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/vxlan.h>
  47 #include <net/arp.h>
  48 #include "en.h"
  49 #include "en_rep.h"
  50 #include "en_tc.h"
  51 #include "eswitch.h"
  52 #include "vxlan.h"
  53
  54 struct mlx5_nic_flow_attr {
  55         u32 action;
  56         u32 flow_tag;
  57         u32 mod_hdr_id;
  58 };
  59
  60 enum {
  61         MLX5E_TC_FLOW_ESWITCH   = BIT(0),
  62         MLX5E_TC_FLOW_NIC       = BIT(1),
  63         MLX5E_TC_FLOW_OFFLOADED = BIT(2),
  64 };
  65
  66 struct mlx5e_tc_flow {
  67         struct rhash_head       node;
  68         u64                     cookie;
  69         u8                      flags;
  70         struct mlx5_flow_handle *rule;
  71         struct list_head        encap; /* flows sharing the same encap */
  72         union {
  73                 struct mlx5_esw_flow_attr esw_attr[0];
  74                 struct mlx5_nic_flow_attr nic_attr[0];
  75         };
  76 };
  77
  78 struct mlx5e_tc_flow_parse_attr {
  79         struct mlx5_flow_spec spec;
  80         int num_mod_hdr_actions;
  81         void *mod_hdr_actions;
  82 };
  83
  84 enum {
  85         MLX5_HEADER_TYPE_VXLAN = 0x0,
  86         MLX5_HEADER_TYPE_NVGRE = 0x1,
  87 };
  88
  89 #define MLX5E_TC_TABLE_NUM_ENTRIES 1024
  90 #define MLX5E_TC_TABLE_NUM_GROUPS 4
  91
  92 static struct mlx5_flow_handle *
  93 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
  94                       struct mlx5e_tc_flow_parse_attr *parse_attr,
  95                       struct mlx5e_tc_flow *flow)
  96 {
  97         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
  98         struct mlx5_core_dev *dev = priv->mdev;
  99         struct mlx5_flow_destination dest = {};
 100         struct mlx5_flow_act flow_act = {
 101                 .action = attr->action,
 102                 .flow_tag = attr->flow_tag,
 103                 .encap_id = 0,
 104         };
 105         struct mlx5_fc *counter = NULL;
 106         struct mlx5_flow_handle *rule;
 107         bool table_created = false;
 108         int err;
 109
 110         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
 111                 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 112                 dest.ft = priv->fs.vlan.ft.t;
 113         } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
 114                 counter = mlx5_fc_create(dev, true);
 115                 if (IS_ERR(counter))
 116                         return ERR_CAST(counter);
 117
 118                 dest.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
 119                 dest.counter = counter;
 120         }
 121
 122         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
 123                 err = mlx5_modify_header_alloc(dev, MLX5_FLOW_NAMESPACE_KERNEL,
 124                                                parse_attr->num_mod_hdr_actions,
 125                                                parse_attr->mod_hdr_actions,
 126                                                &attr->mod_hdr_id);
 127                 flow_act.modify_id = attr->mod_hdr_id;
 128                 kfree(parse_attr->mod_hdr_actions);
 129                 if (err) {
 130                         rule = ERR_PTR(err);
 131                         goto err_create_mod_hdr_id;
 132                 }
 133         }
 134
 135         if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
 136                 priv->fs.tc.t =
 137                         mlx5_create_auto_grouped_flow_table(priv->fs.ns,
 138                                                             MLX5E_TC_PRIO,
 139                                                             MLX5E_TC_TABLE_NUM_ENTRIES,
 140                                                             MLX5E_TC_TABLE_NUM_GROUPS,
 141                                                             0, 0);
 142                 if (IS_ERR(priv->fs.tc.t)) {
 143                         netdev_err(priv->netdev,
 144                                    "Failed to create tc offload table\n");
 145                         rule = ERR_CAST(priv->fs.tc.t);
 146                         goto err_create_ft;
 147                 }
 148
 149                 table_created = true;
 150         }
 151
 152         parse_attr->spec.match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
 153         rule = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
 154                                    &flow_act, &dest, 1);
 155
 156         if (IS_ERR(rule))
 157                 goto err_add_rule;
 158
 159         return rule;
 160
 161 err_add_rule:
 162         if (table_created) {
 163                 mlx5_destroy_flow_table(priv->fs.tc.t);
 164                 priv->fs.tc.t = NULL;
 165         }
 166 err_create_ft:
 167         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 168                 mlx5_modify_header_dealloc(priv->mdev,
 169                                            attr->mod_hdr_id);
 170 err_create_mod_hdr_id:
 171         mlx5_fc_destroy(dev, counter);
 172
 173         return rule;
 174 }
 175
 176 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
 177                                   struct mlx5e_tc_flow *flow)
 178 {
 179         struct mlx5_fc *counter = NULL;
 180
 181         counter = mlx5_flow_rule_counter(flow->rule);
 182         mlx5_del_flow_rules(flow->rule);
 183         mlx5_fc_destroy(priv->mdev, counter);
 184
 185         if (!mlx5e_tc_num_filters(priv) && (priv->fs.tc.t)) {
 186                 mlx5_destroy_flow_table(priv->fs.tc.t);
 187                 priv->fs.tc.t = NULL;
 188         }
 189
 190         if (flow->nic_attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 191                 mlx5_modify_header_dealloc(priv->mdev,
 192                                            flow->nic_attr->mod_hdr_id);
 193 }
 194
 195 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
 196                                struct mlx5e_tc_flow *flow);
 197
 198 static struct mlx5_flow_handle *
 199 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
 200                       struct mlx5e_tc_flow_parse_attr *parse_attr,
 201                       struct mlx5e_tc_flow *flow)
 202 {
 203         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 204         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
 205         struct mlx5_flow_handle *rule;
 206         int err;
 207
 208         err = mlx5_eswitch_add_vlan_action(esw, attr);
 209         if (err) {
 210                 rule = ERR_PTR(err);
 211                 goto err_add_vlan;
 212         }
 213
 214         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
 215                 err = mlx5_modify_header_alloc(priv->mdev, MLX5_FLOW_NAMESPACE_FDB,
 216                                                parse_attr->num_mod_hdr_actions,
 217                                                parse_attr->mod_hdr_actions,
 218                                                &attr->mod_hdr_id);
 219                 kfree(parse_attr->mod_hdr_actions);
 220                 if (err) {
 221                         rule = ERR_PTR(err);
 222                         goto err_mod_hdr;
 223                 }
 224         }
 225
 226         rule = mlx5_eswitch_add_offloaded_rule(esw, &parse_attr->spec, attr);
 227         if (IS_ERR(rule))
 228                 goto err_add_rule;
 229
 230         return rule;
 231
 232 err_add_rule:
 233         if (flow->esw_attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 234                 mlx5_modify_header_dealloc(priv->mdev,
 235                                            attr->mod_hdr_id);
 236 err_mod_hdr:
 237         mlx5_eswitch_del_vlan_action(esw, attr);
 238 err_add_vlan:
 239         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP)
 240                 mlx5e_detach_encap(priv, flow);
 241         return rule;
 242 }
 243
 244 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
 245                                   struct mlx5e_tc_flow *flow)
 246 {
 247         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 248         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
 249
 250         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
 251                 flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
 252                 mlx5_eswitch_del_offloaded_rule(esw, flow->rule, flow->esw_attr);
 253         }
 254
 255         mlx5_eswitch_del_vlan_action(esw, flow->esw_attr);
 256
 257         if (flow->esw_attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP) {
 258                 mlx5e_detach_encap(priv, flow);
 259                 kvfree(flow->esw_attr->parse_attr);
 260         }
 261
 262         if (flow->esw_attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 263                 mlx5_modify_header_dealloc(priv->mdev,
 264                                            attr->mod_hdr_id);
 265 }
 266
 267 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
 268                               struct mlx5e_encap_entry *e)
 269 {
 270         struct mlx5e_tc_flow *flow;
 271         int err;
 272
 273         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
 274                                e->encap_size, e->encap_header,
 275                                &e->encap_id);
 276         if (err) {
 277                 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %d\n",
 278                                err);
 279                 return;
 280         }
 281         e->flags |= MLX5_ENCAP_ENTRY_VALID;
 282         mlx5e_rep_queue_neigh_stats_work(priv);
 283
 284         list_for_each_entry(flow, &e->flows, encap) {
 285                 flow->esw_attr->encap_id = e->encap_id;
 286                 flow->rule = mlx5e_tc_add_fdb_flow(priv,
 287                                                    flow->esw_attr->parse_attr,
 288                                                    flow);
 289                 if (IS_ERR(flow->rule)) {
 290                         err = PTR_ERR(flow->rule);
 291                         mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
 292                                        err);
 293                         continue;
 294                 }
 295                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
 296         }
 297 }
 298
 299 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
 300                               struct mlx5e_encap_entry *e)
 301 {
 302         struct mlx5e_tc_flow *flow;
 303         struct mlx5_fc *counter;
 304
 305         list_for_each_entry(flow, &e->flows, encap) {
 306                 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
 307                         flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
 308                         counter = mlx5_flow_rule_counter(flow->rule);
 309                         mlx5_del_flow_rules(flow->rule);
 310                         mlx5_fc_destroy(priv->mdev, counter);
 311                 }
 312         }
 313
 314         if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
 315                 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
 316                 mlx5_encap_dealloc(priv->mdev, e->encap_id);
 317         }
 318 }
 319
 320 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
