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_handle *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, NULL);
 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, NULL);
 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_rules(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 mlx5e_tir *tir = priv->indir_tir;
 178         struct mlx5_flow_destination dest;
 179         MLX5_DECLARE_FLOW_ACT(flow_act);
 180         struct mlx5_flow_spec *spec;
 181         int err = 0;
 182
 183         spec = mlx5_vzalloc(sizeof(*spec));
 184         if (!spec) {
 185                 netdev_err(priv->netdev, "%s: alloc failed\n", __func__);
 186                 err = -ENOMEM;
 187                 goto out;
 188         }
 189
 190         dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 191         switch (type) {
 192         case ARFS_IPV4_TCP:
 193                 dest.tir_num = tir[MLX5E_TT_IPV4_TCP].tirn;
 194                 break;
 195         case ARFS_IPV4_UDP:
 196                 dest.tir_num = tir[MLX5E_TT_IPV4_UDP].tirn;
 197                 break;
 198         case ARFS_IPV6_TCP:
 199                 dest.tir_num = tir[MLX5E_TT_IPV6_TCP].tirn;
 200                 break;
 201         case ARFS_IPV6_UDP:
 202                 dest.tir_num = tir[MLX5E_TT_IPV6_UDP].tirn;
 203                 break;
 204         default:
 205                 err = -EINVAL;
 206                 goto out;
 207         }
 208
 209         arfs_t->default_rule = mlx5_add_flow_rules(arfs_t->ft.t, spec,
 210                                                    &flow_act,
 211                                                    &dest, 1);
 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         struct mlx5_flow_table_attr ft_attr = {};
 325         int err;
 326
 327         ft->num_groups = 0;
 328
 329         ft_attr.max_fte = MLX5E_ARFS_TABLE_SIZE;
 330         ft_attr.level = MLX5E_ARFS_FT_LEVEL;
 331         ft_attr.prio = MLX5E_NIC_PRIO;
 332
 333         ft->t = mlx5_create_flow_table(priv->fs.ns, &ft_attr);
 334         if (IS_ERR(ft->t)) {
 335                 err = PTR_ERR(ft->t);
 336                 ft->t = NULL;
 337                 return err;
 338         }
 339
 340         err = arfs_create_groups(ft, type);
 341         if (err)
 342                 goto err;
 343
 344         err = arfs_add_default_rule(priv, type);
 345         if (err)
 346                 goto err;
 347
 348         return 0;
 349 err:
 350         mlx5e_destroy_flow_table(ft);
 351         return err;
 352 }
 353
 354 int mlx5e_arfs_create_tables(struct mlx5e_priv *priv)
 355 {
 356         int err = 0;
 357         int i;
 358
 359         if (!(priv->netdev->hw_features & NETIF_F_NTUPLE))
 360                 return 0;
 361
 362         spin_lock_init(&priv->fs.arfs.arfs_lock);
 363         INIT_LIST_HEAD(&priv->fs.arfs.rules);
 364         priv->fs.arfs.wq = create_singlethread_workqueue("mlx5e_arfs");
 365         if (!priv->fs.arfs.wq)
 366                 return -ENOMEM;
 367
 368         for (i = 0; i < ARFS_NUM_TYPES; i++) {
 369                 err = arfs_create_table(priv, i);
 370                 if (err)
 371                         goto err;
 372         }
 373         return 0;
 374 err:
 375         mlx5e_arfs_destroy_tables(priv);
 376         return err;
 377 }
 378
 379 #define MLX5E_ARFS_EXPIRY_QUOTA 60
 380
 381 static void arfs_may_expire_flow(struct mlx5e_priv *priv)
 382 {
 383         struct arfs_rule *arfs_rule;
 384         struct hlist_node *htmp;
 385         int quota = 0;
 386         int i;
 387         int j;
 388
 389         HLIST_HEAD(del_list);
 390         spin_lock_bh(&priv->fs.arfs.arfs_lock);
 391         mlx5e_for_each_arfs_rule(arfs_rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
 392                 if (quota++ > MLX5E_ARFS_EXPIRY_QUOTA)
 393                         break;
 394                 if (!work_pending(&arfs_rule->arfs_work) &&
 395                     rps_may_expire_flow(priv->netdev,
 396                                         arfs_rule->rxq, arfs_rule->flow_id,
 397                                         arfs_rule->filter_id)) {
 398                         hlist_del_init(&arfs_rule->hlist);
 399                         hlist_add_head(&arfs_rule->hlist, &del_list);
 400                 }
 401         }
 402         spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 403         hlist_for_each_entry_safe(arfs_rule, htmp, &del_list, hlist) {
 404                 if (arfs_rule->rule)
 405                         mlx5_del_flow_rules(arfs_rule->rule);
 406                 hlist_del(&arfs_rule->hlist);
 407                 kfree(arfs_rule);
 408         }
 409 }
 410
 411 static void arfs_del_rules(struct mlx5e_priv *priv)
 412 {
 413         struct hlist_node *htmp;
 414         struct arfs_rule *rule;
 415         int i;
 416         int j;
 417
