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 <linux/refcount.h>
  42 #include <linux/completion.h>
  43 #include <net/tc_act/tc_mirred.h>
  44 #include <net/tc_act/tc_vlan.h>
  45 #include <net/tc_act/tc_tunnel_key.h>
  46 #include <net/tc_act/tc_pedit.h>
  47 #include <net/tc_act/tc_csum.h>
  48 #include <net/arp.h>
  49 #include <net/ipv6_stubs.h>
  50 #include "en.h"
  51 #include "en_rep.h"
  52 #include "en_tc.h"
  53 #include "eswitch.h"
  54 #include "esw/chains.h"
  55 #include "fs_core.h"
  56 #include "en/port.h"
  57 #include "en/tc_tun.h"
  58 #include "en/mapping.h"
  59 #include "en/tc_ct.h"
  60 #include "lib/devcom.h"
  61 #include "lib/geneve.h"
  62 #include "diag/en_tc_tracepoint.h"
  63
  64 #define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)
  65
  66 struct mlx5_nic_flow_attr {
  67         u32 action;
  68         u32 flow_tag;
  69         struct mlx5_modify_hdr *modify_hdr;
  70         u32 hairpin_tirn;
  71         u8 match_level;
  72         struct mlx5_flow_table  *hairpin_ft;
  73         struct mlx5_fc          *counter;
  74 };
  75
  76 #define MLX5E_TC_FLOW_BASE (MLX5E_TC_FLAG_LAST_EXPORTED_BIT + 1)
  77
  78 enum {
  79         MLX5E_TC_FLOW_FLAG_INGRESS      = MLX5E_TC_FLAG_INGRESS_BIT,
  80         MLX5E_TC_FLOW_FLAG_EGRESS       = MLX5E_TC_FLAG_EGRESS_BIT,
  81         MLX5E_TC_FLOW_FLAG_ESWITCH      = MLX5E_TC_FLAG_ESW_OFFLOAD_BIT,
  82         MLX5E_TC_FLOW_FLAG_FT           = MLX5E_TC_FLAG_FT_OFFLOAD_BIT,
  83         MLX5E_TC_FLOW_FLAG_NIC          = MLX5E_TC_FLAG_NIC_OFFLOAD_BIT,
  84         MLX5E_TC_FLOW_FLAG_OFFLOADED    = MLX5E_TC_FLOW_BASE,
  85         MLX5E_TC_FLOW_FLAG_HAIRPIN      = MLX5E_TC_FLOW_BASE + 1,
  86         MLX5E_TC_FLOW_FLAG_HAIRPIN_RSS  = MLX5E_TC_FLOW_BASE + 2,
  87         MLX5E_TC_FLOW_FLAG_SLOW         = MLX5E_TC_FLOW_BASE + 3,
  88         MLX5E_TC_FLOW_FLAG_DUP          = MLX5E_TC_FLOW_BASE + 4,
  89         MLX5E_TC_FLOW_FLAG_NOT_READY    = MLX5E_TC_FLOW_BASE + 5,
  90         MLX5E_TC_FLOW_FLAG_DELETED      = MLX5E_TC_FLOW_BASE + 6,
  91         MLX5E_TC_FLOW_FLAG_CT           = MLX5E_TC_FLOW_BASE + 7,
  92 };
  93
  94 #define MLX5E_TC_MAX_SPLITS 1
  95
  96 /* Helper struct for accessing a struct containing list_head array.
  97  * Containing struct
  98  *   |- Helper array
  99  *      [0] Helper item 0
 100  *          |- list_head item 0
 101  *          |- index (0)
 102  *      [1] Helper item 1
 103  *          |- list_head item 1
 104  *          |- index (1)
 105  * To access the containing struct from one of the list_head items:
 106  * 1. Get the helper item from the list_head item using
 107  *    helper item =
 108  *        container_of(list_head item, helper struct type, list_head field)
 109  * 2. Get the contining struct from the helper item and its index in the array:
 110  *    containing struct =
 111  *        container_of(helper item, containing struct type, helper field[index])
 112  */
 113 struct encap_flow_item {
 114         struct mlx5e_encap_entry *e; /* attached encap instance */
 115         struct list_head list;
 116         int index;
 117 };
 118
 119 struct mlx5e_tc_flow {
 120         struct rhash_head       node;
 121         struct mlx5e_priv       *priv;
 122         u64                     cookie;
 123         unsigned long           flags;
 124         struct mlx5_flow_handle *rule[MLX5E_TC_MAX_SPLITS + 1];
 125         /* Flow can be associated with multiple encap IDs.
 126          * The number of encaps is bounded by the number of supported
 127          * destinations.
 128          */
 129         struct encap_flow_item encaps[MLX5_MAX_FLOW_FWD_VPORTS];
 130         struct mlx5e_tc_flow    *peer_flow;
 131         struct mlx5e_mod_hdr_entry *mh; /* attached mod header instance */
 132         struct list_head        mod_hdr; /* flows sharing the same mod hdr ID */
 133         struct mlx5e_hairpin_entry *hpe; /* attached hairpin instance */
 134         struct list_head        hairpin; /* flows sharing the same hairpin */
 135         struct list_head        peer;    /* flows with peer flow */
 136         struct list_head        unready; /* flows not ready to be offloaded (e.g due to missing route) */
 137         int                     tmp_efi_index;
 138         struct list_head        tmp_list; /* temporary flow list used by neigh update */
 139         refcount_t              refcnt;
 140         struct rcu_head         rcu_head;
 141         struct completion       init_done;
 142         int tunnel_id; /* the mapped tunnel id of this flow */
 143
 144         union {
 145                 struct mlx5_esw_flow_attr esw_attr[0];
 146                 struct mlx5_nic_flow_attr nic_attr[0];
 147         };
 148 };
 149
 150 struct mlx5e_tc_flow_parse_attr {
 151         const struct ip_tunnel_info *tun_info[MLX5_MAX_FLOW_FWD_VPORTS];
 152         struct net_device *filter_dev;
 153         struct mlx5_flow_spec spec;
 154         struct mlx5e_tc_mod_hdr_acts mod_hdr_acts;
 155         int mirred_ifindex[MLX5_MAX_FLOW_FWD_VPORTS];
 156 };
 157
 158 #define MLX5E_TC_TABLE_NUM_GROUPS 4
 159 #define MLX5E_TC_TABLE_MAX_GROUP_SIZE BIT(16)
 160
 161 struct tunnel_match_key {
 162         struct flow_dissector_key_control enc_control;
 163         struct flow_dissector_key_keyid enc_key_id;
 164         struct flow_dissector_key_ports enc_tp;
 165         struct flow_dissector_key_ip enc_ip;
 166         union {
 167                 struct flow_dissector_key_ipv4_addrs enc_ipv4;
 168                 struct flow_dissector_key_ipv6_addrs enc_ipv6;
 169         };
 170
 171         int filter_ifindex;
 172 };
 173
 174 /* Tunnel_id mapping is TUNNEL_INFO_BITS + ENC_OPTS_BITS.
 175  * Upper TUNNEL_INFO_BITS for general tunnel info.
 176  * Lower ENC_OPTS_BITS bits for enc_opts.
 177  */
 178 #define TUNNEL_INFO_BITS 6
 179 #define TUNNEL_INFO_BITS_MASK GENMASK(TUNNEL_INFO_BITS - 1, 0)
 180 #define ENC_OPTS_BITS 2
 181 #define ENC_OPTS_BITS_MASK GENMASK(ENC_OPTS_BITS - 1, 0)
 182 #define TUNNEL_ID_BITS (TUNNEL_INFO_BITS + ENC_OPTS_BITS)
 183 #define TUNNEL_ID_MASK GENMASK(TUNNEL_ID_BITS - 1, 0)
 184
 185 struct mlx5e_tc_attr_to_reg_mapping mlx5e_tc_attr_to_reg_mappings[] = {
 186         [CHAIN_TO_REG] = {
 187                 .mfield = MLX5_ACTION_IN_FIELD_METADATA_REG_C_0,
 188                 .moffset = 0,
 189                 .mlen = 2,
 190         },
 191         [TUNNEL_TO_REG] = {
 192                 .mfield = MLX5_ACTION_IN_FIELD_METADATA_REG_C_1,
 193                 .moffset = 3,
 194                 .mlen = 1,
 195                 .soffset = MLX5_BYTE_OFF(fte_match_param,
 196                                          misc_parameters_2.metadata_reg_c_1),
 197         },
 198         [ZONE_TO_REG] = zone_to_reg_ct,
 199         [CTSTATE_TO_REG] = ctstate_to_reg_ct,
 200         [MARK_TO_REG] = mark_to_reg_ct,
 201         [LABELS_TO_REG] = labels_to_reg_ct,
 202         [FTEID_TO_REG] = fteid_to_reg_ct,
 203         [TUPLEID_TO_REG] = tupleid_to_reg_ct,
 204 };
 205
 206 static void mlx5e_put_flow_tunnel_id(struct mlx5e_tc_flow *flow);
 207
 208 void
 209 mlx5e_tc_match_to_reg_match(struct mlx5_flow_spec *spec,
 210                             enum mlx5e_tc_attr_to_reg type,
 211                             u32 data,
 212                             u32 mask)
 213 {
 214         int soffset = mlx5e_tc_attr_to_reg_mappings[type].soffset;
 215         int match_len = mlx5e_tc_attr_to_reg_mappings[type].mlen;
 216         void *headers_c = spec->match_criteria;
 217         void *headers_v = spec->match_value;
 218         void *fmask, *fval;
 219
 220         fmask = headers_c + soffset;
 221         fval = headers_v + soffset;
 222
 223         mask = cpu_to_be32(mask) >> (32 - (match_len * 8));
 224         data = cpu_to_be32(data) >> (32 - (match_len * 8));
 225
 226         memcpy(fmask, &mask, match_len);
 227         memcpy(fval, &data, match_len);
 228
 229         spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS_2;
 230 }
 231
 232 int
 233 mlx5e_tc_match_to_reg_set(struct mlx5_core_dev *mdev,
 234                           struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts,
 235                           enum mlx5e_tc_attr_to_reg type,
 236                           u32 data)
 237 {
 238         int moffset = mlx5e_tc_attr_to_reg_mappings[type].moffset;
 239         int mfield = mlx5e_tc_attr_to_reg_mappings[type].mfield;
 240         int mlen = mlx5e_tc_attr_to_reg_mappings[type].mlen;
 241         char *modact;
 242         int err;
 243
 244         err = alloc_mod_hdr_actions(mdev, MLX5_FLOW_NAMESPACE_FDB,
 245                                     mod_hdr_acts);
 246         if (err)
 247                 return err;
 248
 249         modact = mod_hdr_acts->actions +
 250                  (mod_hdr_acts->num_actions * MLX5_MH_ACT_SZ);
 251
 252         /* Firmware has 5bit length field and 0 means 32bits */
 253         if (mlen == 4)
 254                 mlen = 0;
 255
 256         MLX5_SET(set_action_in, modact, action_type, MLX5_ACTION_TYPE_SET);
 257         MLX5_SET(set_action_in, modact, field, mfield);
 258         MLX5_SET(set_action_in, modact, offset, moffset * 8);
 259         MLX5_SET(set_action_in, modact, length, mlen * 8);
 260         MLX5_SET(set_action_in, modact, data, data);
 261         mod_hdr_acts->num_actions++;
 262
 263         return 0;
 264 }
 265
 266 struct mlx5e_hairpin {
 267         struct mlx5_hairpin *pair;
 268
 269         struct mlx5_core_dev *func_mdev;
 270         struct mlx5e_priv *func_priv;
 271         u32 tdn;
 272         u32 tirn;
 273
 274         int num_channels;
 275         struct mlx5e_rqt indir_rqt;
 276         u32 indir_tirn[MLX5E_NUM_INDIR_TIRS];
 277         struct mlx5e_ttc_table ttc;
 278 };
 279
 280 struct mlx5e_hairpin_entry {
 281         /* a node of a hash table which keeps all the  hairpin entries */
 282         struct hlist_node hairpin_hlist;
 283
 284         /* protects flows list */
 285         spinlock_t flows_lock;
 286         /* flows sharing the same hairpin */
 287         struct list_head flows;
 288         /* hpe's that were not fully initialized when dead peer update event
 289          * function traversed them.
 290          */
 291         struct list_head dead_peer_wait_list;
 292
 293         u16 peer_vhca_id;
 294         u8 prio;
 295         struct mlx5e_hairpin *hp;
 296         refcount_t refcnt;
 297         struct completion res_ready;
 298 };
 299
 300 struct mod_hdr_key {
 301         int num_actions;
 302         void *actions;
 303 };
 304
 305 struct mlx5e_mod_hdr_entry {
 306         /* a node of a hash table which keeps all the mod_hdr entries */
 307         struct hlist_node mod_hdr_hlist;
 308
 309         /* protects flows list */
 310         spinlock_t flows_lock;
 311         /* flows sharing the same mod_hdr entry */
 312         struct list_head flows;
 313
 314         struct mod_hdr_key key;
 315
 316         struct mlx5_modify_hdr *modify_hdr;
 317
 318         refcount_t refcnt;
 319         struct completion res_ready;
 320         int compl_result;
 321 };
 322
 323 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
 324                               struct mlx5e_tc_flow *flow);
 325
 326 static struct mlx5e_tc_flow *mlx5e_flow_get(struct mlx5e_tc_flow *flow)
 327 {
 328         if (!flow || !refcount_inc_not_zero(&flow->refcnt))
 329                 return ERR_PTR(-EINVAL);
 330         return flow;
 331 }
 332
 333 static void mlx5e_flow_put(struct mlx5e_priv *priv,
 334                            struct mlx5e_tc_flow *flow)
 335 {
 336         if (refcount_dec_and_test(&flow->refcnt)) {
 337                 mlx5e_tc_del_flow(priv, flow);
 338                 kfree_rcu(flow, rcu_head);
 339         }
 340 }
 341
 342 static void __flow_flag_set(struct mlx5e_tc_flow *flow, unsigned long flag)
 343 {
 344         /* Complete all memory stores before setting bit. */
 345         smp_mb__before_atomic();
 346         set_bit(flag, &flow->flags);
 347 }
 348
 349 #define flow_flag_set(flow, flag) __flow_flag_set(flow, MLX5E_TC_FLOW_FLAG_##flag)
 350
 351 static bool __flow_flag_test_and_set(struct mlx5e_tc_flow *flow,
 352                                      unsigned long flag)
 353 {
 354         /* test_and_set_bit() provides all necessary barriers */
 355         return test_and_set_bit(flag, &flow->flags);
 356 }
 357
 358 #define flow_flag_test_and_set(flow, flag)                      \
 359         __flow_flag_test_and_set(flow,                          \
 360                                  MLX5E_TC_FLOW_FLAG_##flag)
 361
 362 static void __flow_flag_clear(struct mlx5e_tc_flow *flow, unsigned long flag)
 363 {
 364         /* Complete all memory stores before clearing bit. */
 365         smp_mb__before_atomic();
 366         clear_bit(flag, &flow->flags);
 367 }
 368
 369 #define flow_flag_clear(flow, flag) __flow_flag_clear(flow, \
 370                                                       MLX5E_TC_FLOW_FLAG_##flag)
 371
 372 static bool __flow_flag_test(struct mlx5e_tc_flow *flow, unsigned long flag)
 373 {
 374         bool ret = test_bit(flag, &flow->flags);
 375
 376         /* Read fields of flow structure only after checking flags. */
 377         smp_mb__after_atomic();
 378         return ret;
 379 }
 380
 381 #define flow_flag_test(flow, flag) __flow_flag_test(flow, \
 382                                                     MLX5E_TC_FLOW_FLAG_##flag)
 383
 384 static bool mlx5e_is_eswitch_flow(struct mlx5e_tc_flow *flow)
 385 {
 386         return flow_flag_test(flow, ESWITCH);
 387 }
 388
 389 static bool mlx5e_is_ft_flow(struct mlx5e_tc_flow *flow)
 390 {
 391         return flow_flag_test(flow, FT);
 392 }
 393
 394 static bool mlx5e_is_offloaded_flow(struct mlx5e_tc_flow *flow)
 395 {
 396         return flow_flag_test(flow, OFFLOADED);
 397 }
 398
 399 static inline u32 hash_mod_hdr_info(struct mod_hdr_key *key)
 400 {
 401         return jhash(key->actions,
 402                      key->num_actions * MLX5_MH_ACT_SZ, 0);
 403 }
 404
 405 static inline int cmp_mod_hdr_info(struct mod_hdr_key *a,
 406                                    struct mod_hdr_key *b)
 407 {
 408         if (a->num_actions != b->num_actions)
 409                 return 1;
 410
 411         return memcmp(a->actions, b->actions, a->num_actions * MLX5_MH_ACT_SZ);
 412 }
 413
 414 static struct mod_hdr_tbl *
 415 get_mod_hdr_table(struct mlx5e_priv *priv, int namespace)
 416 {
 417         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 418
 419         return namespace == MLX5_FLOW_NAMESPACE_FDB ? &esw->offloads.mod_hdr :
 420                 &priv->fs.tc.mod_hdr;
 421 }
 422
 423 static struct mlx5e_mod_hdr_entry *
 424 mlx5e_mod_hdr_get(struct mod_hdr_tbl *tbl, struct mod_hdr_key *key, u32 hash_key)
 425 {
 426         struct mlx5e_mod_hdr_entry *mh, *found = NULL;
 427
 428         hash_for_each_possible(tbl->hlist, mh, mod_hdr_hlist, hash_key) {
 429                 if (!cmp_mod_hdr_info(&mh->key, key)) {
 430                         refcount_inc(&mh->refcnt);
 431                         found = mh;
 432                         break;
 433                 }
 434         }
 435
 436         return found;
 437 }
 438
 439 static void mlx5e_mod_hdr_put(struct mlx5e_priv *priv,
 440                               struct mlx5e_mod_hdr_entry *mh,
 441                               int namespace)
 442 {
 443         struct mod_hdr_tbl *tbl = get_mod_hdr_table(priv, namespace);
 444
 445         if (!refcount_dec_and_mutex_lock(&mh->refcnt, &tbl->lock))
 446                 return;
 447         hash_del(&mh->mod_hdr_hlist);
 448         mutex_unlock(&tbl->lock);
 449
 450         WARN_ON(!list_empty(&mh->flows));
 451         if (mh->compl_result > 0)
 452                 mlx5_modify_header_dealloc(priv->mdev, mh->modify_hdr);
 453
 454         kfree(mh);
 455 }
 456
 457 static int get_flow_name_space(struct mlx5e_tc_flow *flow)
 458 {
 459         return mlx5e_is_eswitch_flow(flow) ?
