1 /*
2  * drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
3  * Copyright (c) 2016-2017 Mellanox Technologies. All rights reserved.
4  * Copyright (c) 2016 Jiri Pirko <jiri@mellanox.com>
5  * Copyright (c) 2016 Ido Schimmel <idosch@mellanox.com>
6  * Copyright (c) 2016 Yotam Gigi <yotamg@mellanox.com>
7  * Copyright (c) 2017 Petr Machata <petrm@mellanox.com>
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions are met:
11  *
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the names of the copyright holders nor the names of its
18  *    contributors may be used to endorse or promote products derived from
19  *    this software without specific prior written permission.
20  *
21  * Alternatively, this software may be distributed under the terms of the
22  * GNU General Public License ("GPL") version 2 as published by the Free
23  * Software Foundation.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
26  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
29  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGE.
36  */
37 
38 #include <linux/kernel.h>
39 #include <linux/types.h>
40 #include <linux/rhashtable.h>
41 #include <linux/bitops.h>
42 #include <linux/in6.h>
43 #include <linux/notifier.h>
44 #include <linux/inetdevice.h>
45 #include <linux/netdevice.h>
46 #include <linux/if_bridge.h>
47 #include <linux/socket.h>
48 #include <linux/route.h>
49 #include <linux/gcd.h>
50 #include <linux/random.h>
51 #include <net/netevent.h>
52 #include <net/neighbour.h>
53 #include <net/arp.h>
54 #include <net/ip_fib.h>
55 #include <net/ip6_fib.h>
56 #include <net/fib_rules.h>
57 #include <net/ip_tunnels.h>
58 #include <net/l3mdev.h>
59 #include <net/addrconf.h>
60 #include <net/ndisc.h>
61 #include <net/ipv6.h>
62 #include <net/fib_notifier.h>
63 
64 #include "spectrum.h"
65 #include "core.h"
66 #include "reg.h"
67 #include "spectrum_cnt.h"
68 #include "spectrum_dpipe.h"
69 #include "spectrum_ipip.h"
70 #include "spectrum_mr.h"
71 #include "spectrum_mr_tcam.h"
72 #include "spectrum_router.h"
73 
74 struct mlxsw_sp_vr;
75 struct mlxsw_sp_lpm_tree;
76 struct mlxsw_sp_rif_ops;
77 
78 struct mlxsw_sp_router {
79 	struct mlxsw_sp *mlxsw_sp;
80 	struct mlxsw_sp_rif **rifs;
81 	struct mlxsw_sp_vr *vrs;
82 	struct rhashtable neigh_ht;
83 	struct rhashtable nexthop_group_ht;
84 	struct rhashtable nexthop_ht;
85 	struct list_head nexthop_list;
86 	struct {
87 		struct mlxsw_sp_lpm_tree *trees;
88 		unsigned int tree_count;
89 	} lpm;
90 	struct {
91 		struct delayed_work dw;
92 		unsigned long interval;	/* ms */
93 	} neighs_update;
94 	struct delayed_work nexthop_probe_dw;
95 #define MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL 5000 /* ms */
96 	struct list_head nexthop_neighs_list;
97 	struct list_head ipip_list;
98 	bool aborted;
99 	struct notifier_block fib_nb;
100 	struct notifier_block netevent_nb;
101 	const struct mlxsw_sp_rif_ops **rif_ops_arr;
102 	const struct mlxsw_sp_ipip_ops **ipip_ops_arr;
103 };
104 
105 struct mlxsw_sp_rif {
106 	struct list_head nexthop_list;
107 	struct list_head neigh_list;
108 	struct net_device *dev;
109 	struct mlxsw_sp_fid *fid;
110 	unsigned char addr[ETH_ALEN];
111 	int mtu;
112 	u16 rif_index;
113 	u16 vr_id;
114 	const struct mlxsw_sp_rif_ops *ops;
115 	struct mlxsw_sp *mlxsw_sp;
116 
117 	unsigned int counter_ingress;
118 	bool counter_ingress_valid;
119 	unsigned int counter_egress;
120 	bool counter_egress_valid;
121 };
122 
123 struct mlxsw_sp_rif_params {
124 	struct net_device *dev;
125 	union {
126 		u16 system_port;
127 		u16 lag_id;
128 	};
129 	u16 vid;
130 	bool lag;
131 };
132 
133 struct mlxsw_sp_rif_subport {
134 	struct mlxsw_sp_rif common;
135 	union {
136 		u16 system_port;
137 		u16 lag_id;
138 	};
139 	u16 vid;
140 	bool lag;
141 };
142 
143 struct mlxsw_sp_rif_ipip_lb {
144 	struct mlxsw_sp_rif common;
145 	struct mlxsw_sp_rif_ipip_lb_config lb_config;
146 	u16 ul_vr_id; /* Reserved for Spectrum-2. */
147 };
148 
149 struct mlxsw_sp_rif_params_ipip_lb {
150 	struct mlxsw_sp_rif_params common;
151 	struct mlxsw_sp_rif_ipip_lb_config lb_config;
152 };
153 
154 struct mlxsw_sp_rif_ops {
155 	enum mlxsw_sp_rif_type type;
156 	size_t rif_size;
157 
158 	void (*setup)(struct mlxsw_sp_rif *rif,
159 		      const struct mlxsw_sp_rif_params *params);
160 	int (*configure)(struct mlxsw_sp_rif *rif);
161 	void (*deconfigure)(struct mlxsw_sp_rif *rif);
162 	struct mlxsw_sp_fid * (*fid_get)(struct mlxsw_sp_rif *rif);
163 };
164 
165 static unsigned int *
166 mlxsw_sp_rif_p_counter_get(struct mlxsw_sp_rif *rif,
167 			   enum mlxsw_sp_rif_counter_dir dir)
168 {
169 	switch (dir) {
170 	case MLXSW_SP_RIF_COUNTER_EGRESS:
171 		return &rif->counter_egress;
172 	case MLXSW_SP_RIF_COUNTER_INGRESS:
173 		return &rif->counter_ingress;
174 	}
175 	return NULL;
176 }
177 
178 static bool
179 mlxsw_sp_rif_counter_valid_get(struct mlxsw_sp_rif *rif,
180 			       enum mlxsw_sp_rif_counter_dir dir)
181 {
182 	switch (dir) {
183 	case MLXSW_SP_RIF_COUNTER_EGRESS:
184 		return rif->counter_egress_valid;
185 	case MLXSW_SP_RIF_COUNTER_INGRESS:
186 		return rif->counter_ingress_valid;
187 	}
188 	return false;
189 }
190 
191 static void
192 mlxsw_sp_rif_counter_valid_set(struct mlxsw_sp_rif *rif,
193 			       enum mlxsw_sp_rif_counter_dir dir,
194 			       bool valid)
195 {
196 	switch (dir) {
197 	case MLXSW_SP_RIF_COUNTER_EGRESS:
198 		rif->counter_egress_valid = valid;
199 		break;
200 	case MLXSW_SP_RIF_COUNTER_INGRESS:
201 		rif->counter_ingress_valid = valid;
202 		break;
203 	}
204 }
205 
206 static int mlxsw_sp_rif_counter_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
207 				     unsigned int counter_index, bool enable,
208 				     enum mlxsw_sp_rif_counter_dir dir)
209 {
210 	char ritr_pl[MLXSW_REG_RITR_LEN];
211 	bool is_egress = false;
212 	int err;
213 
214 	if (dir == MLXSW_SP_RIF_COUNTER_EGRESS)
215 		is_egress = true;
216 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
217 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
218 	if (err)
219 		return err;
220 
221 	mlxsw_reg_ritr_counter_pack(ritr_pl, counter_index, enable,
222 				    is_egress);
223 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
224 }
225 
226 int mlxsw_sp_rif_counter_value_get(struct mlxsw_sp *mlxsw_sp,
227 				   struct mlxsw_sp_rif *rif,
228 				   enum mlxsw_sp_rif_counter_dir dir, u64 *cnt)
229 {
230 	char ricnt_pl[MLXSW_REG_RICNT_LEN];
231 	unsigned int *p_counter_index;
232 	bool valid;
233 	int err;
234 
235 	valid = mlxsw_sp_rif_counter_valid_get(rif, dir);
236 	if (!valid)
237 		return -EINVAL;
238 
239 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
240 	if (!p_counter_index)
241 		return -EINVAL;
242 	mlxsw_reg_ricnt_pack(ricnt_pl, *p_counter_index,
243 			     MLXSW_REG_RICNT_OPCODE_NOP);
244 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
245 	if (err)
246 		return err;
247 	*cnt = mlxsw_reg_ricnt_good_unicast_packets_get(ricnt_pl);
248 	return 0;
249 }
250 
251 static int mlxsw_sp_rif_counter_clear(struct mlxsw_sp *mlxsw_sp,
252 				      unsigned int counter_index)
253 {
254 	char ricnt_pl[MLXSW_REG_RICNT_LEN];
255 
256 	mlxsw_reg_ricnt_pack(ricnt_pl, counter_index,
257 			     MLXSW_REG_RICNT_OPCODE_CLEAR);
258 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
259 }
260 
261 int mlxsw_sp_rif_counter_alloc(struct mlxsw_sp *mlxsw_sp,
262 			       struct mlxsw_sp_rif *rif,
263 			       enum mlxsw_sp_rif_counter_dir dir)
264 {
265 	unsigned int *p_counter_index;
266 	int err;
267 
268 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
269 	if (!p_counter_index)
270 		return -EINVAL;
271 	err = mlxsw_sp_counter_alloc(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
272 				     p_counter_index);
273 	if (err)
274 		return err;
275 
276 	err = mlxsw_sp_rif_counter_clear(mlxsw_sp, *p_counter_index);
277 	if (err)
278 		goto err_counter_clear;
279 
280 	err = mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
281 					*p_counter_index, true, dir);
282 	if (err)
283 		goto err_counter_edit;
284 	mlxsw_sp_rif_counter_valid_set(rif, dir, true);
285 	return 0;
286 
287 err_counter_edit:
288 err_counter_clear:
289 	mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
290 			      *p_counter_index);
291 	return err;
292 }
293 
294 void mlxsw_sp_rif_counter_free(struct mlxsw_sp *mlxsw_sp,
295 			       struct mlxsw_sp_rif *rif,
296 			       enum mlxsw_sp_rif_counter_dir dir)
297 {
298 	unsigned int *p_counter_index;
299 
300 	if (!mlxsw_sp_rif_counter_valid_get(rif, dir))
301 		return;
302 
303 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
304 	if (WARN_ON(!p_counter_index))
305 		return;
306 	mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
307 				  *p_counter_index, false, dir);
308 	mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
309 			      *p_counter_index);
310 	mlxsw_sp_rif_counter_valid_set(rif, dir, false);
311 }
312 
313 static void mlxsw_sp_rif_counters_alloc(struct mlxsw_sp_rif *rif)
314 {
315 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
316 	struct devlink *devlink;
317 
318 	devlink = priv_to_devlink(mlxsw_sp->core);
319 	if (!devlink_dpipe_table_counter_enabled(devlink,
320 						 MLXSW_SP_DPIPE_TABLE_NAME_ERIF))
321 		return;
322 	mlxsw_sp_rif_counter_alloc(mlxsw_sp, rif, MLXSW_SP_RIF_COUNTER_EGRESS);
323 }
324 
325 static void mlxsw_sp_rif_counters_free(struct mlxsw_sp_rif *rif)
326 {
327 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
328 
329 	mlxsw_sp_rif_counter_free(mlxsw_sp, rif, MLXSW_SP_RIF_COUNTER_EGRESS);
330 }
331 
332 static struct mlxsw_sp_rif *
333 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
334 			 const struct net_device *dev);
335 
336 #define MLXSW_SP_PREFIX_COUNT (sizeof(struct in6_addr) * BITS_PER_BYTE + 1)
337 
338 struct mlxsw_sp_prefix_usage {
339 	DECLARE_BITMAP(b, MLXSW_SP_PREFIX_COUNT);
340 };
341 
342 #define mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) \
343 	for_each_set_bit(prefix, (prefix_usage)->b, MLXSW_SP_PREFIX_COUNT)
344 
345 static bool
346 mlxsw_sp_prefix_usage_eq(struct mlxsw_sp_prefix_usage *prefix_usage1,
347 			 struct mlxsw_sp_prefix_usage *prefix_usage2)
348 {
349 	return !memcmp(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
350 }
351 
352 static bool
353 mlxsw_sp_prefix_usage_none(struct mlxsw_sp_prefix_usage *prefix_usage)
354 {
355 	struct mlxsw_sp_prefix_usage prefix_usage_none = {{ 0 } };
356 
357 	return mlxsw_sp_prefix_usage_eq(prefix_usage, &prefix_usage_none);
358 }
359 
360 static void
361 mlxsw_sp_prefix_usage_cpy(struct mlxsw_sp_prefix_usage *prefix_usage1,
362 			  struct mlxsw_sp_prefix_usage *prefix_usage2)
363 {
364 	memcpy(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
365 }
366 
367 static void
368 mlxsw_sp_prefix_usage_set(struct mlxsw_sp_prefix_usage *prefix_usage,
369 			  unsigned char prefix_len)
370 {
371 	set_bit(prefix_len, prefix_usage->b);
372 }
373 
374 static void
375 mlxsw_sp_prefix_usage_clear(struct mlxsw_sp_prefix_usage *prefix_usage,
376 			    unsigned char prefix_len)
377 {
378 	clear_bit(prefix_len, prefix_usage->b);
379 }
380 
381 struct mlxsw_sp_fib_key {
382 	unsigned char addr[sizeof(struct in6_addr)];
383 	unsigned char prefix_len;
384 };
385 
386 enum mlxsw_sp_fib_entry_type {
387 	MLXSW_SP_FIB_ENTRY_TYPE_REMOTE,
388 	MLXSW_SP_FIB_ENTRY_TYPE_LOCAL,
389 	MLXSW_SP_FIB_ENTRY_TYPE_TRAP,
390 
391 	/* This is a special case of local delivery, where a packet should be
392 	 * decapsulated on reception. Note that there is no corresponding ENCAP,
393 	 * because that's a type of next hop, not of FIB entry. (There can be
394 	 * several next hops in a REMOTE entry, and some of them may be
395 	 * encapsulating entries.)
396 	 */
397 	MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP,
398 };
399 
400 struct mlxsw_sp_nexthop_group;
401 struct mlxsw_sp_fib;
402 
403 struct mlxsw_sp_fib_node {
404 	struct list_head entry_list;
405 	struct list_head list;
406 	struct rhash_head ht_node;
407 	struct mlxsw_sp_fib *fib;
408 	struct mlxsw_sp_fib_key key;
409 };
410 
411 struct mlxsw_sp_fib_entry_decap {
412 	struct mlxsw_sp_ipip_entry *ipip_entry;
413 	u32 tunnel_index;
414 };
415 
416 struct mlxsw_sp_fib_entry {
417 	struct list_head list;
418 	struct mlxsw_sp_fib_node *fib_node;
419 	enum mlxsw_sp_fib_entry_type type;
420 	struct list_head nexthop_group_node;
421 	struct mlxsw_sp_nexthop_group *nh_group;
422 	struct mlxsw_sp_fib_entry_decap decap; /* Valid for decap entries. */
423 };
424 
425 struct mlxsw_sp_fib4_entry {
426 	struct mlxsw_sp_fib_entry common;
427 	u32 tb_id;
428 	u32 prio;
429 	u8 tos;
430 	u8 type;
431 };
432 
433 struct mlxsw_sp_fib6_entry {
434 	struct mlxsw_sp_fib_entry common;
435 	struct list_head rt6_list;
436 	unsigned int nrt6;
437 };
438 
439 struct mlxsw_sp_rt6 {
440 	struct list_head list;
441 	struct rt6_info *rt;
442 };
443 
444 struct mlxsw_sp_lpm_tree {
445 	u8 id; /* tree ID */
446 	unsigned int ref_count;
447 	enum mlxsw_sp_l3proto proto;
448 	struct mlxsw_sp_prefix_usage prefix_usage;
449 };
450 
451 struct mlxsw_sp_fib {
452 	struct rhashtable ht;
453 	struct list_head node_list;
454 	struct mlxsw_sp_vr *vr;
455 	struct mlxsw_sp_lpm_tree *lpm_tree;
456 	unsigned long prefix_ref_count[MLXSW_SP_PREFIX_COUNT];
457 	struct mlxsw_sp_prefix_usage prefix_usage;
458 	enum mlxsw_sp_l3proto proto;
459 };
460 
461 struct mlxsw_sp_vr {
462 	u16 id; /* virtual router ID */
463 	u32 tb_id; /* kernel fib table id */
464 	unsigned int rif_count;
465 	struct mlxsw_sp_fib *fib4;
466 	struct mlxsw_sp_fib *fib6;
467 	struct mlxsw_sp_mr_table *mr4_table;
468 };
469 
470 static const struct rhashtable_params mlxsw_sp_fib_ht_params;
471 
472 static struct mlxsw_sp_fib *mlxsw_sp_fib_create(struct mlxsw_sp_vr *vr,
473 						enum mlxsw_sp_l3proto proto)
474 {
475 	struct mlxsw_sp_fib *fib;
476 	int err;
477 
478 	fib = kzalloc(sizeof(*fib), GFP_KERNEL);
479 	if (!fib)
480 		return ERR_PTR(-ENOMEM);
481 	err = rhashtable_init(&fib->ht, &mlxsw_sp_fib_ht_params);
482 	if (err)
483 		goto err_rhashtable_init;
484 	INIT_LIST_HEAD(&fib->node_list);
485 	fib->proto = proto;
486 	fib->vr = vr;
487 	return fib;
488 
489 err_rhashtable_init:
490 	kfree(fib);
491 	return ERR_PTR(err);
492 }
493 
494 static void mlxsw_sp_fib_destroy(struct mlxsw_sp_fib *fib)
495 {
496 	WARN_ON(!list_empty(&fib->node_list));
497 	WARN_ON(fib->lpm_tree);
498 	rhashtable_destroy(&fib->ht);
499 	kfree(fib);
500 }
501 
502 static struct mlxsw_sp_lpm_tree *
503 mlxsw_sp_lpm_tree_find_unused(struct mlxsw_sp *mlxsw_sp)
504 {
505 	static struct mlxsw_sp_lpm_tree *lpm_tree;
506 	int i;
507 
508 	for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
509 		lpm_tree = &mlxsw_sp->router->lpm.trees[i];
510 		if (lpm_tree->ref_count == 0)
511 			return lpm_tree;
512 	}
513 	return NULL;
514 }
515 
516 static int mlxsw_sp_lpm_tree_alloc(struct mlxsw_sp *mlxsw_sp,
517 				   struct mlxsw_sp_lpm_tree *lpm_tree)
518 {
519 	char ralta_pl[MLXSW_REG_RALTA_LEN];
520 
521 	mlxsw_reg_ralta_pack(ralta_pl, true,
522 			     (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
523 			     lpm_tree->id);
524 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
525 }
526 
527 static void mlxsw_sp_lpm_tree_free(struct mlxsw_sp *mlxsw_sp,
528 				   struct mlxsw_sp_lpm_tree *lpm_tree)
529 {
530 	char ralta_pl[MLXSW_REG_RALTA_LEN];
531 
532 	mlxsw_reg_ralta_pack(ralta_pl, false,
533 			     (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
534 			     lpm_tree->id);
535 	mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
536 }
537 
538 static int
539 mlxsw_sp_lpm_tree_left_struct_set(struct mlxsw_sp *mlxsw_sp,
540 				  struct mlxsw_sp_prefix_usage *prefix_usage,
541 				  struct mlxsw_sp_lpm_tree *lpm_tree)
542 {
543 	char ralst_pl[MLXSW_REG_RALST_LEN];
544 	u8 root_bin = 0;
545 	u8 prefix;
546 	u8 last_prefix = MLXSW_REG_RALST_BIN_NO_CHILD;
547 
548 	mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage)
549 		root_bin = prefix;
550 
551 	mlxsw_reg_ralst_pack(ralst_pl, root_bin, lpm_tree->id);
552 	mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) {
553 		if (prefix == 0)
554 			continue;
555 		mlxsw_reg_ralst_bin_pack(ralst_pl, prefix, last_prefix,
556 					 MLXSW_REG_RALST_BIN_NO_CHILD);
557 		last_prefix = prefix;
558 	}
559 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
560 }
561 
562 static struct mlxsw_sp_lpm_tree *
563 mlxsw_sp_lpm_tree_create(struct mlxsw_sp *mlxsw_sp,
564 			 struct mlxsw_sp_prefix_usage *prefix_usage,
565 			 enum mlxsw_sp_l3proto proto)
566 {
567 	struct mlxsw_sp_lpm_tree *lpm_tree;
568 	int err;
569 
570 	lpm_tree = mlxsw_sp_lpm_tree_find_unused(mlxsw_sp);
571 	if (!lpm_tree)
572 		return ERR_PTR(-EBUSY);
573 	lpm_tree->proto = proto;
574 	err = mlxsw_sp_lpm_tree_alloc(mlxsw_sp, lpm_tree);
575 	if (err)
576 		return ERR_PTR(err);
577 
578 	err = mlxsw_sp_lpm_tree_left_struct_set(mlxsw_sp, prefix_usage,
579 						lpm_tree);
580 	if (err)
581 		goto err_left_struct_set;
582 	memcpy(&lpm_tree->prefix_usage, prefix_usage,
583 	       sizeof(lpm_tree->prefix_usage));
584 	return lpm_tree;
585 
586 err_left_struct_set:
587 	mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
588 	return ERR_PTR(err);
589 }
590 
591 static void mlxsw_sp_lpm_tree_destroy(struct mlxsw_sp *mlxsw_sp,
592 				      struct mlxsw_sp_lpm_tree *lpm_tree)
593 {
594 	mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
595 }
596 
597 static struct mlxsw_sp_lpm_tree *
598 mlxsw_sp_lpm_tree_get(struct mlxsw_sp *mlxsw_sp,
599 		      struct mlxsw_sp_prefix_usage *prefix_usage,
600 		      enum mlxsw_sp_l3proto proto)
601 {
602 	struct mlxsw_sp_lpm_tree *lpm_tree;
603 	int i;
604 
605 	for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
606 		lpm_tree = &mlxsw_sp->router->lpm.trees[i];
607 		if (lpm_tree->ref_count != 0 &&
608 		    lpm_tree->proto == proto &&
609 		    mlxsw_sp_prefix_usage_eq(&lpm_tree->prefix_usage,
610 					     prefix_usage))
611 			return lpm_tree;
612 	}
613 	return mlxsw_sp_lpm_tree_create(mlxsw_sp, prefix_usage, proto);
614 }
615 
616 static void mlxsw_sp_lpm_tree_hold(struct mlxsw_sp_lpm_tree *lpm_tree)
617 {
618 	lpm_tree->ref_count++;
619 }
620 
621 static void mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
622 				  struct mlxsw_sp_lpm_tree *lpm_tree)
623 {
624 	if (--lpm_tree->ref_count == 0)
625 		mlxsw_sp_lpm_tree_destroy(mlxsw_sp, lpm_tree);
626 }
627 
628 #define MLXSW_SP_LPM_TREE_MIN 1 /* tree 0 is reserved */
629 
630 static int mlxsw_sp_lpm_init(struct mlxsw_sp *mlxsw_sp)
631 {
632 	struct mlxsw_sp_lpm_tree *lpm_tree;
633 	u64 max_trees;
634 	int i;
635 
636 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_LPM_TREES))
637 		return -EIO;
638 
639 	max_trees = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_LPM_TREES);
640 	mlxsw_sp->router->lpm.tree_count = max_trees - MLXSW_SP_LPM_TREE_MIN;
641 	mlxsw_sp->router->lpm.trees = kcalloc(mlxsw_sp->router->lpm.tree_count,
642 					     sizeof(struct mlxsw_sp_lpm_tree),
643 					     GFP_KERNEL);
644 	if (!mlxsw_sp->router->lpm.trees)
645 		return -ENOMEM;
646 
647 	for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
648 		lpm_tree = &mlxsw_sp->router->lpm.trees[i];
649 		lpm_tree->id = i + MLXSW_SP_LPM_TREE_MIN;
650 	}
651 
652 	return 0;
653 }
654 
655 static void mlxsw_sp_lpm_fini(struct mlxsw_sp *mlxsw_sp)
656 {
657 	kfree(mlxsw_sp->router->lpm.trees);
658 }
659 
660 static bool mlxsw_sp_vr_is_used(const struct mlxsw_sp_vr *vr)
661 {
662 	return !!vr->fib4 || !!vr->fib6 || !!vr->mr4_table;
663 }
664 
665 static struct mlxsw_sp_vr *mlxsw_sp_vr_find_unused(struct mlxsw_sp *mlxsw_sp)
666 {
667 	struct mlxsw_sp_vr *vr;
668 	int i;
669 
670 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
671 		vr = &mlxsw_sp->router->vrs[i];