 418         HLIST_HEAD(del_list);
 419         spin_lock_bh(&priv->fs.arfs.arfs_lock);
 420         mlx5e_for_each_arfs_rule(rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
 421                 hlist_del_init(&rule->hlist);
 422                 hlist_add_head(&rule->hlist, &del_list);
 423         }
 424         spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 425
 426         hlist_for_each_entry_safe(rule, htmp, &del_list, hlist) {
 427                 cancel_work_sync(&rule->arfs_work);
 428                 if (rule->rule)
 429                         mlx5_del_flow_rules(rule->rule);
 430                 hlist_del(&rule->hlist);
 431                 kfree(rule);
 432         }
 433 }
 434
 435 static struct hlist_head *
 436 arfs_hash_bucket(struct arfs_table *arfs_t, __be16 src_port,
 437                  __be16 dst_port)
 438 {
 439         unsigned long l;
 440         int bucket_idx;
 441
 442         l = (__force unsigned long)src_port |
 443             ((__force unsigned long)dst_port << 2);
 444
 445         bucket_idx = hash_long(l, ARFS_HASH_SHIFT);
 446
 447         return &arfs_t->rules_hash[bucket_idx];
 448 }
 449
 450 static u8 arfs_get_ip_proto(const struct sk_buff *skb)
 451 {
 452         return (skb->protocol == htons(ETH_P_IP)) ?
 453                 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
 454 }
 455
 456 static struct arfs_table *arfs_get_table(struct mlx5e_arfs_tables *arfs,
 457                                          u8 ip_proto, __be16 etype)
 458 {
 459         if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_TCP)
 460                 return &arfs->arfs_tables[ARFS_IPV4_TCP];
 461         if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_UDP)
 462                 return &arfs->arfs_tables[ARFS_IPV4_UDP];
 463         if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_TCP)
 464                 return &arfs->arfs_tables[ARFS_IPV6_TCP];
 465         if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_UDP)
 466                 return &arfs->arfs_tables[ARFS_IPV6_UDP];
 467
 468         return NULL;
 469 }
 470
 471 static struct mlx5_flow_handle *arfs_add_rule(struct mlx5e_priv *priv,
 472                                               struct arfs_rule *arfs_rule)
 473 {
 474         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 475         struct arfs_tuple *tuple = &arfs_rule->tuple;
 476         struct mlx5_flow_handle *rule = NULL;
 477         struct mlx5_flow_destination dest;
 478         MLX5_DECLARE_FLOW_ACT(flow_act);
 479         struct arfs_table *arfs_table;
 480         struct mlx5_flow_spec *spec;
 481         struct mlx5_flow_table *ft;
 482         int err = 0;
 483
 484         spec = mlx5_vzalloc(sizeof(*spec));
 485         if (!spec) {
 486                 netdev_err(priv->netdev, "%s: alloc failed\n", __func__);
 487                 err = -ENOMEM;
 488                 goto out;
 489         }
 490         spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
 491         MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 492                          outer_headers.ethertype);
 493         MLX5_SET(fte_match_param, spec->match_value, outer_headers.ethertype,
 494                  ntohs(tuple->etype));
 495         arfs_table = arfs_get_table(arfs, tuple->ip_proto, tuple->etype);
 496         if (!arfs_table) {
 497                 err = -EINVAL;
 498                 goto out;
 499         }
 500
 501         ft = arfs_table->ft.t;
 502         if (tuple->ip_proto == IPPROTO_TCP) {
 503                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 504                                  outer_headers.tcp_dport);
 505                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 506                                  outer_headers.tcp_sport);
 507                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_dport,
 508                          ntohs(tuple->dst_port));
 509                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_sport,
 510                          ntohs(tuple->src_port));
 511         } else {
 512                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 513                                  outer_headers.udp_dport);
 514                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 515                                  outer_headers.udp_sport);
 516                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_dport,
 517                          ntohs(tuple->dst_port));
 518                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_sport,
 519                          ntohs(tuple->src_port));
 520         }
 521         if (tuple->etype == htons(ETH_P_IP)) {
 522                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 523                                     outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4),
