create_flow_group_in, in, match_criteria);
 248         outer_headers_c = MLX5_ADDR_OF(fte_match_param, mc,
 249                                        outer_headers);
 250         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, ethertype);
 251         switch (type) {
 252         case ARFS_IPV4_TCP:
 253         case ARFS_IPV6_TCP:
 254                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_dport);
 255                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_sport);
 256                 break;
 257         case ARFS_IPV4_UDP:
 258         case ARFS_IPV6_UDP:
 259                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_dport);
 260                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_sport);
 261                 break;
 262         default:
 263                 err = -EINVAL;
 264                 goto out;
 265         }
 266
 267         switch (type) {
 268         case ARFS_IPV4_TCP:
 269         case ARFS_IPV4_UDP:
 270                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
 271                                  src_ipv4_src_ipv6.ipv4_layout.ipv4);
 272                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
 273                                  dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
 274                 break;
 275         case ARFS_IPV6_TCP:
 276         case ARFS_IPV6_UDP:
 277                 memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
 278                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 279                        0xff, 16);
 280                 memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
 281                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 282                        0xff, 16);
 283                 break;
 284         default:
 285                 err = -EINVAL;
 286                 goto out;
 287         }
 288
 289         MLX5_SET_CFG(in, match_criteria_enable, MLX5_MATCH_OUTER_HEADERS);
 290         MLX5_SET_CFG(in, start_flow_index, ix);
 291         ix += MLX5E_ARFS_GROUP1_SIZE;
 292         MLX5_SET_CFG(in, end_flow_index, ix - 1);
 293         ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
 294         if (IS_ERR(ft->g[ft->num_groups]))
 295                 goto err;
 296         ft->num_groups++;
 297
 298         memset(in, 0, inlen);
 299         MLX5_SET_CFG(in, start_flow_index, ix);
 300         ix += MLX5E_ARFS_GROUP2_SIZE;
 301         MLX5_SET_CFG(in, end_flow_index, ix - 1);
 302         ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
 303         if (IS_ERR(ft->g[ft->num_groups]))
 304                 goto err;
 305         ft->num_groups++;
 306
 307         kvfree(in);
 308         return 0;
 309
 310 err:
 311         err = PTR_ERR(ft->g[ft->num_groups]);
 312         ft->g[ft->num_groups] = NULL;
 313 out:
 314         kvfree(in);
 315
 316         return err;
 317 }
 318
 319 static int arfs_create_table(struct mlx5e_priv *priv,
 320                              enum arfs_type type)
 321 {
 322         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 323         struct mlx5e_flow_table *ft = &arfs->arfs_tables[type].ft;
 324         int err;
 325
 326         ft->t = mlx5_create_flow_table(priv->fs.ns, MLX5E_NIC_PRIO,
 327                                        MLX5E_ARFS_TABLE_SIZE, MLX5E_ARFS_FT_LEVEL);
 328         if (IS_ERR(ft->t)) {
 329                 err = PTR_ERR(ft->t);
 330                 ft->t = NULL;
 331                 return err;
 332         }
 333
 334         err = arfs_create_groups(ft, type);
 335         if (err)
 336                 goto err;
 337
 338         err = arfs_add_default_rule(priv, type);
 339         if (err)
 340                 goto err;
 341
 342         return 0;
 343 err:
 344         mlx5e_destroy_flow_table(ft);
 345         return err;
 346 }
 347
 348 int mlx5e_arfs_create_tables(struct mlx5e_priv *priv)
 349 {
 350         int err = 0;
 351         int i;
 352
 353         if (!(priv->netdev->hw_features & NETIF_F_NTUPLE))
 354                 return 0;
 355
 356         spin_lock_init(&priv->fs.arfs.arfs_lock);
 357         INIT_LIST_HEAD(&priv->fs.arfs.rules);
 358         priv->fs.arfs.wq = create_singlethread_workqueue("mlx5e_arfs");
 359         if (!priv->fs.arfs.wq)
 360                 return -ENOMEM;
 361
 362         for (i = 0; i < ARFS_NUM_TYPES; i++) {
 363                 err = arfs_create_table(priv, i);
 364                 if (err)
 365                         goto err;
 366         }
 367         return 0;
 368 err:
 369         mlx5e_arfs_destroy_tables(priv);
 370         return err;
 371 }
 372
 373 #define MLX5E_ARFS_EXPIRY_QUOTA 60
 374
 375 static void arfs_may_expire_flow(struct mlx5e_priv *priv)
 376 {
 377         struct arfs_rule *arfs_rule;
 378         struct hlist_node *htmp;
 379         int quota = 0;
 380         int i;
 381         int j;
 382
 383         HLIST_HEAD(del_list);
 384         spin_lock_bh(&priv->fs.arfs.arfs_lock);
 385         mlx5e_for_each_arfs_rule(arfs_rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
 386                 if (quota++ > MLX5E_ARFS_EXPIRY_QUOTA)
 387                         break;
 388                 if (!work_pending(&arfs_rule->arfs_work) &&
