drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c

   1 /*
   2  * drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
   3  * Copyright (c) 2016 Mellanox Technologies. All rights reserved.
   4  * Copyright (c) 2016 Jiri Pirko <jiri@mellanox.com>
   5  * Copyright (c) 2016 Ido Schimmel <idosch@mellanox.com>
   6  * Copyright (c) 2016 Yotam Gigi <yotamg@mellanox.com>
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the names of the copyright holders nor the names of its
  17  *    contributors may be used to endorse or promote products derived from
  18  *    this software without specific prior written permission.
  19  *
  20  * Alternatively, this software may be distributed under the terms of the
  21  * GNU General Public License ("GPL") version 2 as published by the Free
  22  * Software Foundation.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  25  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
  28  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  31  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  32  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  33  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  34  * POSSIBILITY OF SUCH DAMAGE.
  35  */
  36
  37 #include <linux/kernel.h>
  38 #include <linux/types.h>
  39 #include <linux/rhashtable.h>
  40 #include <linux/bitops.h>
  41 #include <linux/in6.h>
  42 #include <linux/notifier.h>
  43 #include <net/netevent.h>
  44 #include <net/neighbour.h>
  45 #include <net/arp.h>
  46 #include <net/ip_fib.h>
  47
  48 #include "spectrum.h"
  49 #include "core.h"
  50 #include "reg.h"
  51
  52 #define mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) \
  53         for_each_set_bit(prefix, (prefix_usage)->b, MLXSW_SP_PREFIX_COUNT)
  54
  55 static bool
  56 mlxsw_sp_prefix_usage_subset(struct mlxsw_sp_prefix_usage *prefix_usage1,
  57                              struct mlxsw_sp_prefix_usage *prefix_usage2)
  58 {
  59         unsigned char prefix;
  60
  61         mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage1) {
  62                 if (!test_bit(prefix, prefix_usage2->b))
  63                         return false;
  64         }
  65         return true;
  66 }
  67
  68 static bool
  69 mlxsw_sp_prefix_usage_eq(struct mlxsw_sp_prefix_usage *prefix_usage1,
  70                          struct mlxsw_sp_prefix_usage *prefix_usage2)
  71 {
  72         return !memcmp(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
  73 }
  74
  75 static bool
  76 mlxsw_sp_prefix_usage_none(struct mlxsw_sp_prefix_usage *prefix_usage)
  77 {
  78         struct mlxsw_sp_prefix_usage prefix_usage_none = {{ 0 } };
  79
  80         return mlxsw_sp_prefix_usage_eq(prefix_usage, &prefix_usage_none);
  81 }
  82
  83 static void
  84 mlxsw_sp_prefix_usage_cpy(struct mlxsw_sp_prefix_usage *prefix_usage1,
  85                           struct mlxsw_sp_prefix_usage *prefix_usage2)
  86 {
  87         memcpy(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
  88 }
  89
  90 static void
  91 mlxsw_sp_prefix_usage_zero(struct mlxsw_sp_prefix_usage *prefix_usage)
  92 {
  93         memset(prefix_usage, 0, sizeof(*prefix_usage));
  94 }
  95
  96 static void
  97 mlxsw_sp_prefix_usage_set(struct mlxsw_sp_prefix_usage *prefix_usage,
  98                           unsigned char prefix_len)
  99 {
 100         set_bit(prefix_len, prefix_usage->b);
 101 }
 102
 103 static void
 104 mlxsw_sp_prefix_usage_clear(struct mlxsw_sp_prefix_usage *prefix_usage,
 105                             unsigned char prefix_len)
 106 {
 107         clear_bit(prefix_len, prefix_usage->b);
 108 }
 109
 110 struct mlxsw_sp_fib_key {
 111         struct net_device *dev;
 112         unsigned char addr[sizeof(struct in6_addr)];
 113         unsigned char prefix_len;
 114 };
 115
 116 enum mlxsw_sp_fib_entry_type {
 117         MLXSW_SP_FIB_ENTRY_TYPE_REMOTE,
 118         MLXSW_SP_FIB_ENTRY_TYPE_LOCAL,
 119         MLXSW_SP_FIB_ENTRY_TYPE_TRAP,
 120 };
 121
 122 struct mlxsw_sp_nexthop_group;
 123
 124 struct mlxsw_sp_fib_entry {
 125         struct rhash_head ht_node;
 126         struct list_head list;
 127         struct mlxsw_sp_fib_key key;
 128         enum mlxsw_sp_fib_entry_type type;
 129         unsigned int ref_count;
 130         u16 rif; /* used for action local */
 131         struct mlxsw_sp_vr *vr;
 132         struct fib_info *fi;
 133         struct list_head nexthop_group_node;
 134         struct mlxsw_sp_nexthop_group *nh_group;
 135 };
 136
 137 struct mlxsw_sp_fib {
 138         struct rhashtable ht;
 139         struct list_head entry_list;
 140         unsigned long prefix_ref_count[MLXSW_SP_PREFIX_COUNT];
 141         struct mlxsw_sp_prefix_usage prefix_usage;
 142 };
 143
 144 static const struct rhashtable_params mlxsw_sp_fib_ht_params = {
 145         .key_offset = offsetof(struct mlxsw_sp_fib_entry, key),
 146         .head_offset = offsetof(struct mlxsw_sp_fib_entry, ht_node),
 147         .key_len = sizeof(struct mlxsw_sp_fib_key),
 148         .automatic_shrinking = true,
 149 };
 150
 151 static int mlxsw_sp_fib_entry_insert(struct mlxsw_sp_fib *fib,
 152                                      struct mlxsw_sp_fib_entry *fib_entry)
 153 {
 154         unsigned char prefix_len = fib_entry->key.prefix_len;
 155         int err;
 156
 157         err = rhashtable_insert_fast(&fib->ht, &fib_entry->ht_node,
 158                                      mlxsw_sp_fib_ht_params);
 159         if (err)
 160                 return err;
 161         list_add_tail(&fib_entry->list, &fib->entry_list);
 162         if (fib->prefix_ref_count[prefix_len]++ == 0)
 163                 mlxsw_sp_prefix_usage_set(&fib->prefix_usage, prefix_len);
 164         return 0;
 165 }
 166
 167 static void mlxsw_sp_fib_entry_remove(struct mlxsw_sp_fib *fib,
 168                                       struct mlxsw_sp_fib_entry *fib_entry)
 169 {
 170         unsigned char prefix_len = fib_entry->key.prefix_len;
 171
 172         if (--fib->prefix_ref_count[prefix_len] == 0)
 173                 mlxsw_sp_prefix_usage_clear(&fib->prefix_usage, prefix_len);
 174         list_del(&fib_entry->list);
 175         rhashtable_remove_fast(&fib->ht, &fib_entry->ht_node,
 176                                mlxsw_sp_fib_ht_params);
 177 }
 178
 179 static struct mlxsw_sp_fib_entry *
 180 mlxsw_sp_fib_entry_create(struct mlxsw_sp_fib *fib, const void *addr,
 181                           size_t addr_len, unsigned char prefix_len,
 182                           struct net_device *dev)
 183 {
 184         struct mlxsw_sp_fib_entry *fib_entry;
 185
 186         fib_entry = kzalloc(sizeof(*fib_entry), GFP_KERNEL);
 187         if (!fib_entry)
 188                 return NULL;
 189         fib_entry->key.dev = dev;
 190         memcpy(fib_entry->key.addr, addr, addr_len);
 191         fib_entry->key.prefix_len = prefix_len;
 192         return fib_entry;
 193 }
 194
 195 static void mlxsw_sp_fib_entry_destroy(struct mlxsw_sp_fib_entry *fib_entry)
 196 {
 197         kfree(fib_entry);
 198 }
 199
 200 static struct mlxsw_sp_fib_entry *
 201 mlxsw_sp_fib_entry_lookup(struct mlxsw_sp_fib *fib, const void *addr,
 202                           size_t addr_len, unsigned char prefix_len,
 203                           struct net_device *dev)
 204 {
 205         struct mlxsw_sp_fib_key key;
 206
 207         memset(&key, 0, sizeof(key));
 208         key.dev = dev;
 209         memcpy(key.addr, addr, addr_len);
 210         key.prefix_len = prefix_len;
 211         return rhashtable_lookup_fast(&fib->ht, &key, mlxsw_sp_fib_ht_params);
 212 }
 213
 214 static struct mlxsw_sp_fib *mlxsw_sp_fib_create(void)
 215 {
 216         struct mlxsw_sp_fib *fib;
 217         int err;
 218
 219         fib = kzalloc(sizeof(*fib), GFP_KERNEL);
 220         if (!fib)
 221                 return ERR_PTR(-ENOMEM);
 222         err = rhashtable_init(&fib->ht, &mlxsw_sp_fib_ht_params);
 223         if (err)
 224                 goto err_rhashtable_init;
 225         INIT_LIST_HEAD(&fib->entry_list);
 226         return fib;
 227
 228 err_rhashtable_init:
 229         kfree(fib);
 230         return ERR_PTR(err);
 231 }
 232
 233 static void mlxsw_sp_fib_destroy(struct mlxsw_sp_fib *fib)
 234 {
 235         rhashtable_destroy(&fib->ht);
 236         kfree(fib);
 237 }
 238
 239 static struct mlxsw_sp_lpm_tree *
 240 mlxsw_sp_lpm_tree_find_unused(struct mlxsw_sp *mlxsw_sp, bool one_reserved)
 241 {
 242         static struct mlxsw_sp_lpm_tree *lpm_tree;
 243         int i;
 244
 245         for (i = 0; i < MLXSW_SP_LPM_TREE_COUNT; i++) {
 246                 lpm_tree = &mlxsw_sp->router.lpm_trees[i];
 247                 if (lpm_tree->ref_count == 0) {
 248                         if (one_reserved)
 249                                 one_reserved = false;
 250                         else
 251                                 return lpm_tree;
 252                 }
 253         }
 254         return NULL;
 255 }
 256
 257 static int mlxsw_sp_lpm_tree_alloc(struct mlxsw_sp *mlxsw_sp,
 258                                    struct mlxsw_sp_lpm_tree *lpm_tree)
 259 {
 260         char ralta_pl[MLXSW_REG_RALTA_LEN];
 261
 262         mlxsw_reg_ralta_pack(ralta_pl, true,
 263                              (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
 264                              lpm_tree->id);
 265         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
 266 }
 267
 268 static int mlxsw_sp_lpm_tree_free(struct mlxsw_sp *mlxsw_sp,
 269                                   struct mlxsw_sp_lpm_tree *lpm_tree)