672 		if (!mlxsw_sp_vr_is_used(vr))
673 			return vr;
674 	}
675 	return NULL;
676 }
677 
678 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
679 				     const struct mlxsw_sp_fib *fib, u8 tree_id)
680 {
681 	char raltb_pl[MLXSW_REG_RALTB_LEN];
682 
683 	mlxsw_reg_raltb_pack(raltb_pl, fib->vr->id,
684 			     (enum mlxsw_reg_ralxx_protocol) fib->proto,
685 			     tree_id);
686 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
687 }
688 
689 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
690 				       const struct mlxsw_sp_fib *fib)
691 {
692 	char raltb_pl[MLXSW_REG_RALTB_LEN];
693 
694 	/* Bind to tree 0 which is default */
695 	mlxsw_reg_raltb_pack(raltb_pl, fib->vr->id,
696 			     (enum mlxsw_reg_ralxx_protocol) fib->proto, 0);
697 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
698 }
699 
700 static u32 mlxsw_sp_fix_tb_id(u32 tb_id)
701 {
702 	/* For our purpose, squash main, default and local tables into one */
703 	if (tb_id == RT_TABLE_LOCAL || tb_id == RT_TABLE_DEFAULT)
704 		tb_id = RT_TABLE_MAIN;
705 	return tb_id;
706 }
707 
708 static struct mlxsw_sp_vr *mlxsw_sp_vr_find(struct mlxsw_sp *mlxsw_sp,
709 					    u32 tb_id)
710 {
711 	struct mlxsw_sp_vr *vr;
712 	int i;
713 
714 	tb_id = mlxsw_sp_fix_tb_id(tb_id);
715 
716 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
717 		vr = &mlxsw_sp->router->vrs[i];
718 		if (mlxsw_sp_vr_is_used(vr) && vr->tb_id == tb_id)
719 			return vr;
720 	}
721 	return NULL;
722 }
723 
724 static struct mlxsw_sp_fib *mlxsw_sp_vr_fib(const struct mlxsw_sp_vr *vr,
725 					    enum mlxsw_sp_l3proto proto)
726 {
727 	switch (proto) {
728 	case MLXSW_SP_L3_PROTO_IPV4:
729 		return vr->fib4;
730 	case MLXSW_SP_L3_PROTO_IPV6:
731 		return vr->fib6;
732 	}
733 	return NULL;
734 }
735 
736 static struct mlxsw_sp_vr *mlxsw_sp_vr_create(struct mlxsw_sp *mlxsw_sp,
737 					      u32 tb_id,
738 					      struct netlink_ext_ack *extack)
739 {
740 	struct mlxsw_sp_vr *vr;
741 	int err;
742 
743 	vr = mlxsw_sp_vr_find_unused(mlxsw_sp);
744 	if (!vr) {
745 		NL_SET_ERR_MSG(extack, "spectrum: Exceeded number of supported virtual routers");
746 		return ERR_PTR(-EBUSY);
747 	}
748 	vr->fib4 = mlxsw_sp_fib_create(vr, MLXSW_SP_L3_PROTO_IPV4);
749 	if (IS_ERR(vr->fib4))
750 		return ERR_CAST(vr->fib4);
751 	vr->fib6 = mlxsw_sp_fib_create(vr, MLXSW_SP_L3_PROTO_IPV6);
752 	if (IS_ERR(vr->fib6)) {
753 		err = PTR_ERR(vr->fib6);
754 		goto err_fib6_create;
755 	}
756 	vr->mr4_table = mlxsw_sp_mr_table_create(mlxsw_sp, vr->id,
757 						 MLXSW_SP_L3_PROTO_IPV4);
758 	if (IS_ERR(vr->mr4_table)) {
759 		err = PTR_ERR(vr->mr4_table);
760 		goto err_mr_table_create;
761 	}
762 	vr->tb_id = tb_id;
763 	return vr;
764 
765 err_mr_table_create:
766 	mlxsw_sp_fib_destroy(vr->fib6);
767 	vr->fib6 = NULL;
768 err_fib6_create:
769 	mlxsw_sp_fib_destroy(vr->fib4);
770 	vr->fib4 = NULL;
771 	return ERR_PTR(err);
772 }
773 
774 static void mlxsw_sp_vr_destroy(struct mlxsw_sp_vr *vr)
775 {
776 	mlxsw_sp_mr_table_destroy(vr->mr4_table);
777 	vr->mr4_table = NULL;
778 	mlxsw_sp_fib_destroy(vr->fib6);
779 	vr->fib6 = NULL;
780 	mlxsw_sp_fib_destroy(vr->fib4);
781 	vr->fib4 = NULL;
782 }
783 
784 static struct mlxsw_sp_vr *mlxsw_sp_vr_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
785 					   struct netlink_ext_ack *extack)
786 {
787 	struct mlxsw_sp_vr *vr;
788 
789 	tb_id = mlxsw_sp_fix_tb_id(tb_id);
790 	vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
791 	if (!vr)
792 		vr = mlxsw_sp_vr_create(mlxsw_sp, tb_id, extack);
793 	return vr;
794 }
795 
796 static void mlxsw_sp_vr_put(struct mlxsw_sp_vr *vr)
797 {
798 	if (!vr->rif_count && list_empty(&vr->fib4->node_list) &&
799 	    list_empty(&vr->fib6->node_list) &&
800 	    mlxsw_sp_mr_table_empty(vr->mr4_table))
801 		mlxsw_sp_vr_destroy(vr);
802 }
803 
804 static bool
805 mlxsw_sp_vr_lpm_tree_should_replace(struct mlxsw_sp_vr *vr,
806 				    enum mlxsw_sp_l3proto proto, u8 tree_id)
807 {
808 	struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
809 
810 	if (!mlxsw_sp_vr_is_used(vr))
811 		return false;
812 	if (fib->lpm_tree && fib->lpm_tree->id == tree_id)
813 		return true;
814 	return false;
815 }
816 
817 static int mlxsw_sp_vr_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
818 					struct mlxsw_sp_fib *fib,
819 					struct mlxsw_sp_lpm_tree *new_tree)
820 {
821 	struct mlxsw_sp_lpm_tree *old_tree = fib->lpm_tree;
822 	int err;
823 
824 	err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, new_tree->id);
825 	if (err)
826 		return err;
827 	fib->lpm_tree = new_tree;
828 	mlxsw_sp_lpm_tree_hold(new_tree);
829 	mlxsw_sp_lpm_tree_put(mlxsw_sp, old_tree);
830 	return 0;
831 }
832 
833 static int mlxsw_sp_vrs_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
834 					 struct mlxsw_sp_fib *fib,
835 					 struct mlxsw_sp_lpm_tree *new_tree)
836 {
837 	struct mlxsw_sp_lpm_tree *old_tree = fib->lpm_tree;
838 	enum mlxsw_sp_l3proto proto = fib->proto;
839 	u8 old_id, new_id = new_tree->id;
840 	struct mlxsw_sp_vr *vr;
841 	int i, err;
842 
843 	if (!old_tree)
844 		goto no_replace;
845 	old_id = old_tree->id;
846 
847 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
848 		vr = &mlxsw_sp->router->vrs[i];
849 		if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, old_id))
850 			continue;
851 		err = mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
852 						   mlxsw_sp_vr_fib(vr, proto),
853 						   new_tree);
854 		if (err)
855 			goto err_tree_replace;
856 	}
857 
858 	return 0;
859 
860 err_tree_replace:
861 	for (i--; i >= 0; i--) {
862 		if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, new_id))
863 			continue;
864 		mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
865 					     mlxsw_sp_vr_fib(vr, proto),
866 					     old_tree);
867 	}
868 	return err;
869 
870 no_replace:
871 	err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, new_tree->id);
872 	if (err)
873 		return err;
874 	fib->lpm_tree = new_tree;
875 	mlxsw_sp_lpm_tree_hold(new_tree);
876 	return 0;
877 }
878 
879 static void
880 mlxsw_sp_vrs_prefixes(struct mlxsw_sp *mlxsw_sp,
881 		      enum mlxsw_sp_l3proto proto,
882 		      struct mlxsw_sp_prefix_usage *req_prefix_usage)
883 {
884 	int i;
885 
886 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
887 		struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
888 		struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
889 		unsigned char prefix;
890 
891 		if (!mlxsw_sp_vr_is_used(vr))
892 			continue;
893 		mlxsw_sp_prefix_usage_for_each(prefix, &fib->prefix_usage)
894 			mlxsw_sp_prefix_usage_set(req_prefix_usage, prefix);
895 	}
896 }
897 
898 static int mlxsw_sp_vrs_init(struct mlxsw_sp *mlxsw_sp)
899 {
900 	struct mlxsw_sp_vr *vr;
901 	u64 max_vrs;
902 	int i;
903 
904 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_VRS))
905 		return -EIO;
906 
907 	max_vrs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS);
908 	mlxsw_sp->router->vrs = kcalloc(max_vrs, sizeof(struct mlxsw_sp_vr),
909 					GFP_KERNEL);
910 	if (!mlxsw_sp->router->vrs)
911 		return -ENOMEM;
912 
913 	for (i = 0; i < max_vrs; i++) {
914 		vr = &mlxsw_sp->router->vrs[i];
915 		vr->id = i;
916 	}
917 
918 	return 0;
919 }
920 
921 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp);
922 
923 static void mlxsw_sp_vrs_fini(struct mlxsw_sp *mlxsw_sp)
924 {
925 	/* At this stage we're guaranteed not to have new incoming
926 	 * FIB notifications and the work queue is free from FIBs
927 	 * sitting on top of mlxsw netdevs. However, we can still
928 	 * have other FIBs queued. Flush the queue before flushing
929 	 * the device's tables. No need for locks, as we're the only
930 	 * writer.
931 	 */
932 	mlxsw_core_flush_owq();
933 	mlxsw_sp_router_fib_flush(mlxsw_sp);
934 	kfree(mlxsw_sp->router->vrs);
935 }
936 
937 static struct net_device *
938 __mlxsw_sp_ipip_netdev_ul_dev_get(const struct net_device *ol_dev)
939 {
940 	struct ip_tunnel *tun = netdev_priv(ol_dev);
941 	struct net *net = dev_net(ol_dev);
942 
943 	return __dev_get_by_index(net, tun->parms.link);
944 }
945 
946 u32 mlxsw_sp_ipip_dev_ul_tb_id(const struct net_device *ol_dev)
947 {
948 	struct net_device *d = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
949 
950 	if (d)
951 		return l3mdev_fib_table(d) ? : RT_TABLE_MAIN;
952 	else
953 		return l3mdev_fib_table(ol_dev) ? : RT_TABLE_MAIN;
954 }
955 
956 static struct mlxsw_sp_rif *
957 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
958 		    const struct mlxsw_sp_rif_params *params,
959 		    struct netlink_ext_ack *extack);
960 
961 static struct mlxsw_sp_rif_ipip_lb *
962 mlxsw_sp_ipip_ol_ipip_lb_create(struct mlxsw_sp *mlxsw_sp,
963 				enum mlxsw_sp_ipip_type ipipt,
964 				struct net_device *ol_dev,
965 				struct netlink_ext_ack *extack)
966 {
967 	struct mlxsw_sp_rif_params_ipip_lb lb_params;
968 	const struct mlxsw_sp_ipip_ops *ipip_ops;
969 	struct mlxsw_sp_rif *rif;
970 
971 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
972 	lb_params = (struct mlxsw_sp_rif_params_ipip_lb) {
973 		.common.dev = ol_dev,
974 		.common.lag = false,
975 		.lb_config = ipip_ops->ol_loopback_config(mlxsw_sp, ol_dev),
976 	};
977 
978 	rif = mlxsw_sp_rif_create(mlxsw_sp, &lb_params.common, extack);
979 	if (IS_ERR(rif))
980 		return ERR_CAST(rif);
981 	return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
982 }
983 
984 static struct mlxsw_sp_ipip_entry *
985 mlxsw_sp_ipip_entry_alloc(struct mlxsw_sp *mlxsw_sp,
986 			  enum mlxsw_sp_ipip_type ipipt,
987 			  struct net_device *ol_dev)
988 {
989 	struct mlxsw_sp_ipip_entry *ipip_entry;
990 	struct mlxsw_sp_ipip_entry *ret = NULL;
991 
992 	ipip_entry = kzalloc(sizeof(*ipip_entry), GFP_KERNEL);
993 	if (!ipip_entry)
994 		return ERR_PTR(-ENOMEM);
995 
996 	ipip_entry->ol_lb = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp, ipipt,
997 							    ol_dev, NULL);
998 	if (IS_ERR(ipip_entry->ol_lb)) {
999 		ret = ERR_CAST(ipip_entry->ol_lb);
1000 		goto err_ol_ipip_lb_create;
1001 	}
1002 
1003 	ipip_entry->ipipt = ipipt;
1004 	ipip_entry->ol_dev = ol_dev;
1005 	ipip_entry->parms = mlxsw_sp_ipip_netdev_parms(ol_dev);
1006 
1007 	return ipip_entry;
1008 
1009 err_ol_ipip_lb_create:
1010 	kfree(ipip_entry);
1011 	return ret;
1012 }
1013 
1014 static void
1015 mlxsw_sp_ipip_entry_dealloc(struct mlxsw_sp_ipip_entry *ipip_entry)
1016 {
1017 	mlxsw_sp_rif_destroy(&ipip_entry->ol_lb->common);
1018 	kfree(ipip_entry);
1019 }
1020 
1021 static bool
1022 mlxsw_sp_ipip_entry_saddr_matches(struct mlxsw_sp *mlxsw_sp,
1023 				  const enum mlxsw_sp_l3proto ul_proto,
1024 				  union mlxsw_sp_l3addr saddr,
1025 				  u32 ul_tb_id,
1026 				  struct mlxsw_sp_ipip_entry *ipip_entry)
1027 {
1028 	u32 tun_ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1029 	enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1030 	union mlxsw_sp_l3addr tun_saddr;
1031 
1032 	if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1033 		return false;
1034 
1035 	tun_saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ipip_entry->ol_dev);
1036 	return tun_ul_tb_id == ul_tb_id &&
1037 	       mlxsw_sp_l3addr_eq(&tun_saddr, &saddr);
1038 }
1039 
1040 static int
1041 mlxsw_sp_fib_entry_decap_init(struct mlxsw_sp *mlxsw_sp,
1042 			      struct mlxsw_sp_fib_entry *fib_entry,
1043 			      struct mlxsw_sp_ipip_entry *ipip_entry)
1044 {
1045 	u32 tunnel_index;
1046 	int err;
1047 
1048 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, 1, &tunnel_index);
1049 	if (err)
1050 		return err;
1051 
1052 	ipip_entry->decap_fib_entry = fib_entry;
1053 	fib_entry->decap.ipip_entry = ipip_entry;
1054 	fib_entry->decap.tunnel_index = tunnel_index;
1055 	return 0;
1056 }
1057 
1058 static void mlxsw_sp_fib_entry_decap_fini(struct mlxsw_sp *mlxsw_sp,
1059 					  struct mlxsw_sp_fib_entry *fib_entry)
1060 {
1061 	/* Unlink this node from the IPIP entry that it's the decap entry of. */
1062 	fib_entry->decap.ipip_entry->decap_fib_entry = NULL;
1063 	fib_entry->decap.ipip_entry = NULL;
1064 	mlxsw_sp_kvdl_free(mlxsw_sp, fib_entry->decap.tunnel_index);
1065 }
1066 
1067 static struct mlxsw_sp_fib_node *
1068 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
1069 			 size_t addr_len, unsigned char prefix_len);
1070 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1071 				     struct mlxsw_sp_fib_entry *fib_entry);
1072 
1073 static void
1074 mlxsw_sp_ipip_entry_demote_decap(struct mlxsw_sp *mlxsw_sp,
1075 				 struct mlxsw_sp_ipip_entry *ipip_entry)
1076 {
1077 	struct mlxsw_sp_fib_entry *fib_entry = ipip_entry->decap_fib_entry;
1078 
1079 	mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, fib_entry);
1080 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1081 
1082 	mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1083 }
1084 
1085 static void
1086 mlxsw_sp_ipip_entry_promote_decap(struct mlxsw_sp *mlxsw_sp,
1087 				  struct mlxsw_sp_ipip_entry *ipip_entry,
1088 				  struct mlxsw_sp_fib_entry *decap_fib_entry)
1089 {
1090 	if (mlxsw_sp_fib_entry_decap_init(mlxsw_sp, decap_fib_entry,
1091 					  ipip_entry))
1092 		return;
1093 	decap_fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
1094 
1095 	if (mlxsw_sp_fib_entry_update(mlxsw_sp, decap_fib_entry))
1096 		mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1097 }
1098 
1099 /* Given an IPIP entry, find the corresponding decap route. */
1100 static struct mlxsw_sp_fib_entry *
1101 mlxsw_sp_ipip_entry_find_decap(struct mlxsw_sp *mlxsw_sp,
1102 			       struct mlxsw_sp_ipip_entry *ipip_entry)
1103 {
1104 	static struct mlxsw_sp_fib_node *fib_node;
1105 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1106 	struct mlxsw_sp_fib_entry *fib_entry;
1107 	unsigned char saddr_prefix_len;
1108 	union mlxsw_sp_l3addr saddr;
1109 	struct mlxsw_sp_fib *ul_fib;
1110 	struct mlxsw_sp_vr *ul_vr;
1111 	const void *saddrp;
1112 	size_t saddr_len;
1113 	u32 ul_tb_id;
1114 	u32 saddr4;
1115 
1116 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1117 
1118 	ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1119 	ul_vr = mlxsw_sp_vr_find(mlxsw_sp, ul_tb_id);
1120 	if (!ul_vr)
1121 		return NULL;
1122 
1123 	ul_fib = mlxsw_sp_vr_fib(ul_vr, ipip_ops->ul_proto);
1124 	saddr = mlxsw_sp_ipip_netdev_saddr(ipip_ops->ul_proto,
1125 					   ipip_entry->ol_dev);
1126 
1127 	switch (ipip_ops->ul_proto) {
1128 	case MLXSW_SP_L3_PROTO_IPV4:
1129 		saddr4 = be32_to_cpu(saddr.addr4);
1130 		saddrp = &saddr4;
1131 		saddr_len = 4;
1132 		saddr_prefix_len = 32;
1133 		break;
1134 	case MLXSW_SP_L3_PROTO_IPV6:
1135 		WARN_ON(1);
1136 		return NULL;
1137 	}
1138 
1139 	fib_node = mlxsw_sp_fib_node_lookup(ul_fib, saddrp, saddr_len,
1140 					    saddr_prefix_len);
1141 	if (!fib_node || list_empty(&fib_node->entry_list))
1142 		return NULL;
1143 
1144 	fib_entry = list_first_entry(&fib_node->entry_list,
1145 				     struct mlxsw_sp_fib_entry, list);
1146 	if (fib_entry->type != MLXSW_SP_FIB_ENTRY_TYPE_TRAP)
1147 		return NULL;
1148 
1149 	return fib_entry;
1150 }
1151 
1152 static struct mlxsw_sp_ipip_entry *
1153 mlxsw_sp_ipip_entry_create(struct mlxsw_sp *mlxsw_sp,
1154 			   enum mlxsw_sp_ipip_type ipipt,
1155 			   struct net_device *ol_dev)
1156 {
1157 	struct mlxsw_sp_ipip_entry *ipip_entry;
1158 
1159 	ipip_entry = mlxsw_sp_ipip_entry_alloc(mlxsw_sp, ipipt, ol_dev);
1160 	if (IS_ERR(ipip_entry))
1161 		return ipip_entry;
1162 
1163 	list_add_tail(&ipip_entry->ipip_list_node,
1164 		      &mlxsw_sp->router->ipip_list);
1165 
1166 	return ipip_entry;
1167 }
1168 
1169 static void
1170 mlxsw_sp_ipip_entry_destroy(struct mlxsw_sp *mlxsw_sp,
1171 			    struct mlxsw_sp_ipip_entry *ipip_entry)
1172 {
1173 	list_del(&ipip_entry->ipip_list_node);
1174 	mlxsw_sp_ipip_entry_dealloc(ipip_entry);
1175 }
1176 
1177 static bool
1178 mlxsw_sp_ipip_entry_matches_decap(struct mlxsw_sp *mlxsw_sp,
1179 				  const struct net_device *ul_dev,
1180 				  enum mlxsw_sp_l3proto ul_proto,
1181 				  union mlxsw_sp_l3addr ul_dip,
1182 				  struct mlxsw_sp_ipip_entry *ipip_entry)
1183 {
1184 	u32 ul_tb_id = l3mdev_fib_table(ul_dev) ? : RT_TABLE_MAIN;
1185 	enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1186 	struct net_device *ipip_ul_dev;
1187 
1188 	if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1189 		return false;
1190 
1191 	ipip_ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ipip_entry->ol_dev);
1192 	return mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, ul_dip,
1193 						 ul_tb_id, ipip_entry) &&
1194 	       (!ipip_ul_dev || ipip_ul_dev == ul_dev);
1195 }
1196 
1197 /* Given decap parameters, find the corresponding IPIP entry. */
1198 static struct mlxsw_sp_ipip_entry *
1199 mlxsw_sp_ipip_entry_find_by_decap(struct mlxsw_sp *mlxsw_sp,
1200 				  const struct net_device *ul_dev,
1201 				  enum mlxsw_sp_l3proto ul_proto,
1202 				  union mlxsw_sp_l3addr ul_dip)
1203 {
1204 	struct mlxsw_sp_ipip_entry *ipip_entry;
1205 
1206 	list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1207 			    ipip_list_node)
1208 		if (mlxsw_sp_ipip_entry_matches_decap(mlxsw_sp, ul_dev,
1209 						      ul_proto, ul_dip,
1210 						      ipip_entry))
1211 			return ipip_entry;
1212 
1213 	return NULL;
1214 }
1215 
1216 static bool mlxsw_sp_netdev_ipip_type(const struct mlxsw_sp *mlxsw_sp,
1217 				      const struct net_device *dev,
1218 				      enum mlxsw_sp_ipip_type *p_type)
1219 {
1220 	struct mlxsw_sp_router *router = mlxsw_sp->router;
1221 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1222 	enum mlxsw_sp_ipip_type ipipt;
1223 
1224 	for (ipipt = 0; ipipt < MLXSW_SP_IPIP_TYPE_MAX; ++ipipt) {
1225 		ipip_ops = router->ipip_ops_arr[ipipt];
1226 		if (dev->type == ipip_ops->dev_type) {
1227 			if (p_type)
1228 				*p_type = ipipt;
1229 			return true;
1230 		}
1231 	}
1232 	return false;
1233 }
1234 
1235 bool mlxsw_sp_netdev_is_ipip_ol(const struct mlxsw_sp *mlxsw_sp,
1236 				const struct net_device *dev)
1237 {
1238 	return mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL);
1239 }
1240 
1241 static struct mlxsw_sp_ipip_entry *
1242 mlxsw_sp_ipip_entry_find_by_ol_dev(struct mlxsw_sp *mlxsw_sp,
1243 				   const struct net_device *ol_dev)
1244 {
1245 	struct mlxsw_sp_ipip_entry *ipip_entry;
1246 
1247 	list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1248 			    ipip_list_node)
1249 		if (ipip_entry->ol_dev == ol_dev)
1250 			return ipip_entry;
1251 
1252 	return NULL;
1253 }
1254 
1255 static struct mlxsw_sp_ipip_entry *
1256 mlxsw_sp_ipip_entry_find_by_ul_dev(const struct mlxsw_sp *mlxsw_sp,
1257 				   const struct net_device *ul_dev,
1258 				   struct mlxsw_sp_ipip_entry *start)
1259 {
1260 	struct mlxsw_sp_ipip_entry *ipip_entry;
1261 
1262 	ipip_entry = list_prepare_entry(start, &mlxsw_sp->router->ipip_list,
1263 					ipip_list_node);
1264 	list_for_each_entry_continue(ipip_entry, &mlxsw_sp->router->ipip_list,
1265 				     ipip_list_node) {
1266 		struct net_device *ipip_ul_dev =
1267 			__mlxsw_sp_ipip_netdev_ul_dev_get(ipip_entry->ol_dev);
1268 
1269 		if (ipip_ul_dev == ul_dev)
1270 			return ipip_entry;
1271 	}
1272 
1273 	return NULL;
1274 }
1275 
1276 bool mlxsw_sp_netdev_is_ipip_ul(const struct mlxsw_sp *mlxsw_sp,
1277 				const struct net_device *dev)
1278 {
1279 	return mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp, dev, NULL);
1280 }
1281 
1282 static bool mlxsw_sp_netdevice_ipip_can_offload(struct mlxsw_sp *mlxsw_sp,
1283 						const struct net_device *ol_dev,
1284 						enum mlxsw_sp_ipip_type ipipt)
1285 {
1286 	const struct mlxsw_sp_ipip_ops *ops
1287 		= mlxsw_sp->router->ipip_ops_arr[ipipt];
1288 
1289 	/* For deciding whether decap should be offloaded, we don't care about
1290 	 * overlay protocol, so ask whether either one is supported.
