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         enum mlx5e_traffic_types tt;
 182         int err = 0;
 183
 184         spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
 185         if (!spec) {
 186                 err = -ENOMEM;
 187                 goto out;
 188         }
 189
 190         dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 191         tt = arfs_get_tt(type);
 192         if (tt == -EINVAL) {
 193                 netdev_err(priv->netdev, "%s: bad arfs_type: %d\n",
 194                            __func__, type);
 195                 err = -EINVAL;
 196                 goto out;
 197         }
 198
 199         dest.tir_num = tir[tt].tirn;
 200
 201         arfs_t->default_rule = mlx5_add_flow_rules(arfs_t->ft.t, spec,
 202                                                    &flow_act,
 203                                                    &dest, 1);
 204         if (IS_ERR(arfs_t->default_rule)) {
 205                 err = PTR_ERR(arfs_t->default_rule);
 206                 arfs_t->default_rule = NULL;
 207                 netdev_err(priv->netdev, "%s: add rule failed, arfs type=%d\n",
 208                            __func__, type);
 209         }
 210 out:
 211         kvfree(spec);
 212         return err;
 213 }
 214
 215 #define MLX5E_ARFS_NUM_GROUPS   2
 216 #define MLX5E_ARFS_GROUP1_SIZE  (BIT(16) - 1)
 217 #define MLX5E_ARFS_GROUP2_SIZE  BIT(0)
 218 #define MLX5E_ARFS_TABLE_SIZE   (MLX5E_ARFS_GROUP1_SIZE +\
 219                                  MLX5E_ARFS_GROUP2_SIZE)
 220 static int arfs_create_groups(struct mlx5e_flow_table *ft,
 221                               enum  arfs_type type)
 222 {
 223         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 224         void *outer_headers_c;
 225         int ix = 0;
 226         u32 *in;
 227         int err;
 228         u8 *mc;
 229
 230         ft->g = kcalloc(MLX5E_ARFS_NUM_GROUPS,
 231                         sizeof(*ft->g), GFP_KERNEL);
 232         in = kvzalloc(inlen, GFP_KERNEL);
 233         if  (!in || !ft->g) {
 234                 kvfree(ft->g);
 235                 kvfree(in);
 236                 return -ENOMEM;
 237         }
 238
 239         mc = MLX5_ADDR_OF(create_flow_group_in, in, match_criteria);
 240         outer_headers_c = MLX5_ADDR_OF(fte_match_param, mc,
 241                                        outer_headers);
 242         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, ethertype);
 243         switch (type) {
 244         case ARFS_IPV4_TCP:
 245         case ARFS_IPV6_TCP:
 246                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_dport);
 247                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_sport);
 248                 break;
 249         case ARFS_IPV4_UDP:
 250         case ARFS_IPV6_UDP:
 251                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_dport);
 252                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_sport);
 253                 break;
 254         default:
 255                 err = -EINVAL;
 256                 goto out;
 257         }
 258
 259         switch (type) {
 260         case ARFS_IPV4_TCP:
 261         case ARFS_IPV4_UDP:
 262                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
 263                                  src_ipv4_src_ipv6.ipv4_layout.ipv4);
 264                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
 265                                  dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
 266                 break;
 267         case ARFS_IPV6_TCP:
 268         case ARFS_IPV6_UDP:
 269                 memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
 270                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 271                        0xff, 16);
 272                 memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
 273                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 274                        0xff, 16);
 275                 break;
 276         default:
 277                 err = -EINVAL;
 278                 goto out;
 279         }
 280
 281         MLX5_SET_CFG(in, match_criteria_enable, MLX5_MATCH_OUTER_HEADERS);
 282         MLX5_SET_CFG(in, start_flow_index, ix);
 283         ix += MLX5E_ARFS_GROUP1_SIZE;
 284         MLX5_SET_CFG(in, end_flow_index, ix - 1);
 285         ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
 286         if (IS_ERR(ft->g[ft->num_groups]))
 287                 goto err;
 288         ft->num_groups++;
 289
 290         memset(in, 0, inlen);
 291         MLX5_SET_CFG(in, start_flow_index, ix);
