drivers/net/ethernet/mellanox/mlx5/core/en_tc.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 #include <net/flow_dissector.h>
  34 #include <net/sch_generic.h>
  35 #include <net/pkt_cls.h>
  36 #include <net/tc_act/tc_gact.h>
  37 #include <net/tc_act/tc_skbedit.h>
  38 #include <linux/mlx5/fs.h>
  39 #include <linux/mlx5/device.h>
  40 #include <linux/rhashtable.h>
  41 #include <net/switchdev.h>
  42 #include <net/tc_act/tc_mirred.h>
  43 #include <net/tc_act/tc_vlan.h>
  44 #include <net/tc_act/tc_tunnel_key.h>
  45 #include <net/tc_act/tc_pedit.h>
  46 #include <net/tc_act/tc_csum.h>
  47 #include <net/vxlan.h>
  48 #include <net/arp.h>
  49 #include "en.h"
  50 #include "en_rep.h"
  51 #include "en_tc.h"
  52 #include "eswitch.h"
  53 #include "vxlan.h"
  54
  55 struct mlx5_nic_flow_attr {
  56         u32 action;
  57         u32 flow_tag;
  58         u32 mod_hdr_id;
  59 };
  60
  61 enum {
  62         MLX5E_TC_FLOW_ESWITCH   = BIT(0),
  63         MLX5E_TC_FLOW_NIC       = BIT(1),
  64         MLX5E_TC_FLOW_OFFLOADED = BIT(2),
  65 };
  66
  67 struct mlx5e_tc_flow {
  68         struct rhash_head       node;
  69         u64                     cookie;
  70         u8                      flags;
  71         struct mlx5_flow_handle *rule;
  72         struct list_head        encap;   /* flows sharing the same encap ID */
  73         struct list_head        mod_hdr; /* flows sharing the same mod hdr ID */
  74         union {
  75                 struct mlx5_esw_flow_attr esw_attr[0];
  76                 struct mlx5_nic_flow_attr nic_attr[0];
  77         };
  78 };
  79
  80 struct mlx5e_tc_flow_parse_attr {
  81         struct ip_tunnel_info tun_info;
  82         struct mlx5_flow_spec spec;
  83         int num_mod_hdr_actions;
  84         void *mod_hdr_actions;
  85         int mirred_ifindex;
  86 };
  87
  88 enum {
  89         MLX5_HEADER_TYPE_VXLAN = 0x0,
  90         MLX5_HEADER_TYPE_NVGRE = 0x1,
  91 };
  92
  93 #define MLX5E_TC_TABLE_NUM_ENTRIES 1024
  94 #define MLX5E_TC_TABLE_NUM_GROUPS 4
  95
  96 struct mod_hdr_key {
  97         int num_actions;
  98         void *actions;
  99 };
 100
 101 struct mlx5e_mod_hdr_entry {
 102         /* a node of a hash table which keeps all the mod_hdr entries */
 103         struct hlist_node mod_hdr_hlist;
 104
 105         /* flows sharing the same mod_hdr entry */
 106         struct list_head flows;
 107
 108         struct mod_hdr_key key;
 109
 110         u32 mod_hdr_id;
 111 };
 112
 113 #define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)
 114
 115 static inline u32 hash_mod_hdr_info(struct mod_hdr_key *key)
 116 {
 117         return jhash(key->actions,
 118                      key->num_actions * MLX5_MH_ACT_SZ, 0);
 119 }
 120
 121 static inline int cmp_mod_hdr_info(struct mod_hdr_key *a,
 122                                    struct mod_hdr_key *b)
 123 {
 124         if (a->num_actions != b->num_actions)
 125                 return 1;
 126
 127         return memcmp(a->actions, b->actions, a->num_actions * MLX5_MH_ACT_SZ);
 128 }
 129
 130 static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
 131                                 struct mlx5e_tc_flow *flow,
 132                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
 133 {
 134         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 135         int num_actions, actions_size, namespace, err;
 136         struct mlx5e_mod_hdr_entry *mh;
 137         struct mod_hdr_key key;
 138         bool found = false;
 139         u32 hash_key;
 140
 141         num_actions  = parse_attr->num_mod_hdr_actions;
 142         actions_size = MLX5_MH_ACT_SZ * num_actions;
 143
 144         key.actions = parse_attr->mod_hdr_actions;
 145         key.num_actions = num_actions;
 146
 147         hash_key = hash_mod_hdr_info(&key);
 148
 149         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
 150                 namespace = MLX5_FLOW_NAMESPACE_FDB;
 151                 hash_for_each_possible(esw->offloads.mod_hdr_tbl, mh,
 152                                        mod_hdr_hlist, hash_key) {
 153                         if (!cmp_mod_hdr_info(&mh->key, &key)) {
 154                                 found = true;
 155                                 break;
 156                         }
 157                 }
 158         } else {
 159                 namespace = MLX5_FLOW_NAMESPACE_KERNEL;
 160                 hash_for_each_possible(priv->fs.tc.mod_hdr_tbl, mh,
 161                                        mod_hdr_hlist, hash_key) {
 162                         if (!cmp_mod_hdr_info(&mh->key, &key)) {
 163                                 found = true;
 164                                 break;
 165                         }
 166                 }
 167         }
 168
 169         if (found)
 170                 goto attach_flow;
 171
 172         mh = kzalloc(sizeof(*mh) + actions_size, GFP_KERNEL);
 173         if (!mh)
 174                 return -ENOMEM;
 175
 176         mh->key.actions = (void *)mh + sizeof(*mh);
 177         memcpy(mh->key.actions, key.actions, actions_size);
 178         mh->key.num_actions = num_actions;
 179         INIT_LIST_HEAD(&mh->flows);
 180
 181         err = mlx5_modify_header_alloc(priv->mdev, namespace,
 182                                        mh->key.num_actions,
 183                                        mh->key.actions,
 184                                        &mh->mod_hdr_id);
 185         if (err)
 186                 goto out_err;
 187
 188         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
 189                 hash_add(esw->offloads.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
 190         else
 191                 hash_add(priv->fs.tc.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
 192
 193 attach_flow:
 194         list_add(&flow->mod_hdr, &mh->flows);
 195         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
 196                 flow->esw_attr->mod_hdr_id = mh->mod_hdr_id;
 197         else
 198                 flow->nic_attr->mod_hdr_id = mh->mod_hdr_id;
 199
 200         return 0;
 201
 202 out_err:
 203         kfree(mh);
 204         return err;
 205 }
 206
 207 static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
 208                                  struct mlx5e_tc_flow *flow)
 209 {
 210         struct list_head *next = flow->mod_hdr.next;
 211
 212         list_del(&flow->mod_hdr);
 213
 214         if (list_empty(next)) {
 215                 struct mlx5e_mod_hdr_entry *mh;
 216
 217                 mh = list_entry(next, struct mlx5e_mod_hdr_entry, flows);
 218
 219                 mlx5_modify_header_dealloc(priv->mdev, mh->mod_hdr_id);
 220                 hash_del(&mh->mod_hdr_hlist);
 221                 kfree(mh);
 222         }
 223 }
 224
 225 static struct mlx5_flow_handle *
 226 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
 227                       struct mlx5e_tc_flow_parse_attr *parse_attr,
 228                       struct mlx5e_tc_flow *flow)
 229 {
 230         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
 231         struct mlx5_core_dev *dev = priv->mdev;
 232         struct mlx5_flow_destination dest = {};
 233         struct mlx5_flow_act flow_act = {
 234                 .action = attr->action,
 235                 .flow_tag = attr->flow_tag,
 236                 .encap_id = 0,
 237         };
 238         struct mlx5_fc *counter = NULL;
 239         struct mlx5_flow_handle *rule;
 240         bool table_created = false;
 241         int err;
 242
 243         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
 244                 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 245                 dest.ft = priv->fs.vlan.ft.t;
 246         } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
 247                 counter = mlx5_fc_create(dev, true);
 248                 if (IS_ERR(counter))
 249                         return ERR_CAST(counter);
 250
 251                 dest.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
 252                 dest.counter = counter;
 253         }
 254
 255         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
 256                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
 257                 flow_act.modify_id = attr->mod_hdr_id;
 258                 kfree(parse_attr->mod_hdr_actions);
 259                 if (err) {
 260                         rule = ERR_PTR(err);
 261                         goto err_create_mod_hdr_id;
 262                 }
 263         }
 264
 265         if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
 266                 priv->fs.tc.t =
 267                         mlx5_create_auto_grouped_flow_table(priv->fs.ns,
 268                                                             MLX5E_TC_PRIO,
 269                                                             MLX5E_TC_TABLE_NUM_ENTRIES,
 270                                                             MLX5E_TC_TABLE_NUM_GROUPS,
 271                                                             0, 0);
 272                 if (IS_ERR(priv->fs.tc.t)) {
 273                         netdev_err(priv->netdev,
 274                                    "Failed to create tc offload table\n");
 275                         rule = ERR_CAST(priv->fs.tc.t);
 276                         goto err_create_ft;
 277                 }
 278
 279                 table_created = true;
 280         }
 281
 282         parse_attr->spec.match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
 283         rule = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
 284                                    &flow_act, &dest, 1);
 285
 286         if (IS_ERR(rule))
 287                 goto err_add_rule;
 288
 289         return rule;
 290
 291 err_add_rule:
 292         if (table_created) {
 293                 mlx5_destroy_flow_table(priv->fs.tc.t);
 294                 priv->fs.tc.t = NULL;
 295         }
 296 err_create_ft:
 297         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 298                 mlx5e_detach_mod_hdr(priv, flow);
 299 err_create_mod_hdr_id:
 300         mlx5_fc_destroy(dev, counter);
 301
 302         return rule;
 303 }
 304
 305 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
 306                                   struct mlx5e_tc_flow *flow)
 307 {
 308         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
 309         struct mlx5_fc *counter = NULL;
 310
 311         counter = mlx5_flow_rule_counter(flow->rule);
 312         mlx5_del_flow_rules(flow->rule);
 313         mlx5_fc_destroy(priv->mdev, counter);
 314
 315         if (!mlx5e_tc_num_filters(priv) && (priv->fs.tc.t)) {
 316                 mlx5_destroy_flow_table(priv->fs.tc.t);
 317                 priv->fs.tc.t = NULL;
 318         }
 319
 320         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 321                 mlx5e_detach_mod_hdr(priv, flow);
 322 }
 323
 324 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
 325                                struct mlx5e_tc_flow *flow);
 326
 327 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
 328                               struct ip_tunnel_info *tun_info,
 329                               struct net_device *mirred_dev,
 330                               struct net_device **encap_dev,
 331                               struct mlx5e_tc_flow *flow);
 332
 333 static struct mlx5_flow_handle *
 334 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
 335                       struct mlx5e_tc_flow_parse_attr *parse_attr,
 336                       struct mlx5e_tc_flow *flow)
 337 {
 338         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 339         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
 340         struct net_device *out_dev, *encap_dev = NULL;
 341         struct mlx5_flow_handle *rule = NULL;
 342         struct mlx5e_rep_priv *rpriv;
 343         struct mlx5e_priv *out_priv;
 344         int err;
 345
 346         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP) {
 347                 out_dev = __dev_get_by_index(dev_net(priv->netdev),
 348                                              attr->parse_attr->mirred_ifindex);
 349                 err = mlx5e_attach_encap(priv, &parse_attr->tun_info,
 350                                          out_dev, &encap_dev, flow);
 351                 if (err) {
 352                         rule = ERR_PTR(err);
 353                         if (err != -EAGAIN)
 354                                 goto err_attach_encap;
 355                 }
 356                 out_priv = netdev_priv(encap_dev);