1291 	 */
1292 	return ops->can_offload(mlxsw_sp, ol_dev, MLXSW_SP_L3_PROTO_IPV4) ||
1293 	       ops->can_offload(mlxsw_sp, ol_dev, MLXSW_SP_L3_PROTO_IPV6);
1294 }
1295 
1296 static int mlxsw_sp_netdevice_ipip_ol_reg_event(struct mlxsw_sp *mlxsw_sp,
1297 						struct net_device *ol_dev)
1298 {
1299 	struct mlxsw_sp_ipip_entry *ipip_entry;
1300 	enum mlxsw_sp_l3proto ul_proto;
1301 	enum mlxsw_sp_ipip_type ipipt;
1302 	union mlxsw_sp_l3addr saddr;
1303 	u32 ul_tb_id;
1304 
1305 	mlxsw_sp_netdev_ipip_type(mlxsw_sp, ol_dev, &ipipt);
1306 	if (mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev, ipipt)) {
1307 		ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ol_dev);
1308 		ul_proto = mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto;
1309 		saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ol_dev);
1310 		if (!mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1311 							  saddr, ul_tb_id,
1312 							  NULL)) {
1313 			ipip_entry = mlxsw_sp_ipip_entry_create(mlxsw_sp, ipipt,
1314 								ol_dev);
1315 			if (IS_ERR(ipip_entry))
1316 				return PTR_ERR(ipip_entry);
1317 		}
1318 	}
1319 
1320 	return 0;
1321 }
1322 
1323 static void mlxsw_sp_netdevice_ipip_ol_unreg_event(struct mlxsw_sp *mlxsw_sp,
1324 						   struct net_device *ol_dev)
1325 {
1326 	struct mlxsw_sp_ipip_entry *ipip_entry;
1327 
1328 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1329 	if (ipip_entry)
1330 		mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1331 }
1332 
1333 static void
1334 mlxsw_sp_ipip_entry_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1335 				struct mlxsw_sp_ipip_entry *ipip_entry)
1336 {
1337 	struct mlxsw_sp_fib_entry *decap_fib_entry;
1338 
1339 	decap_fib_entry = mlxsw_sp_ipip_entry_find_decap(mlxsw_sp, ipip_entry);
1340 	if (decap_fib_entry)
1341 		mlxsw_sp_ipip_entry_promote_decap(mlxsw_sp, ipip_entry,
1342 						  decap_fib_entry);
1343 }
1344 
1345 static void mlxsw_sp_netdevice_ipip_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1346 						struct net_device *ol_dev)
1347 {
1348 	struct mlxsw_sp_ipip_entry *ipip_entry;
1349 
1350 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1351 	if (ipip_entry)
1352 		mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1353 }
1354 
1355 static void
1356 mlxsw_sp_ipip_entry_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1357 				  struct mlxsw_sp_ipip_entry *ipip_entry)
1358 {
1359 	if (ipip_entry->decap_fib_entry)
1360 		mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1361 }
1362 
1363 static void mlxsw_sp_netdevice_ipip_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1364 						  struct net_device *ol_dev)
1365 {
1366 	struct mlxsw_sp_ipip_entry *ipip_entry;
1367 
1368 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1369 	if (ipip_entry)
1370 		mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1371 }
1372 
1373 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
1374 					 struct mlxsw_sp_rif *old_rif,
1375 					 struct mlxsw_sp_rif *new_rif);
1376 static int
1377 mlxsw_sp_ipip_entry_ol_lb_update(struct mlxsw_sp *mlxsw_sp,
1378 				 struct mlxsw_sp_ipip_entry *ipip_entry,
1379 				 bool keep_encap,
1380 				 struct netlink_ext_ack *extack)
1381 {
1382 	struct mlxsw_sp_rif_ipip_lb *old_lb_rif = ipip_entry->ol_lb;
1383 	struct mlxsw_sp_rif_ipip_lb *new_lb_rif;
1384 
1385 	new_lb_rif = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp,
1386 						     ipip_entry->ipipt,
1387 						     ipip_entry->ol_dev,
1388 						     extack);
1389 	if (IS_ERR(new_lb_rif))
1390 		return PTR_ERR(new_lb_rif);
1391 	ipip_entry->ol_lb = new_lb_rif;
1392 
1393 	if (keep_encap)
1394 		mlxsw_sp_nexthop_rif_migrate(mlxsw_sp, &old_lb_rif->common,
1395 					     &new_lb_rif->common);
1396 
1397 	mlxsw_sp_rif_destroy(&old_lb_rif->common);
1398 
1399 	return 0;
1400 }
1401 
1402 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
1403 					struct mlxsw_sp_rif *rif);
1404 
1405 /**
1406  * Update the offload related to an IPIP entry. This always updates decap, and
1407  * in addition to that it also:
1408  * @recreate_loopback: recreates the associated loopback RIF
1409  * @keep_encap: updates next hops that use the tunnel netdevice. This is only
1410  *              relevant when recreate_loopback is true.
1411  * @update_nexthops: updates next hops, keeping the current loopback RIF. This
1412  *                   is only relevant when recreate_loopback is false.
1413  */
1414 int __mlxsw_sp_ipip_entry_update_tunnel(struct mlxsw_sp *mlxsw_sp,
1415 					struct mlxsw_sp_ipip_entry *ipip_entry,
1416 					bool recreate_loopback,
1417 					bool keep_encap,
1418 					bool update_nexthops,
1419 					struct netlink_ext_ack *extack)
1420 {
1421 	int err;
1422 
1423 	/* RIFs can't be edited, so to update loopback, we need to destroy and
1424 	 * recreate it. That creates a window of opportunity where RALUE and
1425 	 * RATR registers end up referencing a RIF that's already gone. RATRs
1426 	 * are handled in mlxsw_sp_ipip_entry_ol_lb_update(), and to take care
1427 	 * of RALUE, demote the decap route back.
1428 	 */
1429 	if (ipip_entry->decap_fib_entry)
1430 		mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1431 
1432 	if (recreate_loopback) {
1433 		err = mlxsw_sp_ipip_entry_ol_lb_update(mlxsw_sp, ipip_entry,
1434 						       keep_encap, extack);
1435 		if (err)
1436 			return err;
1437 	} else if (update_nexthops) {
1438 		mlxsw_sp_nexthop_rif_update(mlxsw_sp,
1439 					    &ipip_entry->ol_lb->common);
1440 	}
1441 
1442 	if (ipip_entry->ol_dev->flags & IFF_UP)
1443 		mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1444 
1445 	return 0;
1446 }
1447 
1448 static int mlxsw_sp_netdevice_ipip_ol_vrf_event(struct mlxsw_sp *mlxsw_sp,
1449 						struct net_device *ol_dev,
1450 						struct netlink_ext_ack *extack)
1451 {
1452 	struct mlxsw_sp_ipip_entry *ipip_entry =
1453 		mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1454 	enum mlxsw_sp_l3proto ul_proto;
1455 	union mlxsw_sp_l3addr saddr;
1456 	u32 ul_tb_id;
1457 
1458 	if (!ipip_entry)
1459 		return 0;
1460 
1461 	/* For flat configuration cases, moving overlay to a different VRF might
1462 	 * cause local address conflict, and the conflicting tunnels need to be
1463 	 * demoted.
1464 	 */
1465 	ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ol_dev);
1466 	ul_proto = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt]->ul_proto;
1467 	saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ol_dev);
1468 	if (mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1469 						 saddr, ul_tb_id,
1470 						 ipip_entry)) {
1471 		mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1472 		return 0;
1473 	}
1474 
1475 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1476 						   true, false, false, extack);
1477 }
1478 
1479 static int
1480 mlxsw_sp_netdevice_ipip_ul_vrf_event(struct mlxsw_sp *mlxsw_sp,
1481 				     struct mlxsw_sp_ipip_entry *ipip_entry,
1482 				     struct net_device *ul_dev,
1483 				     struct netlink_ext_ack *extack)
1484 {
1485 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1486 						   true, true, false, extack);
1487 }
1488 
1489 static int
1490 mlxsw_sp_netdevice_ipip_ul_up_event(struct mlxsw_sp *mlxsw_sp,
1491 				    struct mlxsw_sp_ipip_entry *ipip_entry,
1492 				    struct net_device *ul_dev)
1493 {
1494 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1495 						   false, false, true, NULL);
1496 }
1497 
1498 static int
1499 mlxsw_sp_netdevice_ipip_ul_down_event(struct mlxsw_sp *mlxsw_sp,
1500 				      struct mlxsw_sp_ipip_entry *ipip_entry,
1501 				      struct net_device *ul_dev)
1502 {
1503 	/* A down underlay device causes encapsulated packets to not be
1504 	 * forwarded, but decap still works. So refresh next hops without
1505 	 * touching anything else.
1506 	 */
1507 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1508 						   false, false, true, NULL);
1509 }
1510 
1511 static int
1512 mlxsw_sp_netdevice_ipip_ol_change_event(struct mlxsw_sp *mlxsw_sp,
1513 					struct net_device *ol_dev,
1514 					struct netlink_ext_ack *extack)
1515 {
1516 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1517 	struct mlxsw_sp_ipip_entry *ipip_entry;
1518 	int err;
1519 
1520 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1521 	if (!ipip_entry)
1522 		/* A change might make a tunnel eligible for offloading, but
1523 		 * that is currently not implemented. What falls to slow path
1524 		 * stays there.
1525 		 */
1526 		return 0;
1527 
1528 	/* A change might make a tunnel not eligible for offloading. */
1529 	if (!mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev,
1530 						 ipip_entry->ipipt)) {
1531 		mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1532 		return 0;
1533 	}
1534 
1535 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1536 	err = ipip_ops->ol_netdev_change(mlxsw_sp, ipip_entry, extack);
1537 	return err;
1538 }
1539 
1540 void mlxsw_sp_ipip_entry_demote_tunnel(struct mlxsw_sp *mlxsw_sp,
1541 				       struct mlxsw_sp_ipip_entry *ipip_entry)
1542 {
1543 	struct net_device *ol_dev = ipip_entry->ol_dev;
1544 
1545 	if (ol_dev->flags & IFF_UP)
1546 		mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1547 	mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1548 }
1549 
1550 /* The configuration where several tunnels have the same local address in the
1551  * same underlay table needs special treatment in the HW. That is currently not
1552  * implemented in the driver. This function finds and demotes the first tunnel
1553  * with a given source address, except the one passed in in the argument
1554  * `except'.
1555  */
1556 bool
1557 mlxsw_sp_ipip_demote_tunnel_by_saddr(struct mlxsw_sp *mlxsw_sp,
1558 				     enum mlxsw_sp_l3proto ul_proto,
1559 				     union mlxsw_sp_l3addr saddr,
1560 				     u32 ul_tb_id,
1561 				     const struct mlxsw_sp_ipip_entry *except)
1562 {
1563 	struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1564 
1565 	list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1566 				 ipip_list_node) {
1567 		if (ipip_entry != except &&
1568 		    mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, saddr,
1569 						      ul_tb_id, ipip_entry)) {
1570 			mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1571 			return true;
1572 		}
1573 	}
1574 
1575 	return false;
1576 }
1577 
1578 static void mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(struct mlxsw_sp *mlxsw_sp,
1579 						     struct net_device *ul_dev)
1580 {
1581 	struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1582 
1583 	list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1584 				 ipip_list_node) {
1585 		struct net_device *ipip_ul_dev =
1586 			__mlxsw_sp_ipip_netdev_ul_dev_get(ipip_entry->ol_dev);
1587 
1588 		if (ipip_ul_dev == ul_dev)
1589 			mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1590 	}
1591 }
1592 
1593 int mlxsw_sp_netdevice_ipip_ol_event(struct mlxsw_sp *mlxsw_sp,
1594 				     struct net_device *ol_dev,
1595 				     unsigned long event,
1596 				     struct netdev_notifier_info *info)
1597 {
1598 	struct netdev_notifier_changeupper_info *chup;
1599 	struct netlink_ext_ack *extack;
1600 
1601 	switch (event) {
1602 	case NETDEV_REGISTER:
1603 		return mlxsw_sp_netdevice_ipip_ol_reg_event(mlxsw_sp, ol_dev);
1604 	case NETDEV_UNREGISTER:
1605 		mlxsw_sp_netdevice_ipip_ol_unreg_event(mlxsw_sp, ol_dev);
1606 		return 0;
1607 	case NETDEV_UP:
1608 		mlxsw_sp_netdevice_ipip_ol_up_event(mlxsw_sp, ol_dev);
1609 		return 0;
1610 	case NETDEV_DOWN:
1611 		mlxsw_sp_netdevice_ipip_ol_down_event(mlxsw_sp, ol_dev);
1612 		return 0;
1613 	case NETDEV_CHANGEUPPER:
1614 		chup = container_of(info, typeof(*chup), info);
1615 		extack = info->extack;
1616 		if (netif_is_l3_master(chup->upper_dev))
1617 			return mlxsw_sp_netdevice_ipip_ol_vrf_event(mlxsw_sp,
1618 								    ol_dev,
1619 								    extack);
1620 		return 0;
1621 	case NETDEV_CHANGE:
1622 		extack = info->extack;
1623 		return mlxsw_sp_netdevice_ipip_ol_change_event(mlxsw_sp,
1624 							       ol_dev, extack);
1625 	}
1626 	return 0;
1627 }
1628 
1629 static int
1630 __mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1631 				   struct mlxsw_sp_ipip_entry *ipip_entry,
1632 				   struct net_device *ul_dev,
1633 				   unsigned long event,
1634 				   struct netdev_notifier_info *info)
1635 {
1636 	struct netdev_notifier_changeupper_info *chup;
1637 	struct netlink_ext_ack *extack;
1638 
1639 	switch (event) {
1640 	case NETDEV_CHANGEUPPER:
1641 		chup = container_of(info, typeof(*chup), info);
1642 		extack = info->extack;
1643 		if (netif_is_l3_master(chup->upper_dev))
1644 			return mlxsw_sp_netdevice_ipip_ul_vrf_event(mlxsw_sp,
1645 								    ipip_entry,
1646 								    ul_dev,
1647 								    extack);
1648 		break;
1649 
1650 	case NETDEV_UP:
1651 		return mlxsw_sp_netdevice_ipip_ul_up_event(mlxsw_sp, ipip_entry,
1652 							   ul_dev);
1653 	case NETDEV_DOWN:
1654 		return mlxsw_sp_netdevice_ipip_ul_down_event(mlxsw_sp,
1655 							     ipip_entry,
1656 							     ul_dev);
1657 	}
1658 	return 0;
1659 }
1660 
1661 int
1662 mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1663 				 struct net_device *ul_dev,
1664 				 unsigned long event,
1665 				 struct netdev_notifier_info *info)
1666 {
1667 	struct mlxsw_sp_ipip_entry *ipip_entry = NULL;
1668 	int err;
1669 
1670 	while ((ipip_entry = mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp,
1671 								ul_dev,
1672 								ipip_entry))) {
1673 		err = __mlxsw_sp_netdevice_ipip_ul_event(mlxsw_sp, ipip_entry,
1674 							 ul_dev, event, info);
1675 		if (err) {
1676 			mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(mlxsw_sp,
1677 								 ul_dev);
1678 			return err;
1679 		}
1680 	}
1681 
1682 	return 0;
1683 }
1684 
1685 struct mlxsw_sp_neigh_key {
1686 	struct neighbour *n;
1687 };
1688 
1689 struct mlxsw_sp_neigh_entry {
1690 	struct list_head rif_list_node;
1691 	struct rhash_head ht_node;
1692 	struct mlxsw_sp_neigh_key key;
1693 	u16 rif;
1694 	bool connected;
1695 	unsigned char ha[ETH_ALEN];
1696 	struct list_head nexthop_list; /* list of nexthops using
1697 					* this neigh entry
1698 					*/
1699 	struct list_head nexthop_neighs_list_node;
1700 	unsigned int counter_index;
1701 	bool counter_valid;
1702 };
1703 
1704 static const struct rhashtable_params mlxsw_sp_neigh_ht_params = {
1705 	.key_offset = offsetof(struct mlxsw_sp_neigh_entry, key),
1706 	.head_offset = offsetof(struct mlxsw_sp_neigh_entry, ht_node),
1707 	.key_len = sizeof(struct mlxsw_sp_neigh_key),
1708 };
1709 
1710 struct mlxsw_sp_neigh_entry *
1711 mlxsw_sp_rif_neigh_next(struct mlxsw_sp_rif *rif,
1712 			struct mlxsw_sp_neigh_entry *neigh_entry)
1713 {
1714 	if (!neigh_entry) {
1715 		if (list_empty(&rif->neigh_list))
1716 			return NULL;
1717 		else
1718 			return list_first_entry(&rif->neigh_list,
1719 						typeof(*neigh_entry),
1720 						rif_list_node);
1721 	}
1722 	if (list_is_last(&neigh_entry->rif_list_node, &rif->neigh_list))
1723 		return NULL;
1724 	return list_next_entry(neigh_entry, rif_list_node);
1725 }
1726 
1727 int mlxsw_sp_neigh_entry_type(struct mlxsw_sp_neigh_entry *neigh_entry)
1728 {
1729 	return neigh_entry->key.n->tbl->family;
1730 }
1731 
1732 unsigned char *
1733 mlxsw_sp_neigh_entry_ha(struct mlxsw_sp_neigh_entry *neigh_entry)
1734 {
1735 	return neigh_entry->ha;
1736 }
1737 
1738 u32 mlxsw_sp_neigh4_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
1739 {
1740 	struct neighbour *n;
1741 
1742 	n = neigh_entry->key.n;
1743 	return ntohl(*((__be32 *) n->primary_key));
1744 }
1745 
1746 struct in6_addr *
1747 mlxsw_sp_neigh6_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
1748 {
1749 	struct neighbour *n;
1750 
1751 	n = neigh_entry->key.n;
1752 	return (struct in6_addr *) &n->primary_key;
1753 }
1754 
1755 int mlxsw_sp_neigh_counter_get(struct mlxsw_sp *mlxsw_sp,
1756 			       struct mlxsw_sp_neigh_entry *neigh_entry,
1757 			       u64 *p_counter)
1758 {
1759 	if (!neigh_entry->counter_valid)
1760 		return -EINVAL;
1761 
1762 	return mlxsw_sp_flow_counter_get(mlxsw_sp, neigh_entry->counter_index,
1763 					 p_counter, NULL);
1764 }
1765 
1766 static struct mlxsw_sp_neigh_entry *
1767 mlxsw_sp_neigh_entry_alloc(struct mlxsw_sp *mlxsw_sp, struct neighbour *n,
1768 			   u16 rif)
1769 {
1770 	struct mlxsw_sp_neigh_entry *neigh_entry;
1771 
1772 	neigh_entry = kzalloc(sizeof(*neigh_entry), GFP_KERNEL);
1773 	if (!neigh_entry)
1774 		return NULL;
1775 
1776 	neigh_entry->key.n = n;
1777 	neigh_entry->rif = rif;
1778 	INIT_LIST_HEAD(&neigh_entry->nexthop_list);
1779 
1780 	return neigh_entry;
1781 }
1782 
1783 static void mlxsw_sp_neigh_entry_free(struct mlxsw_sp_neigh_entry *neigh_entry)
1784 {
1785 	kfree(neigh_entry);
1786 }
1787 
1788 static int
1789 mlxsw_sp_neigh_entry_insert(struct mlxsw_sp *mlxsw_sp,
1790 			    struct mlxsw_sp_neigh_entry *neigh_entry)
1791 {
1792 	return rhashtable_insert_fast(&mlxsw_sp->router->neigh_ht,
1793 				      &neigh_entry->ht_node,
1794 				      mlxsw_sp_neigh_ht_params);
1795 }
1796 
1797 static void
1798 mlxsw_sp_neigh_entry_remove(struct mlxsw_sp *mlxsw_sp,
1799 			    struct mlxsw_sp_neigh_entry *neigh_entry)
1800 {
1801 	rhashtable_remove_fast(&mlxsw_sp->router->neigh_ht,
1802 			       &neigh_entry->ht_node,
1803 			       mlxsw_sp_neigh_ht_params);
1804 }
1805 
1806 static bool
1807 mlxsw_sp_neigh_counter_should_alloc(struct mlxsw_sp *mlxsw_sp,
1808 				    struct mlxsw_sp_neigh_entry *neigh_entry)
1809 {
1810 	struct devlink *devlink;
1811 	const char *table_name;
1812 
1813 	switch (mlxsw_sp_neigh_entry_type(neigh_entry)) {
1814 	case AF_INET:
1815 		table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST4;
1816 		break;
1817 	case AF_INET6:
1818 		table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST6;
1819 		break;
1820 	default:
1821 		WARN_ON(1);
1822 		return false;
1823 	}
1824 
1825 	devlink = priv_to_devlink(mlxsw_sp->core);
1826 	return devlink_dpipe_table_counter_enabled(devlink, table_name);
1827 }
1828 
1829 static void
1830 mlxsw_sp_neigh_counter_alloc(struct mlxsw_sp *mlxsw_sp,
1831 			     struct mlxsw_sp_neigh_entry *neigh_entry)
1832 {
1833 	if (!mlxsw_sp_neigh_counter_should_alloc(mlxsw_sp, neigh_entry))
1834 		return;
1835 
1836 	if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &neigh_entry->counter_index))
1837 		return;
1838 
1839 	neigh_entry->counter_valid = true;
1840 }
1841 
1842 static void
1843 mlxsw_sp_neigh_counter_free(struct mlxsw_sp *mlxsw_sp,
1844 			    struct mlxsw_sp_neigh_entry *neigh_entry)
1845 {
1846 	if (!neigh_entry->counter_valid)
1847 		return;
1848 	mlxsw_sp_flow_counter_free(mlxsw_sp,
1849 				   neigh_entry->counter_index);
1850 	neigh_entry->counter_valid = false;
1851 }
1852 
1853 static struct mlxsw_sp_neigh_entry *
1854 mlxsw_sp_neigh_entry_create(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
1855 {
1856 	struct mlxsw_sp_neigh_entry *neigh_entry;
1857 	struct mlxsw_sp_rif *rif;
1858 	int err;
1859 
1860 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, n->dev);
1861 	if (!rif)
1862 		return ERR_PTR(-EINVAL);
1863 
1864 	neigh_entry = mlxsw_sp_neigh_entry_alloc(mlxsw_sp, n, rif->rif_index);
1865 	if (!neigh_entry)
1866 		return ERR_PTR(-ENOMEM);
1867 
1868 	err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
1869 	if (err)
1870 		goto err_neigh_entry_insert;
1871 
1872 	mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
1873 	list_add(&neigh_entry->rif_list_node, &rif->neigh_list);
1874 
1875 	return neigh_entry;
1876 
1877 err_neigh_entry_insert:
1878 	mlxsw_sp_neigh_entry_free(neigh_entry);
1879 	return ERR_PTR(err);
1880 }
1881 
1882 static void
1883 mlxsw_sp_neigh_entry_destroy(struct mlxsw_sp *mlxsw_sp,
1884 			     struct mlxsw_sp_neigh_entry *neigh_entry)
1885 {
1886 	list_del(&neigh_entry->rif_list_node);
