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 u16 rif; /* used for action local */
131 struct mlxsw_sp_vr *vr;
133 struct list_head nexthop_group_node;
134 struct mlxsw_sp_nexthop_group *nh_group;
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;
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,
151 static int mlxsw_sp_fib_entry_insert(struct mlxsw_sp_fib *fib,
152 struct mlxsw_sp_fib_entry *fib_entry)
154 unsigned char prefix_len = fib_entry->key.prefix_len;
157 err = rhashtable_insert_fast(&fib->ht, &fib_entry->ht_node,
158 mlxsw_sp_fib_ht_params);
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);
167 static void mlxsw_sp_fib_entry_remove(struct mlxsw_sp_fib *fib,
168 struct mlxsw_sp_fib_entry *fib_entry)
170 unsigned char prefix_len = fib_entry->key.prefix_len;
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);
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)
184 struct mlxsw_sp_fib_entry *fib_entry;
186 fib_entry = kzalloc(sizeof(*fib_entry), GFP_KERNEL);
189 fib_entry->key.dev = dev;
190 memcpy(fib_entry->key.addr, addr, addr_len);
191 fib_entry->key.prefix_len = prefix_len;
195 static void mlxsw_sp_fib_entry_destroy(struct mlxsw_sp_fib_entry *fib_entry)
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)
205 struct mlxsw_sp_fib_key key;
207 memset(&key, 0, sizeof(key));
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);
214 static struct mlxsw_sp_fib *mlxsw_sp_fib_create(void)
216 struct mlxsw_sp_fib *fib;
219 fib = kzalloc(sizeof(*fib), GFP_KERNEL);
221 return ERR_PTR(-ENOMEM);
222 err = rhashtable_init(&fib->ht, &mlxsw_sp_fib_ht_params);
224 goto err_rhashtable_init;
225 INIT_LIST_HEAD(&fib->entry_list);
233 static void mlxsw_sp_fib_destroy(struct mlxsw_sp_fib *fib)
235 rhashtable_destroy(&fib->ht);
239 static struct mlxsw_sp_lpm_tree *
240 mlxsw_sp_lpm_tree_find_unused(struct mlxsw_sp *mlxsw_sp, bool one_reserved)
242 static struct mlxsw_sp_lpm_tree *lpm_tree;
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) {
249 one_reserved = false;
257 static int mlxsw_sp_lpm_tree_alloc(struct mlxsw_sp *mlxsw_sp,
258 struct mlxsw_sp_lpm_tree *lpm_tree)
260 char ralta_pl[MLXSW_REG_RALTA_LEN];
262 mlxsw_reg_ralta_pack(ralta_pl, true,
263 (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
265 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
268 static int mlxsw_sp_lpm_tree_free(struct mlxsw_sp *mlxsw_sp,
269 struct mlxsw_sp_lpm_tree *lpm_tree)
271 char ralta_pl[MLXSW_REG_RALTA_LEN];
273 mlxsw_reg_ralta_pack(ralta_pl, false,
274 (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
276 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
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)
284 char ralst_pl[MLXSW_REG_RALST_LEN];
287 u8 last_prefix = MLXSW_REG_RALST_BIN_NO_CHILD;
289 mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage)
292 mlxsw_reg_ralst_pack(ralst_pl, root_bin, lpm_tree->id);
293 mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) {
296 mlxsw_reg_ralst_bin_pack(ralst_pl, prefix, last_prefix,
297 MLXSW_REG_RALST_BIN_NO_CHILD);
298 last_prefix = prefix;
300 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
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)
308 struct mlxsw_sp_lpm_tree *lpm_tree;
311 lpm_tree = mlxsw_sp_lpm_tree_find_unused(mlxsw_sp, one_reserved);
313 return ERR_PTR(-EBUSY);
314 lpm_tree->proto = proto;
315 err = mlxsw_sp_lpm_tree_alloc(mlxsw_sp, lpm_tree);
319 err = mlxsw_sp_lpm_tree_left_struct_set(mlxsw_sp, prefix_usage,
322 goto err_left_struct_set;
323 memcpy(&lpm_tree->prefix_usage, prefix_usage,
324 sizeof(lpm_tree->prefix_usage));
328 mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
332 static int mlxsw_sp_lpm_tree_destroy(struct mlxsw_sp *mlxsw_sp,
333 struct mlxsw_sp_lpm_tree *lpm_tree)
335 return mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
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)
343 struct mlxsw_sp_lpm_tree *lpm_tree;
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,
354 lpm_tree = mlxsw_sp_lpm_tree_create(mlxsw_sp, prefix_usage,
355 proto, one_reserved);
356 if (IS_ERR(lpm_tree))
360 lpm_tree->ref_count++;
364 static int mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
365 struct mlxsw_sp_lpm_tree *lpm_tree)
367 if (--lpm_tree->ref_count == 0)
368 return mlxsw_sp_lpm_tree_destroy(mlxsw_sp, lpm_tree);
372 static void mlxsw_sp_lpm_init(struct mlxsw_sp *mlxsw_sp)
374 struct mlxsw_sp_lpm_tree *lpm_tree;
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;
383 static struct mlxsw_sp_vr *mlxsw_sp_vr_find_unused(struct mlxsw_sp *mlxsw_sp)
385 struct mlxsw_sp_vr *vr;
388 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
389 vr = &mlxsw_sp->router.vrs[i];
396 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
397 struct mlxsw_sp_vr *vr)
399 char raltb_pl[MLXSW_REG_RALTB_LEN];
401 mlxsw_reg_raltb_pack(raltb_pl, vr->id,
402 (enum mlxsw_reg_ralxx_protocol) vr->proto,
404 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
407 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
408 struct mlxsw_sp_vr *vr)
410 char raltb_pl[MLXSW_REG_RALTB_LEN];
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);
418 static u32 mlxsw_sp_fix_tb_id(u32 tb_id)
420 /* For our purpose, squash main and local table into one */
421 if (tb_id == RT_TABLE_LOCAL)
422 tb_id = RT_TABLE_MAIN;
426 static struct mlxsw_sp_vr *mlxsw_sp_vr_find(struct mlxsw_sp *mlxsw_sp,
428 enum mlxsw_sp_l3proto proto)
430 struct mlxsw_sp_vr *vr;
433 tb_id = mlxsw_sp_fix_tb_id(tb_id);
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)
443 static struct mlxsw_sp_vr *mlxsw_sp_vr_create(struct mlxsw_sp *mlxsw_sp,
444 unsigned char prefix_len,
446 enum mlxsw_sp_l3proto proto)
448 struct mlxsw_sp_prefix_usage req_prefix_usage;
449 struct mlxsw_sp_lpm_tree *lpm_tree;
450 struct mlxsw_sp_vr *vr;
453 vr = mlxsw_sp_vr_find_unused(mlxsw_sp);
455 return ERR_PTR(-EBUSY);
456 vr->fib = mlxsw_sp_fib_create();
458 return ERR_CAST(vr->fib);
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,
466 if (IS_ERR(lpm_tree)) {
467 err = PTR_ERR(lpm_tree);
470 vr->lpm_tree = lpm_tree;
471 err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, vr);
479 mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
481 mlxsw_sp_fib_destroy(vr->fib);
486 static void mlxsw_sp_vr_destroy(struct mlxsw_sp *mlxsw_sp,
487 struct mlxsw_sp_vr *vr)
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);
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)
499 struct mlxsw_sp_lpm_tree *lpm_tree;
501 if (mlxsw_sp_prefix_usage_eq(req_prefix_usage,
502 &vr->lpm_tree->prefix_usage))
505 lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, req_prefix_usage,
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
513 if (mlxsw_sp_prefix_usage_subset(req_prefix_usage,
514 &vr->lpm_tree->prefix_usage))
516 return PTR_ERR(lpm_tree);
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);
525 static struct mlxsw_sp_vr *mlxsw_sp_vr_get(struct mlxsw_sp *mlxsw_sp,
526 unsigned char prefix_len,
528 enum mlxsw_sp_l3proto proto)
530 struct mlxsw_sp_vr *vr;
533 tb_id = mlxsw_sp_fix_tb_id(tb_id);
534 vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id, proto);
536 vr = mlxsw_sp_vr_create(mlxsw_sp, prefix_len, tb_id, proto);
540 struct mlxsw_sp_prefix_usage req_prefix_usage;
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,
554 static void mlxsw_sp_vr_put(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr)
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.
