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>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions are met:
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.
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.
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.
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>
46 #include <net/ip_fib.h>
52 #define mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) \
53 for_each_set_bit(prefix, (prefix_usage)->b, MLXSW_SP_PREFIX_COUNT)
56 mlxsw_sp_prefix_usage_subset(struct mlxsw_sp_prefix_usage *prefix_usage1,
57 struct mlxsw_sp_prefix_usage *prefix_usage2)
61 mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage1) {
62 if (!test_bit(prefix, prefix_usage2->b))
69 mlxsw_sp_prefix_usage_eq(struct mlxsw_sp_prefix_usage *prefix_usage1,
70 struct mlxsw_sp_prefix_usage *prefix_usage2)
72 return !memcmp(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
76 mlxsw_sp_prefix_usage_none(struct mlxsw_sp_prefix_usage *prefix_usage)
78 struct mlxsw_sp_prefix_usage prefix_usage_none = {{ 0 } };
80 return mlxsw_sp_prefix_usage_eq(prefix_usage, &prefix_usage_none);
84 mlxsw_sp_prefix_usage_cpy(struct mlxsw_sp_prefix_usage *prefix_usage1,
85 struct mlxsw_sp_prefix_usage *prefix_usage2)
87 memcpy(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
91 mlxsw_sp_prefix_usage_zero(struct mlxsw_sp_prefix_usage *prefix_usage)
93 memset(prefix_usage, 0, sizeof(*prefix_usage));
97 mlxsw_sp_prefix_usage_set(struct mlxsw_sp_prefix_usage *prefix_usage,
98 unsigned char prefix_len)
100 set_bit(prefix_len, prefix_usage->b);
104 mlxsw_sp_prefix_usage_clear(struct mlxsw_sp_prefix_usage *prefix_usage,
105 unsigned char prefix_len)
107 clear_bit(prefix_len, prefix_usage->b);
110 struct mlxsw_sp_fib_key {
111 struct net_device *dev;
112 unsigned char addr[sizeof(struct in6_addr)];
113 unsigned char prefix_len;
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,
122 struct mlxsw_sp_nexthop_group;
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 struct mlxsw_sp_vr *vr;
131 struct list_head nexthop_group_node;
132 struct mlxsw_sp_nexthop_group *nh_group;
136 struct mlxsw_sp_fib {
137 struct rhashtable ht;
138 struct list_head entry_list;
139 unsigned long prefix_ref_count[MLXSW_SP_PREFIX_COUNT];
140 struct mlxsw_sp_prefix_usage prefix_usage;
143 static const struct rhashtable_params mlxsw_sp_fib_ht_params = {
144 .key_offset = offsetof(struct mlxsw_sp_fib_entry, key),
145 .head_offset = offsetof(struct mlxsw_sp_fib_entry, ht_node),
146 .key_len = sizeof(struct mlxsw_sp_fib_key),
147 .automatic_shrinking = true,
150 static int mlxsw_sp_fib_entry_insert(struct mlxsw_sp_fib *fib,
151 struct mlxsw_sp_fib_entry *fib_entry)
153 unsigned char prefix_len = fib_entry->key.prefix_len;
156 err = rhashtable_insert_fast(&fib->ht, &fib_entry->ht_node,
157 mlxsw_sp_fib_ht_params);
160 list_add_tail(&fib_entry->list, &fib->entry_list);
161 if (fib->prefix_ref_count[prefix_len]++ == 0)
162 mlxsw_sp_prefix_usage_set(&fib->prefix_usage, prefix_len);
166 static void mlxsw_sp_fib_entry_remove(struct mlxsw_sp_fib *fib,
167 struct mlxsw_sp_fib_entry *fib_entry)
169 unsigned char prefix_len = fib_entry->key.prefix_len;
171 if (--fib->prefix_ref_count[prefix_len] == 0)
172 mlxsw_sp_prefix_usage_clear(&fib->prefix_usage, prefix_len);
173 list_del(&fib_entry->list);
174 rhashtable_remove_fast(&fib->ht, &fib_entry->ht_node,
175 mlxsw_sp_fib_ht_params);
178 static struct mlxsw_sp_fib_entry *
179 mlxsw_sp_fib_entry_create(struct mlxsw_sp_fib *fib, const void *addr,
180 size_t addr_len, unsigned char prefix_len,
181 struct net_device *dev)
183 struct mlxsw_sp_fib_entry *fib_entry;
185 fib_entry = kzalloc(sizeof(*fib_entry), GFP_KERNEL);
188 fib_entry->key.dev = dev;
189 memcpy(fib_entry->key.addr, addr, addr_len);
190 fib_entry->key.prefix_len = prefix_len;
194 static void mlxsw_sp_fib_entry_destroy(struct mlxsw_sp_fib_entry *fib_entry)
199 static struct mlxsw_sp_fib_entry *
200 mlxsw_sp_fib_entry_lookup(struct mlxsw_sp_fib *fib, const void *addr,
201 size_t addr_len, unsigned char prefix_len,
202 struct net_device *dev)
204 struct mlxsw_sp_fib_key key;
206 memset(&key, 0, sizeof(key));
208 memcpy(key.addr, addr, addr_len);
209 key.prefix_len = prefix_len;
210 return rhashtable_lookup_fast(&fib->ht, &key, mlxsw_sp_fib_ht_params);
213 static struct mlxsw_sp_fib *mlxsw_sp_fib_create(void)
215 struct mlxsw_sp_fib *fib;
218 fib = kzalloc(sizeof(*fib), GFP_KERNEL);
220 return ERR_PTR(-ENOMEM);
221 err = rhashtable_init(&fib->ht, &mlxsw_sp_fib_ht_params);
223 goto err_rhashtable_init;
224 INIT_LIST_HEAD(&fib->entry_list);
232 static void mlxsw_sp_fib_destroy(struct mlxsw_sp_fib *fib)
234 rhashtable_destroy(&fib->ht);
238 static struct mlxsw_sp_lpm_tree *
239 mlxsw_sp_lpm_tree_find_unused(struct mlxsw_sp *mlxsw_sp, bool one_reserved)
241 static struct mlxsw_sp_lpm_tree *lpm_tree;
244 for (i = 0; i < MLXSW_SP_LPM_TREE_COUNT; i++) {
245 lpm_tree = &mlxsw_sp->router.lpm_trees[i];
246 if (lpm_tree->ref_count == 0) {
248 one_reserved = false;
256 static int mlxsw_sp_lpm_tree_alloc(struct mlxsw_sp *mlxsw_sp,
257 struct mlxsw_sp_lpm_tree *lpm_tree)
259 char ralta_pl[MLXSW_REG_RALTA_LEN];
261 mlxsw_reg_ralta_pack(ralta_pl, true,
262 (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
264 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
267 static int mlxsw_sp_lpm_tree_free(struct mlxsw_sp *mlxsw_sp,
268 struct mlxsw_sp_lpm_tree *lpm_tree)
270 char ralta_pl[MLXSW_REG_RALTA_LEN];
272 mlxsw_reg_ralta_pack(ralta_pl, false,
273 (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
275 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
279 mlxsw_sp_lpm_tree_left_struct_set(struct mlxsw_sp *mlxsw_sp,
280 struct mlxsw_sp_prefix_usage *prefix_usage,
281 struct mlxsw_sp_lpm_tree *lpm_tree)
283 char ralst_pl[MLXSW_REG_RALST_LEN];
286 u8 last_prefix = MLXSW_REG_RALST_BIN_NO_CHILD;
288 mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage)
291 mlxsw_reg_ralst_pack(ralst_pl, root_bin, lpm_tree->id);
292 mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) {
295 mlxsw_reg_ralst_bin_pack(ralst_pl, prefix, last_prefix,
296 MLXSW_REG_RALST_BIN_NO_CHILD);
297 last_prefix = prefix;
299 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