1887 	mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
1888 	mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
1889 	mlxsw_sp_neigh_entry_free(neigh_entry);
1890 }
1891 
1892 static struct mlxsw_sp_neigh_entry *
1893 mlxsw_sp_neigh_entry_lookup(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
1894 {
1895 	struct mlxsw_sp_neigh_key key;
1896 
1897 	key.n = n;
1898 	return rhashtable_lookup_fast(&mlxsw_sp->router->neigh_ht,
1899 				      &key, mlxsw_sp_neigh_ht_params);
1900 }
1901 
1902 static void
1903 mlxsw_sp_router_neighs_update_interval_init(struct mlxsw_sp *mlxsw_sp)
1904 {
1905 	unsigned long interval;
1906 
1907 #if IS_ENABLED(CONFIG_IPV6)
1908 	interval = min_t(unsigned long,
1909 			 NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME),
1910 			 NEIGH_VAR(&nd_tbl.parms, DELAY_PROBE_TIME));
1911 #else
1912 	interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME);
1913 #endif
1914 	mlxsw_sp->router->neighs_update.interval = jiffies_to_msecs(interval);
1915 }
1916 
1917 static void mlxsw_sp_router_neigh_ent_ipv4_process(struct mlxsw_sp *mlxsw_sp,
1918 						   char *rauhtd_pl,
1919 						   int ent_index)
1920 {
1921 	struct net_device *dev;
1922 	struct neighbour *n;
1923 	__be32 dipn;
1924 	u32 dip;
1925 	u16 rif;
1926 
1927 	mlxsw_reg_rauhtd_ent_ipv4_unpack(rauhtd_pl, ent_index, &rif, &dip);
1928 
1929 	if (!mlxsw_sp->router->rifs[rif]) {
1930 		dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
1931 		return;
1932 	}
1933 
1934 	dipn = htonl(dip);
1935 	dev = mlxsw_sp->router->rifs[rif]->dev;
1936 	n = neigh_lookup(&arp_tbl, &dipn, dev);
1937 	if (!n) {
1938 		netdev_err(dev, "Failed to find matching neighbour for IP=%pI4h\n",
1939 			   &dip);
1940 		return;
1941 	}
1942 
1943 	netdev_dbg(dev, "Updating neighbour with IP=%pI4h\n", &dip);
1944 	neigh_event_send(n, NULL);
1945 	neigh_release(n);
1946 }
1947 
1948 #if IS_ENABLED(CONFIG_IPV6)
1949 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
1950 						   char *rauhtd_pl,
1951 						   int rec_index)
1952 {
1953 	struct net_device *dev;
1954 	struct neighbour *n;
1955 	struct in6_addr dip;
1956 	u16 rif;
1957 
1958 	mlxsw_reg_rauhtd_ent_ipv6_unpack(rauhtd_pl, rec_index, &rif,
1959 					 (char *) &dip);
1960 
1961 	if (!mlxsw_sp->router->rifs[rif]) {
1962 		dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
1963 		return;
1964 	}
1965 
1966 	dev = mlxsw_sp->router->rifs[rif]->dev;
1967 	n = neigh_lookup(&nd_tbl, &dip, dev);
1968 	if (!n) {
1969 		netdev_err(dev, "Failed to find matching neighbour for IP=%pI6c\n",
1970 			   &dip);
1971 		return;
1972 	}
1973 
1974 	netdev_dbg(dev, "Updating neighbour with IP=%pI6c\n", &dip);
1975 	neigh_event_send(n, NULL);
1976 	neigh_release(n);
1977 }
1978 #else
1979 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
1980 						   char *rauhtd_pl,
1981 						   int rec_index)
1982 {
1983 }
1984 #endif
1985 
1986 static void mlxsw_sp_router_neigh_rec_ipv4_process(struct mlxsw_sp *mlxsw_sp,
1987 						   char *rauhtd_pl,
1988 						   int rec_index)
1989 {
1990 	u8 num_entries;
1991 	int i;
1992 
1993 	num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
1994 								rec_index);
1995 	/* Hardware starts counting at 0, so add 1. */
1996 	num_entries++;
1997 
1998 	/* Each record consists of several neighbour entries. */
1999 	for (i = 0; i < num_entries; i++) {
2000 		int ent_index;
2001 
2002 		ent_index = rec_index * MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC + i;
2003 		mlxsw_sp_router_neigh_ent_ipv4_process(mlxsw_sp, rauhtd_pl,
2004 						       ent_index);
2005 	}
2006 
2007 }
2008 
2009 static void mlxsw_sp_router_neigh_rec_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2010 						   char *rauhtd_pl,
2011 						   int rec_index)
2012 {
2013 	/* One record contains one entry. */
2014 	mlxsw_sp_router_neigh_ent_ipv6_process(mlxsw_sp, rauhtd_pl,
2015 					       rec_index);
2016 }
2017 
2018 static void mlxsw_sp_router_neigh_rec_process(struct mlxsw_sp *mlxsw_sp,
2019 					      char *rauhtd_pl, int rec_index)
2020 {
2021 	switch (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, rec_index)) {
2022 	case MLXSW_REG_RAUHTD_TYPE_IPV4:
2023 		mlxsw_sp_router_neigh_rec_ipv4_process(mlxsw_sp, rauhtd_pl,
2024 						       rec_index);
2025 		break;
2026 	case MLXSW_REG_RAUHTD_TYPE_IPV6:
2027 		mlxsw_sp_router_neigh_rec_ipv6_process(mlxsw_sp, rauhtd_pl,
2028 						       rec_index);
2029 		break;
2030 	}
2031 }
2032 
2033 static bool mlxsw_sp_router_rauhtd_is_full(char *rauhtd_pl)
2034 {
2035 	u8 num_rec, last_rec_index, num_entries;
2036 
2037 	num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2038 	last_rec_index = num_rec - 1;
2039 
2040 	if (num_rec < MLXSW_REG_RAUHTD_REC_MAX_NUM)
2041 		return false;
2042 	if (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, last_rec_index) ==
2043 	    MLXSW_REG_RAUHTD_TYPE_IPV6)
2044 		return true;
2045 
2046 	num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
2047 								last_rec_index);
2048 	if (++num_entries == MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC)
2049 		return true;
2050 	return false;
2051 }
2052 
2053 static int
2054 __mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp,
2055 				       char *rauhtd_pl,
2056 				       enum mlxsw_reg_rauhtd_type type)
2057 {
2058 	int i, num_rec;
2059 	int err;
2060 
2061 	/* Make sure the neighbour's netdev isn't removed in the
2062 	 * process.
2063 	 */
2064 	rtnl_lock();
2065 	do {
2066 		mlxsw_reg_rauhtd_pack(rauhtd_pl, type);
2067 		err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(rauhtd),
2068 				      rauhtd_pl);
2069 		if (err) {
2070 			dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Failed to dump neighbour table\n");
2071 			break;
2072 		}
2073 		num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2074 		for (i = 0; i < num_rec; i++)
2075 			mlxsw_sp_router_neigh_rec_process(mlxsw_sp, rauhtd_pl,
2076 							  i);
2077 	} while (mlxsw_sp_router_rauhtd_is_full(rauhtd_pl));
2078 	rtnl_unlock();
2079 
2080 	return err;
2081 }
2082 
2083 static int mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp)
2084 {
2085 	enum mlxsw_reg_rauhtd_type type;
2086 	char *rauhtd_pl;
2087 	int err;
2088 
2089 	rauhtd_pl = kmalloc(MLXSW_REG_RAUHTD_LEN, GFP_KERNEL);
2090 	if (!rauhtd_pl)
2091 		return -ENOMEM;
2092 
2093 	type = MLXSW_REG_RAUHTD_TYPE_IPV4;
2094 	err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2095 	if (err)
2096 		goto out;
2097 
2098 	type = MLXSW_REG_RAUHTD_TYPE_IPV6;
2099 	err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2100 out:
2101 	kfree(rauhtd_pl);
2102 	return err;
2103 }
2104 
2105 static void mlxsw_sp_router_neighs_update_nh(struct mlxsw_sp *mlxsw_sp)
2106 {
2107 	struct mlxsw_sp_neigh_entry *neigh_entry;
2108 
2109 	/* Take RTNL mutex here to prevent lists from changes */
2110 	rtnl_lock();
2111 	list_for_each_entry(neigh_entry, &mlxsw_sp->router->nexthop_neighs_list,
2112 			    nexthop_neighs_list_node)
2113 		/* If this neigh have nexthops, make the kernel think this neigh
2114 		 * is active regardless of the traffic.
2115 		 */
2116 		neigh_event_send(neigh_entry->key.n, NULL);
2117 	rtnl_unlock();
2118 }
2119 
2120 static void
2121 mlxsw_sp_router_neighs_update_work_schedule(struct mlxsw_sp *mlxsw_sp)
2122 {
2123 	unsigned long interval = mlxsw_sp->router->neighs_update.interval;
2124 
2125 	mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw,
2126 			       msecs_to_jiffies(interval));
2127 }
2128 
2129 static void mlxsw_sp_router_neighs_update_work(struct work_struct *work)
2130 {
2131 	struct mlxsw_sp_router *router;
2132 	int err;
2133 
2134 	router = container_of(work, struct mlxsw_sp_router,
2135 			      neighs_update.dw.work);
2136 	err = mlxsw_sp_router_neighs_update_rauhtd(router->mlxsw_sp);
2137 	if (err)
2138 		dev_err(router->mlxsw_sp->bus_info->dev, "Could not update kernel for neigh activity");
2139 
2140 	mlxsw_sp_router_neighs_update_nh(router->mlxsw_sp);
2141 
2142 	mlxsw_sp_router_neighs_update_work_schedule(router->mlxsw_sp);
2143 }
2144 
2145 static void mlxsw_sp_router_probe_unresolved_nexthops(struct work_struct *work)
2146 {
2147 	struct mlxsw_sp_neigh_entry *neigh_entry;
2148 	struct mlxsw_sp_router *router;
2149 
2150 	router = container_of(work, struct mlxsw_sp_router,
2151 			      nexthop_probe_dw.work);
2152 	/* Iterate over nexthop neighbours, find those who are unresolved and
2153 	 * send arp on them. This solves the chicken-egg problem when
2154 	 * the nexthop wouldn't get offloaded until the neighbor is resolved
2155 	 * but it wouldn't get resolved ever in case traffic is flowing in HW
2156 	 * using different nexthop.
2157 	 *
2158 	 * Take RTNL mutex here to prevent lists from changes.
2159 	 */
2160 	rtnl_lock();
2161 	list_for_each_entry(neigh_entry, &router->nexthop_neighs_list,
2162 			    nexthop_neighs_list_node)
2163 		if (!neigh_entry->connected)
2164 			neigh_event_send(neigh_entry->key.n, NULL);
2165 	rtnl_unlock();
2166 
2167 	mlxsw_core_schedule_dw(&router->nexthop_probe_dw,
2168 			       MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL);
2169 }
2170 
2171 static void
2172 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
2173 			      struct mlxsw_sp_neigh_entry *neigh_entry,
2174 			      bool removing);
2175 
2176 static enum mlxsw_reg_rauht_op mlxsw_sp_rauht_op(bool adding)
2177 {
2178 	return adding ? MLXSW_REG_RAUHT_OP_WRITE_ADD :
2179 			MLXSW_REG_RAUHT_OP_WRITE_DELETE;
2180 }
2181 
2182 static void
2183 mlxsw_sp_router_neigh_entry_op4(struct mlxsw_sp *mlxsw_sp,
2184 				struct mlxsw_sp_neigh_entry *neigh_entry,
2185 				enum mlxsw_reg_rauht_op op)
2186 {
2187 	struct neighbour *n = neigh_entry->key.n;
2188 	u32 dip = ntohl(*((__be32 *) n->primary_key));
2189 	char rauht_pl[MLXSW_REG_RAUHT_LEN];
2190 
2191 	mlxsw_reg_rauht_pack4(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2192 			      dip);
2193 	if (neigh_entry->counter_valid)
2194 		mlxsw_reg_rauht_pack_counter(rauht_pl,
2195 					     neigh_entry->counter_index);
2196 	mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2197 }
2198 
2199 static void
2200 mlxsw_sp_router_neigh_entry_op6(struct mlxsw_sp *mlxsw_sp,
2201 				struct mlxsw_sp_neigh_entry *neigh_entry,
2202 				enum mlxsw_reg_rauht_op op)
2203 {
2204 	struct neighbour *n = neigh_entry->key.n;
2205 	char rauht_pl[MLXSW_REG_RAUHT_LEN];
2206 	const char *dip = n->primary_key;
2207 
2208 	mlxsw_reg_rauht_pack6(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2209 			      dip);
2210 	if (neigh_entry->counter_valid)
2211 		mlxsw_reg_rauht_pack_counter(rauht_pl,
2212 					     neigh_entry->counter_index);
2213 	mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2214 }
2215 
2216 bool mlxsw_sp_neigh_ipv6_ignore(struct mlxsw_sp_neigh_entry *neigh_entry)
2217 {
2218 	struct neighbour *n = neigh_entry->key.n;
2219 
2220 	/* Packets with a link-local destination address are trapped
2221 	 * after LPM lookup and never reach the neighbour table, so
2222 	 * there is no need to program such neighbours to the device.
2223 	 */
2224 	if (ipv6_addr_type((struct in6_addr *) &n->primary_key) &
2225 	    IPV6_ADDR_LINKLOCAL)
2226 		return true;
2227 	return false;
2228 }
2229 
2230 static void
2231 mlxsw_sp_neigh_entry_update(struct mlxsw_sp *mlxsw_sp,
2232 			    struct mlxsw_sp_neigh_entry *neigh_entry,
2233 			    bool adding)
2234 {
2235 	if (!adding && !neigh_entry->connected)
2236 		return;
2237 	neigh_entry->connected = adding;
2238 	if (neigh_entry->key.n->tbl->family == AF_INET) {
2239 		mlxsw_sp_router_neigh_entry_op4(mlxsw_sp, neigh_entry,
2240 						mlxsw_sp_rauht_op(adding));
2241 	} else if (neigh_entry->key.n->tbl->family == AF_INET6) {
2242 		if (mlxsw_sp_neigh_ipv6_ignore(neigh_entry))
2243 			return;
2244 		mlxsw_sp_router_neigh_entry_op6(mlxsw_sp, neigh_entry,
2245 						mlxsw_sp_rauht_op(adding));
2246 	} else {
2247 		WARN_ON_ONCE(1);
2248 	}
2249 }
2250 
2251 void
2252 mlxsw_sp_neigh_entry_counter_update(struct mlxsw_sp *mlxsw_sp,
2253 				    struct mlxsw_sp_neigh_entry *neigh_entry,
2254 				    bool adding)
2255 {
2256 	if (adding)
2257 		mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
2258 	else
2259 		mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
2260 	mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, true);
2261 }
2262 
2263 struct mlxsw_sp_netevent_work {
2264 	struct work_struct work;
2265 	struct mlxsw_sp *mlxsw_sp;
2266 	struct neighbour *n;
2267 };
2268 
2269 static void mlxsw_sp_router_neigh_event_work(struct work_struct *work)
2270 {
2271 	struct mlxsw_sp_netevent_work *net_work =
2272 		container_of(work, struct mlxsw_sp_netevent_work, work);
2273 	struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2274 	struct mlxsw_sp_neigh_entry *neigh_entry;
2275 	struct neighbour *n = net_work->n;
2276 	unsigned char ha[ETH_ALEN];
2277 	bool entry_connected;
2278 	u8 nud_state, dead;
2279 
2280 	/* If these parameters are changed after we release the lock,
2281 	 * then we are guaranteed to receive another event letting us
2282 	 * know about it.
2283 	 */
2284 	read_lock_bh(&n->lock);
2285 	memcpy(ha, n->ha, ETH_ALEN);
2286 	nud_state = n->nud_state;
2287 	dead = n->dead;
2288 	read_unlock_bh(&n->lock);
2289 
2290 	rtnl_lock();
2291 	entry_connected = nud_state & NUD_VALID && !dead;
2292 	neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
2293 	if (!entry_connected && !neigh_entry)
2294 		goto out;
2295 	if (!neigh_entry) {
2296 		neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
2297 		if (IS_ERR(neigh_entry))
2298 			goto out;
2299 	}
2300 
2301 	memcpy(neigh_entry->ha, ha, ETH_ALEN);
2302 	mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, entry_connected);
2303 	mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, !entry_connected);
2304 
2305 	if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
2306 		mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2307 
2308 out:
2309 	rtnl_unlock();
2310 	neigh_release(n);
2311 	kfree(net_work);
2312 }
2313 
2314 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp);
2315 
2316 static void mlxsw_sp_router_mp_hash_event_work(struct work_struct *work)
2317 {
2318 	struct mlxsw_sp_netevent_work *net_work =
2319 		container_of(work, struct mlxsw_sp_netevent_work, work);
2320 	struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2321 
2322 	mlxsw_sp_mp_hash_init(mlxsw_sp);
2323 	kfree(net_work);
2324 }
2325 
2326 static int mlxsw_sp_router_netevent_event(struct notifier_block *nb,
2327 					  unsigned long event, void *ptr)
2328 {
2329 	struct mlxsw_sp_netevent_work *net_work;
2330 	struct mlxsw_sp_port *mlxsw_sp_port;
2331 	struct mlxsw_sp_router *router;
2332 	struct mlxsw_sp *mlxsw_sp;
2333 	unsigned long interval;
2334 	struct neigh_parms *p;
2335 	struct neighbour *n;
2336 	struct net *net;
2337 
2338 	switch (event) {
2339 	case NETEVENT_DELAY_PROBE_TIME_UPDATE:
2340 		p = ptr;
2341 
2342 		/* We don't care about changes in the default table. */
2343 		if (!p->dev || (p->tbl->family != AF_INET &&
2344 				p->tbl->family != AF_INET6))
2345 			return NOTIFY_DONE;
2346 
2347 		/* We are in atomic context and can't take RTNL mutex,
2348 		 * so use RCU variant to walk the device chain.
2349 		 */
2350 		mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(p->dev);
2351 		if (!mlxsw_sp_port)
2352 			return NOTIFY_DONE;
2353 
2354 		mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2355 		interval = jiffies_to_msecs(NEIGH_VAR(p, DELAY_PROBE_TIME));
2356 		mlxsw_sp->router->neighs_update.interval = interval;
2357 
2358 		mlxsw_sp_port_dev_put(mlxsw_sp_port);
2359 		break;
2360 	case NETEVENT_NEIGH_UPDATE:
2361 		n = ptr;
2362 
2363 		if (n->tbl->family != AF_INET && n->tbl->family != AF_INET6)
2364 			return NOTIFY_DONE;
2365 
2366 		mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(n->dev);
2367 		if (!mlxsw_sp_port)
2368 			return NOTIFY_DONE;
2369 
2370 		net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2371 		if (!net_work) {
2372 			mlxsw_sp_port_dev_put(mlxsw_sp_port);
2373 			return NOTIFY_BAD;
2374 		}
2375 
2376 		INIT_WORK(&net_work->work, mlxsw_sp_router_neigh_event_work);
2377 		net_work->mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2378 		net_work->n = n;
2379 
2380 		/* Take a reference to ensure the neighbour won't be
2381 		 * destructed until we drop the reference in delayed
2382 		 * work.
2383 		 */
2384 		neigh_clone(n);
2385 		mlxsw_core_schedule_work(&net_work->work);
2386 		mlxsw_sp_port_dev_put(mlxsw_sp_port);
2387 		break;
2388 	case NETEVENT_MULTIPATH_HASH_UPDATE:
2389 		net = ptr;
2390 
2391 		if (!net_eq(net, &init_net))
2392 			return NOTIFY_DONE;
2393 
2394 		net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2395 		if (!net_work)
2396 			return NOTIFY_BAD;
2397 
2398 		router = container_of(nb, struct mlxsw_sp_router, netevent_nb);
2399 		INIT_WORK(&net_work->work, mlxsw_sp_router_mp_hash_event_work);
2400 		net_work->mlxsw_sp = router->mlxsw_sp;
2401 		mlxsw_core_schedule_work(&net_work->work);
2402 		break;
2403 	}
2404 
2405 	return NOTIFY_DONE;
2406 }
2407 
2408 static int mlxsw_sp_neigh_init(struct mlxsw_sp *mlxsw_sp)
2409 {
2410 	int err;
2411 
2412 	err = rhashtable_init(&mlxsw_sp->router->neigh_ht,
2413 			      &mlxsw_sp_neigh_ht_params);
2414 	if (err)
2415 		return err;
2416 
2417 	/* Initialize the polling interval according to the default
2418 	 * table.
2419 	 */
2420 	mlxsw_sp_router_neighs_update_interval_init(mlxsw_sp);
2421 
2422 	/* Create the delayed works for the activity_update */
2423 	INIT_DELAYED_WORK(&mlxsw_sp->router->neighs_update.dw,
2424 			  mlxsw_sp_router_neighs_update_work);
2425 	INIT_DELAYED_WORK(&mlxsw_sp->router->nexthop_probe_dw,
2426 			  mlxsw_sp_router_probe_unresolved_nexthops);
2427 	mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw, 0);
2428 	mlxsw_core_schedule_dw(&mlxsw_sp->router->nexthop_probe_dw, 0);
2429 	return 0;
2430 }
2431 
2432 static void mlxsw_sp_neigh_fini(struct mlxsw_sp *mlxsw_sp)
2433 {
2434 	cancel_delayed_work_sync(&mlxsw_sp->router->neighs_update.dw);
2435 	cancel_delayed_work_sync(&mlxsw_sp->router->nexthop_probe_dw);
2436 	rhashtable_destroy(&mlxsw_sp->router->neigh_ht);
2437 }
2438 
2439 static int mlxsw_sp_neigh_rif_flush(struct mlxsw_sp *mlxsw_sp,
2440 				    const struct mlxsw_sp_rif *rif)
2441 {
2442 	char rauht_pl[MLXSW_REG_RAUHT_LEN];
2443 
2444 	mlxsw_reg_rauht_pack(rauht_pl, MLXSW_REG_RAUHT_OP_WRITE_DELETE_ALL,
2445 			     rif->rif_index, rif->addr);
2446 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2447 }
2448 
2449 static void mlxsw_sp_neigh_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
2450 					 struct mlxsw_sp_rif *rif)
2451 {
2452 	struct mlxsw_sp_neigh_entry *neigh_entry, *tmp;
2453 
2454 	mlxsw_sp_neigh_rif_flush(mlxsw_sp, rif);
2455 	list_for_each_entry_safe(neigh_entry, tmp, &rif->neigh_list,
2456 				 rif_list_node)
2457 		mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2458 }
2459 
2460 enum mlxsw_sp_nexthop_type {
2461 	MLXSW_SP_NEXTHOP_TYPE_ETH,
2462 	MLXSW_SP_NEXTHOP_TYPE_IPIP,
2463 };
2464 
2465 struct mlxsw_sp_nexthop_key {
2466 	struct fib_nh *fib_nh;
2467 };
2468 
2469 struct mlxsw_sp_nexthop {
2470 	struct list_head neigh_list_node; /* member of neigh entry list */
2471 	struct list_head rif_list_node;
2472 	struct list_head router_list_node;
2473 	struct mlxsw_sp_nexthop_group *nh_grp; /* pointer back to the group
2474 						* this belongs to
2475 						*/
2476 	struct rhash_head ht_node;
2477 	struct mlxsw_sp_nexthop_key key;
2478 	unsigned char gw_addr[sizeof(struct in6_addr)];
2479 	int ifindex;
2480 	int nh_weight;
2481 	int norm_nh_weight;
2482 	int num_adj_entries;
2483 	struct mlxsw_sp_rif *rif;
2484 	u8 should_offload:1, /* set indicates this neigh is connected and
2485 			      * should be put to KVD linear area of this group.
