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_resources *resources;
386 struct mlxsw_sp_vr *vr;
389 resources = mlxsw_core_resources_get(mlxsw_sp->core);
390 for (i = 0; i < resources->max_virtual_routers; i++) {
391 vr = &mlxsw_sp->router.vrs[i];
398 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
399 struct mlxsw_sp_vr *vr)
401 char raltb_pl[MLXSW_REG_RALTB_LEN];
403 mlxsw_reg_raltb_pack(raltb_pl, vr->id,
404 (enum mlxsw_reg_ralxx_protocol) vr->proto,
406 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
409 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
410 struct mlxsw_sp_vr *vr)
412 char raltb_pl[MLXSW_REG_RALTB_LEN];
414 /* Bind to tree 0 which is default */
415 mlxsw_reg_raltb_pack(raltb_pl, vr->id,
416 (enum mlxsw_reg_ralxx_protocol) vr->proto, 0);
417 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
420 static u32 mlxsw_sp_fix_tb_id(u32 tb_id)
422 /* For our purpose, squash main and local table into one */
423 if (tb_id == RT_TABLE_LOCAL)
424 tb_id = RT_TABLE_MAIN;
428 static struct mlxsw_sp_vr *mlxsw_sp_vr_find(struct mlxsw_sp *mlxsw_sp,
430 enum mlxsw_sp_l3proto proto)
432 struct mlxsw_resources *resources;
433 struct mlxsw_sp_vr *vr;
436 tb_id = mlxsw_sp_fix_tb_id(tb_id);
438 resources = mlxsw_core_resources_get(mlxsw_sp->core);
439 for (i = 0; i < resources->max_virtual_routers; i++) {
440 vr = &mlxsw_sp->router.vrs[i];
441 if (vr->used && vr->proto == proto && vr->tb_id == tb_id)
447 static struct mlxsw_sp_vr *mlxsw_sp_vr_create(struct mlxsw_sp *mlxsw_sp,
448 unsigned char prefix_len,
450 enum mlxsw_sp_l3proto proto)
452 struct mlxsw_sp_prefix_usage req_prefix_usage;
453 struct mlxsw_sp_lpm_tree *lpm_tree;
454 struct mlxsw_sp_vr *vr;
457 vr = mlxsw_sp_vr_find_unused(mlxsw_sp);
459 return ERR_PTR(-EBUSY);
460 vr->fib = mlxsw_sp_fib_create();
462 return ERR_CAST(vr->fib);
466 mlxsw_sp_prefix_usage_zero(&req_prefix_usage);
467 mlxsw_sp_prefix_usage_set(&req_prefix_usage, prefix_len);
468 lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
470 if (IS_ERR(lpm_tree)) {
471 err = PTR_ERR(lpm_tree);
474 vr->lpm_tree = lpm_tree;
475 err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, vr);
483 mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
485 mlxsw_sp_fib_destroy(vr->fib);
490 static void mlxsw_sp_vr_destroy(struct mlxsw_sp *mlxsw_sp,
491 struct mlxsw_sp_vr *vr)
493 mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, vr);
494 mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
495 mlxsw_sp_fib_destroy(vr->fib);
500 mlxsw_sp_vr_lpm_tree_check(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr,
501 struct mlxsw_sp_prefix_usage *req_prefix_usage)
503 struct mlxsw_sp_lpm_tree *lpm_tree;
505 if (mlxsw_sp_prefix_usage_eq(req_prefix_usage,
506 &vr->lpm_tree->prefix_usage))
509 lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, req_prefix_usage,
511 if (IS_ERR(lpm_tree)) {
512 /* We failed to get a tree according to the required
513 * prefix usage. However, the current tree might be still good
514 * for us if our requirement is subset of the prefixes used
517 if (mlxsw_sp_prefix_usage_subset(req_prefix_usage,
518 &vr->lpm_tree->prefix_usage))
520 return PTR_ERR(lpm_tree);
523 mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, vr);
524 mlxsw_sp_lpm_tree_put(mlxsw_sp, vr->lpm_tree);
525 vr->lpm_tree = lpm_tree;
526 return mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, vr);
529 static struct mlxsw_sp_vr *mlxsw_sp_vr_get(struct mlxsw_sp *mlxsw_sp,
530 unsigned char prefix_len,
532 enum mlxsw_sp_l3proto proto)
534 struct mlxsw_sp_vr *vr;
537 tb_id = mlxsw_sp_fix_tb_id(tb_id);
538 vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id, proto);
540 vr = mlxsw_sp_vr_create(mlxsw_sp, prefix_len, tb_id, proto);
544 struct mlxsw_sp_prefix_usage req_prefix_usage;
546 mlxsw_sp_prefix_usage_cpy(&req_prefix_usage,
547 &vr->fib->prefix_usage);
548 mlxsw_sp_prefix_usage_set(&req_prefix_usage, prefix_len);
549 /* Need to replace LPM tree in case new prefix is required. */
550 err = mlxsw_sp_vr_lpm_tree_check(mlxsw_sp, vr,
558 static void mlxsw_sp_vr_put(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr)
560 /* Destroy virtual router entity in case the associated FIB is empty
561 * and allow it to be used for other tables in future. Otherwise,
562 * check if some prefix usage did not disappear and change tree if
563 * that is the case. Note that in case new, smaller tree cannot be
564 * allocated, the original one will be kept being used.