 357                 rpriv = out_priv->ppriv;
 358                 attr->out_rep = rpriv->rep;
 359         }
 360
 361         err = mlx5_eswitch_add_vlan_action(esw, attr);
 362         if (err) {
 363                 rule = ERR_PTR(err);
 364                 goto err_add_vlan;
 365         }
 366
 367         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
 368                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
 369                 kfree(parse_attr->mod_hdr_actions);
 370                 if (err) {
 371                         rule = ERR_PTR(err);
 372                         goto err_mod_hdr;
 373                 }
 374         }
 375
 376         /* we get here if (1) there's no error (rule being null) or when
 377          * (2) there's an encap action and we're on -EAGAIN (no valid neigh)
 378          */
 379         if (rule != ERR_PTR(-EAGAIN)) {
 380                 rule = mlx5_eswitch_add_offloaded_rule(esw, &parse_attr->spec, attr);
 381                 if (IS_ERR(rule))
 382                         goto err_add_rule;
 383         }
 384         return rule;
 385
 386 err_add_rule:
 387         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 388                 mlx5e_detach_mod_hdr(priv, flow);
 389 err_mod_hdr:
 390         mlx5_eswitch_del_vlan_action(esw, attr);
 391 err_add_vlan:
 392         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP)
 393                 mlx5e_detach_encap(priv, flow);
 394 err_attach_encap:
 395         return rule;
 396 }
 397
 398 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
 399                                   struct mlx5e_tc_flow *flow)
 400 {
 401         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 402         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
 403
 404         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
 405                 flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
 406                 mlx5_eswitch_del_offloaded_rule(esw, flow->rule, attr);
 407         }
 408
 409         mlx5_eswitch_del_vlan_action(esw, attr);
 410
 411         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP) {
 412                 mlx5e_detach_encap(priv, flow);
 413                 kvfree(attr->parse_attr);
 414         }
 415
 416         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 417                 mlx5e_detach_mod_hdr(priv, flow);
 418 }
 419
 420 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
 421                               struct mlx5e_encap_entry *e)
 422 {
 423         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 424         struct mlx5_esw_flow_attr *esw_attr;
 425         struct mlx5e_tc_flow *flow;
 426         int err;
 427
 428         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
 429                                e->encap_size, e->encap_header,
 430                                &e->encap_id);
 431         if (err) {
 432                 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %d\n",
 433                                err);
 434                 return;
 435         }
 436         e->flags |= MLX5_ENCAP_ENTRY_VALID;
 437         mlx5e_rep_queue_neigh_stats_work(priv);
 438
 439         list_for_each_entry(flow, &e->flows, encap) {
 440                 esw_attr = flow->esw_attr;
 441                 esw_attr->encap_id = e->encap_id;
 442                 flow->rule = mlx5_eswitch_add_offloaded_rule(esw, &esw_attr->parse_attr->spec, esw_attr);
 443                 if (IS_ERR(flow->rule)) {
 444                         err = PTR_ERR(flow->rule);
 445                         mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
 446                                        err);
 447                         continue;
 448                 }
 449                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
 450         }
 451 }
 452
 453 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
 454                               struct mlx5e_encap_entry *e)
 455 {
 456         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 457         struct mlx5e_tc_flow *flow;
 458
 459         list_for_each_entry(flow, &e->flows, encap) {
 460                 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
 461                         flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
 462                         mlx5_eswitch_del_offloaded_rule(esw, flow->rule, flow->esw_attr);
 463                 }
 464         }
 465
 466         if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
 467                 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
 468                 mlx5_encap_dealloc(priv->mdev, e->encap_id);
 469         }
 470 }
 471
 472 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
 473 {
 474         struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
 475         u64 bytes, packets, lastuse = 0;
 476         struct mlx5e_tc_flow *flow;
 477         struct mlx5e_encap_entry *e;
 478         struct mlx5_fc *counter;
 479         struct neigh_table *tbl;
 480         bool neigh_used = false;
 481         struct neighbour *n;
 482
 483         if (m_neigh->family == AF_INET)
 484                 tbl = &arp_tbl;
 485 #if IS_ENABLED(CONFIG_IPV6)
 486         else if (m_neigh->family == AF_INET6)
 487                 tbl = ipv6_stub->nd_tbl;
 488 #endif
 489         else
 490                 return;
 491
 492         list_for_each_entry(e, &nhe->encap_list, encap_list) {
 493                 if (!(e->flags & MLX5_ENCAP_ENTRY_VALID))
 494                         continue;
 495                 list_for_each_entry(flow, &e->flows, encap) {
 496                         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
 497                                 counter = mlx5_flow_rule_counter(flow->rule);
 498                                 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
 499                                 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
 500                                         neigh_used = true;
 501                                         break;
 502                                 }
 503                         }
 504                 }
 505         }
 506
 507         if (neigh_used) {
 508                 nhe->reported_lastuse = jiffies;
 509
 510                 /* find the relevant neigh according to the cached device and
 511                  * dst ip pair
 512                  */
 513                 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
 514                 if (!n) {
 515                         WARN(1, "The neighbour already freed\n");
 516                         return;
 517                 }
 518
 519                 neigh_event_send(n, NULL);
 520                 neigh_release(n);
 521         }
 522 }
 523
 524 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
 525                                struct mlx5e_tc_flow *flow)
 526 {
 527         struct list_head *next = flow->encap.next;
 528
 529         list_del(&flow->encap);
 530         if (list_empty(next)) {
 531                 struct mlx5e_encap_entry *e;
 532
 533                 e = list_entry(next, struct mlx5e_encap_entry, flows);
 534                 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
 535
 536                 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
 537                         mlx5_encap_dealloc(priv->mdev, e->encap_id);
 538
 539                 hash_del_rcu(&e->encap_hlist);
 540                 kfree(e->encap_header);
 541                 kfree(e);
 542         }
 543 }
 544
 545 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
 546                               struct mlx5e_tc_flow *flow)
 547 {
 548         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
 549                 mlx5e_tc_del_fdb_flow(priv, flow);
 550         else
 551                 mlx5e_tc_del_nic_flow(priv, flow);
 552 }
 553
 554 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
 555                              struct tc_cls_flower_offload *f)
 556 {
 557         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 558                                        outer_headers);
 559         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 560                                        outer_headers);
 561         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 562                                     misc_parameters);
 563         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 564                                     misc_parameters);
 565
 566         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
 567         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
 568
 569         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
 570                 struct flow_dissector_key_keyid *key =
 571                         skb_flow_dissector_target(f->dissector,
 572                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
 573                                                   f->key);
 574                 struct flow_dissector_key_keyid *mask =
 575                         skb_flow_dissector_target(f->dissector,
 576                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
 577                                                   f->mask);
 578                 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
 579                          be32_to_cpu(mask->keyid));
 580                 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
 581                          be32_to_cpu(key->keyid));
 582         }
 583 }
 584
 585 static int parse_tunnel_attr(struct mlx5e_priv *priv,
 586                              struct mlx5_flow_spec *spec,
 587                              struct tc_cls_flower_offload *f)
 588 {
 589         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 590                                        outer_headers);
 591         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 592                                        outer_headers);
 593
 594         struct flow_dissector_key_control *enc_control =
 595                 skb_flow_dissector_target(f->dissector,
 596                                           FLOW_DISSECTOR_KEY_ENC_CONTROL,
 597                                           f->key);
 598
 599         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
 600                 struct flow_dissector_key_ports *key =
 601                         skb_flow_dissector_target(f->dissector,
 602                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
 603                                                   f->key);
 604                 struct flow_dissector_key_ports *mask =
 605                         skb_flow_dissector_target(f->dissector,
 606                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
 607                                                   f->mask);
 608                 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 609                 struct net_device *up_dev = mlx5_eswitch_get_uplink_netdev(esw);
 610                 struct mlx5e_priv *up_priv = netdev_priv(up_dev);
 611
 612                 /* Full udp dst port must be given */
 613                 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
 614                         goto vxlan_match_offload_err;
 615
 616                 if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->dst)) &&
 617                     MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
 618                         parse_vxlan_attr(spec, f);
 619                 else {
 620                         netdev_warn(priv->netdev,
 621                                     "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
 622                         return -EOPNOTSUPP;
 623                 }
 624
 625                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 626                          udp_dport, ntohs(mask->dst));
 627                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 628                          udp_dport, ntohs(key->dst));
 629
 630                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 631                          udp_sport, ntohs(mask->src));
 632                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 633                          udp_sport, ntohs(key->src));
 634         } else { /* udp dst port must be given */
 635 vxlan_match_offload_err:
 636                 netdev_warn(priv->netdev,
 637                             "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
 638                 return -EOPNOTSUPP;