 270 {
 271         char ralta_pl[MLXSW_REG_RALTA_LEN];
 272
 273         mlxsw_reg_ralta_pack(ralta_pl, false,
 274                              (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
 275                              lpm_tree->id);
 276         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
 277 }
 278
 279 static int
 280 mlxsw_sp_lpm_tree_left_struct_set(struct mlxsw_sp *mlxsw_sp,
 281                                   struct mlxsw_sp_prefix_usage *prefix_usage,
 282                                   struct mlxsw_sp_lpm_tree *lpm_tree)
 283 {
 284         char ralst_pl[MLXSW_REG_RALST_LEN];
 285         u8 root_bin = 0;
 286         u8 prefix;
 287         u8 last_prefix = MLXSW_REG_RALST_BIN_NO_CHILD;
 288
 289         mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage)
 290                 root_bin = prefix;
 291
 292         mlxsw_reg_ralst_pack(ralst_pl, root_bin, lpm_tree->id);
 293         mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) {
 294                 if (prefix == 0)
 295                         continue;
 296                 mlxsw_reg_ralst_bin_pack(ralst_pl, prefix, last_prefix,
 297                                          MLXSW_REG_RALST_BIN_NO_CHILD);
 298                 last_prefix = prefix;
 299         }
 300         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
 301 }
 302
 303 static struct mlxsw_sp_lpm_tree *
 304 mlxsw_sp_lpm_tree_create(struct mlxsw_sp *mlxsw_sp,
 305                          struct mlxsw_sp_prefix_usage *prefix_usage,
 306                          enum mlxsw_sp_l3proto proto, bool one_reserved)
 307 {
 308         struct mlxsw_sp_lpm_tree *lpm_tree;
 309         int err;
 310
 311         lpm_tree = mlxsw_sp_lpm_tree_find_unused(mlxsw_sp, one_reserved);
 312         if (!lpm_tree)
 313                 return ERR_PTR(-EBUSY);
 314         lpm_tree->proto = proto;
 315         err = mlxsw_sp_lpm_tree_alloc(mlxsw_sp, lpm_tree);
 316         if (err)
 317                 return ERR_PTR(err);
 318
 319         err = mlxsw_sp_lpm_tree_left_struct_set(mlxsw_sp, prefix_usage,
 320                                                 lpm_tree);
 321         if (err)
 322                 goto err_left_struct_set;
 323         memcpy(&lpm_tree->prefix_usage, prefix_usage,
 324                sizeof(lpm_tree->prefix_usage));
 325         return lpm_tree;
 326
 327 err_left_struct_set:
 328         mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
 329         return ERR_PTR(err);
 330 }
 331
 332 static int mlxsw_sp_lpm_tree_destroy(struct mlxsw_sp *mlxsw_sp,
 333                                      struct mlxsw_sp_lpm_tree *lpm_tree)
 334 {
 335         return mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
 336 }
 337
 338 static struct mlxsw_sp_lpm_tree *
 339 mlxsw_sp_lpm_tree_get(struct mlxsw_sp *mlxsw_sp,
 340                       struct mlxsw_sp_prefix_usage *prefix_usage,
 341                       enum mlxsw_sp_l3proto proto, bool one_reserved)
 342 {
 343         struct mlxsw_sp_lpm_tree *lpm_tree;
 344         int i;
 345
 346         for (i = 0; i < MLXSW_SP_LPM_TREE_COUNT; i++) {
 347                 lpm_tree = &mlxsw_sp->router.lpm_trees[i];
 348                 if (lpm_tree->ref_count != 0 &&
 349                     lpm_tree->proto == proto &&
 350                     mlxsw_sp_prefix_usage_eq(&lpm_tree->prefix_usage,
 351                                              prefix_usage))
 352                         goto inc_ref_count;
 353         }
 354         lpm_tree = mlxsw_sp_lpm_tree_create(mlxsw_sp, prefix_usage,
 355                                             proto, one_reserved);
 356         if (IS_ERR(lpm_tree))
 357                 return lpm_tree;
 358
 359 inc_ref_count:
 360         lpm_tree->ref_count++;
 361         return lpm_tree;
 362 }
 363
 364 static int mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
 365                                  struct mlxsw_sp_lpm_tree *lpm_tree)
 366 {
 367         if (--lpm_tree->ref_count == 0)
 368                 return mlxsw_sp_lpm_tree_destroy(mlxsw_sp, lpm_tree);
 369         return 0;
 370 }
 371
 372 static void mlxsw_sp_lpm_init(struct mlxsw_sp *mlxsw_sp)
 373 {
 374         struct mlxsw_sp_lpm_tree *lpm_tree;
 375         int i;
 376
 377         for (i = 0; i < MLXSW_SP_LPM_TREE_COUNT; i++) {
 378                 lpm_tree = &mlxsw_sp->router.lpm_trees[i];
 379                 lpm_tree->id = i + MLXSW_SP_LPM_TREE_MIN;
 380         }
 381 }
 382
 383 static struct mlxsw_sp_vr *mlxsw_sp_vr_find_unused(struct mlxsw_sp *mlxsw_sp)
 384 {
 385         struct mlxsw_sp_vr *vr;
 386         int i;
 387
 388         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
 389                 vr = &mlxsw_sp->router.vrs[i];
 390                 if (!vr->used)
 391                         return vr;
 392         }
 393         return NULL;
 394 }
 395
 396 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
 397                                      struct mlxsw_sp_vr *vr)
 398 {
 399         char raltb_pl[MLXSW_REG_RALTB_LEN];
 400
 401         mlxsw_reg_raltb_pack(raltb_pl, vr->id,
 402                              (enum mlxsw_reg_ralxx_protocol) vr->proto,
 403                              vr->lpm_tree->id);
 404         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
 405 }
 406
 407 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
 408                                        struct mlxsw_sp_vr *vr)
 409 {
 410         char raltb_pl[MLXSW_REG_RALTB_LEN];
 411
 412         /* Bind to tree 0 which is default */
 413         mlxsw_reg_raltb_pack(raltb_pl, vr->id,
 414                              (enum mlxsw_reg_ralxx_protocol) vr->proto, 0);
 415         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
 416 }
 417
 418 static u32 mlxsw_sp_fix_tb_id(u32 tb_id)
 419 {
 420         /* For our purpose, squash main and local table into one */
 421         if (tb_id == RT_TABLE_LOCAL)
 422                 tb_id = RT_TABLE_MAIN;
 423         return tb_id;
 424 }
 425
 426 static struct mlxsw_sp_vr *mlxsw_sp_vr_find(struct mlxsw_sp *mlxsw_sp,
 427                                             u32 tb_id,
 428                                             enum mlxsw_sp_l3proto proto)
 429 {
 430         struct mlxsw_sp_vr *vr;
 431         int i;
 432
 433         tb_id = mlxsw_sp_fix_tb_id(tb_id);
 434
 435         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
 436                 vr = &mlxsw_sp->router.vrs[i];
 437                 if (vr->used && vr->proto == proto && vr->tb_id == tb_id)
 438                         return vr;
 439         }
 440         return NULL;
 441 }
 442
 443 static struct mlxsw_sp_vr *mlxsw_sp_vr_create(struct mlxsw_sp *mlxsw_sp,
 444                                               unsigned char prefix_len,
 445                                               u32 tb_id,
 446                                               enum mlxsw_sp_l3proto proto)
 447 {
 448         struct mlxsw_sp_prefix_usage req_prefix_usage;
 449         struct mlxsw_sp_lpm_tree *lpm_tree;
 450         struct mlxsw_sp_vr *vr;
 451         int err;
 452
 453         vr = mlxsw_sp_vr_find_unused(mlxsw_sp);
 454         if (!vr)
 455                 return ERR_PTR(-EBUSY);
 456         vr->fib = mlxsw_sp_fib_create();
 457         if (IS_ERR(vr->fib))
 458                 return ERR_CAST(vr->fib);
 459
 460         vr->proto = proto;
 461         vr->tb_id = tb_id;
 462         mlxsw_sp_prefix_usage_zero(&req_prefix_usage);
 463         mlxsw_sp_prefix_usage_set(&req_prefix_usage, prefix_len);
 464         lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
 465                                          proto, true);
 466         if (IS_ERR(lpm_tree)) {
 467                 err = PTR_ERR(lpm_tree);
 468                 goto err_tree_get;
 469         }
 470         vr->lpm_tree = lpm_tree;
 471         err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, vr);
 472         if (err)
 473                 goto err_tree_bind;
 474
 475         vr->used = true;
 476         return vr;
 477
 478 err_tree_bind:
 479         mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
 480 err_tree_get:
 481         mlxsw_sp_fib_destroy(vr->fib);
 482
 483         return ERR_PTR(err);
 484 }
 485
 486 static void mlxsw_sp_vr_destroy(struct mlxsw_sp *mlxsw_sp,
 487                                 struct mlxsw_sp_vr *vr)
 488 {
 489         mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, vr);
 490         mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
 491         mlxsw_sp_fib_destroy(vr->fib);
 492         vr->used = false;
 493 }
 494
 495 static int
 496 mlxsw_sp_vr_lpm_tree_check(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr,
 497                            struct mlxsw_sp_prefix_usage *req_prefix_usage)
 498 {
 499         struct mlxsw_sp_lpm_tree *lpm_tree;
 500
 501         if (mlxsw_sp_prefix_usage_eq(req_prefix_usage,
 502                                      &vr->lpm_tree->prefix_usage))
 503                 return 0;
 504
 505         lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, req_prefix_usage,
 506                                          vr->proto, false);
 507         if (IS_ERR(lpm_tree)) {
 508                 /* We failed to get a tree according to the required
 509                  * prefix usage. However, the current tree might be still good
 510                  * for us if our requirement is subset of the prefixes used
 511                  * in the tree.
