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 #include "fs_core.h"
  55 #include "en/port.h"
  56
  57 struct mlx5_nic_flow_attr {
  58         u32 action;
  59         u32 flow_tag;
  60         u32 mod_hdr_id;
  61         u32 hairpin_tirn;
  62         u8 match_level;
  63         struct mlx5_flow_table  *hairpin_ft;
  64 };
  65
  66 #define MLX5E_TC_FLOW_BASE (MLX5E_TC_LAST_EXPORTED_BIT + 1)
  67
  68 enum {
  69         MLX5E_TC_FLOW_INGRESS   = MLX5E_TC_INGRESS,
  70         MLX5E_TC_FLOW_EGRESS    = MLX5E_TC_EGRESS,
  71         MLX5E_TC_FLOW_ESWITCH   = BIT(MLX5E_TC_FLOW_BASE),
  72         MLX5E_TC_FLOW_NIC       = BIT(MLX5E_TC_FLOW_BASE + 1),
  73         MLX5E_TC_FLOW_OFFLOADED = BIT(MLX5E_TC_FLOW_BASE + 2),
  74         MLX5E_TC_FLOW_HAIRPIN   = BIT(MLX5E_TC_FLOW_BASE + 3),
  75         MLX5E_TC_FLOW_HAIRPIN_RSS = BIT(MLX5E_TC_FLOW_BASE + 4),
  76 };
  77
  78 #define MLX5E_TC_MAX_SPLITS 1
  79
  80 struct mlx5e_tc_flow {
  81         struct rhash_head       node;
  82         struct mlx5e_priv       *priv;
  83         u64                     cookie;
  84         u8                      flags;
  85         struct mlx5_flow_handle *rule[MLX5E_TC_MAX_SPLITS + 1];
  86         struct list_head        encap;   /* flows sharing the same encap ID */
  87         struct list_head        mod_hdr; /* flows sharing the same mod hdr ID */
  88         struct list_head        hairpin; /* flows sharing the same hairpin */
  89         union {
  90                 struct mlx5_esw_flow_attr esw_attr[0];
  91                 struct mlx5_nic_flow_attr nic_attr[0];
  92         };
  93 };
  94
  95 struct mlx5e_tc_flow_parse_attr {
  96         struct ip_tunnel_info tun_info;
  97         struct mlx5_flow_spec spec;
  98         int num_mod_hdr_actions;
  99         void *mod_hdr_actions;
 100         int mirred_ifindex;
 101 };
 102
 103 enum {
 104         MLX5_HEADER_TYPE_VXLAN = 0x0,
 105         MLX5_HEADER_TYPE_NVGRE = 0x1,
 106 };
 107
 108 #define MLX5E_TC_TABLE_NUM_GROUPS 4
 109 #define MLX5E_TC_TABLE_MAX_GROUP_SIZE BIT(16)
 110
 111 struct mlx5e_hairpin {
 112         struct mlx5_hairpin *pair;
 113
 114         struct mlx5_core_dev *func_mdev;
 115         struct mlx5e_priv *func_priv;
 116         u32 tdn;
 117         u32 tirn;
 118
 119         int num_channels;
 120         struct mlx5e_rqt indir_rqt;
 121         u32 indir_tirn[MLX5E_NUM_INDIR_TIRS];
 122         struct mlx5e_ttc_table ttc;
 123 };
 124
 125 struct mlx5e_hairpin_entry {
 126         /* a node of a hash table which keeps all the  hairpin entries */
 127         struct hlist_node hairpin_hlist;
 128
 129         /* flows sharing the same hairpin */
 130         struct list_head flows;
 131
 132         u16 peer_vhca_id;
 133         u8 prio;
 134         struct mlx5e_hairpin *hp;
 135 };
 136
 137 struct mod_hdr_key {
 138         int num_actions;
 139         void *actions;
 140 };
 141
 142 struct mlx5e_mod_hdr_entry {
 143         /* a node of a hash table which keeps all the mod_hdr entries */
 144         struct hlist_node mod_hdr_hlist;
 145
 146         /* flows sharing the same mod_hdr entry */
 147         struct list_head flows;
 148
 149         struct mod_hdr_key key;
 150
 151         u32 mod_hdr_id;
 152 };
 153
 154 #define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)
 155
 156 static inline u32 hash_mod_hdr_info(struct mod_hdr_key *key)
 157 {
 158         return jhash(key->actions,
 159                      key->num_actions * MLX5_MH_ACT_SZ, 0);
 160 }
 161
 162 static inline int cmp_mod_hdr_info(struct mod_hdr_key *a,
 163                                    struct mod_hdr_key *b)
 164 {
 165         if (a->num_actions != b->num_actions)
 166                 return 1;
 167
 168         return memcmp(a->actions, b->actions, a->num_actions * MLX5_MH_ACT_SZ);
 169 }
 170
 171 static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
 172                                 struct mlx5e_tc_flow *flow,
 173                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
 174 {
 175         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 176         int num_actions, actions_size, namespace, err;
 177         struct mlx5e_mod_hdr_entry *mh;
 178         struct mod_hdr_key key;
 179         bool found = false;
 180         u32 hash_key;
 181
 182         num_actions  = parse_attr->num_mod_hdr_actions;
 183         actions_size = MLX5_MH_ACT_SZ * num_actions;
 184
 185         key.actions = parse_attr->mod_hdr_actions;
 186         key.num_actions = num_actions;
 187
 188         hash_key = hash_mod_hdr_info(&key);
 189
 190         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
 191                 namespace = MLX5_FLOW_NAMESPACE_FDB;
 192                 hash_for_each_possible(esw->offloads.mod_hdr_tbl, mh,
 193                                        mod_hdr_hlist, hash_key) {
 194                         if (!cmp_mod_hdr_info(&mh->key, &key)) {
 195                                 found = true;
 196                                 break;
 197                         }
 198                 }
 199         } else {
 200                 namespace = MLX5_FLOW_NAMESPACE_KERNEL;
 201                 hash_for_each_possible(priv->fs.tc.mod_hdr_tbl, mh,
 202                                        mod_hdr_hlist, hash_key) {
 203                         if (!cmp_mod_hdr_info(&mh->key, &key)) {
 204                                 found = true;
 205                                 break;
 206                         }
 207                 }
 208         }
 209
 210         if (found)
 211                 goto attach_flow;
 212
 213         mh = kzalloc(sizeof(*mh) + actions_size, GFP_KERNEL);
 214         if (!mh)
 215                 return -ENOMEM;
 216
 217         mh->key.actions = (void *)mh + sizeof(*mh);
 218         memcpy(mh->key.actions, key.actions, actions_size);
 219         mh->key.num_actions = num_actions;
 220         INIT_LIST_HEAD(&mh->flows);
 221
 222         err = mlx5_modify_header_alloc(priv->mdev, namespace,
 223                                        mh->key.num_actions,
 224                                        mh->key.actions,
 225                                        &mh->mod_hdr_id);
 226         if (err)
 227                 goto out_err;
 228
 229         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
 230                 hash_add(esw->offloads.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
 231         else
 232                 hash_add(priv->fs.tc.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
 233
 234 attach_flow:
 235         list_add(&flow->mod_hdr, &mh->flows);
 236         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
 237                 flow->esw_attr->mod_hdr_id = mh->mod_hdr_id;
 238         else
 239                 flow->nic_attr->mod_hdr_id = mh->mod_hdr_id;
 240
 241         return 0;
 242
 243 out_err:
 244         kfree(mh);
 245         return err;
 246 }
 247
 248 static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
 249                                  struct mlx5e_tc_flow *flow)
 250 {
 251         struct list_head *next = flow->mod_hdr.next;
 252
 253         list_del(&flow->mod_hdr);
 254
 255         if (list_empty(next)) {
 256                 struct mlx5e_mod_hdr_entry *mh;
 257
 258                 mh = list_entry(next, struct mlx5e_mod_hdr_entry, flows);
 259
 260                 mlx5_modify_header_dealloc(priv->mdev, mh->mod_hdr_id);
 261                 hash_del(&mh->mod_hdr_hlist);
 262                 kfree(mh);
 263         }
 264 }
 265
 266 static
 267 struct mlx5_core_dev *mlx5e_hairpin_get_mdev(struct net *net, int ifindex)
 268 {
 269         struct net_device *netdev;
 270         struct mlx5e_priv *priv;
 271
 272         netdev = __dev_get_by_index(net, ifindex);
 273         priv = netdev_priv(netdev);
 274         return priv->mdev;
 275 }
 276
 277 static int mlx5e_hairpin_create_transport(struct mlx5e_hairpin *hp)
 278 {
 279         u32 in[MLX5_ST_SZ_DW(create_tir_in)] = {0};
 280         void *tirc;
 281         int err;
 282
 283         err = mlx5_core_alloc_transport_domain(hp->func_mdev, &hp->tdn);
 284         if (err)
 285                 goto alloc_tdn_err;
 286
 287         tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
 288
 289         MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
 290         MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn[0]);
 291         MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
 292
 293         err = mlx5_core_create_tir(hp->func_mdev, in, MLX5_ST_SZ_BYTES(create_tir_in), &hp->tirn);
 294         if (err)
 295                 goto create_tir_err;
 296
 297         return 0;
 298
 299 create_tir_err:
 300         mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
 301 alloc_tdn_err:
 302         return err;
 303 }
 304
 305 static void mlx5e_hairpin_destroy_transport(struct mlx5e_hairpin *hp)
 306 {
 307         mlx5_core_destroy_tir(hp->func_mdev, hp->tirn);
 308         mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
 309 }
 310
 311 static void mlx5e_hairpin_fill_rqt_rqns(struct mlx5e_hairpin *hp, void *rqtc)
 312 {
 313         u32 indirection_rqt[MLX5E_INDIR_RQT_SIZE], rqn;
 314         struct mlx5e_priv *priv = hp->func_priv;
 315         int i, ix, sz = MLX5E_INDIR_RQT_SIZE;
 316
 317         mlx5e_build_default_indir_rqt(indirection_rqt, sz,
 318                                       hp->num_channels);
 319
 320         for (i = 0; i < sz; i++) {
 321                 ix = i;
 322                 if (priv->channels.params.rss_hfunc == ETH_RSS_HASH_XOR)
 323                         ix = mlx5e_bits_invert(i, ilog2(sz));
 324                 ix = indirection_rqt[ix];
 325                 rqn = hp->pair->rqn[ix];
 326                 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
 327         }
 328 }
 329
 330 static int mlx5e_hairpin_create_indirect_rqt(struct mlx5e_hairpin *hp)
 331 {
 332         int inlen, err, sz = MLX5E_INDIR_RQT_SIZE;
 333         struct mlx5e_priv *priv = hp->func_priv;
 334         struct mlx5_core_dev *mdev = priv->mdev;
 335         void *rqtc;
 336         u32 *in;
 337
 338         inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
 339         in = kvzalloc(inlen, GFP_KERNEL);
 340         if (!in)
 341                 return -ENOMEM;
 342
 343         rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
 344
 345         MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
 346         MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
 347
 348         mlx5e_hairpin_fill_rqt_rqns(hp, rqtc);
 349
 350         err = mlx5_core_create_rqt(mdev, in, inlen, &hp->indir_rqt.rqtn);
 351         if (!err)
 352                 hp->indir_rqt.enabled = true;
 353
 354         kvfree(in);
 355         return err;
 356 }
 357
 358 static int mlx5e_hairpin_create_indirect_tirs(struct mlx5e_hairpin *hp)
 359 {
 360         struct mlx5e_priv *priv = hp->func_priv;
 361         u32 in[MLX5_ST_SZ_DW(create_tir_in)];
 362         int tt, i, err;
 363         void *tirc;
 364
 365         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
 366                 memset(in, 0, MLX5_ST_SZ_BYTES(create_tir_in));
 367                 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
 368
 369                 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
 370                 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
 371                 MLX5_SET(tirc, tirc, indirect_table, hp->indir_rqt.rqtn);
 372                 mlx5e_build_indir_tir_ctx_hash(&priv->channels.params, tt, tirc, false);
 373
 374                 err = mlx5_core_create_tir(hp->func_mdev, in,
 375                                            MLX5_ST_SZ_BYTES(create_tir_in), &hp->indir_tirn[tt]);
 376                 if (err) {
 377                         mlx5_core_warn(hp->func_mdev, "create indirect tirs failed, %d\n", err);
 378                         goto err_destroy_tirs;
 379                 }
 380         }
 381         return 0;
 382
 383 err_destroy_tirs:
 384         for (i = 0; i < tt; i++)
 385                 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[i]);
 386         return err;
 387 }
 388
 389 static void mlx5e_hairpin_destroy_indirect_tirs(struct mlx5e_hairpin *hp)
 390 {
 391         int tt;
 392
 393         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
 394                 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[tt]);
 395 }
 396
 397 static void mlx5e_hairpin_set_ttc_params(struct mlx5e_hairpin *hp,
 398                                          struct ttc_params *ttc_params)
 399 {
 400         struct mlx5_flow_table_attr *ft_attr = &ttc_params->ft_attr;
 401         int tt;
 402
 403         memset(ttc_params, 0, sizeof(*ttc_params));
 404
 405         ttc_params->any_tt_tirn = hp->tirn;
 406
 407         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
 408                 ttc_params->indir_tirn[tt] = hp->indir_tirn[tt];
 409
 410         ft_attr->max_fte = MLX5E_NUM_TT;
 411         ft_attr->level = MLX5E_TC_TTC_FT_LEVEL;
 412         ft_attr->prio = MLX5E_TC_PRIO;
 413 }
 414
 415 static int mlx5e_hairpin_rss_init(struct mlx5e_hairpin *hp)
 416 {
 417         struct mlx5e_priv *priv = hp->func_priv;
 418         struct ttc_params ttc_params;
 419         int err;
 420
 421         err = mlx5e_hairpin_create_indirect_rqt(hp);
 422         if (err)
 423                 return err;
 424
 425         err = mlx5e_hairpin_create_indirect_tirs(hp);
 426         if (err)
 427                 goto err_create_indirect_tirs;
 428
 429         mlx5e_hairpin_set_ttc_params(hp, &ttc_params);
 430         err = mlx5e_create_ttc_table(priv, &ttc_params, &hp->ttc);
 431         if (err)
 432                 goto err_create_ttc_table;
 433
 434         netdev_dbg(priv->netdev, "add hairpin: using %d channels rss ttc table id %x\n",
 435                    hp->num_channels, hp->ttc.ft.t->id);
 436
 437         return 0;
 438
 439 err_create_ttc_table:
 440         mlx5e_hairpin_destroy_indirect_tirs(hp);
 441 err_create_indirect_tirs:
 442         mlx5e_destroy_rqt(priv, &hp->indir_rqt);
 443
 444         return err;
 445 }
 446
 447 static void mlx5e_hairpin_rss_cleanup(struct mlx5e_hairpin *hp)
 448 {
 449         struct mlx5e_priv *priv = hp->func_priv;
 450
 451         mlx5e_destroy_ttc_table(priv, &hp->ttc);
 452         mlx5e_hairpin_destroy_indirect_tirs(hp);
 453         mlx5e_destroy_rqt(priv, &hp->indir_rqt);
 454 }
 455
 456 static struct mlx5e_hairpin *
 457 mlx5e_hairpin_create(struct mlx5e_priv *priv, struct mlx5_hairpin_params *params,
 458                      int peer_ifindex)
 459 {
 460         struct mlx5_core_dev *func_mdev, *peer_mdev;
 461         struct mlx5e_hairpin *hp;
 462         struct mlx5_hairpin *pair;
 463         int err;
 464
 465         hp = kzalloc(sizeof(*hp), GFP_KERNEL);
 466         if (!hp)
 467                 return ERR_PTR(-ENOMEM);
 468
 469         func_mdev = priv->mdev;
 470         peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
 471
 472         pair = mlx5_core_hairpin_create(func_mdev, peer_mdev, params);