 389                     rps_may_expire_flow(priv->netdev,
 390                                         arfs_rule->rxq, arfs_rule->flow_id,
 391                                         arfs_rule->filter_id)) {
 392                         hlist_del_init(&arfs_rule->hlist);
 393                         hlist_add_head(&arfs_rule->hlist, &del_list);
 394                 }
 395         }
 396         spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 397         hlist_for_each_entry_safe(arfs_rule, htmp, &del_list, hlist) {
 398                 if (arfs_rule->rule)
 399                         mlx5_del_flow_rule(arfs_rule->rule);
 400                 hlist_del(&arfs_rule->hlist);
 401                 kfree(arfs_rule);
 402         }
 403 }
 404
 405 static void arfs_del_rules(struct mlx5e_priv *priv)
 406 {
 407         struct hlist_node *htmp;
 408         struct arfs_rule *rule;
 409         int i;
 410         int j;
 411
 412         HLIST_HEAD(del_list);
 413         spin_lock_bh(&priv->fs.arfs.arfs_lock);
 414         mlx5e_for_each_arfs_rule(rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
 415                 hlist_del_init(&rule->hlist);
 416                 hlist_add_head(&rule->hlist, &del_list);
 417         }
 418         spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 419
 420         hlist_for_each_entry_safe(rule, htmp, &del_list, hlist) {
 421                 cancel_work_sync(&rule->arfs_work);
 422                 if (rule->rule)
 423                         mlx5_del_flow_rule(rule->rule);
 424                 hlist_del(&rule->hlist);
 425                 kfree(rule);
 426         }
 427 }
 428
 429 static struct hlist_head *
 430 arfs_hash_bucket(struct arfs_table *arfs_t, __be16 src_port,
 431                  __be16 dst_port)
 432 {
 433         unsigned long l;
 434         int bucket_idx;
 435
 436         l = (__force unsigned long)src_port |
 437             ((__force unsigned long)dst_port << 2);
 438
 439         bucket_idx = hash_long(l, ARFS_HASH_SHIFT);
 440
 441         return &arfs_t->rules_hash[bucket_idx];
 442 }
 443
 444 static u8 arfs_get_ip_proto(const struct sk_buff *skb)
 445 {
 446         return (skb->protocol == htons(ETH_P_IP)) ?
 447                 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
 448 }
 449
 450 static struct arfs_table *arfs_get_table(struct mlx5e_arfs_tables *arfs,
 451                                          u8 ip_proto, __be16 etype)
 452 {
 453         if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_TCP)
 454                 return &arfs->arfs_tables[ARFS_IPV4_TCP];
 455         if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_UDP)
 456                 return &arfs->arfs_tables[ARFS_IPV4_UDP];
 457         if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_TCP)
 458                 return &arfs->arfs_tables[ARFS_IPV6_TCP];
 459         if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_UDP)
 460                 return &arfs->arfs_tables[ARFS_IPV6_UDP];
 461
 462         return NULL;
 463 }
 464
 465 static struct mlx5_flow_rule *arfs_add_rule(struct mlx5e_priv *priv,
 466                                             struct arfs_rule *arfs_rule)
 467 {
 468         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 469         struct arfs_tuple *tuple = &arfs_rule->tuple;
 470         struct mlx5_flow_rule *rule = NULL;
 471         struct mlx5_flow_destination dest;
 472         struct arfs_table *arfs_table;
 473         struct mlx5_flow_spec *spec;
 474         struct mlx5_flow_table *ft;
 475         int err = 0;
 476
 477         spec = mlx5_vzalloc(sizeof(*spec));
 478         if (!spec) {
 479                 netdev_err(priv->netdev, "%s: alloc failed\n", __func__);
 480                 err = -ENOMEM;
 481                 goto out;
 482         }
 483         spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
 484         MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 485                          outer_headers.ethertype);
 486         MLX5_SET(fte_match_param, spec->match_value, outer_headers.ethertype,
 487                  ntohs(tuple->etype));
 488         arfs_table = arfs_get_table(arfs, tuple->ip_proto, tuple->etype);
 489         if (!arfs_table) {
 490                 err = -EINVAL;
 491                 goto out;
 492         }
 493
 494         ft = arfs_table->ft.t;
 495         if (tuple->ip_proto == IPPROTO_TCP) {
 496                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 497                                  outer_headers.tcp_dport);
 498                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 499                                  outer_headers.tcp_sport);
 500                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_dport,
 501                          ntohs(tuple->dst_port));
 502                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_sport,
 503                          ntohs(tuple->src_port));
 504         } else {
 505                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 506                                  outer_headers.udp_dport);
 507                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 508                                  outer_headers.udp_sport);