 292         ix += MLX5E_ARFS_GROUP2_SIZE;
 293         MLX5_SET_CFG(in, end_flow_index, ix - 1);
 294         ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
 295         if (IS_ERR(ft->g[ft->num_groups]))
 296                 goto err;
 297         ft->num_groups++;
 298
 299         kvfree(in);
 300         return 0;
 301
 302 err:
 303         err = PTR_ERR(ft->g[ft->num_groups]);
 304         ft->g[ft->num_groups] = NULL;
 305 out:
 306         kvfree(in);
 307
 308         return err;
 309 }
 310
 311 static int arfs_create_table(struct mlx5e_priv *priv,
 312                              enum arfs_type type)
 313 {
 314         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 315         struct mlx5e_flow_table *ft = &arfs->arfs_tables[type].ft;
 316         struct mlx5_flow_table_attr ft_attr = {};
 317         int err;
 318
 319         ft->num_groups = 0;
 320
 321         ft_attr.max_fte = MLX5E_ARFS_TABLE_SIZE;
 322         ft_attr.level = MLX5E_ARFS_FT_LEVEL;
 323         ft_attr.prio = MLX5E_NIC_PRIO;
 324
 325         ft->t = mlx5_create_flow_table(priv->fs.ns, &ft_attr);
 326         if (IS_ERR(ft->t)) {
 327                 err = PTR_ERR(ft->t);
 328                 ft->t = NULL;
 329                 return err;
 330         }
 331
 332         err = arfs_create_groups(ft, type);
 333         if (err)
 334                 goto err;
 335
 336         err = arfs_add_default_rule(priv, type);
 337         if (err)
 338                 goto err;
 339
 340         return 0;
 341 err:
 342         mlx5e_destroy_flow_table(ft);
 343         return err;
 344 }
 345
 346 int mlx5e_arfs_create_tables(struct mlx5e_priv *priv)
 347 {
 348         int err = 0;
 349         int i;
 350
 351         if (!(priv->netdev->hw_features & NETIF_F_NTUPLE))
 352                 return 0;
 353
 354         spin_lock_init(&priv->fs.arfs.arfs_lock);
 355         INIT_LIST_HEAD(&priv->fs.arfs.rules);
 356         priv->fs.arfs.wq = create_singlethread_workqueue("mlx5e_arfs");
 357         if (!priv->fs.arfs.wq)
 358                 return -ENOMEM;
 359
 360         for (i = 0; i < ARFS_NUM_TYPES; i++) {
 361                 err = arfs_create_table(priv, i);
 362                 if (err)
 363                         goto err;
 364         }
 365         return 0;
 366 err:
 367         mlx5e_arfs_destroy_tables(priv);
 368         return err;
 369 }
 370
 371 #define MLX5E_ARFS_EXPIRY_QUOTA 60
 372
 373 static void arfs_may_expire_flow(struct mlx5e_priv *priv)
 374 {
 375         struct arfs_rule *arfs_rule;
 376         struct hlist_node *htmp;
 377         int quota = 0;
 378         int i;
 379         int j;
 380
 381         HLIST_HEAD(del_list);
 382         spin_lock_bh(&priv->fs.arfs.arfs_lock);
 383         mlx5e_for_each_arfs_rule(arfs_rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
 384                 if (quota++ > MLX5E_ARFS_EXPIRY_QUOTA)
 385                         break;
 386                 if (!work_pending(&arfs_rule->arfs_work) &&
 387                     rps_may_expire_flow(priv->netdev,
 388                                         arfs_rule->rxq, arfs_rule->flow_id,
 389                                         arfs_rule->filter_id)) {
 390                         hlist_del_init(&arfs_rule->hlist);
 391                         hlist_add_head(&arfs_rule->hlist, &del_list);
 392                 }
 393         }
 394         spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 395         hlist_for_each_entry_safe(arfs_rule, htmp, &del_list, hlist) {
 396                 if (arfs_rule->rule)
 397                         mlx5_del_flow_rules(arfs_rule->rule);
 398                 hlist_del(&arfs_rule->hlist);
 399                 kfree(arfs_rule);
 400         }
 401 }
 402
 403 static void arfs_del_rules(struct mlx5e_priv *priv)
 404 {
 405         struct hlist_node *htmp;
 406         struct arfs_rule *rule;
 407         int i;
 408         int j;
 409
 410         HLIST_HEAD(del_list);
 411         spin_lock_bh(&priv->fs.arfs.arfs_lock);
 412         mlx5e_for_each_arfs_rule(rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
 413                 hlist_del_init(&rule->hlist);
 414                 hlist_add_head(&rule->hlist, &del_list);
 415         }
 416         spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 417
 418         hlist_for_each_entry_safe(rule, htmp, &del_list, hlist) {
 419                 cancel_work_sync(&rule->arfs_work);
 420                 if (rule->rule)
 421                         mlx5_del_flow_rules(rule->rule);
 422                 hlist_del(&rule->hlist);
 423                 kfree(rule);
 424         }
 425 }