 473         if (IS_ERR(pair)) {
 474                 err = PTR_ERR(pair);
 475                 goto create_pair_err;
 476         }
 477         hp->pair = pair;
 478         hp->func_mdev = func_mdev;
 479         hp->func_priv = priv;
 480         hp->num_channels = params->num_channels;
 481
 482         err = mlx5e_hairpin_create_transport(hp);
 483         if (err)
 484                 goto create_transport_err;
 485
 486         if (hp->num_channels > 1) {
 487                 err = mlx5e_hairpin_rss_init(hp);
 488                 if (err)
 489                         goto rss_init_err;
 490         }
 491
 492         return hp;
 493
 494 rss_init_err:
 495         mlx5e_hairpin_destroy_transport(hp);
 496 create_transport_err:
 497         mlx5_core_hairpin_destroy(hp->pair);
 498 create_pair_err:
 499         kfree(hp);
 500         return ERR_PTR(err);
 501 }
 502
 503 static void mlx5e_hairpin_destroy(struct mlx5e_hairpin *hp)
 504 {
 505         if (hp->num_channels > 1)
 506                 mlx5e_hairpin_rss_cleanup(hp);
 507         mlx5e_hairpin_destroy_transport(hp);
 508         mlx5_core_hairpin_destroy(hp->pair);
 509         kvfree(hp);
 510 }
 511
 512 static inline u32 hash_hairpin_info(u16 peer_vhca_id, u8 prio)
 513 {
 514         return (peer_vhca_id << 16 | prio);
 515 }
 516
 517 static struct mlx5e_hairpin_entry *mlx5e_hairpin_get(struct mlx5e_priv *priv,
 518                                                      u16 peer_vhca_id, u8 prio)
 519 {
 520         struct mlx5e_hairpin_entry *hpe;
 521         u32 hash_key = hash_hairpin_info(peer_vhca_id, prio);
 522
 523         hash_for_each_possible(priv->fs.tc.hairpin_tbl, hpe,
 524                                hairpin_hlist, hash_key) {
 525                 if (hpe->peer_vhca_id == peer_vhca_id && hpe->prio == prio)
 526                         return hpe;
 527         }
 528
 529         return NULL;
 530 }
 531
 532 #define UNKNOWN_MATCH_PRIO 8
 533
 534 static int mlx5e_hairpin_get_prio(struct mlx5e_priv *priv,
 535                                   struct mlx5_flow_spec *spec, u8 *match_prio)
 536 {
 537         void *headers_c, *headers_v;
 538         u8 prio_val, prio_mask = 0;
 539         bool vlan_present;
 540
 541 #ifdef CONFIG_MLX5_CORE_EN_DCB
 542         if (priv->dcbx_dp.trust_state != MLX5_QPTS_TRUST_PCP) {
 543                 netdev_warn(priv->netdev,
 544                             "only PCP trust state supported for hairpin\n");
 545                 return -EOPNOTSUPP;
 546         }
 547 #endif
 548         headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, outer_headers);
 549         headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
 550
 551         vlan_present = MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag);
 552         if (vlan_present) {
 553                 prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
 554                 prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
 555         }
 556
 557         if (!vlan_present || !prio_mask) {
 558                 prio_val = UNKNOWN_MATCH_PRIO;
 559         } else if (prio_mask != 0x7) {
 560                 netdev_warn(priv->netdev,
 561                             "masked priority match not supported for hairpin\n");
 562                 return -EOPNOTSUPP;
 563         }
 564
 565         *match_prio = prio_val;
 566         return 0;
 567 }
 568
 569 static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
 570                                   struct mlx5e_tc_flow *flow,
 571                                   struct mlx5e_tc_flow_parse_attr *parse_attr)
 572 {
 573         int peer_ifindex = parse_attr->mirred_ifindex;
 574         struct mlx5_hairpin_params params;
 575         struct mlx5_core_dev *peer_mdev;
 576         struct mlx5e_hairpin_entry *hpe;
 577         struct mlx5e_hairpin *hp;
 578         u64 link_speed64;
 579         u32 link_speed;
 580         u8 match_prio;
 581         u16 peer_id;
 582         int err;
 583
 584         peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
 585         if (!MLX5_CAP_GEN(priv->mdev, hairpin) || !MLX5_CAP_GEN(peer_mdev, hairpin)) {
 586                 netdev_warn(priv->netdev, "hairpin is not supported\n");
 587                 return -EOPNOTSUPP;
 588         }
 589
 590         peer_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
 591         err = mlx5e_hairpin_get_prio(priv, &parse_attr->spec, &match_prio);
 592         if (err)
 593                 return err;
 594         hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
 595         if (hpe)
 596                 goto attach_flow;
 597
 598         hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
 599         if (!hpe)
 600                 return -ENOMEM;
 601
 602         INIT_LIST_HEAD(&hpe->flows);
 603         hpe->peer_vhca_id = peer_id;
 604         hpe->prio = match_prio;
 605
 606         params.log_data_size = 15;
 607         params.log_data_size = min_t(u8, params.log_data_size,
 608                                      MLX5_CAP_GEN(priv->mdev, log_max_hairpin_wq_data_sz));
 609         params.log_data_size = max_t(u8, params.log_data_size,
 610                                      MLX5_CAP_GEN(priv->mdev, log_min_hairpin_wq_data_sz));
 611
 612         params.log_num_packets = params.log_data_size -
 613                                  MLX5_MPWRQ_MIN_LOG_STRIDE_SZ(priv->mdev);
 614         params.log_num_packets = min_t(u8, params.log_num_packets,
 615                                        MLX5_CAP_GEN(priv->mdev, log_max_hairpin_num_packets));
 616
 617         params.q_counter = priv->q_counter;
 618         /* set hairpin pair per each 50Gbs share of the link */
 619         mlx5e_port_max_linkspeed(priv->mdev, &link_speed);
 620         link_speed = max_t(u32, link_speed, 50000);
 621         link_speed64 = link_speed;
 622         do_div(link_speed64, 50000);
 623         params.num_channels = link_speed64;
 624
 625         hp = mlx5e_hairpin_create(priv, &params, peer_ifindex);
 626         if (IS_ERR(hp)) {
 627                 err = PTR_ERR(hp);
 628                 goto create_hairpin_err;
 629         }
 630
 631         netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
 632                    hp->tirn, hp->pair->rqn[0], hp->pair->peer_mdev->priv.name,
 633                    hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
 634
 635         hpe->hp = hp;
 636         hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
 637                  hash_hairpin_info(peer_id, match_prio));
 638
 639 attach_flow:
 640         if (hpe->hp->num_channels > 1) {
 641                 flow->flags |= MLX5E_TC_FLOW_HAIRPIN_RSS;
 642                 flow->nic_attr->hairpin_ft = hpe->hp->ttc.ft.t;
 643         } else {
 644                 flow->nic_attr->hairpin_tirn = hpe->hp->tirn;
 645         }
 646         list_add(&flow->hairpin, &hpe->flows);
 647
 648         return 0;
 649
 650 create_hairpin_err:
 651         kfree(hpe);
 652         return err;
 653 }
 654
 655 static void mlx5e_hairpin_flow_del(struct mlx5e_priv *priv,
 656                                    struct mlx5e_tc_flow *flow)
 657 {
 658         struct list_head *next = flow->hairpin.next;
 659
 660         list_del(&flow->hairpin);
 661
 662         /* no more hairpin flows for us, release the hairpin pair */
 663         if (list_empty(next)) {
 664                 struct mlx5e_hairpin_entry *hpe;
 665
 666                 hpe = list_entry(next, struct mlx5e_hairpin_entry, flows);
 667
 668                 netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
 669                            hpe->hp->pair->peer_mdev->priv.name);
 670
 671                 mlx5e_hairpin_destroy(hpe->hp);
 672                 hash_del(&hpe->hairpin_hlist);
 673                 kfree(hpe);
 674         }
 675 }
 676
 677 static struct mlx5_flow_handle *
 678 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
 679                       struct mlx5e_tc_flow_parse_attr *parse_attr,
 680                       struct mlx5e_tc_flow *flow)
 681 {
 682         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
 683         struct mlx5_core_dev *dev = priv->mdev;
 684         struct mlx5_flow_destination dest[2] = {};
 685         struct mlx5_flow_act flow_act = {
 686                 .action = attr->action,
 687                 .has_flow_tag = true,
 688                 .flow_tag = attr->flow_tag,
 689                 .encap_id = 0,
 690         };
 691         struct mlx5_fc *counter = NULL;
 692         struct mlx5_flow_handle *rule;
 693         bool table_created = false;
 694         int err, dest_ix = 0;
 695
 696         if (flow->flags & MLX5E_TC_FLOW_HAIRPIN) {
 697                 err = mlx5e_hairpin_flow_add(priv, flow, parse_attr);
 698                 if (err) {
 699                         rule = ERR_PTR(err);
 700                         goto err_add_hairpin_flow;
 701                 }
 702                 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN_RSS) {
 703                         dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 704                         dest[dest_ix].ft = attr->hairpin_ft;
 705                 } else {
 706                         dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_TIR;
 707                         dest[dest_ix].tir_num = attr->hairpin_tirn;
 708                 }
 709                 dest_ix++;
 710         } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
 711                 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 712                 dest[dest_ix].ft = priv->fs.vlan.ft.t;
 713                 dest_ix++;
 714         }
 715
 716         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
 717                 counter = mlx5_fc_create(dev, true);
 718                 if (IS_ERR(counter)) {
 719                         rule = ERR_CAST(counter);
 720                         goto err_fc_create;
 721                 }
 722                 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
 723                 dest[dest_ix].counter = counter;
 724                 dest_ix++;
 725         }
 726
 727         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
 728                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
 729                 flow_act.modify_id = attr->mod_hdr_id;
 730                 kfree(parse_attr->mod_hdr_actions);
 731                 if (err) {
 732                         rule = ERR_PTR(err);
 733                         goto err_create_mod_hdr_id;
 734                 }
 735         }
 736
 737         if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
 738                 int tc_grp_size, tc_tbl_size;
 739                 u32 max_flow_counter;
 740
 741                 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
 742                                     MLX5_CAP_GEN(dev, max_flow_counter_15_0);
 743
 744                 tc_grp_size = min_t(int, max_flow_counter, MLX5E_TC_TABLE_MAX_GROUP_SIZE);
 745
 746                 tc_tbl_size = min_t(int, tc_grp_size * MLX5E_TC_TABLE_NUM_GROUPS,
 747                                     BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev, log_max_ft_size)));
 748
 749                 priv->fs.tc.t =
 750                         mlx5_create_auto_grouped_flow_table(priv->fs.ns,
 751                                                             MLX5E_TC_PRIO,
 752                                                             tc_tbl_size,
 753                                                             MLX5E_TC_TABLE_NUM_GROUPS,
 754                                                             MLX5E_TC_FT_LEVEL, 0);
 755                 if (IS_ERR(priv->fs.tc.t)) {
 756                         netdev_err(priv->netdev,
 757                                    "Failed to create tc offload table\n");
 758                         rule = ERR_CAST(priv->fs.tc.t);
 759                         goto err_create_ft;
 760                 }
 761
 762                 table_created = true;
 763         }
 764
 765         if (attr->match_level != MLX5_MATCH_NONE)
 766                 parse_attr->spec.match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
 767
 768         rule = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
 769                                    &flow_act, dest, dest_ix);
 770
 771         if (IS_ERR(rule))
 772                 goto err_add_rule;
 773
 774         return rule;
 775
 776 err_add_rule:
 777         if (table_created) {
 778                 mlx5_destroy_flow_table(priv->fs.tc.t);
 779                 priv->fs.tc.t = NULL;
 780         }
 781 err_create_ft:
 782         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 783                 mlx5e_detach_mod_hdr(priv, flow);
 784 err_create_mod_hdr_id:
 785         mlx5_fc_destroy(dev, counter);
 786 err_fc_create:
 787         if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
 788                 mlx5e_hairpin_flow_del(priv, flow);
 789 err_add_hairpin_flow:
 790         return rule;
 791 }
 792
 793 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
 794                                   struct mlx5e_tc_flow *flow)
 795 {
 796         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
 797         struct mlx5_fc *counter = NULL;
 798
 799         counter = mlx5_flow_rule_counter(flow->rule[0]);
 800         mlx5_del_flow_rules(flow->rule[0]);
 801         mlx5_fc_destroy(priv->mdev, counter);
 802
 803         if (!mlx5e_tc_num_filters(priv) && priv->fs.tc.t) {
 804                 mlx5_destroy_flow_table(priv->fs.tc.t);
 805                 priv->fs.tc.t = NULL;
 806         }
 807
 808         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 809                 mlx5e_detach_mod_hdr(priv, flow);
 810
 811         if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
 812                 mlx5e_hairpin_flow_del(priv, flow);
 813 }
 814
 815 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
 816                                struct mlx5e_tc_flow *flow);
 817
 818 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
 819                               struct ip_tunnel_info *tun_info,
 820                               struct net_device *mirred_dev,
 821                               struct net_device **encap_dev,
 822                               struct mlx5e_tc_flow *flow);
 823
 824 static struct mlx5_flow_handle *
 825 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
 826                       struct mlx5e_tc_flow_parse_attr *parse_attr,
 827                       struct mlx5e_tc_flow *flow)
 828 {
 829         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 830         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
 831         struct net_device *out_dev, *encap_dev = NULL;
 832         struct mlx5_flow_handle *rule = NULL;
 833         struct mlx5e_rep_priv *rpriv;
 834         struct mlx5e_priv *out_priv;
 835         int err;
 836
 837         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP) {
 838                 out_dev = __dev_get_by_index(dev_net(priv->netdev),