 639         }
 640
 641         if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
 642                 struct flow_dissector_key_ipv4_addrs *key =
 643                         skb_flow_dissector_target(f->dissector,
 644                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
 645                                                   f->key);
 646                 struct flow_dissector_key_ipv4_addrs *mask =
 647                         skb_flow_dissector_target(f->dissector,
 648                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
 649                                                   f->mask);
 650                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 651                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
 652                          ntohl(mask->src));
 653                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 654                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
 655                          ntohl(key->src));
 656
 657                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 658                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
 659                          ntohl(mask->dst));
 660                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 661                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
 662                          ntohl(key->dst));
 663
 664                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
 665                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
 666         } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
 667                 struct flow_dissector_key_ipv6_addrs *key =
 668                         skb_flow_dissector_target(f->dissector,
 669                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
 670                                                   f->key);
 671                 struct flow_dissector_key_ipv6_addrs *mask =
 672                         skb_flow_dissector_target(f->dissector,
 673                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
 674                                                   f->mask);
 675
 676                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 677                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 678                        &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 679                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 680                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 681                        &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 682
 683                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 684                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 685                        &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 686                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 687                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 688                        &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 689
 690                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
 691                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
 692         }
 693
 694         /* Enforce DMAC when offloading incoming tunneled flows.
 695          * Flow counters require a match on the DMAC.
 696          */
 697         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
 698         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
 699         ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 700                                      dmac_47_16), priv->netdev->dev_addr);
 701
 702         /* let software handle IP fragments */
 703         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
 704         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
 705
 706         return 0;
 707 }
 708
 709 static int __parse_cls_flower(struct mlx5e_priv *priv,
 710                               struct mlx5_flow_spec *spec,
 711                               struct tc_cls_flower_offload *f,
 712                               u8 *min_inline)
 713 {
 714         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 715                                        outer_headers);
 716         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 717                                        outer_headers);
 718         u16 addr_type = 0;
 719         u8 ip_proto = 0;
 720
 721         *min_inline = MLX5_INLINE_MODE_L2;
 722
 723         if (f->dissector->used_keys &
 724             ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
 725               BIT(FLOW_DISSECTOR_KEY_BASIC) |
 726               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
 727               BIT(FLOW_DISSECTOR_KEY_VLAN) |
 728               BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
 729               BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
 730               BIT(FLOW_DISSECTOR_KEY_PORTS) |
 731               BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
 732               BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
 733               BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
 734               BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
 735               BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
 736               BIT(FLOW_DISSECTOR_KEY_TCP) |
 737               BIT(FLOW_DISSECTOR_KEY_IP))) {
 738                 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
 739                             f->dissector->used_keys);
 740                 return -EOPNOTSUPP;
 741         }
 742
 743         if ((dissector_uses_key(f->dissector,
 744                                 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
 745              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
 746              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
 747             dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
 748                 struct flow_dissector_key_control *key =
 749                         skb_flow_dissector_target(f->dissector,
 750                                                   FLOW_DISSECTOR_KEY_ENC_CONTROL,
 751                                                   f->key);
 752                 switch (key->addr_type) {
 753                 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
 754                 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
 755                         if (parse_tunnel_attr(priv, spec, f))
 756                                 return -EOPNOTSUPP;
 757                         break;
 758                 default:
 759                         return -EOPNOTSUPP;
 760                 }
 761
 762                 /* In decap flow, header pointers should point to the inner
 763                  * headers, outer header were already set by parse_tunnel_attr
 764                  */
 765                 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 766                                          inner_headers);
 767                 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 768                                          inner_headers);
 769         }
 770
 771         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
 772                 struct flow_dissector_key_control *key =
 773                         skb_flow_dissector_target(f->dissector,
 774                                                   FLOW_DISSECTOR_KEY_CONTROL,
 775                                                   f->key);
 776
 777                 struct flow_dissector_key_control *mask =
 778                         skb_flow_dissector_target(f->dissector,
 779                                                   FLOW_DISSECTOR_KEY_CONTROL,
 780                                                   f->mask);
 781                 addr_type = key->addr_type;
 782
 783                 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
 784                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
 785                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
 786                                  key->flags & FLOW_DIS_IS_FRAGMENT);
 787
 788                         /* the HW doesn't need L3 inline to match on frag=no */
 789                         if (key->flags & FLOW_DIS_IS_FRAGMENT)
 790                                 *min_inline = MLX5_INLINE_MODE_IP;
 791                 }
 792         }
 793
 794         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
 795                 struct flow_dissector_key_basic *key =
 796                         skb_flow_dissector_target(f->dissector,
 797                                                   FLOW_DISSECTOR_KEY_BASIC,
 798                                                   f->key);
 799                 struct flow_dissector_key_basic *mask =
 800                         skb_flow_dissector_target(f->dissector,
 801                                                   FLOW_DISSECTOR_KEY_BASIC,
 802                                                   f->mask);
 803                 ip_proto = key->ip_proto;
 804
 805                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
 806                          ntohs(mask->n_proto));
 807                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
 808                          ntohs(key->n_proto));
 809
 810                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
 811                          mask->ip_proto);
 812                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
 813                          key->ip_proto);
 814
 815                 if (mask->ip_proto)
 816                         *min_inline = MLX5_INLINE_MODE_IP;
 817         }
 818
 819         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
 820                 struct flow_dissector_key_eth_addrs *key =
 821                         skb_flow_dissector_target(f->dissector,
 822                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
 823                                                   f->key);
 824                 struct flow_dissector_key_eth_addrs *mask =
 825                         skb_flow_dissector_target(f->dissector,
 826                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
 827                                                   f->mask);
 828
 829                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 830                                              dmac_47_16),
 831                                 mask->dst);
 832                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 833                                              dmac_47_16),
 834                                 key->dst);
 835
 836                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 837                                              smac_47_16),
 838                                 mask->src);
 839                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 840                                              smac_47_16),
 841                                 key->src);
 842         }
 843
 844         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
 845                 struct flow_dissector_key_vlan *key =
 846                         skb_flow_dissector_target(f->dissector,
 847                                                   FLOW_DISSECTOR_KEY_VLAN,
 848                                                   f->key);
 849                 struct flow_dissector_key_vlan *mask =
 850                         skb_flow_dissector_target(f->dissector,
 851                                                   FLOW_DISSECTOR_KEY_VLAN,
 852                                                   f->mask);
 853                 if (mask->vlan_id || mask->vlan_priority) {
 854                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
 855                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
 856
 857                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
 858                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
 859
 860                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
 861                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
 862                 }
 863         }
 864
 865         if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
 866                 struct flow_dissector_key_ipv4_addrs *key =
 867                         skb_flow_dissector_target(f->dissector,