 460                 MLX5_FLOW_NAMESPACE_FDB : MLX5_FLOW_NAMESPACE_KERNEL;
 461 }
 462 static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
 463                                 struct mlx5e_tc_flow *flow,
 464                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
 465 {
 466         int num_actions, actions_size, namespace, err;
 467         struct mlx5e_mod_hdr_entry *mh;
 468         struct mod_hdr_tbl *tbl;
 469         struct mod_hdr_key key;
 470         u32 hash_key;
 471
 472         num_actions  = parse_attr->mod_hdr_acts.num_actions;
 473         actions_size = MLX5_MH_ACT_SZ * num_actions;
 474
 475         key.actions = parse_attr->mod_hdr_acts.actions;
 476         key.num_actions = num_actions;
 477
 478         hash_key = hash_mod_hdr_info(&key);
 479
 480         namespace = get_flow_name_space(flow);
 481         tbl = get_mod_hdr_table(priv, namespace);
 482
 483         mutex_lock(&tbl->lock);
 484         mh = mlx5e_mod_hdr_get(tbl, &key, hash_key);
 485         if (mh) {
 486                 mutex_unlock(&tbl->lock);
 487                 wait_for_completion(&mh->res_ready);
 488
 489                 if (mh->compl_result < 0) {
 490                         err = -EREMOTEIO;
 491                         goto attach_header_err;
 492                 }
 493                 goto attach_flow;
 494         }
 495
 496         mh = kzalloc(sizeof(*mh) + actions_size, GFP_KERNEL);
 497         if (!mh) {
 498                 mutex_unlock(&tbl->lock);
 499                 return -ENOMEM;
 500         }
 501
 502         mh->key.actions = (void *)mh + sizeof(*mh);
 503         memcpy(mh->key.actions, key.actions, actions_size);
 504         mh->key.num_actions = num_actions;
 505         spin_lock_init(&mh->flows_lock);
 506         INIT_LIST_HEAD(&mh->flows);
 507         refcount_set(&mh->refcnt, 1);
 508         init_completion(&mh->res_ready);
 509
 510         hash_add(tbl->hlist, &mh->mod_hdr_hlist, hash_key);
 511         mutex_unlock(&tbl->lock);
 512
 513         mh->modify_hdr = mlx5_modify_header_alloc(priv->mdev, namespace,
 514                                                   mh->key.num_actions,
 515                                                   mh->key.actions);
 516         if (IS_ERR(mh->modify_hdr)) {
 517                 err = PTR_ERR(mh->modify_hdr);
 518                 mh->compl_result = err;
 519                 goto alloc_header_err;
 520         }
 521         mh->compl_result = 1;
 522         complete_all(&mh->res_ready);
 523
 524 attach_flow:
 525         flow->mh = mh;
 526         spin_lock(&mh->flows_lock);
 527         list_add(&flow->mod_hdr, &mh->flows);
 528         spin_unlock(&mh->flows_lock);
 529         if (mlx5e_is_eswitch_flow(flow))
 530                 flow->esw_attr->modify_hdr = mh->modify_hdr;
 531         else
 532                 flow->nic_attr->modify_hdr = mh->modify_hdr;
 533
 534         return 0;
 535
 536 alloc_header_err:
 537         complete_all(&mh->res_ready);
 538 attach_header_err:
 539         mlx5e_mod_hdr_put(priv, mh, namespace);
 540         return err;
 541 }
 542
 543 static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
 544                                  struct mlx5e_tc_flow *flow)
 545 {
 546         /* flow wasn't fully initialized */
 547         if (!flow->mh)
 548                 return;
 549
 550         spin_lock(&flow->mh->flows_lock);
 551         list_del(&flow->mod_hdr);
 552         spin_unlock(&flow->mh->flows_lock);
 553
 554         mlx5e_mod_hdr_put(priv, flow->mh, get_flow_name_space(flow));
 555         flow->mh = NULL;
 556 }
 557
 558 static
 559 struct mlx5_core_dev *mlx5e_hairpin_get_mdev(struct net *net, int ifindex)
 560 {
 561         struct net_device *netdev;
 562         struct mlx5e_priv *priv;
 563
 564         netdev = __dev_get_by_index(net, ifindex);
 565         priv = netdev_priv(netdev);
 566         return priv->mdev;
 567 }
 568
 569 static int mlx5e_hairpin_create_transport(struct mlx5e_hairpin *hp)
 570 {
 571         u32 in[MLX5_ST_SZ_DW(create_tir_in)] = {0};
 572         void *tirc;
 573         int err;
 574
 575         err = mlx5_core_alloc_transport_domain(hp->func_mdev, &hp->tdn);
 576         if (err)
 577                 goto alloc_tdn_err;
 578
 579         tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
 580
 581         MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
 582         MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn[0]);
 583         MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
 584
 585         err = mlx5_core_create_tir(hp->func_mdev, in, MLX5_ST_SZ_BYTES(create_tir_in), &hp->tirn);
 586         if (err)
 587                 goto create_tir_err;
 588
 589         return 0;
 590
 591 create_tir_err:
 592         mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
 593 alloc_tdn_err:
 594         return err;
 595 }
 596
 597 static void mlx5e_hairpin_destroy_transport(struct mlx5e_hairpin *hp)
 598 {
 599         mlx5_core_destroy_tir(hp->func_mdev, hp->tirn);
 600         mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
 601 }
 602
 603 static void mlx5e_hairpin_fill_rqt_rqns(struct mlx5e_hairpin *hp, void *rqtc)
 604 {
 605         u32 indirection_rqt[MLX5E_INDIR_RQT_SIZE], rqn;
 606         struct mlx5e_priv *priv = hp->func_priv;
 607         int i, ix, sz = MLX5E_INDIR_RQT_SIZE;
 608
 609         mlx5e_build_default_indir_rqt(indirection_rqt, sz,
 610                                       hp->num_channels);
 611
 612         for (i = 0; i < sz; i++) {
 613                 ix = i;
 614                 if (priv->rss_params.hfunc == ETH_RSS_HASH_XOR)
 615                         ix = mlx5e_bits_invert(i, ilog2(sz));
 616                 ix = indirection_rqt[ix];
 617                 rqn = hp->pair->rqn[ix];
 618                 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
 619         }
 620 }
 621
 622 static int mlx5e_hairpin_create_indirect_rqt(struct mlx5e_hairpin *hp)
 623 {
 624         int inlen, err, sz = MLX5E_INDIR_RQT_SIZE;
 625         struct mlx5e_priv *priv = hp->func_priv;
 626         struct mlx5_core_dev *mdev = priv->mdev;
 627         void *rqtc;
 628         u32 *in;
 629
 630         inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
 631         in = kvzalloc(inlen, GFP_KERNEL);
 632         if (!in)
 633                 return -ENOMEM;
 634
 635         rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
 636
 637         MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
 638         MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
 639
 640         mlx5e_hairpin_fill_rqt_rqns(hp, rqtc);
 641
 642         err = mlx5_core_create_rqt(mdev, in, inlen, &hp->indir_rqt.rqtn);
 643         if (!err)
 644                 hp->indir_rqt.enabled = true;
 645
 646         kvfree(in);
 647         return err;
 648 }
 649
 650 static int mlx5e_hairpin_create_indirect_tirs(struct mlx5e_hairpin *hp)
 651 {
 652         struct mlx5e_priv *priv = hp->func_priv;
 653         u32 in[MLX5_ST_SZ_DW(create_tir_in)];
 654         int tt, i, err;
 655         void *tirc;
 656
 657         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
 658                 struct mlx5e_tirc_config ttconfig = mlx5e_tirc_get_default_config(tt);
 659
 660                 memset(in, 0, MLX5_ST_SZ_BYTES(create_tir_in));
 661                 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
 662
 663                 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
 664                 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
 665                 MLX5_SET(tirc, tirc, indirect_table, hp->indir_rqt.rqtn);
 666                 mlx5e_build_indir_tir_ctx_hash(&priv->rss_params, &ttconfig, tirc, false);
 667
 668                 err = mlx5_core_create_tir(hp->func_mdev, in,
 669                                            MLX5_ST_SZ_BYTES(create_tir_in), &hp->indir_tirn[tt]);
 670                 if (err) {
 671                         mlx5_core_warn(hp->func_mdev, "create indirect tirs failed, %d\n", err);
 672                         goto err_destroy_tirs;
 673                 }
 674         }
 675         return 0;
 676
 677 err_destroy_tirs:
 678         for (i = 0; i < tt; i++)
 679                 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[i]);
 680         return err;
 681 }
 682
 683 static void mlx5e_hairpin_destroy_indirect_tirs(struct mlx5e_hairpin *hp)
 684 {
 685         int tt;
 686
 687         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
 688                 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[tt]);
 689 }
 690
 691 static void mlx5e_hairpin_set_ttc_params(struct mlx5e_hairpin *hp,
 692                                          struct ttc_params *ttc_params)
 693 {
 694         struct mlx5_flow_table_attr *ft_attr = &ttc_params->ft_attr;
 695         int tt;
 696
 697         memset(ttc_params, 0, sizeof(*ttc_params));
 698
 699         ttc_params->any_tt_tirn = hp->tirn;
 700
 701         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
 702                 ttc_params->indir_tirn[tt] = hp->indir_tirn[tt];
 703
 704         ft_attr->max_fte = MLX5E_TTC_TABLE_SIZE;
 705         ft_attr->level = MLX5E_TC_TTC_FT_LEVEL;
 706         ft_attr->prio = MLX5E_TC_PRIO;
 707 }
 708
 709 static int mlx5e_hairpin_rss_init(struct mlx5e_hairpin *hp)
 710 {
 711         struct mlx5e_priv *priv = hp->func_priv;
 712         struct ttc_params ttc_params;
 713         int err;
 714
 715         err = mlx5e_hairpin_create_indirect_rqt(hp);
 716         if (err)
 717                 return err;
 718
 719         err = mlx5e_hairpin_create_indirect_tirs(hp);
 720         if (err)
 721                 goto err_create_indirect_tirs;
 722
 723         mlx5e_hairpin_set_ttc_params(hp, &ttc_params);
 724         err = mlx5e_create_ttc_table(priv, &ttc_params, &hp->ttc);
 725         if (err)
 726                 goto err_create_ttc_table;
 727
 728         netdev_dbg(priv->netdev, "add hairpin: using %d channels rss ttc table id %x\n",
 729                    hp->num_channels, hp->ttc.ft.t->id);
 730
 731         return 0;
 732
 733 err_create_ttc_table:
 734         mlx5e_hairpin_destroy_indirect_tirs(hp);
 735 err_create_indirect_tirs:
 736         mlx5e_destroy_rqt(priv, &hp->indir_rqt);
 737
 738         return err;
 739 }
 740
 741 static void mlx5e_hairpin_rss_cleanup(struct mlx5e_hairpin *hp)
 742 {
 743         struct mlx5e_priv *priv = hp->func_priv;
 744
 745         mlx5e_destroy_ttc_table(priv, &hp->ttc);
 746         mlx5e_hairpin_destroy_indirect_tirs(hp);
 747         mlx5e_destroy_rqt(priv, &hp->indir_rqt);
 748 }
 749
 750 static struct mlx5e_hairpin *
 751 mlx5e_hairpin_create(struct mlx5e_priv *priv, struct mlx5_hairpin_params *params,
 752                      int peer_ifindex)
 753 {
 754         struct mlx5_core_dev *func_mdev, *peer_mdev;
 755         struct mlx5e_hairpin *hp;
 756         struct mlx5_hairpin *pair;
 757         int err;
 758
 759         hp = kzalloc(sizeof(*hp), GFP_KERNEL);
 760         if (!hp)
 761                 return ERR_PTR(-ENOMEM);
 762
 763         func_mdev = priv->mdev;
 764         peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
 765
 766         pair = mlx5_core_hairpin_create(func_mdev, peer_mdev, params);
 767         if (IS_ERR(pair)) {
 768                 err = PTR_ERR(pair);
 769                 goto create_pair_err;
 770         }
 771         hp->pair = pair;
 772         hp->func_mdev = func_mdev;
 773         hp->func_priv = priv;
 774         hp->num_channels = params->num_channels;
 775
 776         err = mlx5e_hairpin_create_transport(hp);
 777         if (err)
 778                 goto create_transport_err;
 779
 780         if (hp->num_channels > 1) {
 781                 err = mlx5e_hairpin_rss_init(hp);
 782                 if (err)
 783                         goto rss_init_err;
 784         }
 785
 786         return hp;
 787
 788 rss_init_err:
 789         mlx5e_hairpin_destroy_transport(hp);
 790 create_transport_err:
 791         mlx5_core_hairpin_destroy(hp->pair);
 792 create_pair_err:
 793         kfree(hp);
 794         return ERR_PTR(err);
 795 }
 796
 797 static void mlx5e_hairpin_destroy(struct mlx5e_hairpin *hp)
 798 {
 799         if (hp->num_channels > 1)
 800                 mlx5e_hairpin_rss_cleanup(hp);
 801         mlx5e_hairpin_destroy_transport(hp);
 802         mlx5_core_hairpin_destroy(hp->pair);
 803         kvfree(hp);
 804 }
 805
 806 static inline u32 hash_hairpin_info(u16 peer_vhca_id, u8 prio)
 807 {
 808         return (peer_vhca_id << 16 | prio);
 809 }
 810
 811 static struct mlx5e_hairpin_entry *mlx5e_hairpin_get(struct mlx5e_priv *priv,
 812                                                      u16 peer_vhca_id, u8 prio)
 813 {
 814         struct mlx5e_hairpin_entry *hpe;
 815         u32 hash_key = hash_hairpin_info(peer_vhca_id, prio);
 816
 817         hash_for_each_possible(priv->fs.tc.hairpin_tbl, hpe,
 818                                hairpin_hlist, hash_key) {
 819                 if (hpe->peer_vhca_id == peer_vhca_id && hpe->prio == prio) {
 820                         refcount_inc(&hpe->refcnt);
 821                         return hpe;
 822                 }
 823         }
 824
 825         return NULL;
 826 }
 827
 828 static void mlx5e_hairpin_put(struct mlx5e_priv *priv,
 829                               struct mlx5e_hairpin_entry *hpe)
 830 {
 831         /* no more hairpin flows for us, release the hairpin pair */
 832         if (!refcount_dec_and_mutex_lock(&hpe->refcnt, &priv->fs.tc.hairpin_tbl_lock))
 833                 return;
 834         hash_del(&hpe->hairpin_hlist);
 835         mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
 836
 837         if (!IS_ERR_OR_NULL(hpe->hp)) {
 838                 netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
 839                            dev_name(hpe->hp->pair->peer_mdev->device));
 840
 841                 mlx5e_hairpin_destroy(hpe->hp);
 842         }
 843
 844         WARN_ON(!list_empty(&hpe->flows));
 845         kfree(hpe);
 846 }
 847
 848 #define UNKNOWN_MATCH_PRIO 8
 849
 850 static int mlx5e_hairpin_get_prio(struct mlx5e_priv *priv,
 851                                   struct mlx5_flow_spec *spec, u8 *match_prio,
 852                                   struct netlink_ext_ack *extack)
 853 {
 854         void *headers_c, *headers_v;
 855         u8 prio_val, prio_mask = 0;
 856         bool vlan_present;
 857
 858 #ifdef CONFIG_MLX5_CORE_EN_DCB
 859         if (priv->dcbx_dp.trust_state != MLX5_QPTS_TRUST_PCP) {
 860                 NL_SET_ERR_MSG_MOD(extack,
 861                                    "only PCP trust state supported for hairpin");
 862                 return -EOPNOTSUPP;
 863         }
 864 #endif
 865         headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, outer_headers);
 866         headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
 867
 868         vlan_present = MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag);
 869         if (vlan_present) {
 870                 prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
 871                 prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
 872         }
 873
 874         if (!vlan_present || !prio_mask) {
 875                 prio_val = UNKNOWN_MATCH_PRIO;
 876         } else if (prio_mask != 0x7) {
 877                 NL_SET_ERR_MSG_MOD(extack,
 878                                    "masked priority match not supported for hairpin");
 879                 return -EOPNOTSUPP;
 880         }
 881
 882         *match_prio = prio_val;
 883         return 0;
 884 }
 885
 886 static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
 887                                   struct mlx5e_tc_flow *flow,
 888                                   struct mlx5e_tc_flow_parse_attr *parse_attr,
 889                                   struct netlink_ext_ack *extack)
 890 {
 891         int peer_ifindex = parse_attr->mirred_ifindex[0];
 892         struct mlx5_hairpin_params params;
 893         struct mlx5_core_dev *peer_mdev;
 894         struct mlx5e_hairpin_entry *hpe;
 895         struct mlx5e_hairpin *hp;
 896         u64 link_speed64;
 897         u32 link_speed;
 898         u8 match_prio;
 899         u16 peer_id;
 900         int err;
 901
 902         peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
 903         if (!MLX5_CAP_GEN(priv->mdev, hairpin) || !MLX5_CAP_GEN(peer_mdev, hairpin)) {
 904                 NL_SET_ERR_MSG_MOD(extack, "hairpin is not supported");
 905                 return -EOPNOTSUPP;
 906         }
 907
 908         peer_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
 909         err = mlx5e_hairpin_get_prio(priv, &parse_attr->spec, &match_prio,
 910                                      extack);
 911         if (err)
 912                 return err;
 913
 914         mutex_lock(&priv->fs.tc.hairpin_tbl_lock);
 915         hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
 916         if (hpe) {
 917                 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
 918                 wait_for_completion(&hpe->res_ready);
 919
 920                 if (IS_ERR(hpe->hp)) {
 921                         err = -EREMOTEIO;
 922                         goto out_err;
 923                 }
 924                 goto attach_flow;
 925         }
 926
 927         hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
 928         if (!hpe) {
 929                 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
 930                 return -ENOMEM;
 931         }
 932
 933         spin_lock_init(&hpe->flows_lock);
 934         INIT_LIST_HEAD(&hpe->flows);
 935         INIT_LIST_HEAD(&hpe->dead_peer_wait_list);
 936         hpe->peer_vhca_id = peer_id;
 937         hpe->prio = match_prio;
 938         refcount_set(&hpe->refcnt, 1);
 939         init_completion(&hpe->res_ready);
 940
 941         hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
 942                  hash_hairpin_info(peer_id, match_prio));
 943         mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
 944
 945         params.log_data_size = 15;
 946         params.log_data_size = min_t(u8, params.log_data_size,
 947                                      MLX5_CAP_GEN(priv->mdev, log_max_hairpin_wq_data_sz));
 948         params.log_data_size = max_t(u8, params.log_data_size,
 949                                      MLX5_CAP_GEN(priv->mdev, log_min_hairpin_wq_data_sz));
 950
 951         params.log_num_packets = params.log_data_size -
 952                                  MLX5_MPWRQ_MIN_LOG_STRIDE_SZ(priv->mdev);
 953         params.log_num_packets = min_t(u8, params.log_num_packets,
 954                                        MLX5_CAP_GEN(priv->mdev, log_max_hairpin_num_packets));
 955
 956         params.q_counter = priv->q_counter;
 957         /* set hairpin pair per each 50Gbs share of the link */
 958         mlx5e_port_max_linkspeed(priv->mdev, &link_speed);
 959         link_speed = max_t(u32, link_speed, 50000);
 960         link_speed64 = link_speed;
 961         do_div(link_speed64, 50000);
 962         params.num_channels = link_speed64;
 963
 964         hp = mlx5e_hairpin_create(priv, &params, peer_ifindex);
 965         hpe->hp = hp;
 966         complete_all(&hpe->res_ready);
 967         if (IS_ERR(hp)) {
 968                 err = PTR_ERR(hp);
 969                 goto out_err;
 970         }
 971
 972         netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
 973                    hp->tirn, hp->pair->rqn[0],
 974                    dev_name(hp->pair->peer_mdev->device),
 975                    hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
 976
 977 attach_flow:
 978         if (hpe->hp->num_channels > 1) {
 979                 flow_flag_set(flow, HAIRPIN_RSS);
 980                 flow->nic_attr->hairpin_ft = hpe->hp->ttc.ft.t;
 981         } else {
 982                 flow->nic_attr->hairpin_tirn = hpe->hp->tirn;
 983         }
 984
 985         flow->hpe = hpe;
 986         spin_lock(&hpe->flows_lock);
 987         list_add(&flow->hairpin, &hpe->flows);
 988         spin_unlock(&hpe->flows_lock);
 989
 990         return 0;
 991
 992 out_err:
 993         mlx5e_hairpin_put(priv, hpe);
 994         return err;
 995 }
 996
 997 static void mlx5e_hairpin_flow_del(struct mlx5e_priv *priv,
 998                                    struct mlx5e_tc_flow *flow)
 999 {
1000         /* flow wasn't fully initialized */
1001         if (!flow->hpe)
1002                 return;
1003
1004         spin_lock(&flow->hpe->flows_lock);
1005         list_del(&flow->hairpin);
1006         spin_unlock(&flow->hpe->flows_lock);
1007
1008         mlx5e_hairpin_put(priv, flow->hpe);
1009         flow->hpe = NULL;
1010 }
1011
1012 static int
1013 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
1014                       struct mlx5e_tc_flow_parse_attr *parse_attr,
1015                       struct mlx5e_tc_flow *flow,
1016                       struct netlink_ext_ack *extack)
1017 {
1018         struct mlx5_flow_context *flow_context = &parse_attr->spec.flow_context;
1019         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
1020         struct mlx5_core_dev *dev = priv->mdev;
1021         struct mlx5_flow_destination dest[2] = {};
1022         struct mlx5_flow_act flow_act = {
1023                 .action = attr->action,
1024                 .flags    = FLOW_ACT_NO_APPEND,
1025         };
1026         struct mlx5_fc *counter = NULL;
1027         int err, dest_ix = 0;
1028
1029         flow_context->flags |= FLOW_CONTEXT_HAS_TAG;
1030         flow_context->flow_tag = attr->flow_tag;
1031
1032         if (flow_flag_test(flow, HAIRPIN)) {
1033                 err = mlx5e_hairpin_flow_add(priv, flow, parse_attr, extack);
1034                 if (err)
1035                         return err;
1036
1037                 if (flow_flag_test(flow, HAIRPIN_RSS)) {
1038                         dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1039                         dest[dest_ix].ft = attr->hairpin_ft;
1040                 } else {
1041                         dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1042                         dest[dest_ix].tir_num = attr->hairpin_tirn;
1043                 }
1044                 dest_ix++;
1045         } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
1046                 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1047                 dest[dest_ix].ft = priv->fs.vlan.ft.t;
1048                 dest_ix++;
1049         }
1050
1051         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1052                 counter = mlx5_fc_create(dev, true);
1053                 if (IS_ERR(counter))
1054                         return PTR_ERR(counter);
1055
1056                 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1057                 dest[dest_ix].counter_id = mlx5_fc_id(counter);
1058                 dest_ix++;
1059                 attr->counter = counter;
1060         }
1061
1062         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
1063                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
1064                 flow_act.modify_hdr = attr->modify_hdr;
1065                 dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
1066                 if (err)
1067                         return err;
1068         }
1069
1070         mutex_lock(&priv->fs.tc.t_lock);
1071         if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
1072                 struct mlx5_flow_table_attr ft_attr = {};
1073                 int tc_grp_size, tc_tbl_size, tc_num_grps;
1074                 u32 max_flow_counter;
1075
1076                 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
1077                                     MLX5_CAP_GEN(dev, max_flow_counter_15_0);
1078
1079                 tc_grp_size = min_t(int, max_flow_counter, MLX5E_TC_TABLE_MAX_GROUP_SIZE);
1080
1081                 tc_tbl_size = min_t(int, tc_grp_size * MLX5E_TC_TABLE_NUM_GROUPS,
1082                                     BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev, log_max_ft_size)));
1083                 tc_num_grps = MLX5E_TC_TABLE_NUM_GROUPS;
1084
1085                 ft_attr.prio = MLX5E_TC_PRIO;
1086                 ft_attr.max_fte = tc_tbl_size;
1087                 ft_attr.level = MLX5E_TC_FT_LEVEL;
1088                 ft_attr.autogroup.max_num_groups = tc_num_grps;
1089                 priv->fs.tc.t =
1090                         mlx5_create_auto_grouped_flow_table(priv->fs.ns,
1091                                                             &ft_attr);
1092                 if (IS_ERR(priv->fs.tc.t)) {
1093                         mutex_unlock(&priv->fs.tc.t_lock);
1094                         NL_SET_ERR_MSG_MOD(extack,
1095                                            "Failed to create tc offload table\n");
1096                         netdev_err(priv->netdev,
1097                                    "Failed to create tc offload table\n");
1098                         return PTR_ERR(priv->fs.tc.t);
1099                 }
1100         }
1101
1102         if (attr->match_level != MLX5_MATCH_NONE)
1103                 parse_attr->spec.match_criteria_enable |= MLX5_MATCH_OUTER_HEADERS;
1104
1105         flow->rule[0] = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
1106                                             &flow_act, dest, dest_ix);
1107         mutex_unlock(&priv->fs.tc.t_lock);
1108
1109         return PTR_ERR_OR_ZERO(flow->rule[0]);
1110 }
1111
1112 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
1113                                   struct mlx5e_tc_flow *flow)
1114 {
1115         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
1116         struct mlx5_fc *counter = NULL;
1117
1118         counter = attr->counter;
1119         if (!IS_ERR_OR_NULL(flow->rule[0]))
1120                 mlx5_del_flow_rules(flow->rule[0]);
1121         mlx5_fc_destroy(priv->mdev, counter);
1122
1123         mutex_lock(&priv->fs.tc.t_lock);
1124         if (!mlx5e_tc_num_filters(priv, MLX5_TC_FLAG(NIC_OFFLOAD)) && priv->fs.tc.t) {
1125                 mlx5_destroy_flow_table(priv->fs.tc.t);
1126                 priv->fs.tc.t = NULL;
1127         }
1128         mutex_unlock(&priv->fs.tc.t_lock);
1129
1130         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1131                 mlx5e_detach_mod_hdr(priv, flow);
1132
1133         if (flow_flag_test(flow, HAIRPIN))
1134                 mlx5e_hairpin_flow_del(priv, flow);
1135 }
1136
1137 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1138                                struct mlx5e_tc_flow *flow, int out_index);
1139
1140 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
1141                               struct mlx5e_tc_flow *flow,
1142                               struct net_device *mirred_dev,
1143                               int out_index,
1144                               struct netlink_ext_ack *extack,
1145                               struct net_device **encap_dev,
1146                               bool *encap_valid);
1147
1148 static struct mlx5_flow_handle *
1149 mlx5e_tc_offload_fdb_rules(struct mlx5_eswitch *esw,
1150                            struct mlx5e_tc_flow *flow,
1151                            struct mlx5_flow_spec *spec,
1152                            struct mlx5_esw_flow_attr *attr)
1153 {
1154         struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts;
1155         struct mlx5_flow_handle *rule;
1156
1157         if (flow_flag_test(flow, CT)) {
1158                 mod_hdr_acts = &attr->parse_attr->mod_hdr_acts;
1159
1160                 return mlx5_tc_ct_flow_offload(flow->priv, flow, spec, attr,
1161                                                mod_hdr_acts);
1162         }
1163
1164         rule = mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
1165         if (IS_ERR(rule))
1166                 return rule;
1167
1168         if (attr->split_count) {
1169                 flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, spec, attr);
1170                 if (IS_ERR(flow->rule[1])) {
1171                         mlx5_eswitch_del_offloaded_rule(esw, rule, attr);
1172                         return flow->rule[1];
1173                 }
1174         }
1175
1176         return rule;
1177 }
1178
1179 static void
1180 mlx5e_tc_unoffload_fdb_rules(struct mlx5_eswitch *esw,
1181                              struct mlx5e_tc_flow *flow,
1182                              struct mlx5_esw_flow_attr *attr)
1183 {
1184         flow_flag_clear(flow, OFFLOADED);
1185
1186         if (flow_flag_test(flow, CT)) {
1187                 mlx5_tc_ct_delete_flow(flow->priv, flow, attr);
1188                 return;
1189         }
1190
1191         if (attr->split_count)
1192                 mlx5_eswitch_del_fwd_rule(esw, flow->rule[1], attr);
1193
1194         mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
1195 }
1196
1197 static struct mlx5_flow_handle *
1198 mlx5e_tc_offload_to_slow_path(struct mlx5_eswitch *esw,
1199                               struct mlx5e_tc_flow *flow,
1200                               struct mlx5_flow_spec *spec)
1201 {
1202         struct mlx5_esw_flow_attr slow_attr;
1203         struct mlx5_flow_handle *rule;
1204
1205         memcpy(&slow_attr, flow->esw_attr, sizeof(slow_attr));
1206         slow_attr.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1207         slow_attr.split_count = 0;
1208         slow_attr.flags |= MLX5_ESW_ATTR_FLAG_SLOW_PATH;
1209
1210         rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, &slow_attr);
1211         if (!IS_ERR(rule))
1212                 flow_flag_set(flow, SLOW);
1213
1214         return rule;
1215 }
1216
1217 static void
1218 mlx5e_tc_unoffload_from_slow_path(struct mlx5_eswitch *esw,
1219                                   struct mlx5e_tc_flow *flow)
1220 {
1221         struct mlx5_esw_flow_attr slow_attr;
1222
1223         memcpy(&slow_attr, flow->esw_attr, sizeof(slow_attr));
1224         slow_attr.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1225         slow_attr.split_count = 0;
1226         slow_attr.flags |= MLX5_ESW_ATTR_FLAG_SLOW_PATH;
1227         mlx5e_tc_unoffload_fdb_rules(esw, flow, &slow_attr);
1228         flow_flag_clear(flow, SLOW);
1229 }
1230
1231 /* Caller must obtain uplink_priv->unready_flows_lock mutex before calling this
1232  * function.