302 static struct mlxsw_sp_lpm_tree *
303 mlxsw_sp_lpm_tree_create(struct mlxsw_sp *mlxsw_sp,
304 struct mlxsw_sp_prefix_usage *prefix_usage,
305 enum mlxsw_sp_l3proto proto, bool one_reserved)
307 struct mlxsw_sp_lpm_tree *lpm_tree;
310 lpm_tree = mlxsw_sp_lpm_tree_find_unused(mlxsw_sp, one_reserved);
312 return ERR_PTR(-EBUSY);
313 lpm_tree->proto = proto;
314 err = mlxsw_sp_lpm_tree_alloc(mlxsw_sp, lpm_tree);
318 err = mlxsw_sp_lpm_tree_left_struct_set(mlxsw_sp, prefix_usage,
321 goto err_left_struct_set;
322 memcpy(&lpm_tree->prefix_usage, prefix_usage,
323 sizeof(lpm_tree->prefix_usage));
327 mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
331 static int mlxsw_sp_lpm_tree_destroy(struct mlxsw_sp *mlxsw_sp,
332 struct mlxsw_sp_lpm_tree *lpm_tree)
334 return mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
337 static struct mlxsw_sp_lpm_tree *
338 mlxsw_sp_lpm_tree_get(struct mlxsw_sp *mlxsw_sp,
339 struct mlxsw_sp_prefix_usage *prefix_usage,
340 enum mlxsw_sp_l3proto proto, bool one_reserved)
342 struct mlxsw_sp_lpm_tree *lpm_tree;
345 for (i = 0; i < MLXSW_SP_LPM_TREE_COUNT; i++) {
346 lpm_tree = &mlxsw_sp->router.lpm_trees[i];
347 if (lpm_tree->ref_count != 0 &&
348 lpm_tree->proto == proto &&
349 mlxsw_sp_prefix_usage_eq(&lpm_tree->prefix_usage,
353 lpm_tree = mlxsw_sp_lpm_tree_create(mlxsw_sp, prefix_usage,
354 proto, one_reserved);
355 if (IS_ERR(lpm_tree))
359 lpm_tree->ref_count++;
363 static int mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
364 struct mlxsw_sp_lpm_tree *lpm_tree)
366 if (--lpm_tree->ref_count == 0)
367 return mlxsw_sp_lpm_tree_destroy(mlxsw_sp, lpm_tree);
371 static void mlxsw_sp_lpm_init(struct mlxsw_sp *mlxsw_sp)
373 struct mlxsw_sp_lpm_tree *lpm_tree;
376 for (i = 0; i < MLXSW_SP_LPM_TREE_COUNT; i++) {
377 lpm_tree = &mlxsw_sp->router.lpm_trees[i];
378 lpm_tree->id = i + MLXSW_SP_LPM_TREE_MIN;
382 static struct mlxsw_sp_vr *mlxsw_sp_vr_find_unused(struct mlxsw_sp *mlxsw_sp)
384 struct mlxsw_sp_vr *vr;
387 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
388 vr = &mlxsw_sp->router.vrs[i];
395 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
396 struct mlxsw_sp_vr *vr)
398 char raltb_pl[MLXSW_REG_RALTB_LEN];
400 mlxsw_reg_raltb_pack(raltb_pl, vr->id,
401 (enum mlxsw_reg_ralxx_protocol) vr->proto,
403 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
406 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
407 struct mlxsw_sp_vr *vr)
409 char raltb_pl[MLXSW_REG_RALTB_LEN];
411 /* Bind to tree 0 which is default */
412 mlxsw_reg_raltb_pack(raltb_pl, vr->id,
413 (enum mlxsw_reg_ralxx_protocol) vr->proto, 0);
414 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
417 static u32 mlxsw_sp_fix_tb_id(u32 tb_id)
419 /* For our purpose, squash main and local table into one */
420 if (tb_id == RT_TABLE_LOCAL)
421 tb_id = RT_TABLE_MAIN;
425 static struct mlxsw_sp_vr *mlxsw_sp_vr_find(struct mlxsw_sp *mlxsw_sp,
427 enum mlxsw_sp_l3proto proto)
429 struct mlxsw_sp_vr *vr;
432 tb_id = mlxsw_sp_fix_tb_id(tb_id);
434 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
435 vr = &mlxsw_sp->router.vrs[i];
436 if (vr->used && vr->proto == proto && vr->tb_id == tb_id)
442 static struct mlxsw_sp_vr *mlxsw_sp_vr_create(struct mlxsw_sp *mlxsw_sp,
443 unsigned char prefix_len,
445 enum mlxsw_sp_l3proto proto)
447 struct mlxsw_sp_prefix_usage req_prefix_usage;
448 struct mlxsw_sp_lpm_tree *lpm_tree;
449 struct mlxsw_sp_vr *vr;
452 vr = mlxsw_sp_vr_find_unused(mlxsw_sp);
454 return ERR_PTR(-EBUSY);
455 vr->fib = mlxsw_sp_fib_create();
457 return ERR_CAST(vr->fib);
461 mlxsw_sp_prefix_usage_zero(&req_prefix_usage);
462 mlxsw_sp_prefix_usage_set(&req_prefix_usage, prefix_len);
463 lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
465 if (IS_ERR(lpm_tree)) {
466 err = PTR_ERR(lpm_tree);
469 vr->lpm_tree = lpm_tree;
470 err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, vr);
478 mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
480 mlxsw_sp_fib_destroy(vr->fib);
485 static void mlxsw_sp_vr_destroy(struct mlxsw_sp *mlxsw_sp,
486 struct mlxsw_sp_vr *vr)
488 mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, vr);
489 mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
490 mlxsw_sp_fib_destroy(vr->fib);
495 mlxsw_sp_vr_lpm_tree_check(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr,
496 struct mlxsw_sp_prefix_usage *req_prefix_usage)
498 struct mlxsw_sp_lpm_tree *lpm_tree;
500 if (mlxsw_sp_prefix_usage_eq(req_prefix_usage,
501 &vr->lpm_tree->prefix_usage))
504 lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, req_prefix_usage,
506 if (IS_ERR(lpm_tree)) {
507 /* We failed to get a tree according to the required
508 * prefix usage. However, the current tree might be still good
509 * for us if our requirement is subset of the prefixes used
512 if (mlxsw_sp_prefix_usage_subset(req_prefix_usage,
513 &vr->lpm_tree->prefix_usage))
515 return PTR_ERR(lpm_tree);
518 mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, vr);
519 mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
520 vr->lpm_tree = lpm_tree;
521 return mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, vr);
524 static struct mlxsw_sp_vr *mlxsw_sp_vr_get(struct mlxsw_sp *mlxsw_sp,
525 unsigned char prefix_len,
527 enum mlxsw_sp_l3proto proto)
529 struct mlxsw_sp_vr *vr;
532 tb_id = mlxsw_sp_fix_tb_id(tb_id);
533 vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id, proto);
535 vr = mlxsw_sp_vr_create(mlxsw_sp, prefix_len, tb_id, proto);
539 struct mlxsw_sp_prefix_usage req_prefix_usage;
541 mlxsw_sp_prefix_usage_cpy(&req_prefix_usage,
542 &vr->fib->prefix_usage);
543 mlxsw_sp_prefix_usage_set(&req_prefix_usage, prefix_len);
544 /* Need to replace LPM tree in case new prefix is required. */
545 err = mlxsw_sp_vr_lpm_tree_check(mlxsw_sp, vr,
553 static void mlxsw_sp_vr_put(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr)
555 /* Destroy virtual router entity in case the associated FIB is empty
556 * and allow it to be used for other tables in future. Otherwise,
557 * check if some prefix usage did not disappear and change tree if
558 * that is the case. Note that in case new, smaller tree cannot be
559 * allocated, the original one will be kept being used.