 839                                              attr->parse_attr->mirred_ifindex);
 840                 err = mlx5e_attach_encap(priv, &parse_attr->tun_info,
 841                                          out_dev, &encap_dev, flow);
 842                 if (err) {
 843                         rule = ERR_PTR(err);
 844                         if (err != -EAGAIN)
 845                                 goto err_attach_encap;
 846                 }
 847                 out_priv = netdev_priv(encap_dev);
 848                 rpriv = out_priv->ppriv;
 849                 attr->out_rep[attr->out_count] = rpriv->rep;
 850                 attr->out_mdev[attr->out_count++] = out_priv->mdev;
 851         }
 852
 853         err = mlx5_eswitch_add_vlan_action(esw, attr);
 854         if (err) {
 855                 rule = ERR_PTR(err);
 856                 goto err_add_vlan;
 857         }
 858
 859         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
 860                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
 861                 kfree(parse_attr->mod_hdr_actions);
 862                 if (err) {
 863                         rule = ERR_PTR(err);
 864                         goto err_mod_hdr;
 865                 }
 866         }
 867
 868         /* we get here if (1) there's no error (rule being null) or when
 869          * (2) there's an encap action and we're on -EAGAIN (no valid neigh)
 870          */
 871         if (rule != ERR_PTR(-EAGAIN)) {
 872                 rule = mlx5_eswitch_add_offloaded_rule(esw, &parse_attr->spec, attr);
 873                 if (IS_ERR(rule))
 874                         goto err_add_rule;
 875
 876                 if (attr->mirror_count) {
 877                         flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, &parse_attr->spec, attr);
 878                         if (IS_ERR(flow->rule[1]))
 879                                 goto err_fwd_rule;
 880                 }
 881         }
 882         return rule;
 883
 884 err_fwd_rule:
 885         mlx5_eswitch_del_offloaded_rule(esw, rule, attr);
 886         rule = flow->rule[1];
 887 err_add_rule:
 888         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 889                 mlx5e_detach_mod_hdr(priv, flow);
 890 err_mod_hdr:
 891         mlx5_eswitch_del_vlan_action(esw, attr);
 892 err_add_vlan:
 893         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP)
 894                 mlx5e_detach_encap(priv, flow);
 895 err_attach_encap:
 896         return rule;
 897 }
 898
 899 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
 900                                   struct mlx5e_tc_flow *flow)
 901 {
 902         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 903         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
 904
 905         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
 906                 flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
 907                 if (attr->mirror_count)
 908                         mlx5_eswitch_del_offloaded_rule(esw, flow->rule[1], attr);
 909                 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
 910         }
 911
 912         mlx5_eswitch_del_vlan_action(esw, attr);
 913
 914         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP) {
 915                 mlx5e_detach_encap(priv, flow);
 916                 kvfree(attr->parse_attr);
 917         }
 918
 919         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 920                 mlx5e_detach_mod_hdr(priv, flow);
 921 }
 922
 923 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
 924                               struct mlx5e_encap_entry *e)
 925 {
 926         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 927         struct mlx5_esw_flow_attr *esw_attr;
 928         struct mlx5e_tc_flow *flow;
 929         int err;
 930
 931         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
 932                                e->encap_size, e->encap_header,
 933                                &e->encap_id);
 934         if (err) {
 935                 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %d\n",
 936                                err);
 937                 return;
 938         }
 939         e->flags |= MLX5_ENCAP_ENTRY_VALID;
 940         mlx5e_rep_queue_neigh_stats_work(priv);
 941
 942         list_for_each_entry(flow, &e->flows, encap) {
 943                 esw_attr = flow->esw_attr;
 944                 esw_attr->encap_id = e->encap_id;
 945                 flow->rule[0] = mlx5_eswitch_add_offloaded_rule(esw, &esw_attr->parse_attr->spec, esw_attr);
 946                 if (IS_ERR(flow->rule[0])) {
 947                         err = PTR_ERR(flow->rule[0]);
 948                         mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
 949                                        err);
 950                         continue;
 951                 }
 952
 953                 if (esw_attr->mirror_count) {
 954                         flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, &esw_attr->parse_attr->spec, esw_attr);
 955                         if (IS_ERR(flow->rule[1])) {
 956                                 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], esw_attr);
 957                                 err = PTR_ERR(flow->rule[1]);
 958                                 mlx5_core_warn(priv->mdev, "Failed to update cached mirror flow, %d\n",
 959                                                err);
 960                                 continue;
 961                         }
 962                 }
 963
 964                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
 965         }
 966 }
 967
 968 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
 969                               struct mlx5e_encap_entry *e)
 970 {
 971         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 972         struct mlx5e_tc_flow *flow;
 973
 974         list_for_each_entry(flow, &e->flows, encap) {
 975                 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
 976                         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
 977
 978                         flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
 979                         if (attr->mirror_count)
 980                                 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[1], attr);
 981                         mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
 982                 }
 983         }
 984
 985         if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
 986                 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
 987                 mlx5_encap_dealloc(priv->mdev, e->encap_id);
 988         }
 989 }
 990
 991 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
 992 {
 993         struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
 994         u64 bytes, packets, lastuse = 0;
 995         struct mlx5e_tc_flow *flow;
 996         struct mlx5e_encap_entry *e;
 997         struct mlx5_fc *counter;
 998         struct neigh_table *tbl;
 999         bool neigh_used = false;
1000         struct neighbour *n;
1001
1002         if (m_neigh->family == AF_INET)
1003                 tbl = &arp_tbl;
1004 #if IS_ENABLED(CONFIG_IPV6)
1005         else if (m_neigh->family == AF_INET6)
1006                 tbl = &nd_tbl;
1007 #endif
1008         else
1009                 return;
1010
1011         list_for_each_entry(e, &nhe->encap_list, encap_list) {
1012                 if (!(e->flags & MLX5_ENCAP_ENTRY_VALID))
1013                         continue;
1014                 list_for_each_entry(flow, &e->flows, encap) {
1015                         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
1016                                 counter = mlx5_flow_rule_counter(flow->rule[0]);
1017                                 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
1018                                 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
1019                                         neigh_used = true;
1020                                         break;
1021                                 }
1022                         }
1023                 }
1024                 if (neigh_used)
1025                         break;
1026         }
1027
1028         if (neigh_used) {
1029                 nhe->reported_lastuse = jiffies;
1030
1031                 /* find the relevant neigh according to the cached device and
1032                  * dst ip pair
1033                  */
1034                 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
1035                 if (!n) {
1036                         WARN(1, "The neighbour already freed\n");
1037                         return;
1038                 }
1039
1040                 neigh_event_send(n, NULL);
1041                 neigh_release(n);
1042         }
1043 }
1044
1045 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1046                                struct mlx5e_tc_flow *flow)
1047 {
1048         struct list_head *next = flow->encap.next;
1049
1050         list_del(&flow->encap);
1051         if (list_empty(next)) {
1052                 struct mlx5e_encap_entry *e;
1053
1054                 e = list_entry(next, struct mlx5e_encap_entry, flows);
1055                 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1056
1057                 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1058                         mlx5_encap_dealloc(priv->mdev, e->encap_id);
1059
1060                 hash_del_rcu(&e->encap_hlist);
1061                 kfree(e->encap_header);
1062                 kfree(e);
1063         }
1064 }
1065
1066 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1067                               struct mlx5e_tc_flow *flow)
1068 {
1069         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1070                 mlx5e_tc_del_fdb_flow(priv, flow);
1071         else
1072                 mlx5e_tc_del_nic_flow(priv, flow);
1073 }
1074
1075 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
1076                              struct tc_cls_flower_offload *f)
1077 {
1078         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1079                                        outer_headers);
1080         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1081                                        outer_headers);
1082         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1083                                     misc_parameters);
1084         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1085                                     misc_parameters);
1086
1087         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
1088         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
1089
1090         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
1091                 struct flow_dissector_key_keyid *key =
1092                         skb_flow_dissector_target(f->dissector,
1093                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
1094                                                   f->key);
1095                 struct flow_dissector_key_keyid *mask =
1096                         skb_flow_dissector_target(f->dissector,
1097                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
1098                                                   f->mask);
1099                 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
1100                          be32_to_cpu(mask->keyid));
1101                 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
1102                          be32_to_cpu(key->keyid));
1103         }
1104 }
1105
1106 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1107                              struct mlx5_flow_spec *spec,
1108                              struct tc_cls_flower_offload *f)
1109 {
1110         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1111                                        outer_headers);
1112         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1113                                        outer_headers);
1114
1115         struct flow_dissector_key_control *enc_control =
1116                 skb_flow_dissector_target(f->dissector,
1117                                           FLOW_DISSECTOR_KEY_ENC_CONTROL,
1118                                           f->key);
1119
1120         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
1121                 struct flow_dissector_key_ports *key =
1122                         skb_flow_dissector_target(f->dissector,
1123                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
1124                                                   f->key);
1125                 struct flow_dissector_key_ports *mask =
1126                         skb_flow_dissector_target(f->dissector,
1127                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
1128                                                   f->mask);
1129                 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1130                 struct mlx5e_rep_priv *uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1131                 struct net_device *up_dev = uplink_rpriv->netdev;
1132                 struct mlx5e_priv *up_priv = netdev_priv(up_dev);
1133
1134                 /* Full udp dst port must be given */
1135                 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
1136                         goto vxlan_match_offload_err;
1137
1138                 if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->dst)) &&
1139                     MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
1140                         parse_vxlan_attr(spec, f);
1141                 else {
1142                         netdev_warn(priv->netdev,
1143                                     "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
1144                         return -EOPNOTSUPP;
1145                 }
1146
1147                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1148                          udp_dport, ntohs(mask->dst));
1149                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1150                          udp_dport, ntohs(key->dst));
1151
1152                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1153                          udp_sport, ntohs(mask->src));
1154                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1155                          udp_sport, ntohs(key->src));
1156         } else { /* udp dst port must be given */
1157 vxlan_match_offload_err:
1158                 netdev_warn(priv->netdev,
1159                             "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
1160                 return -EOPNOTSUPP;
1161         }
1162
1163         if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1164                 struct flow_dissector_key_ipv4_addrs *key =
1165                         skb_flow_dissector_target(f->dissector,
1166                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1167                                                   f->key);