 524                        &tuple->src_ipv4,
 525                        4);
 526                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 527                                     outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 528                        &tuple->dst_ipv4,
 529                        4);
 530                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 531                                  outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4);
 532                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 533                                  outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
 534         } else {
 535                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 536                                     outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
 537                        &tuple->src_ipv6,
 538                        16);
 539                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 540                                     outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 541                        &tuple->dst_ipv6,
 542                        16);
 543                 memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 544                                     outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
 545                        0xff,
 546                        16);
 547                 memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 548                                     outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 549                        0xff,
 550                        16);
 551         }
 552         dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 553         dest.tir_num = priv->direct_tir[arfs_rule->rxq].tirn;
 554         rule = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
 555         if (IS_ERR(rule)) {
 556                 err = PTR_ERR(rule);
 557                 netdev_err(priv->netdev, "%s: add rule(filter id=%d, rq idx=%d) failed, err=%d\n",
 558                            __func__, arfs_rule->filter_id, arfs_rule->rxq, err);
 559         }
 560
 561 out:
 562         kvfree(spec);
 563         return err ? ERR_PTR(err) : rule;
 564 }
 565
 566 static void arfs_modify_rule_rq(struct mlx5e_priv *priv,
 567                                 struct mlx5_flow_handle *rule, u16 rxq)
 568 {
 569         struct mlx5_flow_destination dst;
 570         int err = 0;
 571
 572         dst.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 573         dst.tir_num = priv->direct_tir[rxq].tirn;
 574         err =  mlx5_modify_rule_destination(rule, &dst, NULL);
 575         if (err)
 576                 netdev_warn(priv->netdev,
 577                             "Failed to modfiy aRFS rule destination to rq=%d\n", rxq);
 578 }
 579
 580 static void arfs_handle_work(struct work_struct *work)
 581 {
 582         struct arfs_rule *arfs_rule = container_of(work,
 583                                                    struct arfs_rule,
 584                                                    arfs_work);
 585         struct mlx5e_priv *priv = arfs_rule->priv;
 586         struct mlx5_flow_handle *rule;
 587
 588         mutex_lock(&priv->state_lock);
 589         if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
 590                 spin_lock_bh(&priv->fs.arfs.arfs_lock);
 591                 hlist_del(&arfs_rule->hlist);
 592                 spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 593
 594                 mutex_unlock(&priv->state_lock);
 595                 kfree(arfs_rule);
 596                 goto out;
 597         }
 598         mutex_unlock(&priv->state_lock);
 599
 600         if (!arfs_rule->rule) {
 601                 rule = arfs_add_rule(priv, arfs_rule);
 602                 if (IS_ERR(rule))
 603                         goto out;
 604                 arfs_rule->rule = rule;
 605         } else {
 606                 arfs_modify_rule_rq(priv, arfs_rule->rule,
 607                                     arfs_rule->rxq);
 608         }
 609 out:
 610         arfs_may_expire_flow(priv);
 611 }
 612
 613 /* return L4 destination port from ip4/6 packets */
 614 static __be16 arfs_get_dst_port(const struct sk_buff *skb)
 615 {
 616         char *transport_header;
 617
 618         transport_header = skb_transport_header(skb);
 619         if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
 620                 return ((struct tcphdr *)transport_header)->dest;
 621         return ((struct udphdr *)transport_header)->dest;
 622 }
 623
 624 /* return L4 source port from ip4/6 packets */
 625 static __be16 arfs_get_src_port(const struct sk_buff *skb)
 626 {
 627         char *transport_header;
 628
 629         transport_header = skb_transport_header(skb);
 630         if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