 512                  */
 513                 if (mlxsw_sp_prefix_usage_subset(req_prefix_usage,
 514                                                  &vr->lpm_tree->prefix_usage))
 515                         return 0;
 516                 return PTR_ERR(lpm_tree);
 517         }
 518
 519         mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, vr);
 520         mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
 521         vr->lpm_tree = lpm_tree;
 522         return mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, vr);
 523 }
 524
 525 static struct mlxsw_sp_vr *mlxsw_sp_vr_get(struct mlxsw_sp *mlxsw_sp,
 526                                            unsigned char prefix_len,
 527                                            u32 tb_id,
 528                                            enum mlxsw_sp_l3proto proto)
 529 {
 530         struct mlxsw_sp_vr *vr;
 531         int err;
 532
 533         tb_id = mlxsw_sp_fix_tb_id(tb_id);
 534         vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id, proto);
 535         if (!vr) {
 536                 vr = mlxsw_sp_vr_create(mlxsw_sp, prefix_len, tb_id, proto);
 537                 if (IS_ERR(vr))
 538                         return vr;
 539         } else {
 540                 struct mlxsw_sp_prefix_usage req_prefix_usage;
 541
 542                 mlxsw_sp_prefix_usage_cpy(&req_prefix_usage,
 543                                           &vr->fib->prefix_usage);
 544                 mlxsw_sp_prefix_usage_set(&req_prefix_usage, prefix_len);
 545                 /* Need to replace LPM tree in case new prefix is required. */
 546                 err = mlxsw_sp_vr_lpm_tree_check(mlxsw_sp, vr,
 547                                                  &req_prefix_usage);
 548                 if (err)
 549                         return ERR_PTR(err);
 550         }
 551         return vr;
 552 }
 553
 554 static void mlxsw_sp_vr_put(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr)
 555 {
 556         /* Destroy virtual router entity in case the associated FIB is empty
 557          * and allow it to be used for other tables in future. Otherwise,
 558          * check if some prefix usage did not disappear and change tree if
 559          * that is the case. Note that in case new, smaller tree cannot be
 560          * allocated, the original one will be kept being used.
 561          */
 562         if (mlxsw_sp_prefix_usage_none(&vr->fib->prefix_usage))
 563                 mlxsw_sp_vr_destroy(mlxsw_sp, vr);
 564         else
 565                 mlxsw_sp_vr_lpm_tree_check(mlxsw_sp, vr,
 566                                            &vr->fib->prefix_usage);
 567 }
 568
 569 static int mlxsw_sp_vrs_init(struct mlxsw_sp *mlxsw_sp)
 570 {
 571         struct mlxsw_sp_vr *vr;
 572         u64 max_vrs;
 573         int i;
 574
 575         if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_VRS))
 576                 return -EIO;
 577
 578         max_vrs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS);
 579         mlxsw_sp->router.vrs = kcalloc(max_vrs, sizeof(struct mlxsw_sp_vr),
 580                                        GFP_KERNEL);
 581         if (!mlxsw_sp->router.vrs)
 582                 return -ENOMEM;
 583
 584         for (i = 0; i < max_vrs; i++) {
 585                 vr = &mlxsw_sp->router.vrs[i];
 586                 vr->id = i;
 587         }
 588
 589         return 0;
 590 }
 591
 592 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp);
 593
 594 static void mlxsw_sp_vrs_fini(struct mlxsw_sp *mlxsw_sp)
 595 {
 596         /* At this stage we're guaranteed not to have new incoming
 597          * FIB notifications and the work queue is free from FIBs
 598          * sitting on top of mlxsw netdevs. However, we can still
 599          * have other FIBs queued. Flush the queue before flushing
 600          * the device's tables. No need for locks, as we're the only
 601          * writer.
 602          */
 603         mlxsw_core_flush_owq();
 604         mlxsw_sp_router_fib_flush(mlxsw_sp);
 605         kfree(mlxsw_sp->router.vrs);
 606 }
 607
 608 struct mlxsw_sp_neigh_key {
 609         struct neighbour *n;
 610 };
 611
 612 struct mlxsw_sp_neigh_entry {
 613         struct rhash_head ht_node;
 614         struct mlxsw_sp_neigh_key key;
 615         u16 rif;
 616         bool connected;
 617         unsigned char ha[ETH_ALEN];
 618         struct list_head nexthop_list; /* list of nexthops using
 619                                         * this neigh entry
 620                                         */
 621         struct list_head nexthop_neighs_list_node;
 622 };
 623
 624 static const struct rhashtable_params mlxsw_sp_neigh_ht_params = {
 625         .key_offset = offsetof(struct mlxsw_sp_neigh_entry, key),
 626         .head_offset = offsetof(struct mlxsw_sp_neigh_entry, ht_node),
 627         .key_len = sizeof(struct mlxsw_sp_neigh_key),
 628 };
 629
 630 static struct mlxsw_sp_neigh_entry *
 631 mlxsw_sp_neigh_entry_alloc(struct mlxsw_sp *mlxsw_sp, struct neighbour *n,
 632                            u16 rif)
 633 {
 634         struct mlxsw_sp_neigh_entry *neigh_entry;
 635
 636         neigh_entry = kzalloc(sizeof(*neigh_entry), GFP_KERNEL);
 637         if (!neigh_entry)
 638                 return NULL;
 639
 640         neigh_entry->key.n = n;
 641         neigh_entry->rif = rif;
 642         INIT_LIST_HEAD(&neigh_entry->nexthop_list);
 643
 644         return neigh_entry;
 645 }
 646
 647 static void mlxsw_sp_neigh_entry_free(struct mlxsw_sp_neigh_entry *neigh_entry)
 648 {
 649         kfree(neigh_entry);
 650 }
 651
 652 static int
 653 mlxsw_sp_neigh_entry_insert(struct mlxsw_sp *mlxsw_sp,
 654                             struct mlxsw_sp_neigh_entry *neigh_entry)
 655 {
 656         return rhashtable_insert_fast(&mlxsw_sp->router.neigh_ht,
 657                                       &neigh_entry->ht_node,
 658                                       mlxsw_sp_neigh_ht_params);
 659 }
 660
 661 static void
 662 mlxsw_sp_neigh_entry_remove(struct mlxsw_sp *mlxsw_sp,
 663                             struct mlxsw_sp_neigh_entry *neigh_entry)
 664 {
 665         rhashtable_remove_fast(&mlxsw_sp->router.neigh_ht,
 666                                &neigh_entry->ht_node,
 667                                mlxsw_sp_neigh_ht_params);
 668 }
 669
 670 static struct mlxsw_sp_neigh_entry *
 671 mlxsw_sp_neigh_entry_create(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
 672 {
 673         struct mlxsw_sp_neigh_entry *neigh_entry;
 674         struct mlxsw_sp_rif *r;
 675         int err;
 676
 677         r = mlxsw_sp_rif_find_by_dev(mlxsw_sp, n->dev);
 678         if (!r)
 679                 return ERR_PTR(-EINVAL);
 680
 681         neigh_entry = mlxsw_sp_neigh_entry_alloc(mlxsw_sp, n, r->rif);
 682         if (!neigh_entry)
 683                 return ERR_PTR(-ENOMEM);
 684
 685         err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
 686         if (err)
 687                 goto err_neigh_entry_insert;
 688
 689         return neigh_entry;
 690
 691 err_neigh_entry_insert:
 692         mlxsw_sp_neigh_entry_free(neigh_entry);
 693         return ERR_PTR(err);
 694 }
 695
 696 static void
 697 mlxsw_sp_neigh_entry_destroy(struct mlxsw_sp *mlxsw_sp,
 698                              struct mlxsw_sp_neigh_entry *neigh_entry)
 699 {
 700         mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
 701         mlxsw_sp_neigh_entry_free(neigh_entry);
 702 }
 703
 704 static struct mlxsw_sp_neigh_entry *
 705 mlxsw_sp_neigh_entry_lookup(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
 706 {
 707         struct mlxsw_sp_neigh_key key;
 708
 709         key.n = n;
 710         return rhashtable_lookup_fast(&mlxsw_sp->router.neigh_ht,
 711                                       &key, mlxsw_sp_neigh_ht_params);
 712 }
 713
 714 static void
 715 mlxsw_sp_router_neighs_update_interval_init(struct mlxsw_sp *mlxsw_sp)
 716 {
 717         unsigned long interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME);
 718
 719         mlxsw_sp->router.neighs_update.interval = jiffies_to_msecs(interval);
 720 }
 721
 722 static void mlxsw_sp_router_neigh_ent_ipv4_process(struct mlxsw_sp *mlxsw_sp,
 723                                                    char *rauhtd_pl,
 724                                                    int ent_index)
 725 {
 726         struct net_device *dev;
 727         struct neighbour *n;
 728         __be32 dipn;
 729         u32 dip;
 730         u16 rif;
 731
 732         mlxsw_reg_rauhtd_ent_ipv4_unpack(rauhtd_pl, ent_index, &rif, &dip);
 733
 734         if (!mlxsw_sp->rifs[rif]) {
 735                 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
 736                 return;
 737         }
 738
 739         dipn = htonl(dip);
 740         dev = mlxsw_sp->rifs[rif]->dev;
 741         n = neigh_lookup(&arp_tbl, &dipn, dev);
 742         if (!n) {
 743                 netdev_err(dev, "Failed to find matching neighbour for IP=%pI4h\n",
 744                            &dip);
 745                 return;
 746         }
 747
 748         netdev_dbg(dev, "Updating neighbour with IP=%pI4h\n", &dip);
 749         neigh_event_send(n, NULL);
 750         neigh_release(n);
 751 }
 752
 753 static void mlxsw_sp_router_neigh_rec_ipv4_process(struct mlxsw_sp *mlxsw_sp,
 754                                                    char *rauhtd_pl,
 755                                                    int rec_index)
 756 {
 757         u8 num_entries;
 758         int i;
 759
 760         num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
 761                                                                 rec_index);
 762         /* Hardware starts counting at 0, so add 1. */
 763         num_entries++;
 764
 765         /* Each record consists of several neighbour entries. */
 766         for (i = 0; i < num_entries; i++) {
 767                 int ent_index;
 768
 769                 ent_index = rec_index * MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC + i;
 770                 mlxsw_sp_router_neigh_ent_ipv4_process(mlxsw_sp, rauhtd_pl,
 771                                                        ent_index);
 772         }
 773
 774 }
 775
 776 static void mlxsw_sp_router_neigh_rec_process(struct mlxsw_sp *mlxsw_sp,
 777                                               char *rauhtd_pl, int rec_index)
 778 {
 779         switch (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, rec_index)) {
 780         case MLXSW_REG_RAUHTD_TYPE_IPV4:
 781                 mlxsw_sp_router_neigh_rec_ipv4_process(mlxsw_sp, rauhtd_pl,
 782                                                        rec_index);
 783                 break;
 784         case MLXSW_REG_RAUHTD_TYPE_IPV6:
 785                 WARN_ON_ONCE(1);
 786                 break;
 787         }
 788 }
 789
 790 static bool mlxsw_sp_router_rauhtd_is_full(char *rauhtd_pl)
 791 {
 792         u8 num_rec, last_rec_index, num_entries;
 793
 794         num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
 795         last_rec_index = num_rec - 1;
 796
 797         if (num_rec < MLXSW_REG_RAUHTD_REC_MAX_NUM)
 798                 return false;
 799         if (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, last_rec_index) ==
 800             MLXSW_REG_RAUHTD_TYPE_IPV6)
 801                 return true;
 802
 803         num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
 804                                                                 last_rec_index);
 805         if (++num_entries == MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC)
 806                 return true;
 807         return false;
 808 }
 809
 810 static int mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp)
 811 {
 812         char *rauhtd_pl;
 813         u8 num_rec;
 814         int i, err;
 815
 816         rauhtd_pl = kmalloc(MLXSW_REG_RAUHTD_LEN, GFP_KERNEL);
 817         if (!rauhtd_pl)
 818                 return -ENOMEM;
 819
 820         /* Make sure the neighbour's netdev isn't removed in the
 821          * process.