1168                 struct flow_dissector_key_ipv4_addrs *mask =
1169                         skb_flow_dissector_target(f->dissector,
1170                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1171                                                   f->mask);
1172                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1173                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
1174                          ntohl(mask->src));
1175                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1176                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
1177                          ntohl(key->src));
1178
1179                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1180                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1181                          ntohl(mask->dst));
1182                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1183                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1184                          ntohl(key->dst));
1185
1186                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1187                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
1188         } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1189                 struct flow_dissector_key_ipv6_addrs *key =
1190                         skb_flow_dissector_target(f->dissector,
1191                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1192                                                   f->key);
1193                 struct flow_dissector_key_ipv6_addrs *mask =
1194                         skb_flow_dissector_target(f->dissector,
1195                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1196                                                   f->mask);
1197
1198                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1199                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1200                        &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1201                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1202                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1203                        &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1204
1205                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1206                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1207                        &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1208                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1209                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1210                        &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1211
1212                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1213                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
1214         }
1215
1216         /* Enforce DMAC when offloading incoming tunneled flows.
1217          * Flow counters require a match on the DMAC.
1218          */
1219         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
1220         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
1221         ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1222                                      dmac_47_16), priv->netdev->dev_addr);
1223
1224         /* let software handle IP fragments */
1225         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1226         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
1227
1228         return 0;
1229 }
1230
1231 static int __parse_cls_flower(struct mlx5e_priv *priv,
1232                               struct mlx5_flow_spec *spec,
1233                               struct tc_cls_flower_offload *f,
1234                               u8 *match_level)
1235 {
1236         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1237                                        outer_headers);
1238         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1239                                        outer_headers);
1240         u16 addr_type = 0;
1241         u8 ip_proto = 0;
1242
1243         *match_level = MLX5_MATCH_NONE;
1244
1245         if (f->dissector->used_keys &
1246             ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
1247               BIT(FLOW_DISSECTOR_KEY_BASIC) |
1248               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
1249               BIT(FLOW_DISSECTOR_KEY_VLAN) |
1250               BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
1251               BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
1252               BIT(FLOW_DISSECTOR_KEY_PORTS) |
1253               BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
1254               BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
1255               BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
1256               BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
1257               BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
1258               BIT(FLOW_DISSECTOR_KEY_TCP) |
1259               BIT(FLOW_DISSECTOR_KEY_IP))) {
1260                 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
1261                             f->dissector->used_keys);
1262                 return -EOPNOTSUPP;
1263         }
1264
1265         if ((dissector_uses_key(f->dissector,
1266                                 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
1267              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
1268              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
1269             dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
1270                 struct flow_dissector_key_control *key =
1271                         skb_flow_dissector_target(f->dissector,
1272                                                   FLOW_DISSECTOR_KEY_ENC_CONTROL,
1273                                                   f->key);
1274                 switch (key->addr_type) {
1275                 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
1276                 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
1277                         if (parse_tunnel_attr(priv, spec, f))
1278                                 return -EOPNOTSUPP;
1279                         break;
1280                 default:
1281                         return -EOPNOTSUPP;
1282                 }
1283
1284                 /* In decap flow, header pointers should point to the inner
1285                  * headers, outer header were already set by parse_tunnel_attr
1286                  */
1287                 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1288                                          inner_headers);
1289                 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1290                                          inner_headers);
1291         }
1292
1293         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
1294                 struct flow_dissector_key_eth_addrs *key =
1295                         skb_flow_dissector_target(f->dissector,
1296                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
1297                                                   f->key);
1298                 struct flow_dissector_key_eth_addrs *mask =
1299                         skb_flow_dissector_target(f->dissector,
1300                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
1301                                                   f->mask);
1302
1303                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1304                                              dmac_47_16),
1305                                 mask->dst);
1306                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1307                                              dmac_47_16),
1308                                 key->dst);
1309
1310                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1311                                              smac_47_16),
1312                                 mask->src);
1313                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1314                                              smac_47_16),
1315                                 key->src);
1316
1317                 if (!is_zero_ether_addr(mask->src) || !is_zero_ether_addr(mask->dst))
1318                         *match_level = MLX5_MATCH_L2;
1319         }
1320
1321         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
1322                 struct flow_dissector_key_vlan *key =
1323                         skb_flow_dissector_target(f->dissector,
1324                                                   FLOW_DISSECTOR_KEY_VLAN,
1325                                                   f->key);
1326                 struct flow_dissector_key_vlan *mask =
1327                         skb_flow_dissector_target(f->dissector,
1328                                                   FLOW_DISSECTOR_KEY_VLAN,
1329                                                   f->mask);
1330                 if (mask->vlan_id || mask->vlan_priority) {
1331                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
1332                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
1333
1334                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
1335                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
1336
1337                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
1338                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
1339
1340                         *match_level = MLX5_MATCH_L2;
1341                 }
1342         }
1343
1344         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1345                 struct flow_dissector_key_basic *key =
1346                         skb_flow_dissector_target(f->dissector,
1347                                                   FLOW_DISSECTOR_KEY_BASIC,
1348                                                   f->key);
1349                 struct flow_dissector_key_basic *mask =
1350                         skb_flow_dissector_target(f->dissector,
1351                                                   FLOW_DISSECTOR_KEY_BASIC,
1352                                                   f->mask);
1353                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1354                          ntohs(mask->n_proto));
1355                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1356                          ntohs(key->n_proto));
1357
1358                 if (mask->n_proto)
1359                         *match_level = MLX5_MATCH_L2;
1360         }
1361
1362         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
1363                 struct flow_dissector_key_control *key =
1364                         skb_flow_dissector_target(f->dissector,
1365                                                   FLOW_DISSECTOR_KEY_CONTROL,
1366                                                   f->key);
1367
1368                 struct flow_dissector_key_control *mask =
1369                         skb_flow_dissector_target(f->dissector,
1370                                                   FLOW_DISSECTOR_KEY_CONTROL,
1371                                                   f->mask);
1372                 addr_type = key->addr_type;
1373
1374                 /* the HW doesn't support frag first/later */
1375                 if (mask->flags & FLOW_DIS_FIRST_FRAG)
1376                         return -EOPNOTSUPP;
1377
1378                 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
1379                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1380                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
1381                                  key->flags & FLOW_DIS_IS_FRAGMENT);
1382
1383                         /* the HW doesn't need L3 inline to match on frag=no */
1384                         if (!(key->flags & FLOW_DIS_IS_FRAGMENT))
1385                                 *match_level = MLX5_INLINE_MODE_L2;
1386         /* ***  L2 attributes parsing up to here *** */
1387                         else
1388                                 *match_level = MLX5_INLINE_MODE_IP;
1389                 }
1390         }
1391
1392         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1393                 struct flow_dissector_key_basic *key =
1394                         skb_flow_dissector_target(f->dissector,
1395                                                   FLOW_DISSECTOR_KEY_BASIC,
1396                                                   f->key);
1397                 struct flow_dissector_key_basic *mask =
1398                         skb_flow_dissector_target(f->dissector,
1399                                                   FLOW_DISSECTOR_KEY_BASIC,
1400                                                   f->mask);
1401                 ip_proto = key->ip_proto;
1402
1403                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
1404                          mask->ip_proto);
1405                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
1406                          key->ip_proto);
1407
1408                 if (mask->ip_proto)
1409                         *match_level = MLX5_MATCH_L3;
1410         }
1411
1412         if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1413                 struct flow_dissector_key_ipv4_addrs *key =
1414                         skb_flow_dissector_target(f->dissector,
1415                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1416                                                   f->key);
1417                 struct flow_dissector_key_ipv4_addrs *mask =
1418                         skb_flow_dissector_target(f->dissector,
1419                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1420                                                   f->mask);
1421
1422                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1423                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
1424                        &mask->src, sizeof(mask->src));
1425                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1426                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
1427                        &key->src, sizeof(key->src));
1428                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1429                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1430                        &mask->dst, sizeof(mask->dst));
1431                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1432                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1433                        &key->dst, sizeof(key->dst));
1434
1435                 if (mask->src || mask->dst)
1436                         *match_level = MLX5_MATCH_L3;
1437         }
1438
1439         if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1440                 struct flow_dissector_key_ipv6_addrs *key =
1441                         skb_flow_dissector_target(f->dissector,