566 if (mlxsw_sp_prefix_usage_none(&vr->fib->prefix_usage))
567 mlxsw_sp_vr_destroy(mlxsw_sp, vr);
569 mlxsw_sp_vr_lpm_tree_check(mlxsw_sp, vr,
570 &vr->fib->prefix_usage);
573 static int mlxsw_sp_vrs_init(struct mlxsw_sp *mlxsw_sp)
575 struct mlxsw_resources *resources;
576 struct mlxsw_sp_vr *vr;
579 resources = mlxsw_core_resources_get(mlxsw_sp->core);
580 if (!resources->max_virtual_routers_valid)
583 mlxsw_sp->router.vrs = kcalloc(resources->max_virtual_routers,
584 sizeof(struct mlxsw_sp_vr),
586 if (!mlxsw_sp->router.vrs)
589 for (i = 0; i < resources->max_virtual_routers; i++) {
590 vr = &mlxsw_sp->router.vrs[i];
597 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp);
599 static void mlxsw_sp_vrs_fini(struct mlxsw_sp *mlxsw_sp)
601 mlxsw_sp_router_fib_flush(mlxsw_sp);
602 kfree(mlxsw_sp->router.vrs);
605 struct mlxsw_sp_neigh_key {
609 struct mlxsw_sp_neigh_entry {
610 struct rhash_head ht_node;
611 struct mlxsw_sp_neigh_key key;
614 struct delayed_work dw;
615 struct mlxsw_sp_port *mlxsw_sp_port;
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),
630 mlxsw_sp_neigh_entry_insert(struct mlxsw_sp *mlxsw_sp,
631 struct mlxsw_sp_neigh_entry *neigh_entry)
633 return rhashtable_insert_fast(&mlxsw_sp->router.neigh_ht,
634 &neigh_entry->ht_node,
635 mlxsw_sp_neigh_ht_params);
639 mlxsw_sp_neigh_entry_remove(struct mlxsw_sp *mlxsw_sp,
640 struct mlxsw_sp_neigh_entry *neigh_entry)
642 rhashtable_remove_fast(&mlxsw_sp->router.neigh_ht,
643 &neigh_entry->ht_node,
644 mlxsw_sp_neigh_ht_params);
647 static void mlxsw_sp_router_neigh_update_hw(struct work_struct *work);
649 static struct mlxsw_sp_neigh_entry *
650 mlxsw_sp_neigh_entry_create(struct neighbour *n, u16 rif)
652 struct mlxsw_sp_neigh_entry *neigh_entry;
654 neigh_entry = kzalloc(sizeof(*neigh_entry), GFP_ATOMIC);
657 neigh_entry->key.n = n;
658 neigh_entry->rif = rif;
659 INIT_DELAYED_WORK(&neigh_entry->dw, mlxsw_sp_router_neigh_update_hw);
660 INIT_LIST_HEAD(&neigh_entry->nexthop_list);
665 mlxsw_sp_neigh_entry_destroy(struct mlxsw_sp_neigh_entry *neigh_entry)
670 static struct mlxsw_sp_neigh_entry *
671 mlxsw_sp_neigh_entry_lookup(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
673 struct mlxsw_sp_neigh_key key;
676 return rhashtable_lookup_fast(&mlxsw_sp->router.neigh_ht,
677 &key, mlxsw_sp_neigh_ht_params);
680 int mlxsw_sp_router_neigh_construct(struct net_device *dev,
683 struct mlxsw_sp_port *mlxsw_sp_port = netdev_priv(dev);
684 struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
685 struct mlxsw_sp_neigh_entry *neigh_entry;
686 struct mlxsw_sp_rif *r;
689 if (n->tbl != &arp_tbl)
692 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
696 r = mlxsw_sp_rif_find_by_dev(mlxsw_sp, n->dev);
700 neigh_entry = mlxsw_sp_neigh_entry_create(n, r->rif);
703 err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
705 goto err_neigh_entry_insert;
708 err_neigh_entry_insert:
709 mlxsw_sp_neigh_entry_destroy(neigh_entry);
713 void mlxsw_sp_router_neigh_destroy(struct net_device *dev,
716 struct mlxsw_sp_port *mlxsw_sp_port = netdev_priv(dev);
717 struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
718 struct mlxsw_sp_neigh_entry *neigh_entry;
720 if (n->tbl != &arp_tbl)
723 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
726 mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
727 mlxsw_sp_neigh_entry_destroy(neigh_entry);
731 mlxsw_sp_router_neighs_update_interval_init(struct mlxsw_sp *mlxsw_sp)
733 unsigned long interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME);
735 mlxsw_sp->router.neighs_update.interval = jiffies_to_msecs(interval);
738 static void mlxsw_sp_router_neigh_ent_ipv4_process(struct mlxsw_sp *mlxsw_sp,
742 struct net_device *dev;
748 mlxsw_reg_rauhtd_ent_ipv4_unpack(rauhtd_pl, ent_index, &rif, &dip);
750 if (!mlxsw_sp->rifs[rif]) {
751 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
756 dev = mlxsw_sp->rifs[rif]->dev;
757 n = neigh_lookup(&arp_tbl, &dipn, dev);
759 netdev_err(dev, "Failed to find matching neighbour for IP=%pI4h\n",
764 netdev_dbg(dev, "Updating neighbour with IP=%pI4h\n", &dip);
765 neigh_event_send(n, NULL);
769 static void mlxsw_sp_router_neigh_rec_ipv4_process(struct mlxsw_sp *mlxsw_sp,
776 num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
778 /* Hardware starts counting at 0, so add 1. */
781 /* Each record consists of several neighbour entries. */
782 for (i = 0; i < num_entries; i++) {
785 ent_index = rec_index * MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC + i;
786 mlxsw_sp_router_neigh_ent_ipv4_process(mlxsw_sp, rauhtd_pl,
792 static void mlxsw_sp_router_neigh_rec_process(struct mlxsw_sp *mlxsw_sp,
793 char *rauhtd_pl, int rec_index)
795 switch (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, rec_index)) {
796 case MLXSW_REG_RAUHTD_TYPE_IPV4:
797 mlxsw_sp_router_neigh_rec_ipv4_process(mlxsw_sp, rauhtd_pl,
800 case MLXSW_REG_RAUHTD_TYPE_IPV6:
806 static bool mlxsw_sp_router_rauhtd_is_full(char *rauhtd_pl)
808 u8 num_rec, last_rec_index, num_entries;
810 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
811 last_rec_index = num_rec - 1;
813 if (num_rec < MLXSW_REG_RAUHTD_REC_MAX_NUM)
815 if (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, last_rec_index) ==
816 MLXSW_REG_RAUHTD_TYPE_IPV6)
819 num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
821 if (++num_entries == MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC)
826 static int mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp)
832 rauhtd_pl = kmalloc(MLXSW_REG_RAUHTD_LEN, GFP_KERNEL);
836 /* Make sure the neighbour's netdev isn't removed in the
841 mlxsw_reg_rauhtd_pack(rauhtd_pl, MLXSW_REG_RAUHTD_TYPE_IPV4);
842 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(rauhtd),
845 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Failed to dump neighbour talbe\n");
848 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
849 for (i = 0; i < num_rec; i++)
850 mlxsw_sp_router_neigh_rec_process(mlxsw_sp, rauhtd_pl,
852 } while (mlxsw_sp_router_rauhtd_is_full(rauhtd_pl));
859 static void mlxsw_sp_router_neighs_update_nh(struct mlxsw_sp *mlxsw_sp)
861 struct mlxsw_sp_neigh_entry *neigh_entry;
863 /* Take RTNL mutex here to prevent lists from changes */
865 list_for_each_entry(neigh_entry, &mlxsw_sp->router.nexthop_neighs_list,
866 nexthop_neighs_list_node) {
867 /* If this neigh have nexthops, make the kernel think this neigh
868 * is active regardless of the traffic.