562 if (mlxsw_sp_prefix_usage_none(&vr->fib->prefix_usage))
563 mlxsw_sp_vr_destroy(mlxsw_sp, vr);
565 mlxsw_sp_vr_lpm_tree_check(mlxsw_sp, vr,
566 &vr->fib->prefix_usage);
569 static int mlxsw_sp_vrs_init(struct mlxsw_sp *mlxsw_sp)
571 struct mlxsw_sp_vr *vr;
575 if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_VRS))
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),
581 if (!mlxsw_sp->router.vrs)
584 for (i = 0; i < max_vrs; i++) {
585 vr = &mlxsw_sp->router.vrs[i];
592 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp);
594 static void mlxsw_sp_vrs_fini(struct mlxsw_sp *mlxsw_sp)
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
603 mlxsw_core_flush_owq();
604 mlxsw_sp_router_fib_flush(mlxsw_sp);
605 kfree(mlxsw_sp->router.vrs);
608 struct mlxsw_sp_neigh_key {
612 struct mlxsw_sp_neigh_entry {
613 struct rhash_head ht_node;
614 struct mlxsw_sp_neigh_key key;
617 struct delayed_work dw;
618 struct mlxsw_sp_port *mlxsw_sp_port;
619 unsigned char ha[ETH_ALEN];
620 struct list_head nexthop_list; /* list of nexthops using
623 struct list_head nexthop_neighs_list_node;
626 static const struct rhashtable_params mlxsw_sp_neigh_ht_params = {
627 .key_offset = offsetof(struct mlxsw_sp_neigh_entry, key),
628 .head_offset = offsetof(struct mlxsw_sp_neigh_entry, ht_node),
629 .key_len = sizeof(struct mlxsw_sp_neigh_key),
633 mlxsw_sp_neigh_entry_insert(struct mlxsw_sp *mlxsw_sp,
634 struct mlxsw_sp_neigh_entry *neigh_entry)
636 return rhashtable_insert_fast(&mlxsw_sp->router.neigh_ht,
637 &neigh_entry->ht_node,
638 mlxsw_sp_neigh_ht_params);
642 mlxsw_sp_neigh_entry_remove(struct mlxsw_sp *mlxsw_sp,
643 struct mlxsw_sp_neigh_entry *neigh_entry)
645 rhashtable_remove_fast(&mlxsw_sp->router.neigh_ht,
646 &neigh_entry->ht_node,
647 mlxsw_sp_neigh_ht_params);
650 static void mlxsw_sp_router_neigh_update_hw(struct work_struct *work);
652 static struct mlxsw_sp_neigh_entry *
653 mlxsw_sp_neigh_entry_create(struct neighbour *n, u16 rif)
655 struct mlxsw_sp_neigh_entry *neigh_entry;
657 neigh_entry = kzalloc(sizeof(*neigh_entry), GFP_ATOMIC);
660 neigh_entry->key.n = n;
661 neigh_entry->rif = rif;
662 INIT_DELAYED_WORK(&neigh_entry->dw, mlxsw_sp_router_neigh_update_hw);
663 INIT_LIST_HEAD(&neigh_entry->nexthop_list);
668 mlxsw_sp_neigh_entry_destroy(struct mlxsw_sp_neigh_entry *neigh_entry)
673 static struct mlxsw_sp_neigh_entry *
674 mlxsw_sp_neigh_entry_lookup(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
676 struct mlxsw_sp_neigh_key key;
679 return rhashtable_lookup_fast(&mlxsw_sp->router.neigh_ht,
680 &key, mlxsw_sp_neigh_ht_params);
683 int mlxsw_sp_router_neigh_construct(struct net_device *dev,
686 struct mlxsw_sp_port *mlxsw_sp_port = netdev_priv(dev);
687 struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
688 struct mlxsw_sp_neigh_entry *neigh_entry;
689 struct mlxsw_sp_rif *r;
692 if (n->tbl != &arp_tbl)
695 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
699 r = mlxsw_sp_rif_find_by_dev(mlxsw_sp, n->dev);
703 neigh_entry = mlxsw_sp_neigh_entry_create(n, r->rif);
706 err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
708 goto err_neigh_entry_insert;
711 err_neigh_entry_insert:
712 mlxsw_sp_neigh_entry_destroy(neigh_entry);
716 void mlxsw_sp_router_neigh_destroy(struct net_device *dev,
719 struct mlxsw_sp_port *mlxsw_sp_port = netdev_priv(dev);
720 struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
721 struct mlxsw_sp_neigh_entry *neigh_entry;
723 if (n->tbl != &arp_tbl)
726 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
729 mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
730 mlxsw_sp_neigh_entry_destroy(neigh_entry);
734 mlxsw_sp_router_neighs_update_interval_init(struct mlxsw_sp *mlxsw_sp)
736 unsigned long interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME);
738 mlxsw_sp->router.neighs_update.interval = jiffies_to_msecs(interval);
741 static void mlxsw_sp_router_neigh_ent_ipv4_process(struct mlxsw_sp *mlxsw_sp,
745 struct net_device *dev;
751 mlxsw_reg_rauhtd_ent_ipv4_unpack(rauhtd_pl, ent_index, &rif, &dip);
753 if (!mlxsw_sp->rifs[rif]) {
754 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
759 dev = mlxsw_sp->rifs[rif]->dev;
760 n = neigh_lookup(&arp_tbl, &dipn, dev);
762 netdev_err(dev, "Failed to find matching neighbour for IP=%pI4h\n",
767 netdev_dbg(dev, "Updating neighbour with IP=%pI4h\n", &dip);
768 neigh_event_send(n, NULL);
772 static void mlxsw_sp_router_neigh_rec_ipv4_process(struct mlxsw_sp *mlxsw_sp,
779 num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
781 /* Hardware starts counting at 0, so add 1. */
784 /* Each record consists of several neighbour entries. */
785 for (i = 0; i < num_entries; i++) {
788 ent_index = rec_index * MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC + i;
789 mlxsw_sp_router_neigh_ent_ipv4_process(mlxsw_sp, rauhtd_pl,
795 static void mlxsw_sp_router_neigh_rec_process(struct mlxsw_sp *mlxsw_sp,
796 char *rauhtd_pl, int rec_index)
798 switch (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, rec_index)) {
799 case MLXSW_REG_RAUHTD_TYPE_IPV4:
800 mlxsw_sp_router_neigh_rec_ipv4_process(mlxsw_sp, rauhtd_pl,
803 case MLXSW_REG_RAUHTD_TYPE_IPV6:
809 static bool mlxsw_sp_router_rauhtd_is_full(char *rauhtd_pl)
811 u8 num_rec, last_rec_index, num_entries;
813 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
814 last_rec_index = num_rec - 1;
816 if (num_rec < MLXSW_REG_RAUHTD_REC_MAX_NUM)