 509                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_dport,
 510                          ntohs(tuple->dst_port));
 511                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_sport,
 512                          ntohs(tuple->src_port));
 513         }
 514         if (tuple->etype == htons(ETH_P_IP)) {
 515                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 516                                     outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4),
 517                        &tuple->src_ipv4,
 518                        4);
 519                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 520                                     outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 521                        &tuple->dst_ipv4,
 522                        4);
 523                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 524                                  outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4);
 525                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 526                                  outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
 527         } else {
 528                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 529                                     outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
 530                        &tuple->src_ipv6,
 531                        16);
 532                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 533                                     outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 534                        &tuple->dst_ipv6,
 535                        16);
 536                 memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 537                                     outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
 538                        0xff,
 539                        16);
 540                 memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 541                                     outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 542                        0xff,
 543                        16);
 544         }
 545         dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 546         dest.tir_num = priv->direct_tir[arfs_rule->rxq].tirn;
 547         rule = mlx5_add_flow_rule(ft, spec, MLX5_FLOW_CONTEXT_ACTION_FWD_DEST,
 548                                   MLX5_FS_DEFAULT_FLOW_TAG,
 549                                   &dest);
 550         if (IS_ERR(rule)) {
 551                 err = PTR_ERR(rule);
 552                 netdev_err(priv->netdev, "%s: add rule(filter id=%d, rq idx=%d) failed, err=%d\n",
 553                            __func__, arfs_rule->filter_id, arfs_rule->rxq, err);
 554         }
 555
 556 out:
 557         kvfree(spec);
 558         return err ? ERR_PTR(err) : rule;
 559 }
 560
 561 static void arfs_modify_rule_rq(struct mlx5e_priv *priv,
 562                                 struct mlx5_flow_rule *rule, u16 rxq)
 563 {
 564         struct mlx5_flow_destination dst;
 565         int err = 0;
 566
 567         dst.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 568         dst.tir_num = priv->direct_tir[rxq].tirn;
 569         err =  mlx5_modify_rule_destination(rule, &dst);
 570         if (err)
 571                 netdev_warn(priv->netdev,
 572                             "Failed to modfiy aRFS rule destination to rq=%d\n", rxq);
 573 }
 574
 575 static void arfs_handle_work(struct work_struct *work)
 576 {
 577         struct arfs_rule *arfs_rule = container_of(work,
 578                                                    struct arfs_rule,
 579                                                    arfs_work);
 580         struct mlx5e_priv *priv = arfs_rule->priv;
 581         struct mlx5_flow_rule *rule;
 582
 583         mutex_lock(&priv->state_lock);
 584         if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
 585                 spin_lock_bh(&priv->fs.arfs.arfs_lock);
 586                 hlist_del(&arfs_rule->hlist);
 587                 spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 588
 589                 mutex_unlock(&priv->state_lock);
 590                 kfree(arfs_rule);
 591                 goto out;
 592         }
 593         mutex_unlock(&priv->state_lock);
 594
 595         if (!arfs_rule->rule) {
 596                 rule = arfs_add_rule(priv, arfs_rule);
 597                 if (IS_ERR(rule))
 598                         goto out;
 599                 arfs_rule->rule = rule;
 600         } else {
 601                 arfs_modify_rule_rq(priv, arfs_rule->rule,
 602                                     arfs_rule->rxq);
 603         }
 604 out:
 605         arfs_may_expire_flow(priv);
 606 }
 607
 608 /* return L4 destination port from ip4/6 packets */
 609 static __be16 arfs_get_dst_port(const struct sk_buff *skb)
 610 {
 611         char *transport_header;
 612
 613         transport_header = skb_transport_header(skb);
 614         if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
 615                 return ((struct tcphdr *)transport_header)->dest;