870 if (!list_empty(&neigh_entry->nexthop_list))
871 neigh_event_send(neigh_entry->key.n, NULL);
877 mlxsw_sp_router_neighs_update_work_schedule(struct mlxsw_sp *mlxsw_sp)
879 unsigned long interval = mlxsw_sp->router.neighs_update.interval;
881 mlxsw_core_schedule_dw(&mlxsw_sp->router.neighs_update.dw,
882 msecs_to_jiffies(interval));
885 static void mlxsw_sp_router_neighs_update_work(struct work_struct *work)
887 struct mlxsw_sp *mlxsw_sp = container_of(work, struct mlxsw_sp,
888 router.neighs_update.dw.work);
891 err = mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp);
893 dev_err(mlxsw_sp->bus_info->dev, "Could not update kernel for neigh activity");
895 mlxsw_sp_router_neighs_update_nh(mlxsw_sp);
897 mlxsw_sp_router_neighs_update_work_schedule(mlxsw_sp);
900 static void mlxsw_sp_router_probe_unresolved_nexthops(struct work_struct *work)
902 struct mlxsw_sp_neigh_entry *neigh_entry;
903 struct mlxsw_sp *mlxsw_sp = container_of(work, struct mlxsw_sp,
904 router.nexthop_probe_dw.work);
906 /* Iterate over nexthop neighbours, find those who are unresolved and
907 * send arp on them. This solves the chicken-egg problem when
908 * the nexthop wouldn't get offloaded until the neighbor is resolved
909 * but it wouldn't get resolved ever in case traffic is flowing in HW
910 * using different nexthop.
912 * Take RTNL mutex here to prevent lists from changes.
915 list_for_each_entry(neigh_entry, &mlxsw_sp->router.nexthop_neighs_list,
916 nexthop_neighs_list_node) {
917 if (!(neigh_entry->key.n->nud_state & NUD_VALID) &&
918 !list_empty(&neigh_entry->nexthop_list))
919 neigh_event_send(neigh_entry->key.n, NULL);
923 mlxsw_core_schedule_dw(&mlxsw_sp->router.nexthop_probe_dw,
924 MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL);
928 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
929 struct mlxsw_sp_neigh_entry *neigh_entry,
932 static void mlxsw_sp_router_neigh_update_hw(struct work_struct *work)
934 struct mlxsw_sp_neigh_entry *neigh_entry =
935 container_of(work, struct mlxsw_sp_neigh_entry, dw.work);
936 struct neighbour *n = neigh_entry->key.n;
937 struct mlxsw_sp_port *mlxsw_sp_port = neigh_entry->mlxsw_sp_port;
938 struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
939 char rauht_pl[MLXSW_REG_RAUHT_LEN];
940 struct net_device *dev;
941 bool entry_connected;
949 read_lock_bh(&n->lock);
950 dip = ntohl(*((__be32 *) n->primary_key));
951 memcpy(neigh_entry->ha, n->ha, sizeof(neigh_entry->ha));
952 nud_state = n->nud_state;
954 read_unlock_bh(&n->lock);
956 entry_connected = nud_state & NUD_VALID;
957 adding = (!neigh_entry->offloaded) && entry_connected;
958 updating = neigh_entry->offloaded && entry_connected;
959 removing = neigh_entry->offloaded && !entry_connected;
961 if (adding || updating) {
962 mlxsw_reg_rauht_pack4(rauht_pl, MLXSW_REG_RAUHT_OP_WRITE_ADD,
964 neigh_entry->ha, dip);
965 err = mlxsw_reg_write(mlxsw_sp->core,
966 MLXSW_REG(rauht), rauht_pl);
968 netdev_err(dev, "Could not add neigh %pI4h\n", &dip);
969 neigh_entry->offloaded = false;
971 neigh_entry->offloaded = true;
973 mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, false);
974 } else if (removing) {
975 mlxsw_reg_rauht_pack4(rauht_pl, MLXSW_REG_RAUHT_OP_WRITE_DELETE,
977 neigh_entry->ha, dip);
978 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht),
981 netdev_err(dev, "Could not delete neigh %pI4h\n", &dip);
982 neigh_entry->offloaded = true;
984 neigh_entry->offloaded = false;
986 mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, true);
990 mlxsw_sp_port_dev_put(mlxsw_sp_port);
993 int mlxsw_sp_router_netevent_event(struct notifier_block *unused,
994 unsigned long event, void *ptr)
996 struct mlxsw_sp_neigh_entry *neigh_entry;
997 struct mlxsw_sp_port *mlxsw_sp_port;
998 struct mlxsw_sp *mlxsw_sp;
999 unsigned long interval;
1000 struct net_device *dev;
1001 struct neigh_parms *p;
1002 struct neighbour *n;
1006 case NETEVENT_DELAY_PROBE_TIME_UPDATE:
1009 /* We don't care about changes in the default table. */
1010 if (!p->dev || p->tbl != &arp_tbl)
1013 /* We are in atomic context and can't take RTNL mutex,
1014 * so use RCU variant to walk the device chain.