 822          */
 823         rtnl_lock();
 824         do {
 825                 mlxsw_reg_rauhtd_pack(rauhtd_pl, MLXSW_REG_RAUHTD_TYPE_IPV4);
 826                 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(rauhtd),
 827                                       rauhtd_pl);
 828                 if (err) {
 829                         dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Failed to dump neighbour talbe\n");
 830                         break;
 831                 }
 832                 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
 833                 for (i = 0; i < num_rec; i++)
 834                         mlxsw_sp_router_neigh_rec_process(mlxsw_sp, rauhtd_pl,
 835                                                           i);
 836         } while (mlxsw_sp_router_rauhtd_is_full(rauhtd_pl));
 837         rtnl_unlock();
 838
 839         kfree(rauhtd_pl);
 840         return err;
 841 }
 842
 843 static void mlxsw_sp_router_neighs_update_nh(struct mlxsw_sp *mlxsw_sp)
 844 {
 845         struct mlxsw_sp_neigh_entry *neigh_entry;
 846
 847         /* Take RTNL mutex here to prevent lists from changes */
 848         rtnl_lock();
 849         list_for_each_entry(neigh_entry, &mlxsw_sp->router.nexthop_neighs_list,
 850                             nexthop_neighs_list_node)
 851                 /* If this neigh have nexthops, make the kernel think this neigh
 852                  * is active regardless of the traffic.
 853                  */
 854                 neigh_event_send(neigh_entry->key.n, NULL);
 855         rtnl_unlock();
 856 }
 857
 858 static void
 859 mlxsw_sp_router_neighs_update_work_schedule(struct mlxsw_sp *mlxsw_sp)
 860 {
 861         unsigned long interval = mlxsw_sp->router.neighs_update.interval;
 862
 863         mlxsw_core_schedule_dw(&mlxsw_sp->router.neighs_update.dw,
 864                                msecs_to_jiffies(interval));
 865 }
 866
 867 static void mlxsw_sp_router_neighs_update_work(struct work_struct *work)
 868 {
 869         struct mlxsw_sp *mlxsw_sp = container_of(work, struct mlxsw_sp,
 870                                                  router.neighs_update.dw.work);
 871         int err;
 872
 873         err = mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp);
 874         if (err)
 875                 dev_err(mlxsw_sp->bus_info->dev, "Could not update kernel for neigh activity");
 876
 877         mlxsw_sp_router_neighs_update_nh(mlxsw_sp);
 878
 879         mlxsw_sp_router_neighs_update_work_schedule(mlxsw_sp);
 880 }
 881
 882 static void mlxsw_sp_router_probe_unresolved_nexthops(struct work_struct *work)
 883 {
 884         struct mlxsw_sp_neigh_entry *neigh_entry;
 885         struct mlxsw_sp *mlxsw_sp = container_of(work, struct mlxsw_sp,
 886                                                  router.nexthop_probe_dw.work);
 887
 888         /* Iterate over nexthop neighbours, find those who are unresolved and
 889          * send arp on them. This solves the chicken-egg problem when
 890          * the nexthop wouldn't get offloaded until the neighbor is resolved
 891          * but it wouldn't get resolved ever in case traffic is flowing in HW
 892          * using different nexthop.
 893          *
 894          * Take RTNL mutex here to prevent lists from changes.
 895          */
 896         rtnl_lock();
 897         list_for_each_entry(neigh_entry, &mlxsw_sp->router.nexthop_neighs_list,
 898                             nexthop_neighs_list_node)
 899                 if (!neigh_entry->connected)
 900                         neigh_event_send(neigh_entry->key.n, NULL);
 901         rtnl_unlock();
 902
 903         mlxsw_core_schedule_dw(&mlxsw_sp->router.nexthop_probe_dw,
 904                                MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL);
 905 }
 906
 907 static void
 908 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
 909                               struct mlxsw_sp_neigh_entry *neigh_entry,
 910                               bool removing);
 911
 912 static enum mlxsw_reg_rauht_op mlxsw_sp_rauht_op(bool adding)
 913 {
 914         return adding ? MLXSW_REG_RAUHT_OP_WRITE_ADD :
 915                         MLXSW_REG_RAUHT_OP_WRITE_DELETE;
 916 }
 917
 918 static void
 919 mlxsw_sp_router_neigh_entry_op4(struct mlxsw_sp *mlxsw_sp,
 920                                 struct mlxsw_sp_neigh_entry *neigh_entry,
 921                                 enum mlxsw_reg_rauht_op op)
 922 {
 923         struct neighbour *n = neigh_entry->key.n;
 924         u32 dip = ntohl(*((__be32 *) n->primary_key));
 925         char rauht_pl[MLXSW_REG_RAUHT_LEN];
 926
 927         mlxsw_reg_rauht_pack4(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
 928                               dip);
 929         mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
 930 }
 931
 932 static void
 933 mlxsw_sp_neigh_entry_update(struct mlxsw_sp *mlxsw_sp,
 934                             struct mlxsw_sp_neigh_entry *neigh_entry,
 935                             bool adding)
 936 {
 937         if (!adding && !neigh_entry->connected)
 938                 return;
 939         neigh_entry->connected = adding;
 940         if (neigh_entry->key.n->tbl == &arp_tbl)
 941                 mlxsw_sp_router_neigh_entry_op4(mlxsw_sp, neigh_entry,
 942                                                 mlxsw_sp_rauht_op(adding));
 943         else
 944                 WARN_ON_ONCE(1);
 945 }
 946
 947 struct mlxsw_sp_neigh_event_work {
 948         struct work_struct work;
 949         struct mlxsw_sp *mlxsw_sp;
 950         struct neighbour *n;
 951 };
 952
 953 static void mlxsw_sp_router_neigh_event_work(struct work_struct *work)
 954 {
 955         struct mlxsw_sp_neigh_event_work *neigh_work =
 956                 container_of(work, struct mlxsw_sp_neigh_event_work, work);
 957         struct mlxsw_sp *mlxsw_sp = neigh_work->mlxsw_sp;
 958         struct mlxsw_sp_neigh_entry *neigh_entry;
 959         struct neighbour *n = neigh_work->n;
 960         unsigned char ha[ETH_ALEN];
 961         bool entry_connected;
 962         u8 nud_state, dead;
 963
 964         /* If these parameters are changed after we release the lock,
 965          * then we are guaranteed to receive another event letting us
 966          * know about it.
 967          */
 968         read_lock_bh(&n->lock);
 969         memcpy(ha, n->ha, ETH_ALEN);
 970         nud_state = n->nud_state;
 971         dead = n->dead;
 972         read_unlock_bh(&n->lock);
 973
 974         rtnl_lock();
 975         entry_connected = nud_state & NUD_VALID && !dead;
 976         neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
 977         if (!entry_connected && !neigh_entry)
 978                 goto out;
 979         if (!neigh_entry) {
 980                 neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
 981                 if (IS_ERR(neigh_entry))
 982                         goto out;
 983         }
 984
 985         memcpy(neigh_entry->ha, ha, ETH_ALEN);
 986         mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, entry_connected);
 987         mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, !entry_connected);
 988
 989         if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
 990                 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
 991
 992 out:
 993         rtnl_unlock();
 994         neigh_release(n);
 995         kfree(neigh_work);
 996 }
 997
 998 int mlxsw_sp_router_netevent_event(struct notifier_block *unused,
 999                                    unsigned long event, void *ptr)
1000 {
1001         struct mlxsw_sp_neigh_event_work *neigh_work;
1002         struct mlxsw_sp_port *mlxsw_sp_port;
1003         struct mlxsw_sp *mlxsw_sp;
1004         unsigned long interval;
1005         struct neigh_parms *p;
1006         struct neighbour *n;
1007
1008         switch (event) {
1009         case NETEVENT_DELAY_PROBE_TIME_UPDATE:
1010                 p = ptr;
1011
1012                 /* We don't care about changes in the default table. */
1013                 if (!p->dev || p->tbl != &arp_tbl)
1014                         return NOTIFY_DONE;
1015
1016                 /* We are in atomic context and can't take RTNL mutex,
1017                  * so use RCU variant to walk the device chain.
1018                  */
1019                 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(p->dev);
1020                 if (!mlxsw_sp_port)
1021                         return NOTIFY_DONE;
1022
1023                 mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1024                 interval = jiffies_to_msecs(NEIGH_VAR(p, DELAY_PROBE_TIME));
1025                 mlxsw_sp->router.neighs_update.interval = interval;
1026
1027                 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1028                 break;
1029         case NETEVENT_NEIGH_UPDATE:
1030                 n = ptr;
1031
1032                 if (n->tbl != &arp_tbl)
1033                         return NOTIFY_DONE;
1034
1035                 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(n->dev);
1036                 if (!mlxsw_sp_port)
1037                         return NOTIFY_DONE;
1038
1039                 neigh_work = kzalloc(sizeof(*neigh_work), GFP_ATOMIC);
1040                 if (!neigh_work) {
1041                         mlxsw_sp_port_dev_put(mlxsw_sp_port);
1042                         return NOTIFY_BAD;
1043                 }
1044
1045                 INIT_WORK(&neigh_work->work, mlxsw_sp_router_neigh_event_work);
1046                 neigh_work->mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1047                 neigh_work->n = n;
1048
1049                 /* Take a reference to ensure the neighbour won't be
1050                  * destructed until we drop the reference in delayed
1051                  * work.