 868                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 869                                                   f->key);
 870                 struct flow_dissector_key_ipv4_addrs *mask =
 871                         skb_flow_dissector_target(f->dissector,
 872                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 873                                                   f->mask);
 874
 875                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 876                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
 877                        &mask->src, sizeof(mask->src));
 878                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 879                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
 880                        &key->src, sizeof(key->src));
 881                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 882                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 883                        &mask->dst, sizeof(mask->dst));
 884                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 885                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 886                        &key->dst, sizeof(key->dst));
 887
 888                 if (mask->src || mask->dst)
 889                         *min_inline = MLX5_INLINE_MODE_IP;
 890         }
 891
 892         if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
 893                 struct flow_dissector_key_ipv6_addrs *key =
 894                         skb_flow_dissector_target(f->dissector,
 895                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 896                                                   f->key);
 897                 struct flow_dissector_key_ipv6_addrs *mask =
 898                         skb_flow_dissector_target(f->dissector,
 899                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 900                                                   f->mask);
 901
 902                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 903                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 904                        &mask->src, sizeof(mask->src));
 905                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 906                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 907                        &key->src, sizeof(key->src));
 908
 909                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 910                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 911                        &mask->dst, sizeof(mask->dst));
 912                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 913                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 914                        &key->dst, sizeof(key->dst));
 915
 916                 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
 917                     ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
 918                         *min_inline = MLX5_INLINE_MODE_IP;
 919         }
 920
 921         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_IP)) {
 922                 struct flow_dissector_key_ip *key =
 923                         skb_flow_dissector_target(f->dissector,
 924                                                   FLOW_DISSECTOR_KEY_IP,
 925                                                   f->key);
 926                 struct flow_dissector_key_ip *mask =
 927                         skb_flow_dissector_target(f->dissector,
 928                                                   FLOW_DISSECTOR_KEY_IP,
 929                                                   f->mask);
 930
 931                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn, mask->tos & 0x3);
 932                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, key->tos & 0x3);
 933
 934                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp, mask->tos >> 2);
 935                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, key->tos  >> 2);
 936
 937                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit, mask->ttl);
 938                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit, key->ttl);
 939
 940                 if (mask->ttl &&
 941                     !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
 942                                                 ft_field_support.outer_ipv4_ttl))
 943                         return -EOPNOTSUPP;
 944
 945                 if (mask->tos || mask->ttl)
 946                         *min_inline = MLX5_INLINE_MODE_IP;
 947         }
 948
 949         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
 950                 struct flow_dissector_key_ports *key =
 951                         skb_flow_dissector_target(f->dissector,
 952                                                   FLOW_DISSECTOR_KEY_PORTS,
 953                                                   f->key);
 954                 struct flow_dissector_key_ports *mask =
 955                         skb_flow_dissector_target(f->dissector,
 956                                                   FLOW_DISSECTOR_KEY_PORTS,
 957                                                   f->mask);
 958                 switch (ip_proto) {
 959                 case IPPROTO_TCP:
 960                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 961                                  tcp_sport, ntohs(mask->src));
 962                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 963                                  tcp_sport, ntohs(key->src));
 964
 965                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 966                                  tcp_dport, ntohs(mask->dst));
 967                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 968                                  tcp_dport, ntohs(key->dst));
 969                         break;
 970
 971                 case IPPROTO_UDP:
 972                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 973                                  udp_sport, ntohs(mask->src));
 974                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 975                                  udp_sport, ntohs(key->src));
 976
 977                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 978                                  udp_dport, ntohs(mask->dst));
 979                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 980                                  udp_dport, ntohs(key->dst));
 981                         break;
 982                 default:
 983                         netdev_err(priv->netdev,
 984                                    "Only UDP and TCP transport are supported\n");
 985                         return -EINVAL;
 986                 }
 987
 988                 if (mask->src || mask->dst)
 989                         *min_inline = MLX5_INLINE_MODE_TCP_UDP;
 990         }
 991
 992         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_TCP)) {
 993                 struct flow_dissector_key_tcp *key =
 994                         skb_flow_dissector_target(f->dissector,
 995                                                   FLOW_DISSECTOR_KEY_TCP,
 996                                                   f->key);
 997                 struct flow_dissector_key_tcp *mask =
 998                         skb_flow_dissector_target(f->dissector,
 999                                                   FLOW_DISSECTOR_KEY_TCP,
1000                                                   f->mask);
1001
1002                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
1003                          ntohs(mask->flags));
1004                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
1005                          ntohs(key->flags));
1006
1007                 if (mask->flags)
1008                         *min_inline = MLX5_INLINE_MODE_TCP_UDP;
1009         }
1010
1011         return 0;
1012 }
1013
1014 static int parse_cls_flower(struct mlx5e_priv *priv,
1015                             struct mlx5e_tc_flow *flow,
1016                             struct mlx5_flow_spec *spec,
1017                             struct tc_cls_flower_offload *f)
1018 {
1019         struct mlx5_core_dev *dev = priv->mdev;
1020         struct mlx5_eswitch *esw = dev->priv.eswitch;
1021         struct mlx5e_rep_priv *rpriv = priv->ppriv;
1022         struct mlx5_eswitch_rep *rep;
1023         u8 min_inline;
1024         int err;
1025
1026         err = __parse_cls_flower(priv, spec, f, &min_inline);
1027
1028         if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH)) {
1029                 rep = rpriv->rep;
1030                 if (rep->vport != FDB_UPLINK_VPORT &&
1031                     (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
1032                     esw->offloads.inline_mode < min_inline)) {
1033                         netdev_warn(priv->netdev,
1034                                     "Flow is not offloaded due to min inline setting, required %d actual %d\n",
1035                                     min_inline, esw->offloads.inline_mode);
1036                         return -EOPNOTSUPP;
1037                 }
1038         }
1039
1040         return err;
1041 }
1042
1043 struct pedit_headers {
1044         struct ethhdr  eth;
1045         struct iphdr   ip4;
1046         struct ipv6hdr ip6;
1047         struct tcphdr  tcp;
1048         struct udphdr  udp;
1049 };
1050
1051 static int pedit_header_offsets[] = {
1052         [TCA_PEDIT_KEY_EX_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
1053         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
1054         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
1055         [TCA_PEDIT_KEY_EX_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
1056         [TCA_PEDIT_KEY_EX_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
1057 };
1058
1059 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
1060
1061 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
1062                          struct pedit_headers *masks,
1063                          struct pedit_headers *vals)
1064 {
1065         u32 *curr_pmask, *curr_pval;
1066
1067         if (hdr_type >= __PEDIT_HDR_TYPE_MAX)
1068                 goto out_err;
1069
1070         curr_pmask = (u32 *)(pedit_header(masks, hdr_type) + offset);
1071         curr_pval  = (u32 *)(pedit_header(vals, hdr_type) + offset);
1072
1073         if (*curr_pmask & mask)  /* disallow acting twice on the same location */
1074                 goto out_err;
1075
1076         *curr_pmask |= mask;
1077         *curr_pval  |= (val & mask);
1078
1079         return 0;
1080
1081 out_err:
1082         return -EOPNOTSUPP;
1083 }
1084
1085 struct mlx5_fields {
1086         u8  field;
1087         u8  size;
1088         u32 offset;
1089 };
1090
1091 #define OFFLOAD(fw_field, size, field, off) \
1092                 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, size, offsetof(struct pedit_headers, field) + (off)}
1093
1094 static struct mlx5_fields fields[] = {
1095         OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0),
1096         OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0),
1097         OFFLOAD(DMAC_15_0,  2, eth.h_dest[4], 0),
1098         OFFLOAD(SMAC_47_16, 4, eth.h_source[0], 0),
1099         OFFLOAD(SMAC_15_0,  2, eth.h_source[4], 0),
1100         OFFLOAD(ETHERTYPE,  2, eth.h_proto, 0),
1101
1102         OFFLOAD(IP_TTL, 1, ip4.ttl,   0),
1103         OFFLOAD(SIPV4,  4, ip4.saddr, 0),
1104         OFFLOAD(DIPV4,  4, ip4.daddr, 0),
1105
1106         OFFLOAD(SIPV6_127_96, 4, ip6.saddr.s6_addr32[0], 0),
1107         OFFLOAD(SIPV6_95_64,  4, ip6.saddr.s6_addr32[1], 0),
1108         OFFLOAD(SIPV6_63_32,  4, ip6.saddr.s6_addr32[2], 0),
1109         OFFLOAD(SIPV6_31_0,   4, ip6.saddr.s6_addr32[3], 0),
1110         OFFLOAD(DIPV6_127_96, 4, ip6.daddr.s6_addr32[0], 0),
1111         OFFLOAD(DIPV6_95_64,  4, ip6.daddr.s6_addr32[1], 0),
1112         OFFLOAD(DIPV6_63_32,  4, ip6.daddr.s6_addr32[2], 0),
1113         OFFLOAD(DIPV6_31_0,   4, ip6.daddr.s6_addr32[3], 0),
1114         OFFLOAD(IPV6_HOPLIMIT, 1, ip6.hop_limit, 0),
1115
1116         OFFLOAD(TCP_SPORT, 2, tcp.source,  0),
1117         OFFLOAD(TCP_DPORT, 2, tcp.dest,    0),
1118         OFFLOAD(TCP_FLAGS, 1, tcp.ack_seq, 5),
1119
1120         OFFLOAD(UDP_SPORT, 2, udp.source, 0),
1121         OFFLOAD(UDP_DPORT, 2, udp.dest,   0),
1122 };
1123
1124 /* On input attr->num_mod_hdr_actions tells how many HW actions can be parsed at
1125  * max from the SW pedit action. On success, it says how many HW actions were
1126  * actually parsed.