1016 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(p->dev);
1020 mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1021 interval = jiffies_to_msecs(NEIGH_VAR(p, DELAY_PROBE_TIME));
1022 mlxsw_sp->router.neighs_update.interval = interval;
1024 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1026 case NETEVENT_NEIGH_UPDATE:
1030 if (n->tbl != &arp_tbl)
1033 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(dev);
1037 mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1038 dip = ntohl(*((__be32 *) n->primary_key));
1039 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
1040 if (WARN_ON(!neigh_entry)) {
1041 mlxsw_sp_port_dev_put(mlxsw_sp_port);
1044 neigh_entry->mlxsw_sp_port = mlxsw_sp_port;
1046 /* Take a reference to ensure the neighbour won't be
1047 * destructed until we drop the reference in delayed
1051 if (!mlxsw_core_schedule_dw(&neigh_entry->dw, 0)) {
1053 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 {
1093 struct list_head neigh_list_node; /* member of neigh entry list */
1094 struct mlxsw_sp_nexthop_group *nh_grp; /* pointer back to the group
1097 u8 should_offload:1, /* set indicates this neigh is connected and
1098 * should be put to KVD linear area of this group.
1100 offloaded:1, /* set in case the neigh is actually put into
1101 * KVD linear area of this group.
1103 update:1; /* set indicates that MAC of this neigh should be
1106 struct mlxsw_sp_neigh_entry *neigh_entry;
1109 struct mlxsw_sp_nexthop_group {
1110 struct list_head list; /* node in mlxsw->router.nexthop_group_list */
1111 struct list_head fib_list; /* list of fib entries that use this group */
1112 u8 adj_index_valid:1;
1116 struct mlxsw_sp_nexthop nexthops[0];
1119 static int mlxsw_sp_adj_index_mass_update_vr(struct mlxsw_sp *mlxsw_sp,
1120 struct mlxsw_sp_vr *vr,
1121 u32 adj_index, u16 ecmp_size,
1125 char raleu_pl[MLXSW_REG_RALEU_LEN];
1127 mlxsw_reg_raleu_pack(raleu_pl,
1128 (enum mlxsw_reg_ralxx_protocol) vr->proto, vr->id,
1129 adj_index, ecmp_size, new_adj_index,
1131 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raleu), raleu_pl);
1134 static int mlxsw_sp_adj_index_mass_update(struct mlxsw_sp *mlxsw_sp,
1135 struct mlxsw_sp_nexthop_group *nh_grp,
1136 u32 old_adj_index, u16 old_ecmp_size)
1138 struct mlxsw_sp_fib_entry *fib_entry;
1139 struct mlxsw_sp_vr *vr = NULL;
1142 list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
1143 if (vr == fib_entry->vr)
1146 err = mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp, vr,
1157 static int mlxsw_sp_nexthop_mac_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
1158 struct mlxsw_sp_nexthop *nh)
1160 struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
1161 char ratr_pl[MLXSW_REG_RATR_LEN];
1163 mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
1164 true, adj_index, neigh_entry->rif);
1165 mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
1166 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
1170 mlxsw_sp_nexthop_group_mac_update(struct mlxsw_sp *mlxsw_sp,
1171 struct mlxsw_sp_nexthop_group *nh_grp)
1173 u32 adj_index = nh_grp->adj_index; /* base */
1174 struct mlxsw_sp_nexthop *nh;
1178 for (i = 0; i < nh_grp->count; i++) {
1179 nh = &nh_grp->nexthops[i];
1181 if (!nh->should_offload) {
1187 err = mlxsw_sp_nexthop_mac_update(mlxsw_sp,
1199 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1200 struct mlxsw_sp_fib_entry *fib_entry);
1203 mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
1204 struct mlxsw_sp_nexthop_group *nh_grp)
1206 struct mlxsw_sp_fib_entry *fib_entry;
1209 list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
1210 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1218 mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
1219 struct mlxsw_sp_nexthop_group *nh_grp)
1221 struct mlxsw_sp_nexthop *nh;
1222 bool offload_change = false;
1225 bool old_adj_index_valid;
1232 for (i = 0; i < nh_grp->count; i++) {
1233 nh = &nh_grp->nexthops[i];
1235 if (nh->should_offload ^ nh->offloaded) {
1236 offload_change = true;
1237 if (nh->should_offload)
1240 if (nh->should_offload)
1243 if (!offload_change) {
1244 /* Nothing was added or removed, so no need to reallocate. Just
1245 * update MAC on existing adjacency indexes.
1247 err = mlxsw_sp_nexthop_group_mac_update(mlxsw_sp, nh_grp);
1249 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
1255 /* No neigh of this group is connected so we just set
1256 * the trap and let everthing flow through kernel.
1260 ret = mlxsw_sp_kvdl_alloc(mlxsw_sp, ecmp_size);
1262 /* We ran out of KVD linear space, just set the
1263 * trap and let everything flow through kernel.
1265 dev_warn(mlxsw_sp->bus_info->dev, "Failed to allocate KVD linear area for nexthop group.\n");
1269 old_adj_index_valid = nh_grp->adj_index_valid;
1270 old_adj_index = nh_grp->adj_index;
1271 old_ecmp_size = nh_grp->ecmp_size;
1272 nh_grp->adj_index_valid = 1;
1273 nh_grp->adj_index = adj_index;
1274 nh_grp->ecmp_size = ecmp_size;
1275 err = mlxsw_sp_nexthop_group_mac_update(mlxsw_sp, nh_grp);
1277 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
1281 if (!old_adj_index_valid) {
1282 /* The trap was set for fib entries, so we have to call
1283 * fib entry update to unset it and use adjacency index.