1233  */
1234 static void unready_flow_add(struct mlx5e_tc_flow *flow,
1235                              struct list_head *unready_flows)
1236 {
1237         flow_flag_set(flow, NOT_READY);
1238         list_add_tail(&flow->unready, unready_flows);
1239 }
1240
1241 /* Caller must obtain uplink_priv->unready_flows_lock mutex before calling this
1242  * function.
1243  */
1244 static void unready_flow_del(struct mlx5e_tc_flow *flow)
1245 {
1246         list_del(&flow->unready);
1247         flow_flag_clear(flow, NOT_READY);
1248 }
1249
1250 static void add_unready_flow(struct mlx5e_tc_flow *flow)
1251 {
1252         struct mlx5_rep_uplink_priv *uplink_priv;
1253         struct mlx5e_rep_priv *rpriv;
1254         struct mlx5_eswitch *esw;
1255
1256         esw = flow->priv->mdev->priv.eswitch;
1257         rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1258         uplink_priv = &rpriv->uplink_priv;
1259
1260         mutex_lock(&uplink_priv->unready_flows_lock);
1261         unready_flow_add(flow, &uplink_priv->unready_flows);
1262         mutex_unlock(&uplink_priv->unready_flows_lock);
1263 }
1264
1265 static void remove_unready_flow(struct mlx5e_tc_flow *flow)
1266 {
1267         struct mlx5_rep_uplink_priv *uplink_priv;
1268         struct mlx5e_rep_priv *rpriv;
1269         struct mlx5_eswitch *esw;
1270
1271         esw = flow->priv->mdev->priv.eswitch;
1272         rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1273         uplink_priv = &rpriv->uplink_priv;
1274
1275         mutex_lock(&uplink_priv->unready_flows_lock);
1276         unready_flow_del(flow);
1277         mutex_unlock(&uplink_priv->unready_flows_lock);
1278 }
1279
1280 static int
1281 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
1282                       struct mlx5e_tc_flow *flow,
1283                       struct netlink_ext_ack *extack)
1284 {
1285         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1286         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1287         struct mlx5e_tc_flow_parse_attr *parse_attr = attr->parse_attr;
1288         struct net_device *out_dev, *encap_dev = NULL;
1289         struct mlx5_fc *counter = NULL;
1290         struct mlx5e_rep_priv *rpriv;
1291         struct mlx5e_priv *out_priv;
1292         bool encap_valid = true;
1293         u32 max_prio, max_chain;
1294         int err = 0;
1295         int out_index;
1296
1297         if (!mlx5_esw_chains_prios_supported(esw) && attr->prio != 1) {
1298                 NL_SET_ERR_MSG_MOD(extack,
1299                                    "E-switch priorities unsupported, upgrade FW");
1300                 return -EOPNOTSUPP;
1301         }
1302
1303         /* We check chain range only for tc flows.
1304          * For ft flows, we checked attr->chain was originally 0 and set it to
1305          * FDB_FT_CHAIN which is outside tc range.
1306          * See mlx5e_rep_setup_ft_cb().
1307          */
1308         max_chain = mlx5_esw_chains_get_chain_range(esw);
1309         if (!mlx5e_is_ft_flow(flow) && attr->chain > max_chain) {
1310                 NL_SET_ERR_MSG_MOD(extack,
1311                                    "Requested chain is out of supported range");
1312                 return -EOPNOTSUPP;
1313         }
1314
1315         max_prio = mlx5_esw_chains_get_prio_range(esw);
1316         if (attr->prio > max_prio) {
1317                 NL_SET_ERR_MSG_MOD(extack,
1318                                    "Requested priority is out of supported range");
1319                 return -EOPNOTSUPP;
1320         }
1321
1322         for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++) {
1323                 int mirred_ifindex;
1324
1325                 if (!(attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP))
1326                         continue;
1327
1328                 mirred_ifindex = parse_attr->mirred_ifindex[out_index];
1329                 out_dev = __dev_get_by_index(dev_net(priv->netdev),
1330                                              mirred_ifindex);
1331                 err = mlx5e_attach_encap(priv, flow, out_dev, out_index,
1332                                          extack, &encap_dev, &encap_valid);
1333                 if (err)
1334                         return err;
1335
1336                 out_priv = netdev_priv(encap_dev);
1337                 rpriv = out_priv->ppriv;
1338                 attr->dests[out_index].rep = rpriv->rep;
1339                 attr->dests[out_index].mdev = out_priv->mdev;
1340         }
1341
1342         err = mlx5_eswitch_add_vlan_action(esw, attr);
1343         if (err)
1344                 return err;
1345
1346         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
1347                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
1348                 dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
1349                 if (err)
1350                         return err;
1351         }
1352
1353         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1354                 counter = mlx5_fc_create(attr->counter_dev, true);
1355                 if (IS_ERR(counter))
1356                         return PTR_ERR(counter);
1357
1358                 attr->counter = counter;
1359         }
1360
1361         /* we get here if one of the following takes place:
1362          * (1) there's no error
1363          * (2) there's an encap action and we don't have valid neigh
1364          */
1365         if (!encap_valid)
1366                 flow->rule[0] = mlx5e_tc_offload_to_slow_path(esw, flow, &parse_attr->spec);
1367         else
1368                 flow->rule[0] = mlx5e_tc_offload_fdb_rules(esw, flow, &parse_attr->spec, attr);
1369
1370         if (IS_ERR(flow->rule[0]))
1371                 return PTR_ERR(flow->rule[0]);
1372         else
1373                 flow_flag_set(flow, OFFLOADED);
1374
1375         return 0;
1376 }
1377
1378 static bool mlx5_flow_has_geneve_opt(struct mlx5e_tc_flow *flow)
1379 {
1380         struct mlx5_flow_spec *spec = &flow->esw_attr->parse_attr->spec;
1381         void *headers_v = MLX5_ADDR_OF(fte_match_param,
1382                                        spec->match_value,
1383                                        misc_parameters_3);
1384         u32 geneve_tlv_opt_0_data = MLX5_GET(fte_match_set_misc3,
1385                                              headers_v,
1386                                              geneve_tlv_option_0_data);
1387
1388         return !!geneve_tlv_opt_0_data;
1389 }
1390
1391 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
1392                                   struct mlx5e_tc_flow *flow)
1393 {
1394         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1395         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1396         int out_index;
1397
1398         mlx5e_put_flow_tunnel_id(flow);
1399
1400         if (flow_flag_test(flow, NOT_READY)) {
1401                 remove_unready_flow(flow);
1402                 kvfree(attr->parse_attr);
1403                 return;
1404         }
1405
1406         if (mlx5e_is_offloaded_flow(flow)) {
1407                 if (flow_flag_test(flow, SLOW))
1408                         mlx5e_tc_unoffload_from_slow_path(esw, flow);
1409                 else
1410                         mlx5e_tc_unoffload_fdb_rules(esw, flow, attr);
1411         }
1412
1413         if (mlx5_flow_has_geneve_opt(flow))
1414                 mlx5_geneve_tlv_option_del(priv->mdev->geneve);
1415
1416         mlx5_eswitch_del_vlan_action(esw, attr);
1417
1418         for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++)
1419                 if (attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP) {
1420                         mlx5e_detach_encap(priv, flow, out_index);
1421                         kfree(attr->parse_attr->tun_info[out_index]);
1422                 }
1423         kvfree(attr->parse_attr);
1424
1425         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1426                 mlx5e_detach_mod_hdr(priv, flow);
1427
1428         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
1429                 mlx5_fc_destroy(attr->counter_dev, attr->counter);
1430 }
1431
1432 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
1433                               struct mlx5e_encap_entry *e,
1434                               struct list_head *flow_list)
1435 {
1436         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1437         struct mlx5_esw_flow_attr *esw_attr;
1438         struct mlx5_flow_handle *rule;
1439         struct mlx5_flow_spec *spec;
1440         struct mlx5e_tc_flow *flow;
1441         int err;
1442
1443         e->pkt_reformat = mlx5_packet_reformat_alloc(priv->mdev,
1444                                                      e->reformat_type,
1445                                                      e->encap_size, e->encap_header,
1446                                                      MLX5_FLOW_NAMESPACE_FDB);
1447         if (IS_ERR(e->pkt_reformat)) {
1448                 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %lu\n",
1449                                PTR_ERR(e->pkt_reformat));
1450                 return;
1451         }
1452         e->flags |= MLX5_ENCAP_ENTRY_VALID;
1453         mlx5e_rep_queue_neigh_stats_work(priv);
1454
1455         list_for_each_entry(flow, flow_list, tmp_list) {
1456                 bool all_flow_encaps_valid = true;
1457                 int i;
1458
1459                 if (!mlx5e_is_offloaded_flow(flow))
1460                         continue;
1461                 esw_attr = flow->esw_attr;
1462                 spec = &esw_attr->parse_attr->spec;
1463
1464                 esw_attr->dests[flow->tmp_efi_index].pkt_reformat = e->pkt_reformat;
1465                 esw_attr->dests[flow->tmp_efi_index].flags |= MLX5_ESW_DEST_ENCAP_VALID;
1466                 /* Flow can be associated with multiple encap entries.
1467                  * Before offloading the flow verify that all of them have
1468                  * a valid neighbour.
1469                  */
1470                 for (i = 0; i < MLX5_MAX_FLOW_FWD_VPORTS; i++) {
1471                         if (!(esw_attr->dests[i].flags & MLX5_ESW_DEST_ENCAP))
1472                                 continue;
1473                         if (!(esw_attr->dests[i].flags & MLX5_ESW_DEST_ENCAP_VALID)) {
1474                                 all_flow_encaps_valid = false;
1475                                 break;
1476                         }
1477                 }
1478                 /* Do not offload flows with unresolved neighbors */
1479                 if (!all_flow_encaps_valid)
1480                         continue;
1481                 /* update from slow path rule to encap rule */
1482                 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, esw_attr);
1483                 if (IS_ERR(rule)) {
1484                         err = PTR_ERR(rule);
1485                         mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
1486                                        err);
1487                         continue;
1488                 }
1489
1490                 mlx5e_tc_unoffload_from_slow_path(esw, flow);
1491                 flow->rule[0] = rule;
1492                 /* was unset when slow path rule removed */
1493                 flow_flag_set(flow, OFFLOADED);
1494         }
1495 }
1496
1497 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
1498                               struct mlx5e_encap_entry *e,
1499                               struct list_head *flow_list)
1500 {
1501         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1502         struct mlx5_flow_handle *rule;
1503         struct mlx5_flow_spec *spec;
1504         struct mlx5e_tc_flow *flow;
1505         int err;
1506
1507         list_for_each_entry(flow, flow_list, tmp_list) {
1508                 if (!mlx5e_is_offloaded_flow(flow))
1509                         continue;
1510                 spec = &flow->esw_attr->parse_attr->spec;
1511
1512                 /* update from encap rule to slow path rule */
1513                 rule = mlx5e_tc_offload_to_slow_path(esw, flow, spec);
1514                 /* mark the flow's encap dest as non-valid */
1515                 flow->esw_attr->dests[flow->tmp_efi_index].flags &= ~MLX5_ESW_DEST_ENCAP_VALID;
1516
1517                 if (IS_ERR(rule)) {
1518                         err = PTR_ERR(rule);
1519                         mlx5_core_warn(priv->mdev, "Failed to update slow path (encap) flow, %d\n",
1520                                        err);
1521                         continue;
1522                 }
1523
1524                 mlx5e_tc_unoffload_fdb_rules(esw, flow, flow->esw_attr);
1525                 flow->rule[0] = rule;
1526                 /* was unset when fast path rule removed */
1527                 flow_flag_set(flow, OFFLOADED);
1528         }
1529
1530         /* we know that the encap is valid */
1531         e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
1532         mlx5_packet_reformat_dealloc(priv->mdev, e->pkt_reformat);
1533 }
1534
1535 static struct mlx5_fc *mlx5e_tc_get_counter(struct mlx5e_tc_flow *flow)
1536 {
1537         if (mlx5e_is_eswitch_flow(flow))
1538                 return flow->esw_attr->counter;
1539         else
1540                 return flow->nic_attr->counter;
1541 }
1542
1543 /* Takes reference to all flows attached to encap and adds the flows to
1544  * flow_list using 'tmp_list' list_head in mlx5e_tc_flow.
1545  */
1546 void mlx5e_take_all_encap_flows(struct mlx5e_encap_entry *e, struct list_head *flow_list)
1547 {
1548         struct encap_flow_item *efi;
1549         struct mlx5e_tc_flow *flow;
1550
1551         list_for_each_entry(efi, &e->flows, list) {
1552                 flow = container_of(efi, struct mlx5e_tc_flow, encaps[efi->index]);
1553                 if (IS_ERR(mlx5e_flow_get(flow)))
1554                         continue;
1555                 wait_for_completion(&flow->init_done);
1556
1557                 flow->tmp_efi_index = efi->index;
1558                 list_add(&flow->tmp_list, flow_list);
1559         }
1560 }
1561
1562 /* Iterate over tmp_list of flows attached to flow_list head. */
1563 void mlx5e_put_encap_flow_list(struct mlx5e_priv *priv, struct list_head *flow_list)
1564 {
1565         struct mlx5e_tc_flow *flow, *tmp;
1566
1567         list_for_each_entry_safe(flow, tmp, flow_list, tmp_list)
1568                 mlx5e_flow_put(priv, flow);
1569 }
1570
1571 static struct mlx5e_encap_entry *
1572 mlx5e_get_next_valid_encap(struct mlx5e_neigh_hash_entry *nhe,
1573                            struct mlx5e_encap_entry *e)
1574 {
1575         struct mlx5e_encap_entry *next = NULL;
1576
1577 retry:
1578         rcu_read_lock();
1579
1580         /* find encap with non-zero reference counter value */
1581         for (next = e ?
1582                      list_next_or_null_rcu(&nhe->encap_list,
1583                                            &e->encap_list,
1584                                            struct mlx5e_encap_entry,
1585                                            encap_list) :
1586                      list_first_or_null_rcu(&nhe->encap_list,
1587                                             struct mlx5e_encap_entry,
1588                                             encap_list);
1589              next;
1590              next = list_next_or_null_rcu(&nhe->encap_list,
1591                                           &next->encap_list,
1592                                           struct mlx5e_encap_entry,
1593                                           encap_list))
1594                 if (mlx5e_encap_take(next))
1595                         break;
1596
1597         rcu_read_unlock();
1598
1599         /* release starting encap */
1600         if (e)
1601                 mlx5e_encap_put(netdev_priv(e->out_dev), e);
1602         if (!next)
1603                 return next;
1604
1605         /* wait for encap to be fully initialized */
1606         wait_for_completion(&next->res_ready);
1607         /* continue searching if encap entry is not in valid state after completion */
1608         if (!(next->flags & MLX5_ENCAP_ENTRY_VALID)) {
1609                 e = next;
1610                 goto retry;
1611         }
1612
1613         return next;
1614 }
1615
1616 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
1617 {
1618         struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
1619         struct mlx5e_encap_entry *e = NULL;
1620         struct mlx5e_tc_flow *flow;
1621         struct mlx5_fc *counter;
1622         struct neigh_table *tbl;
1623         bool neigh_used = false;
1624         struct neighbour *n;
1625         u64 lastuse;
1626
1627         if (m_neigh->family == AF_INET)
1628                 tbl = &arp_tbl;
1629 #if IS_ENABLED(CONFIG_IPV6)
1630         else if (m_neigh->family == AF_INET6)
1631                 tbl = ipv6_stub->nd_tbl;
1632 #endif
1633         else
1634                 return;
1635
1636         /* mlx5e_get_next_valid_encap() releases previous encap before returning
1637          * next one.