 321 {
 322         struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
 323         u64 bytes, packets, lastuse = 0;
 324         struct mlx5e_tc_flow *flow;
 325         struct mlx5e_encap_entry *e;
 326         struct mlx5_fc *counter;
 327         struct neigh_table *tbl;
 328         bool neigh_used = false;
 329         struct neighbour *n;
 330
 331         if (m_neigh->family == AF_INET)
 332                 tbl = &arp_tbl;
 333 #if IS_ENABLED(CONFIG_IPV6)
 334         else if (m_neigh->family == AF_INET6)
 335                 tbl = ipv6_stub->nd_tbl;
 336 #endif
 337         else
 338                 return;
 339
 340         list_for_each_entry(e, &nhe->encap_list, encap_list) {
 341                 if (!(e->flags & MLX5_ENCAP_ENTRY_VALID))
 342                         continue;
 343                 list_for_each_entry(flow, &e->flows, encap) {
 344                         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
 345                                 counter = mlx5_flow_rule_counter(flow->rule);
 346                                 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
 347                                 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
 348                                         neigh_used = true;
 349                                         break;
 350                                 }
 351                         }
 352                 }
 353         }
 354
 355         if (neigh_used) {
 356                 nhe->reported_lastuse = jiffies;
 357
 358                 /* find the relevant neigh according to the cached device and
 359                  * dst ip pair
 360                  */
 361                 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
 362                 if (!n) {
 363                         WARN(1, "The neighbour already freed\n");
 364                         return;
 365                 }
 366
 367                 neigh_event_send(n, NULL);
 368                 neigh_release(n);
 369         }
 370 }
 371
 372 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
 373                                struct mlx5e_tc_flow *flow)
 374 {
 375         struct list_head *next = flow->encap.next;
 376
 377         list_del(&flow->encap);
 378         if (list_empty(next)) {
 379                 struct mlx5e_encap_entry *e;
 380
 381                 e = list_entry(next, struct mlx5e_encap_entry, flows);
 382                 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
 383
 384                 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
 385                         mlx5_encap_dealloc(priv->mdev, e->encap_id);
 386
 387                 hlist_del_rcu(&e->encap_hlist);
 388                 kfree(e->encap_header);
 389                 kfree(e);
 390         }
 391 }
 392
 393 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
 394                               struct mlx5e_tc_flow *flow)
 395 {
 396         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
 397                 mlx5e_tc_del_fdb_flow(priv, flow);
 398         else
 399                 mlx5e_tc_del_nic_flow(priv, flow);
 400 }
 401
 402 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
 403                              struct tc_cls_flower_offload *f)
 404 {
 405         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 406                                        outer_headers);
 407         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 408                                        outer_headers);
 409         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 410                                     misc_parameters);
 411         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 412                                     misc_parameters);
 413
 414         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
 415         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
 416
 417         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
 418                 struct flow_dissector_key_keyid *key =
 419                         skb_flow_dissector_target(f->dissector,
 420                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
 421                                                   f->key);
 422                 struct flow_dissector_key_keyid *mask =
 423                         skb_flow_dissector_target(f->dissector,
 424                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
 425                                                   f->mask);
 426                 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
 427                          be32_to_cpu(mask->keyid));
 428                 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
 429                          be32_to_cpu(key->keyid));
 430         }
 431 }
 432
 433 static int parse_tunnel_attr(struct mlx5e_priv *priv,
 434                              struct mlx5_flow_spec *spec,
 435                              struct tc_cls_flower_offload *f)
 436 {
 437         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 438                                        outer_headers);
 439         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 440                                        outer_headers);
 441
 442         struct flow_dissector_key_control *enc_control =
 443                 skb_flow_dissector_target(f->dissector,
 444                                           FLOW_DISSECTOR_KEY_ENC_CONTROL,
 445                                           f->key);
 446
 447         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
 448                 struct flow_dissector_key_ports *key =
 449                         skb_flow_dissector_target(f->dissector,
 450                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
 451                                                   f->key);
 452                 struct flow_dissector_key_ports *mask =
 453                         skb_flow_dissector_target(f->dissector,
 454                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
 455                                                   f->mask);
 456                 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 457                 struct net_device *up_dev = mlx5_eswitch_get_uplink_netdev(esw);
 458                 struct mlx5e_priv *up_priv = netdev_priv(up_dev);
 459
 460                 /* Full udp dst port must be given */
 461                 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
 462                         goto vxlan_match_offload_err;
 463
 464                 if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->dst)) &&
 465                     MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
 466                         parse_vxlan_attr(spec, f);
 467                 else {
 468                         netdev_warn(priv->netdev,
 469                                     "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
 470                         return -EOPNOTSUPP;
 471                 }
 472
 473                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 474                          udp_dport, ntohs(mask->dst));
 475                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 476                          udp_dport, ntohs(key->dst));
 477
 478                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 479                          udp_sport, ntohs(mask->src));
 480                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 481                          udp_sport, ntohs(key->src));
 482         } else { /* udp dst port must be given */
 483 vxlan_match_offload_err:
 484                 netdev_warn(priv->netdev,
 485                             "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
 486                 return -EOPNOTSUPP;
 487         }
 488
 489         if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
 490                 struct flow_dissector_key_ipv4_addrs *key =
 491                         skb_flow_dissector_target(f->dissector,
 492                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
 493                                                   f->key);
 494                 struct flow_dissector_key_ipv4_addrs *mask =
 495                         skb_flow_dissector_target(f->dissector,
 496                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
 497                                                   f->mask);
 498                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 499                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
 500                          ntohl(mask->src));
 501                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 502                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
 503                          ntohl(key->src));
 504
 505                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 506                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
 507                          ntohl(mask->dst));
 508                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 509                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
 510                          ntohl(key->dst));
 511
 512                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
 513                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
 514         } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
 515                 struct flow_dissector_key_ipv6_addrs *key =
 516                         skb_flow_dissector_target(f->dissector,
 517                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
 518                                                   f->key);
 519                 struct flow_dissector_key_ipv6_addrs *mask =
 520                         skb_flow_dissector_target(f->dissector,
 521                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
 522                                                   f->mask);
 523
 524                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 525                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 526                        &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 527                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 528                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 529                        &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 530
 531                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 532                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 533                        &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 534                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 535                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 536                        &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 537
 538                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
 539                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
 540         }
 541
 542         /* Enforce DMAC when offloading incoming tunneled flows.
 543          * Flow counters require a match on the DMAC.