2486 			      */
2487 	   offloaded:1, /* set in case the neigh is actually put into
2488 			 * KVD linear area of this group.
2489 			 */
2490 	   update:1; /* set indicates that MAC of this neigh should be
2491 		      * updated in HW
2492 		      */
2493 	enum mlxsw_sp_nexthop_type type;
2494 	union {
2495 		struct mlxsw_sp_neigh_entry *neigh_entry;
2496 		struct mlxsw_sp_ipip_entry *ipip_entry;
2497 	};
2498 	unsigned int counter_index;
2499 	bool counter_valid;
2500 };
2501 
2502 struct mlxsw_sp_nexthop_group {
2503 	void *priv;
2504 	struct rhash_head ht_node;
2505 	struct list_head fib_list; /* list of fib entries that use this group */
2506 	struct neigh_table *neigh_tbl;
2507 	u8 adj_index_valid:1,
2508 	   gateway:1; /* routes using the group use a gateway */
2509 	u32 adj_index;
2510 	u16 ecmp_size;
2511 	u16 count;
2512 	int sum_norm_weight;
2513 	struct mlxsw_sp_nexthop nexthops[0];
2514 #define nh_rif	nexthops[0].rif
2515 };
2516 
2517 void mlxsw_sp_nexthop_counter_alloc(struct mlxsw_sp *mlxsw_sp,
2518 				    struct mlxsw_sp_nexthop *nh)
2519 {
2520 	struct devlink *devlink;
2521 
2522 	devlink = priv_to_devlink(mlxsw_sp->core);
2523 	if (!devlink_dpipe_table_counter_enabled(devlink,
2524 						 MLXSW_SP_DPIPE_TABLE_NAME_ADJ))
2525 		return;
2526 
2527 	if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &nh->counter_index))
2528 		return;
2529 
2530 	nh->counter_valid = true;
2531 }
2532 
2533 void mlxsw_sp_nexthop_counter_free(struct mlxsw_sp *mlxsw_sp,
2534 				   struct mlxsw_sp_nexthop *nh)
2535 {
2536 	if (!nh->counter_valid)
2537 		return;
2538 	mlxsw_sp_flow_counter_free(mlxsw_sp, nh->counter_index);
2539 	nh->counter_valid = false;
2540 }
2541 
2542 int mlxsw_sp_nexthop_counter_get(struct mlxsw_sp *mlxsw_sp,
2543 				 struct mlxsw_sp_nexthop *nh, u64 *p_counter)
2544 {
2545 	if (!nh->counter_valid)
2546 		return -EINVAL;
2547 
2548 	return mlxsw_sp_flow_counter_get(mlxsw_sp, nh->counter_index,
2549 					 p_counter, NULL);
2550 }
2551 
2552 struct mlxsw_sp_nexthop *mlxsw_sp_nexthop_next(struct mlxsw_sp_router *router,
2553 					       struct mlxsw_sp_nexthop *nh)
2554 {
2555 	if (!nh) {
2556 		if (list_empty(&router->nexthop_list))
2557 			return NULL;
2558 		else
2559 			return list_first_entry(&router->nexthop_list,
2560 						typeof(*nh), router_list_node);
2561 	}
2562 	if (list_is_last(&nh->router_list_node, &router->nexthop_list))
2563 		return NULL;
2564 	return list_next_entry(nh, router_list_node);
2565 }
2566 
2567 bool mlxsw_sp_nexthop_offload(struct mlxsw_sp_nexthop *nh)
2568 {
2569 	return nh->offloaded;
2570 }
2571 
2572 unsigned char *mlxsw_sp_nexthop_ha(struct mlxsw_sp_nexthop *nh)
2573 {
2574 	if (!nh->offloaded)
2575 		return NULL;
2576 	return nh->neigh_entry->ha;
2577 }
2578 
2579 int mlxsw_sp_nexthop_indexes(struct mlxsw_sp_nexthop *nh, u32 *p_adj_index,
2580 			     u32 *p_adj_size, u32 *p_adj_hash_index)
2581 {
2582 	struct mlxsw_sp_nexthop_group *nh_grp = nh->nh_grp;
2583 	u32 adj_hash_index = 0;
2584 	int i;
2585 
2586 	if (!nh->offloaded || !nh_grp->adj_index_valid)
2587 		return -EINVAL;
2588 
2589 	*p_adj_index = nh_grp->adj_index;
2590 	*p_adj_size = nh_grp->ecmp_size;
2591 
2592 	for (i = 0; i < nh_grp->count; i++) {
2593 		struct mlxsw_sp_nexthop *nh_iter = &nh_grp->nexthops[i];
2594 
2595 		if (nh_iter == nh)
2596 			break;
2597 		if (nh_iter->offloaded)
2598 			adj_hash_index += nh_iter->num_adj_entries;
2599 	}
2600 
2601 	*p_adj_hash_index = adj_hash_index;
2602 	return 0;
2603 }
2604 
2605 struct mlxsw_sp_rif *mlxsw_sp_nexthop_rif(struct mlxsw_sp_nexthop *nh)
2606 {
2607 	return nh->rif;
2608 }
2609 
2610 bool mlxsw_sp_nexthop_group_has_ipip(struct mlxsw_sp_nexthop *nh)
2611 {
2612 	struct mlxsw_sp_nexthop_group *nh_grp = nh->nh_grp;
2613 	int i;
2614 
2615 	for (i = 0; i < nh_grp->count; i++) {
2616 		struct mlxsw_sp_nexthop *nh_iter = &nh_grp->nexthops[i];
2617 
2618 		if (nh_iter->type == MLXSW_SP_NEXTHOP_TYPE_IPIP)
2619 			return true;
2620 	}
2621 	return false;
2622 }
2623 
2624 static struct fib_info *
2625 mlxsw_sp_nexthop4_group_fi(const struct mlxsw_sp_nexthop_group *nh_grp)
2626 {
2627 	return nh_grp->priv;
2628 }
2629 
2630 struct mlxsw_sp_nexthop_group_cmp_arg {
2631 	enum mlxsw_sp_l3proto proto;
2632 	union {
2633 		struct fib_info *fi;
2634 		struct mlxsw_sp_fib6_entry *fib6_entry;
2635 	};
2636 };
2637 
2638 static bool
2639 mlxsw_sp_nexthop6_group_has_nexthop(const struct mlxsw_sp_nexthop_group *nh_grp,
2640 				    const struct in6_addr *gw, int ifindex)
2641 {
2642 	int i;
2643 
2644 	for (i = 0; i < nh_grp->count; i++) {
2645 		const struct mlxsw_sp_nexthop *nh;
2646 
2647 		nh = &nh_grp->nexthops[i];
2648 		if (nh->ifindex == ifindex &&
2649 		    ipv6_addr_equal(gw, (struct in6_addr *) nh->gw_addr))
2650 			return true;
2651 	}
2652 
2653 	return false;
2654 }
2655 
2656 static bool
2657 mlxsw_sp_nexthop6_group_cmp(const struct mlxsw_sp_nexthop_group *nh_grp,
2658 			    const struct mlxsw_sp_fib6_entry *fib6_entry)
2659 {
2660 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
2661 
2662 	if (nh_grp->count != fib6_entry->nrt6)
2663 		return false;
2664 
2665 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
2666 		struct in6_addr *gw;
2667 		int ifindex;
2668 
2669 		ifindex = mlxsw_sp_rt6->rt->dst.dev->ifindex;
2670 		gw = &mlxsw_sp_rt6->rt->rt6i_gateway;
2671 		if (!mlxsw_sp_nexthop6_group_has_nexthop(nh_grp, gw, ifindex))
2672 			return false;
2673 	}
2674 
2675 	return true;
2676 }
2677 
2678 static int
2679 mlxsw_sp_nexthop_group_cmp(struct rhashtable_compare_arg *arg, const void *ptr)
2680 {
2681 	const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = arg->key;
2682 	const struct mlxsw_sp_nexthop_group *nh_grp = ptr;
2683 
2684 	switch (cmp_arg->proto) {
2685 	case MLXSW_SP_L3_PROTO_IPV4:
2686 		return cmp_arg->fi != mlxsw_sp_nexthop4_group_fi(nh_grp);
2687 	case MLXSW_SP_L3_PROTO_IPV6:
2688 		return !mlxsw_sp_nexthop6_group_cmp(nh_grp,
2689 						    cmp_arg->fib6_entry);
2690 	default:
2691 		WARN_ON(1);
2692 		return 1;
2693 	}
2694 }
2695 
2696 static int
2697 mlxsw_sp_nexthop_group_type(const struct mlxsw_sp_nexthop_group *nh_grp)
2698 {
2699 	return nh_grp->neigh_tbl->family;
2700 }
2701 
2702 static u32 mlxsw_sp_nexthop_group_hash_obj(const void *data, u32 len, u32 seed)
2703 {
2704 	const struct mlxsw_sp_nexthop_group *nh_grp = data;
2705 	const struct mlxsw_sp_nexthop *nh;
2706 	struct fib_info *fi;
2707 	unsigned int val;
2708 	int i;
2709 
2710 	switch (mlxsw_sp_nexthop_group_type(nh_grp)) {
2711 	case AF_INET:
2712 		fi = mlxsw_sp_nexthop4_group_fi(nh_grp);
2713 		return jhash(&fi, sizeof(fi), seed);
2714 	case AF_INET6:
2715 		val = nh_grp->count;
2716 		for (i = 0; i < nh_grp->count; i++) {
2717 			nh = &nh_grp->nexthops[i];
2718 			val ^= nh->ifindex;
2719 		}
2720 		return jhash(&val, sizeof(val), seed);
2721 	default:
2722 		WARN_ON(1);
2723 		return 0;
2724 	}
2725 }
2726 
2727 static u32
2728 mlxsw_sp_nexthop6_group_hash(struct mlxsw_sp_fib6_entry *fib6_entry, u32 seed)
2729 {
2730 	unsigned int val = fib6_entry->nrt6;
2731 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
2732 	struct net_device *dev;
2733 
2734 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
2735 		dev = mlxsw_sp_rt6->rt->dst.dev;
2736 		val ^= dev->ifindex;
2737 	}
2738 
2739 	return jhash(&val, sizeof(val), seed);
2740 }
2741 
2742 static u32
2743 mlxsw_sp_nexthop_group_hash(const void *data, u32 len, u32 seed)
2744 {
2745 	const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = data;
2746 
2747 	switch (cmp_arg->proto) {
2748 	case MLXSW_SP_L3_PROTO_IPV4:
2749 		return jhash(&cmp_arg->fi, sizeof(cmp_arg->fi), seed);
2750 	case MLXSW_SP_L3_PROTO_IPV6:
2751 		return mlxsw_sp_nexthop6_group_hash(cmp_arg->fib6_entry, seed);
2752 	default:
2753 		WARN_ON(1);
2754 		return 0;
2755 	}
2756 }
2757 
2758 static const struct rhashtable_params mlxsw_sp_nexthop_group_ht_params = {
2759 	.head_offset = offsetof(struct mlxsw_sp_nexthop_group, ht_node),
2760 	.hashfn	     = mlxsw_sp_nexthop_group_hash,
2761 	.obj_hashfn  = mlxsw_sp_nexthop_group_hash_obj,
2762 	.obj_cmpfn   = mlxsw_sp_nexthop_group_cmp,
2763 };
2764 
2765 static int mlxsw_sp_nexthop_group_insert(struct mlxsw_sp *mlxsw_sp,
2766 					 struct mlxsw_sp_nexthop_group *nh_grp)
2767 {
2768 	if (mlxsw_sp_nexthop_group_type(nh_grp) == AF_INET6 &&
2769 	    !nh_grp->gateway)
2770 		return 0;
2771 
2772 	return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_group_ht,
2773 				      &nh_grp->ht_node,
2774 				      mlxsw_sp_nexthop_group_ht_params);
2775 }
2776 
2777 static void mlxsw_sp_nexthop_group_remove(struct mlxsw_sp *mlxsw_sp,
2778 					  struct mlxsw_sp_nexthop_group *nh_grp)
2779 {
2780 	if (mlxsw_sp_nexthop_group_type(nh_grp) == AF_INET6 &&
2781 	    !nh_grp->gateway)
2782 		return;
2783 
2784 	rhashtable_remove_fast(&mlxsw_sp->router->nexthop_group_ht,
2785 			       &nh_grp->ht_node,
2786 			       mlxsw_sp_nexthop_group_ht_params);
2787 }
2788 
2789 static struct mlxsw_sp_nexthop_group *
2790 mlxsw_sp_nexthop4_group_lookup(struct mlxsw_sp *mlxsw_sp,
2791 			       struct fib_info *fi)
2792 {
2793 	struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
2794 
2795 	cmp_arg.proto = MLXSW_SP_L3_PROTO_IPV4;
2796 	cmp_arg.fi = fi;
2797 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
2798 				      &cmp_arg,
2799 				      mlxsw_sp_nexthop_group_ht_params);
2800 }
2801 
2802 static struct mlxsw_sp_nexthop_group *
2803 mlxsw_sp_nexthop6_group_lookup(struct mlxsw_sp *mlxsw_sp,
2804 			       struct mlxsw_sp_fib6_entry *fib6_entry)
2805 {
2806 	struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
2807 
2808 	cmp_arg.proto = MLXSW_SP_L3_PROTO_IPV6;
2809 	cmp_arg.fib6_entry = fib6_entry;
2810 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
2811 				      &cmp_arg,
2812 				      mlxsw_sp_nexthop_group_ht_params);
2813 }
2814 
2815 static const struct rhashtable_params mlxsw_sp_nexthop_ht_params = {
2816 	.key_offset = offsetof(struct mlxsw_sp_nexthop, key),
2817 	.head_offset = offsetof(struct mlxsw_sp_nexthop, ht_node),
2818 	.key_len = sizeof(struct mlxsw_sp_nexthop_key),
2819 };
2820 
2821 static int mlxsw_sp_nexthop_insert(struct mlxsw_sp *mlxsw_sp,
2822 				   struct mlxsw_sp_nexthop *nh)
2823 {
2824 	return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_ht,
2825 				      &nh->ht_node, mlxsw_sp_nexthop_ht_params);
2826 }
2827 
2828 static void mlxsw_sp_nexthop_remove(struct mlxsw_sp *mlxsw_sp,
2829 				    struct mlxsw_sp_nexthop *nh)
2830 {
2831 	rhashtable_remove_fast(&mlxsw_sp->router->nexthop_ht, &nh->ht_node,
2832 			       mlxsw_sp_nexthop_ht_params);
2833 }
2834 
2835 static struct mlxsw_sp_nexthop *
2836 mlxsw_sp_nexthop_lookup(struct mlxsw_sp *mlxsw_sp,
2837 			struct mlxsw_sp_nexthop_key key)
2838 {
2839 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_ht, &key,
2840 				      mlxsw_sp_nexthop_ht_params);
2841 }
2842 
2843 static int mlxsw_sp_adj_index_mass_update_vr(struct mlxsw_sp *mlxsw_sp,
2844 					     const struct mlxsw_sp_fib *fib,
2845 					     u32 adj_index, u16 ecmp_size,
2846 					     u32 new_adj_index,
2847 					     u16 new_ecmp_size)
2848 {
2849 	char raleu_pl[MLXSW_REG_RALEU_LEN];
2850 
2851 	mlxsw_reg_raleu_pack(raleu_pl,
2852 			     (enum mlxsw_reg_ralxx_protocol) fib->proto,
2853 			     fib->vr->id, adj_index, ecmp_size, new_adj_index,
2854 			     new_ecmp_size);
2855 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raleu), raleu_pl);
2856 }
2857 
2858 static int mlxsw_sp_adj_index_mass_update(struct mlxsw_sp *mlxsw_sp,
2859 					  struct mlxsw_sp_nexthop_group *nh_grp,
2860 					  u32 old_adj_index, u16 old_ecmp_size)
2861 {
2862 	struct mlxsw_sp_fib_entry *fib_entry;
2863 	struct mlxsw_sp_fib *fib = NULL;
2864 	int err;
2865 
2866 	list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
2867 		if (fib == fib_entry->fib_node->fib)
2868 			continue;
2869 		fib = fib_entry->fib_node->fib;
2870 		err = mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp, fib,
2871 							old_adj_index,
2872 							old_ecmp_size,
2873 							nh_grp->adj_index,
2874 							nh_grp->ecmp_size);
2875 		if (err)
2876 			return err;
2877 	}
2878 	return 0;
2879 }
2880 
2881 static int __mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
2882 				     struct mlxsw_sp_nexthop *nh)
2883 {
2884 	struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
2885 	char ratr_pl[MLXSW_REG_RATR_LEN];
2886 
2887 	mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
2888 			    true, MLXSW_REG_RATR_TYPE_ETHERNET,
2889 			    adj_index, neigh_entry->rif);
2890 	mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
2891 	if (nh->counter_valid)
2892 		mlxsw_reg_ratr_counter_pack(ratr_pl, nh->counter_index, true);
2893 	else
2894 		mlxsw_reg_ratr_counter_pack(ratr_pl, 0, false);
2895 
2896 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
2897 }
2898 
2899 int mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
2900 			    struct mlxsw_sp_nexthop *nh)
2901 {
2902 	int i;
2903 
2904 	for (i = 0; i < nh->num_adj_entries; i++) {
2905 		int err;
2906 
2907 		err = __mlxsw_sp_nexthop_update(mlxsw_sp, adj_index + i, nh);
2908 		if (err)
2909 			return err;
2910 	}
2911 
2912 	return 0;
2913 }
2914 
2915 static int __mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
2916 					  u32 adj_index,
2917 					  struct mlxsw_sp_nexthop *nh)
2918 {
2919 	const struct mlxsw_sp_ipip_ops *ipip_ops;
2920 
2921 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[nh->ipip_entry->ipipt];
2922 	return ipip_ops->nexthop_update(mlxsw_sp, adj_index, nh->ipip_entry);
2923 }
2924 
2925 static int mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
2926 					u32 adj_index,
2927 					struct mlxsw_sp_nexthop *nh)
2928 {
2929 	int i;
2930 
2931 	for (i = 0; i < nh->num_adj_entries; i++) {
2932 		int err;
2933 
2934 		err = __mlxsw_sp_nexthop_ipip_update(mlxsw_sp, adj_index + i,
2935 						     nh);
2936 		if (err)
2937 			return err;
2938 	}
2939 
2940 	return 0;
2941 }
2942 
2943 static int
2944 mlxsw_sp_nexthop_group_update(struct mlxsw_sp *mlxsw_sp,
2945 			      struct mlxsw_sp_nexthop_group *nh_grp,
2946 			      bool reallocate)
2947 {
2948 	u32 adj_index = nh_grp->adj_index; /* base */
2949 	struct mlxsw_sp_nexthop *nh;
2950 	int i;
2951 	int err;
2952 
2953 	for (i = 0; i < nh_grp->count; i++) {
2954 		nh = &nh_grp->nexthops[i];
2955 
2956 		if (!nh->should_offload) {
2957 			nh->offloaded = 0;
2958 			continue;
2959 		}
2960 
2961 		if (nh->update || reallocate) {
2962 			switch (nh->type) {
2963 			case MLXSW_SP_NEXTHOP_TYPE_ETH:
2964 				err = mlxsw_sp_nexthop_update
2965 					    (mlxsw_sp, adj_index, nh);
2966 				break;
2967 			case MLXSW_SP_NEXTHOP_TYPE_IPIP:
2968 				err = mlxsw_sp_nexthop_ipip_update
2969 					    (mlxsw_sp, adj_index, nh);
2970 				break;
2971 			}
2972 			if (err)
2973 				return err;
2974 			nh->update = 0;
2975 			nh->offloaded = 1;
2976 		}
2977 		adj_index += nh->num_adj_entries;
2978 	}
2979 	return 0;
2980 }
2981 
2982 static bool
2983 mlxsw_sp_fib_node_entry_is_first(const struct mlxsw_sp_fib_node *fib_node,
2984 				 const struct mlxsw_sp_fib_entry *fib_entry);
2985 
2986 static int
2987 mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
2988 				    struct mlxsw_sp_nexthop_group *nh_grp)
2989 {
2990 	struct mlxsw_sp_fib_entry *fib_entry;
2991 	int err;
2992 
2993 	list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
2994 		if (!mlxsw_sp_fib_node_entry_is_first(fib_entry->fib_node,
2995 						      fib_entry))
2996 			continue;
2997 		err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
2998 		if (err)
2999 			return err;
3000 	}
3001 	return 0;
3002 }
3003 
3004 static void
3005 mlxsw_sp_fib_entry_offload_refresh(struct mlxsw_sp_fib_entry *fib_entry,
3006 				   enum mlxsw_reg_ralue_op op, int err);
3007 
3008 static void
3009 mlxsw_sp_nexthop_fib_entries_refresh(struct mlxsw_sp_nexthop_group *nh_grp)
3010 {
3011 	enum mlxsw_reg_ralue_op op = MLXSW_REG_RALUE_OP_WRITE_WRITE;
3012 	struct mlxsw_sp_fib_entry *fib_entry;
3013 
3014 	list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
3015 		if (!mlxsw_sp_fib_node_entry_is_first(fib_entry->fib_node,
3016 						      fib_entry))
3017 			continue;
3018 		mlxsw_sp_fib_entry_offload_refresh(fib_entry, op, 0);
3019 	}
3020 }
3021 
3022 static void mlxsw_sp_adj_grp_size_round_up(u16 *p_adj_grp_size)
3023 {
3024 	/* Valid sizes for an adjacency group are:
3025 	 * 1-64, 512, 1024, 2048 and 4096.
3026 	 */
3027 	if (*p_adj_grp_size <= 64)
3028 		return;
3029 	else if (*p_adj_grp_size <= 512)
3030 		*p_adj_grp_size = 512;
3031 	else if (*p_adj_grp_size <= 1024)
3032 		*p_adj_grp_size = 1024;
3033 	else if (*p_adj_grp_size <= 2048)
3034 		*p_adj_grp_size = 2048;
3035 	else
3036 		*p_adj_grp_size = 4096;
3037 }
3038 
3039 static void mlxsw_sp_adj_grp_size_round_down(u16 *p_adj_grp_size,
3040 					     unsigned int alloc_size)
3041 {
3042 	if (alloc_size >= 4096)
3043 		*p_adj_grp_size = 4096;
3044 	else if (alloc_size >= 2048)
3045 		*p_adj_grp_size = 2048;
3046 	else if (alloc_size >= 1024)
3047 		*p_adj_grp_size = 1024;
3048 	else if (alloc_size >= 512)
3049 		*p_adj_grp_size = 512;
3050 }
3051 
3052 static int mlxsw_sp_fix_adj_grp_size(struct mlxsw_sp *mlxsw_sp,
3053 				     u16 *p_adj_grp_size)
3054 {
3055 	unsigned int alloc_size;
3056 	int err;
3057 
3058 	/* Round up the requested group size to the next size supported
3059 	 * by the device and make sure the request can be satisfied.
3060 	 */
3061 	mlxsw_sp_adj_grp_size_round_up(p_adj_grp_size);
3062 	err = mlxsw_sp_kvdl_alloc_size_query(mlxsw_sp, *p_adj_grp_size,
3063 					     &alloc_size);
3064 	if (err)
3065 		return err;
3066 	/* It is possible the allocation results in more allocated
3067 	 * entries than requested. Try to use as much of them as
3068 	 * possible.