561 if (mlxsw_sp_prefix_usage_none(&vr->fib->prefix_usage))
562 mlxsw_sp_vr_destroy(mlxsw_sp, vr);
564 mlxsw_sp_vr_lpm_tree_check(mlxsw_sp, vr,
565 &vr->fib->prefix_usage);
568 static int mlxsw_sp_vrs_init(struct mlxsw_sp *mlxsw_sp)
570 struct mlxsw_sp_vr *vr;
574 if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_VRS))
577 max_vrs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS);
578 mlxsw_sp->router.vrs = kcalloc(max_vrs, sizeof(struct mlxsw_sp_vr),
580 if (!mlxsw_sp->router.vrs)
583 for (i = 0; i < max_vrs; i++) {
584 vr = &mlxsw_sp->router.vrs[i];
591 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp);
593 static void mlxsw_sp_vrs_fini(struct mlxsw_sp *mlxsw_sp)
595 /* At this stage we're guaranteed not to have new incoming
596 * FIB notifications and the work queue is free from FIBs
597 * sitting on top of mlxsw netdevs. However, we can still
598 * have other FIBs queued. Flush the queue before flushing
599 * the device's tables. No need for locks, as we're the only
602 mlxsw_core_flush_owq();
603 mlxsw_sp_router_fib_flush(mlxsw_sp);
604 kfree(mlxsw_sp->router.vrs);
607 struct mlxsw_sp_neigh_key {
611 struct mlxsw_sp_neigh_entry {
612 struct rhash_head ht_node;
613 struct mlxsw_sp_neigh_key key;
616 unsigned char ha[ETH_ALEN];
617 struct list_head nexthop_list; /* list of nexthops using
620 struct list_head nexthop_neighs_list_node;
623 static const struct rhashtable_params mlxsw_sp_neigh_ht_params = {
624 .key_offset = offsetof(struct mlxsw_sp_neigh_entry, key),
625 .head_offset = offsetof(struct mlxsw_sp_neigh_entry, ht_node),
626 .key_len = sizeof(struct mlxsw_sp_neigh_key),
629 static struct mlxsw_sp_neigh_entry *
630 mlxsw_sp_neigh_entry_alloc(struct mlxsw_sp *mlxsw_sp, struct neighbour *n,
633 struct mlxsw_sp_neigh_entry *neigh_entry;
635 neigh_entry = kzalloc(sizeof(*neigh_entry), GFP_KERNEL);
639 neigh_entry->key.n = n;
640 neigh_entry->rif = rif;
641 INIT_LIST_HEAD(&neigh_entry->nexthop_list);
646 static void mlxsw_sp_neigh_entry_free(struct mlxsw_sp_neigh_entry *neigh_entry)
652 mlxsw_sp_neigh_entry_insert(struct mlxsw_sp *mlxsw_sp,
653 struct mlxsw_sp_neigh_entry *neigh_entry)
655 return rhashtable_insert_fast(&mlxsw_sp->router.neigh_ht,
656 &neigh_entry->ht_node,
657 mlxsw_sp_neigh_ht_params);
661 mlxsw_sp_neigh_entry_remove(struct mlxsw_sp *mlxsw_sp,
662 struct mlxsw_sp_neigh_entry *neigh_entry)
664 rhashtable_remove_fast(&mlxsw_sp->router.neigh_ht,
665 &neigh_entry->ht_node,
666 mlxsw_sp_neigh_ht_params);
669 static struct mlxsw_sp_neigh_entry *
670 mlxsw_sp_neigh_entry_create(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
672 struct mlxsw_sp_neigh_entry *neigh_entry;
673 struct mlxsw_sp_rif *r;
676 r = mlxsw_sp_rif_find_by_dev(mlxsw_sp, n->dev);
678 return ERR_PTR(-EINVAL);
680 neigh_entry = mlxsw_sp_neigh_entry_alloc(mlxsw_sp, n, r->rif);
682 return ERR_PTR(-ENOMEM);
684 err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
686 goto err_neigh_entry_insert;
690 err_neigh_entry_insert:
691 mlxsw_sp_neigh_entry_free(neigh_entry);
696 mlxsw_sp_neigh_entry_destroy(struct mlxsw_sp *mlxsw_sp,
697 struct mlxsw_sp_neigh_entry *neigh_entry)
699 mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
700 mlxsw_sp_neigh_entry_free(neigh_entry);
703 static struct mlxsw_sp_neigh_entry *
704 mlxsw_sp_neigh_entry_lookup(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
706 struct mlxsw_sp_neigh_key key;
709 return rhashtable_lookup_fast(&mlxsw_sp->router.neigh_ht,
710 &key, mlxsw_sp_neigh_ht_params);
714 mlxsw_sp_router_neighs_update_interval_init(struct mlxsw_sp *mlxsw_sp)
716 unsigned long interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME);
718 mlxsw_sp->router.neighs_update.interval = jiffies_to_msecs(interval);
721 static void mlxsw_sp_router_neigh_ent_ipv4_process(struct mlxsw_sp *mlxsw_sp,
725 struct net_device *dev;
731 mlxsw_reg_rauhtd_ent_ipv4_unpack(rauhtd_pl, ent_index, &rif, &dip);
733 if (!mlxsw_sp->rifs[rif]) {
734 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
739 dev = mlxsw_sp->rifs[rif]->dev;
740 n = neigh_lookup(&arp_tbl, &dipn, dev);
742 netdev_err(dev, "Failed to find matching neighbour for IP=%pI4h\n",
747 netdev_dbg(dev, "Updating neighbour with IP=%pI4h\n", &dip);
748 neigh_event_send(n, NULL);
752 static void mlxsw_sp_router_neigh_rec_ipv4_process(struct mlxsw_sp *mlxsw_sp,
759 num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
761 /* Hardware starts counting at 0, so add 1. */
764 /* Each record consists of several neighbour entries. */
765 for (i = 0; i < num_entries; i++) {
768 ent_index = rec_index * MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC + i;
769 mlxsw_sp_router_neigh_ent_ipv4_process(mlxsw_sp, rauhtd_pl,
775 static void mlxsw_sp_router_neigh_rec_process(struct mlxsw_sp *mlxsw_sp,
776 char *rauhtd_pl, int rec_index)
778 switch (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, rec_index)) {
779 case MLXSW_REG_RAUHTD_TYPE_IPV4:
780 mlxsw_sp_router_neigh_rec_ipv4_process(mlxsw_sp, rauhtd_pl,
783 case MLXSW_REG_RAUHTD_TYPE_IPV6:
789 static bool mlxsw_sp_router_rauhtd_is_full(char *rauhtd_pl)
791 u8 num_rec, last_rec_index, num_entries;
793 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
794 last_rec_index = num_rec - 1;
796 if (num_rec < MLXSW_REG_RAUHTD_REC_MAX_NUM)
798 if (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, last_rec_index) ==
799 MLXSW_REG_RAUHTD_TYPE_IPV6)
802 num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
804 if (++num_entries == MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC)
809 static int mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp)
815 rauhtd_pl = kmalloc(MLXSW_REG_RAUHTD_LEN, GFP_KERNEL);
819 /* Make sure the neighbour's netdev isn't removed in the
824 mlxsw_reg_rauhtd_pack(rauhtd_pl, MLXSW_REG_RAUHTD_TYPE_IPV4);
825 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(rauhtd),
828 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Failed to dump neighbour talbe\n");
831 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
832 for (i = 0; i < num_rec; i++)
833 mlxsw_sp_router_neigh_rec_process(mlxsw_sp, rauhtd_pl,
835 } while (mlxsw_sp_router_rauhtd_is_full(rauhtd_pl));
842 static void mlxsw_sp_router_neighs_update_nh(struct mlxsw_sp *mlxsw_sp)
844 struct mlxsw_sp_neigh_entry *neigh_entry;
846 /* Take RTNL mutex here to prevent lists from changes */
848 list_for_each_entry(neigh_entry, &mlxsw_sp->router.nexthop_neighs_list,
849 nexthop_neighs_list_node)
850 /* If this neigh have nexthops, make the kernel think this neigh
851 * is active regardless of the traffic.
853 neigh_event_send(neigh_entry->key.n, NULL);
858 mlxsw_sp_router_neighs_update_work_schedule(struct mlxsw_sp *mlxsw_sp)
860 unsigned long interval = mlxsw_sp->router.neighs_update.interval;
862 mlxsw_core_schedule_dw(&mlxsw_sp->router.neighs_update.dw,
863 msecs_to_jiffies(interval));
866 static void mlxsw_sp_router_neighs_update_work(struct work_struct *work)
868 struct mlxsw_sp *mlxsw_sp = container_of(work, struct mlxsw_sp,
869 router.neighs_update.dw.work);
872 err = mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp);
874 dev_err(mlxsw_sp->bus_info->dev, "Could not update kernel for neigh activity");
876 mlxsw_sp_router_neighs_update_nh(mlxsw_sp);
878 mlxsw_sp_router_neighs_update_work_schedule(mlxsw_sp);
881 static void mlxsw_sp_router_probe_unresolved_nexthops(struct work_struct *work)
883 struct mlxsw_sp_neigh_entry *neigh_entry;
884 struct mlxsw_sp *mlxsw_sp = container_of(work, struct mlxsw_sp,
885 router.nexthop_probe_dw.work);
887 /* Iterate over nexthop neighbours, find those who are unresolved and
888 * send arp on them. This solves the chicken-egg problem when
889 * the nexthop wouldn't get offloaded until the neighbor is resolved
890 * but it wouldn't get resolved ever in case traffic is flowing in HW
891 * using different nexthop.
893 * Take RTNL mutex here to prevent lists from changes.