1442                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1443                                                   f->key);
1444                 struct flow_dissector_key_ipv6_addrs *mask =
1445                         skb_flow_dissector_target(f->dissector,
1446                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1447                                                   f->mask);
1448
1449                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1450                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1451                        &mask->src, sizeof(mask->src));
1452                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1453                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1454                        &key->src, sizeof(key->src));
1455
1456                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1457                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1458                        &mask->dst, sizeof(mask->dst));
1459                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1460                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1461                        &key->dst, sizeof(key->dst));
1462
1463                 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
1464                     ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
1465                         *match_level = MLX5_MATCH_L3;
1466         }
1467
1468         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_IP)) {
1469                 struct flow_dissector_key_ip *key =
1470                         skb_flow_dissector_target(f->dissector,
1471                                                   FLOW_DISSECTOR_KEY_IP,
1472                                                   f->key);
1473                 struct flow_dissector_key_ip *mask =
1474                         skb_flow_dissector_target(f->dissector,
1475                                                   FLOW_DISSECTOR_KEY_IP,
1476                                                   f->mask);
1477
1478                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn, mask->tos & 0x3);
1479                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, key->tos & 0x3);
1480
1481                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp, mask->tos >> 2);
1482                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, key->tos  >> 2);
1483
1484                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit, mask->ttl);
1485                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit, key->ttl);
1486
1487                 if (mask->ttl &&
1488                     !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
1489                                                 ft_field_support.outer_ipv4_ttl))
1490                         return -EOPNOTSUPP;
1491
1492                 if (mask->tos || mask->ttl)
1493                         *match_level = MLX5_MATCH_L3;
1494         }
1495
1496         /* ***  L3 attributes parsing up to here *** */
1497
1498         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
1499                 struct flow_dissector_key_ports *key =
1500                         skb_flow_dissector_target(f->dissector,
1501                                                   FLOW_DISSECTOR_KEY_PORTS,
1502                                                   f->key);
1503                 struct flow_dissector_key_ports *mask =
1504                         skb_flow_dissector_target(f->dissector,
1505                                                   FLOW_DISSECTOR_KEY_PORTS,
1506                                                   f->mask);
1507                 switch (ip_proto) {
1508                 case IPPROTO_TCP:
1509                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1510                                  tcp_sport, ntohs(mask->src));
1511                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1512                                  tcp_sport, ntohs(key->src));
1513
1514                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1515                                  tcp_dport, ntohs(mask->dst));
1516                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1517                                  tcp_dport, ntohs(key->dst));
1518                         break;
1519
1520                 case IPPROTO_UDP:
1521                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1522                                  udp_sport, ntohs(mask->src));
1523                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1524                                  udp_sport, ntohs(key->src));
1525
1526                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1527                                  udp_dport, ntohs(mask->dst));
1528                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1529                                  udp_dport, ntohs(key->dst));
1530                         break;
1531                 default:
1532                         netdev_err(priv->netdev,
1533                                    "Only UDP and TCP transport are supported\n");
1534                         return -EINVAL;
1535                 }
1536
1537                 if (mask->src || mask->dst)
1538                         *match_level = MLX5_MATCH_L4;
1539         }
1540
1541         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_TCP)) {
1542                 struct flow_dissector_key_tcp *key =
1543                         skb_flow_dissector_target(f->dissector,
1544                                                   FLOW_DISSECTOR_KEY_TCP,
1545                                                   f->key);
1546                 struct flow_dissector_key_tcp *mask =
1547                         skb_flow_dissector_target(f->dissector,
1548                                                   FLOW_DISSECTOR_KEY_TCP,
1549                                                   f->mask);
1550
1551                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
1552                          ntohs(mask->flags));
1553                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
1554                          ntohs(key->flags));
1555
1556                 if (mask->flags)
1557                         *match_level = MLX5_MATCH_L4;
1558         }
1559
1560         return 0;
1561 }
1562
1563 static int parse_cls_flower(struct mlx5e_priv *priv,
1564                             struct mlx5e_tc_flow *flow,
1565                             struct mlx5_flow_spec *spec,
1566                             struct tc_cls_flower_offload *f)
1567 {
1568         struct mlx5_core_dev *dev = priv->mdev;
1569         struct mlx5_eswitch *esw = dev->priv.eswitch;
1570         struct mlx5e_rep_priv *rpriv = priv->ppriv;
1571         struct mlx5_eswitch_rep *rep;
1572         u8 match_level;
1573         int err;
1574
1575         err = __parse_cls_flower(priv, spec, f, &match_level);
1576
1577         if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH)) {
1578                 rep = rpriv->rep;
1579                 if (rep->vport != FDB_UPLINK_VPORT &&
1580                     (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
1581                     esw->offloads.inline_mode < match_level)) {
1582                         netdev_warn(priv->netdev,
1583                                     "Flow is not offloaded due to min inline setting, required %d actual %d\n",
1584                                     match_level, esw->offloads.inline_mode);
1585                         return -EOPNOTSUPP;
1586                 }
1587         }
1588
1589         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1590                 flow->esw_attr->match_level = match_level;
1591         else
1592                 flow->nic_attr->match_level = match_level;
1593
1594         return err;
1595 }
1596
1597 struct pedit_headers {
1598         struct ethhdr  eth;
1599         struct iphdr   ip4;
1600         struct ipv6hdr ip6;
1601         struct tcphdr  tcp;
1602         struct udphdr  udp;
1603 };
1604
1605 static int pedit_header_offsets[] = {
1606         [TCA_PEDIT_KEY_EX_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
1607         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
1608         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
1609         [TCA_PEDIT_KEY_EX_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
1610         [TCA_PEDIT_KEY_EX_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
1611 };
1612
1613 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
1614
1615 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
1616                          struct pedit_headers *masks,
1617                          struct pedit_headers *vals)
1618 {
1619         u32 *curr_pmask, *curr_pval;
1620
1621         if (hdr_type >= __PEDIT_HDR_TYPE_MAX)
1622                 goto out_err;
1623
1624         curr_pmask = (u32 *)(pedit_header(masks, hdr_type) + offset);
1625         curr_pval  = (u32 *)(pedit_header(vals, hdr_type) + offset);
1626
1627         if (*curr_pmask & mask)  /* disallow acting twice on the same location */
1628                 goto out_err;
1629
1630         *curr_pmask |= mask;
1631         *curr_pval  |= (val & mask);
1632
1633         return 0;
1634
1635 out_err:
1636         return -EOPNOTSUPP;
1637 }
1638
1639 struct mlx5_fields {
1640         u8  field;
1641         u8  size;
1642         u32 offset;
1643 };
1644
1645 #define OFFLOAD(fw_field, size, field, off) \
1646                 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, size, offsetof(struct pedit_headers, field) + (off)}
1647
1648 static struct mlx5_fields fields[] = {
1649         OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0),
1650         OFFLOAD(DMAC_15_0,  2, eth.h_dest[4], 0),
1651         OFFLOAD(SMAC_47_16, 4, eth.h_source[0], 0),
1652         OFFLOAD(SMAC_15_0,  2, eth.h_source[4], 0),
1653         OFFLOAD(ETHERTYPE,  2, eth.h_proto, 0),
1654
1655         OFFLOAD(IP_TTL, 1, ip4.ttl,   0),
1656         OFFLOAD(SIPV4,  4, ip4.saddr, 0),
1657         OFFLOAD(DIPV4,  4, ip4.daddr, 0),
1658
1659         OFFLOAD(SIPV6_127_96, 4, ip6.saddr.s6_addr32[0], 0),
1660         OFFLOAD(SIPV6_95_64,  4, ip6.saddr.s6_addr32[1], 0),
1661         OFFLOAD(SIPV6_63_32,  4, ip6.saddr.s6_addr32[2], 0),
1662         OFFLOAD(SIPV6_31_0,   4, ip6.saddr.s6_addr32[3], 0),
1663         OFFLOAD(DIPV6_127_96, 4, ip6.daddr.s6_addr32[0], 0),
1664         OFFLOAD(DIPV6_95_64,  4, ip6.daddr.s6_addr32[1], 0),
1665         OFFLOAD(DIPV6_63_32,  4, ip6.daddr.s6_addr32[2], 0),
1666         OFFLOAD(DIPV6_31_0,   4, ip6.daddr.s6_addr32[3], 0),
1667         OFFLOAD(IPV6_HOPLIMIT, 1, ip6.hop_limit, 0),
1668
1669         OFFLOAD(TCP_SPORT, 2, tcp.source,  0),
1670         OFFLOAD(TCP_DPORT, 2, tcp.dest,    0),
1671         OFFLOAD(TCP_FLAGS, 1, tcp.ack_seq, 5),
1672
1673         OFFLOAD(UDP_SPORT, 2, udp.source, 0),
1674         OFFLOAD(UDP_DPORT, 2, udp.dest,   0),
1675 };
1676
1677 /* On input attr->num_mod_hdr_actions tells how many HW actions can be parsed at
1678  * max from the SW pedit action. On success, it says how many HW actions were
1679  * actually parsed.
1680  */
1681 static int offload_pedit_fields(struct pedit_headers *masks,
1682                                 struct pedit_headers *vals,
1683                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
1684 {
1685         struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
1686         int i, action_size, nactions, max_actions, first, last, next_z;
1687         void *s_masks_p, *a_masks_p, *vals_p;
1688         struct mlx5_fields *f;
1689         u8 cmd, field_bsize;
1690         u32 s_mask, a_mask;
1691         unsigned long mask;
1692         __be32 mask_be32;
1693         __be16 mask_be16;
1694         void *action;
1695
1696         set_masks = &masks[TCA_PEDIT_KEY_EX_CMD_SET];
1697         add_masks = &masks[TCA_PEDIT_KEY_EX_CMD_ADD];
1698         set_vals = &vals[TCA_PEDIT_KEY_EX_CMD_SET];
1699         add_vals = &vals[TCA_PEDIT_KEY_EX_CMD_ADD];
1700
1701         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1702         action = parse_attr->mod_hdr_actions;
1703         max_actions = parse_attr->num_mod_hdr_actions;
1704         nactions = 0;
1705
1706         for (i = 0; i < ARRAY_SIZE(fields); i++) {
1707                 f = &fields[i];
1708                 /* avoid seeing bits set from previous iterations */
1709                 s_mask = 0;
1710                 a_mask = 0;
1711
1712                 s_masks_p = (void *)set_masks + f->offset;
1713                 a_masks_p = (void *)add_masks + f->offset;
1714
1715                 memcpy(&s_mask, s_masks_p, f->size);
1716                 memcpy(&a_mask, a_masks_p, f->size);
1717
1718                 if (!s_mask && !a_mask) /* nothing to offload here */
1719                         continue;
1720
1721                 if (s_mask && a_mask) {
1722                         printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
1723                         return -EOPNOTSUPP;
1724                 }
1725
1726                 if (nactions == max_actions) {
1727                         printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
1728                         return -EOPNOTSUPP;
1729                 }
1730
1731                 if (s_mask) {
1732                         cmd  = MLX5_ACTION_TYPE_SET;
1733                         mask = s_mask;
1734                         vals_p = (void *)set_vals + f->offset;
1735                         /* clear to denote we consumed this field */
1736                         memset(s_masks_p, 0, f->size);
1737                 } else {
1738                         cmd  = MLX5_ACTION_TYPE_ADD;
1739                         mask = a_mask;
1740                         vals_p = (void *)add_vals + f->offset;
1741                         /* clear to denote we consumed this field */
1742                         memset(a_masks_p, 0, f->size);
1743                 }
1744
1745                 field_bsize = f->size * BITS_PER_BYTE;
1746
1747                 if (field_bsize == 32) {
1748                         mask_be32 = *(__be32 *)&mask;
1749                         mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
1750                 } else if (field_bsize == 16) {
1751                         mask_be16 = *(__be16 *)&mask;
1752                         mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
1753                 }
1754
1755                 first = find_first_bit(&mask, field_bsize);
1756                 next_z = find_next_zero_bit(&mask, field_bsize, first);
1757                 last  = find_last_bit(&mask, field_bsize);
1758                 if (first < next_z && next_z < last) {
1759                         printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
1760                                mask);
1761                         return -EOPNOTSUPP;
1762                 }
1763
1764                 MLX5_SET(set_action_in, action, action_type, cmd);
1765                 MLX5_SET(set_action_in, action, field, f->field);