3069 	 */
3070 	mlxsw_sp_adj_grp_size_round_down(p_adj_grp_size, alloc_size);
3071 
3072 	return 0;
3073 }
3074 
3075 static void
3076 mlxsw_sp_nexthop_group_normalize(struct mlxsw_sp_nexthop_group *nh_grp)
3077 {
3078 	int i, g = 0, sum_norm_weight = 0;
3079 	struct mlxsw_sp_nexthop *nh;
3080 
3081 	for (i = 0; i < nh_grp->count; i++) {
3082 		nh = &nh_grp->nexthops[i];
3083 
3084 		if (!nh->should_offload)
3085 			continue;
3086 		if (g > 0)
3087 			g = gcd(nh->nh_weight, g);
3088 		else
3089 			g = nh->nh_weight;
3090 	}
3091 
3092 	for (i = 0; i < nh_grp->count; i++) {
3093 		nh = &nh_grp->nexthops[i];
3094 
3095 		if (!nh->should_offload)
3096 			continue;
3097 		nh->norm_nh_weight = nh->nh_weight / g;
3098 		sum_norm_weight += nh->norm_nh_weight;
3099 	}
3100 
3101 	nh_grp->sum_norm_weight = sum_norm_weight;
3102 }
3103 
3104 static void
3105 mlxsw_sp_nexthop_group_rebalance(struct mlxsw_sp_nexthop_group *nh_grp)
3106 {
3107 	int total = nh_grp->sum_norm_weight;
3108 	u16 ecmp_size = nh_grp->ecmp_size;
3109 	int i, weight = 0, lower_bound = 0;
3110 
3111 	for (i = 0; i < nh_grp->count; i++) {
3112 		struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
3113 		int upper_bound;
3114 
3115 		if (!nh->should_offload)
3116 			continue;
3117 		weight += nh->norm_nh_weight;
3118 		upper_bound = DIV_ROUND_CLOSEST(ecmp_size * weight, total);
3119 		nh->num_adj_entries = upper_bound - lower_bound;
3120 		lower_bound = upper_bound;
3121 	}
3122 }
3123 
3124 static void
3125 mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
3126 			       struct mlxsw_sp_nexthop_group *nh_grp)
3127 {
3128 	u16 ecmp_size, old_ecmp_size;
3129 	struct mlxsw_sp_nexthop *nh;
3130 	bool offload_change = false;
3131 	u32 adj_index;
3132 	bool old_adj_index_valid;
3133 	u32 old_adj_index;
3134 	int i;
3135 	int err;
3136 
3137 	if (!nh_grp->gateway) {
3138 		mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3139 		return;
3140 	}
3141 
3142 	for (i = 0; i < nh_grp->count; i++) {
3143 		nh = &nh_grp->nexthops[i];
3144 
3145 		if (nh->should_offload != nh->offloaded) {
3146 			offload_change = true;
3147 			if (nh->should_offload)
3148 				nh->update = 1;
3149 		}
3150 	}
3151 	if (!offload_change) {
3152 		/* Nothing was added or removed, so no need to reallocate. Just
3153 		 * update MAC on existing adjacency indexes.
3154 		 */
3155 		err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nh_grp, false);
3156 		if (err) {
3157 			dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3158 			goto set_trap;
3159 		}
3160 		return;
3161 	}
3162 	mlxsw_sp_nexthop_group_normalize(nh_grp);
3163 	if (!nh_grp->sum_norm_weight)
3164 		/* No neigh of this group is connected so we just set
3165 		 * the trap and let everthing flow through kernel.
3166 		 */
3167 		goto set_trap;
3168 
3169 	ecmp_size = nh_grp->sum_norm_weight;
3170 	err = mlxsw_sp_fix_adj_grp_size(mlxsw_sp, &ecmp_size);
3171 	if (err)
3172 		/* No valid allocation size available. */
3173 		goto set_trap;
3174 
3175 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, ecmp_size, &adj_index);
3176 	if (err) {
3177 		/* We ran out of KVD linear space, just set the
3178 		 * trap and let everything flow through kernel.
3179 		 */
3180 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to allocate KVD linear area for nexthop group.\n");
3181 		goto set_trap;
3182 	}
3183 	old_adj_index_valid = nh_grp->adj_index_valid;
3184 	old_adj_index = nh_grp->adj_index;
3185 	old_ecmp_size = nh_grp->ecmp_size;
3186 	nh_grp->adj_index_valid = 1;
3187 	nh_grp->adj_index = adj_index;
3188 	nh_grp->ecmp_size = ecmp_size;
3189 	mlxsw_sp_nexthop_group_rebalance(nh_grp);
3190 	err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nh_grp, true);
3191 	if (err) {
3192 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3193 		goto set_trap;
3194 	}
3195 
3196 	if (!old_adj_index_valid) {
3197 		/* The trap was set for fib entries, so we have to call
3198 		 * fib entry update to unset it and use adjacency index.
3199 		 */
3200 		err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3201 		if (err) {
3202 			dev_warn(mlxsw_sp->bus_info->dev, "Failed to add adjacency index to fib entries.\n");
3203 			goto set_trap;
3204 		}
3205 		return;
3206 	}
3207 
3208 	err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, nh_grp,
3209 					     old_adj_index, old_ecmp_size);
3210 	mlxsw_sp_kvdl_free(mlxsw_sp, old_adj_index);
3211 	if (err) {
3212 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to mass-update adjacency index for nexthop group.\n");
3213 		goto set_trap;
3214 	}
3215 
3216 	/* Offload state within the group changed, so update the flags. */
3217 	mlxsw_sp_nexthop_fib_entries_refresh(nh_grp);
3218 
3219 	return;
3220 
3221 set_trap:
3222 	old_adj_index_valid = nh_grp->adj_index_valid;
3223 	nh_grp->adj_index_valid = 0;
3224 	for (i = 0; i < nh_grp->count; i++) {
3225 		nh = &nh_grp->nexthops[i];
3226 		nh->offloaded = 0;
3227 	}
3228 	err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3229 	if (err)
3230 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to set traps for fib entries.\n");
3231 	if (old_adj_index_valid)
3232 		mlxsw_sp_kvdl_free(mlxsw_sp, nh_grp->adj_index);
3233 }
3234 
3235 static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
3236 					    bool removing)
3237 {
3238 	if (!removing)
3239 		nh->should_offload = 1;
3240 	else if (nh->offloaded)
3241 		nh->should_offload = 0;
3242 	nh->update = 1;
3243 }
3244 
3245 static void
3246 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
3247 			      struct mlxsw_sp_neigh_entry *neigh_entry,
3248 			      bool removing)
3249 {
3250 	struct mlxsw_sp_nexthop *nh;
3251 
3252 	list_for_each_entry(nh, &neigh_entry->nexthop_list,
3253 			    neigh_list_node) {
3254 		__mlxsw_sp_nexthop_neigh_update(nh, removing);
3255 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
3256 	}
3257 }
3258 
3259 static void mlxsw_sp_nexthop_rif_init(struct mlxsw_sp_nexthop *nh,
3260 				      struct mlxsw_sp_rif *rif)
3261 {
3262 	if (nh->rif)
3263 		return;
3264 
3265 	nh->rif = rif;
3266 	list_add(&nh->rif_list_node, &rif->nexthop_list);
3267 }
3268 
3269 static void mlxsw_sp_nexthop_rif_fini(struct mlxsw_sp_nexthop *nh)
3270 {
3271 	if (!nh->rif)
3272 		return;
3273 
3274 	list_del(&nh->rif_list_node);
3275 	nh->rif = NULL;
3276 }
3277 
3278 static int mlxsw_sp_nexthop_neigh_init(struct mlxsw_sp *mlxsw_sp,
3279 				       struct mlxsw_sp_nexthop *nh)
3280 {
3281 	struct mlxsw_sp_neigh_entry *neigh_entry;
3282 	struct neighbour *n;
3283 	u8 nud_state, dead;
3284 	int err;
3285 
3286 	if (!nh->nh_grp->gateway || nh->neigh_entry)
3287 		return 0;
3288 
3289 	/* Take a reference of neigh here ensuring that neigh would
3290 	 * not be destructed before the nexthop entry is finished.
3291 	 * The reference is taken either in neigh_lookup() or
3292 	 * in neigh_create() in case n is not found.
3293 	 */
3294 	n = neigh_lookup(nh->nh_grp->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3295 	if (!n) {
3296 		n = neigh_create(nh->nh_grp->neigh_tbl, &nh->gw_addr,
3297 				 nh->rif->dev);
3298 		if (IS_ERR(n))
3299 			return PTR_ERR(n);
3300 		neigh_event_send(n, NULL);
3301 	}
3302 	neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
3303 	if (!neigh_entry) {
3304 		neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
3305 		if (IS_ERR(neigh_entry)) {
3306 			err = -EINVAL;
3307 			goto err_neigh_entry_create;
3308 		}
3309 	}
3310 
3311 	/* If that is the first nexthop connected to that neigh, add to
3312 	 * nexthop_neighs_list
3313 	 */
3314 	if (list_empty(&neigh_entry->nexthop_list))
3315 		list_add_tail(&neigh_entry->nexthop_neighs_list_node,
3316 			      &mlxsw_sp->router->nexthop_neighs_list);
3317 
3318 	nh->neigh_entry = neigh_entry;
3319 	list_add_tail(&nh->neigh_list_node, &neigh_entry->nexthop_list);
3320 	read_lock_bh(&n->lock);
3321 	nud_state = n->nud_state;
3322 	dead = n->dead;
3323 	read_unlock_bh(&n->lock);
3324 	__mlxsw_sp_nexthop_neigh_update(nh, !(nud_state & NUD_VALID && !dead));
3325 
3326 	return 0;
3327 
3328 err_neigh_entry_create:
3329 	neigh_release(n);
3330 	return err;
3331 }
3332 
3333 static void mlxsw_sp_nexthop_neigh_fini(struct mlxsw_sp *mlxsw_sp,
3334 					struct mlxsw_sp_nexthop *nh)
3335 {
3336 	struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
3337 	struct neighbour *n;
3338 
3339 	if (!neigh_entry)
3340 		return;
3341 	n = neigh_entry->key.n;
3342 
3343 	__mlxsw_sp_nexthop_neigh_update(nh, true);
3344 	list_del(&nh->neigh_list_node);
3345 	nh->neigh_entry = NULL;
3346 
3347 	/* If that is the last nexthop connected to that neigh, remove from
3348 	 * nexthop_neighs_list
3349 	 */
3350 	if (list_empty(&neigh_entry->nexthop_list))
3351 		list_del(&neigh_entry->nexthop_neighs_list_node);
3352 
3353 	if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
3354 		mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
3355 
3356 	neigh_release(n);
3357 }
3358 
3359 static bool mlxsw_sp_ipip_netdev_ul_up(struct net_device *ol_dev)
3360 {
3361 	struct net_device *ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
3362 
3363 	return ul_dev ? (ul_dev->flags & IFF_UP) : true;
3364 }
3365 
3366 static void mlxsw_sp_nexthop_ipip_init(struct mlxsw_sp *mlxsw_sp,
3367 				       struct mlxsw_sp_nexthop *nh,
3368 				       struct mlxsw_sp_ipip_entry *ipip_entry)
3369 {
3370 	bool removing;
3371 
3372 	if (!nh->nh_grp->gateway || nh->ipip_entry)
3373 		return;
3374 
3375 	nh->ipip_entry = ipip_entry;
3376 	removing = !mlxsw_sp_ipip_netdev_ul_up(ipip_entry->ol_dev);
3377 	__mlxsw_sp_nexthop_neigh_update(nh, removing);
3378 	mlxsw_sp_nexthop_rif_init(nh, &ipip_entry->ol_lb->common);
3379 }
3380 
3381 static void mlxsw_sp_nexthop_ipip_fini(struct mlxsw_sp *mlxsw_sp,
3382 				       struct mlxsw_sp_nexthop *nh)
3383 {
3384 	struct mlxsw_sp_ipip_entry *ipip_entry = nh->ipip_entry;
3385 
3386 	if (!ipip_entry)
3387 		return;
3388 
3389 	__mlxsw_sp_nexthop_neigh_update(nh, true);
3390 	nh->ipip_entry = NULL;
3391 }
3392 
3393 static bool mlxsw_sp_nexthop4_ipip_type(const struct mlxsw_sp *mlxsw_sp,
3394 					const struct fib_nh *fib_nh,
3395 					enum mlxsw_sp_ipip_type *p_ipipt)
3396 {
3397 	struct net_device *dev = fib_nh->nh_dev;
3398 
3399 	return dev &&
3400 	       fib_nh->nh_parent->fib_type == RTN_UNICAST &&
3401 	       mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, p_ipipt);
3402 }
3403 
3404 static void mlxsw_sp_nexthop_type_fini(struct mlxsw_sp *mlxsw_sp,
3405 				       struct mlxsw_sp_nexthop *nh)
3406 {
3407 	switch (nh->type) {
3408 	case MLXSW_SP_NEXTHOP_TYPE_ETH:
3409 		mlxsw_sp_nexthop_neigh_fini(mlxsw_sp, nh);
3410 		mlxsw_sp_nexthop_rif_fini(nh);
3411 		break;
3412 	case MLXSW_SP_NEXTHOP_TYPE_IPIP:
3413 		mlxsw_sp_nexthop_rif_fini(nh);
3414 		mlxsw_sp_nexthop_ipip_fini(mlxsw_sp, nh);
3415 		break;
3416 	}
3417 }
3418 
3419 static int mlxsw_sp_nexthop4_type_init(struct mlxsw_sp *mlxsw_sp,
3420 				       struct mlxsw_sp_nexthop *nh,
3421 				       struct fib_nh *fib_nh)
3422 {
3423 	const struct mlxsw_sp_ipip_ops *ipip_ops;
3424 	struct net_device *dev = fib_nh->nh_dev;
3425 	struct mlxsw_sp_ipip_entry *ipip_entry;
3426 	struct mlxsw_sp_rif *rif;
3427 	int err;
3428 
3429 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, dev);
3430 	if (ipip_entry) {
3431 		ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
3432 		if (ipip_ops->can_offload(mlxsw_sp, dev,
3433 					  MLXSW_SP_L3_PROTO_IPV4)) {
3434 			nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
3435 			mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, ipip_entry);
3436 			return 0;
3437 		}
3438 	}
3439 
3440 	nh->type = MLXSW_SP_NEXTHOP_TYPE_ETH;
3441 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
3442 	if (!rif)
3443 		return 0;
3444 
3445 	mlxsw_sp_nexthop_rif_init(nh, rif);
3446 	err = mlxsw_sp_nexthop_neigh_init(mlxsw_sp, nh);
3447 	if (err)
3448 		goto err_neigh_init;
3449 
3450 	return 0;
3451 
3452 err_neigh_init:
3453 	mlxsw_sp_nexthop_rif_fini(nh);
3454 	return err;
3455 }
3456 
3457 static void mlxsw_sp_nexthop4_type_fini(struct mlxsw_sp *mlxsw_sp,
3458 					struct mlxsw_sp_nexthop *nh)
3459 {
3460 	mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
3461 }
3462 
3463 static int mlxsw_sp_nexthop4_init(struct mlxsw_sp *mlxsw_sp,
3464 				  struct mlxsw_sp_nexthop_group *nh_grp,
3465 				  struct mlxsw_sp_nexthop *nh,
3466 				  struct fib_nh *fib_nh)
3467 {
3468 	struct net_device *dev = fib_nh->nh_dev;
3469 	struct in_device *in_dev;
3470 	int err;
3471 
3472 	nh->nh_grp = nh_grp;
3473 	nh->key.fib_nh = fib_nh;
3474 #ifdef CONFIG_IP_ROUTE_MULTIPATH
3475 	nh->nh_weight = fib_nh->nh_weight;
3476 #else
3477 	nh->nh_weight = 1;
3478 #endif
3479 	memcpy(&nh->gw_addr, &fib_nh->nh_gw, sizeof(fib_nh->nh_gw));
3480 	err = mlxsw_sp_nexthop_insert(mlxsw_sp, nh);
3481 	if (err)
3482 		return err;
3483 
3484 	mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
3485 	list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
3486 
3487 	if (!dev)
3488 		return 0;
3489 
3490 	in_dev = __in_dev_get_rtnl(dev);
3491 	if (in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
3492 	    fib_nh->nh_flags & RTNH_F_LINKDOWN)
3493 		return 0;
3494 
3495 	err = mlxsw_sp_nexthop4_type_init(mlxsw_sp, nh, fib_nh);
3496 	if (err)
3497 		goto err_nexthop_neigh_init;
3498 
3499 	return 0;
3500 
3501 err_nexthop_neigh_init:
3502 	mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
3503 	return err;
3504 }
3505 
3506 static void mlxsw_sp_nexthop4_fini(struct mlxsw_sp *mlxsw_sp,
3507 				   struct mlxsw_sp_nexthop *nh)
3508 {
3509 	mlxsw_sp_nexthop4_type_fini(mlxsw_sp, nh);
3510 	list_del(&nh->router_list_node);
3511 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
3512 	mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
3513 }
3514 
3515 static void mlxsw_sp_nexthop4_event(struct mlxsw_sp *mlxsw_sp,
3516 				    unsigned long event, struct fib_nh *fib_nh)
3517 {
3518 	struct mlxsw_sp_nexthop_key key;
3519 	struct mlxsw_sp_nexthop *nh;
3520 
3521 	if (mlxsw_sp->router->aborted)
3522 		return;
3523 
3524 	key.fib_nh = fib_nh;
3525 	nh = mlxsw_sp_nexthop_lookup(mlxsw_sp, key);
3526 	if (WARN_ON_ONCE(!nh))
3527 		return;
3528 
3529 	switch (event) {
3530 	case FIB_EVENT_NH_ADD:
3531 		mlxsw_sp_nexthop4_type_init(mlxsw_sp, nh, fib_nh);
3532 		break;
3533 	case FIB_EVENT_NH_DEL:
3534 		mlxsw_sp_nexthop4_type_fini(mlxsw_sp, nh);
3535 		break;
3536 	}
3537 
3538 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
3539 }
3540 
3541 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
3542 					struct mlxsw_sp_rif *rif)
3543 {
3544 	struct mlxsw_sp_nexthop *nh;
3545 	bool removing;
3546 
3547 	list_for_each_entry(nh, &rif->nexthop_list, rif_list_node) {
3548 		switch (nh->type) {
3549 		case MLXSW_SP_NEXTHOP_TYPE_ETH:
3550 			removing = false;
3551 			break;
3552 		case MLXSW_SP_NEXTHOP_TYPE_IPIP:
3553 			removing = !mlxsw_sp_ipip_netdev_ul_up(rif->dev);
3554 			break;
3555 		default:
3556 			WARN_ON(1);
3557 			continue;
3558 		}
3559 
3560 		__mlxsw_sp_nexthop_neigh_update(nh, removing);
3561 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
3562 	}
3563 }
3564 
3565 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
3566 					 struct mlxsw_sp_rif *old_rif,
3567 					 struct mlxsw_sp_rif *new_rif)
3568 {
3569 	struct mlxsw_sp_nexthop *nh;
3570 
3571 	list_splice_init(&old_rif->nexthop_list, &new_rif->nexthop_list);
3572 	list_for_each_entry(nh, &new_rif->nexthop_list, rif_list_node)
3573 		nh->rif = new_rif;
3574 	mlxsw_sp_nexthop_rif_update(mlxsw_sp, new_rif);
3575 }
3576 
3577 static void mlxsw_sp_nexthop_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
3578 					   struct mlxsw_sp_rif *rif)
3579 {
3580 	struct mlxsw_sp_nexthop *nh, *tmp;
3581 
3582 	list_for_each_entry_safe(nh, tmp, &rif->nexthop_list, rif_list_node) {
3583 		mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
3584 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
3585 	}
3586 }
3587 
3588 static bool mlxsw_sp_fi_is_gateway(const struct mlxsw_sp *mlxsw_sp,
3589 				   const struct fib_info *fi)
3590 {
3591 	return fi->fib_nh->nh_scope == RT_SCOPE_LINK ||
3592 	       mlxsw_sp_nexthop4_ipip_type(mlxsw_sp, fi->fib_nh, NULL);
3593 }
3594 
3595 static struct mlxsw_sp_nexthop_group *
3596 mlxsw_sp_nexthop4_group_create(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
3597 {
3598 	struct mlxsw_sp_nexthop_group *nh_grp;
3599 	struct mlxsw_sp_nexthop *nh;
3600 	struct fib_nh *fib_nh;
3601 	size_t alloc_size;
3602 	int i;
3603 	int err;
3604 
3605 	alloc_size = sizeof(*nh_grp) +
3606 		     fi->fib_nhs * sizeof(struct mlxsw_sp_nexthop);
3607 	nh_grp = kzalloc(alloc_size, GFP_KERNEL);
3608 	if (!nh_grp)
3609 		return ERR_PTR(-ENOMEM);
3610 	nh_grp->priv = fi;
3611 	INIT_LIST_HEAD(&nh_grp->fib_list);
3612 	nh_grp->neigh_tbl = &arp_tbl;
3613 
3614 	nh_grp->gateway = mlxsw_sp_fi_is_gateway(mlxsw_sp, fi);
3615 	nh_grp->count = fi->fib_nhs;
3616 	fib_info_hold(fi);
3617 	for (i = 0; i < nh_grp->count; i++) {
3618 		nh = &nh_grp->nexthops[i];
3619 		fib_nh = &fi->fib_nh[i];
3620 		err = mlxsw_sp_nexthop4_init(mlxsw_sp, nh_grp, nh, fib_nh);
3621 		if (err)
3622 			goto err_nexthop4_init;
3623 	}
3624 	err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
3625 	if (err)
3626 		goto err_nexthop_group_insert;
3627 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
3628 	return nh_grp;
3629 
3630 err_nexthop_group_insert:
3631 err_nexthop4_init:
3632 	for (i--; i >= 0; i--) {
3633 		nh = &nh_grp->nexthops[i];
3634 		mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
3635 	}
3636 	fib_info_put(fi);
3637 	kfree(nh_grp);
3638 	return ERR_PTR(err);
3639 }
3640 
3641 static void
3642 mlxsw_sp_nexthop4_group_destroy(struct mlxsw_sp *mlxsw_sp,
3643 				struct mlxsw_sp_nexthop_group *nh_grp)
3644 {
3645 	struct mlxsw_sp_nexthop *nh;
3646 	int i;
3647 
3648 	mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
3649 	for (i = 0; i < nh_grp->count; i++) {
3650 		nh = &nh_grp->nexthops[i];
3651 		mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
3652 	}
3653 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
3654 	WARN_ON_ONCE(nh_grp->adj_index_valid);
3655 	fib_info_put(mlxsw_sp_nexthop4_group_fi(nh_grp));
3656 	kfree(nh_grp);
3657 }
3658 
3659 static int mlxsw_sp_nexthop4_group_get(struct mlxsw_sp *mlxsw_sp,
3660 				       struct mlxsw_sp_fib_entry *fib_entry,
3661 				       struct fib_info *fi)
3662 {
3663 	struct mlxsw_sp_nexthop_group *nh_grp;
3664 
3665 	nh_grp = mlxsw_sp_nexthop4_group_lookup(mlxsw_sp, fi);