1638          */
1639         while ((e = mlx5e_get_next_valid_encap(nhe, e)) != NULL) {
1640                 struct mlx5e_priv *priv = netdev_priv(e->out_dev);
1641                 struct encap_flow_item *efi, *tmp;
1642                 struct mlx5_eswitch *esw;
1643                 LIST_HEAD(flow_list);
1644
1645                 esw = priv->mdev->priv.eswitch;
1646                 mutex_lock(&esw->offloads.encap_tbl_lock);
1647                 list_for_each_entry_safe(efi, tmp, &e->flows, list) {
1648                         flow = container_of(efi, struct mlx5e_tc_flow,
1649                                             encaps[efi->index]);
1650                         if (IS_ERR(mlx5e_flow_get(flow)))
1651                                 continue;
1652                         list_add(&flow->tmp_list, &flow_list);
1653
1654                         if (mlx5e_is_offloaded_flow(flow)) {
1655                                 counter = mlx5e_tc_get_counter(flow);
1656                                 lastuse = mlx5_fc_query_lastuse(counter);
1657                                 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
1658                                         neigh_used = true;
1659                                         break;
1660                                 }
1661                         }
1662                 }
1663                 mutex_unlock(&esw->offloads.encap_tbl_lock);
1664
1665                 mlx5e_put_encap_flow_list(priv, &flow_list);
1666                 if (neigh_used) {
1667                         /* release current encap before breaking the loop */
1668                         mlx5e_encap_put(priv, e);
1669                         break;
1670                 }
1671         }
1672
1673         trace_mlx5e_tc_update_neigh_used_value(nhe, neigh_used);
1674
1675         if (neigh_used) {
1676                 nhe->reported_lastuse = jiffies;
1677
1678                 /* find the relevant neigh according to the cached device and
1679                  * dst ip pair
1680                  */
1681                 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
1682                 if (!n)
1683                         return;
1684
1685                 neigh_event_send(n, NULL);
1686                 neigh_release(n);
1687         }
1688 }
1689
1690 static void mlx5e_encap_dealloc(struct mlx5e_priv *priv, struct mlx5e_encap_entry *e)
1691 {
1692         WARN_ON(!list_empty(&e->flows));
1693
1694         if (e->compl_result > 0) {
1695                 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1696
1697                 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1698                         mlx5_packet_reformat_dealloc(priv->mdev, e->pkt_reformat);
1699         }
1700
1701         kfree(e->tun_info);
1702         kfree(e->encap_header);
1703         kfree_rcu(e, rcu);
1704 }
1705
1706 void mlx5e_encap_put(struct mlx5e_priv *priv, struct mlx5e_encap_entry *e)
1707 {
1708         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1709
1710         if (!refcount_dec_and_mutex_lock(&e->refcnt, &esw->offloads.encap_tbl_lock))
1711                 return;
1712         hash_del_rcu(&e->encap_hlist);
1713         mutex_unlock(&esw->offloads.encap_tbl_lock);
1714
1715         mlx5e_encap_dealloc(priv, e);
1716 }
1717
1718 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1719                                struct mlx5e_tc_flow *flow, int out_index)
1720 {
1721         struct mlx5e_encap_entry *e = flow->encaps[out_index].e;
1722         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1723
1724         /* flow wasn't fully initialized */
1725         if (!e)
1726                 return;
1727
1728         mutex_lock(&esw->offloads.encap_tbl_lock);
1729         list_del(&flow->encaps[out_index].list);
1730         flow->encaps[out_index].e = NULL;
1731         if (!refcount_dec_and_test(&e->refcnt)) {
1732                 mutex_unlock(&esw->offloads.encap_tbl_lock);
1733                 return;
1734         }
1735         hash_del_rcu(&e->encap_hlist);
1736         mutex_unlock(&esw->offloads.encap_tbl_lock);
1737
1738         mlx5e_encap_dealloc(priv, e);
1739 }
1740
1741 static void __mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1742 {
1743         struct mlx5_eswitch *esw = flow->priv->mdev->priv.eswitch;
1744
1745         if (!flow_flag_test(flow, ESWITCH) ||
1746             !flow_flag_test(flow, DUP))
1747                 return;
1748
1749         mutex_lock(&esw->offloads.peer_mutex);
1750         list_del(&flow->peer);
1751         mutex_unlock(&esw->offloads.peer_mutex);
1752
1753         flow_flag_clear(flow, DUP);
1754
1755         if (refcount_dec_and_test(&flow->peer_flow->refcnt)) {
1756                 mlx5e_tc_del_fdb_flow(flow->peer_flow->priv, flow->peer_flow);
1757                 kfree(flow->peer_flow);
1758         }
1759
1760         flow->peer_flow = NULL;
1761 }
1762
1763 static void mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1764 {
1765         struct mlx5_core_dev *dev = flow->priv->mdev;
1766         struct mlx5_devcom *devcom = dev->priv.devcom;
1767         struct mlx5_eswitch *peer_esw;
1768
1769         peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1770         if (!peer_esw)
1771                 return;
1772
1773         __mlx5e_tc_del_fdb_peer_flow(flow);
1774         mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1775 }
1776
1777 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1778                               struct mlx5e_tc_flow *flow)
1779 {
1780         if (mlx5e_is_eswitch_flow(flow)) {
1781                 mlx5e_tc_del_fdb_peer_flow(flow);
1782                 mlx5e_tc_del_fdb_flow(priv, flow);
1783         } else {
1784                 mlx5e_tc_del_nic_flow(priv, flow);
1785         }
1786 }
1787
1788 static int flow_has_tc_fwd_action(struct flow_cls_offload *f)
1789 {
1790         struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1791         struct flow_action *flow_action = &rule->action;
1792         const struct flow_action_entry *act;
1793         int i;
1794
1795         flow_action_for_each(i, act, flow_action) {
1796                 switch (act->id) {
1797                 case FLOW_ACTION_GOTO:
1798                         return true;
1799                 default:
1800                         continue;
1801                 }
1802         }
1803
1804         return false;
1805 }
1806
1807 static int
1808 enc_opts_is_dont_care_or_full_match(struct mlx5e_priv *priv,
1809                                     struct flow_dissector_key_enc_opts *opts,
1810                                     struct netlink_ext_ack *extack,
1811                                     bool *dont_care)
1812 {
1813         struct geneve_opt *opt;
1814         int off = 0;
1815
1816         *dont_care = true;
1817
1818         while (opts->len > off) {
1819                 opt = (struct geneve_opt *)&opts->data[off];
1820
1821                 if (!(*dont_care) || opt->opt_class || opt->type ||
1822                     memchr_inv(opt->opt_data, 0, opt->length * 4)) {
1823                         *dont_care = false;
1824
1825                         if (opt->opt_class != U16_MAX ||
1826                             opt->type != U8_MAX ||
1827                             memchr_inv(opt->opt_data, 0xFF,
1828                                        opt->length * 4)) {
1829                                 NL_SET_ERR_MSG(extack,
1830                                                "Partial match of tunnel options in chain > 0 isn't supported");
1831                                 netdev_warn(priv->netdev,
1832                                             "Partial match of tunnel options in chain > 0 isn't supported");
1833                                 return -EOPNOTSUPP;
1834                         }
1835                 }
1836
1837                 off += sizeof(struct geneve_opt) + opt->length * 4;
1838         }
1839
1840         return 0;
1841 }
1842
1843 #define COPY_DISSECTOR(rule, diss_key, dst)\
1844 ({ \
1845         struct flow_rule *__rule = (rule);\
1846         typeof(dst) __dst = dst;\
1847 \
1848         memcpy(__dst,\
1849                skb_flow_dissector_target(__rule->match.dissector,\
1850                                          diss_key,\
1851                                          __rule->match.key),\
1852                sizeof(*__dst));\
1853 })
1854
1855 static int mlx5e_get_flow_tunnel_id(struct mlx5e_priv *priv,
1856                                     struct mlx5e_tc_flow *flow,
1857                                     struct flow_cls_offload *f,
1858                                     struct net_device *filter_dev)
1859 {
1860         struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1861         struct netlink_ext_ack *extack = f->common.extack;
1862         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1863         struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts;
1864         struct flow_match_enc_opts enc_opts_match;
1865         struct mlx5_rep_uplink_priv *uplink_priv;
1866         struct mlx5e_rep_priv *uplink_rpriv;
1867         struct tunnel_match_key tunnel_key;
1868         bool enc_opts_is_dont_care = true;
1869         u32 tun_id, enc_opts_id = 0;
1870         struct mlx5_eswitch *esw;
1871         u32 value, mask;
1872         int err;
1873
1874         esw = priv->mdev->priv.eswitch;
1875         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1876         uplink_priv = &uplink_rpriv->uplink_priv;
1877
1878         memset(&tunnel_key, 0, sizeof(tunnel_key));
1879         COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL,
1880                        &tunnel_key.enc_control);
1881         if (tunnel_key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS)
1882                 COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1883                                &tunnel_key.enc_ipv4);
1884         else
1885                 COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1886                                &tunnel_key.enc_ipv6);
1887         COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_IP, &tunnel_key.enc_ip);
1888         COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_PORTS,
1889                        &tunnel_key.enc_tp);
1890         COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_KEYID,
1891                        &tunnel_key.enc_key_id);
1892         tunnel_key.filter_ifindex = filter_dev->ifindex;
1893
1894         err = mapping_add(uplink_priv->tunnel_mapping, &tunnel_key, &tun_id);
1895         if (err)
1896                 return err;
1897
1898         flow_rule_match_enc_opts(rule, &enc_opts_match);
1899         err = enc_opts_is_dont_care_or_full_match(priv,
1900                                                   enc_opts_match.mask,
1901                                                   extack,
1902                                                   &enc_opts_is_dont_care);
1903         if (err)
1904                 goto err_enc_opts;
1905
1906         if (!enc_opts_is_dont_care) {
1907                 err = mapping_add(uplink_priv->tunnel_enc_opts_mapping,
1908                                   enc_opts_match.key, &enc_opts_id);
1909                 if (err)
1910                         goto err_enc_opts;
1911         }
1912
1913         value = tun_id << ENC_OPTS_BITS | enc_opts_id;
1914         mask = enc_opts_id ? TUNNEL_ID_MASK :
1915                              (TUNNEL_ID_MASK & ~ENC_OPTS_BITS_MASK);
1916
1917         if (attr->chain) {
1918                 mlx5e_tc_match_to_reg_match(&attr->parse_attr->spec,
1919                                             TUNNEL_TO_REG, value, mask);
1920         } else {
1921                 mod_hdr_acts = &attr->parse_attr->mod_hdr_acts;
1922                 err = mlx5e_tc_match_to_reg_set(priv->mdev,
1923                                                 mod_hdr_acts,
1924                                                 TUNNEL_TO_REG, value);
1925                 if (err)
1926                         goto err_set;
1927
1928                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
1929         }
1930
1931         flow->tunnel_id = value;
1932         return 0;
1933
1934 err_set:
1935         if (enc_opts_id)
1936                 mapping_remove(uplink_priv->tunnel_enc_opts_mapping,
1937                                enc_opts_id);
1938 err_enc_opts:
1939         mapping_remove(uplink_priv->tunnel_mapping, tun_id);
1940         return err;
1941 }
1942
1943 static void mlx5e_put_flow_tunnel_id(struct mlx5e_tc_flow *flow)
1944 {
1945         u32 enc_opts_id = flow->tunnel_id & ENC_OPTS_BITS_MASK;
1946         u32 tun_id = flow->tunnel_id >> ENC_OPTS_BITS;
1947         struct mlx5_rep_uplink_priv *uplink_priv;
1948         struct mlx5e_rep_priv *uplink_rpriv;
1949         struct mlx5_eswitch *esw;
1950
1951         esw = flow->priv->mdev->priv.eswitch;
1952         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1953         uplink_priv = &uplink_rpriv->uplink_priv;
1954
1955         if (tun_id)
1956                 mapping_remove(uplink_priv->tunnel_mapping, tun_id);
1957         if (enc_opts_id)
1958                 mapping_remove(uplink_priv->tunnel_enc_opts_mapping,
1959                                enc_opts_id);
1960 }
1961
1962 u32 mlx5e_tc_get_flow_tun_id(struct mlx5e_tc_flow *flow)
1963 {
1964         return flow->tunnel_id;
1965 }
1966
1967 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1968                              struct mlx5e_tc_flow *flow,
1969                              struct mlx5_flow_spec *spec,
1970                              struct flow_cls_offload *f,
1971                              struct net_device *filter_dev,
1972                              u8 *match_level,
1973                              bool *match_inner)
1974 {
1975         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1976         struct netlink_ext_ack *extack = f->common.extack;
1977         bool needs_mapping, sets_mapping;
1978         int err;
1979
1980         if (!mlx5e_is_eswitch_flow(flow))
1981                 return -EOPNOTSUPP;
1982
1983         needs_mapping = !!flow->esw_attr->chain;
1984         sets_mapping = !flow->esw_attr->chain && flow_has_tc_fwd_action(f);
1985         *match_inner = !needs_mapping;
1986
1987         if ((needs_mapping || sets_mapping) &&
1988             !mlx5_eswitch_reg_c1_loopback_enabled(esw)) {
1989                 NL_SET_ERR_MSG(extack,
1990                                "Chains on tunnel devices isn't supported without register loopback support");
1991                 netdev_warn(priv->netdev,
1992                             "Chains on tunnel devices isn't supported without register loopback support");
1993                 return -EOPNOTSUPP;
1994         }
1995
1996         if (!flow->esw_attr->chain) {
1997                 err = mlx5e_tc_tun_parse(filter_dev, priv, spec, f,
1998                                          match_level);
1999                 if (err) {
2000                         NL_SET_ERR_MSG_MOD(extack,
2001                                            "Failed to parse tunnel attributes");
2002                         netdev_warn(priv->netdev,
2003                                     "Failed to parse tunnel attributes");
2004                         return err;
2005                 }
2006
2007                 flow->esw_attr->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
2008         }
2009
2010         if (!needs_mapping && !sets_mapping)
2011                 return 0;
2012
2013         return mlx5e_get_flow_tunnel_id(priv, flow, f, filter_dev);
2014 }
2015
2016 static void *get_match_inner_headers_criteria(struct mlx5_flow_spec *spec)
2017 {
2018         return MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
2019                             inner_headers);
2020 }
2021
2022 static void *get_match_inner_headers_value(struct mlx5_flow_spec *spec)
2023 {
2024         return MLX5_ADDR_OF(fte_match_param, spec->match_value,
2025                             inner_headers);
2026 }
2027
2028 static void *get_match_outer_headers_criteria(struct mlx5_flow_spec *spec)
2029 {
2030         return MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
2031                             outer_headers);
2032 }
2033
2034 static void *get_match_outer_headers_value(struct mlx5_flow_spec *spec)
2035 {
2036         return MLX5_ADDR_OF(fte_match_param, spec->match_value,
2037                             outer_headers);
2038 }
2039
2040 static void *get_match_headers_value(u32 flags,
2041                                      struct mlx5_flow_spec *spec)
2042 {
2043         return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
2044                 get_match_inner_headers_value(spec) :
2045                 get_match_outer_headers_value(spec);
2046 }
2047
2048 static void *get_match_headers_criteria(u32 flags,
2049                                         struct mlx5_flow_spec *spec)
2050 {
2051         return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
2052                 get_match_inner_headers_criteria(spec) :
2053                 get_match_outer_headers_criteria(spec);
2054 }
2055
2056 static int mlx5e_flower_parse_meta(struct net_device *filter_dev,
2057                                    struct flow_cls_offload *f)
2058 {
2059         struct flow_rule *rule = flow_cls_offload_flow_rule(f);
2060         struct netlink_ext_ack *extack = f->common.extack;
2061         struct net_device *ingress_dev;
2062         struct flow_match_meta match;
2063
2064         if (!flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_META))
2065                 return 0;
2066
2067         flow_rule_match_meta(rule, &match);
2068         if (match.mask->ingress_ifindex != 0xFFFFFFFF) {
2069                 NL_SET_ERR_MSG_MOD(extack, "Unsupported ingress ifindex mask");
2070                 return -EINVAL;
2071         }
2072
2073         ingress_dev = __dev_get_by_index(dev_net(filter_dev),
2074                                          match.key->ingress_ifindex);
2075         if (!ingress_dev) {
2076                 NL_SET_ERR_MSG_MOD(extack,
2077                                    "Can't find the ingress port to match on");
2078                 return -EINVAL;
2079         }
2080
2081         if (ingress_dev != filter_dev) {
2082                 NL_SET_ERR_MSG_MOD(extack,
2083                                    "Can't match on the ingress filter port");
2084                 return -EINVAL;
2085         }
2086
2087         return 0;
2088 }
2089
2090 static int __parse_cls_flower(struct mlx5e_priv *priv,
2091                               struct mlx5e_tc_flow *flow,
2092                               struct mlx5_flow_spec *spec,
2093                               struct flow_cls_offload *f,
2094                               struct net_device *filter_dev,
2095                               u8 *inner_match_level, u8 *outer_match_level)
2096 {
2097         struct netlink_ext_ack *extack = f->common.extack;
2098         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
2099                                        outer_headers);
2100         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
2101                                        outer_headers);
2102         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
2103                                     misc_parameters);
2104         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
2105                                     misc_parameters);
2106         struct flow_rule *rule = flow_cls_offload_flow_rule(f);
2107         struct flow_dissector *dissector = rule->match.dissector;
2108         u16 addr_type = 0;
2109         u8 ip_proto = 0;
2110         u8 *match_level;
2111         int err;
2112
2113         match_level = outer_match_level;
2114
2115         if (dissector->used_keys &
2116             ~(BIT(FLOW_DISSECTOR_KEY_META) |
2117               BIT(FLOW_DISSECTOR_KEY_CONTROL) |
2118               BIT(FLOW_DISSECTOR_KEY_BASIC) |
2119               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
2120               BIT(FLOW_DISSECTOR_KEY_VLAN) |
2121               BIT(FLOW_DISSECTOR_KEY_CVLAN) |
2122               BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
2123               BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
2124               BIT(FLOW_DISSECTOR_KEY_PORTS) |
2125               BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
2126               BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
2127               BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
2128               BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
2129               BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
2130               BIT(FLOW_DISSECTOR_KEY_TCP) |
2131               BIT(FLOW_DISSECTOR_KEY_IP)  |
2132               BIT(FLOW_DISSECTOR_KEY_CT) |
2133               BIT(FLOW_DISSECTOR_KEY_ENC_IP) |
2134               BIT(FLOW_DISSECTOR_KEY_ENC_OPTS))) {
2135                 NL_SET_ERR_MSG_MOD(extack, "Unsupported key");
2136                 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
2137                             dissector->used_keys);
2138                 return -EOPNOTSUPP;
2139         }
2140
2141         if (mlx5e_get_tc_tun(filter_dev)) {
2142                 bool match_inner = false;
2143
2144                 err = parse_tunnel_attr(priv, flow, spec, f, filter_dev,
2145                                         outer_match_level, &match_inner);
2146                 if (err)
2147                         return err;
2148
2149                 if (match_inner) {
2150                         /* header pointers should point to the inner headers
2151                          * if the packet was decapsulated already.
2152                          * outer headers are set by parse_tunnel_attr.
2153                          */
2154                         match_level = inner_match_level;
2155                         headers_c = get_match_inner_headers_criteria(spec);
2156                         headers_v = get_match_inner_headers_value(spec);
2157                 }
2158         }
2159
2160         err = mlx5e_flower_parse_meta(filter_dev, f);
2161         if (err)
2162                 return err;
2163
2164         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2165                 struct flow_match_basic match;
2166
2167                 flow_rule_match_basic(rule, &match);
2168                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
2169                          ntohs(match.mask->n_proto));
2170                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
2171                          ntohs(match.key->n_proto));
2172
2173                 if (match.mask->n_proto)
2174                         *match_level = MLX5_MATCH_L2;
2175         }
2176         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN) ||
2177             is_vlan_dev(filter_dev)) {
2178                 struct flow_dissector_key_vlan filter_dev_mask;
2179                 struct flow_dissector_key_vlan filter_dev_key;
2180                 struct flow_match_vlan match;
2181
2182                 if (is_vlan_dev(filter_dev)) {
2183                         match.key = &filter_dev_key;
2184                         match.key->vlan_id = vlan_dev_vlan_id(filter_dev);
2185                         match.key->vlan_tpid = vlan_dev_vlan_proto(filter_dev);
2186                         match.key->vlan_priority = 0;
2187                         match.mask = &filter_dev_mask;
2188                         memset(match.mask, 0xff, sizeof(*match.mask));
2189                         match.mask->vlan_priority = 0;
2190                 } else {
2191                         flow_rule_match_vlan(rule, &match);
2192                 }
2193                 if (match.mask->vlan_id ||
2194                     match.mask->vlan_priority ||
2195                     match.mask->vlan_tpid) {
2196                         if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
2197                                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2198                                          svlan_tag, 1);
2199                                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2200                                          svlan_tag, 1);
2201                         } else {
2202                                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2203                                          cvlan_tag, 1);
2204                                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2205                                          cvlan_tag, 1);
2206                         }
2207
2208                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid,
2209                                  match.mask->vlan_id);
2210                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid,
2211                                  match.key->vlan_id);
2212
2213                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio,
2214                                  match.mask->vlan_priority);
2215                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio,
2216                                  match.key->vlan_priority);
2217
2218                         *match_level = MLX5_MATCH_L2;
2219                 }
2220         } else if (*match_level != MLX5_MATCH_NONE) {
2221                 /* cvlan_tag enabled in match criteria and
2222                  * disabled in match value means both S & C tags
2223                  * don't exist (untagged of both)
2224                  */
2225                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
2226                 *match_level = MLX5_MATCH_L2;
2227         }
2228
2229         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
2230                 struct flow_match_vlan match;
2231
2232                 flow_rule_match_cvlan(rule, &match);
2233                 if (match.mask->vlan_id ||
2234                     match.mask->vlan_priority ||
2235                     match.mask->vlan_tpid) {
2236                         if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
2237                                 MLX5_SET(fte_match_set_misc, misc_c,
2238                                          outer_second_svlan_tag, 1);
2239                                 MLX5_SET(fte_match_set_misc, misc_v,
2240                                          outer_second_svlan_tag, 1);
2241                         } else {
2242                                 MLX5_SET(fte_match_set_misc, misc_c,
2243                                          outer_second_cvlan_tag, 1);
2244                                 MLX5_SET(fte_match_set_misc, misc_v,
2245                                          outer_second_cvlan_tag, 1);
2246                         }
2247
2248                         MLX5_SET(fte_match_set_misc, misc_c, outer_second_vid,
2249                                  match.mask->vlan_id);
2250                         MLX5_SET(fte_match_set_misc, misc_v, outer_second_vid,
2251                                  match.key->vlan_id);
2252                         MLX5_SET(fte_match_set_misc, misc_c, outer_second_prio,
2253                                  match.mask->vlan_priority);
2254                         MLX5_SET(fte_match_set_misc, misc_v, outer_second_prio,
2255                                  match.key->vlan_priority);
2256
2257                         *match_level = MLX5_MATCH_L2;
2258                 }
2259         }
2260
2261         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2262                 struct flow_match_eth_addrs match;
2263
2264                 flow_rule_match_eth_addrs(rule, &match);
2265                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2266                                              dmac_47_16),
2267                                 match.mask->dst);
2268                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2269                                              dmac_47_16),
2270                                 match.key->dst);
2271
2272                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2273                                              smac_47_16),
2274                                 match.mask->src);
2275                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2276                                              smac_47_16),
2277                                 match.key->src);
2278
2279                 if (!is_zero_ether_addr(match.mask->src) ||
2280                     !is_zero_ether_addr(match.mask->dst))
2281                         *match_level = MLX5_MATCH_L2;
2282         }
2283
2284         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
2285                 struct flow_match_control match;
2286
2287                 flow_rule_match_control(rule, &match);
2288                 addr_type = match.key->addr_type;
2289
2290                 /* the HW doesn't support frag first/later */
2291                 if (match.mask->flags & FLOW_DIS_FIRST_FRAG)
2292                         return -EOPNOTSUPP;
2293
2294                 if (match.mask->flags & FLOW_DIS_IS_FRAGMENT) {
2295                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
2296                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
2297                                  match.key->flags & FLOW_DIS_IS_FRAGMENT);
2298
2299                         /* the HW doesn't need L3 inline to match on frag=no */
2300                         if (!(match.key->flags & FLOW_DIS_IS_FRAGMENT))
2301                                 *match_level = MLX5_MATCH_L2;
2302         /* ***  L2 attributes parsing up to here *** */
2303                         else
2304                                 *match_level = MLX5_MATCH_L3;
2305                 }
2306         }
2307
2308         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2309                 struct flow_match_basic match;
2310
2311                 flow_rule_match_basic(rule, &match);
2312                 ip_proto = match.key->ip_proto;
2313
2314                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
2315                          match.mask->ip_proto);
2316                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
2317                          match.key->ip_proto);
2318
2319                 if (match.mask->ip_proto)
2320                         *match_level = MLX5_MATCH_L3;
2321         }
2322
2323         if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2324                 struct flow_match_ipv4_addrs match;
2325
2326                 flow_rule_match_ipv4_addrs(rule, &match);
2327                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2328                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
2329                        &match.mask->src, sizeof(match.mask->src));
2330                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2331                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
2332                        &match.key->src, sizeof(match.key->src));
2333                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2334                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2335                        &match.mask->dst, sizeof(match.mask->dst));
2336                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2337                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2338                        &match.key->dst, sizeof(match.key->dst));
2339
2340                 if (match.mask->src || match.mask->dst)
2341                         *match_level = MLX5_MATCH_L3;
2342         }
2343
2344         if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2345                 struct flow_match_ipv6_addrs match;
2346
2347                 flow_rule_match_ipv6_addrs(rule, &match);
2348                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2349                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
2350                        &match.mask->src, sizeof(match.mask->src));
2351                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2352                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
2353                        &match.key->src, sizeof(match.key->src));
2354
2355                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2356                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2357                        &match.mask->dst, sizeof(match.mask->dst));
2358                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2359                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2360                        &match.key->dst, sizeof(match.key->dst));
2361
2362                 if (ipv6_addr_type(&match.mask->src) != IPV6_ADDR_ANY ||
2363                     ipv6_addr_type(&match.mask->dst) != IPV6_ADDR_ANY)
2364                         *match_level = MLX5_MATCH_L3;
2365         }
2366
2367         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
2368                 struct flow_match_ip match;
2369
2370                 flow_rule_match_ip(rule, &match);
2371                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn,
2372                          match.mask->tos & 0x3);
2373                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn,
2374                          match.key->tos & 0x3);
2375
2376                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp,
2377                          match.mask->tos >> 2);
2378                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp,
2379                          match.key->tos  >> 2);
2380
2381                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit,
2382                          match.mask->ttl);
2383                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit,
2384                          match.key->ttl);
2385
2386                 if (match.mask->ttl &&
2387                     !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
2388                                                 ft_field_support.outer_ipv4_ttl)) {
2389                         NL_SET_ERR_MSG_MOD(extack,
2390                                            "Matching on TTL is not supported");
2391                         return -EOPNOTSUPP;
2392                 }
2393
2394                 if (match.mask->tos || match.mask->ttl)
2395                         *match_level = MLX5_MATCH_L3;
2396         }
2397
2398         /* ***  L3 attributes parsing up to here *** */
2399
2400         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) {
2401                 struct flow_match_ports match;
2402
2403                 flow_rule_match_ports(rule, &match);
2404                 switch (ip_proto) {
2405                 case IPPROTO_TCP:
2406                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2407                                  tcp_sport, ntohs(match.mask->src));
2408                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2409                                  tcp_sport, ntohs(match.key->src));
2410
2411                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2412                                  tcp_dport, ntohs(match.mask->dst));
2413                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2414                                  tcp_dport, ntohs(match.key->dst));
2415                         break;
2416
2417                 case IPPROTO_UDP:
2418                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2419                                  udp_sport, ntohs(match.mask->src));
2420                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2421                                  udp_sport, ntohs(match.key->src));
2422
2423                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2424                                  udp_dport, ntohs(match.mask->dst));
2425                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2426                                  udp_dport, ntohs(match.key->dst));
2427                         break;
2428                 default:
2429                         NL_SET_ERR_MSG_MOD(extack,
2430                                            "Only UDP and TCP transports are supported for L4 matching");
2431                         netdev_err(priv->netdev,
2432                                    "Only UDP and TCP transport are supported\n");
2433                         return -EINVAL;
2434                 }
2435
2436                 if (match.mask->src || match.mask->dst)
2437                         *match_level = MLX5_MATCH_L4;
2438         }
2439
2440         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_TCP)) {
2441                 struct flow_match_tcp match;
2442
2443                 flow_rule_match_tcp(rule, &match);
2444                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
2445                          ntohs(match.mask->flags));
2446                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
2447                          ntohs(match.key->flags));
2448
2449                 if (match.mask->flags)
2450                         *match_level = MLX5_MATCH_L4;
2451         }
2452
2453         return 0;
2454 }
2455
2456 static int parse_cls_flower(struct mlx5e_priv *priv,
2457                             struct mlx5e_tc_flow *flow,
2458                             struct mlx5_flow_spec *spec,
2459                             struct flow_cls_offload *f,
2460                             struct net_device *filter_dev)
2461 {
2462         u8 inner_match_level, outer_match_level, non_tunnel_match_level;
2463         struct netlink_ext_ack *extack = f->common.extack;
2464         struct mlx5_core_dev *dev = priv->mdev;
2465         struct mlx5_eswitch *esw = dev->priv.eswitch;
2466         struct mlx5e_rep_priv *rpriv = priv->ppriv;
2467         struct mlx5_eswitch_rep *rep;
2468         bool is_eswitch_flow;
2469         int err;
2470
2471         inner_match_level = MLX5_MATCH_NONE;
2472         outer_match_level = MLX5_MATCH_NONE;
2473
2474         err = __parse_cls_flower(priv, flow, spec, f, filter_dev,
2475                                  &inner_match_level, &outer_match_level);
2476         non_tunnel_match_level = (inner_match_level == MLX5_MATCH_NONE) ?