1052                  */
1053                 neigh_clone(n);
1054                 mlxsw_core_schedule_work(&neigh_work->work);
1055                 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1056                 break;
1057         }
1058
1059         return NOTIFY_DONE;
1060 }
1061
1062 static int mlxsw_sp_neigh_init(struct mlxsw_sp *mlxsw_sp)
1063 {
1064         int err;
1065
1066         err = rhashtable_init(&mlxsw_sp->router.neigh_ht,
1067                               &mlxsw_sp_neigh_ht_params);
1068         if (err)
1069                 return err;
1070
1071         /* Initialize the polling interval according to the default
1072          * table.
1073          */
1074         mlxsw_sp_router_neighs_update_interval_init(mlxsw_sp);
1075
1076         /* Create the delayed works for the activity_update */
1077         INIT_DELAYED_WORK(&mlxsw_sp->router.neighs_update.dw,
1078                           mlxsw_sp_router_neighs_update_work);
1079         INIT_DELAYED_WORK(&mlxsw_sp->router.nexthop_probe_dw,
1080                           mlxsw_sp_router_probe_unresolved_nexthops);
1081         mlxsw_core_schedule_dw(&mlxsw_sp->router.neighs_update.dw, 0);
1082         mlxsw_core_schedule_dw(&mlxsw_sp->router.nexthop_probe_dw, 0);
1083         return 0;
1084 }
1085
1086 static void mlxsw_sp_neigh_fini(struct mlxsw_sp *mlxsw_sp)
1087 {
1088         cancel_delayed_work_sync(&mlxsw_sp->router.neighs_update.dw);
1089         cancel_delayed_work_sync(&mlxsw_sp->router.nexthop_probe_dw);
1090         rhashtable_destroy(&mlxsw_sp->router.neigh_ht);
1091 }
1092
1093 struct mlxsw_sp_nexthop {
1094         struct list_head neigh_list_node; /* member of neigh entry list */
1095         struct mlxsw_sp_nexthop_group *nh_grp; /* pointer back to the group
1096                                                 * this belongs to
1097                                                 */
1098         u8 should_offload:1, /* set indicates this neigh is connected and
1099                               * should be put to KVD linear area of this group.
1100                               */
1101            offloaded:1, /* set in case the neigh is actually put into
1102                          * KVD linear area of this group.
1103                          */
1104            update:1; /* set indicates that MAC of this neigh should be
1105                       * updated in HW
1106                       */
1107         struct mlxsw_sp_neigh_entry *neigh_entry;
1108 };
1109
1110 struct mlxsw_sp_nexthop_group {
1111         struct list_head list; /* node in mlxsw->router.nexthop_group_list */
1112         struct list_head fib_list; /* list of fib entries that use this group */
1113         u8 adj_index_valid:1;
1114         u32 adj_index;
1115         u16 ecmp_size;
1116         u16 count;
1117         struct mlxsw_sp_nexthop nexthops[0];
1118 };
1119
1120 static int mlxsw_sp_adj_index_mass_update_vr(struct mlxsw_sp *mlxsw_sp,
1121                                              struct mlxsw_sp_vr *vr,
1122                                              u32 adj_index, u16 ecmp_size,
1123                                              u32 new_adj_index,
1124                                              u16 new_ecmp_size)
1125 {
1126         char raleu_pl[MLXSW_REG_RALEU_LEN];
1127
1128         mlxsw_reg_raleu_pack(raleu_pl,
1129                              (enum mlxsw_reg_ralxx_protocol) vr->proto, vr->id,
1130                              adj_index, ecmp_size, new_adj_index,
1131                              new_ecmp_size);
1132         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raleu), raleu_pl);
1133 }
1134
1135 static int mlxsw_sp_adj_index_mass_update(struct mlxsw_sp *mlxsw_sp,
1136                                           struct mlxsw_sp_nexthop_group *nh_grp,
1137                                           u32 old_adj_index, u16 old_ecmp_size)
1138 {
1139         struct mlxsw_sp_fib_entry *fib_entry;
1140         struct mlxsw_sp_vr *vr = NULL;
1141         int err;
1142
1143         list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
1144                 if (vr == fib_entry->vr)
1145                         continue;
1146                 vr = fib_entry->vr;
1147                 err = mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp, vr,
1148                                                         old_adj_index,
1149                                                         old_ecmp_size,
1150                                                         nh_grp->adj_index,
1151                                                         nh_grp->ecmp_size);
1152                 if (err)
1153                         return err;
1154         }
1155         return 0;
1156 }
1157
1158 static int mlxsw_sp_nexthop_mac_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
1159                                        struct mlxsw_sp_nexthop *nh)
1160 {
1161         struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
1162         char ratr_pl[MLXSW_REG_RATR_LEN];
1163
1164         mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
1165                             true, adj_index, neigh_entry->rif);
1166         mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
1167         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
1168 }
1169
1170 static int
1171 mlxsw_sp_nexthop_group_mac_update(struct mlxsw_sp *mlxsw_sp,
1172                                   struct mlxsw_sp_nexthop_group *nh_grp,
1173                                   bool reallocate)
1174 {
1175         u32 adj_index = nh_grp->adj_index; /* base */
1176         struct mlxsw_sp_nexthop *nh;
1177         int i;
1178         int err;
1179
1180         for (i = 0; i < nh_grp->count; i++) {
1181                 nh = &nh_grp->nexthops[i];
1182
1183                 if (!nh->should_offload) {
1184                         nh->offloaded = 0;
1185                         continue;
1186                 }
1187
1188                 if (nh->update || reallocate) {
1189                         err = mlxsw_sp_nexthop_mac_update(mlxsw_sp,
1190                                                           adj_index, nh);
1191                         if (err)
1192                                 return err;
1193                         nh->update = 0;
1194                         nh->offloaded = 1;
1195                 }
1196                 adj_index++;
1197         }
1198         return 0;
1199 }
1200
1201 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1202                                      struct mlxsw_sp_fib_entry *fib_entry);
1203
1204 static int
1205 mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
1206                                     struct mlxsw_sp_nexthop_group *nh_grp)
1207 {
1208         struct mlxsw_sp_fib_entry *fib_entry;
1209         int err;
1210
1211         list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
1212                 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1213                 if (err)
1214                         return err;
1215         }
1216         return 0;
1217 }
1218
1219 static void
1220 mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
1221                                struct mlxsw_sp_nexthop_group *nh_grp)
1222 {
1223         struct mlxsw_sp_nexthop *nh;
1224         bool offload_change = false;
1225         u32 adj_index;
1226         u16 ecmp_size = 0;
1227         bool old_adj_index_valid;
1228         u32 old_adj_index;
1229         u16 old_ecmp_size;
1230         int ret;
1231         int i;
1232         int err;
1233
1234         for (i = 0; i < nh_grp->count; i++) {
1235                 nh = &nh_grp->nexthops[i];
1236
1237                 if (nh->should_offload ^ nh->offloaded) {
1238                         offload_change = true;
1239                         if (nh->should_offload)
1240                                 nh->update = 1;
1241                 }
1242                 if (nh->should_offload)
1243                         ecmp_size++;
1244         }
1245         if (!offload_change) {
1246                 /* Nothing was added or removed, so no need to reallocate. Just
1247                  * update MAC on existing adjacency indexes.
1248                  */
1249                 err = mlxsw_sp_nexthop_group_mac_update(mlxsw_sp, nh_grp,
1250                                                         false);
1251                 if (err) {
1252                         dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
1253                         goto set_trap;
1254                 }
1255                 return;
1256         }
1257         if (!ecmp_size)
1258                 /* No neigh of this group is connected so we just set
1259                  * the trap and let everthing flow through kernel.
1260                  */
1261                 goto set_trap;
1262
1263         ret = mlxsw_sp_kvdl_alloc(mlxsw_sp, ecmp_size);
1264         if (ret < 0) {
1265                 /* We ran out of KVD linear space, just set the
1266                  * trap and let everything flow through kernel.
1267                  */
1268                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to allocate KVD linear area for nexthop group.\n");
1269                 goto set_trap;
1270         }
1271         adj_index = ret;
1272         old_adj_index_valid = nh_grp->adj_index_valid;
1273         old_adj_index = nh_grp->adj_index;
1274         old_ecmp_size = nh_grp->ecmp_size;
1275         nh_grp->adj_index_valid = 1;
1276         nh_grp->adj_index = adj_index;
1277         nh_grp->ecmp_size = ecmp_size;
1278         err = mlxsw_sp_nexthop_group_mac_update(mlxsw_sp, nh_grp, true);
1279         if (err) {
1280                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
1281                 goto set_trap;
1282         }
1283
1284         if (!old_adj_index_valid) {
1285                 /* The trap was set for fib entries, so we have to call
1286                  * fib entry update to unset it and use adjacency index.
1287                  */
1288                 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
1289                 if (err) {
1290                         dev_warn(mlxsw_sp->bus_info->dev, "Failed to add adjacency index to fib entries.\n");
1291                         goto set_trap;
1292                 }
1293                 return;
1294         }
1295
1296         err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, nh_grp,
1297                                              old_adj_index, old_ecmp_size);
1298         mlxsw_sp_kvdl_free(mlxsw_sp, old_adj_index);
1299         if (err) {
1300                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to mass-update adjacency index for nexthop group.\n");
1301                 goto set_trap;
1302         }
1303         return;
1304
1305 set_trap:
1306         old_adj_index_valid = nh_grp->adj_index_valid;
1307         nh_grp->adj_index_valid = 0;
1308         for (i = 0; i < nh_grp->count; i++) {
1309                 nh = &nh_grp->nexthops[i];
1310                 nh->offloaded = 0;
1311         }
1312         err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
1313         if (err)
1314                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set traps for fib entries.\n");
1315         if (old_adj_index_valid)
1316                 mlxsw_sp_kvdl_free(mlxsw_sp, nh_grp->adj_index);
1317 }
1318
1319 static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
1320                                             bool removing)
1321 {
1322         if (!removing && !nh->should_offload)
1323                 nh->should_offload = 1;
1324         else if (removing && nh->offloaded)
1325                 nh->should_offload = 0;
1326         nh->update = 1;
1327 }
1328
1329 static void
1330 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
1331                               struct mlxsw_sp_neigh_entry *neigh_entry,
1332                               bool removing)
1333 {
1334         struct mlxsw_sp_nexthop *nh;
1335
1336         list_for_each_entry(nh, &neigh_entry->nexthop_list,
1337                             neigh_list_node) {
1338                 __mlxsw_sp_nexthop_neigh_update(nh, removing);
1339                 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
1340         }
1341 }
1342
1343 static int mlxsw_sp_nexthop_init(struct mlxsw_sp *mlxsw_sp,
1344                                  struct mlxsw_sp_nexthop_group *nh_grp,
1345                                  struct mlxsw_sp_nexthop *nh,
1346                                  struct fib_nh *fib_nh)
1347 {
1348         struct mlxsw_sp_neigh_entry *neigh_entry;
1349         struct net_device *dev = fib_nh->nh_dev;
1350         struct neighbour *n;
1351         u8 nud_state, dead;
1352
1353         /* Take a reference of neigh here ensuring that neigh would
1354          * not be detructed before the nexthop entry is finished.