 631                 return ((struct tcphdr *)transport_header)->source;
 632         return ((struct udphdr *)transport_header)->source;
 633 }
 634
 635 static struct arfs_rule *arfs_alloc_rule(struct mlx5e_priv *priv,
 636                                          struct arfs_table *arfs_t,
 637                                          const struct sk_buff *skb,
 638                                          u16 rxq, u32 flow_id)
 639 {
 640         struct arfs_rule *rule;
 641         struct arfs_tuple *tuple;
 642
 643         rule = kzalloc(sizeof(*rule), GFP_ATOMIC);
 644         if (!rule)
 645                 return NULL;
 646
 647         rule->priv = priv;
 648         rule->rxq = rxq;
 649         INIT_WORK(&rule->arfs_work, arfs_handle_work);
 650
 651         tuple = &rule->tuple;
 652         tuple->etype = skb->protocol;
 653         if (tuple->etype == htons(ETH_P_IP)) {
 654                 tuple->src_ipv4 = ip_hdr(skb)->saddr;
 655                 tuple->dst_ipv4 = ip_hdr(skb)->daddr;
 656         } else {
 657                 memcpy(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
 658                        sizeof(struct in6_addr));
 659                 memcpy(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
 660                        sizeof(struct in6_addr));
 661         }
 662         tuple->ip_proto = arfs_get_ip_proto(skb);
 663         tuple->src_port = arfs_get_src_port(skb);
 664         tuple->dst_port = arfs_get_dst_port(skb);
 665
 666         rule->flow_id = flow_id;
 667         rule->filter_id = priv->fs.arfs.last_filter_id++ % RPS_NO_FILTER;
 668
 669         hlist_add_head(&rule->hlist,
 670                        arfs_hash_bucket(arfs_t, tuple->src_port,
 671                                         tuple->dst_port));
 672         return rule;
 673 }
 674
 675 static bool arfs_cmp_ips(struct arfs_tuple *tuple,
 676                          const struct sk_buff *skb)
 677 {
 678         if (tuple->etype == htons(ETH_P_IP) &&
 679             tuple->src_ipv4 == ip_hdr(skb)->saddr &&
 680             tuple->dst_ipv4 == ip_hdr(skb)->daddr)
 681                 return true;
 682         if (tuple->etype == htons(ETH_P_IPV6) &&
 683             (!memcmp(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
 684                      sizeof(struct in6_addr))) &&
 685             (!memcmp(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
 686                      sizeof(struct in6_addr))))
 687                 return true;
 688         return false;
 689 }
 690
 691 static struct arfs_rule *arfs_find_rule(struct arfs_table *arfs_t,
 692                                         const struct sk_buff *skb)
 693 {
 694         struct arfs_rule *arfs_rule;
 695         struct hlist_head *head;
 696         __be16 src_port = arfs_get_src_port(skb);
 697         __be16 dst_port = arfs_get_dst_port(skb);
 698
 699         head = arfs_hash_bucket(arfs_t, src_port, dst_port);
 700         hlist_for_each_entry(arfs_rule, head, hlist) {
 701                 if (arfs_rule->tuple.src_port == src_port &&
 702                     arfs_rule->tuple.dst_port == dst_port &&
 703                     arfs_cmp_ips(&arfs_rule->tuple, skb)) {
 704                         return arfs_rule;
 705                 }
 706         }
 707
 708         return NULL;
 709 }
 710
 711 int mlx5e_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
 712                         u16 rxq_index, u32 flow_id)
 713 {
 714         struct mlx5e_priv *priv = netdev_priv(dev);
 715         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 716         struct arfs_table *arfs_t;
 717         struct arfs_rule *arfs_rule;
 718
 719         if (skb->protocol != htons(ETH_P_IP) &&
 720             skb->protocol != htons(ETH_P_IPV6))
 721                 return -EPROTONOSUPPORT;
 722
 723         arfs_t = arfs_get_table(arfs, arfs_get_ip_proto(skb), skb->protocol);
 724         if (!arfs_t)
 725                 return -EPROTONOSUPPORT;
 726
 727         spin_lock_bh(&arfs->arfs_lock);
 728         arfs_rule = arfs_find_rule(arfs_t, skb);
 729         if (arfs_rule) {
 730                 if (arfs_rule->rxq == rxq_index) {
 731                         spin_unlock_bh(&arfs->arfs_lock);
 732                         return arfs_rule->filter_id;
 733                 }
 734                 arfs_rule->rxq = rxq_index;
 735         } else {
 736                 arfs_rule = arfs_alloc_rule(priv, arfs_t, skb,
 737                                             rxq_index, flow_id);
 738                 if (!arfs_rule) {
 739                         spin_unlock_bh(&arfs->arfs_lock);
 740                         return -ENOMEM;
 741                 }
 742         }
 743         queue_work(priv->fs.arfs.wq, &arfs_rule->arfs_work);
 744         spin_unlock_bh(&arfs->arfs_lock);
 745         return arfs_rule->filter_id;
 746 }
 747 #endif