2477                                  outer_match_level : inner_match_level;
2478
2479         is_eswitch_flow = mlx5e_is_eswitch_flow(flow);
2480         if (!err && is_eswitch_flow) {
2481                 rep = rpriv->rep;
2482                 if (rep->vport != MLX5_VPORT_UPLINK &&
2483                     (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
2484                     esw->offloads.inline_mode < non_tunnel_match_level)) {
2485                         NL_SET_ERR_MSG_MOD(extack,
2486                                            "Flow is not offloaded due to min inline setting");
2487                         netdev_warn(priv->netdev,
2488                                     "Flow is not offloaded due to min inline setting, required %d actual %d\n",
2489                                     non_tunnel_match_level, esw->offloads.inline_mode);
2490                         return -EOPNOTSUPP;
2491                 }
2492         }
2493
2494         if (is_eswitch_flow) {
2495                 flow->esw_attr->inner_match_level = inner_match_level;
2496                 flow->esw_attr->outer_match_level = outer_match_level;
2497         } else {
2498                 flow->nic_attr->match_level = non_tunnel_match_level;
2499         }
2500
2501         return err;
2502 }
2503
2504 struct pedit_headers {
2505         struct ethhdr  eth;
2506         struct vlan_hdr vlan;
2507         struct iphdr   ip4;
2508         struct ipv6hdr ip6;
2509         struct tcphdr  tcp;
2510         struct udphdr  udp;
2511 };
2512
2513 struct pedit_headers_action {
2514         struct pedit_headers    vals;
2515         struct pedit_headers    masks;
2516         u32                     pedits;
2517 };
2518
2519 static int pedit_header_offsets[] = {
2520         [FLOW_ACT_MANGLE_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
2521         [FLOW_ACT_MANGLE_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
2522         [FLOW_ACT_MANGLE_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
2523         [FLOW_ACT_MANGLE_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
2524         [FLOW_ACT_MANGLE_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
2525 };
2526
2527 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
2528
2529 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
2530                          struct pedit_headers_action *hdrs)
2531 {
2532         u32 *curr_pmask, *curr_pval;
2533
2534         curr_pmask = (u32 *)(pedit_header(&hdrs->masks, hdr_type) + offset);
2535         curr_pval  = (u32 *)(pedit_header(&hdrs->vals, hdr_type) + offset);
2536
2537         if (*curr_pmask & mask)  /* disallow acting twice on the same location */
2538                 goto out_err;
2539
2540         *curr_pmask |= mask;
2541         *curr_pval  |= (val & mask);
2542
2543         return 0;
2544
2545 out_err:
2546         return -EOPNOTSUPP;
2547 }
2548
2549 struct mlx5_fields {
2550         u8  field;
2551         u8  field_bsize;
2552         u32 field_mask;
2553         u32 offset;
2554         u32 match_offset;
2555 };
2556
2557 #define OFFLOAD(fw_field, field_bsize, field_mask, field, off, match_field) \
2558                 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, field_bsize, field_mask, \
2559                  offsetof(struct pedit_headers, field) + (off), \
2560                  MLX5_BYTE_OFF(fte_match_set_lyr_2_4, match_field)}
2561
2562 /* masked values are the same and there are no rewrites that do not have a
2563  * match.
2564  */
2565 #define SAME_VAL_MASK(type, valp, maskp, matchvalp, matchmaskp) ({ \
2566         type matchmaskx = *(type *)(matchmaskp); \
2567         type matchvalx = *(type *)(matchvalp); \
2568         type maskx = *(type *)(maskp); \
2569         type valx = *(type *)(valp); \
2570         \
2571         (valx & maskx) == (matchvalx & matchmaskx) && !(maskx & (maskx ^ \
2572                                                                  matchmaskx)); \
2573 })
2574
2575 static bool cmp_val_mask(void *valp, void *maskp, void *matchvalp,
2576                          void *matchmaskp, u8 bsize)
2577 {
2578         bool same = false;
2579
2580         switch (bsize) {
2581         case 8:
2582                 same = SAME_VAL_MASK(u8, valp, maskp, matchvalp, matchmaskp);
2583                 break;
2584         case 16:
2585                 same = SAME_VAL_MASK(u16, valp, maskp, matchvalp, matchmaskp);
2586                 break;
2587         case 32:
2588                 same = SAME_VAL_MASK(u32, valp, maskp, matchvalp, matchmaskp);
2589                 break;
2590         }
2591
2592         return same;
2593 }
2594
2595 static struct mlx5_fields fields[] = {
2596         OFFLOAD(DMAC_47_16, 32, U32_MAX, eth.h_dest[0], 0, dmac_47_16),
2597         OFFLOAD(DMAC_15_0,  16, U16_MAX, eth.h_dest[4], 0, dmac_15_0),
2598         OFFLOAD(SMAC_47_16, 32, U32_MAX, eth.h_source[0], 0, smac_47_16),
2599         OFFLOAD(SMAC_15_0,  16, U16_MAX, eth.h_source[4], 0, smac_15_0),
2600         OFFLOAD(ETHERTYPE,  16, U16_MAX, eth.h_proto, 0, ethertype),
2601         OFFLOAD(FIRST_VID,  16, U16_MAX, vlan.h_vlan_TCI, 0, first_vid),
2602
2603         OFFLOAD(IP_DSCP, 8,    0xfc, ip4.tos,   0, ip_dscp),
2604         OFFLOAD(IP_TTL,  8,  U8_MAX, ip4.ttl,   0, ttl_hoplimit),
2605         OFFLOAD(SIPV4,  32, U32_MAX, ip4.saddr, 0, src_ipv4_src_ipv6.ipv4_layout.ipv4),
2606         OFFLOAD(DIPV4,  32, U32_MAX, ip4.daddr, 0, dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2607
2608         OFFLOAD(SIPV6_127_96, 32, U32_MAX, ip6.saddr.s6_addr32[0], 0,
2609                 src_ipv4_src_ipv6.ipv6_layout.ipv6[0]),
2610         OFFLOAD(SIPV6_95_64,  32, U32_MAX, ip6.saddr.s6_addr32[1], 0,
2611                 src_ipv4_src_ipv6.ipv6_layout.ipv6[4]),
2612         OFFLOAD(SIPV6_63_32,  32, U32_MAX, ip6.saddr.s6_addr32[2], 0,
2613                 src_ipv4_src_ipv6.ipv6_layout.ipv6[8]),
2614         OFFLOAD(SIPV6_31_0,   32, U32_MAX, ip6.saddr.s6_addr32[3], 0,
2615                 src_ipv4_src_ipv6.ipv6_layout.ipv6[12]),
2616         OFFLOAD(DIPV6_127_96, 32, U32_MAX, ip6.daddr.s6_addr32[0], 0,
2617                 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[0]),
2618         OFFLOAD(DIPV6_95_64,  32, U32_MAX, ip6.daddr.s6_addr32[1], 0,
2619                 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[4]),
2620         OFFLOAD(DIPV6_63_32,  32, U32_MAX, ip6.daddr.s6_addr32[2], 0,
2621                 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[8]),
2622         OFFLOAD(DIPV6_31_0,   32, U32_MAX, ip6.daddr.s6_addr32[3], 0,
2623                 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[12]),
2624         OFFLOAD(IPV6_HOPLIMIT, 8,  U8_MAX, ip6.hop_limit, 0, ttl_hoplimit),
2625
2626         OFFLOAD(TCP_SPORT, 16, U16_MAX, tcp.source,  0, tcp_sport),
2627         OFFLOAD(TCP_DPORT, 16, U16_MAX, tcp.dest,    0, tcp_dport),
2628         /* in linux iphdr tcp_flags is 8 bits long */
2629         OFFLOAD(TCP_FLAGS,  8,  U8_MAX, tcp.ack_seq, 5, tcp_flags),
2630
2631         OFFLOAD(UDP_SPORT, 16, U16_MAX, udp.source, 0, udp_sport),
2632         OFFLOAD(UDP_DPORT, 16, U16_MAX, udp.dest,   0, udp_dport),
2633 };
2634
2635 static int offload_pedit_fields(struct mlx5e_priv *priv,
2636                                 int namespace,
2637                                 struct pedit_headers_action *hdrs,
2638                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
2639                                 u32 *action_flags,
2640                                 struct netlink_ext_ack *extack)
2641 {
2642         struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
2643         int i, action_size, first, last, next_z;
2644         void *headers_c, *headers_v, *action, *vals_p;
2645         u32 *s_masks_p, *a_masks_p, s_mask, a_mask;
2646         struct mlx5e_tc_mod_hdr_acts *mod_acts;
2647         struct mlx5_fields *f;
2648         unsigned long mask;
2649         __be32 mask_be32;
2650         __be16 mask_be16;
2651         int err;
2652         u8 cmd;
2653
2654         mod_acts = &parse_attr->mod_hdr_acts;
2655         headers_c = get_match_headers_criteria(*action_flags, &parse_attr->spec);
2656         headers_v = get_match_headers_value(*action_flags, &parse_attr->spec);
2657
2658         set_masks = &hdrs[0].masks;
2659         add_masks = &hdrs[1].masks;
2660         set_vals = &hdrs[0].vals;
2661         add_vals = &hdrs[1].vals;
2662
2663         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2664
2665         for (i = 0; i < ARRAY_SIZE(fields); i++) {
2666                 bool skip;
2667
2668                 f = &fields[i];
2669                 /* avoid seeing bits set from previous iterations */
2670                 s_mask = 0;
2671                 a_mask = 0;
2672
2673                 s_masks_p = (void *)set_masks + f->offset;
2674                 a_masks_p = (void *)add_masks + f->offset;
2675
2676                 s_mask = *s_masks_p & f->field_mask;
2677                 a_mask = *a_masks_p & f->field_mask;
2678
2679                 if (!s_mask && !a_mask) /* nothing to offload here */
2680                         continue;
2681
2682                 if (s_mask && a_mask) {
2683                         NL_SET_ERR_MSG_MOD(extack,
2684                                            "can't set and add to the same HW field");
2685                         printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
2686                         return -EOPNOTSUPP;
2687                 }
2688
2689                 skip = false;
2690                 if (s_mask) {
2691                         void *match_mask = headers_c + f->match_offset;
2692                         void *match_val = headers_v + f->match_offset;
2693
2694                         cmd  = MLX5_ACTION_TYPE_SET;
2695                         mask = s_mask;
2696                         vals_p = (void *)set_vals + f->offset;
2697                         /* don't rewrite if we have a match on the same value */
2698                         if (cmp_val_mask(vals_p, s_masks_p, match_val,
2699                                          match_mask, f->field_bsize))
2700                                 skip = true;
2701                         /* clear to denote we consumed this field */
2702                         *s_masks_p &= ~f->field_mask;
2703                 } else {
2704                         cmd  = MLX5_ACTION_TYPE_ADD;
2705                         mask = a_mask;
2706                         vals_p = (void *)add_vals + f->offset;
2707                         /* add 0 is no change */
2708                         if ((*(u32 *)vals_p & f->field_mask) == 0)
2709                                 skip = true;
2710                         /* clear to denote we consumed this field */
2711                         *a_masks_p &= ~f->field_mask;
2712                 }
2713                 if (skip)
2714                         continue;
2715
2716                 if (f->field_bsize == 32) {
2717                         mask_be32 = (__be32)mask;
2718                         mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
2719                 } else if (f->field_bsize == 16) {
2720                         mask_be32 = (__be32)mask;
2721                         mask_be16 = *(__be16 *)&mask_be32;
2722                         mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
2723                 }
2724
2725                 first = find_first_bit(&mask, f->field_bsize);
2726                 next_z = find_next_zero_bit(&mask, f->field_bsize, first);
2727                 last  = find_last_bit(&mask, f->field_bsize);
2728                 if (first < next_z && next_z < last) {
2729                         NL_SET_ERR_MSG_MOD(extack,
2730                                            "rewrite of few sub-fields isn't supported");
2731                         printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
2732                                mask);
2733                         return -EOPNOTSUPP;
2734                 }
2735
2736                 err = alloc_mod_hdr_actions(priv->mdev, namespace, mod_acts);
2737                 if (err) {
2738                         NL_SET_ERR_MSG_MOD(extack,
2739                                            "too many pedit actions, can't offload");
2740                         mlx5_core_warn(priv->mdev,
2741                                        "mlx5: parsed %d pedit actions, can't do more\n",
2742                                        mod_acts->num_actions);
2743                         return err;
2744                 }
2745
2746                 action = mod_acts->actions +
2747                          (mod_acts->num_actions * action_size);
2748                 MLX5_SET(set_action_in, action, action_type, cmd);
2749                 MLX5_SET(set_action_in, action, field, f->field);
2750
2751                 if (cmd == MLX5_ACTION_TYPE_SET) {
2752                         int start;
2753
2754                         /* if field is bit sized it can start not from first bit */
2755                         start = find_first_bit((unsigned long *)&f->field_mask,
2756                                                f->field_bsize);
2757
2758                         MLX5_SET(set_action_in, action, offset, first - start);
2759                         /* length is num of bits to be written, zero means length of 32 */
2760                         MLX5_SET(set_action_in, action, length, (last - first + 1));
2761                 }
2762
2763                 if (f->field_bsize == 32)
2764                         MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
2765                 else if (f->field_bsize == 16)
2766                         MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
2767                 else if (f->field_bsize == 8)
2768                         MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
2769
2770                 ++mod_acts->num_actions;
2771         }
2772
2773         return 0;
2774 }
2775
2776 static int mlx5e_flow_namespace_max_modify_action(struct mlx5_core_dev *mdev,
2777                                                   int namespace)
2778 {
2779         if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
2780                 return MLX5_CAP_ESW_FLOWTABLE_FDB(mdev, max_modify_header_actions);
2781         else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
2782                 return MLX5_CAP_FLOWTABLE_NIC_RX(mdev, max_modify_header_actions);
2783 }
2784
2785 int alloc_mod_hdr_actions(struct mlx5_core_dev *mdev,
2786                           int namespace,
2787                           struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts)
2788 {
2789         int action_size, new_num_actions, max_hw_actions;
2790         size_t new_sz, old_sz;
2791         void *ret;
2792
2793         if (mod_hdr_acts->num_actions < mod_hdr_acts->max_actions)
2794                 return 0;
2795
2796         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2797
2798         max_hw_actions = mlx5e_flow_namespace_max_modify_action(mdev,
2799                                                                 namespace);
2800         new_num_actions = min(max_hw_actions,
2801                               mod_hdr_acts->actions ?
2802                               mod_hdr_acts->max_actions * 2 : 1);
2803         if (mod_hdr_acts->max_actions == new_num_actions)
2804                 return -ENOSPC;
2805
2806         new_sz = action_size * new_num_actions;
2807         old_sz = mod_hdr_acts->max_actions * action_size;
2808         ret = krealloc(mod_hdr_acts->actions, new_sz, GFP_KERNEL);
2809         if (!ret)
2810                 return -ENOMEM;
2811
2812         memset(ret + old_sz, 0, new_sz - old_sz);
2813         mod_hdr_acts->actions = ret;
2814         mod_hdr_acts->max_actions = new_num_actions;
2815
2816         return 0;
2817 }
2818
2819 void dealloc_mod_hdr_actions(struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts)
2820 {
2821         kfree(mod_hdr_acts->actions);
2822         mod_hdr_acts->actions = NULL;
2823         mod_hdr_acts->num_actions = 0;
2824         mod_hdr_acts->max_actions = 0;
2825 }
2826
2827 static const struct pedit_headers zero_masks = {};
2828
2829 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
2830                                  const struct flow_action_entry *act, int namespace,
2831                                  struct pedit_headers_action *hdrs,
2832                                  struct netlink_ext_ack *extack)
2833 {
2834         u8 cmd = (act->id == FLOW_ACTION_MANGLE) ? 0 : 1;
2835         int err = -EOPNOTSUPP;
2836         u32 mask, val, offset;
2837         u8 htype;
2838
2839         htype = act->mangle.htype;
2840         err = -EOPNOTSUPP; /* can't be all optimistic */
2841
2842         if (htype == FLOW_ACT_MANGLE_UNSPEC) {
2843                 NL_SET_ERR_MSG_MOD(extack, "legacy pedit isn't offloaded");
2844                 goto out_err;
2845         }
2846
2847         if (!mlx5e_flow_namespace_max_modify_action(priv->mdev, namespace)) {
2848                 NL_SET_ERR_MSG_MOD(extack,
2849                                    "The pedit offload action is not supported");
2850                 goto out_err;
2851         }
2852
2853         mask = act->mangle.mask;
2854         val = act->mangle.val;
2855         offset = act->mangle.offset;
2856
2857         err = set_pedit_val(htype, ~mask, val, offset, &hdrs[cmd]);
2858         if (err)
2859                 goto out_err;
2860
2861         hdrs[cmd].pedits++;
2862
2863         return 0;
2864 out_err:
2865         return err;
2866 }
2867
2868 static int alloc_tc_pedit_action(struct mlx5e_priv *priv, int namespace,
2869                                  struct mlx5e_tc_flow_parse_attr *parse_attr,
2870                                  struct pedit_headers_action *hdrs,
2871                                  u32 *action_flags,
2872                                  struct netlink_ext_ack *extack)
2873 {
2874         struct pedit_headers *cmd_masks;
2875         int err;
2876         u8 cmd;
2877
2878         err = offload_pedit_fields(priv, namespace, hdrs, parse_attr,
2879                                    action_flags, extack);
2880         if (err < 0)
2881                 goto out_dealloc_parsed_actions;
2882
2883         for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
2884                 cmd_masks = &hdrs[cmd].masks;
2885                 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
2886                         NL_SET_ERR_MSG_MOD(extack,
2887                                            "attempt to offload an unsupported field");
2888                         netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
2889                         print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
2890                                        16, 1, cmd_masks, sizeof(zero_masks), true);
2891                         err = -EOPNOTSUPP;
2892                         goto out_dealloc_parsed_actions;
2893                 }
2894         }
2895
2896         return 0;
2897
2898 out_dealloc_parsed_actions:
2899         dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
2900         return err;
2901 }
2902
2903 static bool csum_offload_supported(struct mlx5e_priv *priv,
2904                                    u32 action,
2905                                    u32 update_flags,
2906                                    struct netlink_ext_ack *extack)
2907 {
2908         u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
2909                          TCA_CSUM_UPDATE_FLAG_UDP;
2910
2911         /*  The HW recalcs checksums only if re-writing headers */
2912         if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
2913                 NL_SET_ERR_MSG_MOD(extack,
2914                                    "TC csum action is only offloaded with pedit");
2915                 netdev_warn(priv->netdev,
2916                             "TC csum action is only offloaded with pedit\n");
2917                 return false;
2918         }
2919
2920         if (update_flags & ~prot_flags) {
2921                 NL_SET_ERR_MSG_MOD(extack,
2922                                    "can't offload TC csum action for some header/s");
2923                 netdev_warn(priv->netdev,
2924                             "can't offload TC csum action for some header/s - flags %#x\n",
2925                             update_flags);
2926                 return false;
2927         }
2928
2929         return true;
2930 }
2931
2932 struct ip_ttl_word {
2933         __u8    ttl;
2934         __u8    protocol;
2935         __sum16 check;
2936 };
2937
2938 struct ipv6_hoplimit_word {
2939         __be16  payload_len;
2940         __u8    nexthdr;
2941         __u8    hop_limit;
2942 };
2943
2944 static int is_action_keys_supported(const struct flow_action_entry *act,
2945                                     bool ct_flow, bool *modify_ip_header,
2946                                     struct netlink_ext_ack *extack)
2947 {
2948         u32 mask, offset;
2949         u8 htype;
2950
2951         htype = act->mangle.htype;
2952         offset = act->mangle.offset;
2953         mask = ~act->mangle.mask;
2954         /* For IPv4 & IPv6 header check 4 byte word,
2955          * to determine that modified fields
2956          * are NOT ttl & hop_limit only.
2957          */
2958         if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP4) {
2959                 struct ip_ttl_word *ttl_word =
2960                         (struct ip_ttl_word *)&mask;
2961
2962                 if (offset != offsetof(struct iphdr, ttl) ||
2963                     ttl_word->protocol ||
2964                     ttl_word->check) {
2965                         *modify_ip_header = true;
2966                 }
2967
2968                 if (ct_flow && offset >= offsetof(struct iphdr, saddr)) {
2969                         NL_SET_ERR_MSG_MOD(extack,
2970                                            "can't offload re-write of ipv4 address with action ct");
2971                         return -EOPNOTSUPP;
2972                 }
2973         } else if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP6) {
2974                 struct ipv6_hoplimit_word *hoplimit_word =
2975                         (struct ipv6_hoplimit_word *)&mask;
2976
2977                 if (offset != offsetof(struct ipv6hdr, payload_len) ||
2978                     hoplimit_word->payload_len ||
2979                     hoplimit_word->nexthdr) {
2980                         *modify_ip_header = true;
2981                 }
2982
2983                 if (ct_flow && offset >= offsetof(struct ipv6hdr, saddr)) {
2984                         NL_SET_ERR_MSG_MOD(extack,
2985                                            "can't offload re-write of ipv6 address with action ct");
2986                         return -EOPNOTSUPP;
2987                 }
2988         } else if (ct_flow && (htype == FLOW_ACT_MANGLE_HDR_TYPE_TCP ||
2989                                htype == FLOW_ACT_MANGLE_HDR_TYPE_UDP)) {
2990                 NL_SET_ERR_MSG_MOD(extack,
2991                                    "can't offload re-write of transport header ports with action ct");
2992                 return -EOPNOTSUPP;
2993         }
2994
2995         return 0;
2996 }
2997
2998 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
2999                                           struct flow_action *flow_action,
3000                                           u32 actions, bool ct_flow,
3001                                           struct netlink_ext_ack *extack)
3002 {
3003         const struct flow_action_entry *act;
3004         bool modify_ip_header;
3005         void *headers_v;
3006         u16 ethertype;
3007         u8 ip_proto;
3008         int i, err;
3009
3010         headers_v = get_match_headers_value(actions, spec);
3011         ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
3012
3013         /* for non-IP we only re-write MACs, so we're okay */
3014         if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
3015                 goto out_ok;
3016
3017         modify_ip_header = false;
3018         flow_action_for_each(i, act, flow_action) {
3019                 if (act->id != FLOW_ACTION_MANGLE &&
3020                     act->id != FLOW_ACTION_ADD)
3021                         continue;
3022
3023                 err = is_action_keys_supported(act, ct_flow,
3024                                                &modify_ip_header, extack);
3025                 if (err)
3026                         return err;
3027         }
3028
3029         ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
3030         if (modify_ip_header && ip_proto != IPPROTO_TCP &&
3031             ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
3032                 NL_SET_ERR_MSG_MOD(extack,
3033                                    "can't offload re-write of non TCP/UDP");
3034                 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
3035                 return false;
3036         }
3037
3038 out_ok:
3039         return true;
3040 }
3041
3042 static bool actions_match_supported(struct mlx5e_priv *priv,
3043                                     struct flow_action *flow_action,
3044                                     struct mlx5e_tc_flow_parse_attr *parse_attr,
3045                                     struct mlx5e_tc_flow *flow,
3046                                     struct netlink_ext_ack *extack)
3047 {
3048         bool ct_flow;
3049         u32 actions;
3050
3051         ct_flow = flow_flag_test(flow, CT);
3052         if (mlx5e_is_eswitch_flow(flow)) {
3053                 actions = flow->esw_attr->action;
3054
3055                 if (flow->esw_attr->split_count && ct_flow) {
3056                         /* All registers used by ct are cleared when using
3057                          * split rules.