896 list_for_each_entry(neigh_entry, &mlxsw_sp->router.nexthop_neighs_list,
897 nexthop_neighs_list_node)
898 if (!neigh_entry->connected)
899 neigh_event_send(neigh_entry->key.n, NULL);
902 mlxsw_core_schedule_dw(&mlxsw_sp->router.nexthop_probe_dw,
903 MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL);
907 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
908 struct mlxsw_sp_neigh_entry *neigh_entry,
911 static enum mlxsw_reg_rauht_op mlxsw_sp_rauht_op(bool adding)
913 return adding ? MLXSW_REG_RAUHT_OP_WRITE_ADD :
914 MLXSW_REG_RAUHT_OP_WRITE_DELETE;
918 mlxsw_sp_router_neigh_entry_op4(struct mlxsw_sp *mlxsw_sp,
919 struct mlxsw_sp_neigh_entry *neigh_entry,
920 enum mlxsw_reg_rauht_op op)
922 struct neighbour *n = neigh_entry->key.n;
923 u32 dip = ntohl(*((__be32 *) n->primary_key));
924 char rauht_pl[MLXSW_REG_RAUHT_LEN];
926 mlxsw_reg_rauht_pack4(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
928 mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
932 mlxsw_sp_neigh_entry_update(struct mlxsw_sp *mlxsw_sp,
933 struct mlxsw_sp_neigh_entry *neigh_entry,
936 if (!adding && !neigh_entry->connected)
938 neigh_entry->connected = adding;
939 if (neigh_entry->key.n->tbl == &arp_tbl)
940 mlxsw_sp_router_neigh_entry_op4(mlxsw_sp, neigh_entry,
941 mlxsw_sp_rauht_op(adding));
946 struct mlxsw_sp_neigh_event_work {
947 struct work_struct work;
948 struct mlxsw_sp *mlxsw_sp;
952 static void mlxsw_sp_router_neigh_event_work(struct work_struct *work)
954 struct mlxsw_sp_neigh_event_work *neigh_work =
955 container_of(work, struct mlxsw_sp_neigh_event_work, work);
956 struct mlxsw_sp *mlxsw_sp = neigh_work->mlxsw_sp;
957 struct mlxsw_sp_neigh_entry *neigh_entry;
958 struct neighbour *n = neigh_work->n;
959 unsigned char ha[ETH_ALEN];
960 bool entry_connected;
963 /* If these parameters are changed after we release the lock,
964 * then we are guaranteed to receive another event letting us
967 read_lock_bh(&n->lock);
968 memcpy(ha, n->ha, ETH_ALEN);
969 nud_state = n->nud_state;
971 read_unlock_bh(&n->lock);
974 entry_connected = nud_state & NUD_VALID && !dead;
975 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
976 if (!entry_connected && !neigh_entry)
979 neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
980 if (IS_ERR(neigh_entry))
984 memcpy(neigh_entry->ha, ha, ETH_ALEN);
985 mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, entry_connected);
986 mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, !entry_connected);
988 if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
989 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
997 int mlxsw_sp_router_netevent_event(struct notifier_block *unused,
998 unsigned long event, void *ptr)
1000 struct mlxsw_sp_neigh_event_work *neigh_work;
1001 struct mlxsw_sp_port *mlxsw_sp_port;
1002 struct mlxsw_sp *mlxsw_sp;
1003 unsigned long interval;
1004 struct neigh_parms *p;
1005 struct neighbour *n;
1008 case NETEVENT_DELAY_PROBE_TIME_UPDATE:
1011 /* We don't care about changes in the default table. */
1012 if (!p->dev || p->tbl != &arp_tbl)
1015 /* We are in atomic context and can't take RTNL mutex,
1016 * so use RCU variant to walk the device chain.
1018 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(p->dev);
1022 mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1023 interval = jiffies_to_msecs(NEIGH_VAR(p, DELAY_PROBE_TIME));
1024 mlxsw_sp->router.neighs_update.interval = interval;
1026 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1028 case NETEVENT_NEIGH_UPDATE:
1031 if (n->tbl != &arp_tbl)
1034 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(n->dev);
1038 neigh_work = kzalloc(sizeof(*neigh_work), GFP_ATOMIC);
1040 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1044 INIT_WORK(&neigh_work->work, mlxsw_sp_router_neigh_event_work);
1045 neigh_work->mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1048 /* Take a reference to ensure the neighbour won't be
1049 * destructed until we drop the reference in delayed
1053 mlxsw_core_schedule_work(&neigh_work->work);
1054 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1061 static int mlxsw_sp_neigh_init(struct mlxsw_sp *mlxsw_sp)
1065 err = rhashtable_init(&mlxsw_sp->router.neigh_ht,
1066 &mlxsw_sp_neigh_ht_params);
1070 /* Initialize the polling interval according to the default
1073 mlxsw_sp_router_neighs_update_interval_init(mlxsw_sp);
1075 /* Create the delayed works for the activity_update */
1076 INIT_DELAYED_WORK(&mlxsw_sp->router.neighs_update.dw,
1077 mlxsw_sp_router_neighs_update_work);
1078 INIT_DELAYED_WORK(&mlxsw_sp->router.nexthop_probe_dw,
1079 mlxsw_sp_router_probe_unresolved_nexthops);
1080 mlxsw_core_schedule_dw(&mlxsw_sp->router.neighs_update.dw, 0);
1081 mlxsw_core_schedule_dw(&mlxsw_sp->router.nexthop_probe_dw, 0);
1085 static void mlxsw_sp_neigh_fini(struct mlxsw_sp *mlxsw_sp)
1087 cancel_delayed_work_sync(&mlxsw_sp->router.neighs_update.dw);
1088 cancel_delayed_work_sync(&mlxsw_sp->router.nexthop_probe_dw);
1089 rhashtable_destroy(&mlxsw_sp->router.neigh_ht);
1092 struct mlxsw_sp_nexthop_key {
1093 struct fib_nh *fib_nh;
1096 struct mlxsw_sp_nexthop {
1097 struct list_head neigh_list_node; /* member of neigh entry list */
1098 struct mlxsw_sp_nexthop_group *nh_grp; /* pointer back to the group
1101 struct rhash_head ht_node;
1102 struct mlxsw_sp_nexthop_key key;
1103 struct mlxsw_sp_rif *r;
1104 u8 should_offload:1, /* set indicates this neigh is connected and
1105 * should be put to KVD linear area of this group.
1107 offloaded:1, /* set in case the neigh is actually put into
1108 * KVD linear area of this group.