1127  */
1128 static int offload_pedit_fields(struct pedit_headers *masks,
1129                                 struct pedit_headers *vals,
1130                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
1131 {
1132         struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
1133         int i, action_size, nactions, max_actions, first, last, next_z;
1134         void *s_masks_p, *a_masks_p, *vals_p;
1135         struct mlx5_fields *f;
1136         u8 cmd, field_bsize;
1137         u32 s_mask, a_mask;
1138         unsigned long mask;
1139         __be32 mask_be32;
1140         __be16 mask_be16;
1141         void *action;
1142
1143         set_masks = &masks[TCA_PEDIT_KEY_EX_CMD_SET];
1144         add_masks = &masks[TCA_PEDIT_KEY_EX_CMD_ADD];
1145         set_vals = &vals[TCA_PEDIT_KEY_EX_CMD_SET];
1146         add_vals = &vals[TCA_PEDIT_KEY_EX_CMD_ADD];
1147
1148         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1149         action = parse_attr->mod_hdr_actions;
1150         max_actions = parse_attr->num_mod_hdr_actions;
1151         nactions = 0;
1152
1153         for (i = 0; i < ARRAY_SIZE(fields); i++) {
1154                 f = &fields[i];
1155                 /* avoid seeing bits set from previous iterations */
1156                 s_mask = 0;
1157                 a_mask = 0;
1158
1159                 s_masks_p = (void *)set_masks + f->offset;
1160                 a_masks_p = (void *)add_masks + f->offset;
1161
1162                 memcpy(&s_mask, s_masks_p, f->size);
1163                 memcpy(&a_mask, a_masks_p, f->size);
1164
1165                 if (!s_mask && !a_mask) /* nothing to offload here */
1166                         continue;
1167
1168                 if (s_mask && a_mask) {
1169                         printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
1170                         return -EOPNOTSUPP;
1171                 }
1172
1173                 if (nactions == max_actions) {
1174                         printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
1175                         return -EOPNOTSUPP;
1176                 }
1177
1178                 if (s_mask) {
1179                         cmd  = MLX5_ACTION_TYPE_SET;
1180                         mask = s_mask;
1181                         vals_p = (void *)set_vals + f->offset;
1182                         /* clear to denote we consumed this field */
1183                         memset(s_masks_p, 0, f->size);
1184                 } else {
1185                         cmd  = MLX5_ACTION_TYPE_ADD;
1186                         mask = a_mask;
1187                         vals_p = (void *)add_vals + f->offset;
1188                         /* clear to denote we consumed this field */
1189                         memset(a_masks_p, 0, f->size);
1190                 }
1191
1192                 field_bsize = f->size * BITS_PER_BYTE;
1193
1194                 if (field_bsize == 32) {
1195                         mask_be32 = *(__be32 *)&mask;
1196                         mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
1197                 } else if (field_bsize == 16) {
1198                         mask_be16 = *(__be16 *)&mask;
1199                         mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
1200                 }
1201
1202                 first = find_first_bit(&mask, field_bsize);
1203                 next_z = find_next_zero_bit(&mask, field_bsize, first);
1204                 last  = find_last_bit(&mask, field_bsize);
1205                 if (first < next_z && next_z < last) {
1206                         printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
1207                                mask);
1208                         return -EOPNOTSUPP;
1209                 }
1210
1211                 MLX5_SET(set_action_in, action, action_type, cmd);
1212                 MLX5_SET(set_action_in, action, field, f->field);
1213
1214                 if (cmd == MLX5_ACTION_TYPE_SET) {
1215                         MLX5_SET(set_action_in, action, offset, first);
1216                         /* length is num of bits to be written, zero means length of 32 */
1217                         MLX5_SET(set_action_in, action, length, (last - first + 1));
1218                 }
1219
1220                 if (field_bsize == 32)
1221                         MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
1222                 else if (field_bsize == 16)
1223                         MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
1224                 else if (field_bsize == 8)
1225                         MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
1226
1227                 action += action_size;
1228                 nactions++;
1229         }
1230
1231         parse_attr->num_mod_hdr_actions = nactions;
1232         return 0;
1233 }
1234
1235 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
1236                                  const struct tc_action *a, int namespace,
1237                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1238 {
1239         int nkeys, action_size, max_actions;
1240
1241         nkeys = tcf_pedit_nkeys(a);
1242         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1243
1244         if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
1245                 max_actions = MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev, max_modify_header_actions);
1246         else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
1247                 max_actions = MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, max_modify_header_actions);
1248
1249         /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
1250         max_actions = min(max_actions, nkeys * 16);
1251
1252         parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
1253         if (!parse_attr->mod_hdr_actions)
1254                 return -ENOMEM;
1255
1256         parse_attr->num_mod_hdr_actions = max_actions;
1257         return 0;
1258 }
1259
1260 static const struct pedit_headers zero_masks = {};
1261
1262 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
1263                                  const struct tc_action *a, int namespace,
1264                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1265 {
1266         struct pedit_headers masks[__PEDIT_CMD_MAX], vals[__PEDIT_CMD_MAX], *cmd_masks;
1267         int nkeys, i, err = -EOPNOTSUPP;
1268         u32 mask, val, offset;
1269         u8 cmd, htype;
1270
1271         nkeys = tcf_pedit_nkeys(a);
1272
1273         memset(masks, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1274         memset(vals,  0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1275
1276         for (i = 0; i < nkeys; i++) {
1277                 htype = tcf_pedit_htype(a, i);
1278                 cmd = tcf_pedit_cmd(a, i);
1279                 err = -EOPNOTSUPP; /* can't be all optimistic */
1280
1281                 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK) {
1282                         printk(KERN_WARNING "mlx5: legacy pedit isn't offloaded\n");
1283                         goto out_err;
1284                 }
1285
1286                 if (cmd != TCA_PEDIT_KEY_EX_CMD_SET && cmd != TCA_PEDIT_KEY_EX_CMD_ADD) {
1287                         printk(KERN_WARNING "mlx5: pedit cmd %d isn't offloaded\n", cmd);
1288                         goto out_err;
1289                 }
1290
1291                 mask = tcf_pedit_mask(a, i);
1292                 val = tcf_pedit_val(a, i);
1293                 offset = tcf_pedit_offset(a, i);
1294
1295                 err = set_pedit_val(htype, ~mask, val, offset, &masks[cmd], &vals[cmd]);
1296                 if (err)
1297                         goto out_err;
1298         }
1299
1300         err = alloc_mod_hdr_actions(priv, a, namespace, parse_attr);
1301         if (err)
1302                 goto out_err;
1303
1304         err = offload_pedit_fields(masks, vals, parse_attr);
1305         if (err < 0)
1306                 goto out_dealloc_parsed_actions;
1307
1308         for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
1309                 cmd_masks = &masks[cmd];
1310                 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
1311                         printk(KERN_WARNING "mlx5: attempt to offload an unsupported field (cmd %d)\n",
1312                                cmd);
1313                         print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
1314                                        16, 1, cmd_masks, sizeof(zero_masks), true);
1315                         err = -EOPNOTSUPP;
1316                         goto out_dealloc_parsed_actions;
1317                 }
1318         }
1319
1320         return 0;
1321
1322 out_dealloc_parsed_actions:
1323         kfree(parse_attr->mod_hdr_actions);
1324 out_err:
1325         return err;
1326 }
1327
1328 static bool csum_offload_supported(struct mlx5e_priv *priv, u32 action, u32 update_flags)
1329 {
1330         u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
1331                          TCA_CSUM_UPDATE_FLAG_UDP;
1332
1333         /*  The HW recalcs checksums only if re-writing headers */
1334         if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
1335                 netdev_warn(priv->netdev,
1336                             "TC csum action is only offloaded with pedit\n");
1337                 return false;
1338         }
1339
1340         if (update_flags & ~prot_flags) {
1341                 netdev_warn(priv->netdev,
1342                             "can't offload TC csum action for some header/s - flags %#x\n",
1343                             update_flags);
1344                 return false;
1345         }
1346
1347         return true;
1348 }
1349
1350 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
1351                                           struct tcf_exts *exts)
1352 {
1353         const struct tc_action *a;
1354         bool modify_ip_header;
1355         LIST_HEAD(actions);
1356         u8 htype, ip_proto;
1357         void *headers_v;
1358         u16 ethertype;
1359         int nkeys, i;
1360
1361         headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
1362         ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
1363
1364         /* for non-IP we only re-write MACs, so we're okay */
1365         if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
1366                 goto out_ok;
1367
1368         modify_ip_header = false;
1369         tcf_exts_to_list(exts, &actions);
1370         list_for_each_entry(a, &actions, list) {
1371                 if (!is_tcf_pedit(a))
1372                         continue;
1373
1374                 nkeys = tcf_pedit_nkeys(a);
1375                 for (i = 0; i < nkeys; i++) {
1376                         htype = tcf_pedit_htype(a, i);
1377                         if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP4 ||
1378                             htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP6) {
1379                                 modify_ip_header = true;
1380                                 break;
1381                         }
1382                 }
1383         }
1384
1385         ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
1386         if (modify_ip_header && ip_proto != IPPROTO_TCP && ip_proto != IPPROTO_UDP) {
1387                 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
1388                 return false;
1389         }
1390
1391 out_ok:
1392         return true;
1393 }
1394
1395 static bool actions_match_supported(struct mlx5e_priv *priv,