818 if (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, last_rec_index) ==
819 MLXSW_REG_RAUHTD_TYPE_IPV6)
822 num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
824 if (++num_entries == MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC)
829 static int mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp)
835 rauhtd_pl = kmalloc(MLXSW_REG_RAUHTD_LEN, GFP_KERNEL);
839 /* Make sure the neighbour's netdev isn't removed in the
844 mlxsw_reg_rauhtd_pack(rauhtd_pl, MLXSW_REG_RAUHTD_TYPE_IPV4);
845 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(rauhtd),
848 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Failed to dump neighbour talbe\n");
851 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
852 for (i = 0; i < num_rec; i++)
853 mlxsw_sp_router_neigh_rec_process(mlxsw_sp, rauhtd_pl,
855 } while (mlxsw_sp_router_rauhtd_is_full(rauhtd_pl));
862 static void mlxsw_sp_router_neighs_update_nh(struct mlxsw_sp *mlxsw_sp)
864 struct mlxsw_sp_neigh_entry *neigh_entry;
866 /* Take RTNL mutex here to prevent lists from changes */
868 list_for_each_entry(neigh_entry, &mlxsw_sp->router.nexthop_neighs_list,
869 nexthop_neighs_list_node) {
870 /* If this neigh have nexthops, make the kernel think this neigh
871 * is active regardless of the traffic.
873 if (!list_empty(&neigh_entry->nexthop_list))
874 neigh_event_send(neigh_entry->key.n, NULL);
880 mlxsw_sp_router_neighs_update_work_schedule(struct mlxsw_sp *mlxsw_sp)
882 unsigned long interval = mlxsw_sp->router.neighs_update.interval;
884 mlxsw_core_schedule_dw(&mlxsw_sp->router.neighs_update.dw,
885 msecs_to_jiffies(interval));
888 static void mlxsw_sp_router_neighs_update_work(struct work_struct *work)
890 struct mlxsw_sp *mlxsw_sp = container_of(work, struct mlxsw_sp,
891 router.neighs_update.dw.work);
894 err = mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp);
896 dev_err(mlxsw_sp->bus_info->dev, "Could not update kernel for neigh activity");
898 mlxsw_sp_router_neighs_update_nh(mlxsw_sp);
900 mlxsw_sp_router_neighs_update_work_schedule(mlxsw_sp);
903 static void mlxsw_sp_router_probe_unresolved_nexthops(struct work_struct *work)
905 struct mlxsw_sp_neigh_entry *neigh_entry;
906 struct mlxsw_sp *mlxsw_sp = container_of(work, struct mlxsw_sp,
907 router.nexthop_probe_dw.work);
909 /* Iterate over nexthop neighbours, find those who are unresolved and
910 * send arp on them. This solves the chicken-egg problem when
911 * the nexthop wouldn't get offloaded until the neighbor is resolved
912 * but it wouldn't get resolved ever in case traffic is flowing in HW
913 * using different nexthop.
915 * Take RTNL mutex here to prevent lists from changes.
918 list_for_each_entry(neigh_entry, &mlxsw_sp->router.nexthop_neighs_list,
919 nexthop_neighs_list_node) {
920 if (!(neigh_entry->key.n->nud_state & NUD_VALID) &&
921 !list_empty(&neigh_entry->nexthop_list))
922 neigh_event_send(neigh_entry->key.n, NULL);
926 mlxsw_core_schedule_dw(&mlxsw_sp->router.nexthop_probe_dw,
927 MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL);
931 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
932 struct mlxsw_sp_neigh_entry *neigh_entry,
935 static void mlxsw_sp_router_neigh_update_hw(struct work_struct *work)
937 struct mlxsw_sp_neigh_entry *neigh_entry =
938 container_of(work, struct mlxsw_sp_neigh_entry, dw.work);
939 struct neighbour *n = neigh_entry->key.n;
940 struct mlxsw_sp_port *mlxsw_sp_port = neigh_entry->mlxsw_sp_port;
941 struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
942 char rauht_pl[MLXSW_REG_RAUHT_LEN];
943 struct net_device *dev;
944 bool entry_connected;
952 read_lock_bh(&n->lock);
953 dip = ntohl(*((__be32 *) n->primary_key));
954 memcpy(neigh_entry->ha, n->ha, sizeof(neigh_entry->ha));
955 nud_state = n->nud_state;
958 read_unlock_bh(&n->lock);
960 entry_connected = nud_state & NUD_VALID && !dead;
961 adding = (!neigh_entry->offloaded) && entry_connected;
962 updating = neigh_entry->offloaded && entry_connected;
963 removing = neigh_entry->offloaded && !entry_connected;
965 if (adding || updating) {
966 mlxsw_reg_rauht_pack4(rauht_pl, MLXSW_REG_RAUHT_OP_WRITE_ADD,
968 neigh_entry->ha, dip);
969 err = mlxsw_reg_write(mlxsw_sp->core,
970 MLXSW_REG(rauht), rauht_pl);
972 netdev_err(dev, "Could not add neigh %pI4h\n", &dip);
973 neigh_entry->offloaded = false;
975 neigh_entry->offloaded = true;
977 mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, false);
978 } else if (removing) {
979 mlxsw_reg_rauht_pack4(rauht_pl, MLXSW_REG_RAUHT_OP_WRITE_DELETE,
981 neigh_entry->ha, dip);
982 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht),
985 netdev_err(dev, "Could not delete neigh %pI4h\n", &dip);
986 neigh_entry->offloaded = true;
988 neigh_entry->offloaded = false;
990 mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, true);
994 mlxsw_sp_port_dev_put(mlxsw_sp_port);
997 int mlxsw_sp_router_netevent_event(struct notifier_block *unused,
998 unsigned long event, void *ptr)
1000 struct mlxsw_sp_neigh_entry *neigh_entry;
1001 struct mlxsw_sp_port *mlxsw_sp_port;
1002 struct mlxsw_sp *mlxsw_sp;
1003 unsigned long interval;
1004 struct net_device *dev;
1005 struct neigh_parms *p;
1006 struct neighbour *n;
1010 case NETEVENT_DELAY_PROBE_TIME_UPDATE:
1013 /* We don't care about changes in the default table. */
1014 if (!p->dev || p->tbl != &arp_tbl)
1017 /* We are in atomic context and can't take RTNL mutex,
1018 * so use RCU variant to walk the device chain.