1110 update:1; /* set indicates that MAC of this neigh should be
1113 struct mlxsw_sp_neigh_entry *neigh_entry;
1116 struct mlxsw_sp_nexthop_group_key {
1117 struct fib_info *fi;
1120 struct mlxsw_sp_nexthop_group {
1121 struct rhash_head ht_node;
1122 struct list_head fib_list; /* list of fib entries that use this group */
1123 struct mlxsw_sp_nexthop_group_key key;
1124 u8 adj_index_valid:1,
1125 gateway:1; /* routes using the group use a gateway */
1129 struct mlxsw_sp_nexthop nexthops[0];
1130 #define nh_rif nexthops[0].r
1133 static const struct rhashtable_params mlxsw_sp_nexthop_group_ht_params = {
1134 .key_offset = offsetof(struct mlxsw_sp_nexthop_group, key),
1135 .head_offset = offsetof(struct mlxsw_sp_nexthop_group, ht_node),
1136 .key_len = sizeof(struct mlxsw_sp_nexthop_group_key),
1139 static int mlxsw_sp_nexthop_group_insert(struct mlxsw_sp *mlxsw_sp,
1140 struct mlxsw_sp_nexthop_group *nh_grp)
1142 return rhashtable_insert_fast(&mlxsw_sp->router.nexthop_group_ht,
1144 mlxsw_sp_nexthop_group_ht_params);
1147 static void mlxsw_sp_nexthop_group_remove(struct mlxsw_sp *mlxsw_sp,
1148 struct mlxsw_sp_nexthop_group *nh_grp)
1150 rhashtable_remove_fast(&mlxsw_sp->router.nexthop_group_ht,
1152 mlxsw_sp_nexthop_group_ht_params);
1155 static struct mlxsw_sp_nexthop_group *
1156 mlxsw_sp_nexthop_group_lookup(struct mlxsw_sp *mlxsw_sp,
1157 struct mlxsw_sp_nexthop_group_key key)
1159 return rhashtable_lookup_fast(&mlxsw_sp->router.nexthop_group_ht, &key,
1160 mlxsw_sp_nexthop_group_ht_params);
1163 static const struct rhashtable_params mlxsw_sp_nexthop_ht_params = {
1164 .key_offset = offsetof(struct mlxsw_sp_nexthop, key),
1165 .head_offset = offsetof(struct mlxsw_sp_nexthop, ht_node),
1166 .key_len = sizeof(struct mlxsw_sp_nexthop_key),
1169 static int mlxsw_sp_nexthop_insert(struct mlxsw_sp *mlxsw_sp,
1170 struct mlxsw_sp_nexthop *nh)
1172 return rhashtable_insert_fast(&mlxsw_sp->router.nexthop_ht,
1173 &nh->ht_node, mlxsw_sp_nexthop_ht_params);
1176 static void mlxsw_sp_nexthop_remove(struct mlxsw_sp *mlxsw_sp,
1177 struct mlxsw_sp_nexthop *nh)
1179 rhashtable_remove_fast(&mlxsw_sp->router.nexthop_ht, &nh->ht_node,
1180 mlxsw_sp_nexthop_ht_params);
1183 static int mlxsw_sp_adj_index_mass_update_vr(struct mlxsw_sp *mlxsw_sp,
1184 struct mlxsw_sp_vr *vr,
1185 u32 adj_index, u16 ecmp_size,
1189 char raleu_pl[MLXSW_REG_RALEU_LEN];
1191 mlxsw_reg_raleu_pack(raleu_pl,
1192 (enum mlxsw_reg_ralxx_protocol) vr->proto, vr->id,
1193 adj_index, ecmp_size, new_adj_index,
1195 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raleu), raleu_pl);
1198 static int mlxsw_sp_adj_index_mass_update(struct mlxsw_sp *mlxsw_sp,
1199 struct mlxsw_sp_nexthop_group *nh_grp,
1200 u32 old_adj_index, u16 old_ecmp_size)
1202 struct mlxsw_sp_fib_entry *fib_entry;
1203 struct mlxsw_sp_vr *vr = NULL;
1206 list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
1207 if (vr == fib_entry->vr)
1210 err = mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp, vr,
1221 static int mlxsw_sp_nexthop_mac_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
1222 struct mlxsw_sp_nexthop *nh)
1224 struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
1225 char ratr_pl[MLXSW_REG_RATR_LEN];
1227 mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
1228 true, adj_index, neigh_entry->rif);
1229 mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
1230 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
1234 mlxsw_sp_nexthop_group_mac_update(struct mlxsw_sp *mlxsw_sp,
1235 struct mlxsw_sp_nexthop_group *nh_grp,
1238 u32 adj_index = nh_grp->adj_index; /* base */
1239 struct mlxsw_sp_nexthop *nh;
1243 for (i = 0; i < nh_grp->count; i++) {
1244 nh = &nh_grp->nexthops[i];
1246 if (!nh->should_offload) {
1251 if (nh->update || reallocate) {
1252 err = mlxsw_sp_nexthop_mac_update(mlxsw_sp,
1264 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1265 struct mlxsw_sp_fib_entry *fib_entry);
1268 mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
1269 struct mlxsw_sp_nexthop_group *nh_grp)
1271 struct mlxsw_sp_fib_entry *fib_entry;
1274 list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
1275 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1283 mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
1284 struct mlxsw_sp_nexthop_group *nh_grp)
1286 struct mlxsw_sp_nexthop *nh;
1287 bool offload_change = false;
1290 bool old_adj_index_valid;
1297 if (!nh_grp->gateway) {
1298 mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
1302 for (i = 0; i < nh_grp->count; i++) {
1303 nh = &nh_grp->nexthops[i];
1305 if (nh->should_offload ^ nh->offloaded) {
1306 offload_change = true;
1307 if (nh->should_offload)
1310 if (nh->should_offload)
1313 if (!offload_change) {
1314 /* Nothing was added or removed, so no need to reallocate. Just
1315 * update MAC on existing adjacency indexes.
1317 err = mlxsw_sp_nexthop_group_mac_update(mlxsw_sp, nh_grp,
1320 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
1326 /* No neigh of this group is connected so we just set
1327 * the trap and let everthing flow through kernel.
1331 ret = mlxsw_sp_kvdl_alloc(mlxsw_sp, ecmp_size);
1333 /* We ran out of KVD linear space, just set the
1334 * trap and let everything flow through kernel.
1336 dev_warn(mlxsw_sp->bus_info->dev, "Failed to allocate KVD linear area for nexthop group.\n");
1340 old_adj_index_valid = nh_grp->adj_index_valid;
1341 old_adj_index = nh_grp->adj_index;
1342 old_ecmp_size = nh_grp->ecmp_size;
1343 nh_grp->adj_index_valid = 1;
1344 nh_grp->adj_index = adj_index;
1345 nh_grp->ecmp_size = ecmp_size;
1346 err = mlxsw_sp_nexthop_group_mac_update(mlxsw_sp, nh_grp, true);
1348 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
1352 if (!old_adj_index_valid) {
1353 /* The trap was set for fib entries, so we have to call
1354 * fib entry update to unset it and use adjacency index.
1356 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
1358 dev_warn(mlxsw_sp->bus_info->dev, "Failed to add adjacency index to fib entries.\n");
1364 err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, nh_grp,
1365 old_adj_index, old_ecmp_size);
1366 mlxsw_sp_kvdl_free(mlxsw_sp, old_adj_index);
1368 dev_warn(mlxsw_sp->bus_info->dev, "Failed to mass-update adjacency index for nexthop group.\n");
1374 old_adj_index_valid = nh_grp->adj_index_valid;
1375 nh_grp->adj_index_valid = 0;
1376 for (i = 0; i < nh_grp->count; i++) {
1377 nh = &nh_grp->nexthops[i];
1380 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
1382 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set traps for fib entries.\n");
1383 if (old_adj_index_valid)
1384 mlxsw_sp_kvdl_free(mlxsw_sp, nh_grp->adj_index);
1387 static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
1390 if (!removing && !nh->should_offload)
1391 nh->should_offload = 1;
1392 else if (removing && nh->offloaded)
1393 nh->should_offload = 0;
1398 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
1399 struct mlxsw_sp_neigh_entry *neigh_entry,
1402 struct mlxsw_sp_nexthop *nh;
1404 list_for_each_entry(nh, &neigh_entry->nexthop_list,
1406 __mlxsw_sp_nexthop_neigh_update(nh, removing);
1407 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
1411 static int mlxsw_sp_nexthop_neigh_init(struct mlxsw_sp *mlxsw_sp,
1412 struct mlxsw_sp_nexthop *nh)
1414 struct mlxsw_sp_neigh_entry *neigh_entry;
1415 struct fib_nh *fib_nh = nh->key.fib_nh;
1416 struct neighbour *n;
1420 if (!nh->nh_grp->gateway)
1423 /* Take a reference of neigh here ensuring that neigh would
1424 * not be detructed before the nexthop entry is finished.
1425 * The reference is taken either in neigh_lookup() or
1426 * in neigh_create() in case n is not found.