 426
 427 static struct hlist_head *
 428 arfs_hash_bucket(struct arfs_table *arfs_t, __be16 src_port,
 429                  __be16 dst_port)
 430 {
 431         unsigned long l;
 432         int bucket_idx;
 433
 434         l = (__force unsigned long)src_port |
 435             ((__force unsigned long)dst_port << 2);
 436
 437         bucket_idx = hash_long(l, ARFS_HASH_SHIFT);
 438
 439         return &arfs_t->rules_hash[bucket_idx];
 440 }
 441
 442 static u8 arfs_get_ip_proto(const struct sk_buff *skb)
 443 {
 444         return (skb->protocol == htons(ETH_P_IP)) ?
 445                 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
 446 }
 447
 448 static struct arfs_table *arfs_get_table(struct mlx5e_arfs_tables *arfs,
 449                                          u8 ip_proto, __be16 etype)
 450 {
 451         if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_TCP)
 452                 return &arfs->arfs_tables[ARFS_IPV4_TCP];
 453         if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_UDP)
 454                 return &arfs->arfs_tables[ARFS_IPV4_UDP];
 455         if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_TCP)
 456                 return &arfs->arfs_tables[ARFS_IPV6_TCP];
 457         if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_UDP)
 458                 return &arfs->arfs_tables[ARFS_IPV6_UDP];
 459
 460         return NULL;
 461 }
 462
 463 static struct mlx5_flow_handle *arfs_add_rule(struct mlx5e_priv *priv,
 464                                               struct arfs_rule *arfs_rule)
 465 {
 466         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 467         struct arfs_tuple *tuple = &arfs_rule->tuple;
 468         struct mlx5_flow_handle *rule = NULL;
 469         struct mlx5_flow_destination dest = {};
 470         MLX5_DECLARE_FLOW_ACT(flow_act);
 471         struct arfs_table *arfs_table;
 472         struct mlx5_flow_spec *spec;
 473         struct mlx5_flow_table *ft;
 474         int err = 0;
 475
 476         spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
 477         if (!spec) {
 478                 err = -ENOMEM;
 479                 goto out;
 480         }
 481         spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
 482         MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 483                          outer_headers.ethertype);
 484         MLX5_SET(fte_match_param, spec->match_value, outer_headers.ethertype,
 485                  ntohs(tuple->etype));
 486         arfs_table = arfs_get_table(arfs, tuple->ip_proto, tuple->etype);
 487         if (!arfs_table) {
 488                 err = -EINVAL;
 489                 goto out;
 490         }
 491
 492         ft = arfs_table->ft.t;
 493         if (tuple->ip_proto == IPPROTO_TCP) {
 494                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 495                                  outer_headers.tcp_dport);
 496                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 497                                  outer_headers.tcp_sport);
 498                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_dport,
 499                          ntohs(tuple->dst_port));
 500                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_sport,
 501                          ntohs(tuple->src_port));
 502         } else {
 503                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 504                                  outer_headers.udp_dport);
 505                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 506                                  outer_headers.udp_sport);
 507                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_dport,
 508                          ntohs(tuple->dst_port));
 509                 MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_sport,
 510                          ntohs(tuple->src_port));
 511         }
 512         if (tuple->etype == htons(ETH_P_IP)) {
 513                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 514                                     outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4),
 515                        &tuple->src_ipv4,
 516                        4);
 517                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 518                                     outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 519                        &tuple->dst_ipv4,
 520                        4);
 521                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 522                                  outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4);