1766
1767                 if (cmd == MLX5_ACTION_TYPE_SET) {
1768                         MLX5_SET(set_action_in, action, offset, first);
1769                         /* length is num of bits to be written, zero means length of 32 */
1770                         MLX5_SET(set_action_in, action, length, (last - first + 1));
1771                 }
1772
1773                 if (field_bsize == 32)
1774                         MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
1775                 else if (field_bsize == 16)
1776                         MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
1777                 else if (field_bsize == 8)
1778                         MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
1779
1780                 action += action_size;
1781                 nactions++;
1782         }
1783
1784         parse_attr->num_mod_hdr_actions = nactions;
1785         return 0;
1786 }
1787
1788 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
1789                                  const struct tc_action *a, int namespace,
1790                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1791 {
1792         int nkeys, action_size, max_actions;
1793
1794         nkeys = tcf_pedit_nkeys(a);
1795         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1796
1797         if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
1798                 max_actions = MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev, max_modify_header_actions);
1799         else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
1800                 max_actions = MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, max_modify_header_actions);
1801
1802         /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
1803         max_actions = min(max_actions, nkeys * 16);
1804
1805         parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
1806         if (!parse_attr->mod_hdr_actions)
1807                 return -ENOMEM;
1808
1809         parse_attr->num_mod_hdr_actions = max_actions;
1810         return 0;
1811 }
1812
1813 static const struct pedit_headers zero_masks = {};
1814
1815 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
1816                                  const struct tc_action *a, int namespace,
1817                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1818 {
1819         struct pedit_headers masks[__PEDIT_CMD_MAX], vals[__PEDIT_CMD_MAX], *cmd_masks;
1820         int nkeys, i, err = -EOPNOTSUPP;
1821         u32 mask, val, offset;
1822         u8 cmd, htype;
1823
1824         nkeys = tcf_pedit_nkeys(a);
1825
1826         memset(masks, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1827         memset(vals,  0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1828
1829         for (i = 0; i < nkeys; i++) {
1830                 htype = tcf_pedit_htype(a, i);
1831                 cmd = tcf_pedit_cmd(a, i);
1832                 err = -EOPNOTSUPP; /* can't be all optimistic */
1833
1834                 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK) {
1835                         netdev_warn(priv->netdev, "legacy pedit isn't offloaded\n");
1836                         goto out_err;
1837                 }
1838
1839                 if (cmd != TCA_PEDIT_KEY_EX_CMD_SET && cmd != TCA_PEDIT_KEY_EX_CMD_ADD) {
1840                         netdev_warn(priv->netdev, "pedit cmd %d isn't offloaded\n", cmd);
1841                         goto out_err;
1842                 }
1843
1844                 mask = tcf_pedit_mask(a, i);
1845                 val = tcf_pedit_val(a, i);
1846                 offset = tcf_pedit_offset(a, i);
1847
1848                 err = set_pedit_val(htype, ~mask, val, offset, &masks[cmd], &vals[cmd]);
1849                 if (err)
1850                         goto out_err;
1851         }
1852
1853         err = alloc_mod_hdr_actions(priv, a, namespace, parse_attr);
1854         if (err)
1855                 goto out_err;
1856
1857         err = offload_pedit_fields(masks, vals, parse_attr);
1858         if (err < 0)
1859                 goto out_dealloc_parsed_actions;
1860
1861         for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
1862                 cmd_masks = &masks[cmd];
1863                 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
1864                         netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
1865                         print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
1866                                        16, 1, cmd_masks, sizeof(zero_masks), true);
1867                         err = -EOPNOTSUPP;
1868                         goto out_dealloc_parsed_actions;
1869                 }
1870         }
1871
1872         return 0;
1873
1874 out_dealloc_parsed_actions:
1875         kfree(parse_attr->mod_hdr_actions);
1876 out_err:
1877         return err;
1878 }
1879
1880 static bool csum_offload_supported(struct mlx5e_priv *priv, u32 action, u32 update_flags)
1881 {
1882         u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
1883                          TCA_CSUM_UPDATE_FLAG_UDP;
1884
1885         /*  The HW recalcs checksums only if re-writing headers */
1886         if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
1887                 netdev_warn(priv->netdev,
1888                             "TC csum action is only offloaded with pedit\n");
1889                 return false;
1890         }
1891
1892         if (update_flags & ~prot_flags) {
1893                 netdev_warn(priv->netdev,
1894                             "can't offload TC csum action for some header/s - flags %#x\n",
1895                             update_flags);
1896                 return false;
1897         }
1898
1899         return true;
1900 }
1901
1902 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
1903                                           struct tcf_exts *exts)
1904 {
1905         const struct tc_action *a;
1906         bool modify_ip_header;
1907         LIST_HEAD(actions);
1908         u8 htype, ip_proto;
1909         void *headers_v;
1910         u16 ethertype;
1911         int nkeys, i;
1912
1913         headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
1914         ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
1915
1916         /* for non-IP we only re-write MACs, so we're okay */
1917         if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
1918                 goto out_ok;
1919
1920         modify_ip_header = false;
1921         tcf_exts_to_list(exts, &actions);
1922         list_for_each_entry(a, &actions, list) {
1923                 if (!is_tcf_pedit(a))
1924                         continue;
1925
1926                 nkeys = tcf_pedit_nkeys(a);
1927                 for (i = 0; i < nkeys; i++) {
1928                         htype = tcf_pedit_htype(a, i);
1929                         if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP4 ||
1930                             htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP6) {
1931                                 modify_ip_header = true;
1932                                 break;
1933                         }
1934                 }
1935         }
1936
1937         ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
1938         if (modify_ip_header && ip_proto != IPPROTO_TCP &&
1939             ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
1940                 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
1941                 return false;
1942         }
1943
1944 out_ok:
1945         return true;
1946 }
1947
1948 static bool actions_match_supported(struct mlx5e_priv *priv,
1949                                     struct tcf_exts *exts,
1950                                     struct mlx5e_tc_flow_parse_attr *parse_attr,
1951                                     struct mlx5e_tc_flow *flow)
1952 {
1953         u32 actions;
1954
1955         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1956                 actions = flow->esw_attr->action;
1957         else
1958                 actions = flow->nic_attr->action;
1959
1960         if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1961                 return modify_header_match_supported(&parse_attr->spec, exts);
1962
1963         return true;
1964 }
1965
1966 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
1967 {
1968         struct mlx5_core_dev *fmdev, *pmdev;
1969         u16 func_id, peer_id;
1970
1971         fmdev = priv->mdev;
1972         pmdev = peer_priv->mdev;
1973
1974         func_id = (u16)((fmdev->pdev->bus->number << 8) | PCI_SLOT(fmdev->pdev->devfn));
1975         peer_id = (u16)((pmdev->pdev->bus->number << 8) | PCI_SLOT(pmdev->pdev->devfn));
1976
1977         return (func_id == peer_id);
1978 }
1979
1980 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
1981                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
1982                                 struct mlx5e_tc_flow *flow)
1983 {
1984         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
1985         const struct tc_action *a;
1986         LIST_HEAD(actions);
1987         u32 action = 0;
1988         int err;
1989
1990         if (!tcf_exts_has_actions(exts))
1991                 return -EINVAL;
1992
1993         attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
1994
1995         tcf_exts_to_list(exts, &actions);
1996         list_for_each_entry(a, &actions, list) {
1997                 if (is_tcf_gact_shot(a)) {
1998                         action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
1999                         if (MLX5_CAP_FLOWTABLE(priv->mdev,
2000                                                flow_table_properties_nic_receive.flow_counter))
2001                                 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
2002                         continue;
2003                 }
2004
2005                 if (is_tcf_pedit(a)) {
2006                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_KERNEL,
2007                                                     parse_attr);
2008                         if (err)
2009                                 return err;
2010
2011                         action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
2012                                   MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2013                         continue;
2014                 }
2015
2016                 if (is_tcf_csum(a)) {
2017                         if (csum_offload_supported(priv, action,
2018                                                    tcf_csum_update_flags(a)))
2019                                 continue;
2020
2021                         return -EOPNOTSUPP;
2022                 }
2023
2024                 if (is_tcf_mirred_egress_redirect(a)) {
2025                         struct net_device *peer_dev = tcf_mirred_dev(a);
2026
2027                         if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
2028                             same_hw_devs(priv, netdev_priv(peer_dev))) {
2029                                 parse_attr->mirred_ifindex = peer_dev->ifindex;
2030                                 flow->flags |= MLX5E_TC_FLOW_HAIRPIN;
2031                                 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2032                                           MLX5_FLOW_CONTEXT_ACTION_COUNT;
2033                         } else {
2034                                 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
2035                                             peer_dev->name);
2036                                 return -EINVAL;
2037                         }
2038                         continue;
2039                 }
2040
2041                 if (is_tcf_skbedit_mark(a)) {
2042                         u32 mark = tcf_skbedit_mark(a);
2043
2044                         if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
2045                                 netdev_warn(priv->netdev, "Bad flow mark - only 16 bit is supported: 0x%x\n",
2046                                             mark);
2047                                 return -EINVAL;
2048                         }
2049
2050                         attr->flow_tag = mark;
2051                         action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2052                         continue;
2053                 }
2054
2055                 return -EINVAL;
2056         }
2057
2058         attr->action = action;
2059         if (!actions_match_supported(priv, exts, parse_attr, flow))
2060                 return -EOPNOTSUPP;
2061
2062         return 0;
2063 }
2064
2065 static inline int cmp_encap_info(struct ip_tunnel_key *a,
2066                                  struct ip_tunnel_key *b)
2067 {
2068         return memcmp(a, b, sizeof(*a));
2069 }
2070
2071 static inline int hash_encap_info(struct ip_tunnel_key *key)
2072 {
2073         return jhash(key, sizeof(*key), 0);
2074 }
2075
2076 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
2077                                    struct net_device *mirred_dev,
2078                                    struct net_device **out_dev,
2079                                    struct flowi4 *fl4,
2080                                    struct neighbour **out_n,
2081                                    int *out_ttl)
2082 {
2083         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2084         struct mlx5e_rep_priv *uplink_rpriv;
2085         struct rtable *rt;
2086         struct neighbour *n = NULL;
2087
2088 #if IS_ENABLED(CONFIG_INET)
2089         int ret;
2090
2091         rt = ip_route_output_key(dev_net(mirred_dev), fl4);
2092         ret = PTR_ERR_OR_ZERO(rt);
2093         if (ret)
2094                 return ret;
2095 #else
2096         return -EOPNOTSUPP;
2097 #endif
2098         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2099         /* if the egress device isn't on the same HW e-switch, we use the uplink */
2100         if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
2101                 *out_dev = uplink_rpriv->netdev;
2102         else
2103                 *out_dev = rt->dst.dev;
2104
2105         *out_ttl = ip4_dst_hoplimit(&rt->dst);
2106         n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
2107         ip_rt_put(rt);
2108         if (!n)
2109                 return -ENOMEM;
2110
2111         *out_n = n;
2112         return 0;
2113 }
2114
2115 static bool is_merged_eswitch_dev(struct mlx5e_priv *priv,
2116                                   struct net_device *peer_netdev)
2117 {
2118         struct mlx5e_priv *peer_priv;
2119
2120         peer_priv = netdev_priv(peer_netdev);
2121
2122         return (MLX5_CAP_ESW(priv->mdev, merged_eswitch) &&
2123                 (priv->netdev->netdev_ops == peer_netdev->netdev_ops) &&
2124                 same_hw_devs(priv, peer_priv) &&
2125                 MLX5_VPORT_MANAGER(peer_priv->mdev) &&
2126                 (peer_priv->mdev->priv.eswitch->mode == SRIOV_OFFLOADS));
2127 }
2128
2129 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