 544          */
 545         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
 546         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
 547         ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 548                                      dmac_47_16), priv->netdev->dev_addr);
 549
 550         /* let software handle IP fragments */
 551         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
 552         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
 553
 554         return 0;
 555 }
 556
 557 static int __parse_cls_flower(struct mlx5e_priv *priv,
 558                               struct mlx5_flow_spec *spec,
 559                               struct tc_cls_flower_offload *f,
 560                               u8 *min_inline)
 561 {
 562         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 563                                        outer_headers);
 564         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 565                                        outer_headers);
 566         u16 addr_type = 0;
 567         u8 ip_proto = 0;
 568
 569         *min_inline = MLX5_INLINE_MODE_L2;
 570
 571         if (f->dissector->used_keys &
 572             ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
 573               BIT(FLOW_DISSECTOR_KEY_BASIC) |
 574               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
 575               BIT(FLOW_DISSECTOR_KEY_VLAN) |
 576               BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
 577               BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
 578               BIT(FLOW_DISSECTOR_KEY_PORTS) |
 579               BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
 580               BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
 581               BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
 582               BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
 583               BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL))) {
 584                 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
 585                             f->dissector->used_keys);
 586                 return -EOPNOTSUPP;
 587         }
 588
 589         if ((dissector_uses_key(f->dissector,
 590                                 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
 591              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
 592              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
 593             dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
 594                 struct flow_dissector_key_control *key =
 595                         skb_flow_dissector_target(f->dissector,
 596                                                   FLOW_DISSECTOR_KEY_ENC_CONTROL,
 597                                                   f->key);
 598                 switch (key->addr_type) {
 599                 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
 600                 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
 601                         if (parse_tunnel_attr(priv, spec, f))
 602                                 return -EOPNOTSUPP;
 603                         break;
 604                 default:
 605                         return -EOPNOTSUPP;
 606                 }
 607
 608                 /* In decap flow, header pointers should point to the inner
 609                  * headers, outer header were already set by parse_tunnel_attr
 610                  */
 611                 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 612                                          inner_headers);
 613                 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 614                                          inner_headers);
 615         }
 616
 617         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
 618                 struct flow_dissector_key_control *key =
 619                         skb_flow_dissector_target(f->dissector,
 620                                                   FLOW_DISSECTOR_KEY_CONTROL,
 621                                                   f->key);
 622
 623                 struct flow_dissector_key_control *mask =
 624                         skb_flow_dissector_target(f->dissector,
 625                                                   FLOW_DISSECTOR_KEY_CONTROL,
 626                                                   f->mask);
 627                 addr_type = key->addr_type;
 628
 629                 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
 630                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
 631                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
 632                                  key->flags & FLOW_DIS_IS_FRAGMENT);
 633
 634                         /* the HW doesn't need L3 inline to match on frag=no */
 635                         if (key->flags & FLOW_DIS_IS_FRAGMENT)
 636                                 *min_inline = MLX5_INLINE_MODE_IP;
 637                 }
 638         }
 639
 640         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
 641                 struct flow_dissector_key_basic *key =
 642                         skb_flow_dissector_target(f->dissector,
 643                                                   FLOW_DISSECTOR_KEY_BASIC,
 644                                                   f->key);
 645                 struct flow_dissector_key_basic *mask =
 646                         skb_flow_dissector_target(f->dissector,
 647                                                   FLOW_DISSECTOR_KEY_BASIC,
 648                                                   f->mask);
 649                 ip_proto = key->ip_proto;
 650
 651                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
 652                          ntohs(mask->n_proto));
 653                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
 654                          ntohs(key->n_proto));
 655
 656                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
 657                          mask->ip_proto);
 658                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
 659                          key->ip_proto);
 660
 661                 if (mask->ip_proto)
 662                         *min_inline = MLX5_INLINE_MODE_IP;
 663         }
 664
 665         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
 666                 struct flow_dissector_key_eth_addrs *key =
 667                         skb_flow_dissector_target(f->dissector,
 668                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
 669                                                   f->key);
 670                 struct flow_dissector_key_eth_addrs *mask =
 671                         skb_flow_dissector_target(f->dissector,
 672                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
 673                                                   f->mask);
 674
 675                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 676                                              dmac_47_16),
 677                                 mask->dst);
 678                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 679                                              dmac_47_16),
 680                                 key->dst);
 681
 682                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 683                                              smac_47_16),
 684                                 mask->src);
 685                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 686                                              smac_47_16),
 687                                 key->src);
 688         }
 689
 690         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
 691                 struct flow_dissector_key_vlan *key =
 692                         skb_flow_dissector_target(f->dissector,
 693                                                   FLOW_DISSECTOR_KEY_VLAN,
 694                                                   f->key);
 695                 struct flow_dissector_key_vlan *mask =
 696                         skb_flow_dissector_target(f->dissector,
 697                                                   FLOW_DISSECTOR_KEY_VLAN,
 698                                                   f->mask);
 699                 if (mask->vlan_id || mask->vlan_priority) {
 700                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
 701                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
 702
 703                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
 704                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
 705
 706                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
 707                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
 708                 }
 709         }
 710
 711         if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
 712                 struct flow_dissector_key_ipv4_addrs *key =
 713                         skb_flow_dissector_target(f->dissector,
 714                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 715                                                   f->key);
 716                 struct flow_dissector_key_ipv4_addrs *mask =
 717                         skb_flow_dissector_target(f->dissector,
 718                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 719                                                   f->mask);
 720
 721                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 722                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
 723                        &mask->src, sizeof(mask->src));
 724                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 725                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
 726                        &key->src, sizeof(key->src));