1020 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(p->dev);
1024 mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1025 interval = jiffies_to_msecs(NEIGH_VAR(p, DELAY_PROBE_TIME));
1026 mlxsw_sp->router.neighs_update.interval = interval;
1028 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1030 case NETEVENT_NEIGH_UPDATE:
1034 if (n->tbl != &arp_tbl)
1037 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(dev);
1041 mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1042 dip = ntohl(*((__be32 *) n->primary_key));
1043 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
1044 if (WARN_ON(!neigh_entry)) {
1045 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1048 neigh_entry->mlxsw_sp_port = mlxsw_sp_port;
1050 /* Take a reference to ensure the neighbour won't be
1051 * destructed until we drop the reference in delayed
1055 if (!mlxsw_core_schedule_dw(&neigh_entry->dw, 0)) {
1057 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1065 static int mlxsw_sp_neigh_init(struct mlxsw_sp *mlxsw_sp)
1069 err = rhashtable_init(&mlxsw_sp->router.neigh_ht,
1070 &mlxsw_sp_neigh_ht_params);
1074 /* Initialize the polling interval according to the default
1077 mlxsw_sp_router_neighs_update_interval_init(mlxsw_sp);
1079 /* Create the delayed works for the activity_update */
1080 INIT_DELAYED_WORK(&mlxsw_sp->router.neighs_update.dw,
1081 mlxsw_sp_router_neighs_update_work);
1082 INIT_DELAYED_WORK(&mlxsw_sp->router.nexthop_probe_dw,
1083 mlxsw_sp_router_probe_unresolved_nexthops);
1084 mlxsw_core_schedule_dw(&mlxsw_sp->router.neighs_update.dw, 0);
1085 mlxsw_core_schedule_dw(&mlxsw_sp->router.nexthop_probe_dw, 0);
1089 static void mlxsw_sp_neigh_fini(struct mlxsw_sp *mlxsw_sp)
1091 cancel_delayed_work_sync(&mlxsw_sp->router.neighs_update.dw);
1092 cancel_delayed_work_sync(&mlxsw_sp->router.nexthop_probe_dw);
1093 rhashtable_destroy(&mlxsw_sp->router.neigh_ht);
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 u8 should_offload:1, /* set indicates this neigh is connected and
1102 * should be put to KVD linear area of this group.
1104 offloaded:1, /* set in case the neigh is actually put into
1105 * KVD linear area of this group.
1107 update:1; /* set indicates that MAC of this neigh should be
1110 struct mlxsw_sp_neigh_entry *neigh_entry;
1113 struct mlxsw_sp_nexthop_group {
1114 struct list_head list; /* node in mlxsw->router.nexthop_group_list */
1115 struct list_head fib_list; /* list of fib entries that use this group */
1116 u8 adj_index_valid:1;
1120 struct mlxsw_sp_nexthop nexthops[0];
1123 static int mlxsw_sp_adj_index_mass_update_vr(struct mlxsw_sp *mlxsw_sp,
1124 struct mlxsw_sp_vr *vr,
1125 u32 adj_index, u16 ecmp_size,
1129 char raleu_pl[MLXSW_REG_RALEU_LEN];
1131 mlxsw_reg_raleu_pack(raleu_pl,
1132 (enum mlxsw_reg_ralxx_protocol) vr->proto, vr->id,
1133 adj_index, ecmp_size, new_adj_index,
1135 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raleu), raleu_pl);
1138 static int mlxsw_sp_adj_index_mass_update(struct mlxsw_sp *mlxsw_sp,
1139 struct mlxsw_sp_nexthop_group *nh_grp,
1140 u32 old_adj_index, u16 old_ecmp_size)
1142 struct mlxsw_sp_fib_entry *fib_entry;
1143 struct mlxsw_sp_vr *vr = NULL;
1146 list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
1147 if (vr == fib_entry->vr)
1150 err = mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp, vr,
1161 static int mlxsw_sp_nexthop_mac_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
1162 struct mlxsw_sp_nexthop *nh)
1164 struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
1165 char ratr_pl[MLXSW_REG_RATR_LEN];
1167 mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
1168 true, adj_index, neigh_entry->rif);
1169 mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
1170 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
1174 mlxsw_sp_nexthop_group_mac_update(struct mlxsw_sp *mlxsw_sp,
1175 struct mlxsw_sp_nexthop_group *nh_grp,
1178 u32 adj_index = nh_grp->adj_index; /* base */
1179 struct mlxsw_sp_nexthop *nh;
1183 for (i = 0; i < nh_grp->count; i++) {
1184 nh = &nh_grp->nexthops[i];
1186 if (!nh->should_offload) {
1191 if (nh->update || reallocate) {
1192 err = mlxsw_sp_nexthop_mac_update(mlxsw_sp,
1204 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1205 struct mlxsw_sp_fib_entry *fib_entry);
1208 mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
1209 struct mlxsw_sp_nexthop_group *nh_grp)
1211 struct mlxsw_sp_fib_entry *fib_entry;
1214 list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
1215 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1223 mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
1224 struct mlxsw_sp_nexthop_group *nh_grp)
1226 struct mlxsw_sp_nexthop *nh;
1227 bool offload_change = false;
1230 bool old_adj_index_valid;
1237 for (i = 0; i < nh_grp->count; i++) {
1238 nh = &nh_grp->nexthops[i];
1240 if (nh->should_offload ^ nh->offloaded) {
1241 offload_change = true;
1242 if (nh->should_offload)
1245 if (nh->should_offload)
1248 if (!offload_change) {
1249 /* Nothing was added or removed, so no need to reallocate. Just
1250 * update MAC on existing adjacency indexes.
1252 err = mlxsw_sp_nexthop_group_mac_update(mlxsw_sp, nh_grp,
1255 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
1261 /* No neigh of this group is connected so we just set
1262 * the trap and let everthing flow through kernel.