2130                                    struct net_device *mirred_dev,
2131                                    struct net_device **out_dev,
2132                                    struct flowi6 *fl6,
2133                                    struct neighbour **out_n,
2134                                    int *out_ttl)
2135 {
2136         struct neighbour *n = NULL;
2137         struct dst_entry *dst;
2138
2139 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
2140         struct mlx5e_rep_priv *uplink_rpriv;
2141         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2142         int ret;
2143
2144         ret = ipv6_stub->ipv6_dst_lookup(dev_net(mirred_dev), NULL, &dst,
2145                                          fl6);
2146         if (ret < 0)
2147                 return ret;
2148
2149         *out_ttl = ip6_dst_hoplimit(dst);
2150
2151         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2152         /* if the egress device isn't on the same HW e-switch, we use the uplink */
2153         if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
2154                 *out_dev = uplink_rpriv->netdev;
2155         else
2156                 *out_dev = dst->dev;
2157 #else
2158         return -EOPNOTSUPP;
2159 #endif
2160
2161         n = dst_neigh_lookup(dst, &fl6->daddr);
2162         dst_release(dst);
2163         if (!n)
2164                 return -ENOMEM;
2165
2166         *out_n = n;
2167         return 0;
2168 }
2169
2170 static void gen_vxlan_header_ipv4(struct net_device *out_dev,
2171                                   char buf[], int encap_size,
2172                                   unsigned char h_dest[ETH_ALEN],
2173                                   int ttl,
2174                                   __be32 daddr,
2175                                   __be32 saddr,
2176                                   __be16 udp_dst_port,
2177                                   __be32 vx_vni)
2178 {
2179         struct ethhdr *eth = (struct ethhdr *)buf;
2180         struct iphdr  *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
2181         struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
2182         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2183
2184         memset(buf, 0, encap_size);
2185
2186         ether_addr_copy(eth->h_dest, h_dest);
2187         ether_addr_copy(eth->h_source, out_dev->dev_addr);
2188         eth->h_proto = htons(ETH_P_IP);
2189
2190         ip->daddr = daddr;
2191         ip->saddr = saddr;
2192
2193         ip->ttl = ttl;
2194         ip->protocol = IPPROTO_UDP;
2195         ip->version = 0x4;
2196         ip->ihl = 0x5;
2197
2198         udp->dest = udp_dst_port;
2199         vxh->vx_flags = VXLAN_HF_VNI;
2200         vxh->vx_vni = vxlan_vni_field(vx_vni);
2201 }
2202
2203 static void gen_vxlan_header_ipv6(struct net_device *out_dev,
2204                                   char buf[], int encap_size,
2205                                   unsigned char h_dest[ETH_ALEN],
2206                                   int ttl,
2207                                   struct in6_addr *daddr,
2208                                   struct in6_addr *saddr,
2209                                   __be16 udp_dst_port,
2210                                   __be32 vx_vni)
2211 {
2212         struct ethhdr *eth = (struct ethhdr *)buf;
2213         struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
2214         struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
2215         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2216
2217         memset(buf, 0, encap_size);
2218
2219         ether_addr_copy(eth->h_dest, h_dest);
2220         ether_addr_copy(eth->h_source, out_dev->dev_addr);
2221         eth->h_proto = htons(ETH_P_IPV6);
2222
2223         ip6_flow_hdr(ip6h, 0, 0);
2224         /* the HW fills up ipv6 payload len */
2225         ip6h->nexthdr     = IPPROTO_UDP;
2226         ip6h->hop_limit   = ttl;
2227         ip6h->daddr       = *daddr;
2228         ip6h->saddr       = *saddr;
2229
2230         udp->dest = udp_dst_port;
2231         vxh->vx_flags = VXLAN_HF_VNI;
2232         vxh->vx_vni = vxlan_vni_field(vx_vni);
2233 }
2234
2235 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
2236                                           struct net_device *mirred_dev,
2237                                           struct mlx5e_encap_entry *e)
2238 {
2239         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2240         int ipv4_encap_size = ETH_HLEN + sizeof(struct iphdr) + VXLAN_HLEN;
2241         struct ip_tunnel_key *tun_key = &e->tun_info.key;
2242         struct net_device *out_dev;
2243         struct neighbour *n = NULL;
2244         struct flowi4 fl4 = {};
2245         char *encap_header;
2246         int ttl, err;
2247         u8 nud_state;
2248
2249         if (max_encap_size < ipv4_encap_size) {
2250                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2251                                ipv4_encap_size, max_encap_size);
2252                 return -EOPNOTSUPP;
2253         }
2254
2255         encap_header = kzalloc(ipv4_encap_size, GFP_KERNEL);
2256         if (!encap_header)
2257                 return -ENOMEM;
2258
2259         switch (e->tunnel_type) {
2260         case MLX5_HEADER_TYPE_VXLAN:
2261                 fl4.flowi4_proto = IPPROTO_UDP;
2262                 fl4.fl4_dport = tun_key->tp_dst;
2263                 break;
2264         default:
2265                 err = -EOPNOTSUPP;
2266                 goto free_encap;
2267         }
2268         fl4.flowi4_tos = tun_key->tos;
2269         fl4.daddr = tun_key->u.ipv4.dst;
2270         fl4.saddr = tun_key->u.ipv4.src;
2271
2272         err = mlx5e_route_lookup_ipv4(priv, mirred_dev, &out_dev,
2273                                       &fl4, &n, &ttl);
2274         if (err)
2275                 goto free_encap;
2276
2277         /* used by mlx5e_detach_encap to lookup a neigh hash table
2278          * entry in the neigh hash table when a user deletes a rule
2279          */
2280         e->m_neigh.dev = n->dev;
2281         e->m_neigh.family = n->ops->family;
2282         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2283         e->out_dev = out_dev;
2284
2285         /* It's importent to add the neigh to the hash table before checking
2286          * the neigh validity state. So if we'll get a notification, in case the
2287          * neigh changes it's validity state, we would find the relevant neigh
2288          * in the hash.
2289          */
2290         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2291         if (err)
2292                 goto free_encap;
2293
2294         read_lock_bh(&n->lock);
2295         nud_state = n->nud_state;
2296         ether_addr_copy(e->h_dest, n->ha);
2297         read_unlock_bh(&n->lock);
2298
2299         switch (e->tunnel_type) {
2300         case MLX5_HEADER_TYPE_VXLAN:
2301                 gen_vxlan_header_ipv4(out_dev, encap_header,
2302                                       ipv4_encap_size, e->h_dest, ttl,
2303                                       fl4.daddr,
2304                                       fl4.saddr, tun_key->tp_dst,
2305                                       tunnel_id_to_key32(tun_key->tun_id));
2306                 break;
2307         default:
2308                 err = -EOPNOTSUPP;
2309                 goto destroy_neigh_entry;
2310         }
2311         e->encap_size = ipv4_encap_size;
2312         e->encap_header = encap_header;
2313
2314         if (!(nud_state & NUD_VALID)) {
2315                 neigh_event_send(n, NULL);
2316                 err = -EAGAIN;
2317                 goto out;
2318         }
2319
2320         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
2321                                ipv4_encap_size, encap_header, &e->encap_id);
2322         if (err)
2323                 goto destroy_neigh_entry;
2324
2325         e->flags |= MLX5_ENCAP_ENTRY_VALID;
2326         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2327         neigh_release(n);
2328         return err;
2329
2330 destroy_neigh_entry:
2331         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2332 free_encap:
2333         kfree(encap_header);
2334 out:
2335         if (n)
2336                 neigh_release(n);
2337         return err;
2338 }
2339
2340 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
2341                                           struct net_device *mirred_dev,
2342                                           struct mlx5e_encap_entry *e)
2343 {
2344         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2345         int ipv6_encap_size = ETH_HLEN + sizeof(struct ipv6hdr) + VXLAN_HLEN;
2346         struct ip_tunnel_key *tun_key = &e->tun_info.key;
2347         struct net_device *out_dev;
2348         struct neighbour *n = NULL;
2349         struct flowi6 fl6 = {};
2350         char *encap_header;
2351         int err, ttl = 0;
2352         u8 nud_state;
2353
2354         if (max_encap_size < ipv6_encap_size) {
2355                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2356                                ipv6_encap_size, max_encap_size);
2357                 return -EOPNOTSUPP;
2358         }
2359
2360         encap_header = kzalloc(ipv6_encap_size, GFP_KERNEL);
2361         if (!encap_header)
2362                 return -ENOMEM;
2363
2364         switch (e->tunnel_type) {
2365         case MLX5_HEADER_TYPE_VXLAN:
2366                 fl6.flowi6_proto = IPPROTO_UDP;
2367                 fl6.fl6_dport = tun_key->tp_dst;
2368                 break;
2369         default:
2370                 err = -EOPNOTSUPP;
2371                 goto free_encap;
2372         }
2373
2374         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
2375         fl6.daddr = tun_key->u.ipv6.dst;
2376         fl6.saddr = tun_key->u.ipv6.src;
2377
2378         err = mlx5e_route_lookup_ipv6(priv, mirred_dev, &out_dev,
2379                                       &fl6, &n, &ttl);
2380         if (err)
2381                 goto free_encap;
2382
2383         /* used by mlx5e_detach_encap to lookup a neigh hash table
2384          * entry in the neigh hash table when a user deletes a rule
2385          */
2386         e->m_neigh.dev = n->dev;
2387         e->m_neigh.family = n->ops->family;
2388         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2389         e->out_dev = out_dev;
2390
2391         /* It's importent to add the neigh to the hash table before checking
2392          * the neigh validity state. So if we'll get a notification, in case the
2393          * neigh changes it's validity state, we would find the relevant neigh
2394          * in the hash.
2395          */
2396         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2397         if (err)
2398                 goto free_encap;
2399
2400         read_lock_bh(&n->lock);
2401         nud_state = n->nud_state;
2402         ether_addr_copy(e->h_dest, n->ha);
2403         read_unlock_bh(&n->lock);
2404
2405         switch (e->tunnel_type) {
2406         case MLX5_HEADER_TYPE_VXLAN:
2407                 gen_vxlan_header_ipv6(out_dev, encap_header,
2408                                       ipv6_encap_size, e->h_dest, ttl,
2409                                       &fl6.daddr,
2410                                       &fl6.saddr, tun_key->tp_dst,
2411                                       tunnel_id_to_key32(tun_key->tun_id));
2412                 break;
2413         default:
2414                 err = -EOPNOTSUPP;
2415                 goto destroy_neigh_entry;
2416         }
2417
2418         e->encap_size = ipv6_encap_size;
2419         e->encap_header = encap_header;
2420
2421         if (!(nud_state & NUD_VALID)) {
2422                 neigh_event_send(n, NULL);
2423                 err = -EAGAIN;
2424                 goto out;
2425         }
2426
2427         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
2428                                ipv6_encap_size, encap_header, &e->encap_id);
2429         if (err)
2430                 goto destroy_neigh_entry;
2431
2432         e->flags |= MLX5_ENCAP_ENTRY_VALID;
2433         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2434         neigh_release(n);
2435         return err;
2436
2437 destroy_neigh_entry:
2438         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2439 free_encap:
2440         kfree(encap_header);
2441 out:
2442         if (n)
2443                 neigh_release(n);
2444         return err;
2445 }
2446
2447 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
2448                               struct ip_tunnel_info *tun_info,
2449                               struct net_device *mirred_dev,
2450                               struct net_device **encap_dev,
2451                               struct mlx5e_tc_flow *flow)
2452 {
2453         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2454         struct mlx5e_rep_priv *uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw,
2455                                                                            REP_ETH);
2456         struct net_device *up_dev = uplink_rpriv->netdev;
2457         unsigned short family = ip_tunnel_info_af(tun_info);
2458         struct mlx5e_priv *up_priv = netdev_priv(up_dev);
2459         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2460         struct ip_tunnel_key *key = &tun_info->key;
2461         struct mlx5e_encap_entry *e;
2462         int tunnel_type, err = 0;
2463         uintptr_t hash_key;
2464         bool found = false;
2465
2466         /* udp dst port must be set */
2467         if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
2468                 goto vxlan_encap_offload_err;
2469
2470         /* setting udp src port isn't supported */
2471         if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
2472 vxlan_encap_offload_err:
2473                 netdev_warn(priv->netdev,
2474                             "must set udp dst port and not set udp src port\n");
2475                 return -EOPNOTSUPP;
2476         }
2477
2478         if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->tp_dst)) &&
2479             MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
2480                 tunnel_type = MLX5_HEADER_TYPE_VXLAN;
2481         } else {
2482                 netdev_warn(priv->netdev,
2483                             "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
2484                 return -EOPNOTSUPP;
2485         }
2486
2487         hash_key = hash_encap_info(key);
2488
2489         hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
2490                                    encap_hlist, hash_key) {
2491                 if (!cmp_encap_info(&e->tun_info.key, key)) {
2492                         found = true;
2493                         break;
2494                 }
2495         }
2496
2497         /* must verify if encap is valid or not */
2498         if (found)
2499                 goto attach_flow;
2500
2501         e = kzalloc(sizeof(*e), GFP_KERNEL);
2502         if (!e)
2503                 return -ENOMEM;
2504
2505         e->tun_info = *tun_info;
2506         e->tunnel_type = tunnel_type;