 727                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 728                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 729                        &mask->dst, sizeof(mask->dst));
 730                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 731                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 732                        &key->dst, sizeof(key->dst));
 733
 734                 if (mask->src || mask->dst)
 735                         *min_inline = MLX5_INLINE_MODE_IP;
 736         }
 737
 738         if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
 739                 struct flow_dissector_key_ipv6_addrs *key =
 740                         skb_flow_dissector_target(f->dissector,
 741                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 742                                                   f->key);
 743                 struct flow_dissector_key_ipv6_addrs *mask =
 744                         skb_flow_dissector_target(f->dissector,
 745                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 746                                                   f->mask);
 747
 748                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 749                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 750                        &mask->src, sizeof(mask->src));
 751                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 752                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 753                        &key->src, sizeof(key->src));
 754
 755                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 756                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 757                        &mask->dst, sizeof(mask->dst));
 758                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 759                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 760                        &key->dst, sizeof(key->dst));
 761
 762                 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
 763                     ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
 764                         *min_inline = MLX5_INLINE_MODE_IP;
 765         }
 766
 767         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
 768                 struct flow_dissector_key_ports *key =
 769                         skb_flow_dissector_target(f->dissector,
 770                                                   FLOW_DISSECTOR_KEY_PORTS,
 771                                                   f->key);
 772                 struct flow_dissector_key_ports *mask =
 773                         skb_flow_dissector_target(f->dissector,
 774                                                   FLOW_DISSECTOR_KEY_PORTS,
 775                                                   f->mask);
 776                 switch (ip_proto) {
 777                 case IPPROTO_TCP:
 778                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 779                                  tcp_sport, ntohs(mask->src));
 780                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 781                                  tcp_sport, ntohs(key->src));
 782
 783                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 784                                  tcp_dport, ntohs(mask->dst));
 785                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 786                                  tcp_dport, ntohs(key->dst));
 787                         break;
 788
 789                 case IPPROTO_UDP:
 790                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 791                                  udp_sport, ntohs(mask->src));
 792                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 793                                  udp_sport, ntohs(key->src));
 794
 795                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 796                                  udp_dport, ntohs(mask->dst));
 797                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 798                                  udp_dport, ntohs(key->dst));
 799                         break;
 800                 default:
 801                         netdev_err(priv->netdev,
 802                                    "Only UDP and TCP transport are supported\n");
 803                         return -EINVAL;
 804                 }
 805
 806                 if (mask->src || mask->dst)
 807                         *min_inline = MLX5_INLINE_MODE_TCP_UDP;
 808         }
 809
 810         return 0;
 811 }
 812
 813 static int parse_cls_flower(struct mlx5e_priv *priv,
 814                             struct mlx5e_tc_flow *flow,
 815                             struct mlx5_flow_spec *spec,
 816                             struct tc_cls_flower_offload *f)
 817 {
 818         struct mlx5_core_dev *dev = priv->mdev;
 819         struct mlx5_eswitch *esw = dev->priv.eswitch;
 820         struct mlx5e_rep_priv *rpriv = priv->ppriv;
 821         struct mlx5_eswitch_rep *rep;
 822         u8 min_inline;
 823         int err;
 824
 825         err = __parse_cls_flower(priv, spec, f, &min_inline);
 826
 827         if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH)) {
 828                 rep = rpriv->rep;
 829                 if (rep->vport != FDB_UPLINK_VPORT &&
 830                     (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
 831                     esw->offloads.inline_mode < min_inline)) {
 832                         netdev_warn(priv->netdev,
 833                                     "Flow is not offloaded due to min inline setting, required %d actual %d\n",
 834                                     min_inline, esw->offloads.inline_mode);
 835                         return -EOPNOTSUPP;
 836                 }
 837         }
 838
 839         return err;
 840 }
 841
 842 struct pedit_headers {
 843         struct ethhdr  eth;
 844         struct iphdr   ip4;
 845         struct ipv6hdr ip6;
 846         struct tcphdr  tcp;
 847         struct udphdr  udp;
 848 };
 849
 850 static int pedit_header_offsets[] = {
 851         [TCA_PEDIT_KEY_EX_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
 852         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
 853         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
 854         [TCA_PEDIT_KEY_EX_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
 855         [TCA_PEDIT_KEY_EX_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
 856 };
 857
 858 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
 859
 860 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
 861                          struct pedit_headers *masks,
 862                          struct pedit_headers *vals)
 863 {
 864         u32 *curr_pmask, *curr_pval;
 865
 866         if (hdr_type >= __PEDIT_HDR_TYPE_MAX)
 867                 goto out_err;
 868
 869         curr_pmask = (u32 *)(pedit_header(masks, hdr_type) + offset);
 870         curr_pval  = (u32 *)(pedit_header(vals, hdr_type) + offset);
 871
 872         if (*curr_pmask & mask)  /* disallow acting twice on the same location */
 873                 goto out_err;
 874
 875         *curr_pmask |= mask;
 876         *curr_pval  |= (val & mask);
 877
 878         return 0;
 879
 880 out_err:
 881         return -EOPNOTSUPP;
 882 }
 883
 884 struct mlx5_fields {
 885         u8  field;
 886         u8  size;
 887         u32 offset;
 888 };
 889
 890 static struct mlx5_fields fields[] = {
 891         {MLX5_ACTION_IN_FIELD_OUT_DMAC_47_16, 4, offsetof(struct pedit_headers, eth.h_dest[0])},
 892         {MLX5_ACTION_IN_FIELD_OUT_DMAC_15_0,  2, offsetof(struct pedit_headers, eth.h_dest[4])},
 893         {MLX5_ACTION_IN_FIELD_OUT_SMAC_47_16, 4, offsetof(struct pedit_headers, eth.h_source[0])},
 894         {MLX5_ACTION_IN_FIELD_OUT_SMAC_15_0,  2, offsetof(struct pedit_headers, eth.h_source[4])},
 895         {MLX5_ACTION_IN_FIELD_OUT_ETHERTYPE,  2, offsetof(struct pedit_headers, eth.h_proto)},
 896
 897         {MLX5_ACTION_IN_FIELD_OUT_IP_DSCP, 1, offsetof(struct pedit_headers, ip4.tos)},
 898         {MLX5_ACTION_IN_FIELD_OUT_IP_TTL,  1, offsetof(struct pedit_headers, ip4.ttl)},
 899         {MLX5_ACTION_IN_FIELD_OUT_SIPV4,   4, offsetof(struct pedit_headers, ip4.saddr)},
 900         {MLX5_ACTION_IN_FIELD_OUT_DIPV4,   4, offsetof(struct pedit_headers, ip4.daddr)},
 901
 902         {MLX5_ACTION_IN_FIELD_OUT_SIPV6_127_96, 4, offsetof(struct pedit_headers, ip6.saddr.s6_addr32[0])},
 903         {MLX5_ACTION_IN_FIELD_OUT_SIPV6_95_64,  4, offsetof(struct pedit_headers, ip6.saddr.s6_addr32[1])},
 904         {MLX5_ACTION_IN_FIELD_OUT_SIPV6_63_32,  4, offsetof(struct pedit_headers, ip6.saddr.s6_addr32[2])},
 905         {MLX5_ACTION_IN_FIELD_OUT_SIPV6_31_0,   4, offsetof(struct pedit_headers, ip6.saddr.s6_addr32[3])},
 906         {MLX5_ACTION_IN_FIELD_OUT_DIPV6_127_96, 4, offsetof(struct pedit_headers, ip6.daddr.s6_addr32[0])},
 907         {MLX5_ACTION_IN_FIELD_OUT_DIPV6_95_64,  4, offsetof(struct pedit_headers, ip6.daddr.s6_addr32[1])},
 908         {MLX5_ACTION_IN_FIELD_OUT_DIPV6_63_32,  4, offsetof(struct pedit_headers, ip6.daddr.s6_addr32[2])},
 909         {MLX5_ACTION_IN_FIELD_OUT_DIPV6_31_0,   4, offsetof(struct pedit_headers, ip6.daddr.s6_addr32[3])},
 910
 911         {MLX5_ACTION_IN_FIELD_OUT_TCP_SPORT, 2, offsetof(struct pedit_headers, tcp.source)},
 912         {MLX5_ACTION_IN_FIELD_OUT_TCP_DPORT, 2, offsetof(struct pedit_headers, tcp.dest)},
 913         {MLX5_ACTION_IN_FIELD_OUT_TCP_FLAGS, 1, offsetof(struct pedit_headers, tcp.ack_seq) + 5},
 914
 915         {MLX5_ACTION_IN_FIELD_OUT_UDP_SPORT, 2, offsetof(struct pedit_headers, udp.source)},
 916         {MLX5_ACTION_IN_FIELD_OUT_UDP_DPORT, 2, offsetof(struct pedit_headers, udp.dest)},
 917 };
 918
 919 /* On input attr->num_mod_hdr_actions tells how many HW actions can be parsed at
 920  * max from the SW pedit action. On success, it says how many HW actions were
 921  * actually parsed.