1266 ret = mlxsw_sp_kvdl_alloc(mlxsw_sp, ecmp_size);
1268 /* We ran out of KVD linear space, just set the
1269 * trap and let everything flow through kernel.
1271 dev_warn(mlxsw_sp->bus_info->dev, "Failed to allocate KVD linear area for nexthop group.\n");
1275 old_adj_index_valid = nh_grp->adj_index_valid;
1276 old_adj_index = nh_grp->adj_index;
1277 old_ecmp_size = nh_grp->ecmp_size;
1278 nh_grp->adj_index_valid = 1;
1279 nh_grp->adj_index = adj_index;
1280 nh_grp->ecmp_size = ecmp_size;
1281 err = mlxsw_sp_nexthop_group_mac_update(mlxsw_sp, nh_grp, true);
1283 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
1287 if (!old_adj_index_valid) {
1288 /* The trap was set for fib entries, so we have to call
1289 * fib entry update to unset it and use adjacency index.
1291 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
1293 dev_warn(mlxsw_sp->bus_info->dev, "Failed to add adjacency index to fib entries.\n");
1299 err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, nh_grp,
1300 old_adj_index, old_ecmp_size);
1301 mlxsw_sp_kvdl_free(mlxsw_sp, old_adj_index);
1303 dev_warn(mlxsw_sp->bus_info->dev, "Failed to mass-update adjacency index for nexthop group.\n");
1309 old_adj_index_valid = nh_grp->adj_index_valid;
1310 nh_grp->adj_index_valid = 0;
1311 for (i = 0; i < nh_grp->count; i++) {
1312 nh = &nh_grp->nexthops[i];
1315 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
1317 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set traps for fib entries.\n");
1318 if (old_adj_index_valid)
1319 mlxsw_sp_kvdl_free(mlxsw_sp, nh_grp->adj_index);
1322 static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
1325 if (!removing && !nh->should_offload)
1326 nh->should_offload = 1;
1327 else if (removing && nh->offloaded)
1328 nh->should_offload = 0;
1333 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
1334 struct mlxsw_sp_neigh_entry *neigh_entry,
1337 struct mlxsw_sp_nexthop *nh;
1339 /* Take RTNL mutex here to prevent lists from changes */
1341 list_for_each_entry(nh, &neigh_entry->nexthop_list,
1343 __mlxsw_sp_nexthop_neigh_update(nh, removing);
1344 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
1349 static int mlxsw_sp_nexthop_init(struct mlxsw_sp *mlxsw_sp,
1350 struct mlxsw_sp_nexthop_group *nh_grp,
1351 struct mlxsw_sp_nexthop *nh,
1352 struct fib_nh *fib_nh)
1354 struct mlxsw_sp_neigh_entry *neigh_entry;
1355 struct net_device *dev = fib_nh->nh_dev;
1356 struct neighbour *n;
1359 /* Take a reference of neigh here ensuring that neigh would
1360 * not be detructed before the nexthop entry is finished.
1361 * The reference is taken either in neigh_lookup() or
1362 * in neith_create() in case n is not found.
1364 n = neigh_lookup(&arp_tbl, &fib_nh->nh_gw, dev);
1366 n = neigh_create(&arp_tbl, &fib_nh->nh_gw, dev);
1369 neigh_event_send(n, NULL);
1371 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
1377 /* If that is the first nexthop connected to that neigh, add to
1378 * nexthop_neighs_list
1380 if (list_empty(&neigh_entry->nexthop_list))
1381 list_add_tail(&neigh_entry->nexthop_neighs_list_node,
1382 &mlxsw_sp->router.nexthop_neighs_list);
1384 nh->nh_grp = nh_grp;
1385 nh->neigh_entry = neigh_entry;
1386 list_add_tail(&nh->neigh_list_node, &neigh_entry->nexthop_list);
1387 read_lock_bh(&n->lock);
1388 nud_state = n->nud_state;
1390 read_unlock_bh(&n->lock);
1391 __mlxsw_sp_nexthop_neigh_update(nh, !(nud_state & NUD_VALID && !dead));
1396 static void mlxsw_sp_nexthop_fini(struct mlxsw_sp *mlxsw_sp,
1397 struct mlxsw_sp_nexthop *nh)
1399 struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
1401 __mlxsw_sp_nexthop_neigh_update(nh, true);
1402 list_del(&nh->neigh_list_node);
1404 /* If that is the last nexthop connected to that neigh, remove from
1405 * nexthop_neighs_list
1407 if (list_empty(&nh->neigh_entry->nexthop_list))
1408 list_del(&nh->neigh_entry->nexthop_neighs_list_node);
1410 neigh_release(neigh_entry->key.n);
1413 static struct mlxsw_sp_nexthop_group *
1414 mlxsw_sp_nexthop_group_create(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
1416 struct mlxsw_sp_nexthop_group *nh_grp;
1417 struct mlxsw_sp_nexthop *nh;
1418 struct fib_nh *fib_nh;
1423 alloc_size = sizeof(*nh_grp) +
1424 fi->fib_nhs * sizeof(struct mlxsw_sp_nexthop);
1425 nh_grp = kzalloc(alloc_size, GFP_KERNEL);
1427 return ERR_PTR(-ENOMEM);
1428 INIT_LIST_HEAD(&nh_grp->fib_list);
1429 nh_grp->count = fi->fib_nhs;
1430 for (i = 0; i < nh_grp->count; i++) {
1431 nh = &nh_grp->nexthops[i];
1432 fib_nh = &fi->fib_nh[i];
1433 err = mlxsw_sp_nexthop_init(mlxsw_sp, nh_grp, nh, fib_nh);
1435 goto err_nexthop_init;
1437 list_add_tail(&nh_grp->list, &mlxsw_sp->router.nexthop_group_list);
1438 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
1442 for (i--; i >= 0; i--)
1443 mlxsw_sp_nexthop_fini(mlxsw_sp, nh);
1445 return ERR_PTR(err);
1449 mlxsw_sp_nexthop_group_destroy(struct mlxsw_sp *mlxsw_sp,
1450 struct mlxsw_sp_nexthop_group *nh_grp)
1452 struct mlxsw_sp_nexthop *nh;
1455 list_del(&nh_grp->list);
1456 for (i = 0; i < nh_grp->count; i++) {
1457 nh = &nh_grp->nexthops[i];
1458 mlxsw_sp_nexthop_fini(mlxsw_sp, nh);
1460 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
1461 WARN_ON_ONCE(nh_grp->adj_index_valid);
1465 static bool mlxsw_sp_nexthop_match(struct mlxsw_sp_nexthop *nh,
1466 struct fib_info *fi)
1470 for (i = 0; i < fi->fib_nhs; i++) {
1471 struct fib_nh *fib_nh = &fi->fib_nh[i];
1472 struct neighbour *n = nh->neigh_entry->key.n;
1474 if (memcmp(n->primary_key, &fib_nh->nh_gw,
1475 sizeof(fib_nh->nh_gw)) == 0 &&
1476 n->dev == fib_nh->nh_dev)
1482 static bool mlxsw_sp_nexthop_group_match(struct mlxsw_sp_nexthop_group *nh_grp,