3666 	if (!nh_grp) {
3667 		nh_grp = mlxsw_sp_nexthop4_group_create(mlxsw_sp, fi);
3668 		if (IS_ERR(nh_grp))
3669 			return PTR_ERR(nh_grp);
3670 	}
3671 	list_add_tail(&fib_entry->nexthop_group_node, &nh_grp->fib_list);
3672 	fib_entry->nh_group = nh_grp;
3673 	return 0;
3674 }
3675 
3676 static void mlxsw_sp_nexthop4_group_put(struct mlxsw_sp *mlxsw_sp,
3677 					struct mlxsw_sp_fib_entry *fib_entry)
3678 {
3679 	struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
3680 
3681 	list_del(&fib_entry->nexthop_group_node);
3682 	if (!list_empty(&nh_grp->fib_list))
3683 		return;
3684 	mlxsw_sp_nexthop4_group_destroy(mlxsw_sp, nh_grp);
3685 }
3686 
3687 static bool
3688 mlxsw_sp_fib4_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
3689 {
3690 	struct mlxsw_sp_fib4_entry *fib4_entry;
3691 
3692 	fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
3693 				  common);
3694 	return !fib4_entry->tos;
3695 }
3696 
3697 static bool
3698 mlxsw_sp_fib_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
3699 {
3700 	struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
3701 
3702 	switch (fib_entry->fib_node->fib->proto) {
3703 	case MLXSW_SP_L3_PROTO_IPV4:
3704 		if (!mlxsw_sp_fib4_entry_should_offload(fib_entry))
3705 			return false;
3706 		break;
3707 	case MLXSW_SP_L3_PROTO_IPV6:
3708 		break;
3709 	}
3710 
3711 	switch (fib_entry->type) {
3712 	case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
3713 		return !!nh_group->adj_index_valid;
3714 	case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
3715 		return !!nh_group->nh_rif;
3716 	case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
3717 		return true;
3718 	default:
3719 		return false;
3720 	}
3721 }
3722 
3723 static struct mlxsw_sp_nexthop *
3724 mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
3725 		     const struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
3726 {
3727 	int i;
3728 
3729 	for (i = 0; i < nh_grp->count; i++) {
3730 		struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
3731 		struct rt6_info *rt = mlxsw_sp_rt6->rt;
3732 
3733 		if (nh->rif && nh->rif->dev == rt->dst.dev &&
3734 		    ipv6_addr_equal((const struct in6_addr *) &nh->gw_addr,
3735 				    &rt->rt6i_gateway))
3736 			return nh;
3737 		continue;
3738 	}
3739 
3740 	return NULL;
3741 }
3742 
3743 static void
3744 mlxsw_sp_fib4_entry_offload_set(struct mlxsw_sp_fib_entry *fib_entry)
3745 {
3746 	struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
3747 	int i;
3748 
3749 	if (fib_entry->type == MLXSW_SP_FIB_ENTRY_TYPE_LOCAL ||
3750 	    fib_entry->type == MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP) {
3751 		nh_grp->nexthops->key.fib_nh->nh_flags |= RTNH_F_OFFLOAD;
3752 		return;
3753 	}
3754 
3755 	for (i = 0; i < nh_grp->count; i++) {
3756 		struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
3757 
3758 		if (nh->offloaded)
3759 			nh->key.fib_nh->nh_flags |= RTNH_F_OFFLOAD;
3760 		else
3761 			nh->key.fib_nh->nh_flags &= ~RTNH_F_OFFLOAD;
3762 	}
3763 }
3764 
3765 static void
3766 mlxsw_sp_fib4_entry_offload_unset(struct mlxsw_sp_fib_entry *fib_entry)
3767 {
3768 	struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
3769 	int i;
3770 
3771 	for (i = 0; i < nh_grp->count; i++) {
3772 		struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
3773 
3774 		nh->key.fib_nh->nh_flags &= ~RTNH_F_OFFLOAD;
3775 	}
3776 }
3777 
3778 static void
3779 mlxsw_sp_fib6_entry_offload_set(struct mlxsw_sp_fib_entry *fib_entry)
3780 {
3781 	struct mlxsw_sp_fib6_entry *fib6_entry;
3782 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3783 
3784 	fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
3785 				  common);
3786 
3787 	if (fib_entry->type == MLXSW_SP_FIB_ENTRY_TYPE_LOCAL) {
3788 		list_first_entry(&fib6_entry->rt6_list, struct mlxsw_sp_rt6,
3789 				 list)->rt->rt6i_nh_flags |= RTNH_F_OFFLOAD;
3790 		return;
3791 	}
3792 
3793 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3794 		struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
3795 		struct mlxsw_sp_nexthop *nh;
3796 
3797 		nh = mlxsw_sp_rt6_nexthop(nh_grp, mlxsw_sp_rt6);
3798 		if (nh && nh->offloaded)
3799 			mlxsw_sp_rt6->rt->rt6i_nh_flags |= RTNH_F_OFFLOAD;
3800 		else
3801 			mlxsw_sp_rt6->rt->rt6i_nh_flags &= ~RTNH_F_OFFLOAD;
3802 	}
3803 }
3804 
3805 static void
3806 mlxsw_sp_fib6_entry_offload_unset(struct mlxsw_sp_fib_entry *fib_entry)
3807 {
3808 	struct mlxsw_sp_fib6_entry *fib6_entry;
3809 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3810 
3811 	fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
3812 				  common);
3813 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3814 		struct rt6_info *rt = mlxsw_sp_rt6->rt;
3815 
3816 		rt->rt6i_nh_flags &= ~RTNH_F_OFFLOAD;
3817 	}
3818 }
3819 
3820 static void mlxsw_sp_fib_entry_offload_set(struct mlxsw_sp_fib_entry *fib_entry)
3821 {
3822 	switch (fib_entry->fib_node->fib->proto) {
3823 	case MLXSW_SP_L3_PROTO_IPV4:
3824 		mlxsw_sp_fib4_entry_offload_set(fib_entry);
3825 		break;
3826 	case MLXSW_SP_L3_PROTO_IPV6:
3827 		mlxsw_sp_fib6_entry_offload_set(fib_entry);
3828 		break;
3829 	}
3830 }
3831 
3832 static void
3833 mlxsw_sp_fib_entry_offload_unset(struct mlxsw_sp_fib_entry *fib_entry)
3834 {
3835 	switch (fib_entry->fib_node->fib->proto) {
3836 	case MLXSW_SP_L3_PROTO_IPV4:
3837 		mlxsw_sp_fib4_entry_offload_unset(fib_entry);
3838 		break;
3839 	case MLXSW_SP_L3_PROTO_IPV6:
3840 		mlxsw_sp_fib6_entry_offload_unset(fib_entry);
3841 		break;
3842 	}
3843 }
3844 
3845 static void
3846 mlxsw_sp_fib_entry_offload_refresh(struct mlxsw_sp_fib_entry *fib_entry,
3847 				   enum mlxsw_reg_ralue_op op, int err)
3848 {
3849 	switch (op) {
3850 	case MLXSW_REG_RALUE_OP_WRITE_DELETE:
3851 		return mlxsw_sp_fib_entry_offload_unset(fib_entry);
3852 	case MLXSW_REG_RALUE_OP_WRITE_WRITE:
3853 		if (err)
3854 			return;
3855 		if (mlxsw_sp_fib_entry_should_offload(fib_entry))
3856 			mlxsw_sp_fib_entry_offload_set(fib_entry);
3857 		else
3858 			mlxsw_sp_fib_entry_offload_unset(fib_entry);
3859 		return;
3860 	default:
3861 		return;
3862 	}
3863 }
3864 
3865 static void
3866 mlxsw_sp_fib_entry_ralue_pack(char *ralue_pl,
3867 			      const struct mlxsw_sp_fib_entry *fib_entry,
3868 			      enum mlxsw_reg_ralue_op op)
3869 {
3870 	struct mlxsw_sp_fib *fib = fib_entry->fib_node->fib;
3871 	enum mlxsw_reg_ralxx_protocol proto;
3872 	u32 *p_dip;
3873 
3874 	proto = (enum mlxsw_reg_ralxx_protocol) fib->proto;
3875 
3876 	switch (fib->proto) {
3877 	case MLXSW_SP_L3_PROTO_IPV4:
3878 		p_dip = (u32 *) fib_entry->fib_node->key.addr;
3879 		mlxsw_reg_ralue_pack4(ralue_pl, proto, op, fib->vr->id,
3880 				      fib_entry->fib_node->key.prefix_len,
3881 				      *p_dip);
3882 		break;
3883 	case MLXSW_SP_L3_PROTO_IPV6:
3884 		mlxsw_reg_ralue_pack6(ralue_pl, proto, op, fib->vr->id,
3885 				      fib_entry->fib_node->key.prefix_len,
3886 				      fib_entry->fib_node->key.addr);
3887 		break;
3888 	}
3889 }
3890 
3891 static int mlxsw_sp_fib_entry_op_remote(struct mlxsw_sp *mlxsw_sp,
3892 					struct mlxsw_sp_fib_entry *fib_entry,
3893 					enum mlxsw_reg_ralue_op op)
3894 {
3895 	char ralue_pl[MLXSW_REG_RALUE_LEN];
3896 	enum mlxsw_reg_ralue_trap_action trap_action;
3897 	u16 trap_id = 0;
3898 	u32 adjacency_index = 0;
3899 	u16 ecmp_size = 0;
3900 
3901 	/* In case the nexthop group adjacency index is valid, use it
3902 	 * with provided ECMP size. Otherwise, setup trap and pass
3903 	 * traffic to kernel.
3904 	 */
3905 	if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
3906 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
3907 		adjacency_index = fib_entry->nh_group->adj_index;
3908 		ecmp_size = fib_entry->nh_group->ecmp_size;
3909 	} else {
3910 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
3911 		trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
3912 	}
3913 
3914 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
3915 	mlxsw_reg_ralue_act_remote_pack(ralue_pl, trap_action, trap_id,
3916 					adjacency_index, ecmp_size);
3917 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
3918 }
3919 
3920 static int mlxsw_sp_fib_entry_op_local(struct mlxsw_sp *mlxsw_sp,
3921 				       struct mlxsw_sp_fib_entry *fib_entry,
3922 				       enum mlxsw_reg_ralue_op op)
3923 {
3924 	struct mlxsw_sp_rif *rif = fib_entry->nh_group->nh_rif;
3925 	enum mlxsw_reg_ralue_trap_action trap_action;
3926 	char ralue_pl[MLXSW_REG_RALUE_LEN];
3927 	u16 trap_id = 0;
3928 	u16 rif_index = 0;
3929 
3930 	if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
3931 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
3932 		rif_index = rif->rif_index;
3933 	} else {
3934 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
3935 		trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
3936 	}
3937 
3938 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
3939 	mlxsw_reg_ralue_act_local_pack(ralue_pl, trap_action, trap_id,
3940 				       rif_index);
3941 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
3942 }
3943 
3944 static int mlxsw_sp_fib_entry_op_trap(struct mlxsw_sp *mlxsw_sp,
3945 				      struct mlxsw_sp_fib_entry *fib_entry,
3946 				      enum mlxsw_reg_ralue_op op)
3947 {
3948 	char ralue_pl[MLXSW_REG_RALUE_LEN];
3949 
3950 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
3951 	mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
3952 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
3953 }
3954 
3955 static int
3956 mlxsw_sp_fib_entry_op_ipip_decap(struct mlxsw_sp *mlxsw_sp,
3957 				 struct mlxsw_sp_fib_entry *fib_entry,
3958 				 enum mlxsw_reg_ralue_op op)
3959 {
3960 	struct mlxsw_sp_ipip_entry *ipip_entry = fib_entry->decap.ipip_entry;
3961 	const struct mlxsw_sp_ipip_ops *ipip_ops;
3962 
3963 	if (WARN_ON(!ipip_entry))
3964 		return -EINVAL;
3965 
3966 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
3967 	return ipip_ops->fib_entry_op(mlxsw_sp, ipip_entry, op,
3968 				      fib_entry->decap.tunnel_index);
3969 }
3970 
3971 static int __mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
3972 				   struct mlxsw_sp_fib_entry *fib_entry,
3973 				   enum mlxsw_reg_ralue_op op)
3974 {
3975 	switch (fib_entry->type) {
3976 	case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
3977 		return mlxsw_sp_fib_entry_op_remote(mlxsw_sp, fib_entry, op);
3978 	case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
3979 		return mlxsw_sp_fib_entry_op_local(mlxsw_sp, fib_entry, op);
3980 	case MLXSW_SP_FIB_ENTRY_TYPE_TRAP:
3981 		return mlxsw_sp_fib_entry_op_trap(mlxsw_sp, fib_entry, op);
3982 	case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
3983 		return mlxsw_sp_fib_entry_op_ipip_decap(mlxsw_sp,
3984 							fib_entry, op);
3985 	}
3986 	return -EINVAL;
3987 }
3988 
3989 static int mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
3990 				 struct mlxsw_sp_fib_entry *fib_entry,
3991 				 enum mlxsw_reg_ralue_op op)
3992 {
3993 	int err = __mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry, op);
3994 
3995 	mlxsw_sp_fib_entry_offload_refresh(fib_entry, op, err);
3996 
3997 	return err;
3998 }
3999 
4000 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
4001 				     struct mlxsw_sp_fib_entry *fib_entry)
4002 {
4003 	return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
4004 				     MLXSW_REG_RALUE_OP_WRITE_WRITE);
4005 }
4006 
4007 static int mlxsw_sp_fib_entry_del(struct mlxsw_sp *mlxsw_sp,
4008 				  struct mlxsw_sp_fib_entry *fib_entry)
4009 {
4010 	return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
4011 				     MLXSW_REG_RALUE_OP_WRITE_DELETE);
4012 }
4013 
4014 static int
4015 mlxsw_sp_fib4_entry_type_set(struct mlxsw_sp *mlxsw_sp,
4016 			     const struct fib_entry_notifier_info *fen_info,
4017 			     struct mlxsw_sp_fib_entry *fib_entry)
4018 {
4019 	union mlxsw_sp_l3addr dip = { .addr4 = htonl(fen_info->dst) };
4020 	struct net_device *dev = fen_info->fi->fib_dev;
4021 	struct mlxsw_sp_ipip_entry *ipip_entry;
4022 	struct fib_info *fi = fen_info->fi;
4023 
4024 	switch (fen_info->type) {
4025 	case RTN_LOCAL:
4026 		ipip_entry = mlxsw_sp_ipip_entry_find_by_decap(mlxsw_sp, dev,
4027 						 MLXSW_SP_L3_PROTO_IPV4, dip);
4028 		if (ipip_entry && ipip_entry->ol_dev->flags & IFF_UP) {
4029 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
4030 			return mlxsw_sp_fib_entry_decap_init(mlxsw_sp,
4031 							     fib_entry,
4032 							     ipip_entry);
4033 		}
4034 		/* fall through */
4035 	case RTN_BROADCAST:
4036 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
4037 		return 0;
4038 	case RTN_UNREACHABLE: /* fall through */
4039 	case RTN_BLACKHOLE: /* fall through */
4040 	case RTN_PROHIBIT:
4041 		/* Packets hitting these routes need to be trapped, but
4042 		 * can do so with a lower priority than packets directed
4043 		 * at the host, so use action type local instead of trap.
4044 		 */
4045 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
4046 		return 0;
4047 	case RTN_UNICAST:
4048 		if (mlxsw_sp_fi_is_gateway(mlxsw_sp, fi))
4049 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
4050 		else
4051 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
4052 		return 0;
4053 	default:
4054 		return -EINVAL;
4055 	}
4056 }
4057 
4058 static struct mlxsw_sp_fib4_entry *
4059 mlxsw_sp_fib4_entry_create(struct mlxsw_sp *mlxsw_sp,
4060 			   struct mlxsw_sp_fib_node *fib_node,
4061 			   const struct fib_entry_notifier_info *fen_info)
4062 {
4063 	struct mlxsw_sp_fib4_entry *fib4_entry;
4064 	struct mlxsw_sp_fib_entry *fib_entry;
4065 	int err;
4066 
4067 	fib4_entry = kzalloc(sizeof(*fib4_entry), GFP_KERNEL);
4068 	if (!fib4_entry)
4069 		return ERR_PTR(-ENOMEM);
4070 	fib_entry = &fib4_entry->common;
4071 
4072 	err = mlxsw_sp_fib4_entry_type_set(mlxsw_sp, fen_info, fib_entry);
4073 	if (err)
4074 		goto err_fib4_entry_type_set;
4075 
4076 	err = mlxsw_sp_nexthop4_group_get(mlxsw_sp, fib_entry, fen_info->fi);
4077 	if (err)
4078 		goto err_nexthop4_group_get;
4079 
4080 	fib4_entry->prio = fen_info->fi->fib_priority;
4081 	fib4_entry->tb_id = fen_info->tb_id;
4082 	fib4_entry->type = fen_info->type;
4083 	fib4_entry->tos = fen_info->tos;
4084 
4085 	fib_entry->fib_node = fib_node;
4086 
4087 	return fib4_entry;
4088 
4089 err_nexthop4_group_get:
4090 err_fib4_entry_type_set:
4091 	kfree(fib4_entry);
4092 	return ERR_PTR(err);
4093 }
4094 
4095 static void mlxsw_sp_fib4_entry_destroy(struct mlxsw_sp *mlxsw_sp,
4096 					struct mlxsw_sp_fib4_entry *fib4_entry)
4097 {
4098 	mlxsw_sp_nexthop4_group_put(mlxsw_sp, &fib4_entry->common);
4099 	kfree(fib4_entry);
4100 }
4101 
4102 static struct mlxsw_sp_fib4_entry *
4103 mlxsw_sp_fib4_entry_lookup(struct mlxsw_sp *mlxsw_sp,
4104 			   const struct fib_entry_notifier_info *fen_info)
4105 {
4106 	struct mlxsw_sp_fib4_entry *fib4_entry;
4107 	struct mlxsw_sp_fib_node *fib_node;
4108 	struct mlxsw_sp_fib *fib;
4109 	struct mlxsw_sp_vr *vr;
4110 
4111 	vr = mlxsw_sp_vr_find(mlxsw_sp, fen_info->tb_id);
4112 	if (!vr)
4113 		return NULL;
4114 	fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV4);
4115 
4116 	fib_node = mlxsw_sp_fib_node_lookup(fib, &fen_info->dst,
4117 					    sizeof(fen_info->dst),
4118 					    fen_info->dst_len);
4119 	if (!fib_node)
4120 		return NULL;
4121 
4122 	list_for_each_entry(fib4_entry, &fib_node->entry_list, common.list) {
4123 		if (fib4_entry->tb_id == fen_info->tb_id &&
4124 		    fib4_entry->tos == fen_info->tos &&
4125 		    fib4_entry->type == fen_info->type &&
4126 		    mlxsw_sp_nexthop4_group_fi(fib4_entry->common.nh_group) ==
4127 		    fen_info->fi) {
4128 			return fib4_entry;
4129 		}
4130 	}
4131 
4132 	return NULL;
4133 }
4134 
4135 static const struct rhashtable_params mlxsw_sp_fib_ht_params = {
4136 	.key_offset = offsetof(struct mlxsw_sp_fib_node, key),
4137 	.head_offset = offsetof(struct mlxsw_sp_fib_node, ht_node),
4138 	.key_len = sizeof(struct mlxsw_sp_fib_key),
4139 	.automatic_shrinking = true,
4140 };
4141 
4142 static int mlxsw_sp_fib_node_insert(struct mlxsw_sp_fib *fib,
4143 				    struct mlxsw_sp_fib_node *fib_node)
4144 {
4145 	return rhashtable_insert_fast(&fib->ht, &fib_node->ht_node,
4146 				      mlxsw_sp_fib_ht_params);
4147 }
4148 
4149 static void mlxsw_sp_fib_node_remove(struct mlxsw_sp_fib *fib,
4150 				     struct mlxsw_sp_fib_node *fib_node)
4151 {
4152 	rhashtable_remove_fast(&fib->ht, &fib_node->ht_node,
4153 			       mlxsw_sp_fib_ht_params);
4154 }
4155 
4156 static struct mlxsw_sp_fib_node *
4157 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
4158 			 size_t addr_len, unsigned char prefix_len)
4159 {
4160 	struct mlxsw_sp_fib_key key;
4161 
4162 	memset(&key, 0, sizeof(key));
4163 	memcpy(key.addr, addr, addr_len);
4164 	key.prefix_len = prefix_len;
4165 	return rhashtable_lookup_fast(&fib->ht, &key, mlxsw_sp_fib_ht_params);
4166 }
4167 
4168 static struct mlxsw_sp_fib_node *
4169 mlxsw_sp_fib_node_create(struct mlxsw_sp_fib *fib, const void *addr,
4170 			 size_t addr_len, unsigned char prefix_len)
4171 {
4172 	struct mlxsw_sp_fib_node *fib_node;
4173 
4174 	fib_node = kzalloc(sizeof(*fib_node), GFP_KERNEL);
4175 	if (!fib_node)
4176 		return NULL;
4177 
4178 	INIT_LIST_HEAD(&fib_node->entry_list);
4179 	list_add(&fib_node->list, &fib->node_list);
4180 	memcpy(fib_node->key.addr, addr, addr_len);
4181 	fib_node->key.prefix_len = prefix_len;
4182 
4183 	return fib_node;
4184 }
4185 
4186 static void mlxsw_sp_fib_node_destroy(struct mlxsw_sp_fib_node *fib_node)
4187 {
4188 	list_del(&fib_node->list);
4189 	WARN_ON(!list_empty(&fib_node->entry_list));
4190 	kfree(fib_node);
4191 }
4192 
4193 static bool
4194 mlxsw_sp_fib_node_entry_is_first(const struct mlxsw_sp_fib_node *fib_node,
4195 				 const struct mlxsw_sp_fib_entry *fib_entry)
4196 {
4197 	return list_first_entry(&fib_node->entry_list,
4198 				struct mlxsw_sp_fib_entry, list) == fib_entry;
4199 }
4200 
4201 static int mlxsw_sp_fib_lpm_tree_link(struct mlxsw_sp *mlxsw_sp,
4202 				      struct mlxsw_sp_fib *fib,
4203 				      struct mlxsw_sp_fib_node *fib_node)
4204 {
4205 	struct mlxsw_sp_prefix_usage req_prefix_usage = {{ 0 } };
4206 	struct mlxsw_sp_lpm_tree *lpm_tree;
4207 	int err;
4208 
4209 	/* Since the tree is shared between all virtual routers we must
4210 	 * make sure it contains all the required prefix lengths. This
4211 	 * can be computed by either adding the new prefix length to the
4212 	 * existing prefix usage of a bound tree, or by aggregating the
4213 	 * prefix lengths across all virtual routers and adding the new
4214 	 * one as well.