1285 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
1287 dev_warn(mlxsw_sp->bus_info->dev, "Failed to add adjacency index to fib entries.\n");
1293 err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, nh_grp,
1294 old_adj_index, old_ecmp_size);
1295 mlxsw_sp_kvdl_free(mlxsw_sp, old_adj_index);
1297 dev_warn(mlxsw_sp->bus_info->dev, "Failed to mass-update adjacency index for nexthop group.\n");
1303 old_adj_index_valid = nh_grp->adj_index_valid;
1304 nh_grp->adj_index_valid = 0;
1305 for (i = 0; i < nh_grp->count; i++) {
1306 nh = &nh_grp->nexthops[i];
1309 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
1311 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set traps for fib entries.\n");
1312 if (old_adj_index_valid)
1313 mlxsw_sp_kvdl_free(mlxsw_sp, nh_grp->adj_index);
1316 static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
1319 if (!removing && !nh->should_offload)
1320 nh->should_offload = 1;
1321 else if (removing && nh->offloaded)
1322 nh->should_offload = 0;
1327 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
1328 struct mlxsw_sp_neigh_entry *neigh_entry,
1331 struct mlxsw_sp_nexthop *nh;
1333 /* Take RTNL mutex here to prevent lists from changes */
1335 list_for_each_entry(nh, &neigh_entry->nexthop_list,
1337 __mlxsw_sp_nexthop_neigh_update(nh, removing);
1338 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
1343 static int mlxsw_sp_nexthop_init(struct mlxsw_sp *mlxsw_sp,
1344 struct mlxsw_sp_nexthop_group *nh_grp,
1345 struct mlxsw_sp_nexthop *nh,
1346 struct fib_nh *fib_nh)
1348 struct mlxsw_sp_neigh_entry *neigh_entry;
1349 struct net_device *dev = fib_nh->nh_dev;
1350 struct neighbour *n;
1353 /* Take a reference of neigh here ensuring that neigh would
1354 * not be detructed before the nexthop entry is finished.
1355 * The reference is taken either in neigh_lookup() or
1356 * in neith_create() in case n is not found.
1358 n = neigh_lookup(&arp_tbl, &fib_nh->nh_gw, dev);
1360 n = neigh_create(&arp_tbl, &fib_nh->nh_gw, dev);
1363 neigh_event_send(n, NULL);
1365 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
1371 /* If that is the first nexthop connected to that neigh, add to
1372 * nexthop_neighs_list
1374 if (list_empty(&neigh_entry->nexthop_list))
1375 list_add_tail(&neigh_entry->nexthop_neighs_list_node,
1376 &mlxsw_sp->router.nexthop_neighs_list);
1378 nh->nh_grp = nh_grp;
1379 nh->neigh_entry = neigh_entry;
1380 list_add_tail(&nh->neigh_list_node, &neigh_entry->nexthop_list);
1381 read_lock_bh(&n->lock);
1382 nud_state = n->nud_state;
1383 read_unlock_bh(&n->lock);
1384 __mlxsw_sp_nexthop_neigh_update(nh, !(nud_state & NUD_VALID));
1389 static void mlxsw_sp_nexthop_fini(struct mlxsw_sp *mlxsw_sp,
1390 struct mlxsw_sp_nexthop *nh)
1392 struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
1394 list_del(&nh->neigh_list_node);
1396 /* If that is the last nexthop connected to that neigh, remove from
1397 * nexthop_neighs_list
1399 if (list_empty(&nh->neigh_entry->nexthop_list))
1400 list_del(&nh->neigh_entry->nexthop_neighs_list_node);
1402 neigh_release(neigh_entry->key.n);
1405 static struct mlxsw_sp_nexthop_group *
1406 mlxsw_sp_nexthop_group_create(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
1408 struct mlxsw_sp_nexthop_group *nh_grp;
1409 struct mlxsw_sp_nexthop *nh;
1410 struct fib_nh *fib_nh;
1415 alloc_size = sizeof(*nh_grp) +
1416 fi->fib_nhs * sizeof(struct mlxsw_sp_nexthop);
1417 nh_grp = kzalloc(alloc_size, GFP_KERNEL);
1419 return ERR_PTR(-ENOMEM);
1420 INIT_LIST_HEAD(&nh_grp->fib_list);
1421 nh_grp->count = fi->fib_nhs;
1422 for (i = 0; i < nh_grp->count; i++) {
1423 nh = &nh_grp->nexthops[i];
1424 fib_nh = &fi->fib_nh[i];
1425 err = mlxsw_sp_nexthop_init(mlxsw_sp, nh_grp, nh, fib_nh);
1427 goto err_nexthop_init;
1429 list_add_tail(&nh_grp->list, &mlxsw_sp->router.nexthop_group_list);
1430 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
1434 for (i--; i >= 0; i--)
1435 mlxsw_sp_nexthop_fini(mlxsw_sp, nh);
1437 return ERR_PTR(err);
1441 mlxsw_sp_nexthop_group_destroy(struct mlxsw_sp *mlxsw_sp,
1442 struct mlxsw_sp_nexthop_group *nh_grp)
1444 struct mlxsw_sp_nexthop *nh;
1447 list_del(&nh_grp->list);
1448 for (i = 0; i < nh_grp->count; i++) {
1449 nh = &nh_grp->nexthops[i];
1450 mlxsw_sp_nexthop_fini(mlxsw_sp, nh);
1455 static bool mlxsw_sp_nexthop_match(struct mlxsw_sp_nexthop *nh,
1456 struct fib_info *fi)
1460 for (i = 0; i < fi->fib_nhs; i++) {
1461 struct fib_nh *fib_nh = &fi->fib_nh[i];
1462 struct neighbour *n = nh->neigh_entry->key.n;
1464 if (memcmp(n->primary_key, &fib_nh->nh_gw,
1465 sizeof(fib_nh->nh_gw)) == 0 &&
1466 n->dev == fib_nh->nh_dev)
1472 static bool mlxsw_sp_nexthop_group_match(struct mlxsw_sp_nexthop_group *nh_grp,
1473 struct fib_info *fi)
1477 if (nh_grp->count != fi->fib_nhs)
1479 for (i = 0; i < nh_grp->count; i++) {
1480 struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
1482 if (!mlxsw_sp_nexthop_match(nh, fi))
1488 static struct mlxsw_sp_nexthop_group *
1489 mlxsw_sp_nexthop_group_find(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
1491 struct mlxsw_sp_nexthop_group *nh_grp;
1493 list_for_each_entry(nh_grp, &mlxsw_sp->router.nexthop_group_list,