1396                                     struct tcf_exts *exts,
1397                                     struct mlx5e_tc_flow_parse_attr *parse_attr,
1398                                     struct mlx5e_tc_flow *flow)
1399 {
1400         u32 actions;
1401
1402         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1403                 actions = flow->esw_attr->action;
1404         else
1405                 actions = flow->nic_attr->action;
1406
1407         if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1408                 return modify_header_match_supported(&parse_attr->spec, exts);
1409
1410         return true;
1411 }
1412
1413 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
1414                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
1415                                 struct mlx5e_tc_flow *flow)
1416 {
1417         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
1418         const struct tc_action *a;
1419         LIST_HEAD(actions);
1420         int err;
1421
1422         if (!tcf_exts_has_actions(exts))
1423                 return -EINVAL;
1424
1425         attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
1426         attr->action = 0;
1427
1428         tcf_exts_to_list(exts, &actions);
1429         list_for_each_entry(a, &actions, list) {
1430                 if (is_tcf_gact_shot(a)) {
1431                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
1432                         if (MLX5_CAP_FLOWTABLE(priv->mdev,
1433                                                flow_table_properties_nic_receive.flow_counter))
1434                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
1435                         continue;
1436                 }
1437
1438                 if (is_tcf_pedit(a)) {
1439                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_KERNEL,
1440                                                     parse_attr);
1441                         if (err)
1442                                 return err;
1443
1444                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
1445                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1446                         continue;
1447                 }
1448
1449                 if (is_tcf_csum(a)) {
1450                         if (csum_offload_supported(priv, attr->action,
1451                                                    tcf_csum_update_flags(a)))
1452                                 continue;
1453
1454                         return -EOPNOTSUPP;
1455                 }
1456
1457                 if (is_tcf_skbedit_mark(a)) {
1458                         u32 mark = tcf_skbedit_mark(a);
1459
1460                         if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
1461                                 netdev_warn(priv->netdev, "Bad flow mark - only 16 bit is supported: 0x%x\n",
1462                                             mark);
1463                                 return -EINVAL;
1464                         }
1465
1466                         attr->flow_tag = mark;
1467                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1468                         continue;
1469                 }
1470
1471                 return -EINVAL;
1472         }
1473
1474         if (!actions_match_supported(priv, exts, parse_attr, flow))
1475                 return -EOPNOTSUPP;
1476
1477         return 0;
1478 }
1479
1480 static inline int cmp_encap_info(struct ip_tunnel_key *a,
1481                                  struct ip_tunnel_key *b)
1482 {
1483         return memcmp(a, b, sizeof(*a));
1484 }
1485
1486 static inline int hash_encap_info(struct ip_tunnel_key *key)
1487 {
1488         return jhash(key, sizeof(*key), 0);
1489 }
1490
1491 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
1492                                    struct net_device *mirred_dev,
1493                                    struct net_device **out_dev,
1494                                    struct flowi4 *fl4,
1495                                    struct neighbour **out_n,
1496                                    int *out_ttl)
1497 {
1498         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1499         struct rtable *rt;
1500         struct neighbour *n = NULL;
1501
1502 #if IS_ENABLED(CONFIG_INET)
1503         int ret;
1504
1505         rt = ip_route_output_key(dev_net(mirred_dev), fl4);
1506         ret = PTR_ERR_OR_ZERO(rt);
1507         if (ret)
1508                 return ret;
1509 #else
1510         return -EOPNOTSUPP;
1511 #endif
1512         /* if the egress device isn't on the same HW e-switch, we use the uplink */
1513         if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
1514                 *out_dev = mlx5_eswitch_get_uplink_netdev(esw);
1515         else
1516                 *out_dev = rt->dst.dev;
1517
1518         *out_ttl = ip4_dst_hoplimit(&rt->dst);
1519         n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
1520         ip_rt_put(rt);
1521         if (!n)
1522                 return -ENOMEM;
1523
1524         *out_n = n;
1525         return 0;
1526 }
1527
1528 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
1529                                    struct net_device *mirred_dev,
1530                                    struct net_device **out_dev,
1531                                    struct flowi6 *fl6,
1532                                    struct neighbour **out_n,
1533                                    int *out_ttl)
1534 {
1535         struct neighbour *n = NULL;
1536         struct dst_entry *dst;
1537
1538 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
1539         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1540         int ret;
1541
1542         ret = ipv6_stub->ipv6_dst_lookup(dev_net(mirred_dev), NULL, &dst,
1543                                          fl6);
1544         if (ret < 0)
1545                 return ret;
1546
1547         *out_ttl = ip6_dst_hoplimit(dst);
1548
1549         /* if the egress device isn't on the same HW e-switch, we use the uplink */
1550         if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
1551                 *out_dev = mlx5_eswitch_get_uplink_netdev(esw);
1552         else
1553                 *out_dev = dst->dev;
1554 #else
1555         return -EOPNOTSUPP;
1556 #endif
1557
1558         n = dst_neigh_lookup(dst, &fl6->daddr);
1559         dst_release(dst);
1560         if (!n)
1561                 return -ENOMEM;
1562
1563         *out_n = n;
1564         return 0;
1565 }
1566
1567 static void gen_vxlan_header_ipv4(struct net_device *out_dev,
1568                                   char buf[], int encap_size,
1569                                   unsigned char h_dest[ETH_ALEN],
1570                                   int ttl,
1571                                   __be32 daddr,
1572                                   __be32 saddr,
1573                                   __be16 udp_dst_port,
1574                                   __be32 vx_vni)
1575 {
1576         struct ethhdr *eth = (struct ethhdr *)buf;
1577         struct iphdr  *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
1578         struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
1579         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
1580
1581         memset(buf, 0, encap_size);
1582
1583         ether_addr_copy(eth->h_dest, h_dest);
1584         ether_addr_copy(eth->h_source, out_dev->dev_addr);
1585         eth->h_proto = htons(ETH_P_IP);
1586
1587         ip->daddr = daddr;
1588         ip->saddr = saddr;
1589
1590         ip->ttl = ttl;
1591         ip->protocol = IPPROTO_UDP;
1592         ip->version = 0x4;
1593         ip->ihl = 0x5;
1594
1595         udp->dest = udp_dst_port;
1596         vxh->vx_flags = VXLAN_HF_VNI;
1597         vxh->vx_vni = vxlan_vni_field(vx_vni);
1598 }
1599
1600 static void gen_vxlan_header_ipv6(struct net_device *out_dev,
1601                                   char buf[], int encap_size,
1602                                   unsigned char h_dest[ETH_ALEN],
1603                                   int ttl,
1604                                   struct in6_addr *daddr,
1605                                   struct in6_addr *saddr,
1606                                   __be16 udp_dst_port,
1607                                   __be32 vx_vni)
1608 {
1609         struct ethhdr *eth = (struct ethhdr *)buf;
1610         struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
1611         struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
1612         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
1613
1614         memset(buf, 0, encap_size);
1615
1616         ether_addr_copy(eth->h_dest, h_dest);
1617         ether_addr_copy(eth->h_source, out_dev->dev_addr);
1618         eth->h_proto = htons(ETH_P_IPV6);
1619
1620         ip6_flow_hdr(ip6h, 0, 0);
1621         /* the HW fills up ipv6 payload len */
1622         ip6h->nexthdr     = IPPROTO_UDP;
1623         ip6h->hop_limit   = ttl;
1624         ip6h->daddr       = *daddr;
1625         ip6h->saddr       = *saddr;
1626
1627         udp->dest = udp_dst_port;
1628         vxh->vx_flags = VXLAN_HF_VNI;
1629         vxh->vx_vni = vxlan_vni_field(vx_vni);
1630 }
1631
1632 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
1633                                           struct net_device *mirred_dev,
1634                                           struct mlx5e_encap_entry *e)
1635 {
1636         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
1637         int ipv4_encap_size = ETH_HLEN + sizeof(struct iphdr) + VXLAN_HLEN;
1638         struct ip_tunnel_key *tun_key = &e->tun_info.key;
1639         struct net_device *out_dev;
1640         struct neighbour *n = NULL;
1641         struct flowi4 fl4 = {};
1642         char *encap_header;
1643         int ttl, err;
1644         u8 nud_state;
1645
1646         if (max_encap_size < ipv4_encap_size) {
1647                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
1648                                ipv4_encap_size, max_encap_size);
1649                 return -EOPNOTSUPP;
1650         }
1651
1652         encap_header = kzalloc(ipv4_encap_size, GFP_KERNEL);
1653         if (!encap_header)
1654                 return -ENOMEM;
1655
1656         switch (e->tunnel_type) {
1657         case MLX5_HEADER_TYPE_VXLAN:
1658                 fl4.flowi4_proto = IPPROTO_UDP;
1659                 fl4.fl4_dport = tun_key->tp_dst;
1660                 break;
1661         default:
1662                 err = -EOPNOTSUPP;
1663                 goto free_encap;
1664         }
1665         fl4.flowi4_tos = tun_key->tos;
1666         fl4.daddr = tun_key->u.ipv4.dst;
1667         fl4.saddr = tun_key->u.ipv4.src;
1668
1669         err = mlx5e_route_lookup_ipv4(priv, mirred_dev, &out_dev,
1670                                       &fl4, &n, &ttl);
1671         if (err)
1672                 goto free_encap;
1673
1674         /* used by mlx5e_detach_encap to lookup a neigh hash table
1675          * entry in the neigh hash table when a user deletes a rule
1676          */
1677         e->m_neigh.dev = n->dev;
1678         e->m_neigh.family = n->ops->family;
1679         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
1680         e->out_dev = out_dev;
1681
1682         /* It's importent to add the neigh to the hash table before checking
1683          * the neigh validity state. So if we'll get a notification, in case the
1684          * neigh changes it's validity state, we would find the relevant neigh
1685          * in the hash.