2507         INIT_LIST_HEAD(&e->flows);
2508
2509         if (family == AF_INET)
2510                 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e);
2511         else if (family == AF_INET6)
2512                 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e);
2513
2514         if (err && err != -EAGAIN)
2515                 goto out_err;
2516
2517         hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
2518
2519 attach_flow:
2520         list_add(&flow->encap, &e->flows);
2521         *encap_dev = e->out_dev;
2522         if (e->flags & MLX5_ENCAP_ENTRY_VALID)
2523                 attr->encap_id = e->encap_id;
2524         else
2525                 err = -EAGAIN;
2526
2527         return err;
2528
2529 out_err:
2530         kfree(e);
2531         return err;
2532 }
2533
2534 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
2535                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
2536                                 struct mlx5e_tc_flow *flow)
2537 {
2538         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2539         struct mlx5e_rep_priv *rpriv = priv->ppriv;
2540         struct ip_tunnel_info *info = NULL;
2541         const struct tc_action *a;
2542         LIST_HEAD(actions);
2543         bool encap = false;
2544         u32 action = 0;
2545
2546         if (!tcf_exts_has_actions(exts))
2547                 return -EINVAL;
2548
2549         attr->in_rep = rpriv->rep;
2550         attr->in_mdev = priv->mdev;
2551
2552         tcf_exts_to_list(exts, &actions);
2553         list_for_each_entry(a, &actions, list) {
2554                 if (is_tcf_gact_shot(a)) {
2555                         action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
2556                                   MLX5_FLOW_CONTEXT_ACTION_COUNT;
2557                         continue;
2558                 }
2559
2560                 if (is_tcf_pedit(a)) {
2561                         int err;
2562
2563                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_FDB,
2564                                                     parse_attr);
2565                         if (err)
2566                                 return err;
2567
2568                         action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2569                         attr->mirror_count = attr->out_count;
2570                         continue;
2571                 }
2572
2573                 if (is_tcf_csum(a)) {
2574                         if (csum_offload_supported(priv, action,
2575                                                    tcf_csum_update_flags(a)))
2576                                 continue;
2577
2578                         return -EOPNOTSUPP;
2579                 }
2580
2581                 if (is_tcf_mirred_egress_redirect(a) || is_tcf_mirred_egress_mirror(a)) {
2582                         struct mlx5e_priv *out_priv;
2583                         struct net_device *out_dev;
2584
2585                         out_dev = tcf_mirred_dev(a);
2586
2587                         if (attr->out_count >= MLX5_MAX_FLOW_FWD_VPORTS) {
2588                                 pr_err("can't support more than %d output ports, can't offload forwarding\n",
2589                                        attr->out_count);
2590                                 return -EOPNOTSUPP;
2591                         }
2592
2593                         if (switchdev_port_same_parent_id(priv->netdev,
2594                                                           out_dev) ||
2595                             is_merged_eswitch_dev(priv, out_dev)) {
2596                                 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2597                                           MLX5_FLOW_CONTEXT_ACTION_COUNT;
2598                                 out_priv = netdev_priv(out_dev);
2599                                 rpriv = out_priv->ppriv;
2600                                 attr->out_rep[attr->out_count] = rpriv->rep;
2601                                 attr->out_mdev[attr->out_count++] = out_priv->mdev;
2602                         } else if (encap) {
2603                                 parse_attr->mirred_ifindex = out_dev->ifindex;
2604                                 parse_attr->tun_info = *info;
2605                                 attr->parse_attr = parse_attr;
2606                                 action |= MLX5_FLOW_CONTEXT_ACTION_ENCAP |
2607                                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2608                                           MLX5_FLOW_CONTEXT_ACTION_COUNT;
2609                                 /* attr->out_rep is resolved when we handle encap */
2610                         } else {
2611                                 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
2612                                        priv->netdev->name, out_dev->name);
2613                                 return -EINVAL;
2614                         }
2615                         continue;
2616                 }
2617
2618                 if (is_tcf_tunnel_set(a)) {
2619                         info = tcf_tunnel_info(a);
2620                         if (info)
2621                                 encap = true;
2622                         else
2623                                 return -EOPNOTSUPP;
2624                         attr->mirror_count = attr->out_count;
2625                         continue;
2626                 }
2627
2628                 if (is_tcf_vlan(a)) {
2629                         if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
2630                                 action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2631                         } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
2632                                 action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
2633                                 attr->vlan_vid = tcf_vlan_push_vid(a);
2634                                 if (mlx5_eswitch_vlan_actions_supported(priv->mdev)) {
2635                                         attr->vlan_prio = tcf_vlan_push_prio(a);
2636                                         attr->vlan_proto = tcf_vlan_push_proto(a);
2637                                         if (!attr->vlan_proto)
2638                                                 attr->vlan_proto = htons(ETH_P_8021Q);
2639                                 } else if (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q) ||
2640                                            tcf_vlan_push_prio(a)) {
2641                                         return -EOPNOTSUPP;
2642                                 }
2643                         } else { /* action is TCA_VLAN_ACT_MODIFY */
2644                                 return -EOPNOTSUPP;
2645                         }
2646                         attr->mirror_count = attr->out_count;
2647                         continue;
2648                 }
2649
2650                 if (is_tcf_tunnel_release(a)) {
2651                         action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
2652                         continue;
2653                 }
2654
2655                 return -EINVAL;
2656         }
2657
2658         attr->action = action;
2659         if (!actions_match_supported(priv, exts, parse_attr, flow))
2660                 return -EOPNOTSUPP;
2661
2662         if (attr->out_count > 1 && !mlx5_esw_has_fwd_fdb(priv->mdev)) {
2663                 netdev_warn_once(priv->netdev, "current firmware doesn't support split rule for port mirroring\n");
2664                 return -EOPNOTSUPP;
2665         }
2666
2667         return 0;
2668 }
2669
2670 static void get_flags(int flags, u8 *flow_flags)
2671 {
2672         u8 __flow_flags = 0;
2673
2674         if (flags & MLX5E_TC_INGRESS)
2675                 __flow_flags |= MLX5E_TC_FLOW_INGRESS;
2676         if (flags & MLX5E_TC_EGRESS)
2677                 __flow_flags |= MLX5E_TC_FLOW_EGRESS;
2678
2679         *flow_flags = __flow_flags;
2680 }
2681
2682 static const struct rhashtable_params tc_ht_params = {
2683         .head_offset = offsetof(struct mlx5e_tc_flow, node),
2684         .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
2685         .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
2686         .automatic_shrinking = true,
2687 };
2688
2689 static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv)
2690 {
2691         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2692         struct mlx5e_rep_priv *uplink_rpriv;
2693
2694         if (MLX5_VPORT_MANAGER(priv->mdev) && esw->mode == SRIOV_OFFLOADS) {
2695                 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2696                 return &uplink_rpriv->tc_ht;
2697         } else
2698                 return &priv->fs.tc.ht;
2699 }
2700
2701 int mlx5e_configure_flower(struct mlx5e_priv *priv,
2702                            struct tc_cls_flower_offload *f, int flags)
2703 {
2704         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2705         struct mlx5e_tc_flow_parse_attr *parse_attr;
2706         struct rhashtable *tc_ht = get_tc_ht(priv);
2707         struct mlx5e_tc_flow *flow;
2708         int attr_size, err = 0;
2709         u8 flow_flags = 0;
2710
2711         get_flags(flags, &flow_flags);
2712
2713         flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
2714         if (flow) {
2715                 netdev_warn_once(priv->netdev, "flow cookie %lx already exists, ignoring\n", f->cookie);
2716                 return 0;
2717         }
2718
2719         if (esw && esw->mode == SRIOV_OFFLOADS) {
2720                 flow_flags |= MLX5E_TC_FLOW_ESWITCH;
2721                 attr_size  = sizeof(struct mlx5_esw_flow_attr);
2722         } else {
2723                 flow_flags |= MLX5E_TC_FLOW_NIC;
2724                 attr_size  = sizeof(struct mlx5_nic_flow_attr);
2725         }
2726
2727         flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
2728         parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
2729         if (!parse_attr || !flow) {
2730                 err = -ENOMEM;
2731                 goto err_free;
2732         }
2733
2734         flow->cookie = f->cookie;
2735         flow->flags = flow_flags;
2736         flow->priv = priv;
2737
2738         err = parse_cls_flower(priv, flow, &parse_attr->spec, f);
2739         if (err < 0)
2740                 goto err_free;
2741
2742         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
2743                 err = parse_tc_fdb_actions(priv, f->exts, parse_attr, flow);
2744                 if (err < 0)
2745                         goto err_free;
2746                 flow->rule[0] = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow);
2747         } else {
2748                 err = parse_tc_nic_actions(priv, f->exts, parse_attr, flow);
2749                 if (err < 0)
2750                         goto err_free;
2751                 flow->rule[0] = mlx5e_tc_add_nic_flow(priv, parse_attr, flow);
2752         }
2753
2754         if (IS_ERR(flow->rule[0])) {
2755                 err = PTR_ERR(flow->rule[0]);
2756                 if (err != -EAGAIN)
2757                         goto err_free;
2758         }
2759
2760         if (err != -EAGAIN)
2761                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
2762
2763         if (!(flow->flags & MLX5E_TC_FLOW_ESWITCH) ||
2764             !(flow->esw_attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP))
2765                 kvfree(parse_attr);
2766
2767         err = rhashtable_insert_fast(tc_ht, &flow->node, tc_ht_params);
2768         if (err) {
2769                 mlx5e_tc_del_flow(priv, flow);
2770                 kfree(flow);
2771         }
2772
2773         return err;
2774
2775 err_free:
2776         kvfree(parse_attr);
2777         kfree(flow);
2778         return err;
2779 }
2780
2781 #define DIRECTION_MASK (MLX5E_TC_INGRESS | MLX5E_TC_EGRESS)
2782 #define FLOW_DIRECTION_MASK (MLX5E_TC_FLOW_INGRESS | MLX5E_TC_FLOW_EGRESS)
2783
2784 static bool same_flow_direction(struct mlx5e_tc_flow *flow, int flags)
2785 {
2786         if ((flow->flags & FLOW_DIRECTION_MASK) == (flags & DIRECTION_MASK))
2787                 return true;
2788
2789         return false;
2790 }
2791
2792 int mlx5e_delete_flower(struct mlx5e_priv *priv,
2793                         struct tc_cls_flower_offload *f, int flags)
2794 {
2795         struct rhashtable *tc_ht = get_tc_ht(priv);
2796         struct mlx5e_tc_flow *flow;
2797
2798         flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
2799         if (!flow || !same_flow_direction(flow, flags))
2800                 return -EINVAL;
2801
2802         rhashtable_remove_fast(tc_ht, &flow->node, tc_ht_params);
2803
2804         mlx5e_tc_del_flow(priv, flow);
2805
2806         kfree(flow);
2807
2808         return 0;
2809 }
2810
2811 int mlx5e_stats_flower(struct mlx5e_priv *priv,
2812                        struct tc_cls_flower_offload *f, int flags)
2813 {
2814         struct rhashtable *tc_ht = get_tc_ht(priv);
2815         struct mlx5e_tc_flow *flow;
2816         struct mlx5_fc *counter;
2817         u64 bytes;
2818         u64 packets;
2819         u64 lastuse;
2820
2821         flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
2822         if (!flow || !same_flow_direction(flow, flags))
2823                 return -EINVAL;
2824
2825         if (!(flow->flags & MLX5E_TC_FLOW_OFFLOADED))
2826                 return 0;
2827
2828         counter = mlx5_flow_rule_counter(flow->rule[0]);
2829         if (!counter)
2830                 return 0;
2831
2832         mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
2833
2834         tcf_exts_stats_update(f->exts, bytes, packets, lastuse);
2835
2836         return 0;
2837 }
2838
2839 int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
2840 {
2841         struct mlx5e_tc_table *tc = &priv->fs.tc;
2842
2843         hash_init(tc->mod_hdr_tbl);
2844         hash_init(tc->hairpin_tbl);
2845
2846         return rhashtable_init(&tc->ht, &tc_ht_params);
2847 }
2848
2849 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
2850 {
2851         struct mlx5e_tc_flow *flow = ptr;
2852         struct mlx5e_priv *priv = flow->priv;
2853
2854         mlx5e_tc_del_flow(priv, flow);
2855         kfree(flow);
2856 }
2857
2858 void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
2859 {
2860         struct mlx5e_tc_table *tc = &priv->fs.tc;
2861
2862         rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, NULL);
2863
2864         if (!IS_ERR_OR_NULL(tc->t)) {
2865                 mlx5_destroy_flow_table(tc->t);
2866                 tc->t = NULL;
2867         }
2868 }
2869
2870 int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
2871 {
2872         return rhashtable_init(tc_ht, &tc_ht_params);
2873 }
2874
2875 void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
2876 {
2877         rhashtable_free_and_destroy(tc_ht, _mlx5e_tc_del_flow, NULL);
2878 }
2879
2880 int mlx5e_tc_num_filters(struct mlx5e_priv *priv)
2881 {
2882         struct rhashtable *tc_ht = get_tc_ht(priv);
2883
2884         return atomic_read(&tc_ht->nelems);
2885 }