1483 struct fib_info *fi)
1487 if (nh_grp->count != fi->fib_nhs)
1489 for (i = 0; i < nh_grp->count; i++) {
1490 struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
1492 if (!mlxsw_sp_nexthop_match(nh, fi))
1498 static struct mlxsw_sp_nexthop_group *
1499 mlxsw_sp_nexthop_group_find(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
1501 struct mlxsw_sp_nexthop_group *nh_grp;
1503 list_for_each_entry(nh_grp, &mlxsw_sp->router.nexthop_group_list,
1505 if (mlxsw_sp_nexthop_group_match(nh_grp, fi))
1511 static int mlxsw_sp_nexthop_group_get(struct mlxsw_sp *mlxsw_sp,
1512 struct mlxsw_sp_fib_entry *fib_entry,
1513 struct fib_info *fi)
1515 struct mlxsw_sp_nexthop_group *nh_grp;
1517 nh_grp = mlxsw_sp_nexthop_group_find(mlxsw_sp, fi);
1519 nh_grp = mlxsw_sp_nexthop_group_create(mlxsw_sp, fi);
1521 return PTR_ERR(nh_grp);
1523 list_add_tail(&fib_entry->nexthop_group_node, &nh_grp->fib_list);
1524 fib_entry->nh_group = nh_grp;
1528 static void mlxsw_sp_nexthop_group_put(struct mlxsw_sp *mlxsw_sp,
1529 struct mlxsw_sp_fib_entry *fib_entry)
1531 struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
1533 list_del(&fib_entry->nexthop_group_node);
1534 if (!list_empty(&nh_grp->fib_list))
1536 mlxsw_sp_nexthop_group_destroy(mlxsw_sp, nh_grp);
1539 static int mlxsw_sp_fib_entry_op4_remote(struct mlxsw_sp *mlxsw_sp,
1540 struct mlxsw_sp_fib_entry *fib_entry,
1541 enum mlxsw_reg_ralue_op op)
1543 char ralue_pl[MLXSW_REG_RALUE_LEN];
1544 u32 *p_dip = (u32 *) fib_entry->key.addr;
1545 struct mlxsw_sp_vr *vr = fib_entry->vr;
1546 enum mlxsw_reg_ralue_trap_action trap_action;
1548 u32 adjacency_index = 0;
1551 /* In case the nexthop group adjacency index is valid, use it
1552 * with provided ECMP size. Otherwise, setup trap and pass
1553 * traffic to kernel.
1555 if (fib_entry->nh_group->adj_index_valid) {
1556 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
1557 adjacency_index = fib_entry->nh_group->adj_index;
1558 ecmp_size = fib_entry->nh_group->ecmp_size;
1560 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
1561 trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
1564 mlxsw_reg_ralue_pack4(ralue_pl,
1565 (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1566 vr->id, fib_entry->key.prefix_len, *p_dip);
1567 mlxsw_reg_ralue_act_remote_pack(ralue_pl, trap_action, trap_id,
1568 adjacency_index, ecmp_size);
1569 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1572 static int mlxsw_sp_fib_entry_op4_local(struct mlxsw_sp *mlxsw_sp,
1573 struct mlxsw_sp_fib_entry *fib_entry,
1574 enum mlxsw_reg_ralue_op op)
1576 char ralue_pl[MLXSW_REG_RALUE_LEN];
1577 u32 *p_dip = (u32 *) fib_entry->key.addr;
1578 struct mlxsw_sp_vr *vr = fib_entry->vr;
1580 mlxsw_reg_ralue_pack4(ralue_pl,
1581 (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1582 vr->id, fib_entry->key.prefix_len, *p_dip);
1583 mlxsw_reg_ralue_act_local_pack(ralue_pl,
1584 MLXSW_REG_RALUE_TRAP_ACTION_NOP, 0,
1586 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1589 static int mlxsw_sp_fib_entry_op4_trap(struct mlxsw_sp *mlxsw_sp,
1590 struct mlxsw_sp_fib_entry *fib_entry,
1591 enum mlxsw_reg_ralue_op op)
1593 char ralue_pl[MLXSW_REG_RALUE_LEN];
1594 u32 *p_dip = (u32 *) fib_entry->key.addr;
1595 struct mlxsw_sp_vr *vr = fib_entry->vr;
1597 mlxsw_reg_ralue_pack4(ralue_pl,
1598 (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1599 vr->id, fib_entry->key.prefix_len, *p_dip);
1600 mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
1601 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1604 static int mlxsw_sp_fib_entry_op4(struct mlxsw_sp *mlxsw_sp,
1605 struct mlxsw_sp_fib_entry *fib_entry,
1606 enum mlxsw_reg_ralue_op op)
1608 switch (fib_entry->type) {
1609 case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
1610 return mlxsw_sp_fib_entry_op4_remote(mlxsw_sp, fib_entry, op);
1611 case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
1612 return mlxsw_sp_fib_entry_op4_local(mlxsw_sp, fib_entry, op);
1613 case MLXSW_SP_FIB_ENTRY_TYPE_TRAP:
1614 return mlxsw_sp_fib_entry_op4_trap(mlxsw_sp, fib_entry, op);
1619 static int mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
1620 struct mlxsw_sp_fib_entry *fib_entry,
1621 enum mlxsw_reg_ralue_op op)
1623 switch (fib_entry->vr->proto) {
1624 case MLXSW_SP_L3_PROTO_IPV4:
1625 return mlxsw_sp_fib_entry_op4(mlxsw_sp, fib_entry, op);
1626 case MLXSW_SP_L3_PROTO_IPV6:
1632 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1633 struct mlxsw_sp_fib_entry *fib_entry)
1635 return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
1636 MLXSW_REG_RALUE_OP_WRITE_WRITE);
1639 static int mlxsw_sp_fib_entry_del(struct mlxsw_sp *mlxsw_sp,
1640 struct mlxsw_sp_fib_entry *fib_entry)
1642 return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
1643 MLXSW_REG_RALUE_OP_WRITE_DELETE);
1647 mlxsw_sp_router_fib4_entry_init(struct mlxsw_sp *mlxsw_sp,
1648 const struct fib_entry_notifier_info *fen_info,
1649 struct mlxsw_sp_fib_entry *fib_entry)
1651 struct fib_info *fi = fen_info->fi;
1652 struct mlxsw_sp_rif *r = NULL;
1656 if (fen_info->type == RTN_LOCAL || fen_info->type == RTN_BROADCAST) {
1657 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1660 if (fen_info->type != RTN_UNICAST)
1663 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1664 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1668 r = mlxsw_sp_rif_find_by_dev(mlxsw_sp, nh->nh_dev);
1670 /* In case router interface is not found for
1671 * at least one of the nexthops, that means
1672 * the nexthop points to some device unrelated
1673 * to us. Set trap and pass the packets for
1674 * this prefix to kernel.