3058                          */
3059                         NL_SET_ERR_MSG_MOD(extack,
3060                                            "Can't offload mirroring with action ct");
3061                         return false;
3062                 }
3063         } else {
3064                 actions = flow->nic_attr->action;
3065         }
3066
3067         if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
3068                 return modify_header_match_supported(&parse_attr->spec,
3069                                                      flow_action, actions,
3070                                                      ct_flow, extack);
3071
3072         return true;
3073 }
3074
3075 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
3076 {
3077         struct mlx5_core_dev *fmdev, *pmdev;
3078         u64 fsystem_guid, psystem_guid;
3079
3080         fmdev = priv->mdev;
3081         pmdev = peer_priv->mdev;
3082
3083         fsystem_guid = mlx5_query_nic_system_image_guid(fmdev);
3084         psystem_guid = mlx5_query_nic_system_image_guid(pmdev);
3085
3086         return (fsystem_guid == psystem_guid);
3087 }
3088
3089 static int add_vlan_rewrite_action(struct mlx5e_priv *priv, int namespace,
3090                                    const struct flow_action_entry *act,
3091                                    struct mlx5e_tc_flow_parse_attr *parse_attr,
3092                                    struct pedit_headers_action *hdrs,
3093                                    u32 *action, struct netlink_ext_ack *extack)
3094 {
3095         u16 mask16 = VLAN_VID_MASK;
3096         u16 val16 = act->vlan.vid & VLAN_VID_MASK;
3097         const struct flow_action_entry pedit_act = {
3098                 .id = FLOW_ACTION_MANGLE,
3099                 .mangle.htype = FLOW_ACT_MANGLE_HDR_TYPE_ETH,
3100                 .mangle.offset = offsetof(struct vlan_ethhdr, h_vlan_TCI),
3101                 .mangle.mask = ~(u32)be16_to_cpu(*(__be16 *)&mask16),
3102                 .mangle.val = (u32)be16_to_cpu(*(__be16 *)&val16),
3103         };
3104         u8 match_prio_mask, match_prio_val;
3105         void *headers_c, *headers_v;
3106         int err;
3107
3108         headers_c = get_match_headers_criteria(*action, &parse_attr->spec);
3109         headers_v = get_match_headers_value(*action, &parse_attr->spec);
3110
3111         if (!(MLX5_GET(fte_match_set_lyr_2_4, headers_c, cvlan_tag) &&
3112               MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag))) {
3113                 NL_SET_ERR_MSG_MOD(extack,
3114                                    "VLAN rewrite action must have VLAN protocol match");
3115                 return -EOPNOTSUPP;
3116         }
3117
3118         match_prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
3119         match_prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
3120         if (act->vlan.prio != (match_prio_val & match_prio_mask)) {
3121                 NL_SET_ERR_MSG_MOD(extack,
3122                                    "Changing VLAN prio is not supported");
3123                 return -EOPNOTSUPP;
3124         }
3125
3126         err = parse_tc_pedit_action(priv, &pedit_act, namespace, hdrs, NULL);
3127         *action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3128
3129         return err;
3130 }
3131
3132 static int
3133 add_vlan_prio_tag_rewrite_action(struct mlx5e_priv *priv,
3134                                  struct mlx5e_tc_flow_parse_attr *parse_attr,
3135                                  struct pedit_headers_action *hdrs,
3136                                  u32 *action, struct netlink_ext_ack *extack)
3137 {
3138         const struct flow_action_entry prio_tag_act = {
3139                 .vlan.vid = 0,
3140                 .vlan.prio =
3141                         MLX5_GET(fte_match_set_lyr_2_4,
3142                                  get_match_headers_value(*action,
3143                                                          &parse_attr->spec),
3144                                  first_prio) &
3145                         MLX5_GET(fte_match_set_lyr_2_4,
3146                                  get_match_headers_criteria(*action,
3147                                                             &parse_attr->spec),
3148                                  first_prio),
3149         };
3150
3151         return add_vlan_rewrite_action(priv, MLX5_FLOW_NAMESPACE_FDB,
3152                                        &prio_tag_act, parse_attr, hdrs, action,
3153                                        extack);
3154 }
3155
3156 static int parse_tc_nic_actions(struct mlx5e_priv *priv,
3157                                 struct flow_action *flow_action,
3158                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
3159                                 struct mlx5e_tc_flow *flow,
3160                                 struct netlink_ext_ack *extack)
3161 {
3162         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
3163         struct pedit_headers_action hdrs[2] = {};
3164         const struct flow_action_entry *act;
3165         u32 action = 0;
3166         int err, i;
3167
3168         if (!flow_action_has_entries(flow_action))
3169                 return -EINVAL;
3170
3171         if (!flow_action_hw_stats_check(flow_action, extack,
3172                                         FLOW_ACTION_HW_STATS_DELAYED_BIT))
3173                 return -EOPNOTSUPP;
3174
3175         attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
3176
3177         flow_action_for_each(i, act, flow_action) {
3178                 switch (act->id) {
3179                 case FLOW_ACTION_ACCEPT:
3180                         action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
3181                                   MLX5_FLOW_CONTEXT_ACTION_COUNT;
3182                         break;
3183                 case FLOW_ACTION_DROP:
3184                         action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
3185                         if (MLX5_CAP_FLOWTABLE(priv->mdev,
3186                                                flow_table_properties_nic_receive.flow_counter))
3187                                 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
3188                         break;
3189                 case FLOW_ACTION_MANGLE:
3190                 case FLOW_ACTION_ADD:
3191                         err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_KERNEL,
3192                                                     hdrs, extack);
3193                         if (err)
3194                                 return err;
3195
3196                         action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
3197                                   MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3198                         break;
3199                 case FLOW_ACTION_VLAN_MANGLE:
3200                         err = add_vlan_rewrite_action(priv,
3201                                                       MLX5_FLOW_NAMESPACE_KERNEL,
3202                                                       act, parse_attr, hdrs,
3203                                                       &action, extack);
3204                         if (err)
3205                                 return err;
3206
3207                         break;
3208                 case FLOW_ACTION_CSUM:
3209                         if (csum_offload_supported(priv, action,
3210                                                    act->csum_flags,
3211                                                    extack))
3212                                 break;
3213
3214                         return -EOPNOTSUPP;
3215                 case FLOW_ACTION_REDIRECT: {
3216                         struct net_device *peer_dev = act->dev;
3217
3218                         if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
3219                             same_hw_devs(priv, netdev_priv(peer_dev))) {
3220                                 parse_attr->mirred_ifindex[0] = peer_dev->ifindex;
3221                                 flow_flag_set(flow, HAIRPIN);
3222                                 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
3223                                           MLX5_FLOW_CONTEXT_ACTION_COUNT;
3224                         } else {
3225                                 NL_SET_ERR_MSG_MOD(extack,
3226                                                    "device is not on same HW, can't offload");
3227                                 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
3228                                             peer_dev->name);
3229                                 return -EINVAL;
3230                         }
3231                         }
3232                         break;
3233                 case FLOW_ACTION_MARK: {
3234                         u32 mark = act->mark;
3235
3236                         if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
3237                                 NL_SET_ERR_MSG_MOD(extack,
3238                                                    "Bad flow mark - only 16 bit is supported");
3239                                 return -EINVAL;
3240                         }
3241
3242                         attr->flow_tag = mark;
3243                         action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3244                         }
3245                         break;
3246                 default:
3247                         NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
3248                         return -EOPNOTSUPP;
3249                 }
3250         }
3251
3252         if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
3253             hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
3254                 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_KERNEL,
3255                                             parse_attr, hdrs, &action, extack);
3256                 if (err)
3257                         return err;
3258                 /* in case all pedit actions are skipped, remove the MOD_HDR
3259                  * flag.
3260                  */
3261                 if (parse_attr->mod_hdr_acts.num_actions == 0) {
3262                         action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3263                         dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
3264                 }
3265         }
3266
3267         attr->action = action;
3268         if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
3269                 return -EOPNOTSUPP;
3270
3271         return 0;
3272 }
3273
3274 struct encap_key {
3275         const struct ip_tunnel_key *ip_tun_key;
3276         struct mlx5e_tc_tunnel *tc_tunnel;
3277 };
3278
3279 static inline int cmp_encap_info(struct encap_key *a,
3280                                  struct encap_key *b)
3281 {
3282         return memcmp(a->ip_tun_key, b->ip_tun_key, sizeof(*a->ip_tun_key)) ||
3283                a->tc_tunnel->tunnel_type != b->tc_tunnel->tunnel_type;
3284 }
3285
3286 static inline int hash_encap_info(struct encap_key *key)
3287 {
3288         return jhash(key->ip_tun_key, sizeof(*key->ip_tun_key),
3289                      key->tc_tunnel->tunnel_type);
3290 }
3291
3292
3293 static bool is_merged_eswitch_dev(struct mlx5e_priv *priv,
3294                                   struct net_device *peer_netdev)
3295 {
3296         struct mlx5e_priv *peer_priv;
3297
3298         peer_priv = netdev_priv(peer_netdev);
3299
3300         return (MLX5_CAP_ESW(priv->mdev, merged_eswitch) &&
3301                 mlx5e_eswitch_rep(priv->netdev) &&
3302                 mlx5e_eswitch_rep(peer_netdev) &&
3303                 same_hw_devs(priv, peer_priv));
3304 }
3305
3306
3307
3308 bool mlx5e_encap_take(struct mlx5e_encap_entry *e)
3309 {
3310         return refcount_inc_not_zero(&e->refcnt);
3311 }
3312
3313 static struct mlx5e_encap_entry *
3314 mlx5e_encap_get(struct mlx5e_priv *priv, struct encap_key *key,
3315                 uintptr_t hash_key)
3316 {
3317         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3318         struct mlx5e_encap_entry *e;
3319         struct encap_key e_key;
3320
3321         hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
3322                                    encap_hlist, hash_key) {
3323                 e_key.ip_tun_key = &e->tun_info->key;
3324                 e_key.tc_tunnel = e->tunnel;
3325                 if (!cmp_encap_info(&e_key, key) &&
3326                     mlx5e_encap_take(e))
3327                         return e;
3328         }
3329
3330         return NULL;
3331 }
3332
3333 static struct ip_tunnel_info *dup_tun_info(const struct ip_tunnel_info *tun_info)
3334 {
3335         size_t tun_size = sizeof(*tun_info) + tun_info->options_len;
3336
3337         return kmemdup(tun_info, tun_size, GFP_KERNEL);
3338 }
3339
3340 static bool is_duplicated_encap_entry(struct mlx5e_priv *priv,
3341                                       struct mlx5e_tc_flow *flow,
3342                                       int out_index,
3343                                       struct mlx5e_encap_entry *e,
3344                                       struct netlink_ext_ack *extack)
3345 {
3346         int i;
3347
3348         for (i = 0; i < out_index; i++) {
3349                 if (flow->encaps[i].e != e)
3350                         continue;
3351                 NL_SET_ERR_MSG_MOD(extack, "can't duplicate encap action");
3352                 netdev_err(priv->netdev, "can't duplicate encap action\n");
3353                 return true;
3354         }
3355
3356         return false;
3357 }
3358
3359 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
3360                               struct mlx5e_tc_flow *flow,
3361                               struct net_device *mirred_dev,
3362                               int out_index,
3363                               struct netlink_ext_ack *extack,
3364                               struct net_device **encap_dev,
3365                               bool *encap_valid)
3366 {
3367         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3368         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3369         struct mlx5e_tc_flow_parse_attr *parse_attr;
3370         const struct ip_tunnel_info *tun_info;
3371         struct encap_key key;
3372         struct mlx5e_encap_entry *e;
3373         unsigned short family;
3374         uintptr_t hash_key;
3375         int err = 0;
3376
3377         parse_attr = attr->parse_attr;
3378         tun_info = parse_attr->tun_info[out_index];
3379         family = ip_tunnel_info_af(tun_info);
3380         key.ip_tun_key = &tun_info->key;
3381         key.tc_tunnel = mlx5e_get_tc_tun(mirred_dev);
3382         if (!key.tc_tunnel) {
3383                 NL_SET_ERR_MSG_MOD(extack, "Unsupported tunnel");
3384                 return -EOPNOTSUPP;
3385         }
3386
3387         hash_key = hash_encap_info(&key);
3388
3389         mutex_lock(&esw->offloads.encap_tbl_lock);
3390         e = mlx5e_encap_get(priv, &key, hash_key);
3391
3392         /* must verify if encap is valid or not */
3393         if (e) {
3394                 /* Check that entry was not already attached to this flow */
3395                 if (is_duplicated_encap_entry(priv, flow, out_index, e, extack)) {
3396                         err = -EOPNOTSUPP;
3397                         goto out_err;
3398                 }
3399
3400                 mutex_unlock(&esw->offloads.encap_tbl_lock);
3401                 wait_for_completion(&e->res_ready);
3402
3403                 /* Protect against concurrent neigh update. */
3404                 mutex_lock(&esw->offloads.encap_tbl_lock);
3405                 if (e->compl_result < 0) {
3406                         err = -EREMOTEIO;
3407                         goto out_err;
3408                 }
3409                 goto attach_flow;
3410         }
3411
3412         e = kzalloc(sizeof(*e), GFP_KERNEL);
3413         if (!e) {
3414                 err = -ENOMEM;
3415                 goto out_err;
3416         }
3417
3418         refcount_set(&e->refcnt, 1);
3419         init_completion(&e->res_ready);
3420
3421         tun_info = dup_tun_info(tun_info);
3422         if (!tun_info) {
3423                 err = -ENOMEM;
3424                 goto out_err_init;
3425         }
3426         e->tun_info = tun_info;
3427         err = mlx5e_tc_tun_init_encap_attr(mirred_dev, priv, e, extack);
3428         if (err)
3429                 goto out_err_init;
3430
3431         INIT_LIST_HEAD(&e->flows);
3432         hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
3433         mutex_unlock(&esw->offloads.encap_tbl_lock);
3434
3435         if (family == AF_INET)
3436                 err = mlx5e_tc_tun_create_header_ipv4(priv, mirred_dev, e);
3437         else if (family == AF_INET6)
3438                 err = mlx5e_tc_tun_create_header_ipv6(priv, mirred_dev, e);
3439
3440         /* Protect against concurrent neigh update. */
3441         mutex_lock(&esw->offloads.encap_tbl_lock);
3442         complete_all(&e->res_ready);
3443         if (err) {
3444                 e->compl_result = err;
3445                 goto out_err;
3446         }
3447         e->compl_result = 1;
3448
3449 attach_flow:
3450         flow->encaps[out_index].e = e;
3451         list_add(&flow->encaps[out_index].list, &e->flows);
3452         flow->encaps[out_index].index = out_index;
3453         *encap_dev = e->out_dev;
3454         if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
3455                 attr->dests[out_index].pkt_reformat = e->pkt_reformat;
3456                 attr->dests[out_index].flags |= MLX5_ESW_DEST_ENCAP_VALID;
3457                 *encap_valid = true;
3458         } else {
3459                 *encap_valid = false;
3460         }
3461         mutex_unlock(&esw->offloads.encap_tbl_lock);
3462
3463         return err;
3464
3465 out_err:
3466         mutex_unlock(&esw->offloads.encap_tbl_lock);
3467         if (e)
3468                 mlx5e_encap_put(priv, e);
3469         return err;
3470
3471 out_err_init:
3472         mutex_unlock(&esw->offloads.encap_tbl_lock);
3473         kfree(tun_info);
3474         kfree(e);
3475         return err;
3476 }
3477
3478 static int parse_tc_vlan_action(struct mlx5e_priv *priv,
3479                                 const struct flow_action_entry *act,
3480                                 struct mlx5_esw_flow_attr *attr,
3481                                 u32 *action)
3482 {
3483         u8 vlan_idx = attr->total_vlan;
3484
3485         if (vlan_idx >= MLX5_FS_VLAN_DEPTH)
3486                 return -EOPNOTSUPP;
3487
3488         switch (act->id) {
3489         case FLOW_ACTION_VLAN_POP:
3490                 if (vlan_idx) {
3491                         if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
3492                                                                  MLX5_FS_VLAN_DEPTH))
3493                                 return -EOPNOTSUPP;
3494
3495                         *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2;
3496                 } else {
3497                         *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3498                 }
3499                 break;
3500         case FLOW_ACTION_VLAN_PUSH:
3501                 attr->vlan_vid[vlan_idx] = act->vlan.vid;
3502                 attr->vlan_prio[vlan_idx] = act->vlan.prio;
3503                 attr->vlan_proto[vlan_idx] = act->vlan.proto;
3504                 if (!attr->vlan_proto[vlan_idx])
3505                         attr->vlan_proto[vlan_idx] = htons(ETH_P_8021Q);
3506
3507                 if (vlan_idx) {
3508                         if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
3509                                                                  MLX5_FS_VLAN_DEPTH))
3510                                 return -EOPNOTSUPP;
3511
3512                         *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2;
3513                 } else {
3514                         if (!mlx5_eswitch_vlan_actions_supported(priv->mdev, 1) &&
3515                             (act->vlan.proto != htons(ETH_P_8021Q) ||
3516                              act->vlan.prio))
3517                                 return -EOPNOTSUPP;
3518
3519                         *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
3520                 }
3521                 break;
3522         default:
3523                 return -EINVAL;
3524         }
3525
3526         attr->total_vlan = vlan_idx + 1;
3527
3528         return 0;
3529 }
3530
3531 static int add_vlan_push_action(struct mlx5e_priv *priv,
3532                                 struct mlx5_esw_flow_attr *attr,
3533                                 struct net_device **out_dev,
3534                                 u32 *action)
3535 {
3536         struct net_device *vlan_dev = *out_dev;
3537         struct flow_action_entry vlan_act = {
3538                 .id = FLOW_ACTION_VLAN_PUSH,
3539                 .vlan.vid = vlan_dev_vlan_id(vlan_dev),
3540                 .vlan.proto = vlan_dev_vlan_proto(vlan_dev),
3541                 .vlan.prio = 0,
3542         };
3543         int err;
3544
3545         err = parse_tc_vlan_action(priv, &vlan_act, attr, action);
3546         if (err)
3547                 return err;
3548
3549         *out_dev = dev_get_by_index_rcu(dev_net(vlan_dev),
3550                                         dev_get_iflink(vlan_dev));
3551         if (is_vlan_dev(*out_dev))
3552                 err = add_vlan_push_action(priv, attr, out_dev, action);
3553
3554         return err;
3555 }
3556
3557 static int add_vlan_pop_action(struct mlx5e_priv *priv,
3558                                struct mlx5_esw_flow_attr *attr,
3559                                u32 *action)
3560 {
3561         int nest_level = attr->parse_attr->filter_dev->lower_level;
3562         struct flow_action_entry vlan_act = {
3563                 .id = FLOW_ACTION_VLAN_POP,
3564         };
3565         int err = 0;
3566
3567         while (nest_level--) {
3568                 err = parse_tc_vlan_action(priv, &vlan_act, attr, action);
3569                 if (err)
3570                         return err;
3571         }
3572
3573         return err;
3574 }
3575
3576 bool mlx5e_is_valid_eswitch_fwd_dev(struct mlx5e_priv *priv,
3577                                     struct net_device *out_dev)
3578 {
3579         if (is_merged_eswitch_dev(priv, out_dev))
3580                 return true;
3581
3582         return mlx5e_eswitch_rep(out_dev) &&
3583                same_hw_devs(priv, netdev_priv(out_dev));
3584 }
3585
3586 static bool is_duplicated_output_device(struct net_device *dev,
3587                                         struct net_device *out_dev,
3588                                         int *ifindexes, int if_count,
3589                                         struct netlink_ext_ack *extack)
3590 {
3591         int i;
3592
3593         for (i = 0; i < if_count; i++) {
3594                 if (ifindexes[i] == out_dev->ifindex) {
3595                         NL_SET_ERR_MSG_MOD(extack,
3596                                            "can't duplicate output to same device");
3597                         netdev_err(dev, "can't duplicate output to same device: %s\n",
3598                                    out_dev->name);
3599                         return true;
3600                 }
3601         }
3602
3603         return false;
3604 }
3605
3606 static int mlx5_validate_goto_chain(struct mlx5_eswitch *esw,
3607                                     struct mlx5e_tc_flow *flow,
3608                                     const struct flow_action_entry *act,
3609                                     u32 actions,
3610                                     struct netlink_ext_ack *extack)
3611 {
3612         u32 max_chain = mlx5_esw_chains_get_chain_range(esw);
3613         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3614         bool ft_flow = mlx5e_is_ft_flow(flow);
3615         u32 dest_chain = act->chain_index;
3616
3617         if (ft_flow) {
3618                 NL_SET_ERR_MSG_MOD(extack, "Goto action is not supported");
3619                 return -EOPNOTSUPP;
3620         }
3621
3622         if (!mlx5_esw_chains_backwards_supported(esw) &&
3623             dest_chain <= attr->chain) {
3624                 NL_SET_ERR_MSG_MOD(extack,
3625                                    "Goto lower numbered chain isn't supported");
3626                 return -EOPNOTSUPP;
3627         }
3628         if (dest_chain > max_chain) {
3629                 NL_SET_ERR_MSG_MOD(extack,
3630                                    "Requested destination chain is out of supported range");
3631                 return -EOPNOTSUPP;
3632         }
3633
3634         if (actions & (MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
3635                        MLX5_FLOW_CONTEXT_ACTION_DECAP) &&
3636             !MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, reformat_and_fwd_to_table)) {
3637                 NL_SET_ERR_MSG_MOD(extack,
3638                                    "Goto chain is not allowed if action has reformat or decap");
3639                 return -EOPNOTSUPP;
3640         }
3641
3642         return 0;
3643 }
3644
3645 static int verify_uplink_forwarding(struct mlx5e_priv *priv,
3646                                     struct mlx5e_tc_flow *flow,
3647                                     struct net_device *out_dev,
3648                                     struct netlink_ext_ack *extack)
3649 {
3650         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3651         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3652         struct mlx5e_rep_priv *rep_priv;
3653
3654         /* Forwarding non encapsulated traffic between
3655          * uplink ports is allowed only if
3656          * termination_table_raw_traffic cap is set.
3657          *
3658          * Input vport was stored esw_attr->in_rep.
3659          * In LAG case, *priv* is the private data of
3660          * uplink which may be not the input vport.
3661          */
3662         rep_priv = mlx5e_rep_to_rep_priv(attr->in_rep);
3663
3664         if (!(mlx5e_eswitch_uplink_rep(rep_priv->netdev) &&
3665               mlx5e_eswitch_uplink_rep(out_dev)))
3666                 return 0;
3667
3668         if (!MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev,
3669                                         termination_table_raw_traffic)) {
3670                 NL_SET_ERR_MSG_MOD(extack,
3671                                    "devices are both uplink, can't offload forwarding");
3672                         pr_err("devices %s %s are both uplink, can't offload forwarding\n",
3673                                priv->netdev->name, out_dev->name);
3674                         return -EOPNOTSUPP;
3675         } else if (out_dev != rep_priv->netdev) {
3676                 NL_SET_ERR_MSG_MOD(extack,
3677                                    "devices are not the same uplink, can't offload forwarding");
3678                 pr_err("devices %s %s are both uplink but not the same, can't offload forwarding\n",
3679                        priv->netdev->name, out_dev->name);
3680                 return -EOPNOTSUPP;
3681         }
3682         return 0;
3683 }
3684
3685 static int parse_tc_fdb_actions(struct mlx5e_priv *priv,
3686                                 struct flow_action *flow_action,
3687                                 struct mlx5e_tc_flow *flow,
3688                                 struct netlink_ext_ack *extack)
3689 {
3690         struct pedit_headers_action hdrs[2] = {};
3691         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3692         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3693         struct mlx5e_tc_flow_parse_attr *parse_attr = attr->parse_attr;
3694         struct mlx5e_rep_priv *rpriv = priv->ppriv;
3695         const struct ip_tunnel_info *info = NULL;
3696         int ifindexes[MLX5_MAX_FLOW_FWD_VPORTS];
3697         bool ft_flow = mlx5e_is_ft_flow(flow);
3698         const struct flow_action_entry *act;
3699         bool encap = false, decap = false;
3700         u32 action = attr->action;
3701         int err, i, if_count = 0;
3702
3703         if (!flow_action_has_entries(flow_action))
3704                 return -EINVAL;
3705
3706         if (!flow_action_hw_stats_check(flow_action, extack,
3707                                         FLOW_ACTION_HW_STATS_DELAYED_BIT))
3708                 return -EOPNOTSUPP;
3709
3710         flow_action_for_each(i, act, flow_action) {
3711                 switch (act->id) {
3712                 case FLOW_ACTION_DROP:
3713                         action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
3714                                   MLX5_FLOW_CONTEXT_ACTION_COUNT;
3715                         break;
3716                 case FLOW_ACTION_MANGLE:
3717                 case FLOW_ACTION_ADD:
3718                         err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_FDB,
3719                                                     hdrs, extack);
3720                         if (err)
3721                                 return err;
3722
3723                         action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3724                         attr->split_count = attr->out_count;
3725                         break;
3726                 case FLOW_ACTION_CSUM:
3727                         if (csum_offload_supported(priv, action,
3728                                                    act->csum_flags, extack))
3729                                 break;
3730
3731                         return -EOPNOTSUPP;
3732                 case FLOW_ACTION_REDIRECT:
3733                 case FLOW_ACTION_MIRRED: {
3734                         struct mlx5e_priv *out_priv;
3735                         struct net_device *out_dev;
3736
3737                         out_dev = act->dev;
3738                         if (!out_dev) {
3739                                 /* out_dev is NULL when filters with
3740                                  * non-existing mirred device are replayed to
3741                                  * the driver.