 523                 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
 524                                  outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
 525         } else {
 526                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 527                                     outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
 528                        &tuple->src_ipv6,
 529                        16);
 530                 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
 531                                     outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 532                        &tuple->dst_ipv6,
 533                        16);
 534                 memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 535                                     outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
 536                        0xff,
 537                        16);
 538                 memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 539                                     outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 540                        0xff,
 541                        16);
 542         }
 543         dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 544         dest.tir_num = priv->direct_tir[arfs_rule->rxq].tirn;
 545         rule = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
 546         if (IS_ERR(rule)) {
 547                 err = PTR_ERR(rule);
 548                 netdev_err(priv->netdev, "%s: add rule(filter id=%d, rq idx=%d) failed, err=%d\n",
 549                            __func__, arfs_rule->filter_id, arfs_rule->rxq, err);
 550         }
 551
 552 out:
 553         kvfree(spec);
 554         return err ? ERR_PTR(err) : rule;
 555 }
 556
 557 static void arfs_modify_rule_rq(struct mlx5e_priv *priv,
 558                                 struct mlx5_flow_handle *rule, u16 rxq)
 559 {
 560         struct mlx5_flow_destination dst = {};
 561         int err = 0;
 562
 563         dst.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 564         dst.tir_num = priv->direct_tir[rxq].tirn;
 565         err =  mlx5_modify_rule_destination(rule, &dst, NULL);
 566         if (err)
 567                 netdev_warn(priv->netdev,
 568                             "Failed to modify aRFS rule destination to rq=%d\n", rxq);
 569 }
 570
 571 static void arfs_handle_work(struct work_struct *work)
 572 {
 573         struct arfs_rule *arfs_rule = container_of(work,
 574                                                    struct arfs_rule,
 575                                                    arfs_work);
 576         struct mlx5e_priv *priv = arfs_rule->priv;
 577         struct mlx5_flow_handle *rule;
 578
 579         mutex_lock(&priv->state_lock);
 580         if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
 581                 spin_lock_bh(&priv->fs.arfs.arfs_lock);
 582                 hlist_del(&arfs_rule->hlist);
 583                 spin_unlock_bh(&priv->fs.arfs.arfs_lock);
 584
 585                 mutex_unlock(&priv->state_lock);
 586                 kfree(arfs_rule);
 587                 goto out;
 588         }
 589         mutex_unlock(&priv->state_lock);
 590
 591         if (!arfs_rule->rule) {
 592                 rule = arfs_add_rule(priv, arfs_rule);
 593                 if (IS_ERR(rule))
 594                         goto out;
 595                 arfs_rule->rule = rule;
 596         } else {
 597                 arfs_modify_rule_rq(priv, arfs_rule->rule,
 598                                     arfs_rule->rxq);
 599         }
 600 out:
 601         arfs_may_expire_flow(priv);
 602 }
 603
 604 /* return L4 destination port from ip4/6 packets */
 605 static __be16 arfs_get_dst_port(const struct sk_buff *skb)
 606 {
 607         char *transport_header;
 608
 609         transport_header = skb_transport_header(skb);
 610         if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
 611                 return ((struct tcphdr *)transport_header)->dest;
 612         return ((struct udphdr *)transport_header)->dest;
 613 }
 614
 615 /* return L4 source port from ip4/6 packets */
 616 static __be16 arfs_get_src_port(const struct sk_buff *skb)
 617 {
 618         char *transport_header;
 619
 620         transport_header = skb_transport_header(skb);
 621         if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
 622                 return ((struct tcphdr *)transport_header)->source;
 623         return ((struct udphdr *)transport_header)->source;
 624 }
 625
 626 static struct arfs_rule *arfs_alloc_rule(struct mlx5e_priv *priv,