 922  */
 923 static int offload_pedit_fields(struct pedit_headers *masks,
 924                                 struct pedit_headers *vals,
 925                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
 926 {
 927         struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
 928         int i, action_size, nactions, max_actions, first, last;
 929         void *s_masks_p, *a_masks_p, *vals_p;
 930         u32 s_mask, a_mask, val;
 931         struct mlx5_fields *f;
 932         u8 cmd, field_bsize;
 933         unsigned long mask;
 934         void *action;
 935
 936         set_masks = &masks[TCA_PEDIT_KEY_EX_CMD_SET];
 937         add_masks = &masks[TCA_PEDIT_KEY_EX_CMD_ADD];
 938         set_vals = &vals[TCA_PEDIT_KEY_EX_CMD_SET];
 939         add_vals = &vals[TCA_PEDIT_KEY_EX_CMD_ADD];
 940
 941         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
 942         action = parse_attr->mod_hdr_actions;
 943         max_actions = parse_attr->num_mod_hdr_actions;
 944         nactions = 0;
 945
 946         for (i = 0; i < ARRAY_SIZE(fields); i++) {
 947                 f = &fields[i];
 948                 /* avoid seeing bits set from previous iterations */
 949                 s_mask = a_mask = mask = val = 0;
 950
 951                 s_masks_p = (void *)set_masks + f->offset;
 952                 a_masks_p = (void *)add_masks + f->offset;
 953
 954                 memcpy(&s_mask, s_masks_p, f->size);
 955                 memcpy(&a_mask, a_masks_p, f->size);
 956
 957                 if (!s_mask && !a_mask) /* nothing to offload here */
 958                         continue;
 959
 960                 if (s_mask && a_mask) {
 961                         printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
 962                         return -EOPNOTSUPP;
 963                 }
 964
 965                 if (nactions == max_actions) {
 966                         printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
 967                         return -EOPNOTSUPP;
 968                 }
 969
 970                 if (s_mask) {
 971                         cmd  = MLX5_ACTION_TYPE_SET;
 972                         mask = s_mask;
 973                         vals_p = (void *)set_vals + f->offset;
 974                         /* clear to denote we consumed this field */
 975                         memset(s_masks_p, 0, f->size);
 976                 } else {
 977                         cmd  = MLX5_ACTION_TYPE_ADD;
 978                         mask = a_mask;
 979                         vals_p = (void *)add_vals + f->offset;
 980                         /* clear to denote we consumed this field */
 981                         memset(a_masks_p, 0, f->size);
 982                 }
 983
 984                 memcpy(&val, vals_p, f->size);
 985
 986                 field_bsize = f->size * BITS_PER_BYTE;
 987                 first = find_first_bit(&mask, field_bsize);
 988                 last  = find_last_bit(&mask, field_bsize);
 989                 if (first > 0 || last != (field_bsize - 1)) {
 990                         printk(KERN_WARNING "mlx5: partial rewrite (mask %lx) is currently not offloaded\n",
 991                                mask);
 992                         return -EOPNOTSUPP;
 993                 }
 994
 995                 MLX5_SET(set_action_in, action, action_type, cmd);
 996                 MLX5_SET(set_action_in, action, field, f->field);
 997
 998                 if (cmd == MLX5_ACTION_TYPE_SET) {
 999                         MLX5_SET(set_action_in, action, offset, 0);
1000                         /* length is num of bits to be written, zero means length of 32 */
1001                         MLX5_SET(set_action_in, action, length, field_bsize);
1002                 }
1003
1004                 if (field_bsize == 32)
1005                         MLX5_SET(set_action_in, action, data, ntohl(val));
1006                 else if (field_bsize == 16)
1007                         MLX5_SET(set_action_in, action, data, ntohs(val));
1008                 else if (field_bsize == 8)
1009                         MLX5_SET(set_action_in, action, data, val);
1010
1011                 action += action_size;
1012                 nactions++;
1013         }
1014
1015         parse_attr->num_mod_hdr_actions = nactions;
1016         return 0;
1017 }
1018
1019 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
1020                                  const struct tc_action *a, int namespace,
1021                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1022 {
1023         int nkeys, action_size, max_actions;
1024
1025         nkeys = tcf_pedit_nkeys(a);
1026         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1027
1028         if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
1029                 max_actions = MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev, max_modify_header_actions);
1030         else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
1031                 max_actions = MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, max_modify_header_actions);
1032
1033         /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
1034         max_actions = min(max_actions, nkeys * 16);
1035
1036         parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
1037         if (!parse_attr->mod_hdr_actions)
1038                 return -ENOMEM;
1039
1040         parse_attr->num_mod_hdr_actions = max_actions;
1041         return 0;
1042 }
1043
1044 static const struct pedit_headers zero_masks = {};
1045
1046 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
1047                                  const struct tc_action *a, int namespace,
1048                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1049 {
1050         struct pedit_headers masks[__PEDIT_CMD_MAX], vals[__PEDIT_CMD_MAX], *cmd_masks;
1051         int nkeys, i, err = -EOPNOTSUPP;
1052         u32 mask, val, offset;
1053         u8 cmd, htype;
1054
1055         nkeys = tcf_pedit_nkeys(a);
1056
1057         memset(masks, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1058         memset(vals,  0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1059
1060         for (i = 0; i < nkeys; i++) {
1061                 htype = tcf_pedit_htype(a, i);
1062                 cmd = tcf_pedit_cmd(a, i);
1063                 err = -EOPNOTSUPP; /* can't be all optimistic */
1064
1065                 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK) {
1066                         printk(KERN_WARNING "mlx5: legacy pedit isn't offloaded\n");
1067                         goto out_err;
1068                 }
1069
1070                 if (cmd != TCA_PEDIT_KEY_EX_CMD_SET && cmd != TCA_PEDIT_KEY_EX_CMD_ADD) {
1071                         printk(KERN_WARNING "mlx5: pedit cmd %d isn't offloaded\n", cmd);
1072                         goto out_err;
1073                 }
1074
1075                 mask = tcf_pedit_mask(a, i);
1076                 val = tcf_pedit_val(a, i);
1077                 offset = tcf_pedit_offset(a, i);
1078
1079                 err = set_pedit_val(htype, ~mask, val, offset, &masks[cmd], &vals[cmd]);
1080                 if (err)
1081                         goto out_err;
1082         }
1083
1084         err = alloc_mod_hdr_actions(priv, a, namespace, parse_attr);
1085         if (err)
1086                 goto out_err;
1087
1088         err = offload_pedit_fields(masks, vals, parse_attr);
1089         if (err < 0)
1090                 goto out_dealloc_parsed_actions;
1091
1092         for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
1093                 cmd_masks = &masks[cmd];
1094                 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
1095                         printk(KERN_WARNING "mlx5: attempt to offload an unsupported field (cmd %d)\n",
1096                                cmd);
1097                         print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
1098                                        16, 1, cmd_masks, sizeof(zero_masks), true);
1099                         err = -EOPNOTSUPP;
1100                         goto out_dealloc_parsed_actions;
1101                 }
1102         }
1103
1104         return 0;
1105
1106 out_dealloc_parsed_actions:
1107         kfree(parse_attr->mod_hdr_actions);
1108 out_err:
1109         return err;
1110 }
1111
1112 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
1113                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
1114                                 struct mlx5e_tc_flow *flow)
1115 {
1116         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
1117         const struct tc_action *a;
1118         LIST_HEAD(actions);
1119         int err;
1120
1121         if (tc_no_actions(exts))
1122                 return -EINVAL;
1123
1124         attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
1125         attr->action = 0;
1126
1127         tcf_exts_to_list(exts, &actions);
1128         list_for_each_entry(a, &actions, list) {
1129                 /* Only support a single action per rule */
1130                 if (attr->action)
1131                         return -EINVAL;
1132
1133                 if (is_tcf_gact_shot(a)) {
1134                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
1135                         if (MLX5_CAP_FLOWTABLE(priv->mdev,
1136                                                flow_table_properties_nic_receive.flow_counter))
1137                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