1495 if (mlxsw_sp_nexthop_group_match(nh_grp, fi))
1501 static int mlxsw_sp_nexthop_group_get(struct mlxsw_sp *mlxsw_sp,
1502 struct mlxsw_sp_fib_entry *fib_entry,
1503 struct fib_info *fi)
1505 struct mlxsw_sp_nexthop_group *nh_grp;
1507 nh_grp = mlxsw_sp_nexthop_group_find(mlxsw_sp, fi);
1509 nh_grp = mlxsw_sp_nexthop_group_create(mlxsw_sp, fi);
1511 return PTR_ERR(nh_grp);
1513 list_add_tail(&fib_entry->nexthop_group_node, &nh_grp->fib_list);
1514 fib_entry->nh_group = nh_grp;
1518 static void mlxsw_sp_nexthop_group_put(struct mlxsw_sp *mlxsw_sp,
1519 struct mlxsw_sp_fib_entry *fib_entry)
1521 struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
1523 list_del(&fib_entry->nexthop_group_node);
1524 if (!list_empty(&nh_grp->fib_list))
1526 mlxsw_sp_nexthop_group_destroy(mlxsw_sp, nh_grp);
1529 static int mlxsw_sp_fib_entry_op4_remote(struct mlxsw_sp *mlxsw_sp,
1530 struct mlxsw_sp_fib_entry *fib_entry,
1531 enum mlxsw_reg_ralue_op op)
1533 char ralue_pl[MLXSW_REG_RALUE_LEN];
1534 u32 *p_dip = (u32 *) fib_entry->key.addr;
1535 struct mlxsw_sp_vr *vr = fib_entry->vr;
1536 enum mlxsw_reg_ralue_trap_action trap_action;
1538 u32 adjacency_index = 0;
1541 /* In case the nexthop group adjacency index is valid, use it
1542 * with provided ECMP size. Otherwise, setup trap and pass
1543 * traffic to kernel.
1545 if (fib_entry->nh_group->adj_index_valid) {
1546 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
1547 adjacency_index = fib_entry->nh_group->adj_index;
1548 ecmp_size = fib_entry->nh_group->ecmp_size;
1550 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
1551 trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
1554 mlxsw_reg_ralue_pack4(ralue_pl,
1555 (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1556 vr->id, fib_entry->key.prefix_len, *p_dip);
1557 mlxsw_reg_ralue_act_remote_pack(ralue_pl, trap_action, trap_id,
1558 adjacency_index, ecmp_size);
1559 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1562 static int mlxsw_sp_fib_entry_op4_local(struct mlxsw_sp *mlxsw_sp,
1563 struct mlxsw_sp_fib_entry *fib_entry,
1564 enum mlxsw_reg_ralue_op op)
1566 char ralue_pl[MLXSW_REG_RALUE_LEN];
1567 u32 *p_dip = (u32 *) fib_entry->key.addr;
1568 struct mlxsw_sp_vr *vr = fib_entry->vr;
1570 mlxsw_reg_ralue_pack4(ralue_pl,
1571 (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1572 vr->id, fib_entry->key.prefix_len, *p_dip);
1573 mlxsw_reg_ralue_act_local_pack(ralue_pl,
1574 MLXSW_REG_RALUE_TRAP_ACTION_NOP, 0,
1576 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1579 static int mlxsw_sp_fib_entry_op4_trap(struct mlxsw_sp *mlxsw_sp,
1580 struct mlxsw_sp_fib_entry *fib_entry,
1581 enum mlxsw_reg_ralue_op op)
1583 char ralue_pl[MLXSW_REG_RALUE_LEN];
1584 u32 *p_dip = (u32 *) fib_entry->key.addr;
1585 struct mlxsw_sp_vr *vr = fib_entry->vr;
1587 mlxsw_reg_ralue_pack4(ralue_pl,
1588 (enum mlxsw_reg_ralxx_protocol) vr->proto, op,
1589 vr->id, fib_entry->key.prefix_len, *p_dip);
1590 mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
1591 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1594 static int mlxsw_sp_fib_entry_op4(struct mlxsw_sp *mlxsw_sp,
1595 struct mlxsw_sp_fib_entry *fib_entry,
1596 enum mlxsw_reg_ralue_op op)
1598 switch (fib_entry->type) {
1599 case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
1600 return mlxsw_sp_fib_entry_op4_remote(mlxsw_sp, fib_entry, op);
1601 case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
1602 return mlxsw_sp_fib_entry_op4_local(mlxsw_sp, fib_entry, op);
1603 case MLXSW_SP_FIB_ENTRY_TYPE_TRAP:
1604 return mlxsw_sp_fib_entry_op4_trap(mlxsw_sp, fib_entry, op);
1609 static int mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
1610 struct mlxsw_sp_fib_entry *fib_entry,
1611 enum mlxsw_reg_ralue_op op)
1613 switch (fib_entry->vr->proto) {
1614 case MLXSW_SP_L3_PROTO_IPV4:
1615 return mlxsw_sp_fib_entry_op4(mlxsw_sp, fib_entry, op);
1616 case MLXSW_SP_L3_PROTO_IPV6:
1622 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1623 struct mlxsw_sp_fib_entry *fib_entry)
1625 return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
1626 MLXSW_REG_RALUE_OP_WRITE_WRITE);
1629 static int mlxsw_sp_fib_entry_del(struct mlxsw_sp *mlxsw_sp,
1630 struct mlxsw_sp_fib_entry *fib_entry)
1632 return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
1633 MLXSW_REG_RALUE_OP_WRITE_DELETE);
1637 mlxsw_sp_router_fib4_entry_init(struct mlxsw_sp *mlxsw_sp,
1638 const struct fib_entry_notifier_info *fen_info,
1639 struct mlxsw_sp_fib_entry *fib_entry)
1641 struct fib_info *fi = fen_info->fi;
1642 struct mlxsw_sp_rif *r = NULL;
1646 if (fen_info->type == RTN_LOCAL || fen_info->type == RTN_BROADCAST) {
1647 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1650 if (fen_info->type != RTN_UNICAST)
1653 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1654 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1658 r = mlxsw_sp_rif_find_by_dev(mlxsw_sp, nh->nh_dev);
1660 /* In case router interface is not found for
1661 * at least one of the nexthops, that means
1662 * the nexthop points to some device unrelated
1663 * to us. Set trap and pass the packets for
1664 * this prefix to kernel.