3742                                  */
3743                                 return -EINVAL;
3744                         }
3745
3746                         if (ft_flow && out_dev == priv->netdev) {
3747                                 /* Ignore forward to self rules generated
3748                                  * by adding both mlx5 devs to the flow table
3749                                  * block on a normal nft offload setup.
3750                                  */
3751                                 return -EOPNOTSUPP;
3752                         }
3753
3754                         if (attr->out_count >= MLX5_MAX_FLOW_FWD_VPORTS) {
3755                                 NL_SET_ERR_MSG_MOD(extack,
3756                                                    "can't support more output ports, can't offload forwarding");
3757                                 netdev_warn(priv->netdev,
3758                                             "can't support more than %d output ports, can't offload forwarding\n",
3759                                             attr->out_count);
3760                                 return -EOPNOTSUPP;
3761                         }
3762
3763                         action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
3764                                   MLX5_FLOW_CONTEXT_ACTION_COUNT;
3765                         if (encap) {
3766                                 parse_attr->mirred_ifindex[attr->out_count] =
3767                                         out_dev->ifindex;
3768                                 parse_attr->tun_info[attr->out_count] = dup_tun_info(info);
3769                                 if (!parse_attr->tun_info[attr->out_count])
3770                                         return -ENOMEM;
3771                                 encap = false;
3772                                 attr->dests[attr->out_count].flags |=
3773                                         MLX5_ESW_DEST_ENCAP;
3774                                 attr->out_count++;
3775                                 /* attr->dests[].rep is resolved when we
3776                                  * handle encap
3777                                  */
3778                         } else if (netdev_port_same_parent_id(priv->netdev, out_dev)) {
3779                                 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3780                                 struct net_device *uplink_dev = mlx5_eswitch_uplink_get_proto_dev(esw, REP_ETH);
3781                                 struct net_device *uplink_upper;
3782
3783                                 if (is_duplicated_output_device(priv->netdev,
3784                                                                 out_dev,
3785                                                                 ifindexes,
3786                                                                 if_count,
3787                                                                 extack))
3788                                         return -EOPNOTSUPP;
3789
3790                                 ifindexes[if_count] = out_dev->ifindex;
3791                                 if_count++;
3792
3793                                 rcu_read_lock();
3794                                 uplink_upper =
3795                                         netdev_master_upper_dev_get_rcu(uplink_dev);
3796                                 if (uplink_upper &&
3797                                     netif_is_lag_master(uplink_upper) &&
3798                                     uplink_upper == out_dev)
3799                                         out_dev = uplink_dev;
3800                                 rcu_read_unlock();
3801
3802                                 if (is_vlan_dev(out_dev)) {
3803                                         err = add_vlan_push_action(priv, attr,
3804                                                                    &out_dev,
3805                                                                    &action);
3806                                         if (err)
3807                                                 return err;
3808                                 }
3809
3810                                 if (is_vlan_dev(parse_attr->filter_dev)) {
3811                                         err = add_vlan_pop_action(priv, attr,
3812                                                                   &action);
3813                                         if (err)
3814                                                 return err;
3815                                 }
3816
3817                                 err = verify_uplink_forwarding(priv, flow, out_dev, extack);
3818                                 if (err)
3819                                         return err;
3820
3821                                 if (!mlx5e_is_valid_eswitch_fwd_dev(priv, out_dev)) {
3822                                         NL_SET_ERR_MSG_MOD(extack,
3823                                                            "devices are not on same switch HW, can't offload forwarding");
3824                                         netdev_warn(priv->netdev,
3825                                                     "devices %s %s not on same switch HW, can't offload forwarding\n",
3826                                                     priv->netdev->name,
3827                                                     out_dev->name);
3828                                         return -EOPNOTSUPP;
3829                                 }
3830
3831                                 out_priv = netdev_priv(out_dev);
3832                                 rpriv = out_priv->ppriv;
3833                                 attr->dests[attr->out_count].rep = rpriv->rep;
3834                                 attr->dests[attr->out_count].mdev = out_priv->mdev;
3835                                 attr->out_count++;
3836                         } else if (parse_attr->filter_dev != priv->netdev) {
3837                                 /* All mlx5 devices are called to configure
3838                                  * high level device filters. Therefore, the
3839                                  * *attempt* to  install a filter on invalid
3840                                  * eswitch should not trigger an explicit error
3841                                  */
3842                                 return -EINVAL;
3843                         } else {
3844                                 NL_SET_ERR_MSG_MOD(extack,
3845                                                    "devices are not on same switch HW, can't offload forwarding");
3846                                 netdev_warn(priv->netdev,
3847                                             "devices %s %s not on same switch HW, can't offload forwarding\n",
3848                                             priv->netdev->name,
3849                                             out_dev->name);
3850                                 return -EINVAL;
3851                         }
3852                         }
3853                         break;
3854                 case FLOW_ACTION_TUNNEL_ENCAP:
3855                         info = act->tunnel;
3856                         if (info)
3857                                 encap = true;
3858                         else
3859                                 return -EOPNOTSUPP;
3860
3861                         break;
3862                 case FLOW_ACTION_VLAN_PUSH:
3863                 case FLOW_ACTION_VLAN_POP:
3864                         if (act->id == FLOW_ACTION_VLAN_PUSH &&
3865                             (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP)) {
3866                                 /* Replace vlan pop+push with vlan modify */
3867                                 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3868                                 err = add_vlan_rewrite_action(priv,
3869                                                               MLX5_FLOW_NAMESPACE_FDB,
3870                                                               act, parse_attr, hdrs,
3871                                                               &action, extack);
3872                         } else {
3873                                 err = parse_tc_vlan_action(priv, act, attr, &action);
3874                         }
3875                         if (err)
3876                                 return err;
3877
3878                         attr->split_count = attr->out_count;
3879                         break;
3880                 case FLOW_ACTION_VLAN_MANGLE:
3881                         err = add_vlan_rewrite_action(priv,
3882                                                       MLX5_FLOW_NAMESPACE_FDB,
3883                                                       act, parse_attr, hdrs,
3884                                                       &action, extack);
3885                         if (err)
3886                                 return err;
3887
3888                         attr->split_count = attr->out_count;
3889                         break;
3890                 case FLOW_ACTION_TUNNEL_DECAP:
3891                         decap = true;
3892                         break;
3893                 case FLOW_ACTION_GOTO:
3894                         err = mlx5_validate_goto_chain(esw, flow, act, action,
3895                                                        extack);
3896                         if (err)
3897                                 return err;
3898
3899                         action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
3900                         attr->dest_chain = act->chain_index;
3901                         break;
3902                 case FLOW_ACTION_CT:
3903                         err = mlx5_tc_ct_parse_action(priv, attr, act, extack);
3904                         if (err)
3905                                 return err;
3906
3907                         flow_flag_set(flow, CT);
3908                         break;
3909                 default:
3910                         NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
3911                         return -EOPNOTSUPP;
3912                 }
3913         }
3914
3915         if (MLX5_CAP_GEN(esw->dev, prio_tag_required) &&
3916             action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) {
3917                 /* For prio tag mode, replace vlan pop with rewrite vlan prio
3918                  * tag rewrite.
3919                  */
3920                 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3921                 err = add_vlan_prio_tag_rewrite_action(priv, parse_attr, hdrs,
3922                                                        &action, extack);
3923                 if (err)
3924                         return err;
3925         }
3926
3927         if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
3928             hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
3929                 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_FDB,
3930                                             parse_attr, hdrs, &action, extack);
3931                 if (err)
3932                         return err;
3933                 /* in case all pedit actions are skipped, remove the MOD_HDR
3934                  * flag. we might have set split_count either by pedit or
3935                  * pop/push. if there is no pop/push either, reset it too.
3936                  */
3937                 if (parse_attr->mod_hdr_acts.num_actions == 0) {
3938                         action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3939                         dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
3940                         if (!((action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) ||
3941                               (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH)))
3942                                 attr->split_count = 0;
3943                 }
3944         }
3945
3946         attr->action = action;
3947         if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
3948                 return -EOPNOTSUPP;
3949
3950         if (attr->dest_chain) {
3951                 if (decap) {
3952                         /* It can be supported if we'll create a mapping for
3953                          * the tunnel device only (without tunnel), and set
3954                          * this tunnel id with this decap flow.
3955                          *
3956                          * On restore (miss), we'll just set this saved tunnel
3957                          * device.
3958                          */
3959
3960                         NL_SET_ERR_MSG(extack,
3961                                        "Decap with goto isn't supported");
3962                         netdev_warn(priv->netdev,
3963                                     "Decap with goto isn't supported");
3964                         return -EOPNOTSUPP;
3965                 }
3966
3967                 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
3968                         NL_SET_ERR_MSG_MOD(extack,
3969                                            "Mirroring goto chain rules isn't supported");
3970                         return -EOPNOTSUPP;
3971                 }
3972                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3973         }
3974
3975         if (!(attr->action &
3976               (MLX5_FLOW_CONTEXT_ACTION_FWD_DEST | MLX5_FLOW_CONTEXT_ACTION_DROP))) {
3977                 NL_SET_ERR_MSG_MOD(extack,
3978                                    "Rule must have at least one forward/drop action");
3979                 return -EOPNOTSUPP;
3980         }
3981
3982         if (attr->split_count > 0 && !mlx5_esw_has_fwd_fdb(priv->mdev)) {
3983                 NL_SET_ERR_MSG_MOD(extack,
3984                                    "current firmware doesn't support split rule for port mirroring");
3985                 netdev_warn_once(priv->netdev, "current firmware doesn't support split rule for port mirroring\n");
3986                 return -EOPNOTSUPP;
3987         }
3988
3989         return 0;
3990 }
3991
3992 static void get_flags(int flags, unsigned long *flow_flags)
3993 {
3994         unsigned long __flow_flags = 0;
3995
3996         if (flags & MLX5_TC_FLAG(INGRESS))
3997                 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_INGRESS);
3998         if (flags & MLX5_TC_FLAG(EGRESS))
3999                 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_EGRESS);
4000
4001         if (flags & MLX5_TC_FLAG(ESW_OFFLOAD))
4002                 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_ESWITCH);
4003         if (flags & MLX5_TC_FLAG(NIC_OFFLOAD))
4004                 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_NIC);
4005         if (flags & MLX5_TC_FLAG(FT_OFFLOAD))
4006                 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_FT);
4007
4008         *flow_flags = __flow_flags;
4009 }
4010
4011 static const struct rhashtable_params tc_ht_params = {
4012         .head_offset = offsetof(struct mlx5e_tc_flow, node),
4013         .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
4014         .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
4015         .automatic_shrinking = true,
4016 };
4017
4018 static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv,
4019                                     unsigned long flags)
4020 {
4021         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
4022         struct mlx5e_rep_priv *uplink_rpriv;
4023
4024         if (flags & MLX5_TC_FLAG(ESW_OFFLOAD)) {
4025                 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
4026                 return &uplink_rpriv->uplink_priv.tc_ht;
4027         } else /* NIC offload */
4028                 return &priv->fs.tc.ht;
4029 }
4030
4031 static bool is_peer_flow_needed(struct mlx5e_tc_flow *flow)
4032 {
4033         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
4034         bool is_rep_ingress = attr->in_rep->vport != MLX5_VPORT_UPLINK &&
4035                 flow_flag_test(flow, INGRESS);
4036         bool act_is_encap = !!(attr->action &
4037                                MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT);
4038         bool esw_paired = mlx5_devcom_is_paired(attr->in_mdev->priv.devcom,
4039                                                 MLX5_DEVCOM_ESW_OFFLOADS);
4040
4041         if (!esw_paired)
4042                 return false;
4043
4044         if ((mlx5_lag_is_sriov(attr->in_mdev) ||
4045              mlx5_lag_is_multipath(attr->in_mdev)) &&
4046             (is_rep_ingress || act_is_encap))
4047                 return true;
4048
4049         return false;
4050 }
4051
4052 static int
4053 mlx5e_alloc_flow(struct mlx5e_priv *priv, int attr_size,
4054                  struct flow_cls_offload *f, unsigned long flow_flags,
4055                  struct mlx5e_tc_flow_parse_attr **__parse_attr,
4056                  struct mlx5e_tc_flow **__flow)
4057 {
4058         struct mlx5e_tc_flow_parse_attr *parse_attr;
4059         struct mlx5e_tc_flow *flow;
4060         int out_index, err;
4061
4062         flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
4063         parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
4064         if (!parse_attr || !flow) {
4065                 err = -ENOMEM;
4066                 goto err_free;
4067         }
4068
4069         flow->cookie = f->cookie;
4070         flow->flags = flow_flags;
4071         flow->priv = priv;
4072         for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++)
4073                 INIT_LIST_HEAD(&flow->encaps[out_index].list);
4074         INIT_LIST_HEAD(&flow->mod_hdr);
4075         INIT_LIST_HEAD(&flow->hairpin);
4076         refcount_set(&flow->refcnt, 1);
4077         init_completion(&flow->init_done);
4078
4079         *__flow = flow;
4080         *__parse_attr = parse_attr;
4081
4082         return 0;
4083
4084 err_free:
4085         kfree(flow);
4086         kvfree(parse_attr);
4087         return err;
4088 }
4089
4090 static void
4091 mlx5e_flow_esw_attr_init(struct mlx5_esw_flow_attr *esw_attr,
4092                          struct mlx5e_priv *priv,
4093                          struct mlx5e_tc_flow_parse_attr *parse_attr,
4094                          struct flow_cls_offload *f,
4095                          struct mlx5_eswitch_rep *in_rep,
4096                          struct mlx5_core_dev *in_mdev)
4097 {
4098         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
4099
4100         esw_attr->parse_attr = parse_attr;
4101         esw_attr->chain = f->common.chain_index;
4102         esw_attr->prio = f->common.prio;
4103
4104         esw_attr->in_rep = in_rep;
4105         esw_attr->in_mdev = in_mdev;
4106
4107         if (MLX5_CAP_ESW(esw->dev, counter_eswitch_affinity) ==
4108             MLX5_COUNTER_SOURCE_ESWITCH)
4109                 esw_attr->counter_dev = in_mdev;
4110         else
4111                 esw_attr->counter_dev = priv->mdev;
4112 }
4113
4114 static struct mlx5e_tc_flow *
4115 __mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
4116                      struct flow_cls_offload *f,
4117                      unsigned long flow_flags,
4118                      struct net_device *filter_dev,
4119                      struct mlx5_eswitch_rep *in_rep,
4120                      struct mlx5_core_dev *in_mdev)
4121 {
4122         struct flow_rule *rule = flow_cls_offload_flow_rule(f);
4123         struct netlink_ext_ack *extack = f->common.extack;
4124         struct mlx5e_tc_flow_parse_attr *parse_attr;
4125         struct mlx5e_tc_flow *flow;
4126         int attr_size, err;
4127
4128         flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_ESWITCH);
4129         attr_size  = sizeof(struct mlx5_esw_flow_attr);
4130         err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
4131                                &parse_attr, &flow);
4132         if (err)
4133                 goto out;
4134
4135         parse_attr->filter_dev = filter_dev;
4136         mlx5e_flow_esw_attr_init(flow->esw_attr,
4137                                  priv, parse_attr,
4138                                  f, in_rep, in_mdev);
4139
4140         err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
4141                                f, filter_dev);
4142         if (err)
4143                 goto err_free;
4144
4145         err = parse_tc_fdb_actions(priv, &rule->action, flow, extack);
4146         if (err)
4147                 goto err_free;
4148
4149         err = mlx5_tc_ct_parse_match(priv, &parse_attr->spec, f, extack);
4150         if (err)
4151                 goto err_free;
4152
4153         err = mlx5e_tc_add_fdb_flow(priv, flow, extack);
4154         complete_all(&flow->init_done);
4155         if (err) {
4156                 if (!(err == -ENETUNREACH && mlx5_lag_is_multipath(in_mdev)))
4157                         goto err_free;
4158
4159                 add_unready_flow(flow);
4160         }
4161
4162         return flow;
4163
4164 err_free:
4165         mlx5e_flow_put(priv, flow);
4166 out:
4167         return ERR_PTR(err);
4168 }
4169
4170 static int mlx5e_tc_add_fdb_peer_flow(struct flow_cls_offload *f,
4171                                       struct mlx5e_tc_flow *flow,
4172                                       unsigned long flow_flags)
4173 {
4174         struct mlx5e_priv *priv = flow->priv, *peer_priv;
4175         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch, *peer_esw;
4176         struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
4177         struct mlx5e_tc_flow_parse_attr *parse_attr;
4178         struct mlx5e_rep_priv *peer_urpriv;
4179         struct mlx5e_tc_flow *peer_flow;
4180         struct mlx5_core_dev *in_mdev;
4181         int err = 0;
4182
4183         peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4184         if (!peer_esw)
4185                 return -ENODEV;
4186
4187         peer_urpriv = mlx5_eswitch_get_uplink_priv(peer_esw, REP_ETH);
4188         peer_priv = netdev_priv(peer_urpriv->netdev);
4189
4190         /* in_mdev is assigned of which the packet originated from.
4191          * So packets redirected to uplink use the same mdev of the
4192          * original flow and packets redirected from uplink use the
4193          * peer mdev.
4194          */
4195         if (flow->esw_attr->in_rep->vport == MLX5_VPORT_UPLINK)
4196                 in_mdev = peer_priv->mdev;
4197         else
4198                 in_mdev = priv->mdev;
4199
4200         parse_attr = flow->esw_attr->parse_attr;
4201         peer_flow = __mlx5e_add_fdb_flow(peer_priv, f, flow_flags,
4202                                          parse_attr->filter_dev,
4203                                          flow->esw_attr->in_rep, in_mdev);
4204         if (IS_ERR(peer_flow)) {
4205                 err = PTR_ERR(peer_flow);
4206                 goto out;
4207         }
4208
4209         flow->peer_flow = peer_flow;
4210         flow_flag_set(flow, DUP);
4211         mutex_lock(&esw->offloads.peer_mutex);
4212         list_add_tail(&flow->peer, &esw->offloads.peer_flows);
4213         mutex_unlock(&esw->offloads.peer_mutex);
4214
4215 out:
4216         mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4217         return err;
4218 }
4219
4220 static int
4221 mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
4222                    struct flow_cls_offload *f,
4223                    unsigned long flow_flags,
4224                    struct net_device *filter_dev,
4225                    struct mlx5e_tc_flow **__flow)
4226 {
4227         struct mlx5e_rep_priv *rpriv = priv->ppriv;
4228         struct mlx5_eswitch_rep *in_rep = rpriv->rep;
4229         struct mlx5_core_dev *in_mdev = priv->mdev;
4230         struct mlx5e_tc_flow *flow;
4231         int err;
4232
4233         flow = __mlx5e_add_fdb_flow(priv, f, flow_flags, filter_dev, in_rep,
4234                                     in_mdev);
4235         if (IS_ERR(flow))
4236                 return PTR_ERR(flow);
4237
4238         if (is_peer_flow_needed(flow)) {
4239                 err = mlx5e_tc_add_fdb_peer_flow(f, flow, flow_flags);
4240                 if (err) {
4241                         mlx5e_tc_del_fdb_flow(priv, flow);
4242                         goto out;
4243                 }
4244         }
4245
4246         *__flow = flow;
4247
4248         return 0;
4249
4250 out:
4251         return err;
4252 }
4253
4254 static int
4255 mlx5e_add_nic_flow(struct mlx5e_priv *priv,
4256                    struct flow_cls_offload *f,
4257                    unsigned long flow_flags,
4258                    struct net_device *filter_dev,
4259                    struct mlx5e_tc_flow **__flow)
4260 {
4261         struct flow_rule *rule = flow_cls_offload_flow_rule(f);
4262         struct netlink_ext_ack *extack = f->common.extack;
4263         struct mlx5e_tc_flow_parse_attr *parse_attr;
4264         struct mlx5e_tc_flow *flow;
4265         int attr_size, err;
4266
4267         /* multi-chain not supported for NIC rules */
4268         if (!tc_cls_can_offload_and_chain0(priv->netdev, &f->common))
4269                 return -EOPNOTSUPP;
4270
4271         flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_NIC);
4272         attr_size  = sizeof(struct mlx5_nic_flow_attr);
4273         err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
4274                                &parse_attr, &flow);
4275         if (err)
4276                 goto out;
4277
4278         parse_attr->filter_dev = filter_dev;
4279         err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
4280                                f, filter_dev);
4281         if (err)
4282                 goto err_free;
4283
4284         err = parse_tc_nic_actions(priv, &rule->action, parse_attr, flow, extack);
4285         if (err)
4286                 goto err_free;
4287
4288         err = mlx5e_tc_add_nic_flow(priv, parse_attr, flow, extack);
4289         if (err)
4290                 goto err_free;
4291
4292         flow_flag_set(flow, OFFLOADED);
4293         kvfree(parse_attr);
4294         *__flow = flow;
4295
4296         return 0;
4297
4298 err_free:
4299         mlx5e_flow_put(priv, flow);
4300         kvfree(parse_attr);
4301 out:
4302         return err;
4303 }
4304
4305 static int
4306 mlx5e_tc_add_flow(struct mlx5e_priv *priv,
4307                   struct flow_cls_offload *f,
4308                   unsigned long flags,
4309                   struct net_device *filter_dev,
4310                   struct mlx5e_tc_flow **flow)
4311 {
4312         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
4313         unsigned long flow_flags;
4314         int err;
4315
4316         get_flags(flags, &flow_flags);
4317
4318         if (!tc_can_offload_extack(priv->netdev, f->common.extack))
4319                 return -EOPNOTSUPP;
4320
4321         if (esw && esw->mode == MLX5_ESWITCH_OFFLOADS)
4322                 err = mlx5e_add_fdb_flow(priv, f, flow_flags,
4323                                          filter_dev, flow);
4324         else
4325                 err = mlx5e_add_nic_flow(priv, f, flow_flags,
4326                                          filter_dev, flow);
4327
4328         return err;
4329 }
4330
4331 int mlx5e_configure_flower(struct net_device *dev, struct mlx5e_priv *priv,
4332                            struct flow_cls_offload *f, unsigned long flags)
4333 {
4334         struct netlink_ext_ack *extack = f->common.extack;
4335         struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4336         struct mlx5e_tc_flow *flow;
4337         int err = 0;
4338
4339         rcu_read_lock();
4340         flow = rhashtable_lookup(tc_ht, &f->cookie, tc_ht_params);
4341         rcu_read_unlock();
4342         if (flow) {
4343                 NL_SET_ERR_MSG_MOD(extack,
4344                                    "flow cookie already exists, ignoring");
4345                 netdev_warn_once(priv->netdev,
4346                                  "flow cookie %lx already exists, ignoring\n",
4347                                  f->cookie);
4348                 err = -EEXIST;
4349                 goto out;
4350         }
4351
4352         trace_mlx5e_configure_flower(f);
4353         err = mlx5e_tc_add_flow(priv, f, flags, dev, &flow);
4354         if (err)
4355                 goto out;
4356
4357         err = rhashtable_lookup_insert_fast(tc_ht, &flow->node, tc_ht_params);
4358         if (err)
4359                 goto err_free;
4360
4361         return 0;
4362
4363 err_free:
4364         mlx5e_flow_put(priv, flow);
4365 out:
4366         return err;
4367 }
4368
4369 static bool same_flow_direction(struct mlx5e_tc_flow *flow, int flags)
4370 {
4371         bool dir_ingress = !!(flags & MLX5_TC_FLAG(INGRESS));
4372         bool dir_egress = !!(flags & MLX5_TC_FLAG(EGRESS));
4373
4374         return flow_flag_test(flow, INGRESS) == dir_ingress &&
4375                 flow_flag_test(flow, EGRESS) == dir_egress;
4376 }
4377
4378 int mlx5e_delete_flower(struct net_device *dev, struct mlx5e_priv *priv,
4379                         struct flow_cls_offload *f, unsigned long flags)
4380 {
4381         struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4382         struct mlx5e_tc_flow *flow;
4383         int err;
4384
4385         rcu_read_lock();
4386         flow = rhashtable_lookup(tc_ht, &f->cookie, tc_ht_params);
4387         if (!flow || !same_flow_direction(flow, flags)) {
4388                 err = -EINVAL;
4389                 goto errout;
4390         }
4391
4392         /* Only delete the flow if it doesn't have MLX5E_TC_FLOW_DELETED flag
4393          * set.