1686          */
1687         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
1688         if (err)
1689                 goto free_encap;
1690
1691         read_lock_bh(&n->lock);
1692         nud_state = n->nud_state;
1693         ether_addr_copy(e->h_dest, n->ha);
1694         read_unlock_bh(&n->lock);
1695
1696         switch (e->tunnel_type) {
1697         case MLX5_HEADER_TYPE_VXLAN:
1698                 gen_vxlan_header_ipv4(out_dev, encap_header,
1699                                       ipv4_encap_size, e->h_dest, ttl,
1700                                       fl4.daddr,
1701                                       fl4.saddr, tun_key->tp_dst,
1702                                       tunnel_id_to_key32(tun_key->tun_id));
1703                 break;
1704         default:
1705                 err = -EOPNOTSUPP;
1706                 goto destroy_neigh_entry;
1707         }
1708         e->encap_size = ipv4_encap_size;
1709         e->encap_header = encap_header;
1710
1711         if (!(nud_state & NUD_VALID)) {
1712                 neigh_event_send(n, NULL);
1713                 err = -EAGAIN;
1714                 goto out;
1715         }
1716
1717         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
1718                                ipv4_encap_size, encap_header, &e->encap_id);
1719         if (err)
1720                 goto destroy_neigh_entry;
1721
1722         e->flags |= MLX5_ENCAP_ENTRY_VALID;
1723         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
1724         neigh_release(n);
1725         return err;
1726
1727 destroy_neigh_entry:
1728         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1729 free_encap:
1730         kfree(encap_header);
1731 out:
1732         if (n)
1733                 neigh_release(n);
1734         return err;
1735 }
1736
1737 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
1738                                           struct net_device *mirred_dev,
1739                                           struct mlx5e_encap_entry *e)
1740 {
1741         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
1742         int ipv6_encap_size = ETH_HLEN + sizeof(struct ipv6hdr) + VXLAN_HLEN;
1743         struct ip_tunnel_key *tun_key = &e->tun_info.key;
1744         struct net_device *out_dev;
1745         struct neighbour *n = NULL;
1746         struct flowi6 fl6 = {};
1747         char *encap_header;
1748         int err, ttl = 0;
1749         u8 nud_state;
1750
1751         if (max_encap_size < ipv6_encap_size) {
1752                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
1753                                ipv6_encap_size, max_encap_size);
1754                 return -EOPNOTSUPP;
1755         }
1756
1757         encap_header = kzalloc(ipv6_encap_size, GFP_KERNEL);
1758         if (!encap_header)
1759                 return -ENOMEM;
1760
1761         switch (e->tunnel_type) {
1762         case MLX5_HEADER_TYPE_VXLAN:
1763                 fl6.flowi6_proto = IPPROTO_UDP;
1764                 fl6.fl6_dport = tun_key->tp_dst;
1765                 break;
1766         default:
1767                 err = -EOPNOTSUPP;
1768                 goto free_encap;
1769         }
1770
1771         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
1772         fl6.daddr = tun_key->u.ipv6.dst;
1773         fl6.saddr = tun_key->u.ipv6.src;
1774
1775         err = mlx5e_route_lookup_ipv6(priv, mirred_dev, &out_dev,
1776                                       &fl6, &n, &ttl);
1777         if (err)
1778                 goto free_encap;
1779
1780         /* used by mlx5e_detach_encap to lookup a neigh hash table
1781          * entry in the neigh hash table when a user deletes a rule
1782          */
1783         e->m_neigh.dev = n->dev;
1784         e->m_neigh.family = n->ops->family;
1785         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
1786         e->out_dev = out_dev;
1787
1788         /* It's importent to add the neigh to the hash table before checking
1789          * the neigh validity state. So if we'll get a notification, in case the
1790          * neigh changes it's validity state, we would find the relevant neigh
1791          * in the hash.
1792          */
1793         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
1794         if (err)
1795                 goto free_encap;
1796
1797         read_lock_bh(&n->lock);
1798         nud_state = n->nud_state;
1799         ether_addr_copy(e->h_dest, n->ha);
1800         read_unlock_bh(&n->lock);
1801
1802         switch (e->tunnel_type) {
1803         case MLX5_HEADER_TYPE_VXLAN:
1804                 gen_vxlan_header_ipv6(out_dev, encap_header,
1805                                       ipv6_encap_size, e->h_dest, ttl,
1806                                       &fl6.daddr,
1807                                       &fl6.saddr, tun_key->tp_dst,
1808                                       tunnel_id_to_key32(tun_key->tun_id));
1809                 break;
1810         default:
1811                 err = -EOPNOTSUPP;
1812                 goto destroy_neigh_entry;
1813         }
1814
1815         e->encap_size = ipv6_encap_size;
1816         e->encap_header = encap_header;
1817
1818         if (!(nud_state & NUD_VALID)) {
1819                 neigh_event_send(n, NULL);
1820                 err = -EAGAIN;
1821                 goto out;
1822         }
1823
1824         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
1825                                ipv6_encap_size, encap_header, &e->encap_id);
1826         if (err)
1827                 goto destroy_neigh_entry;
1828
1829         e->flags |= MLX5_ENCAP_ENTRY_VALID;
1830         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
1831         neigh_release(n);
1832         return err;
1833
1834 destroy_neigh_entry:
1835         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1836 free_encap:
1837         kfree(encap_header);
1838 out:
1839         if (n)
1840                 neigh_release(n);
1841         return err;
1842 }
1843
1844 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
1845                               struct ip_tunnel_info *tun_info,
1846                               struct net_device *mirred_dev,
1847                               struct net_device **encap_dev,
1848                               struct mlx5e_tc_flow *flow)
1849 {
1850         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1851         struct net_device *up_dev = mlx5_eswitch_get_uplink_netdev(esw);
1852         unsigned short family = ip_tunnel_info_af(tun_info);
1853         struct mlx5e_priv *up_priv = netdev_priv(up_dev);
1854         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1855         struct ip_tunnel_key *key = &tun_info->key;
1856         struct mlx5e_encap_entry *e;
1857         int tunnel_type, err = 0;
1858         uintptr_t hash_key;
1859         bool found = false;
1860
1861         /* udp dst port must be set */
1862         if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
1863                 goto vxlan_encap_offload_err;
1864
1865         /* setting udp src port isn't supported */
1866         if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
1867 vxlan_encap_offload_err:
1868                 netdev_warn(priv->netdev,
1869                             "must set udp dst port and not set udp src port\n");
1870                 return -EOPNOTSUPP;
1871         }
1872
1873         if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->tp_dst)) &&
1874             MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
1875                 tunnel_type = MLX5_HEADER_TYPE_VXLAN;
1876         } else {
1877                 netdev_warn(priv->netdev,
1878                             "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
1879                 return -EOPNOTSUPP;
1880         }
1881
1882         hash_key = hash_encap_info(key);
1883
1884         hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
1885                                    encap_hlist, hash_key) {
1886                 if (!cmp_encap_info(&e->tun_info.key, key)) {
1887                         found = true;
1888                         break;
1889                 }
1890         }
1891
1892         /* must verify if encap is valid or not */
1893         if (found)
1894                 goto attach_flow;
1895
1896         e = kzalloc(sizeof(*e), GFP_KERNEL);
1897         if (!e)
1898                 return -ENOMEM;
1899
1900         e->tun_info = *tun_info;
1901         e->tunnel_type = tunnel_type;
1902         INIT_LIST_HEAD(&e->flows);
1903
1904         if (family == AF_INET)
1905                 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e);
1906         else if (family == AF_INET6)
1907                 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e);
1908
1909         if (err && err != -EAGAIN)
1910                 goto out_err;
1911
1912         hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
1913
1914 attach_flow:
1915         list_add(&flow->encap, &e->flows);
1916         *encap_dev = e->out_dev;
1917         if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1918                 attr->encap_id = e->encap_id;
1919         else
1920                 err = -EAGAIN;
1921
1922         return err;
1923
1924 out_err:
1925         kfree(e);
1926         return err;
1927 }
1928
1929 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
1930                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
1931                                 struct mlx5e_tc_flow *flow)
1932 {
1933         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1934         struct mlx5e_rep_priv *rpriv = priv->ppriv;
1935         struct ip_tunnel_info *info = NULL;
1936         const struct tc_action *a;
1937         LIST_HEAD(actions);
1938         bool encap = false;
1939         int err = 0;
1940
1941         if (!tcf_exts_has_actions(exts))
1942                 return -EINVAL;
1943
1944         memset(attr, 0, sizeof(*attr));