1671 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1675 if (fi->fib_scope != RT_SCOPE_UNIVERSE) {
1676 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
1677 fib_entry->rif = r->rif;
1679 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
1680 err = mlxsw_sp_nexthop_group_get(mlxsw_sp, fib_entry, fi);
1684 fib_info_offload_inc(fen_info->fi);
1689 mlxsw_sp_router_fib4_entry_fini(struct mlxsw_sp *mlxsw_sp,
1690 struct mlxsw_sp_fib_entry *fib_entry)
1692 if (fib_entry->type != MLXSW_SP_FIB_ENTRY_TYPE_TRAP)
1693 fib_info_offload_dec(fib_entry->fi);
1694 if (fib_entry->type == MLXSW_SP_FIB_ENTRY_TYPE_REMOTE)
1695 mlxsw_sp_nexthop_group_put(mlxsw_sp, fib_entry);
1698 static struct mlxsw_sp_fib_entry *
1699 mlxsw_sp_fib_entry_get(struct mlxsw_sp *mlxsw_sp,
1700 const struct fib_entry_notifier_info *fen_info)
1702 struct mlxsw_sp_fib_entry *fib_entry;
1703 struct fib_info *fi = fen_info->fi;
1704 struct mlxsw_sp_vr *vr;
1707 vr = mlxsw_sp_vr_get(mlxsw_sp, fen_info->dst_len, fen_info->tb_id,
1708 MLXSW_SP_L3_PROTO_IPV4);
1710 return ERR_CAST(vr);
1712 fib_entry = mlxsw_sp_fib_entry_lookup(vr->fib, &fen_info->dst,
1713 sizeof(fen_info->dst),
1714 fen_info->dst_len, fi->fib_dev);
1716 /* Already exists, just take a reference */
1717 fib_entry->ref_count++;
1720 fib_entry = mlxsw_sp_fib_entry_create(vr->fib, &fen_info->dst,
1721 sizeof(fen_info->dst),
1722 fen_info->dst_len, fi->fib_dev);
1725 goto err_fib_entry_create;
1729 fib_entry->ref_count = 1;
1731 err = mlxsw_sp_router_fib4_entry_init(mlxsw_sp, fen_info, fib_entry);
1733 goto err_fib4_entry_init;
1737 err_fib4_entry_init:
1738 mlxsw_sp_fib_entry_destroy(fib_entry);
1739 err_fib_entry_create:
1740 mlxsw_sp_vr_put(mlxsw_sp, vr);
1742 return ERR_PTR(err);
1745 static struct mlxsw_sp_fib_entry *
1746 mlxsw_sp_fib_entry_find(struct mlxsw_sp *mlxsw_sp,
1747 const struct fib_entry_notifier_info *fen_info)
1749 struct mlxsw_sp_vr *vr;
1751 vr = mlxsw_sp_vr_find(mlxsw_sp, fen_info->tb_id,
1752 MLXSW_SP_L3_PROTO_IPV4);
1756 return mlxsw_sp_fib_entry_lookup(vr->fib, &fen_info->dst,
1757 sizeof(fen_info->dst),
1759 fen_info->fi->fib_dev);
1762 static void mlxsw_sp_fib_entry_put(struct mlxsw_sp *mlxsw_sp,
1763 struct mlxsw_sp_fib_entry *fib_entry)
1765 struct mlxsw_sp_vr *vr = fib_entry->vr;
1767 if (--fib_entry->ref_count == 0) {
1768 mlxsw_sp_router_fib4_entry_fini(mlxsw_sp, fib_entry);
1769 mlxsw_sp_fib_entry_destroy(fib_entry);
1771 mlxsw_sp_vr_put(mlxsw_sp, vr);
1774 static void mlxsw_sp_fib_entry_put_all(struct mlxsw_sp *mlxsw_sp,
1775 struct mlxsw_sp_fib_entry *fib_entry)
1777 unsigned int last_ref_count;
1780 last_ref_count = fib_entry->ref_count;
1781 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1782 } while (last_ref_count != 1);
1785 static int mlxsw_sp_router_fib4_add(struct mlxsw_sp *mlxsw_sp,
1786 struct fib_entry_notifier_info *fen_info)
1788 struct mlxsw_sp_fib_entry *fib_entry;
1789 struct mlxsw_sp_vr *vr;
1792 if (mlxsw_sp->router.aborted)
1795 fib_entry = mlxsw_sp_fib_entry_get(mlxsw_sp, fen_info);
1796 if (IS_ERR(fib_entry)) {
1797 dev_warn(mlxsw_sp->bus_info->dev, "Failed to get FIB4 entry being added.\n");
1798 return PTR_ERR(fib_entry);
1801 if (fib_entry->ref_count != 1)
1805 err = mlxsw_sp_fib_entry_insert(vr->fib, fib_entry);
1807 dev_warn(mlxsw_sp->bus_info->dev, "Failed to insert FIB4 entry being added.\n");
1808 goto err_fib_entry_insert;
1810 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1812 goto err_fib_entry_add;
1816 mlxsw_sp_fib_entry_remove(vr->fib, fib_entry);
1817 err_fib_entry_insert:
1818 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1822 static void mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
1823 struct fib_entry_notifier_info *fen_info)
1825 struct mlxsw_sp_fib_entry *fib_entry;
1827 if (mlxsw_sp->router.aborted)
1830 fib_entry = mlxsw_sp_fib_entry_find(mlxsw_sp, fen_info);
1834 if (fib_entry->ref_count == 1) {
1835 mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
1836 mlxsw_sp_fib_entry_remove(fib_entry->vr->fib, fib_entry);
1839 mlxsw_sp_fib_entry_put(mlxsw_sp, fib_entry);
1842 static int mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp)