1355          * The reference is taken either in neigh_lookup() or
1356          * in neigh_create() in case n is not found.
1357          */
1358         n = neigh_lookup(&arp_tbl, &fib_nh->nh_gw, dev);
1359         if (!n) {
1360                 n = neigh_create(&arp_tbl, &fib_nh->nh_gw, dev);
1361                 if (IS_ERR(n))
1362                         return PTR_ERR(n);
1363                 neigh_event_send(n, NULL);
1364         }
1365         neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
1366         if (!neigh_entry) {
1367                 neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
1368                 if (IS_ERR(neigh_entry)) {
1369                         neigh_release(n);
1370                         return -EINVAL;
1371                 }
1372         }
1373
1374         /* If that is the first nexthop connected to that neigh, add to
1375          * nexthop_neighs_list
1376          */
1377         if (list_empty(&neigh_entry->nexthop_list))
1378                 list_add_tail(&neigh_entry->nexthop_neighs_list_node,
1379                               &mlxsw_sp->router.nexthop_neighs_list);
1380
1381         nh->nh_grp = nh_grp;
1382         nh->neigh_entry = neigh_entry;
1383         list_add_tail(&nh->neigh_list_node, &neigh_entry->nexthop_list);
1384         read_lock_bh(&n->lock);
1385         nud_state = n->nud_state;
1386         dead = n->dead;
1387         read_unlock_bh(&n->lock);
1388         __mlxsw_sp_nexthop_neigh_update(nh, !(nud_state & NUD_VALID && !dead));
1389
1390         return 0;
1391 }
1392
1393 static void mlxsw_sp_nexthop_fini(struct mlxsw_sp *mlxsw_sp,
1394                                   struct mlxsw_sp_nexthop *nh)
1395 {
1396         struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
1397         struct neighbour *n = neigh_entry->key.n;
1398
1399         __mlxsw_sp_nexthop_neigh_update(nh, true);
1400         list_del(&nh->neigh_list_node);
1401
1402         /* If that is the last nexthop connected to that neigh, remove from
1403          * nexthop_neighs_list
1404          */
1405         if (list_empty(&nh->neigh_entry->nexthop_list))
1406                 list_del(&nh->neigh_entry->nexthop_neighs_list_node);
1407
1408         if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
1409                 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
1410
1411         neigh_release(n);
1412 }
1413
1414 static struct mlxsw_sp_nexthop_group *
1415 mlxsw_sp_nexthop_group_create(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
1416 {
1417         struct mlxsw_sp_nexthop_group *nh_grp;
1418         struct mlxsw_sp_nexthop *nh;
1419         struct fib_nh *fib_nh;
1420         size_t alloc_size;
1421         int i;
1422         int err;
1423
1424         alloc_size = sizeof(*nh_grp) +
1425                      fi->fib_nhs * sizeof(struct mlxsw_sp_nexthop);
1426         nh_grp = kzalloc(alloc_size, GFP_KERNEL);
1427         if (!nh_grp)
1428                 return ERR_PTR(-ENOMEM);
1429         INIT_LIST_HEAD(&nh_grp->fib_list);
1430         nh_grp->count = fi->fib_nhs;
1431         for (i = 0; i < nh_grp->count; i++) {
1432                 nh = &nh_grp->nexthops[i];
1433                 fib_nh = &fi->fib_nh[i];
1434                 err = mlxsw_sp_nexthop_init(mlxsw_sp, nh_grp, nh, fib_nh);
1435                 if (err)
1436                         goto err_nexthop_init;
1437         }
1438         list_add_tail(&nh_grp->list, &mlxsw_sp->router.nexthop_group_list);
1439         mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
1440         return nh_grp;
1441
1442 err_nexthop_init:
1443         for (i--; i >= 0; i--)
1444                 mlxsw_sp_nexthop_fini(mlxsw_sp, nh);
1445         kfree(nh_grp);
1446         return ERR_PTR(err);
1447 }
1448
1449 static void
1450 mlxsw_sp_nexthop_group_destroy(struct mlxsw_sp *mlxsw_sp,
1451                                struct mlxsw_sp_nexthop_group *nh_grp)
1452 {
1453         struct mlxsw_sp_nexthop *nh;
1454         int i;
1455
1456         list_del(&nh_grp->list);
1457         for (i = 0; i < nh_grp->count; i++) {
1458                 nh = &nh_grp->nexthops[i];
1459                 mlxsw_sp_nexthop_fini(mlxsw_sp, nh);
1460         }
1461         mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
1462         WARN_ON_ONCE(nh_grp->adj_index_valid);
1463         kfree(nh_grp);
1464 }
1465
1466 static bool mlxsw_sp_nexthop_match(struct mlxsw_sp_nexthop *nh,
1467                                    struct fib_info *fi)
1468 {
1469         int i;
1470
1471         for (i = 0; i < fi->fib_nhs; i++) {
1472                 struct fib_nh *fib_nh = &fi->fib_nh[i];
1473                 struct neighbour *n = nh->neigh_entry->key.n;
1474
1475                 if (memcmp(n->primary_key, &fib_nh->nh_gw,
1476                            sizeof(fib_nh->nh_gw)) == 0 &&
1477                     n->dev == fib_nh->nh_dev)
1478                         return true;
1479         }
1480         return false;
1481 }
1482
1483 static bool mlxsw_sp_nexthop_group_match(struct mlxsw_sp_nexthop_group *nh_grp,
1484                                          struct fib_info *fi)
1485 {
1486         int i;
1487
1488         if (nh_grp->count != fi->fib_nhs)
1489                 return false;
1490         for (i = 0; i < nh_grp->count; i++) {
1491                 struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
1492
1493                 if (!mlxsw_sp_nexthop_match(nh, fi))
1494                         return false;
1495         }
1496         return true;
1497 }
1498
1499 static struct mlxsw_sp_nexthop_group *
1500 mlxsw_sp_nexthop_group_find(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
1501 {
1502         struct mlxsw_sp_nexthop_group *nh_grp;
1503
1504         list_for_each_entry(nh_grp, &mlxsw_sp->router.nexthop_group_list,
1505                             list) {
1506                 if (mlxsw_sp_nexthop_group_match(nh_grp, fi))
1507                         return nh_grp;
1508         }
1509         return NULL;
1510 }
1511
1512 static int mlxsw_sp_nexthop_group_get(struct mlxsw_sp *mlxsw_sp,
1513                                       struct mlxsw_sp_fib_entry *fib_entry,
1514                                       struct fib_info *fi)
1515 {
1516         struct mlxsw_sp_nexthop_group *nh_grp;
1517
1518         nh_grp = mlxsw_sp_nexthop_group_find(mlxsw_sp, fi);
1519         if (!nh_grp) {
1520                 nh_grp = mlxsw_sp_nexthop_group_create(mlxsw_sp, fi);
1521                 if (IS_ERR(nh_grp))
1522                         return PTR_ERR(nh_grp);
1523         }
1524         list_add_tail(&fib_entry->nexthop_group_node, &nh_grp->fib_list);
1525         fib_entry->nh_group = nh_grp;
1526         return 0;
1527 }
1528
1529 static void mlxsw_sp_nexthop_group_put(struct mlxsw_sp *mlxsw_sp,
1530                                        struct mlxsw_sp_fib_entry *fib_entry)
1531 {
1532         struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
1533
1534         list_del(&fib_entry->nexthop_group_node);
1535         if (!list_empty(&nh_grp->fib_list))
1536                 return;
1537         mlxsw_sp_nexthop_group_destroy(mlxsw_sp, nh_grp);
1538 }
1539
1540 static int mlxsw_sp_fib_entry_op4_remote(struct mlxsw_sp *mlxsw_sp,
1541                                          struct mlxsw_sp_fib_entry *fib_entry,
1542                                          enum mlxsw_reg_ralue_op op)
1543 {
1544         char ralue_pl[MLXSW_REG_RALUE_LEN];
1545         u32 *p_dip = (u32 *) fib_entry->key.addr;
1546         struct mlxsw_sp_vr *vr = fib_entry->vr;
1547         enum mlxsw_reg_ralue_trap_action trap_action;
1548         u16 trap_id = 0;
1549         u32 adjacency_index = 0;
1550         u16 ecmp_size = 0;
1551
1552         /* In case the nexthop group adjacency index is valid, use it
1553          * with provided ECMP size. Otherwise, setup trap and pass
1554          * traffic to kernel.