 627                                          struct arfs_table *arfs_t,
 628                                          const struct sk_buff *skb,
 629                                          u16 rxq, u32 flow_id)
 630 {
 631         struct arfs_rule *rule;
 632         struct arfs_tuple *tuple;
 633
 634         rule = kzalloc(sizeof(*rule), GFP_ATOMIC);
 635         if (!rule)
 636                 return NULL;
 637
 638         rule->priv = priv;
 639         rule->rxq = rxq;
 640         INIT_WORK(&rule->arfs_work, arfs_handle_work);
 641
 642         tuple = &rule->tuple;
 643         tuple->etype = skb->protocol;
 644         if (tuple->etype == htons(ETH_P_IP)) {
 645                 tuple->src_ipv4 = ip_hdr(skb)->saddr;
 646                 tuple->dst_ipv4 = ip_hdr(skb)->daddr;
 647         } else {
 648                 memcpy(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
 649                        sizeof(struct in6_addr));
 650                 memcpy(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
 651                        sizeof(struct in6_addr));
 652         }
 653         tuple->ip_proto = arfs_get_ip_proto(skb);
 654         tuple->src_port = arfs_get_src_port(skb);
 655         tuple->dst_port = arfs_get_dst_port(skb);
 656
 657         rule->flow_id = flow_id;
 658         rule->filter_id = priv->fs.arfs.last_filter_id++ % RPS_NO_FILTER;
 659
 660         hlist_add_head(&rule->hlist,
 661                        arfs_hash_bucket(arfs_t, tuple->src_port,
 662                                         tuple->dst_port));
 663         return rule;
 664 }
 665
 666 static bool arfs_cmp_ips(struct arfs_tuple *tuple,
 667                          const struct sk_buff *skb)
 668 {
 669         if (tuple->etype == htons(ETH_P_IP) &&
 670             tuple->src_ipv4 == ip_hdr(skb)->saddr &&
 671             tuple->dst_ipv4 == ip_hdr(skb)->daddr)
 672                 return true;
 673         if (tuple->etype == htons(ETH_P_IPV6) &&
 674             (!memcmp(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
 675                      sizeof(struct in6_addr))) &&
 676             (!memcmp(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
 677                      sizeof(struct in6_addr))))
 678                 return true;
 679         return false;
 680 }
 681
 682 static struct arfs_rule *arfs_find_rule(struct arfs_table *arfs_t,
 683                                         const struct sk_buff *skb)
 684 {
 685         struct arfs_rule *arfs_rule;
 686         struct hlist_head *head;
 687         __be16 src_port = arfs_get_src_port(skb);
 688         __be16 dst_port = arfs_get_dst_port(skb);
 689
 690         head = arfs_hash_bucket(arfs_t, src_port, dst_port);
 691         hlist_for_each_entry(arfs_rule, head, hlist) {
 692                 if (arfs_rule->tuple.src_port == src_port &&
 693                     arfs_rule->tuple.dst_port == dst_port &&
 694                     arfs_cmp_ips(&arfs_rule->tuple, skb)) {
 695                         return arfs_rule;
 696                 }
 697         }
 698
 699         return NULL;
 700 }
 701
 702 int mlx5e_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
 703                         u16 rxq_index, u32 flow_id)
 704 {
 705         struct mlx5e_priv *priv = netdev_priv(dev);
 706         struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
 707         struct arfs_table *arfs_t;
 708         struct arfs_rule *arfs_rule;
 709
 710         if (skb->protocol != htons(ETH_P_IP) &&
 711             skb->protocol != htons(ETH_P_IPV6))
 712                 return -EPROTONOSUPPORT;
 713
 714         arfs_t = arfs_get_table(arfs, arfs_get_ip_proto(skb), skb->protocol);
 715         if (!arfs_t)
 716                 return -EPROTONOSUPPORT;
 717
 718         spin_lock_bh(&arfs->arfs_lock);
 719         arfs_rule = arfs_find_rule(arfs_t, skb);
 720         if (arfs_rule) {
 721                 if (arfs_rule->rxq == rxq_index) {
 722                         spin_unlock_bh(&arfs->arfs_lock);
 723                         return arfs_rule->filter_id;
 724                 }
 725                 arfs_rule->rxq = rxq_index;
 726         } else {
 727                 arfs_rule = arfs_alloc_rule(priv, arfs_t, skb,
 728                                             rxq_index, flow_id);
 729                 if (!arfs_rule) {
 730                         spin_unlock_bh(&arfs->arfs_lock);
 731                         return -ENOMEM;
 732                 }
 733         }
 734         queue_work(priv->fs.arfs.wq, &arfs_rule->arfs_work);
 735         spin_unlock_bh(&arfs->arfs_lock);
 736         return arfs_rule->filter_id;
 737 }
 738 #endif