1428 n = neigh_lookup(&arp_tbl, &fib_nh->nh_gw, fib_nh->nh_dev);
1430 n = neigh_create(&arp_tbl, &fib_nh->nh_gw, fib_nh->nh_dev);
1433 neigh_event_send(n, NULL);
1435 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
1437 neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
1438 if (IS_ERR(neigh_entry)) {
1440 goto err_neigh_entry_create;
1444 /* If that is the first nexthop connected to that neigh, add to
1445 * nexthop_neighs_list
1447 if (list_empty(&neigh_entry->nexthop_list))
1448 list_add_tail(&neigh_entry->nexthop_neighs_list_node,
1449 &mlxsw_sp->router.nexthop_neighs_list);
1451 nh->neigh_entry = neigh_entry;
1452 list_add_tail(&nh->neigh_list_node, &neigh_entry->nexthop_list);
1453 read_lock_bh(&n->lock);
1454 nud_state = n->nud_state;
1456 read_unlock_bh(&n->lock);
1457 __mlxsw_sp_nexthop_neigh_update(nh, !(nud_state & NUD_VALID && !dead));
1461 err_neigh_entry_create:
1466 static void mlxsw_sp_nexthop_neigh_fini(struct mlxsw_sp *mlxsw_sp,
1467 struct mlxsw_sp_nexthop *nh)
1469 struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
1470 struct neighbour *n;
1474 n = neigh_entry->key.n;
1476 __mlxsw_sp_nexthop_neigh_update(nh, true);
1477 list_del(&nh->neigh_list_node);
1478 nh->neigh_entry = NULL;
1480 /* If that is the last nexthop connected to that neigh, remove from
1481 * nexthop_neighs_list
1483 if (list_empty(&neigh_entry->nexthop_list))
1484 list_del(&neigh_entry->nexthop_neighs_list_node);
1486 if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
1487 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
1492 static int mlxsw_sp_nexthop_init(struct mlxsw_sp *mlxsw_sp,
1493 struct mlxsw_sp_nexthop_group *nh_grp,
1494 struct mlxsw_sp_nexthop *nh,
1495 struct fib_nh *fib_nh)
1497 struct net_device *dev = fib_nh->nh_dev;
1498 struct mlxsw_sp_rif *r;
1501 nh->nh_grp = nh_grp;
1502 nh->key.fib_nh = fib_nh;
1503 err = mlxsw_sp_nexthop_insert(mlxsw_sp, nh);
1507 r = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
1512 err = mlxsw_sp_nexthop_neigh_init(mlxsw_sp, nh);
1514 goto err_nexthop_neigh_init;
1518 err_nexthop_neigh_init:
1519 mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
1523 static void mlxsw_sp_nexthop_fini(struct mlxsw_sp *mlxsw_sp,
1524 struct mlxsw_sp_nexthop *nh)
1526 mlxsw_sp_nexthop_neigh_fini(mlxsw_sp, nh);
1527 mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
1530 static struct mlxsw_sp_nexthop_group *
1531 mlxsw_sp_nexthop_group_create(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
1533 struct mlxsw_sp_nexthop_group *nh_grp;
1534 struct mlxsw_sp_nexthop *nh;
1535 struct fib_nh *fib_nh;
1540 alloc_size = sizeof(*nh_grp) +
1541 fi->fib_nhs * sizeof(struct mlxsw_sp_nexthop);
1542 nh_grp = kzalloc(alloc_size, GFP_KERNEL);
1544 return ERR_PTR(-ENOMEM);
1545 INIT_LIST_HEAD(&nh_grp->fib_list);
1546 nh_grp->gateway = fi->fib_nh->nh_scope == RT_SCOPE_LINK;
1547 nh_grp->count = fi->fib_nhs;
1548 nh_grp->key.fi = fi;
1549 for (i = 0; i < nh_grp->count; i++) {
1550 nh = &nh_grp->nexthops[i];
1551 fib_nh = &fi->fib_nh[i];
1552 err = mlxsw_sp_nexthop_init(mlxsw_sp, nh_grp, nh, fib_nh);
1554 goto err_nexthop_init;
1556 err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
1558 goto err_nexthop_group_insert;
1559 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
1562 err_nexthop_group_insert:
1564 for (i--; i >= 0; i--)
1565 mlxsw_sp_nexthop_fini(mlxsw_sp, nh);
1567 return ERR_PTR(err);
1571 mlxsw_sp_nexthop_group_destroy(struct mlxsw_sp *mlxsw_sp,
1572 struct mlxsw_sp_nexthop_group *nh_grp)
1574 struct mlxsw_sp_nexthop *nh;
1577 mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
1578 for (i = 0; i < nh_grp->count; i++) {
1579 nh = &nh_grp->nexthops[i];
1580 mlxsw_sp_nexthop_fini(mlxsw_sp, nh);
1582 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
1583 WARN_ON_ONCE(nh_grp->adj_index_valid);
1587 static int mlxsw_sp_nexthop_group_get(struct mlxsw_sp *mlxsw_sp,
1588 struct mlxsw_sp_fib_entry *fib_entry,
1589 struct fib_info *fi)
1591 struct mlxsw_sp_nexthop_group_key key;
1592 struct mlxsw_sp_nexthop_group *nh_grp;
1595 nh_grp = mlxsw_sp_nexthop_group_lookup(mlxsw_sp, key);
1597 nh_grp = mlxsw_sp_nexthop_group_create(mlxsw_sp, fi);
1599 return PTR_ERR(nh_grp);
1601 list_add_tail(&fib_entry->nexthop_group_node, &nh_grp->fib_list);
1602 fib_entry->nh_group = nh_grp;
1606 static void mlxsw_sp_nexthop_group_put(struct mlxsw_sp *mlxsw_sp,
1607 struct mlxsw_sp_fib_entry *fib_entry)
1609 struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
1611 list_del(&fib_entry->nexthop_group_node);
1612 if (!list_empty(&nh_grp->fib_list))
1614 mlxsw_sp_nexthop_group_destroy(mlxsw_sp, nh_grp);
1618 mlxsw_sp_fib_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
1620 struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
1622 switch (fib_entry->type) {
1623 case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
1624 return !!nh_group->adj_index_valid;
1625 case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
1632 static void mlxsw_sp_fib_entry_offload_set(struct mlxsw_sp_fib_entry *fib_entry)
1634 fib_entry->offloaded = true;
1636 switch (fib_entry->vr->proto) {
1637 case MLXSW_SP_L3_PROTO_IPV4:
1638 fib_info_offload_inc(fib_entry->nh_group->key.fi);
1640 case MLXSW_SP_L3_PROTO_IPV6:
1646 mlxsw_sp_fib_entry_offload_unset(struct mlxsw_sp_fib_entry *fib_entry)
1648 switch (fib_entry->vr->proto) {
1649 case MLXSW_SP_L3_PROTO_IPV4:
1650 fib_info_offload_dec(fib_entry->nh_group->key.fi);
1652 case MLXSW_SP_L3_PROTO_IPV6:
1656 fib_entry->offloaded = false;
1660 mlxsw_sp_fib_entry_offload_refresh(struct mlxsw_sp_fib_entry *fib_entry,
1661 enum mlxsw_reg_ralue_op op, int err)
1664 case MLXSW_REG_RALUE_OP_WRITE_DELETE:
1665 if (!fib_entry->offloaded)
1667 return mlxsw_sp_fib_entry_offload_unset(fib_entry);
1668 case MLXSW_REG_RALUE_OP_WRITE_WRITE:
1671 if (mlxsw_sp_fib_entry_should_offload(fib_entry) &&
1672 !fib_entry->offloaded)
1673 mlxsw_sp_fib_entry_offload_set(fib_entry);
1674 else if (!mlxsw_sp_fib_entry_should_offload(fib_entry) &&
1675 fib_entry->offloaded)
1676 mlxsw_sp_fib_entry_offload_unset(fib_entry);
1683 static int mlxsw_sp_fib_entry_op4_remote(struct mlxsw_sp *mlxsw_sp,
1684 struct mlxsw_sp_fib_entry *fib_entry,
1685 enum mlxsw_reg_ralue_op op)
1687 char ralue_pl[MLXSW_REG_RALUE_LEN];
1688 u32 *p_dip = (u32 *) fib_entry->key.addr;
1689 struct mlxsw_sp_vr *vr = fib_entry->vr;
1690 enum mlxsw_reg_ralue_trap_action trap_action;
1692 u32 adjacency_index = 0;
1695 /* In case the nexthop group adjacency index is valid, use it
1696 * with provided ECMP size. Otherwise, setup trap and pass
1697 * traffic to kernel.