1555          */
1556         if (fib_entry->nh_group->adj_index_valid) {
1557                 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
1558                 adjacency_index = fib_entry->nh_group->adj_index;
1559                 ecmp_size = fib_entry->nh_group->ecmp_size;
1560         } else {
1561                 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
1562                 trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
1563         }
1564
1565         mlxsw_reg_ralue_pack4(ralue_pl,
1566                               (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1567                               vr->id, fib_entry->key.prefix_len, *p_dip);
1568         mlxsw_reg_ralue_act_remote_pack(ralue_pl, trap_action, trap_id,
1569                                         adjacency_index, ecmp_size);
1570         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1571 }
1572
1573 static int mlxsw_sp_fib_entry_op4_local(struct mlxsw_sp *mlxsw_sp,
1574                                         struct mlxsw_sp_fib_entry *fib_entry,
1575                                         enum mlxsw_reg_ralue_op op)
1576 {
1577         char ralue_pl[MLXSW_REG_RALUE_LEN];
1578         u32 *p_dip = (u32 *) fib_entry->key.addr;
1579         struct mlxsw_sp_vr *vr = fib_entry->vr;
1580
1581         mlxsw_reg_ralue_pack4(ralue_pl,
1582                               (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1583                               vr->id, fib_entry->key.prefix_len, *p_dip);
1584         mlxsw_reg_ralue_act_local_pack(ralue_pl,
1585                                        MLXSW_REG_RALUE_TRAP_ACTION_NOP, 0,
1586                                        fib_entry->rif);
1587         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1588 }
1589
1590 static int mlxsw_sp_fib_entry_op4_trap(struct mlxsw_sp *mlxsw_sp,
1591                                        struct mlxsw_sp_fib_entry *fib_entry,
1592                                        enum mlxsw_reg_ralue_op op)
1593 {
1594         char ralue_pl[MLXSW_REG_RALUE_LEN];
1595         u32 *p_dip = (u32 *) fib_entry->key.addr;
1596         struct mlxsw_sp_vr *vr = fib_entry->vr;
1597
1598         mlxsw_reg_ralue_pack4(ralue_pl,
1599                               (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1600                               vr->id, fib_entry->key.prefix_len, *p_dip);
1601         mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
1602         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1603 }
1604
1605 static int mlxsw_sp_fib_entry_op4(struct mlxsw_sp *mlxsw_sp,
1606                                   struct mlxsw_sp_fib_entry *fib_entry,
1607                                   enum mlxsw_reg_ralue_op op)
1608 {
1609         switch (fib_entry->type) {
1610         case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
1611                 return mlxsw_sp_fib_entry_op4_remote(mlxsw_sp, fib_entry, op);
1612         case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
1613                 return mlxsw_sp_fib_entry_op4_local(mlxsw_sp, fib_entry, op);
1614         case MLXSW_SP_FIB_ENTRY_TYPE_TRAP:
1615                 return mlxsw_sp_fib_entry_op4_trap(mlxsw_sp, fib_entry, op);
1616         }
1617         return -EINVAL;
1618 }
1619
1620 static int mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
1621                                  struct mlxsw_sp_fib_entry *fib_entry,
1622                                  enum mlxsw_reg_ralue_op op)
1623 {
1624         switch (fib_entry->vr->proto) {
1625         case MLXSW_SP_L3_PROTO_IPV4:
1626                 return mlxsw_sp_fib_entry_op4(mlxsw_sp, fib_entry, op);
1627         case MLXSW_SP_L3_PROTO_IPV6:
1628                 return -EINVAL;
1629         }
1630         return -EINVAL;
1631 }
1632
1633 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1634                                      struct mlxsw_sp_fib_entry *fib_entry)
1635 {
1636         return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
1637                                      MLXSW_REG_RALUE_OP_WRITE_WRITE);
1638 }
1639
1640 static int mlxsw_sp_fib_entry_del(struct mlxsw_sp *mlxsw_sp,
1641                                   struct mlxsw_sp_fib_entry *fib_entry)
1642 {
1643         return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
1644                                      MLXSW_REG_RALUE_OP_WRITE_DELETE);
1645 }
1646
1647 static int
1648 mlxsw_sp_router_fib4_entry_init(struct mlxsw_sp *mlxsw_sp,
1649                                 const struct fib_entry_notifier_info *fen_info,
1650                                 struct mlxsw_sp_fib_entry *fib_entry)
1651 {
1652         struct fib_info *fi = fen_info->fi;
1653         struct mlxsw_sp_rif *r = NULL;
1654         int nhsel;
1655         int err;
1656
1657         if (fen_info->type == RTN_LOCAL || fen_info->type == RTN_BROADCAST) {
1658                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1659                 return 0;
1660         }
1661         if (fen_info->type != RTN_UNICAST)
1662                 return -EINVAL;
1663
1664         for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1665                 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1666
1667                 if (!nh->nh_dev)
1668                         continue;
1669                 r = mlxsw_sp_rif_find_by_dev(mlxsw_sp, nh->nh_dev);
1670                 if (!r) {
1671                         /* In case router interface is not found for
1672                          * at least one of the nexthops, that means
1673                          * the nexthop points to some device unrelated
1674                          * to us. Set trap and pass the packets for
1675                          * this prefix to kernel.
1676                          */
1677                         break;
1678                 }
1679         }
1680
1681         if (!r) {
1682                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1683                 return 0;
1684         }
1685
1686         if (fi->fib_scope != RT_SCOPE_UNIVERSE) {
1687                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
1688                 fib_entry->rif = r->rif;
1689         } else {
1690                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
1691                 err = mlxsw_sp_nexthop_group_get(mlxsw_sp, fib_entry, fi);
1692                 if (err)
1693                         return err;
1694         }
1695         fib_info_offload_inc(fen_info->fi);
1696         return 0;
1697 }
1698
1699 static void
1700 mlxsw_sp_router_fib4_entry_fini(struct mlxsw_sp *mlxsw_sp,
1701                                 struct mlxsw_sp_fib_entry *fib_entry)
1702 {
1703         if (fib_entry->type != MLXSW_SP_FIB_ENTRY_TYPE_TRAP)
1704                 fib_info_offload_dec(fib_entry->fi);
1705         if (fib_entry->type == MLXSW_SP_FIB_ENTRY_TYPE_REMOTE)
1706                 mlxsw_sp_nexthop_group_put(mlxsw_sp, fib_entry);
1707 }
1708
1709 static struct mlxsw_sp_fib_entry *
1710 mlxsw_sp_fib_entry_get(struct mlxsw_sp *mlxsw_sp,
1711                        const struct fib_entry_notifier_info *fen_info)
1712 {
1713         struct mlxsw_sp_fib_entry *fib_entry;
1714         struct fib_info *fi = fen_info->fi;
1715         struct mlxsw_sp_vr *vr;
1716         int err;
1717
1718         vr = mlxsw_sp_vr_get(mlxsw_sp, fen_info->dst_len, fen_info->tb_id,
1719                              MLXSW_SP_L3_PROTO_IPV4);
1720         if (IS_ERR(vr))
1721                 return ERR_CAST(vr);
1722
1723         fib_entry = mlxsw_sp_fib_entry_lookup(vr->fib, &fen_info->dst,
1724                                               sizeof(fen_info->dst),
1725                                               fen_info->dst_len, fi->fib_dev);
1726         if (fib_entry) {
1727                 /* Already exists, just take a reference */
1728                 fib_entry->ref_count++;
1729                 return fib_entry;
1730         }
1731         fib_entry = mlxsw_sp_fib_entry_create(vr->fib, &fen_info->dst,
1732                                               sizeof(fen_info->dst),
1733                                               fen_info->dst_len, fi->fib_dev);
1734         if (!fib_entry) {
1735                 err = -ENOMEM;
1736                 goto err_fib_entry_create;
1737         }
1738         fib_entry->vr = vr;
1739         fib_entry->fi = fi;
1740         fib_entry->ref_count = 1;
1741
1742         err = mlxsw_sp_router_fib4_entry_init(mlxsw_sp, fen_info, fib_entry);
1743         if (err)
1744                 goto err_fib4_entry_init;
1745
1746         return fib_entry;
1747
1748 err_fib4_entry_init:
1749         mlxsw_sp_fib_entry_destroy(fib_entry);
1750 err_fib_entry_create:
1751         mlxsw_sp_vr_put(mlxsw_sp, vr);
1752
1753         return ERR_PTR(err);
1754 }
1755
1756 static struct mlxsw_sp_fib_entry *
1757 mlxsw_sp_fib_entry_find(struct mlxsw_sp *mlxsw_sp,
1758                         const struct fib_entry_notifier_info *fen_info)
1759 {
1760         struct mlxsw_sp_vr *vr;
1761
1762         vr = mlxsw_sp_vr_find(mlxsw_sp, fen_info->tb_id,
1763                               MLXSW_SP_L3_PROTO_IPV4);
1764         if (!vr)
1765                 return NULL;
1766
1767         return mlxsw_sp_fib_entry_lookup(vr->fib, &fen_info->dst,
1768                                          sizeof(fen_info->dst),
1769                                          fen_info->dst_len,
1770                                          fen_info->fi->fib_dev);
1771 }
1772
1773 static void mlxsw_sp_fib_entry_put(struct mlxsw_sp *mlxsw_sp,
1774                                    struct mlxsw_sp_fib_entry *fib_entry)
1775 {
1776         struct mlxsw_sp_vr *vr = fib_entry->vr;
1777
1778         if (--fib_entry->ref_count == 0) {
1779                 mlxsw_sp_router_fib4_entry_fini(mlxsw_sp, fib_entry);
1780                 mlxsw_sp_fib_entry_destroy(fib_entry);
1781         }
1782         mlxsw_sp_vr_put(mlxsw_sp, vr);
1783 }
1784
1785 static void mlxsw_sp_fib_entry_put_all(struct mlxsw_sp *mlxsw_sp,
1786                                        struct mlxsw_sp_fib_entry *fib_entry)
1787 {
1788         unsigned int last_ref_count;
1789
1790         do {
1791                 last_ref_count = fib_entry->ref_count;
1792                 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1793         } while (last_ref_count != 1);
1794 }
1795
1796 static int mlxsw_sp_router_fib4_add(struct mlxsw_sp *mlxsw_sp,
1797                                     struct fib_entry_notifier_info *fen_info)
1798 {
1799         struct mlxsw_sp_fib_entry *fib_entry;
1800         struct mlxsw_sp_vr *vr;
1801         int err;
1802
1803         if (mlxsw_sp->router.aborted)
1804                 return 0;
1805
1806         fib_entry = mlxsw_sp_fib_entry_get(mlxsw_sp, fen_info);