1681 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1685 if (fi->fib_scope != RT_SCOPE_UNIVERSE) {
1686 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
1687 fib_entry->rif = r->rif;
1689 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
1690 err = mlxsw_sp_nexthop_group_get(mlxsw_sp, fib_entry, fi);
1694 fib_info_offload_inc(fen_info->fi);
1699 mlxsw_sp_router_fib4_entry_fini(struct mlxsw_sp *mlxsw_sp,
1700 struct mlxsw_sp_fib_entry *fib_entry)
1702 if (fib_entry->type != MLXSW_SP_FIB_ENTRY_TYPE_TRAP)
1703 fib_info_offload_dec(fib_entry->fi);
1704 if (fib_entry->type == MLXSW_SP_FIB_ENTRY_TYPE_REMOTE)
1705 mlxsw_sp_nexthop_group_put(mlxsw_sp, fib_entry);
1708 static struct mlxsw_sp_fib_entry *
1709 mlxsw_sp_fib_entry_get(struct mlxsw_sp *mlxsw_sp,
1710 const struct fib_entry_notifier_info *fen_info)
1712 struct mlxsw_sp_fib_entry *fib_entry;
1713 struct fib_info *fi = fen_info->fi;
1714 struct mlxsw_sp_vr *vr;
1717 vr = mlxsw_sp_vr_get(mlxsw_sp, fen_info->dst_len, fen_info->tb_id,
1718 MLXSW_SP_L3_PROTO_IPV4);
1720 return ERR_CAST(vr);
1722 fib_entry = mlxsw_sp_fib_entry_lookup(vr->fib, &fen_info->dst,
1723 sizeof(fen_info->dst),
1724 fen_info->dst_len, fi->fib_dev);
1726 /* Already exists, just take a reference */
1727 fib_entry->ref_count++;
1730 fib_entry = mlxsw_sp_fib_entry_create(vr->fib, &fen_info->dst,
1731 sizeof(fen_info->dst),
1732 fen_info->dst_len, fi->fib_dev);
1735 goto err_fib_entry_create;
1739 fib_entry->ref_count = 1;
1741 err = mlxsw_sp_router_fib4_entry_init(mlxsw_sp, fen_info, fib_entry);
1743 goto err_fib4_entry_init;
1747 err_fib4_entry_init:
1748 mlxsw_sp_fib_entry_destroy(fib_entry);
1749 err_fib_entry_create:
1750 mlxsw_sp_vr_put(mlxsw_sp, vr);
1752 return ERR_PTR(err);
1755 static struct mlxsw_sp_fib_entry *
1756 mlxsw_sp_fib_entry_find(struct mlxsw_sp *mlxsw_sp,
1757 const struct fib_entry_notifier_info *fen_info)
1759 struct mlxsw_sp_vr *vr;
1761 vr = mlxsw_sp_vr_find(mlxsw_sp, fen_info->tb_id,
1762 MLXSW_SP_L3_PROTO_IPV4);
1766 return mlxsw_sp_fib_entry_lookup(vr->fib, &fen_info->dst,
1767 sizeof(fen_info->dst),
1769 fen_info->fi->fib_dev);
1772 static void mlxsw_sp_fib_entry_put(struct mlxsw_sp *mlxsw_sp,
1773 struct mlxsw_sp_fib_entry *fib_entry)
1775 struct mlxsw_sp_vr *vr = fib_entry->vr;
1777 if (--fib_entry->ref_count == 0) {
1778 mlxsw_sp_router_fib4_entry_fini(mlxsw_sp, fib_entry);
1779 mlxsw_sp_fib_entry_destroy(fib_entry);
1781 mlxsw_sp_vr_put(mlxsw_sp, vr);
1784 static void mlxsw_sp_fib_entry_put_all(struct mlxsw_sp *mlxsw_sp,
1785 struct mlxsw_sp_fib_entry *fib_entry)
1787 unsigned int last_ref_count;
1790 last_ref_count = fib_entry->ref_count;
1791 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1792 } while (last_ref_count != 1);
1795 static int mlxsw_sp_router_fib4_add(struct mlxsw_sp *mlxsw_sp,
1796 struct fib_entry_notifier_info *fen_info)
1798 struct mlxsw_sp_fib_entry *fib_entry;
1799 struct mlxsw_sp_vr *vr;
1802 if (mlxsw_sp->router.aborted)
1805 fib_entry = mlxsw_sp_fib_entry_get(mlxsw_sp, fen_info);
1806 if (IS_ERR(fib_entry)) {
1807 dev_warn(mlxsw_sp->bus_info->dev, "Failed to get FIB4 entry being added.\n");
1808 return PTR_ERR(fib_entry);
1811 if (fib_entry->ref_count != 1)
1815 err = mlxsw_sp_fib_entry_insert(vr->fib, fib_entry);
1817 dev_warn(mlxsw_sp->bus_info->dev, "Failed to insert FIB4 entry being added.\n");
1818 goto err_fib_entry_insert;
1820 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1822 goto err_fib_entry_add;
1826 mlxsw_sp_fib_entry_remove(vr->fib, fib_entry);
1827 err_fib_entry_insert:
1828 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1832 static void mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
1833 struct fib_entry_notifier_info *fen_info)
1835 struct mlxsw_sp_fib_entry *fib_entry;
1837 if (mlxsw_sp->router.aborted)
1840 fib_entry = mlxsw_sp_fib_entry_find(mlxsw_sp, fen_info);
1844 if (fib_entry->ref_count == 1) {
1845 mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
1846 mlxsw_sp_fib_entry_remove(fib_entry->vr->fib, fib_entry);
1849 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1852 static int mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp)
1854 char ralta_pl[MLXSW_REG_RALTA_LEN];
1855 char ralst_pl[MLXSW_REG_RALST_LEN];
1856 char raltb_pl[MLXSW_REG_RALTB_LEN];
1857 char ralue_pl[MLXSW_REG_RALUE_LEN];
1860 mlxsw_reg_ralta_pack(ralta_pl, true, MLXSW_REG_RALXX_PROTOCOL_IPV4,
1861 MLXSW_SP_LPM_TREE_MIN);
1862 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
1866 mlxsw_reg_ralst_pack(ralst_pl, 0xff, MLXSW_SP_LPM_TREE_MIN);
1867 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
1871 mlxsw_reg_raltb_pack(raltb_pl, 0, MLXSW_REG_RALXX_PROTOCOL_IPV4,
1872 MLXSW_SP_LPM_TREE_MIN);
1873 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
1877 mlxsw_reg_ralue_pack4(ralue_pl, MLXSW_SP_L3_PROTO_IPV4,
1878 MLXSW_REG_RALUE_OP_WRITE_WRITE, 0, 0, 0);
1879 mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
1880 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1883 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp)