1138                         continue;
1139                 }
1140
1141                 if (is_tcf_pedit(a)) {
1142                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_KERNEL,
1143                                                     parse_attr);
1144                         if (err)
1145                                 return err;
1146
1147                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
1148                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1149                         continue;
1150                 }
1151
1152                 if (is_tcf_skbedit_mark(a)) {
1153                         u32 mark = tcf_skbedit_mark(a);
1154
1155                         if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
1156                                 netdev_warn(priv->netdev, "Bad flow mark - only 16 bit is supported: 0x%x\n",
1157                                             mark);
1158                                 return -EINVAL;
1159                         }
1160
1161                         attr->flow_tag = mark;
1162                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1163                         continue;
1164                 }
1165
1166                 return -EINVAL;
1167         }
1168
1169         return 0;
1170 }
1171
1172 static inline int cmp_encap_info(struct ip_tunnel_key *a,
1173                                  struct ip_tunnel_key *b)
1174 {
1175         return memcmp(a, b, sizeof(*a));
1176 }
1177
1178 static inline int hash_encap_info(struct ip_tunnel_key *key)
1179 {
1180         return jhash(key, sizeof(*key), 0);
1181 }
1182
1183 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
1184                                    struct net_device *mirred_dev,
1185                                    struct net_device **out_dev,
1186                                    struct flowi4 *fl4,
1187                                    struct neighbour **out_n,
1188                                    int *out_ttl)
1189 {
1190         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1191         struct rtable *rt;
1192         struct neighbour *n = NULL;
1193
1194 #if IS_ENABLED(CONFIG_INET)
1195         int ret;
1196
1197         rt = ip_route_output_key(dev_net(mirred_dev), fl4);
1198         ret = PTR_ERR_OR_ZERO(rt);
1199         if (ret)
1200                 return ret;
1201 #else
1202         return -EOPNOTSUPP;
1203 #endif
1204         /* if the egress device isn't on the same HW e-switch, we use the uplink */
1205         if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
1206                 *out_dev = mlx5_eswitch_get_uplink_netdev(esw);
1207         else
1208                 *out_dev = rt->dst.dev;
1209
1210         *out_ttl = ip4_dst_hoplimit(&rt->dst);
1211         n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
1212         ip_rt_put(rt);
1213         if (!n)
1214                 return -ENOMEM;
1215
1216         *out_n = n;
1217         return 0;
1218 }
1219
1220 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
1221                                    struct net_device *mirred_dev,
1222                                    struct net_device **out_dev,
1223                                    struct flowi6 *fl6,
1224                                    struct neighbour **out_n,
1225                                    int *out_ttl)
1226 {
1227         struct neighbour *n = NULL;
1228         struct dst_entry *dst;
1229
1230 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
1231         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1232         int ret;
1233
1234         dst = ip6_route_output(dev_net(mirred_dev), NULL, fl6);
1235         ret = dst->error;
1236         if (ret) {
1237                 dst_release(dst);
1238                 return ret;
1239         }
1240
1241         *out_ttl = ip6_dst_hoplimit(dst);
1242
1243         /* if the egress device isn't on the same HW e-switch, we use the uplink */
1244         if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
1245                 *out_dev = mlx5_eswitch_get_uplink_netdev(esw);
1246         else
1247                 *out_dev = dst->dev;
1248 #else
1249         return -EOPNOTSUPP;
1250 #endif
1251
1252         n = dst_neigh_lookup(dst, &fl6->daddr);
1253         dst_release(dst);
1254         if (!n)
1255                 return -ENOMEM;
1256
1257         *out_n = n;
1258         return 0;
1259 }
1260
1261 static void gen_vxlan_header_ipv4(struct net_device *out_dev,
1262                                   char buf[], int encap_size,
1263                                   unsigned char h_dest[ETH_ALEN],
1264                                   int ttl,
1265                                   __be32 daddr,
1266                                   __be32 saddr,
1267                                   __be16 udp_dst_port,
1268                                   __be32 vx_vni)
1269 {
1270         struct ethhdr *eth = (struct ethhdr *)buf;
1271         struct iphdr  *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
1272         struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
1273         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
1274
1275         memset(buf, 0, encap_size);
1276
1277         ether_addr_copy(eth->h_dest, h_dest);
1278         ether_addr_copy(eth->h_source, out_dev->dev_addr);
1279         eth->h_proto = htons(ETH_P_IP);
1280
1281         ip->daddr = daddr;
1282         ip->saddr = saddr;
1283
1284         ip->ttl = ttl;
1285         ip->protocol = IPPROTO_UDP;
1286         ip->version = 0x4;
1287         ip->ihl = 0x5;
1288
1289         udp->dest = udp_dst_port;
1290         vxh->vx_flags = VXLAN_HF_VNI;
1291         vxh->vx_vni = vxlan_vni_field(vx_vni);
1292 }
1293
1294 static void gen_vxlan_header_ipv6(struct net_device *out_dev,
1295                                   char buf[], int encap_size,
1296                                   unsigned char h_dest[ETH_ALEN],
1297                                   int ttl,
1298                                   struct in6_addr *daddr,
1299                                   struct in6_addr *saddr,
1300                                   __be16 udp_dst_port,
1301                                   __be32 vx_vni)
1302 {
1303         struct ethhdr *eth = (struct ethhdr *)buf;
1304         struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
1305         struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
1306         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
1307
1308         memset(buf, 0, encap_size);
1309
1310         ether_addr_copy(eth->h_dest, h_dest);
1311         ether_addr_copy(eth->h_source, out_dev->dev_addr);
1312         eth->h_proto = htons(ETH_P_IPV6);
1313
1314         ip6_flow_hdr(ip6h, 0, 0);
1315         /* the HW fills up ipv6 payload len */
1316         ip6h->nexthdr     = IPPROTO_UDP;
1317         ip6h->hop_limit   = ttl;
1318         ip6h->daddr       = *daddr;
1319         ip6h->saddr       = *saddr;
1320
1321         udp->dest = udp_dst_port;
1322         vxh->vx_flags = VXLAN_HF_VNI;
1323         vxh->vx_vni = vxlan_vni_field(vx_vni);
1324 }
1325
1326 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
1327                                           struct net_device *mirred_dev,
1328                                           struct mlx5e_encap_entry *e)
1329 {
1330         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
1331         int ipv4_encap_size = ETH_HLEN + sizeof(struct iphdr) + VXLAN_HLEN;
1332         struct ip_tunnel_key *tun_key = &e->tun_info.key;
1333         struct net_device *out_dev;
1334         struct neighbour *n = NULL;
1335         struct flowi4 fl4 = {};
1336         char *encap_header;
1337         int ttl, err;
1338         u8 nud_state;
1339
1340         if (max_encap_size < ipv4_encap_size) {
1341                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
1342                                ipv4_encap_size, max_encap_size);
1343                 return -EOPNOTSUPP;
1344         }
1345
1346         encap_header = kzalloc(ipv4_encap_size, GFP_KERNEL);
1347         if (!encap_header)
1348                 return -ENOMEM;
1349
1350         switch (e->tunnel_type) {
1351         case MLX5_HEADER_TYPE_VXLAN:
1352                 fl4.flowi4_proto = IPPROTO_UDP;
1353                 fl4.fl4_dport = tun_key->tp_dst;
1354                 break;
1355         default:
1356                 err = -EOPNOTSUPP;
1357                 goto out;
1358         }
1359         fl4.flowi4_tos = tun_key->tos;
1360         fl4.daddr = tun_key->u.ipv4.dst;
1361         fl4.saddr = tun_key->u.ipv4.src;
1362
1363         err = mlx5e_route_lookup_ipv4(priv, mirred_dev, &out_dev,
1364                                       &fl4, &n, &ttl);
1365         if (err)
1366                 goto out;
1367
1368         /* used by mlx5e_detach_encap to lookup a neigh hash table
1369          * entry in the neigh hash table when a user deletes a rule
1370          */
1371         e->m_neigh.dev = n->dev;
1372         e->m_neigh.family = n->ops->family;
1373         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
1374         e->out_dev = out_dev;
1375
1376         /* It's importent to add the neigh to the hash table before checking
1377          * the neigh validity state. So if we'll get a notification, in case the
1378          * neigh changes it's validity state, we would find the relevant neigh
1379          * in the hash.