1807         if (IS_ERR(fib_entry)) {
1808                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to get FIB4 entry being added.\n");
1809                 return PTR_ERR(fib_entry);
1810         }
1811
1812         if (fib_entry->ref_count != 1)
1813                 return 0;
1814
1815         vr = fib_entry->vr;
1816         err = mlxsw_sp_fib_entry_insert(vr->fib, fib_entry);
1817         if (err) {
1818                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to insert FIB4 entry being added.\n");
1819                 goto err_fib_entry_insert;
1820         }
1821         err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1822         if (err)
1823                 goto err_fib_entry_add;
1824         return 0;
1825
1826 err_fib_entry_add:
1827         mlxsw_sp_fib_entry_remove(vr->fib, fib_entry);
1828 err_fib_entry_insert:
1829         mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1830         return err;
1831 }
1832
1833 static void mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
1834                                      struct fib_entry_notifier_info *fen_info)
1835 {
1836         struct mlxsw_sp_fib_entry *fib_entry;
1837
1838         if (mlxsw_sp->router.aborted)
1839                 return;
1840
1841         fib_entry = mlxsw_sp_fib_entry_find(mlxsw_sp, fen_info);
1842         if (!fib_entry)
1843                 return;
1844
1845         if (fib_entry->ref_count == 1) {
1846                 mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
1847                 mlxsw_sp_fib_entry_remove(fib_entry->vr->fib, fib_entry);
1848         }
1849
1850         mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1851 }
1852
1853 static int mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp)
1854 {
1855         char ralta_pl[MLXSW_REG_RALTA_LEN];
1856         char ralst_pl[MLXSW_REG_RALST_LEN];
1857         char raltb_pl[MLXSW_REG_RALTB_LEN];
1858         char ralue_pl[MLXSW_REG_RALUE_LEN];
1859         int err;
1860
1861         mlxsw_reg_ralta_pack(ralta_pl, true, MLXSW_REG_RALXX_PROTOCOL_IPV4,
1862                              MLXSW_SP_LPM_TREE_MIN);
1863         err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
1864         if (err)
1865                 return err;
1866
1867         mlxsw_reg_ralst_pack(ralst_pl, 0xff, MLXSW_SP_LPM_TREE_MIN);
1868         err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
1869         if (err)
1870                 return err;
1871
1872         mlxsw_reg_raltb_pack(raltb_pl, 0, MLXSW_REG_RALXX_PROTOCOL_IPV4,
1873                              MLXSW_SP_LPM_TREE_MIN);
1874         err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
1875         if (err)
1876                 return err;
1877
1878         mlxsw_reg_ralue_pack4(ralue_pl, MLXSW_SP_L3_PROTO_IPV4,
1879                               MLXSW_REG_RALUE_OP_WRITE_WRITE, 0, 0, 0);
1880         mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
1881         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1882 }
1883
1884 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp)
1885 {
1886         struct mlxsw_sp_fib_entry *fib_entry;
1887         struct mlxsw_sp_fib_entry *tmp;
1888         struct mlxsw_sp_vr *vr;
1889         int i;
1890
1891         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
1892                 vr = &mlxsw_sp->router.vrs[i];
1893
1894                 if (!vr->used)
1895                         continue;
1896
1897                 list_for_each_entry_safe(fib_entry, tmp,
1898                                          &vr->fib->entry_list, list) {
1899                         bool do_break = &tmp->list == &vr->fib->entry_list;
1900
1901                         mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
1902                         mlxsw_sp_fib_entry_remove(fib_entry->vr->fib,
1903                                                   fib_entry);
1904                         mlxsw_sp_fib_entry_put_all(mlxsw_sp, fib_entry);
1905                         if (do_break)
1906                                 break;
1907                 }
1908         }
1909 }
1910
1911 static void mlxsw_sp_router_fib4_abort(struct mlxsw_sp *mlxsw_sp)
1912 {
1913         int err;
1914
1915         if (mlxsw_sp->router.aborted)
1916                 return;
1917         dev_warn(mlxsw_sp->bus_info->dev, "FIB abort triggered. Note that FIB entries are no longer being offloaded to this device.\n");
1918         mlxsw_sp_router_fib_flush(mlxsw_sp);
1919         mlxsw_sp->router.aborted = true;
1920         err = mlxsw_sp_router_set_abort_trap(mlxsw_sp);
1921         if (err)
1922                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set abort trap.\n");
1923 }
1924
1925 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
1926 {
1927         char rgcr_pl[MLXSW_REG_RGCR_LEN];
1928         u64 max_rifs;
1929         int err;
1930
1931         if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_RIFS))
1932                 return -EIO;
1933
1934         max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
1935         mlxsw_sp->rifs = kcalloc(max_rifs, sizeof(struct mlxsw_sp_rif *),
1936                                  GFP_KERNEL);
1937         if (!mlxsw_sp->rifs)
1938                 return -ENOMEM;
1939
1940         mlxsw_reg_rgcr_pack(rgcr_pl, true);
1941         mlxsw_reg_rgcr_max_router_interfaces_set(rgcr_pl, max_rifs);
1942         err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
1943         if (err)
1944                 goto err_rgcr_fail;
1945
1946         return 0;
1947
1948 err_rgcr_fail:
1949         kfree(mlxsw_sp->rifs);
1950         return err;
1951 }
1952
1953 static void __mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
1954 {
1955         char rgcr_pl[MLXSW_REG_RGCR_LEN];
1956         int i;
1957
1958         mlxsw_reg_rgcr_pack(rgcr_pl, false);
1959         mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
1960
1961         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
1962                 WARN_ON_ONCE(mlxsw_sp->rifs[i]);
1963
1964         kfree(mlxsw_sp->rifs);
1965 }
1966
1967 struct mlxsw_sp_fib_event_work {
1968         struct work_struct work;
1969         struct fib_entry_notifier_info fen_info;
1970         struct mlxsw_sp *mlxsw_sp;
1971         unsigned long event;
1972 };
1973
1974 static void mlxsw_sp_router_fib_event_work(struct work_struct *work)
1975 {
1976         struct mlxsw_sp_fib_event_work *fib_work =
1977                 container_of(work, struct mlxsw_sp_fib_event_work, work);
1978         struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
1979         int err;
1980
1981         /* Protect internal structures from changes */
1982         rtnl_lock();
1983         switch (fib_work->event) {
1984         case FIB_EVENT_ENTRY_ADD:
1985                 err = mlxsw_sp_router_fib4_add(mlxsw_sp, &fib_work->fen_info);
1986                 if (err)
1987                         mlxsw_sp_router_fib4_abort(mlxsw_sp);
1988                 fib_info_put(fib_work->fen_info.fi);
1989                 break;
1990         case FIB_EVENT_ENTRY_DEL:
1991                 mlxsw_sp_router_fib4_del(mlxsw_sp, &fib_work->fen_info);
1992                 fib_info_put(fib_work->fen_info.fi);
1993                 break;
1994         case FIB_EVENT_RULE_ADD: /* fall through */
1995         case FIB_EVENT_RULE_DEL:
1996                 mlxsw_sp_router_fib4_abort(mlxsw_sp);
1997                 break;
1998         }
1999         rtnl_unlock();
2000         kfree(fib_work);
2001 }
2002
2003 /* Called with rcu_read_lock() */
2004 static int mlxsw_sp_router_fib_event(struct notifier_block *nb,
2005                                      unsigned long event, void *ptr)
2006 {
2007         struct mlxsw_sp *mlxsw_sp = container_of(nb, struct mlxsw_sp, fib_nb);
2008         struct mlxsw_sp_fib_event_work *fib_work;
2009         struct fib_notifier_info *info = ptr;
2010
2011         if (!net_eq(info->net, &init_net))
2012                 return NOTIFY_DONE;
2013
2014         fib_work = kzalloc(sizeof(*fib_work), GFP_ATOMIC);
2015         if (WARN_ON(!fib_work))
2016                 return NOTIFY_BAD;
2017
2018         INIT_WORK(&fib_work->work, mlxsw_sp_router_fib_event_work);
2019         fib_work->mlxsw_sp = mlxsw_sp;
2020         fib_work->event = event;
2021
2022         switch (event) {
2023         case FIB_EVENT_ENTRY_ADD: /* fall through */
2024         case FIB_EVENT_ENTRY_DEL:
2025                 memcpy(&fib_work->fen_info, ptr, sizeof(fib_work->fen_info));
2026                 /* Take referece on fib_info to prevent it from being
2027                  * freed while work is queued. Release it afterwards.
2028                  */
2029                 fib_info_hold(fib_work->fen_info.fi);
2030                 break;
2031         }
2032
2033         mlxsw_core_schedule_work(&fib_work->work);
2034
2035         return NOTIFY_DONE;
2036 }
2037
2038 static void mlxsw_sp_router_fib_dump_flush(struct notifier_block *nb)
2039 {
2040         struct mlxsw_sp *mlxsw_sp = container_of(nb, struct mlxsw_sp, fib_nb);
2041
2042         /* Flush pending FIB notifications and then flush the device's
2043          * table before requesting another dump. The FIB notification
2044          * block is unregistered, so no need to take RTNL.
2045          */
2046         mlxsw_core_flush_owq();
2047         mlxsw_sp_router_fib_flush(mlxsw_sp);
2048 }
2049
2050 int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
2051 {
2052         int err;
2053
2054         INIT_LIST_HEAD(&mlxsw_sp->router.nexthop_neighs_list);
2055         INIT_LIST_HEAD(&mlxsw_sp->router.nexthop_group_list);
2056         err = __mlxsw_sp_router_init(mlxsw_sp);
2057         if (err)
2058                 return err;
2059
2060         mlxsw_sp_lpm_init(mlxsw_sp);
2061         err = mlxsw_sp_vrs_init(mlxsw_sp);
2062         if (err)
2063                 goto err_vrs_init;
2064
2065         err = mlxsw_sp_neigh_init(mlxsw_sp);
2066         if (err)
2067                 goto err_neigh_init;
2068
2069         mlxsw_sp->fib_nb.notifier_call = mlxsw_sp_router_fib_event;
2070         err = register_fib_notifier(&mlxsw_sp->fib_nb,
2071                                     mlxsw_sp_router_fib_dump_flush);
2072         if (err)
2073                 goto err_register_fib_notifier;
2074
2075         return 0;
2076
2077 err_register_fib_notifier:
2078         mlxsw_sp_neigh_fini(mlxsw_sp);
2079 err_neigh_init:
2080         mlxsw_sp_vrs_fini(mlxsw_sp);
2081 err_vrs_init:
2082         __mlxsw_sp_router_fini(mlxsw_sp);
2083         return err;
2084 }
2085
2086 void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
2087 {
2088         unregister_fib_notifier(&mlxsw_sp->fib_nb);
2089         mlxsw_sp_neigh_fini(mlxsw_sp);
2090         mlxsw_sp_vrs_fini(mlxsw_sp);
2091         __mlxsw_sp_router_fini(mlxsw_sp);
2092 }