1844 char ralta_pl[MLXSW_REG_RALTA_LEN];
1845 char ralst_pl[MLXSW_REG_RALST_LEN];
1846 char raltb_pl[MLXSW_REG_RALTB_LEN];
1847 char ralue_pl[MLXSW_REG_RALUE_LEN];
1850 mlxsw_reg_ralta_pack(ralta_pl, true, MLXSW_REG_RALXX_PROTOCOL_IPV4,
1851 MLXSW_SP_LPM_TREE_MIN);
1852 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
1856 mlxsw_reg_ralst_pack(ralst_pl, 0xff, MLXSW_SP_LPM_TREE_MIN);
1857 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
1861 mlxsw_reg_raltb_pack(raltb_pl, 0, MLXSW_REG_RALXX_PROTOCOL_IPV4,
1862 MLXSW_SP_LPM_TREE_MIN);
1863 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
1867 mlxsw_reg_ralue_pack4(ralue_pl, MLXSW_SP_L3_PROTO_IPV4,
1868 MLXSW_REG_RALUE_OP_WRITE_WRITE, 0, 0, 0);
1869 mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
1870 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
1873 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp)
1875 struct mlxsw_resources *resources;
1876 struct mlxsw_sp_fib_entry *fib_entry;
1877 struct mlxsw_sp_fib_entry *tmp;
1878 struct mlxsw_sp_vr *vr;
1881 resources = mlxsw_core_resources_get(mlxsw_sp->core);
1882 for (i = 0; i < resources->max_virtual_routers; i++) {
1883 vr = &mlxsw_sp->router.vrs[i];
1888 list_for_each_entry_safe(fib_entry, tmp,
1889 &vr->fib->entry_list, list) {
1890 bool do_break = &tmp->list == &vr->fib->entry_list;
1892 mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
1893 mlxsw_sp_fib_entry_remove(fib_entry->vr->fib,
1895 mlxsw_sp_fib_entry_put_all(mlxsw_sp, fib_entry);
1902 static void mlxsw_sp_router_fib4_abort(struct mlxsw_sp *mlxsw_sp)
1906 mlxsw_sp_router_fib_flush(mlxsw_sp);
1907 mlxsw_sp->router.aborted = true;
1908 err = mlxsw_sp_router_set_abort_trap(mlxsw_sp);
1910 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set abort trap.\n");
1913 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
1915 struct mlxsw_resources *resources;
1916 char rgcr_pl[MLXSW_REG_RGCR_LEN];
1919 resources = mlxsw_core_resources_get(mlxsw_sp->core);
1920 if (!resources->max_rif_valid)
1923 mlxsw_sp->rifs = kcalloc(resources->max_rif,
1924 sizeof(struct mlxsw_sp_rif *), GFP_KERNEL);
1925 if (!mlxsw_sp->rifs)
1928 mlxsw_reg_rgcr_pack(rgcr_pl, true);
1929 mlxsw_reg_rgcr_max_router_interfaces_set(rgcr_pl, resources->max_rif);
1930 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
1937 kfree(mlxsw_sp->rifs);
1941 static void __mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
1943 struct mlxsw_resources *resources;
1944 char rgcr_pl[MLXSW_REG_RGCR_LEN];
1947 mlxsw_reg_rgcr_pack(rgcr_pl, false);
1948 mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
1950 resources = mlxsw_core_resources_get(mlxsw_sp->core);
1951 for (i = 0; i < resources->max_rif; i++)
1952 WARN_ON_ONCE(mlxsw_sp->rifs[i]);
1954 kfree(mlxsw_sp->rifs);
1957 static int mlxsw_sp_router_fib_event(struct notifier_block *nb,
1958 unsigned long event, void *ptr)
1960 struct mlxsw_sp *mlxsw_sp = container_of(nb, struct mlxsw_sp, fib_nb);
1961 struct fib_entry_notifier_info *fen_info = ptr;
1964 if (!net_eq(fen_info->info.net, &init_net))
1968 case FIB_EVENT_ENTRY_ADD:
1969 err = mlxsw_sp_router_fib4_add(mlxsw_sp, fen_info);
1971 mlxsw_sp_router_fib4_abort(mlxsw_sp);
1973 case FIB_EVENT_ENTRY_DEL:
1974 mlxsw_sp_router_fib4_del(mlxsw_sp, fen_info);
1976 case FIB_EVENT_RULE_ADD: /* fall through */
1977 case FIB_EVENT_RULE_DEL:
1978 mlxsw_sp_router_fib4_abort(mlxsw_sp);
1984 int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
1988 INIT_LIST_HEAD(&mlxsw_sp->router.nexthop_neighs_list);
1989 INIT_LIST_HEAD(&mlxsw_sp->router.nexthop_group_list);
1990 err = __mlxsw_sp_router_init(mlxsw_sp);
1994 mlxsw_sp_lpm_init(mlxsw_sp);
1995 err = mlxsw_sp_vrs_init(mlxsw_sp);
1999 err = mlxsw_sp_neigh_init(mlxsw_sp);
2001 goto err_neigh_init;
2003 mlxsw_sp->fib_nb.notifier_call = mlxsw_sp_router_fib_event;
2004 register_fib_notifier(&mlxsw_sp->fib_nb);
2008 mlxsw_sp_vrs_fini(mlxsw_sp);
2010 __mlxsw_sp_router_fini(mlxsw_sp);
2014 void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
2016 unregister_fib_notifier(&mlxsw_sp->fib_nb);
2017 mlxsw_sp_neigh_fini(mlxsw_sp);
2018 mlxsw_sp_vrs_fini(mlxsw_sp);
2019 __mlxsw_sp_router_fini(mlxsw_sp);