1699 if (fib_entry->nh_group->adj_index_valid) {
1700 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
1701 adjacency_index = fib_entry->nh_group->adj_index;
1702 ecmp_size = fib_entry->nh_group->ecmp_size;
1704 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
1705 trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
1708 mlxsw_reg_ralue_pack4(ralue_pl,
1709 (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1710 vr->id, fib_entry->key.prefix_len, *p_dip);
1711 mlxsw_reg_ralue_act_remote_pack(ralue_pl, trap_action, trap_id,
1712 adjacency_index, ecmp_size);
1713 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1716 static int mlxsw_sp_fib_entry_op4_local(struct mlxsw_sp *mlxsw_sp,
1717 struct mlxsw_sp_fib_entry *fib_entry,
1718 enum mlxsw_reg_ralue_op op)
1720 struct mlxsw_sp_rif *r = fib_entry->nh_group->nh_rif;
1721 char ralue_pl[MLXSW_REG_RALUE_LEN];
1722 u32 *p_dip = (u32 *) fib_entry->key.addr;
1723 struct mlxsw_sp_vr *vr = fib_entry->vr;
1725 mlxsw_reg_ralue_pack4(ralue_pl,
1726 (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1727 vr->id, fib_entry->key.prefix_len, *p_dip);
1728 mlxsw_reg_ralue_act_local_pack(ralue_pl,
1729 MLXSW_REG_RALUE_TRAP_ACTION_NOP, 0,
1731 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1734 static int mlxsw_sp_fib_entry_op4_trap(struct mlxsw_sp *mlxsw_sp,
1735 struct mlxsw_sp_fib_entry *fib_entry,
1736 enum mlxsw_reg_ralue_op op)
1738 char ralue_pl[MLXSW_REG_RALUE_LEN];
1739 u32 *p_dip = (u32 *) fib_entry->key.addr;
1740 struct mlxsw_sp_vr *vr = fib_entry->vr;
1742 mlxsw_reg_ralue_pack4(ralue_pl,
1743 (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1744 vr->id, fib_entry->key.prefix_len, *p_dip);
1745 mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
1746 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1749 static int mlxsw_sp_fib_entry_op4(struct mlxsw_sp *mlxsw_sp,
1750 struct mlxsw_sp_fib_entry *fib_entry,
1751 enum mlxsw_reg_ralue_op op)
1753 switch (fib_entry->type) {
1754 case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
1755 return mlxsw_sp_fib_entry_op4_remote(mlxsw_sp, fib_entry, op);
1756 case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
1757 return mlxsw_sp_fib_entry_op4_local(mlxsw_sp, fib_entry, op);
1758 case MLXSW_SP_FIB_ENTRY_TYPE_TRAP:
1759 return mlxsw_sp_fib_entry_op4_trap(mlxsw_sp, fib_entry, op);
1764 static int mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
1765 struct mlxsw_sp_fib_entry *fib_entry,
1766 enum mlxsw_reg_ralue_op op)
1770 switch (fib_entry->vr->proto) {
1771 case MLXSW_SP_L3_PROTO_IPV4:
1772 err = mlxsw_sp_fib_entry_op4(mlxsw_sp, fib_entry, op);
1774 case MLXSW_SP_L3_PROTO_IPV6:
1777 mlxsw_sp_fib_entry_offload_refresh(fib_entry, op, err);
1781 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1782 struct mlxsw_sp_fib_entry *fib_entry)
1784 return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
1785 MLXSW_REG_RALUE_OP_WRITE_WRITE);
1788 static int mlxsw_sp_fib_entry_del(struct mlxsw_sp *mlxsw_sp,
1789 struct mlxsw_sp_fib_entry *fib_entry)
1791 return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
1792 MLXSW_REG_RALUE_OP_WRITE_DELETE);
1796 mlxsw_sp_fib4_entry_type_set(struct mlxsw_sp *mlxsw_sp,
1797 const struct fib_entry_notifier_info *fen_info,
1798 struct mlxsw_sp_fib_entry *fib_entry)
1800 struct fib_info *fi = fen_info->fi;
1801 struct mlxsw_sp_rif *r = NULL;
1804 if (fen_info->type == RTN_LOCAL || fen_info->type == RTN_BROADCAST) {
1805 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1808 if (fen_info->type != RTN_UNICAST)
1811 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1812 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1816 r = mlxsw_sp_rif_find_by_dev(mlxsw_sp, nh->nh_dev);
1818 /* In case router interface is not found for
1819 * at least one of the nexthops, that means
1820 * the nexthop points to some device unrelated
1821 * to us. Set trap and pass the packets for
1822 * this prefix to kernel.
1829 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1833 if (fi->fib_nh->nh_scope != RT_SCOPE_LINK)
1834 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
1836 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
1840 static struct mlxsw_sp_fib_entry *
1841 mlxsw_sp_fib_entry_get(struct mlxsw_sp *mlxsw_sp,
1842 const struct fib_entry_notifier_info *fen_info)
1844 struct mlxsw_sp_fib_entry *fib_entry;
1845 struct fib_info *fi = fen_info->fi;
1846 struct mlxsw_sp_vr *vr;
1849 vr = mlxsw_sp_vr_get(mlxsw_sp, fen_info->dst_len, fen_info->tb_id,
1850 MLXSW_SP_L3_PROTO_IPV4);
1852 return ERR_CAST(vr);
1854 fib_entry = mlxsw_sp_fib_entry_lookup(vr->fib, &fen_info->dst,
1855 sizeof(fen_info->dst),
1856 fen_info->dst_len, fi->fib_dev);
1858 /* Already exists, just take a reference */
1859 fib_entry->ref_count++;
1862 fib_entry = mlxsw_sp_fib_entry_create(vr->fib, &fen_info->dst,
1863 sizeof(fen_info->dst),
1864 fen_info->dst_len, fi->fib_dev);
1867 goto err_fib_entry_create;
1870 fib_entry->ref_count = 1;
1872 err = mlxsw_sp_fib4_entry_type_set(mlxsw_sp, fen_info, fib_entry);
1874 goto err_fib4_entry_type_set;
1876 err = mlxsw_sp_nexthop_group_get(mlxsw_sp, fib_entry, fi);
1878 goto err_nexthop_group_get;
1882 err_nexthop_group_get:
1883 err_fib4_entry_type_set:
1884 mlxsw_sp_fib_entry_destroy(fib_entry);
1885 err_fib_entry_create:
1886 mlxsw_sp_vr_put(mlxsw_sp, vr);
1888 return ERR_PTR(err);
1891 static struct mlxsw_sp_fib_entry *
1892 mlxsw_sp_fib_entry_find(struct mlxsw_sp *mlxsw_sp,
1893 const struct fib_entry_notifier_info *fen_info)
1895 struct mlxsw_sp_vr *vr;
1897 vr = mlxsw_sp_vr_find(mlxsw_sp, fen_info->tb_id,
1898 MLXSW_SP_L3_PROTO_IPV4);
1902 return mlxsw_sp_fib_entry_lookup(vr->fib, &fen_info->dst,
1903 sizeof(fen_info->dst),
1905 fen_info->fi->fib_dev);
1908 static void mlxsw_sp_fib_entry_put(struct mlxsw_sp *mlxsw_sp,
1909 struct mlxsw_sp_fib_entry *fib_entry)
1911 struct mlxsw_sp_vr *vr = fib_entry->vr;
1913 if (--fib_entry->ref_count == 0) {
1914 mlxsw_sp_nexthop_group_put(mlxsw_sp, fib_entry);
1915 mlxsw_sp_fib_entry_destroy(fib_entry);
1917 mlxsw_sp_vr_put(mlxsw_sp, vr);
1920 static void mlxsw_sp_fib_entry_put_all(struct mlxsw_sp *mlxsw_sp,
1921 struct mlxsw_sp_fib_entry *fib_entry)
1923 unsigned int last_ref_count;
1926 last_ref_count = fib_entry->ref_count;
1927 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1928 } while (last_ref_count != 1);
1931 static int mlxsw_sp_router_fib4_add(struct mlxsw_sp *mlxsw_sp,
1932 struct fib_entry_notifier_info *fen_info)
1934 struct mlxsw_sp_fib_entry *fib_entry;
1935 struct mlxsw_sp_vr *vr;
1938 if (mlxsw_sp->router.aborted)
1941 fib_entry = mlxsw_sp_fib_entry_get(mlxsw_sp, fen_info);
1942 if (IS_ERR(fib_entry)) {
1943 dev_warn(mlxsw_sp->bus_info->dev, "Failed to get FIB4 entry being added.\n");
1944 return PTR_ERR(fib_entry);
1947 if (fib_entry->ref_count != 1)
1951 err = mlxsw_sp_fib_entry_insert(vr->fib, fib_entry);
1953 dev_warn(mlxsw_sp->bus_info->dev, "Failed to insert FIB4 entry being added.\n");
1954 goto err_fib_entry_insert;
1956 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1958 goto err_fib_entry_add;