1945         attr->in_rep = rpriv->rep;
1946
1947         tcf_exts_to_list(exts, &actions);
1948         list_for_each_entry(a, &actions, list) {
1949                 if (is_tcf_gact_shot(a)) {
1950                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
1951                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1952                         continue;
1953                 }
1954
1955                 if (is_tcf_pedit(a)) {
1956                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_FDB,
1957                                                     parse_attr);
1958                         if (err)
1959                                 return err;
1960
1961                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
1962                         continue;
1963                 }
1964
1965                 if (is_tcf_csum(a)) {
1966                         if (csum_offload_supported(priv, attr->action,
1967                                                    tcf_csum_update_flags(a)))
1968                                 continue;
1969
1970                         return -EOPNOTSUPP;
1971                 }
1972
1973                 if (is_tcf_mirred_egress_redirect(a)) {
1974                         int ifindex = tcf_mirred_ifindex(a);
1975                         struct net_device *out_dev;
1976                         struct mlx5e_priv *out_priv;
1977
1978                         out_dev = __dev_get_by_index(dev_net(priv->netdev), ifindex);
1979
1980                         if (switchdev_port_same_parent_id(priv->netdev,
1981                                                           out_dev)) {
1982                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1983                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1984                                 out_priv = netdev_priv(out_dev);
1985                                 rpriv = out_priv->ppriv;
1986                                 attr->out_rep = rpriv->rep;
1987                         } else if (encap) {
1988                                 parse_attr->mirred_ifindex = ifindex;
1989                                 parse_attr->tun_info = *info;
1990                                 attr->parse_attr = parse_attr;
1991                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_ENCAP |
1992                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1993                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1994                                 /* attr->out_rep is resolved when we handle encap */
1995                         } else {
1996                                 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
1997                                        priv->netdev->name, out_dev->name);
1998                                 return -EINVAL;
1999                         }
2000                         continue;
2001                 }
2002
2003                 if (is_tcf_tunnel_set(a)) {
2004                         info = tcf_tunnel_info(a);
2005                         if (info)
2006                                 encap = true;
2007                         else
2008                                 return -EOPNOTSUPP;
2009                         continue;
2010                 }
2011
2012                 if (is_tcf_vlan(a)) {
2013                         if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
2014                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2015                         } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
2016                                 if (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q))
2017                                         return -EOPNOTSUPP;
2018
2019                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
2020                                 attr->vlan = tcf_vlan_push_vid(a);
2021                         } else { /* action is TCA_VLAN_ACT_MODIFY */
2022                                 return -EOPNOTSUPP;
2023                         }
2024                         continue;
2025                 }
2026
2027                 if (is_tcf_tunnel_release(a)) {
2028                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
2029                         continue;
2030                 }
2031
2032                 return -EINVAL;
2033         }
2034
2035         if (!actions_match_supported(priv, exts, parse_attr, flow))
2036                 return -EOPNOTSUPP;
2037
2038         return err;
2039 }
2040
2041 int mlx5e_configure_flower(struct mlx5e_priv *priv,
2042                            struct tc_cls_flower_offload *f)
2043 {
2044         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2045         struct mlx5e_tc_flow_parse_attr *parse_attr;
2046         struct mlx5e_tc_table *tc = &priv->fs.tc;
2047         struct mlx5e_tc_flow *flow;
2048         int attr_size, err = 0;
2049         u8 flow_flags = 0;
2050
2051         if (esw && esw->mode == SRIOV_OFFLOADS) {
2052                 flow_flags = MLX5E_TC_FLOW_ESWITCH;
2053                 attr_size  = sizeof(struct mlx5_esw_flow_attr);
2054         } else {
2055                 flow_flags = MLX5E_TC_FLOW_NIC;
2056                 attr_size  = sizeof(struct mlx5_nic_flow_attr);
2057         }
2058
2059         flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
2060         parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
2061         if (!parse_attr || !flow) {
2062                 err = -ENOMEM;
2063                 goto err_free;
2064         }
2065
2066         flow->cookie = f->cookie;
2067         flow->flags = flow_flags;
2068
2069         err = parse_cls_flower(priv, flow, &parse_attr->spec, f);
2070         if (err < 0)
2071                 goto err_free;
2072
2073         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
2074                 err = parse_tc_fdb_actions(priv, f->exts, parse_attr, flow);
2075                 if (err < 0)
2076                         goto err_free;
2077                 flow->rule = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow);
2078         } else {
2079                 err = parse_tc_nic_actions(priv, f->exts, parse_attr, flow);
2080                 if (err < 0)
2081                         goto err_free;
2082                 flow->rule = mlx5e_tc_add_nic_flow(priv, parse_attr, flow);
2083         }
2084
2085         if (IS_ERR(flow->rule)) {
2086                 err = PTR_ERR(flow->rule);
2087                 if (err != -EAGAIN)
2088                         goto err_free;
2089         }
2090
2091         if (err != -EAGAIN)
2092                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
2093
2094         err = rhashtable_insert_fast(&tc->ht, &flow->node,
2095                                      tc->ht_params);
2096         if (err)
2097                 goto err_del_rule;
2098
2099         if (flow->flags & MLX5E_TC_FLOW_ESWITCH &&
2100             !(flow->esw_attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP))
2101                 kvfree(parse_attr);
2102         return err;
2103
2104 err_del_rule:
2105         mlx5e_tc_del_flow(priv, flow);
2106
2107 err_free:
2108         kvfree(parse_attr);
2109         kfree(flow);
2110         return err;
2111 }
2112
2113 int mlx5e_delete_flower(struct mlx5e_priv *priv,
2114                         struct tc_cls_flower_offload *f)
2115 {
2116         struct mlx5e_tc_flow *flow;
2117         struct mlx5e_tc_table *tc = &priv->fs.tc;
2118
2119         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
2120                                       tc->ht_params);
2121         if (!flow)
2122                 return -EINVAL;
2123
2124         rhashtable_remove_fast(&tc->ht, &flow->node, tc->ht_params);
2125
2126         mlx5e_tc_del_flow(priv, flow);
2127
2128         kfree(flow);
2129
2130         return 0;
2131 }
2132
2133 int mlx5e_stats_flower(struct mlx5e_priv *priv,
2134                        struct tc_cls_flower_offload *f)
2135 {
2136         struct mlx5e_tc_table *tc = &priv->fs.tc;
2137         struct mlx5e_tc_flow *flow;
2138         struct mlx5_fc *counter;
2139         u64 bytes;
2140         u64 packets;
2141         u64 lastuse;
2142
2143         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
2144                                       tc->ht_params);
2145         if (!flow)
2146                 return -EINVAL;
2147
2148         if (!(flow->flags & MLX5E_TC_FLOW_OFFLOADED))
2149                 return 0;
2150
2151         counter = mlx5_flow_rule_counter(flow->rule);
2152         if (!counter)
2153                 return 0;
2154
2155         mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
2156
2157         tcf_exts_stats_update(f->exts, bytes, packets, lastuse);
2158
2159         return 0;
2160 }
2161
2162 static const struct rhashtable_params mlx5e_tc_flow_ht_params = {
2163         .head_offset = offsetof(struct mlx5e_tc_flow, node),
2164         .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
2165         .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
2166         .automatic_shrinking = true,
2167 };
2168
2169 int mlx5e_tc_init(struct mlx5e_priv *priv)
2170 {
2171         struct mlx5e_tc_table *tc = &priv->fs.tc;
2172
2173         hash_init(tc->mod_hdr_tbl);
2174
2175         tc->ht_params = mlx5e_tc_flow_ht_params;
2176         return rhashtable_init(&tc->ht, &tc->ht_params);
2177 }
2178
2179 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
2180 {
2181         struct mlx5e_tc_flow *flow = ptr;
2182         struct mlx5e_priv *priv = arg;
2183
2184         mlx5e_tc_del_flow(priv, flow);
2185         kfree(flow);
2186 }
2187
2188 void mlx5e_tc_cleanup(struct mlx5e_priv *priv)
2189 {
2190         struct mlx5e_tc_table *tc = &priv->fs.tc;
2191
2192         rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, priv);
2193
2194         if (!IS_ERR_OR_NULL(tc->t)) {
2195                 mlx5_destroy_flow_table(tc->t);
2196                 tc->t = NULL;
2197         }
2198 }