1885 struct mlxsw_sp_fib_entry *fib_entry;
1886 struct mlxsw_sp_fib_entry *tmp;
1887 struct mlxsw_sp_vr *vr;
1890 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
1891 vr = &mlxsw_sp->router.vrs[i];
1896 list_for_each_entry_safe(fib_entry, tmp,
1897 &vr->fib->entry_list, list) {
1898 bool do_break = &tmp->list == &vr->fib->entry_list;
1900 mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
1901 mlxsw_sp_fib_entry_remove(fib_entry->vr->fib,
1903 mlxsw_sp_fib_entry_put_all(mlxsw_sp, fib_entry);
1910 static void mlxsw_sp_router_fib4_abort(struct mlxsw_sp *mlxsw_sp)
1914 if (mlxsw_sp->router.aborted)
1916 dev_warn(mlxsw_sp->bus_info->dev, "FIB abort triggered. Note that FIB entries are no longer being offloaded to this device.\n");
1917 mlxsw_sp_router_fib_flush(mlxsw_sp);
1918 mlxsw_sp->router.aborted = true;
1919 err = mlxsw_sp_router_set_abort_trap(mlxsw_sp);
1921 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set abort trap.\n");
1924 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
1926 char rgcr_pl[MLXSW_REG_RGCR_LEN];
1930 if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_RIFS))
1933 max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
1934 mlxsw_sp->rifs = kcalloc(max_rifs, sizeof(struct mlxsw_sp_rif *),
1936 if (!mlxsw_sp->rifs)
1939 mlxsw_reg_rgcr_pack(rgcr_pl, true);
1940 mlxsw_reg_rgcr_max_router_interfaces_set(rgcr_pl, max_rifs);
1941 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
1948 kfree(mlxsw_sp->rifs);
1952 static void __mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
1954 char rgcr_pl[MLXSW_REG_RGCR_LEN];
1957 mlxsw_reg_rgcr_pack(rgcr_pl, false);
1958 mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
1960 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
1961 WARN_ON_ONCE(mlxsw_sp->rifs[i]);
1963 kfree(mlxsw_sp->rifs);
1966 struct mlxsw_sp_fib_event_work {
1967 struct delayed_work dw;
1968 struct fib_entry_notifier_info fen_info;
1969 struct mlxsw_sp *mlxsw_sp;
1970 unsigned long event;
1973 static void mlxsw_sp_router_fib_event_work(struct work_struct *work)
1975 struct mlxsw_sp_fib_event_work *fib_work =
1976 container_of(work, struct mlxsw_sp_fib_event_work, dw.work);
1977 struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
1980 /* Protect internal structures from changes */
1982 switch (fib_work->event) {
1983 case FIB_EVENT_ENTRY_ADD:
1984 err = mlxsw_sp_router_fib4_add(mlxsw_sp, &fib_work->fen_info);
1986 mlxsw_sp_router_fib4_abort(mlxsw_sp);
1987 fib_info_put(fib_work->fen_info.fi);
1989 case FIB_EVENT_ENTRY_DEL:
1990 mlxsw_sp_router_fib4_del(mlxsw_sp, &fib_work->fen_info);
1991 fib_info_put(fib_work->fen_info.fi);
1993 case FIB_EVENT_RULE_ADD: /* fall through */
1994 case FIB_EVENT_RULE_DEL:
1995 mlxsw_sp_router_fib4_abort(mlxsw_sp);
2002 /* Called with rcu_read_lock() */
2003 static int mlxsw_sp_router_fib_event(struct notifier_block *nb,
2004 unsigned long event, void *ptr)
2006 struct mlxsw_sp *mlxsw_sp = container_of(nb, struct mlxsw_sp, fib_nb);
2007 struct mlxsw_sp_fib_event_work *fib_work;
2008 struct fib_notifier_info *info = ptr;
2010 if (!net_eq(info->net, &init_net))
2013 fib_work = kzalloc(sizeof(*fib_work), GFP_ATOMIC);
2014 if (WARN_ON(!fib_work))
2017 INIT_DELAYED_WORK(&fib_work->dw, mlxsw_sp_router_fib_event_work);
2018 fib_work->mlxsw_sp = mlxsw_sp;
2019 fib_work->event = event;
2022 case FIB_EVENT_ENTRY_ADD: /* fall through */
2023 case FIB_EVENT_ENTRY_DEL:
2024 memcpy(&fib_work->fen_info, ptr, sizeof(fib_work->fen_info));
2025 /* Take referece on fib_info to prevent it from being
2026 * freed while work is queued. Release it afterwards.
2028 fib_info_hold(fib_work->fen_info.fi);
2032 mlxsw_core_schedule_odw(&fib_work->dw, 0);
2037 static void mlxsw_sp_router_fib_dump_flush(struct notifier_block *nb)
2039 struct mlxsw_sp *mlxsw_sp = container_of(nb, struct mlxsw_sp, fib_nb);
2041 /* Flush pending FIB notifications and then flush the device's
2042 * table before requesting another dump. The FIB notification
2043 * block is unregistered, so no need to take RTNL.
2045 mlxsw_core_flush_owq();
2046 mlxsw_sp_router_fib_flush(mlxsw_sp);
2049 int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
2053 INIT_LIST_HEAD(&mlxsw_sp->router.nexthop_neighs_list);
2054 INIT_LIST_HEAD(&mlxsw_sp->router.nexthop_group_list);
2055 err = __mlxsw_sp_router_init(mlxsw_sp);
2059 mlxsw_sp_lpm_init(mlxsw_sp);
2060 err = mlxsw_sp_vrs_init(mlxsw_sp);
2064 err = mlxsw_sp_neigh_init(mlxsw_sp);
2066 goto err_neigh_init;
2068 mlxsw_sp->fib_nb.notifier_call = mlxsw_sp_router_fib_event;
2069 err = register_fib_notifier(&mlxsw_sp->fib_nb,
2070 mlxsw_sp_router_fib_dump_flush);
2072 goto err_register_fib_notifier;
2076 err_register_fib_notifier:
2077 mlxsw_sp_neigh_fini(mlxsw_sp);
2079 mlxsw_sp_vrs_fini(mlxsw_sp);
2081 __mlxsw_sp_router_fini(mlxsw_sp);
2085 void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
2087 unregister_fib_notifier(&mlxsw_sp->fib_nb);
2088 mlxsw_sp_neigh_fini(mlxsw_sp);
2089 mlxsw_sp_vrs_fini(mlxsw_sp);
2090 __mlxsw_sp_router_fini(mlxsw_sp);