1962 mlxsw_sp_fib_entry_remove(vr->fib, fib_entry);
1963 err_fib_entry_insert:
1964 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1968 static void mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
1969 struct fib_entry_notifier_info *fen_info)
1971 struct mlxsw_sp_fib_entry *fib_entry;
1973 if (mlxsw_sp->router.aborted)
1976 fib_entry = mlxsw_sp_fib_entry_find(mlxsw_sp, fen_info);
1980 if (fib_entry->ref_count == 1) {
1981 mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
1982 mlxsw_sp_fib_entry_remove(fib_entry->vr->fib, fib_entry);
1985 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1988 static int mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp)
1990 char ralta_pl[MLXSW_REG_RALTA_LEN];
1991 char ralst_pl[MLXSW_REG_RALST_LEN];
1992 char raltb_pl[MLXSW_REG_RALTB_LEN];
1993 char ralue_pl[MLXSW_REG_RALUE_LEN];
1996 mlxsw_reg_ralta_pack(ralta_pl, true, MLXSW_REG_RALXX_PROTOCOL_IPV4,
1997 MLXSW_SP_LPM_TREE_MIN);
1998 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
2002 mlxsw_reg_ralst_pack(ralst_pl, 0xff, MLXSW_SP_LPM_TREE_MIN);
2003 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
2007 mlxsw_reg_raltb_pack(raltb_pl, 0, MLXSW_REG_RALXX_PROTOCOL_IPV4,
2008 MLXSW_SP_LPM_TREE_MIN);
2009 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
2013 mlxsw_reg_ralue_pack4(ralue_pl, MLXSW_SP_L3_PROTO_IPV4,
2014 MLXSW_REG_RALUE_OP_WRITE_WRITE, 0, 0, 0);
2015 mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
2016 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
2019 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp)
2021 struct mlxsw_sp_fib_entry *fib_entry;
2022 struct mlxsw_sp_fib_entry *tmp;
2023 struct mlxsw_sp_vr *vr;
2026 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
2027 vr = &mlxsw_sp->router.vrs[i];
2032 list_for_each_entry_safe(fib_entry, tmp,
2033 &vr->fib->entry_list, list) {
2034 bool do_break = &tmp->list == &vr->fib->entry_list;
2036 mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
2037 mlxsw_sp_fib_entry_remove(fib_entry->vr->fib,
2039 mlxsw_sp_fib_entry_put_all(mlxsw_sp, fib_entry);
2046 static void mlxsw_sp_router_fib4_abort(struct mlxsw_sp *mlxsw_sp)
2050 if (mlxsw_sp->router.aborted)
2052 dev_warn(mlxsw_sp->bus_info->dev, "FIB abort triggered. Note that FIB entries are no longer being offloaded to this device.\n");
2053 mlxsw_sp_router_fib_flush(mlxsw_sp);
2054 mlxsw_sp->router.aborted = true;
2055 err = mlxsw_sp_router_set_abort_trap(mlxsw_sp);
2057 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set abort trap.\n");
2060 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
2062 char rgcr_pl[MLXSW_REG_RGCR_LEN];
2066 if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_RIFS))
2069 max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
2070 mlxsw_sp->rifs = kcalloc(max_rifs, sizeof(struct mlxsw_sp_rif *),
2072 if (!mlxsw_sp->rifs)
2075 mlxsw_reg_rgcr_pack(rgcr_pl, true);
2076 mlxsw_reg_rgcr_max_router_interfaces_set(rgcr_pl, max_rifs);
2077 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
2084 kfree(mlxsw_sp->rifs);
2088 static void __mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
2090 char rgcr_pl[MLXSW_REG_RGCR_LEN];
2093 mlxsw_reg_rgcr_pack(rgcr_pl, false);
2094 mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
2096 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
2097 WARN_ON_ONCE(mlxsw_sp->rifs[i]);
2099 kfree(mlxsw_sp->rifs);
2102 struct mlxsw_sp_fib_event_work {
2103 struct work_struct work;
2104 struct fib_entry_notifier_info fen_info;
2105 struct mlxsw_sp *mlxsw_sp;
2106 unsigned long event;
2109 static void mlxsw_sp_router_fib_event_work(struct work_struct *work)
2111 struct mlxsw_sp_fib_event_work *fib_work =
2112 container_of(work, struct mlxsw_sp_fib_event_work, work);
2113 struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
2116 /* Protect internal structures from changes */
2118 switch (fib_work->event) {
2119 case FIB_EVENT_ENTRY_ADD:
2120 err = mlxsw_sp_router_fib4_add(mlxsw_sp, &fib_work->fen_info);
2122 mlxsw_sp_router_fib4_abort(mlxsw_sp);
2123 fib_info_put(fib_work->fen_info.fi);
2125 case FIB_EVENT_ENTRY_DEL:
2126 mlxsw_sp_router_fib4_del(mlxsw_sp, &fib_work->fen_info);
2127 fib_info_put(fib_work->fen_info.fi);
2129 case FIB_EVENT_RULE_ADD: /* fall through */
2130 case FIB_EVENT_RULE_DEL:
2131 mlxsw_sp_router_fib4_abort(mlxsw_sp);
2138 /* Called with rcu_read_lock() */
2139 static int mlxsw_sp_router_fib_event(struct notifier_block *nb,
2140 unsigned long event, void *ptr)
2142 struct mlxsw_sp *mlxsw_sp = container_of(nb, struct mlxsw_sp, fib_nb);
2143 struct mlxsw_sp_fib_event_work *fib_work;
2144 struct fib_notifier_info *info = ptr;
2146 if (!net_eq(info->net, &init_net))
2149 fib_work = kzalloc(sizeof(*fib_work), GFP_ATOMIC);
2150 if (WARN_ON(!fib_work))
2153 INIT_WORK(&fib_work->work, mlxsw_sp_router_fib_event_work);
2154 fib_work->mlxsw_sp = mlxsw_sp;
2155 fib_work->event = event;
2158 case FIB_EVENT_ENTRY_ADD: /* fall through */
2159 case FIB_EVENT_ENTRY_DEL:
2160 memcpy(&fib_work->fen_info, ptr, sizeof(fib_work->fen_info));
2161 /* Take referece on fib_info to prevent it from being
2162 * freed while work is queued. Release it afterwards.
2164 fib_info_hold(fib_work->fen_info.fi);
2168 mlxsw_core_schedule_work(&fib_work->work);
2173 static void mlxsw_sp_router_fib_dump_flush(struct notifier_block *nb)
2175 struct mlxsw_sp *mlxsw_sp = container_of(nb, struct mlxsw_sp, fib_nb);
2177 /* Flush pending FIB notifications and then flush the device's
2178 * table before requesting another dump. The FIB notification
2179 * block is unregistered, so no need to take RTNL.
2181 mlxsw_core_flush_owq();
2182 mlxsw_sp_router_fib_flush(mlxsw_sp);
2185 int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
2189 INIT_LIST_HEAD(&mlxsw_sp->router.nexthop_neighs_list);
2190 err = __mlxsw_sp_router_init(mlxsw_sp);
2194 err = rhashtable_init(&mlxsw_sp->router.nexthop_ht,
2195 &mlxsw_sp_nexthop_ht_params);
2197 goto err_nexthop_ht_init;
2199 err = rhashtable_init(&mlxsw_sp->router.nexthop_group_ht,
2200 &mlxsw_sp_nexthop_group_ht_params);
2202 goto err_nexthop_group_ht_init;
2204 mlxsw_sp_lpm_init(mlxsw_sp);
2205 err = mlxsw_sp_vrs_init(mlxsw_sp);
2209 err = mlxsw_sp_neigh_init(mlxsw_sp);
2211 goto err_neigh_init;
2213 mlxsw_sp->fib_nb.notifier_call = mlxsw_sp_router_fib_event;
2214 err = register_fib_notifier(&mlxsw_sp->fib_nb,
2215 mlxsw_sp_router_fib_dump_flush);
2217 goto err_register_fib_notifier;
2221 err_register_fib_notifier:
2222 mlxsw_sp_neigh_fini(mlxsw_sp);
2224 mlxsw_sp_vrs_fini(mlxsw_sp);
2226 rhashtable_destroy(&mlxsw_sp->router.nexthop_group_ht);
2227 err_nexthop_group_ht_init:
2228 rhashtable_destroy(&mlxsw_sp->router.nexthop_ht);
2229 err_nexthop_ht_init:
2230 __mlxsw_sp_router_fini(mlxsw_sp);
2234 void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
2236 unregister_fib_notifier(&mlxsw_sp->fib_nb);
2237 mlxsw_sp_neigh_fini(mlxsw_sp);
2238 mlxsw_sp_vrs_fini(mlxsw_sp);
2239 rhashtable_destroy(&mlxsw_sp->router.nexthop_group_ht);
2240 rhashtable_destroy(&mlxsw_sp->router.nexthop_ht);
2241 __mlxsw_sp_router_fini(mlxsw_sp);