4394          */
4395         if (flow_flag_test_and_set(flow, DELETED)) {
4396                 err = -EINVAL;
4397                 goto errout;
4398         }
4399         rhashtable_remove_fast(tc_ht, &flow->node, tc_ht_params);
4400         rcu_read_unlock();
4401
4402         trace_mlx5e_delete_flower(f);
4403         mlx5e_flow_put(priv, flow);
4404
4405         return 0;
4406
4407 errout:
4408         rcu_read_unlock();
4409         return err;
4410 }
4411
4412 int mlx5e_stats_flower(struct net_device *dev, struct mlx5e_priv *priv,
4413                        struct flow_cls_offload *f, unsigned long flags)
4414 {
4415         struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
4416         struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4417         struct mlx5_eswitch *peer_esw;
4418         struct mlx5e_tc_flow *flow;
4419         struct mlx5_fc *counter;
4420         u64 lastuse = 0;
4421         u64 packets = 0;
4422         u64 bytes = 0;
4423         int err = 0;
4424
4425         rcu_read_lock();
4426         flow = mlx5e_flow_get(rhashtable_lookup(tc_ht, &f->cookie,
4427                                                 tc_ht_params));
4428         rcu_read_unlock();
4429         if (IS_ERR(flow))
4430                 return PTR_ERR(flow);
4431
4432         if (!same_flow_direction(flow, flags)) {
4433                 err = -EINVAL;
4434                 goto errout;
4435         }
4436
4437         if (mlx5e_is_offloaded_flow(flow) || flow_flag_test(flow, CT)) {
4438                 counter = mlx5e_tc_get_counter(flow);
4439                 if (!counter)
4440                         goto errout;
4441
4442                 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
4443         }
4444
4445         /* Under multipath it's possible for one rule to be currently
4446          * un-offloaded while the other rule is offloaded.
4447          */
4448         peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4449         if (!peer_esw)
4450                 goto out;
4451
4452         if (flow_flag_test(flow, DUP) &&
4453             flow_flag_test(flow->peer_flow, OFFLOADED)) {
4454                 u64 bytes2;
4455                 u64 packets2;
4456                 u64 lastuse2;
4457
4458                 counter = mlx5e_tc_get_counter(flow->peer_flow);
4459                 if (!counter)
4460                         goto no_peer_counter;
4461                 mlx5_fc_query_cached(counter, &bytes2, &packets2, &lastuse2);
4462
4463                 bytes += bytes2;
4464                 packets += packets2;
4465                 lastuse = max_t(u64, lastuse, lastuse2);
4466         }
4467
4468 no_peer_counter:
4469         mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4470 out:
4471         flow_stats_update(&f->stats, bytes, packets, lastuse,
4472                           FLOW_ACTION_HW_STATS_DELAYED);
4473         trace_mlx5e_stats_flower(f);
4474 errout:
4475         mlx5e_flow_put(priv, flow);
4476         return err;
4477 }
4478
4479 static int apply_police_params(struct mlx5e_priv *priv, u32 rate,
4480                                struct netlink_ext_ack *extack)
4481 {
4482         struct mlx5e_rep_priv *rpriv = priv->ppriv;
4483         struct mlx5_eswitch *esw;
4484         u16 vport_num;
4485         u32 rate_mbps;
4486         int err;
4487
4488         vport_num = rpriv->rep->vport;
4489         if (vport_num >= MLX5_VPORT_ECPF) {
4490                 NL_SET_ERR_MSG_MOD(extack,
4491                                    "Ingress rate limit is supported only for Eswitch ports connected to VFs");
4492                 return -EOPNOTSUPP;
4493         }
4494
4495         esw = priv->mdev->priv.eswitch;
4496         /* rate is given in bytes/sec.
4497          * First convert to bits/sec and then round to the nearest mbit/secs.
4498          * mbit means million bits.
4499          * Moreover, if rate is non zero we choose to configure to a minimum of
4500          * 1 mbit/sec.
4501          */
4502         rate_mbps = rate ? max_t(u32, (rate * 8 + 500000) / 1000000, 1) : 0;
4503         err = mlx5_esw_modify_vport_rate(esw, vport_num, rate_mbps);
4504         if (err)
4505                 NL_SET_ERR_MSG_MOD(extack, "failed applying action to hardware");
4506
4507         return err;
4508 }
4509
4510 static int scan_tc_matchall_fdb_actions(struct mlx5e_priv *priv,
4511                                         struct flow_action *flow_action,
4512                                         struct netlink_ext_ack *extack)
4513 {
4514         struct mlx5e_rep_priv *rpriv = priv->ppriv;
4515         const struct flow_action_entry *act;
4516         int err;
4517         int i;
4518
4519         if (!flow_action_has_entries(flow_action)) {
4520                 NL_SET_ERR_MSG_MOD(extack, "matchall called with no action");
4521                 return -EINVAL;
4522         }
4523
4524         if (!flow_offload_has_one_action(flow_action)) {
4525                 NL_SET_ERR_MSG_MOD(extack, "matchall policing support only a single action");
4526                 return -EOPNOTSUPP;
4527         }
4528
4529         if (!flow_action_basic_hw_stats_check(flow_action, extack))
4530                 return -EOPNOTSUPP;
4531
4532         flow_action_for_each(i, act, flow_action) {
4533                 switch (act->id) {
4534                 case FLOW_ACTION_POLICE:
4535                         err = apply_police_params(priv, act->police.rate_bytes_ps, extack);
4536                         if (err)
4537                                 return err;
4538
4539                         rpriv->prev_vf_vport_stats = priv->stats.vf_vport;
4540                         break;
4541                 default:
4542                         NL_SET_ERR_MSG_MOD(extack, "mlx5 supports only police action for matchall");
4543                         return -EOPNOTSUPP;
4544                 }
4545         }
4546
4547         return 0;
4548 }
4549
4550 int mlx5e_tc_configure_matchall(struct mlx5e_priv *priv,
4551                                 struct tc_cls_matchall_offload *ma)
4552 {
4553         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
4554         struct netlink_ext_ack *extack = ma->common.extack;
4555
4556         if (!mlx5_esw_qos_enabled(esw)) {
4557                 NL_SET_ERR_MSG_MOD(extack, "QoS is not supported on this device");
4558                 return -EOPNOTSUPP;
4559         }
4560
4561         if (ma->common.prio != 1) {
4562                 NL_SET_ERR_MSG_MOD(extack, "only priority 1 is supported");
4563                 return -EINVAL;
4564         }
4565
4566         return scan_tc_matchall_fdb_actions(priv, &ma->rule->action, extack);
4567 }
4568
4569 int mlx5e_tc_delete_matchall(struct mlx5e_priv *priv,
4570                              struct tc_cls_matchall_offload *ma)
4571 {
4572         struct netlink_ext_ack *extack = ma->common.extack;
4573
4574         return apply_police_params(priv, 0, extack);
4575 }
4576
4577 void mlx5e_tc_stats_matchall(struct mlx5e_priv *priv,
4578                              struct tc_cls_matchall_offload *ma)
4579 {
4580         struct mlx5e_rep_priv *rpriv = priv->ppriv;
4581         struct rtnl_link_stats64 cur_stats;
4582         u64 dbytes;
4583         u64 dpkts;
4584
4585         cur_stats = priv->stats.vf_vport;
4586         dpkts = cur_stats.rx_packets - rpriv->prev_vf_vport_stats.rx_packets;
4587         dbytes = cur_stats.rx_bytes - rpriv->prev_vf_vport_stats.rx_bytes;
4588         rpriv->prev_vf_vport_stats = cur_stats;
4589         flow_stats_update(&ma->stats, dpkts, dbytes, jiffies,
4590                           FLOW_ACTION_HW_STATS_DELAYED);
4591 }
4592
4593 static void mlx5e_tc_hairpin_update_dead_peer(struct mlx5e_priv *priv,
4594                                               struct mlx5e_priv *peer_priv)
4595 {
4596         struct mlx5_core_dev *peer_mdev = peer_priv->mdev;
4597         struct mlx5e_hairpin_entry *hpe, *tmp;
4598         LIST_HEAD(init_wait_list);
4599         u16 peer_vhca_id;
4600         int bkt;
4601
4602         if (!same_hw_devs(priv, peer_priv))
4603                 return;
4604
4605         peer_vhca_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
4606
4607         mutex_lock(&priv->fs.tc.hairpin_tbl_lock);
4608         hash_for_each(priv->fs.tc.hairpin_tbl, bkt, hpe, hairpin_hlist)
4609                 if (refcount_inc_not_zero(&hpe->refcnt))
4610                         list_add(&hpe->dead_peer_wait_list, &init_wait_list);
4611         mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
4612
4613         list_for_each_entry_safe(hpe, tmp, &init_wait_list, dead_peer_wait_list) {
4614                 wait_for_completion(&hpe->res_ready);
4615                 if (!IS_ERR_OR_NULL(hpe->hp) && hpe->peer_vhca_id == peer_vhca_id)
4616                         hpe->hp->pair->peer_gone = true;
4617
4618                 mlx5e_hairpin_put(priv, hpe);
4619         }
4620 }
4621
4622 static int mlx5e_tc_netdev_event(struct notifier_block *this,
4623                                  unsigned long event, void *ptr)
4624 {
4625         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4626         struct mlx5e_flow_steering *fs;
4627         struct mlx5e_priv *peer_priv;
4628         struct mlx5e_tc_table *tc;
4629         struct mlx5e_priv *priv;
4630
4631         if (ndev->netdev_ops != &mlx5e_netdev_ops ||
4632             event != NETDEV_UNREGISTER ||
4633             ndev->reg_state == NETREG_REGISTERED)
4634                 return NOTIFY_DONE;
4635
4636         tc = container_of(this, struct mlx5e_tc_table, netdevice_nb);
4637         fs = container_of(tc, struct mlx5e_flow_steering, tc);
4638         priv = container_of(fs, struct mlx5e_priv, fs);
4639         peer_priv = netdev_priv(ndev);
4640         if (priv == peer_priv ||
4641             !(priv->netdev->features & NETIF_F_HW_TC))
4642                 return NOTIFY_DONE;
4643
4644         mlx5e_tc_hairpin_update_dead_peer(priv, peer_priv);
4645
4646         return NOTIFY_DONE;
4647 }
4648
4649 int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
4650 {
4651         struct mlx5e_tc_table *tc = &priv->fs.tc;
4652         int err;
4653
4654         mutex_init(&tc->t_lock);
4655         mutex_init(&tc->mod_hdr.lock);
4656         hash_init(tc->mod_hdr.hlist);
4657         mutex_init(&tc->hairpin_tbl_lock);
4658         hash_init(tc->hairpin_tbl);
4659
4660         err = rhashtable_init(&tc->ht, &tc_ht_params);
4661         if (err)
4662                 return err;
4663
4664         tc->netdevice_nb.notifier_call = mlx5e_tc_netdev_event;
4665         err = register_netdevice_notifier_dev_net(priv->netdev,
4666                                                   &tc->netdevice_nb,
4667                                                   &tc->netdevice_nn);
4668         if (err) {
4669                 tc->netdevice_nb.notifier_call = NULL;
4670                 mlx5_core_warn(priv->mdev, "Failed to register netdev notifier\n");
4671         }
4672
4673         return err;
4674 }
4675
4676 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
4677 {
4678         struct mlx5e_tc_flow *flow = ptr;
4679         struct mlx5e_priv *priv = flow->priv;
4680
4681         mlx5e_tc_del_flow(priv, flow);
4682         kfree(flow);
4683 }
4684
4685 void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
4686 {
4687         struct mlx5e_tc_table *tc = &priv->fs.tc;
4688
4689         if (tc->netdevice_nb.notifier_call)
4690                 unregister_netdevice_notifier_dev_net(priv->netdev,
4691                                                       &tc->netdevice_nb,
4692                                                       &tc->netdevice_nn);
4693
4694         mutex_destroy(&tc->mod_hdr.lock);
4695         mutex_destroy(&tc->hairpin_tbl_lock);
4696
4697         rhashtable_destroy(&tc->ht);
4698
4699         if (!IS_ERR_OR_NULL(tc->t)) {
4700                 mlx5_destroy_flow_table(tc->t);
4701                 tc->t = NULL;
4702         }
4703         mutex_destroy(&tc->t_lock);
4704 }
4705
4706 int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
4707 {
4708         const size_t sz_enc_opts = sizeof(struct flow_dissector_key_enc_opts);
4709         struct mlx5_rep_uplink_priv *uplink_priv;
4710         struct mlx5e_rep_priv *priv;
4711         struct mapping_ctx *mapping;
4712         int err;
4713
4714         uplink_priv = container_of(tc_ht, struct mlx5_rep_uplink_priv, tc_ht);
4715         priv = container_of(uplink_priv, struct mlx5e_rep_priv, uplink_priv);
4716
4717         err = mlx5_tc_ct_init(uplink_priv);
4718         if (err)
4719                 goto err_ct;
4720
4721         mapping = mapping_create(sizeof(struct tunnel_match_key),
4722                                  TUNNEL_INFO_BITS_MASK, true);
4723         if (IS_ERR(mapping)) {
4724                 err = PTR_ERR(mapping);
4725                 goto err_tun_mapping;
4726         }
4727         uplink_priv->tunnel_mapping = mapping;
4728
4729         mapping = mapping_create(sz_enc_opts, ENC_OPTS_BITS_MASK, true);
4730         if (IS_ERR(mapping)) {
4731                 err = PTR_ERR(mapping);
4732                 goto err_enc_opts_mapping;
4733         }
4734         uplink_priv->tunnel_enc_opts_mapping = mapping;
4735
4736         err = rhashtable_init(tc_ht, &tc_ht_params);
4737         if (err)
4738                 goto err_ht_init;
4739
4740         return err;
4741
4742 err_ht_init:
4743         mapping_destroy(uplink_priv->tunnel_enc_opts_mapping);
4744 err_enc_opts_mapping:
4745         mapping_destroy(uplink_priv->tunnel_mapping);
4746 err_tun_mapping:
4747         mlx5_tc_ct_clean(uplink_priv);
4748 err_ct:
4749         netdev_warn(priv->netdev,
4750                     "Failed to initialize tc (eswitch), err: %d", err);
4751         return err;
4752 }
4753
4754 void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
4755 {
4756         struct mlx5_rep_uplink_priv *uplink_priv;
4757
4758         rhashtable_free_and_destroy(tc_ht, _mlx5e_tc_del_flow, NULL);
4759
4760         uplink_priv = container_of(tc_ht, struct mlx5_rep_uplink_priv, tc_ht);
4761         mapping_destroy(uplink_priv->tunnel_enc_opts_mapping);
4762         mapping_destroy(uplink_priv->tunnel_mapping);
4763
4764         mlx5_tc_ct_clean(uplink_priv);
4765 }
4766
4767 int mlx5e_tc_num_filters(struct mlx5e_priv *priv, unsigned long flags)
4768 {
4769         struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4770
4771         return atomic_read(&tc_ht->nelems);
4772 }
4773
4774 void mlx5e_tc_clean_fdb_peer_flows(struct mlx5_eswitch *esw)
4775 {
4776         struct mlx5e_tc_flow *flow, *tmp;
4777
4778         list_for_each_entry_safe(flow, tmp, &esw->offloads.peer_flows, peer)
4779                 __mlx5e_tc_del_fdb_peer_flow(flow);
4780 }
4781
4782 void mlx5e_tc_reoffload_flows_work(struct work_struct *work)
4783 {
4784         struct mlx5_rep_uplink_priv *rpriv =
4785                 container_of(work, struct mlx5_rep_uplink_priv,
4786                              reoffload_flows_work);
4787         struct mlx5e_tc_flow *flow, *tmp;
4788
4789         mutex_lock(&rpriv->unready_flows_lock);
4790         list_for_each_entry_safe(flow, tmp, &rpriv->unready_flows, unready) {
4791                 if (!mlx5e_tc_add_fdb_flow(flow->priv, flow, NULL))
4792                         unready_flow_del(flow);
4793         }
4794         mutex_unlock(&rpriv->unready_flows_lock);
4795 }
4796
4797 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4798 static bool mlx5e_restore_tunnel(struct mlx5e_priv *priv, struct sk_buff *skb,
4799                                  struct mlx5e_tc_update_priv *tc_priv,
4800                                  u32 tunnel_id)
4801 {
4802         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
4803         struct flow_dissector_key_enc_opts enc_opts = {};
4804         struct mlx5_rep_uplink_priv *uplink_priv;
4805         struct mlx5e_rep_priv *uplink_rpriv;
4806         struct metadata_dst *tun_dst;
4807         struct tunnel_match_key key;
4808         u32 tun_id, enc_opts_id;
4809         struct net_device *dev;
4810         int err;
4811
4812         enc_opts_id = tunnel_id & ENC_OPTS_BITS_MASK;
4813         tun_id = tunnel_id >> ENC_OPTS_BITS;
4814
4815         if (!tun_id)
4816                 return true;
4817
4818         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
4819         uplink_priv = &uplink_rpriv->uplink_priv;
4820
4821         err = mapping_find(uplink_priv->tunnel_mapping, tun_id, &key);
4822         if (err) {
4823                 WARN_ON_ONCE(true);
4824                 netdev_dbg(priv->netdev,
4825                            "Couldn't find tunnel for tun_id: %d, err: %d\n",
4826                            tun_id, err);
4827                 return false;
4828         }
4829
4830         if (enc_opts_id) {
4831                 err = mapping_find(uplink_priv->tunnel_enc_opts_mapping,
4832                                    enc_opts_id, &enc_opts);
4833                 if (err) {
4834                         netdev_dbg(priv->netdev,
4835                                    "Couldn't find tunnel (opts) for tun_id: %d, err: %d\n",
4836                                    enc_opts_id, err);
4837                         return false;
4838                 }
4839         }
4840
4841         tun_dst = tun_rx_dst(enc_opts.len);
4842         if (!tun_dst) {
4843                 WARN_ON_ONCE(true);
4844                 return false;
4845         }
4846
4847         ip_tunnel_key_init(&tun_dst->u.tun_info.key,
4848                            key.enc_ipv4.src, key.enc_ipv4.dst,
4849                            key.enc_ip.tos, key.enc_ip.ttl,
4850                            0, /* label */
4851                            key.enc_tp.src, key.enc_tp.dst,
4852                            key32_to_tunnel_id(key.enc_key_id.keyid),
4853                            TUNNEL_KEY);
4854
4855         if (enc_opts.len)
4856                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info, enc_opts.data,
4857                                         enc_opts.len, enc_opts.dst_opt_type);
4858
4859         skb_dst_set(skb, (struct dst_entry *)tun_dst);
4860         dev = dev_get_by_index(&init_net, key.filter_ifindex);
4861         if (!dev) {
4862                 netdev_dbg(priv->netdev,
4863                            "Couldn't find tunnel device with ifindex: %d\n",
4864                            key.filter_ifindex);
4865                 return false;
4866         }
4867
4868         /* Set tun_dev so we do dev_put() after datapath */
4869         tc_priv->tun_dev = dev;
4870
4871         skb->dev = dev;
4872
4873         return true;
4874 }
4875 #endif /* CONFIG_NET_TC_SKB_EXT */
4876
4877 bool mlx5e_tc_rep_update_skb(struct mlx5_cqe64 *cqe,
4878                              struct sk_buff *skb,
4879                              struct mlx5e_tc_update_priv *tc_priv)
4880 {
4881 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4882         u32 chain = 0, reg_c0, reg_c1, tunnel_id, tuple_id;
4883         struct mlx5_rep_uplink_priv *uplink_priv;
4884         struct mlx5e_rep_priv *uplink_rpriv;
4885         struct tc_skb_ext *tc_skb_ext;
4886         struct mlx5_eswitch *esw;
4887         struct mlx5e_priv *priv;
4888         int tunnel_moffset;
4889         int err;
4890
4891         reg_c0 = (be32_to_cpu(cqe->sop_drop_qpn) & MLX5E_TC_FLOW_ID_MASK);
4892         if (reg_c0 == MLX5_FS_DEFAULT_FLOW_TAG)
4893                 reg_c0 = 0;
4894         reg_c1 = be32_to_cpu(cqe->imm_inval_pkey);
4895
4896         if (!reg_c0)
4897                 return true;
4898
4899         priv = netdev_priv(skb->dev);
4900         esw = priv->mdev->priv.eswitch;
4901
4902         err = mlx5_eswitch_get_chain_for_tag(esw, reg_c0, &chain);
4903         if (err) {
4904                 netdev_dbg(priv->netdev,
4905                            "Couldn't find chain for chain tag: %d, err: %d\n",
4906                            reg_c0, err);
4907                 return false;
4908         }
4909
4910         if (chain) {
4911                 tc_skb_ext = skb_ext_add(skb, TC_SKB_EXT);
4912                 if (!tc_skb_ext) {
4913                         WARN_ON(1);
4914                         return false;
4915                 }
4916
4917                 tc_skb_ext->chain = chain;
4918
4919                 tuple_id = reg_c1 & TUPLE_ID_MAX;
4920
4921                 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
4922                 uplink_priv = &uplink_rpriv->uplink_priv;
4923                 if (!mlx5e_tc_ct_restore_flow(uplink_priv, skb, tuple_id))
4924                         return false;
4925         }
4926
4927         tunnel_moffset = mlx5e_tc_attr_to_reg_mappings[TUNNEL_TO_REG].moffset;
4928         tunnel_id = reg_c1 >> (8 * tunnel_moffset);
4929         return mlx5e_restore_tunnel(priv, skb, tc_priv, tunnel_id);
4930 #endif /* CONFIG_NET_TC_SKB_EXT */
4931
4932         return true;
4933 }
4934
4935 void mlx5_tc_rep_post_napi_receive(struct mlx5e_tc_update_priv *tc_priv)
4936 {
4937         if (tc_priv->tun_dev)
4938                 dev_put(tc_priv->tun_dev);
4939 }