1380          */
1381         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
1382         if (err)
1383                 goto out;
1384
1385         read_lock_bh(&n->lock);
1386         nud_state = n->nud_state;
1387         ether_addr_copy(e->h_dest, n->ha);
1388         read_unlock_bh(&n->lock);
1389
1390         switch (e->tunnel_type) {
1391         case MLX5_HEADER_TYPE_VXLAN:
1392                 gen_vxlan_header_ipv4(out_dev, encap_header,
1393                                       ipv4_encap_size, e->h_dest, ttl,
1394                                       fl4.daddr,
1395                                       fl4.saddr, tun_key->tp_dst,
1396                                       tunnel_id_to_key32(tun_key->tun_id));
1397                 break;
1398         default:
1399                 err = -EOPNOTSUPP;
1400                 goto destroy_neigh_entry;
1401         }
1402         e->encap_size = ipv4_encap_size;
1403         e->encap_header = encap_header;
1404
1405         if (!(nud_state & NUD_VALID)) {
1406                 neigh_event_send(n, NULL);
1407                 neigh_release(n);
1408                 return -EAGAIN;
1409         }
1410
1411         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
1412                                ipv4_encap_size, encap_header, &e->encap_id);
1413         if (err)
1414                 goto destroy_neigh_entry;
1415
1416         e->flags |= MLX5_ENCAP_ENTRY_VALID;
1417         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
1418         neigh_release(n);
1419         return err;
1420
1421 destroy_neigh_entry:
1422         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1423 out:
1424         kfree(encap_header);
1425         if (n)
1426                 neigh_release(n);
1427         return err;
1428 }
1429
1430 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
1431                                           struct net_device *mirred_dev,
1432                                           struct mlx5e_encap_entry *e)
1433 {
1434         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
1435         int ipv6_encap_size = ETH_HLEN + sizeof(struct ipv6hdr) + VXLAN_HLEN;
1436         struct ip_tunnel_key *tun_key = &e->tun_info.key;
1437         struct net_device *out_dev;
1438         struct neighbour *n = NULL;
1439         struct flowi6 fl6 = {};
1440         char *encap_header;
1441         int err, ttl = 0;
1442         u8 nud_state;
1443
1444         if (max_encap_size < ipv6_encap_size) {
1445                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
1446                                ipv6_encap_size, max_encap_size);
1447                 return -EOPNOTSUPP;
1448         }
1449
1450         encap_header = kzalloc(ipv6_encap_size, GFP_KERNEL);
1451         if (!encap_header)
1452                 return -ENOMEM;
1453
1454         switch (e->tunnel_type) {
1455         case MLX5_HEADER_TYPE_VXLAN:
1456                 fl6.flowi6_proto = IPPROTO_UDP;
1457                 fl6.fl6_dport = tun_key->tp_dst;
1458                 break;
1459         default:
1460                 err = -EOPNOTSUPP;
1461                 goto out;
1462         }
1463
1464         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
1465         fl6.daddr = tun_key->u.ipv6.dst;
1466         fl6.saddr = tun_key->u.ipv6.src;
1467
1468         err = mlx5e_route_lookup_ipv6(priv, mirred_dev, &out_dev,
1469                                       &fl6, &n, &ttl);
1470         if (err)
1471                 goto out;
1472
1473         /* used by mlx5e_detach_encap to lookup a neigh hash table
1474          * entry in the neigh hash table when a user deletes a rule
1475          */
1476         e->m_neigh.dev = n->dev;
1477         e->m_neigh.family = n->ops->family;
1478         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
1479         e->out_dev = out_dev;
1480
1481         /* It's importent to add the neigh to the hash table before checking
1482          * the neigh validity state. So if we'll get a notification, in case the
1483          * neigh changes it's validity state, we would find the relevant neigh
1484          * in the hash.
1485          */
1486         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
1487         if (err)
1488                 goto out;
1489
1490         read_lock_bh(&n->lock);
1491         nud_state = n->nud_state;
1492         ether_addr_copy(e->h_dest, n->ha);
1493         read_unlock_bh(&n->lock);
1494
1495         switch (e->tunnel_type) {
1496         case MLX5_HEADER_TYPE_VXLAN:
1497                 gen_vxlan_header_ipv6(out_dev, encap_header,
1498                                       ipv6_encap_size, e->h_dest, ttl,
1499                                       &fl6.daddr,
1500                                       &fl6.saddr, tun_key->tp_dst,
1501                                       tunnel_id_to_key32(tun_key->tun_id));
1502                 break;
1503         default:
1504                 err = -EOPNOTSUPP;
1505                 goto destroy_neigh_entry;
1506         }
1507
1508         e->encap_size = ipv6_encap_size;
1509         e->encap_header = encap_header;
1510
1511         if (!(nud_state & NUD_VALID)) {
1512                 neigh_event_send(n, NULL);
1513                 neigh_release(n);
1514                 return -EAGAIN;
1515         }
1516
1517         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
1518                                ipv6_encap_size, encap_header, &e->encap_id);
1519         if (err)
1520                 goto destroy_neigh_entry;
1521
1522         e->flags |= MLX5_ENCAP_ENTRY_VALID;
1523         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
1524         neigh_release(n);
1525         return err;
1526
1527 destroy_neigh_entry:
1528         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1529 out:
1530         kfree(encap_header);
1531         if (n)
1532                 neigh_release(n);
1533         return err;
1534 }
1535
1536 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
1537                               struct ip_tunnel_info *tun_info,
1538                               struct net_device *mirred_dev,
1539                               struct net_device **encap_dev,
1540                               struct mlx5e_tc_flow *flow)
1541 {
1542         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1543         struct net_device *up_dev = mlx5_eswitch_get_uplink_netdev(esw);
1544         unsigned short family = ip_tunnel_info_af(tun_info);
1545         struct mlx5e_priv *up_priv = netdev_priv(up_dev);
1546         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1547         struct ip_tunnel_key *key = &tun_info->key;
1548         struct mlx5e_encap_entry *e;
1549         int tunnel_type, err = 0;
1550         uintptr_t hash_key;
1551         bool found = false;
1552
1553         /* udp dst port must be set */
1554         if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
1555                 goto vxlan_encap_offload_err;
1556
1557         /* setting udp src port isn't supported */
1558         if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
1559 vxlan_encap_offload_err:
1560                 netdev_warn(priv->netdev,
1561                             "must set udp dst port and not set udp src port\n");
1562                 return -EOPNOTSUPP;
1563         }
1564
1565         if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->tp_dst)) &&
1566             MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
1567                 tunnel_type = MLX5_HEADER_TYPE_VXLAN;
1568         } else {
1569                 netdev_warn(priv->netdev,
1570                             "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
1571                 return -EOPNOTSUPP;
1572         }
1573
1574         hash_key = hash_encap_info(key);
1575
1576         hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
1577                                    encap_hlist, hash_key) {
1578                 if (!cmp_encap_info(&e->tun_info.key, key)) {
1579                         found = true;
1580                         break;
1581                 }
1582         }
1583
1584         if (found)
1585                 goto attach_flow;
1586
1587         e = kzalloc(sizeof(*e), GFP_KERNEL);
1588         if (!e)
1589                 return -ENOMEM;
1590
1591         e->tun_info = *tun_info;
1592         e->tunnel_type = tunnel_type;
1593         INIT_LIST_HEAD(&e->flows);
1594
1595         if (family == AF_INET)
1596                 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e);
1597         else if (family == AF_INET6)
1598                 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e);
1599
1600         if (err && err != -EAGAIN)
1601                 goto out_err;
1602
1603         hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
1604
1605 attach_flow:
1606         list_add(&flow->encap, &e->flows);
1607         *encap_dev = e->out_dev;
1608         if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1609                 attr->encap_id = e->encap_id;
1610
1611         return err;
1612
1613 out_err:
1614         kfree(e);
1615         return err;
1616 }
1617
1618 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
1619                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
1620                                 struct mlx5e_tc_flow *flow)
1621 {
1622         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1623         struct mlx5e_rep_priv *rpriv = priv->ppriv;
1624         struct ip_tunnel_info *info = NULL;
1625         const struct tc_action *a;
1626         LIST_HEAD(actions);
1627         bool encap = false;
1628         int err = 0;
1629
1630         if (tc_no_actions(exts))
1631                 return -EINVAL;
1632
1633         memset(attr, 0, sizeof(*attr));
1634         attr->in_rep = rpriv->rep;
1635
1636         tcf_exts_to_list(exts, &actions);
1637         list_for_each_entry(a, &actions, list) {
1638                 if (is_tcf_gact_shot(a)) {
1639                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