 616         return ((struct udphdr *)transport_header)->dest;
 617 }
 618
 619 /* return L4 source port from ip4/6 packets */
 620 static __be16 arfs_get_src_port(const struct sk_buff *skb)
 621 {
 622         char *transport_header;
 623
 624         transport_header = skb_transport_header(skb);
 625         if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
 626                 return ((struct tcphdr *)transport_header)->source;
 627         return ((struct udphdr *)transport_header)->source;
 628 }
 629
 630 static struct arfs_rule *arfs_alloc_rule(struct mlx5e_priv *priv,
 631                                          struct arfs_table *arfs_t,
 632                                          const struct sk_buff *skb,
 633                                          u16 rxq, u32 flow_id)
 634 {
 635         struct arfs_rule *rule;
 636         struct arfs_tuple *tuple;
 637
 638         rule = kzalloc(sizeof(*rule), GFP_ATOMIC);
 639         if (!rule)
 640                 return NULL;
 641
 642         rule->priv = priv;
 643         rule->rxq = rxq;
 644         INIT_WORK(&rule->arfs_work, arfs_handle_work);
 645
 646         tuple = &rule->tuple;
 647         tuple->etype = skb->protocol;
 648         if (tuple->etype == htons(ETH_P_IP)) {
 649                 tuple->src_ipv4 = ip_hdr(skb)->saddr;
 650                 tuple->dst_ipv4 = ip_hdr(skb)->daddr;
 651         } else {
 652                 memcpy(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
 653                        sizeof(struct in6_addr));
 654                 memcpy(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
 655                        sizeof(struct in6_addr));
 656         }
 657         tuple->ip_proto = arfs_get_ip_proto(skb);
 658         tuple->src_port = arfs_get_src_port(skb);
 659         tuple->dst_port = arfs_get_dst_port(skb);
 660
 661         rule->flow_id = flow_id;
 662         rule->filter_id = priv->fs.arfs.last_filter_id++ % RPS_NO_FILTER;
 663
 664         hlist_add_head(&rule->hlist,
 665                        arfs_hash_bucket(arfs_t, tuple->src_port,
 666                                         tuple->dst_port));
 667         return rule;
 668 }
 669
 670 static bool arfs_cmp_ips(struct arfs_tuple *tuple,
 671                          const struct sk_buff *skb)
 672 {
 673         if (tuple->etype == htons(ETH_P_IP) &&
 674             tuple->src_ipv4 == ip_hdr(skb)->saddr &&
 675             tuple->dst_ipv4 == ip_hdr(skb)->daddr)
 676                 return true;
 677         if (tuple->etype == htons(ETH_P_IPV6) &&
 678             (!memcmp(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
 679                      sizeof(struct in6_addr))) &&
 680             (!memcmp(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
 681                      sizeof(struct in6_addr))))
 682                 return true;
 683         return false;
 684 }
 685
 686 static struct arfs_rule *arfs_find_rule(struct arfs_table *arfs_t,
 687                                         const struct sk_buff *skb)
 688 {
 689         struct arfs_rule *arfs_rule;
 690         struct hlist_head *head;
 691         __be16 src_port = arfs_get_src_port(skb);
 692         __be16 dst_port = arfs_get_dst_port(skb);
 693
 694         head = arfs_hash_bucket(arfs_t, src_port, dst_port);
 695         hlist_for_each_entry(arfs_rule, head, hlist) {
 696                 if (arfs_rule->tuple.src_port == src_port &&
 697                     arfs_rule->tuple.dst_port == dst_port &&
 698                     arfs_cmp_ips(&arfs_rule->tuple, skb)) {
 699                         return arfs_rule;
 700                 }
 701         }
 702
 703         return NULL;
 704 }
 705
 706 int mlx5e_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
 707                         u16 rxq_index, u32 flow_id)
 708 {
 709         struct mlx5e_priv *priv = netdev_priv(dev);
 710         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 711         struct arfs_table *arfs_t;
 712         struct arfs_rule *arfs_rule;
 713
 714         if (skb->protocol != htons(ETH_P_IP) &&
 715             skb->protocol != htons(ETH_P_IPV6))
 716                 return -EPROTONOSUPPORT;
 717
 718         arfs_t = arfs_get_table(arfs, arfs_get_ip_proto(skb), skb->protocol);
 719         if (!arfs_t)
 720                 return -EPROTONOSUPPORT;
 721
 722         spin_lock_bh(&arfs->arfs_lock);
 723         arfs_rule = arfs_find_rule(arfs_t, skb);
 724         if (arfs_rule) {
 725                 if (arfs_rule->rxq == rxq_index) {
 726                         spin_unlock_bh(&arfs->arfs_lock);
 727                         return arfs_rule->filter_id;
 728                 }
 729                 arfs_rule->rxq = rxq_index;
 730         } else {
 731                 arfs_rule = arfs_alloc_rule(priv, arfs_t, skb,
 732                                             rxq_index, flow_id);
 733                 if (!arfs_rule) {
 734                         spin_unlock_bh(&arfs->arfs_lock);
 735                         return -ENOMEM;
 736                 }
 737         }
 738         queue_work(priv->fs.arfs.wq, &arfs_rule->arfs_work);
 739         spin_unlock_bh(&arfs->arfs_lock);
 740         return arfs_rule->filter_id;
 741 }
 742 #endif