1640                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1641                         continue;
1642                 }
1643
1644                 if (is_tcf_pedit(a)) {
1645                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_FDB,
1646                                                     parse_attr);
1647                         if (err)
1648                                 return err;
1649
1650                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
1651                         continue;
1652                 }
1653
1654                 if (is_tcf_mirred_egress_redirect(a)) {
1655                         int ifindex = tcf_mirred_ifindex(a);
1656                         struct net_device *out_dev, *encap_dev = NULL;
1657                         struct mlx5e_priv *out_priv;
1658
1659                         out_dev = __dev_get_by_index(dev_net(priv->netdev), ifindex);
1660
1661                         if (switchdev_port_same_parent_id(priv->netdev,
1662                                                           out_dev)) {
1663                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1664                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1665                                 out_priv = netdev_priv(out_dev);
1666                                 rpriv = out_priv->ppriv;
1667                                 attr->out_rep = rpriv->rep;
1668                         } else if (encap) {
1669                                 err = mlx5e_attach_encap(priv, info,
1670                                                          out_dev, &encap_dev, flow);
1671                                 if (err && err != -EAGAIN)
1672                                         return err;
1673                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_ENCAP |
1674                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1675                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1676                                 out_priv = netdev_priv(encap_dev);
1677                                 rpriv = out_priv->ppriv;
1678                                 attr->out_rep = rpriv->rep;
1679                                 attr->parse_attr = parse_attr;
1680                         } else {
1681                                 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
1682                                        priv->netdev->name, out_dev->name);
1683                                 return -EINVAL;
1684                         }
1685                         continue;
1686                 }
1687
1688                 if (is_tcf_tunnel_set(a)) {
1689                         info = tcf_tunnel_info(a);
1690                         if (info)
1691                                 encap = true;
1692                         else
1693                                 return -EOPNOTSUPP;
1694                         continue;
1695                 }
1696
1697                 if (is_tcf_vlan(a)) {
1698                         if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
1699                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
1700                         } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
1701                                 if (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q))
1702                                         return -EOPNOTSUPP;
1703
1704                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
1705                                 attr->vlan = tcf_vlan_push_vid(a);
1706                         } else { /* action is TCA_VLAN_ACT_MODIFY */
1707                                 return -EOPNOTSUPP;
1708                         }
1709                         continue;
1710                 }
1711
1712                 if (is_tcf_tunnel_release(a)) {
1713                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
1714                         continue;
1715                 }
1716
1717                 return -EINVAL;
1718         }
1719         return err;
1720 }
1721
1722 int mlx5e_configure_flower(struct mlx5e_priv *priv, __be16 protocol,
1723                            struct tc_cls_flower_offload *f)
1724 {
1725         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1726         struct mlx5e_tc_flow_parse_attr *parse_attr;
1727         struct mlx5e_tc_table *tc = &priv->fs.tc;
1728         struct mlx5e_tc_flow *flow;
1729         int attr_size, err = 0;
1730         u8 flow_flags = 0;
1731
1732         if (esw && esw->mode == SRIOV_OFFLOADS) {
1733                 flow_flags = MLX5E_TC_FLOW_ESWITCH;
1734                 attr_size  = sizeof(struct mlx5_esw_flow_attr);
1735         } else {
1736                 flow_flags = MLX5E_TC_FLOW_NIC;
1737                 attr_size  = sizeof(struct mlx5_nic_flow_attr);
1738         }
1739
1740         flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
1741         parse_attr = mlx5_vzalloc(sizeof(*parse_attr));
1742         if (!parse_attr || !flow) {
1743                 err = -ENOMEM;
1744                 goto err_free;
1745         }
1746
1747         flow->cookie = f->cookie;
1748         flow->flags = flow_flags;
1749
1750         err = parse_cls_flower(priv, flow, &parse_attr->spec, f);
1751         if (err < 0)
1752                 goto err_free;
1753
1754         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
1755                 err = parse_tc_fdb_actions(priv, f->exts, parse_attr, flow);
1756                 if (err < 0)
1757                         goto err_handle_encap_flow;
1758                 flow->rule = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow);
1759         } else {
1760                 err = parse_tc_nic_actions(priv, f->exts, parse_attr, flow);
1761                 if (err < 0)
1762                         goto err_free;
1763                 flow->rule = mlx5e_tc_add_nic_flow(priv, parse_attr, flow);
1764         }
1765
1766         if (IS_ERR(flow->rule)) {
1767                 err = PTR_ERR(flow->rule);
1768                 goto err_free;
1769         }
1770
1771         flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
1772         err = rhashtable_insert_fast(&tc->ht, &flow->node,
1773                                      tc->ht_params);
1774         if (err)
1775                 goto err_del_rule;
1776
1777         if (flow->flags & MLX5E_TC_FLOW_ESWITCH &&
1778             !(flow->esw_attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP))
1779                 kvfree(parse_attr);
1780         return err;
1781
1782 err_del_rule:
1783         mlx5e_tc_del_flow(priv, flow);
1784
1785 err_handle_encap_flow:
1786         if (err == -EAGAIN) {
1787                 err = rhashtable_insert_fast(&tc->ht, &flow->node,
1788                                              tc->ht_params);
1789                 if (err)
1790                         mlx5e_tc_del_flow(priv, flow);
1791                 else
1792                         return 0;
1793         }
1794
1795 err_free:
1796         kvfree(parse_attr);
1797         kfree(flow);
1798         return err;
1799 }
1800
1801 int mlx5e_delete_flower(struct mlx5e_priv *priv,
1802                         struct tc_cls_flower_offload *f)
1803 {
1804         struct mlx5e_tc_flow *flow;
1805         struct mlx5e_tc_table *tc = &priv->fs.tc;
1806
1807         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
1808                                       tc->ht_params);
1809         if (!flow)
1810                 return -EINVAL;
1811
1812         rhashtable_remove_fast(&tc->ht, &flow->node, tc->ht_params);
1813
1814         mlx5e_tc_del_flow(priv, flow);
1815
1816         kfree(flow);
1817
1818         return 0;
1819 }
1820
1821 int mlx5e_stats_flower(struct mlx5e_priv *priv,
1822                        struct tc_cls_flower_offload *f)
1823 {
1824         struct mlx5e_tc_table *tc = &priv->fs.tc;
1825         struct mlx5e_tc_flow *flow;
1826         struct tc_action *a;
1827         struct mlx5_fc *counter;
1828         LIST_HEAD(actions);
1829         u64 bytes;
1830         u64 packets;
1831         u64 lastuse;
1832
1833         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
1834                                       tc->ht_params);
1835         if (!flow)
1836                 return -EINVAL;
1837
1838         if (!(flow->flags & MLX5E_TC_FLOW_OFFLOADED))
1839                 return 0;
1840
1841         counter = mlx5_flow_rule_counter(flow->rule);
1842         if (!counter)
1843                 return 0;
1844
1845         mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
1846
1847         preempt_disable();
1848
1849         tcf_exts_to_list(f->exts, &actions);
1850         list_for_each_entry(a, &actions, list)
1851                 tcf_action_stats_update(a, bytes, packets, lastuse);
1852
1853         preempt_enable();
1854
1855         return 0;
1856 }
1857
1858 static const struct rhashtable_params mlx5e_tc_flow_ht_params = {
1859         .head_offset = offsetof(struct mlx5e_tc_flow, node),
1860         .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
1861         .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
1862         .automatic_shrinking = true,
1863 };
1864
1865 int mlx5e_tc_init(struct mlx5e_priv *priv)
1866 {
1867         struct mlx5e_tc_table *tc = &priv->fs.tc;
1868
1869         tc->ht_params = mlx5e_tc_flow_ht_params;
1870         return rhashtable_init(&tc->ht, &tc->ht_params);
1871 }
1872
1873 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
1874 {
1875         struct mlx5e_tc_flow *flow = ptr;
1876         struct mlx5e_priv *priv = arg;
1877
1878         mlx5e_tc_del_flow(priv, flow);
1879         kfree(flow);
1880 }
1881
1882 void mlx5e_tc_cleanup(struct mlx5e_priv *priv)
1883 {
1884         struct mlx5e_tc_table *tc = &priv->fs.tc;
1885
1886         rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, priv);
1887
1888         if (!IS_ERR_OR_NULL(tc->t)) {
1889                 mlx5_destroy_flow_table(tc->t);
1890                 tc->t = NULL;
1891         }
1892 }