4215 	 */
4216 	if (fib->lpm_tree)
4217 		mlxsw_sp_prefix_usage_cpy(&req_prefix_usage,
4218 					  &fib->lpm_tree->prefix_usage);
4219 	else
4220 		mlxsw_sp_vrs_prefixes(mlxsw_sp, fib->proto, &req_prefix_usage);
4221 	mlxsw_sp_prefix_usage_set(&req_prefix_usage, fib_node->key.prefix_len);
4222 
4223 	lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
4224 					 fib->proto);
4225 	if (IS_ERR(lpm_tree))
4226 		return PTR_ERR(lpm_tree);
4227 
4228 	if (fib->lpm_tree && fib->lpm_tree->id == lpm_tree->id)
4229 		return 0;
4230 
4231 	err = mlxsw_sp_vrs_lpm_tree_replace(mlxsw_sp, fib, lpm_tree);
4232 	if (err)
4233 		return err;
4234 
4235 	return 0;
4236 }
4237 
4238 static void mlxsw_sp_fib_lpm_tree_unlink(struct mlxsw_sp *mlxsw_sp,
4239 					 struct mlxsw_sp_fib *fib)
4240 {
4241 	if (!mlxsw_sp_prefix_usage_none(&fib->prefix_usage))
4242 		return;
4243 	mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, fib);
4244 	mlxsw_sp_lpm_tree_put(mlxsw_sp, fib->lpm_tree);
4245 	fib->lpm_tree = NULL;
4246 }
4247 
4248 static void mlxsw_sp_fib_node_prefix_inc(struct mlxsw_sp_fib_node *fib_node)
4249 {
4250 	unsigned char prefix_len = fib_node->key.prefix_len;
4251 	struct mlxsw_sp_fib *fib = fib_node->fib;
4252 
4253 	if (fib->prefix_ref_count[prefix_len]++ == 0)
4254 		mlxsw_sp_prefix_usage_set(&fib->prefix_usage, prefix_len);
4255 }
4256 
4257 static void mlxsw_sp_fib_node_prefix_dec(struct mlxsw_sp_fib_node *fib_node)
4258 {
4259 	unsigned char prefix_len = fib_node->key.prefix_len;
4260 	struct mlxsw_sp_fib *fib = fib_node->fib;
4261 
4262 	if (--fib->prefix_ref_count[prefix_len] == 0)
4263 		mlxsw_sp_prefix_usage_clear(&fib->prefix_usage, prefix_len);
4264 }
4265 
4266 static int mlxsw_sp_fib_node_init(struct mlxsw_sp *mlxsw_sp,
4267 				  struct mlxsw_sp_fib_node *fib_node,
4268 				  struct mlxsw_sp_fib *fib)
4269 {
4270 	int err;
4271 
4272 	err = mlxsw_sp_fib_node_insert(fib, fib_node);
4273 	if (err)
4274 		return err;
4275 	fib_node->fib = fib;
4276 
4277 	err = mlxsw_sp_fib_lpm_tree_link(mlxsw_sp, fib, fib_node);
4278 	if (err)
4279 		goto err_fib_lpm_tree_link;
4280 
4281 	mlxsw_sp_fib_node_prefix_inc(fib_node);
4282 
4283 	return 0;
4284 
4285 err_fib_lpm_tree_link:
4286 	fib_node->fib = NULL;
4287 	mlxsw_sp_fib_node_remove(fib, fib_node);
4288 	return err;
4289 }
4290 
4291 static void mlxsw_sp_fib_node_fini(struct mlxsw_sp *mlxsw_sp,
4292 				   struct mlxsw_sp_fib_node *fib_node)
4293 {
4294 	struct mlxsw_sp_fib *fib = fib_node->fib;
4295 
4296 	mlxsw_sp_fib_node_prefix_dec(fib_node);
4297 	mlxsw_sp_fib_lpm_tree_unlink(mlxsw_sp, fib);
4298 	fib_node->fib = NULL;
4299 	mlxsw_sp_fib_node_remove(fib, fib_node);
4300 }
4301 
4302 static struct mlxsw_sp_fib_node *
4303 mlxsw_sp_fib_node_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id, const void *addr,
4304 		      size_t addr_len, unsigned char prefix_len,
4305 		      enum mlxsw_sp_l3proto proto)
4306 {
4307 	struct mlxsw_sp_fib_node *fib_node;
4308 	struct mlxsw_sp_fib *fib;
4309 	struct mlxsw_sp_vr *vr;
4310 	int err;
4311 
4312 	vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id, NULL);
4313 	if (IS_ERR(vr))
4314 		return ERR_CAST(vr);
4315 	fib = mlxsw_sp_vr_fib(vr, proto);
4316 
4317 	fib_node = mlxsw_sp_fib_node_lookup(fib, addr, addr_len, prefix_len);
4318 	if (fib_node)
4319 		return fib_node;
4320 
4321 	fib_node = mlxsw_sp_fib_node_create(fib, addr, addr_len, prefix_len);
4322 	if (!fib_node) {
4323 		err = -ENOMEM;
4324 		goto err_fib_node_create;
4325 	}
4326 
4327 	err = mlxsw_sp_fib_node_init(mlxsw_sp, fib_node, fib);
4328 	if (err)
4329 		goto err_fib_node_init;
4330 
4331 	return fib_node;
4332 
4333 err_fib_node_init:
4334 	mlxsw_sp_fib_node_destroy(fib_node);
4335 err_fib_node_create:
4336 	mlxsw_sp_vr_put(vr);
4337 	return ERR_PTR(err);
4338 }
4339 
4340 static void mlxsw_sp_fib_node_put(struct mlxsw_sp *mlxsw_sp,
4341 				  struct mlxsw_sp_fib_node *fib_node)
4342 {
4343 	struct mlxsw_sp_vr *vr = fib_node->fib->vr;
4344 
4345 	if (!list_empty(&fib_node->entry_list))
4346 		return;
4347 	mlxsw_sp_fib_node_fini(mlxsw_sp, fib_node);
4348 	mlxsw_sp_fib_node_destroy(fib_node);
4349 	mlxsw_sp_vr_put(vr);
4350 }
4351 
4352 static struct mlxsw_sp_fib4_entry *
4353 mlxsw_sp_fib4_node_entry_find(const struct mlxsw_sp_fib_node *fib_node,
4354 			      const struct mlxsw_sp_fib4_entry *new4_entry)
4355 {
4356 	struct mlxsw_sp_fib4_entry *fib4_entry;
4357 
4358 	list_for_each_entry(fib4_entry, &fib_node->entry_list, common.list) {
4359 		if (fib4_entry->tb_id > new4_entry->tb_id)
4360 			continue;
4361 		if (fib4_entry->tb_id != new4_entry->tb_id)
4362 			break;
4363 		if (fib4_entry->tos > new4_entry->tos)
4364 			continue;
4365 		if (fib4_entry->prio >= new4_entry->prio ||
4366 		    fib4_entry->tos < new4_entry->tos)
4367 			return fib4_entry;
4368 	}
4369 
4370 	return NULL;
4371 }
4372 
4373 static int
4374 mlxsw_sp_fib4_node_list_append(struct mlxsw_sp_fib4_entry *fib4_entry,
4375 			       struct mlxsw_sp_fib4_entry *new4_entry)
4376 {
4377 	struct mlxsw_sp_fib_node *fib_node;
4378 
4379 	if (WARN_ON(!fib4_entry))
4380 		return -EINVAL;
4381 
4382 	fib_node = fib4_entry->common.fib_node;
4383 	list_for_each_entry_from(fib4_entry, &fib_node->entry_list,
4384 				 common.list) {
4385 		if (fib4_entry->tb_id != new4_entry->tb_id ||
4386 		    fib4_entry->tos != new4_entry->tos ||
4387 		    fib4_entry->prio != new4_entry->prio)
4388 			break;
4389 	}
4390 
4391 	list_add_tail(&new4_entry->common.list, &fib4_entry->common.list);
4392 	return 0;
4393 }
4394 
4395 static int
4396 mlxsw_sp_fib4_node_list_insert(struct mlxsw_sp_fib4_entry *new4_entry,
4397 			       bool replace, bool append)
4398 {
4399 	struct mlxsw_sp_fib_node *fib_node = new4_entry->common.fib_node;
4400 	struct mlxsw_sp_fib4_entry *fib4_entry;
4401 
4402 	fib4_entry = mlxsw_sp_fib4_node_entry_find(fib_node, new4_entry);
4403 
4404 	if (append)
4405 		return mlxsw_sp_fib4_node_list_append(fib4_entry, new4_entry);
4406 	if (replace && WARN_ON(!fib4_entry))
4407 		return -EINVAL;
4408 
4409 	/* Insert new entry before replaced one, so that we can later
4410 	 * remove the second.
4411 	 */
4412 	if (fib4_entry) {
4413 		list_add_tail(&new4_entry->common.list,
4414 			      &fib4_entry->common.list);
4415 	} else {
4416 		struct mlxsw_sp_fib4_entry *last;
4417 
4418 		list_for_each_entry(last, &fib_node->entry_list, common.list) {
4419 			if (new4_entry->tb_id > last->tb_id)
4420 				break;
4421 			fib4_entry = last;
4422 		}
4423 
4424 		if (fib4_entry)
4425 			list_add(&new4_entry->common.list,
4426 				 &fib4_entry->common.list);
4427 		else
4428 			list_add(&new4_entry->common.list,
4429 				 &fib_node->entry_list);
4430 	}
4431 
4432 	return 0;
4433 }
4434 
4435 static void
4436 mlxsw_sp_fib4_node_list_remove(struct mlxsw_sp_fib4_entry *fib4_entry)
4437 {
4438 	list_del(&fib4_entry->common.list);
4439 }
4440 
4441 static int mlxsw_sp_fib_node_entry_add(struct mlxsw_sp *mlxsw_sp,
4442 				       struct mlxsw_sp_fib_entry *fib_entry)
4443 {
4444 	struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
4445 
4446 	if (!mlxsw_sp_fib_node_entry_is_first(fib_node, fib_entry))
4447 		return 0;
4448 
4449 	/* To prevent packet loss, overwrite the previously offloaded
4450 	 * entry.
4451 	 */
4452 	if (!list_is_singular(&fib_node->entry_list)) {
4453 		enum mlxsw_reg_ralue_op op = MLXSW_REG_RALUE_OP_WRITE_DELETE;
4454 		struct mlxsw_sp_fib_entry *n = list_next_entry(fib_entry, list);
4455 
4456 		mlxsw_sp_fib_entry_offload_refresh(n, op, 0);
4457 	}
4458 
4459 	return mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
4460 }
4461 
4462 static void mlxsw_sp_fib_node_entry_del(struct mlxsw_sp *mlxsw_sp,
4463 					struct mlxsw_sp_fib_entry *fib_entry)
4464 {
4465 	struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
4466 
4467 	if (!mlxsw_sp_fib_node_entry_is_first(fib_node, fib_entry))
4468 		return;
4469 
4470 	/* Promote the next entry by overwriting the deleted entry */
4471 	if (!list_is_singular(&fib_node->entry_list)) {
4472 		struct mlxsw_sp_fib_entry *n = list_next_entry(fib_entry, list);
4473 		enum mlxsw_reg_ralue_op op = MLXSW_REG_RALUE_OP_WRITE_DELETE;
4474 
4475 		mlxsw_sp_fib_entry_update(mlxsw_sp, n);
4476 		mlxsw_sp_fib_entry_offload_refresh(fib_entry, op, 0);
4477 		return;
4478 	}
4479 
4480 	mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
4481 }
4482 
4483 static int mlxsw_sp_fib4_node_entry_link(struct mlxsw_sp *mlxsw_sp,
4484 					 struct mlxsw_sp_fib4_entry *fib4_entry,
4485 					 bool replace, bool append)
4486 {
4487 	int err;
4488 
4489 	err = mlxsw_sp_fib4_node_list_insert(fib4_entry, replace, append);
4490 	if (err)
4491 		return err;
4492 
4493 	err = mlxsw_sp_fib_node_entry_add(mlxsw_sp, &fib4_entry->common);
4494 	if (err)
4495 		goto err_fib_node_entry_add;
4496 
4497 	return 0;
4498 
4499 err_fib_node_entry_add:
4500 	mlxsw_sp_fib4_node_list_remove(fib4_entry);
4501 	return err;
4502 }
4503 
4504 static void
4505 mlxsw_sp_fib4_node_entry_unlink(struct mlxsw_sp *mlxsw_sp,
4506 				struct mlxsw_sp_fib4_entry *fib4_entry)
4507 {
4508 	mlxsw_sp_fib_node_entry_del(mlxsw_sp, &fib4_entry->common);
4509 	mlxsw_sp_fib4_node_list_remove(fib4_entry);
4510 
4511 	if (fib4_entry->common.type == MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP)
4512 		mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, &fib4_entry->common);
4513 }
4514 
4515 static void mlxsw_sp_fib4_entry_replace(struct mlxsw_sp *mlxsw_sp,
4516 					struct mlxsw_sp_fib4_entry *fib4_entry,
4517 					bool replace)
4518 {
4519 	struct mlxsw_sp_fib_node *fib_node = fib4_entry->common.fib_node;
4520 	struct mlxsw_sp_fib4_entry *replaced;
4521 
4522 	if (!replace)
4523 		return;
4524 
4525 	/* We inserted the new entry before replaced one */
4526 	replaced = list_next_entry(fib4_entry, common.list);
4527 
4528 	mlxsw_sp_fib4_node_entry_unlink(mlxsw_sp, replaced);
4529 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, replaced);
4530 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
4531 }
4532 
4533 static int
4534 mlxsw_sp_router_fib4_add(struct mlxsw_sp *mlxsw_sp,
4535 			 const struct fib_entry_notifier_info *fen_info,
4536 			 bool replace, bool append)
4537 {
4538 	struct mlxsw_sp_fib4_entry *fib4_entry;
4539 	struct mlxsw_sp_fib_node *fib_node;
4540 	int err;
4541 
4542 	if (mlxsw_sp->router->aborted)
4543 		return 0;
4544 
4545 	fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, fen_info->tb_id,
4546 					 &fen_info->dst, sizeof(fen_info->dst),
4547 					 fen_info->dst_len,
4548 					 MLXSW_SP_L3_PROTO_IPV4);
4549 	if (IS_ERR(fib_node)) {
4550 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to get FIB node\n");
4551 		return PTR_ERR(fib_node);
4552 	}
4553 
4554 	fib4_entry = mlxsw_sp_fib4_entry_create(mlxsw_sp, fib_node, fen_info);
4555 	if (IS_ERR(fib4_entry)) {
4556 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to create FIB entry\n");
4557 		err = PTR_ERR(fib4_entry);
4558 		goto err_fib4_entry_create;
4559 	}
4560 
4561 	err = mlxsw_sp_fib4_node_entry_link(mlxsw_sp, fib4_entry, replace,
4562 					    append);
4563 	if (err) {
4564 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to link FIB entry to node\n");
4565 		goto err_fib4_node_entry_link;
4566 	}
4567 
4568 	mlxsw_sp_fib4_entry_replace(mlxsw_sp, fib4_entry, replace);
4569 
4570 	return 0;
4571 
4572 err_fib4_node_entry_link:
4573 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
4574 err_fib4_entry_create:
4575 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
4576 	return err;
4577 }
4578 
4579 static void mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
4580 				     struct fib_entry_notifier_info *fen_info)
4581 {
4582 	struct mlxsw_sp_fib4_entry *fib4_entry;
4583 	struct mlxsw_sp_fib_node *fib_node;
4584 
4585 	if (mlxsw_sp->router->aborted)
4586 		return;
4587 
4588 	fib4_entry = mlxsw_sp_fib4_entry_lookup(mlxsw_sp, fen_info);
4589 	if (WARN_ON(!fib4_entry))
4590 		return;
4591 	fib_node = fib4_entry->common.fib_node;
4592 
4593 	mlxsw_sp_fib4_node_entry_unlink(mlxsw_sp, fib4_entry);
4594 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
4595 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
4596 }
4597 
4598 static bool mlxsw_sp_fib6_rt_should_ignore(const struct rt6_info *rt)
4599 {
4600 	/* Packets with link-local destination IP arriving to the router
4601 	 * are trapped to the CPU, so no need to program specific routes
4602 	 * for them.
4603 	 */
4604 	if (ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LINKLOCAL)
4605 		return true;
4606 
4607 	/* Multicast routes aren't supported, so ignore them. Neighbour
4608 	 * Discovery packets are specifically trapped.
4609 	 */
4610 	if (ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_MULTICAST)
4611 		return true;
4612 
4613 	/* Cloned routes are irrelevant in the forwarding path. */
4614 	if (rt->rt6i_flags & RTF_CACHE)
4615 		return true;
4616 
4617 	return false;
4618 }
4619 
4620 static struct mlxsw_sp_rt6 *mlxsw_sp_rt6_create(struct rt6_info *rt)
4621 {
4622 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4623 
4624 	mlxsw_sp_rt6 = kzalloc(sizeof(*mlxsw_sp_rt6), GFP_KERNEL);
4625 	if (!mlxsw_sp_rt6)
4626 		return ERR_PTR(-ENOMEM);
4627 
4628 	/* In case of route replace, replaced route is deleted with
4629 	 * no notification. Take reference to prevent accessing freed
4630 	 * memory.
4631 	 */
4632 	mlxsw_sp_rt6->rt = rt;
4633 	rt6_hold(rt);
4634 
4635 	return mlxsw_sp_rt6;
4636 }
4637 
4638 #if IS_ENABLED(CONFIG_IPV6)
4639 static void mlxsw_sp_rt6_release(struct rt6_info *rt)
4640 {
4641 	rt6_release(rt);
4642 }
4643 #else
4644 static void mlxsw_sp_rt6_release(struct rt6_info *rt)
4645 {
4646 }
4647 #endif
4648 
4649 static void mlxsw_sp_rt6_destroy(struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
4650 {
4651 	mlxsw_sp_rt6_release(mlxsw_sp_rt6->rt);
4652 	kfree(mlxsw_sp_rt6);
4653 }
4654 
4655 static bool mlxsw_sp_fib6_rt_can_mp(const struct rt6_info *rt)
4656 {
4657 	/* RTF_CACHE routes are ignored */
4658 	return (rt->rt6i_flags & (RTF_GATEWAY | RTF_ADDRCONF)) == RTF_GATEWAY;
4659 }
4660 
4661 static struct rt6_info *
4662 mlxsw_sp_fib6_entry_rt(const struct mlxsw_sp_fib6_entry *fib6_entry)
4663 {
4664 	return list_first_entry(&fib6_entry->rt6_list, struct mlxsw_sp_rt6,
4665 				list)->rt;
4666 }
4667 
4668 static struct mlxsw_sp_fib6_entry *
4669 mlxsw_sp_fib6_node_mp_entry_find(const struct mlxsw_sp_fib_node *fib_node,
4670 				 const struct rt6_info *nrt, bool replace)
4671 {
4672 	struct mlxsw_sp_fib6_entry *fib6_entry;
4673 
4674 	if (!mlxsw_sp_fib6_rt_can_mp(nrt) || replace)
4675 		return NULL;
4676 
4677 	list_for_each_entry(fib6_entry, &fib_node->entry_list, common.list) {
4678 		struct rt6_info *rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
4679 
4680 		/* RT6_TABLE_LOCAL and RT6_TABLE_MAIN share the same
4681 		 * virtual router.
4682 		 */
4683 		if (rt->rt6i_table->tb6_id > nrt->rt6i_table->tb6_id)
4684 			continue;
4685 		if (rt->rt6i_table->tb6_id != nrt->rt6i_table->tb6_id)
4686 			break;
4687 		if (rt->rt6i_metric < nrt->rt6i_metric)
4688 			continue;
4689 		if (rt->rt6i_metric == nrt->rt6i_metric &&
4690 		    mlxsw_sp_fib6_rt_can_mp(rt))
4691 			return fib6_entry;
4692 		if (rt->rt6i_metric > nrt->rt6i_metric)
4693 			break;
4694 	}
4695 
4696 	return NULL;
4697 }
4698 
4699 static struct mlxsw_sp_rt6 *
4700 mlxsw_sp_fib6_entry_rt_find(const struct mlxsw_sp_fib6_entry *fib6_entry,
4701 			    const struct rt6_info *rt)
4702 {
4703 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4704 
4705 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
4706 		if (mlxsw_sp_rt6->rt == rt)
4707 			return mlxsw_sp_rt6;
4708 	}
4709 
4710 	return NULL;
4711 }
4712 
4713 static bool mlxsw_sp_nexthop6_ipip_type(const struct mlxsw_sp *mlxsw_sp,
4714 					const struct rt6_info *rt,
4715 					enum mlxsw_sp_ipip_type *ret)
4716 {
4717 	return rt->dst.dev &&
4718 	       mlxsw_sp_netdev_ipip_type(mlxsw_sp, rt->dst.dev, ret);
4719 }
4720 
4721 static int mlxsw_sp_nexthop6_type_init(struct mlxsw_sp *mlxsw_sp,
4722 				       struct mlxsw_sp_nexthop_group *nh_grp,
4723 				       struct mlxsw_sp_nexthop *nh,
4724 				       const struct rt6_info *rt)
4725 {
4726 	const struct mlxsw_sp_ipip_ops *ipip_ops;
4727 	struct mlxsw_sp_ipip_entry *ipip_entry;
4728 	struct net_device *dev = rt->dst.dev;
4729 	struct mlxsw_sp_rif *rif;
4730 	int err;
4731 
4732 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, dev);
4733 	if (ipip_entry) {
4734 		ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
4735 		if (ipip_ops->can_offload(mlxsw_sp, dev,
4736 					  MLXSW_SP_L3_PROTO_IPV6)) {
4737 			nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
4738 			mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, ipip_entry);
4739 			return 0;
4740 		}
4741 	}
4742 
4743 	nh->type = MLXSW_SP_NEXTHOP_TYPE_ETH;
4744 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
4745 	if (!rif)
4746 		return 0;
4747 	mlxsw_sp_nexthop_rif_init(nh, rif);
4748 
4749 	err = mlxsw_sp_nexthop_neigh_init(mlxsw_sp, nh);
4750 	if (err)
4751 		goto err_nexthop_neigh_init;
4752 
4753 	return 0;
4754 
4755 err_nexthop_neigh_init:
4756 	mlxsw_sp_nexthop_rif_fini(nh);
4757 	return err;
4758 }
4759 
4760 static void mlxsw_sp_nexthop6_type_fini(struct mlxsw_sp *mlxsw_sp,
4761 					struct mlxsw_sp_nexthop *nh)
4762 {
4763 	mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4764 }
4765 
4766 static int mlxsw_sp_nexthop6_init(struct mlxsw_sp *mlxsw_sp,
4767 				  struct mlxsw_sp_nexthop_group *nh_grp,
4768 				  struct mlxsw_sp_nexthop *nh,
4769 				  const struct rt6_info *rt)
4770 {
4771 	struct net_device *dev = rt->dst.dev;
4772 
4773 	nh->nh_grp = nh_grp;
4774 	nh->nh_weight = 1;
4775 	memcpy(&nh->gw_addr, &rt->rt6i_gateway, sizeof(nh->gw_addr));
4776 	mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
4777 
4778 	list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
4779 
4780 	if (!dev)
4781 		return 0;
4782 	nh->ifindex = dev->ifindex;
4783 
4784 	return mlxsw_sp_nexthop6_type_init(mlxsw_sp, nh_grp, nh, rt);
4785 }
4786 
4787 static void mlxsw_sp_nexthop6_fini(struct mlxsw_sp *mlxsw_sp,
4788 				   struct mlxsw_sp_nexthop *nh)
4789 {
4790 	mlxsw_sp_nexthop6_type_fini(mlxsw_sp, nh);
4791 	list_del(&nh->router_list_node);
4792 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
4793 }
4794 
4795 static bool mlxsw_sp_rt6_is_gateway(const struct mlxsw_sp *mlxsw_sp,
4796 				    const struct rt6_info *rt)
4797 {
4798 	return rt->rt6i_flags & RTF_GATEWAY ||
4799 	       mlxsw_sp_nexthop6_ipip_type(mlxsw_sp, rt, NULL);
4800 }
4801 
4802 static struct mlxsw_sp_nexthop_group *
4803 mlxsw_sp_nexthop6_group_create(struct mlxsw_sp *mlxsw_sp,
4804 			       struct mlxsw_sp_fib6_entry *fib6_entry)
4805 {
4806 	struct mlxsw_sp_nexthop_group *nh_grp;
4807 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4808 	struct mlxsw_sp_nexthop *nh;
4809 	size_t alloc_size;
4810 	int i = 0;
4811 	int err;
4812 
4813 	alloc_size = sizeof(*nh_grp) +
4814 		     fib6_entry->nrt6 * sizeof(struct mlxsw_sp_nexthop);
4815 	nh_grp = kzalloc(alloc_size, GFP_KERNEL);
4816 	if (!nh_grp)
4817 		return ERR_PTR(-ENOMEM);
4818 	INIT_LIST_HEAD(&nh_grp->fib_list);
4819 #if IS_ENABLED(CONFIG_IPV6)
4820 	nh_grp->neigh_tbl = &nd_tbl;
4821 #endif
4822 	mlxsw_sp_rt6 = list_first_entry(&fib6_entry->rt6_list,
4823 					struct mlxsw_sp_rt6, list);
4824 	nh_grp->gateway = mlxsw_sp_rt6_is_gateway(mlxsw_sp, mlxsw_sp_rt6->rt);
4825 	nh_grp->count = fib6_entry->nrt6;
4826 	for (i = 0; i < nh_grp->count; i++) {
4827 		struct rt6_info *rt = mlxsw_sp_rt6->rt;
4828 
4829 		nh = &nh_grp->nexthops[i];
4830 		err = mlxsw_sp_nexthop6_init(mlxsw_sp, nh_grp, nh, rt);
4831 		if (err)
4832 			goto err_nexthop6_init;
4833 		mlxsw_sp_rt6 = list_next_entry(mlxsw_sp_rt6, list);
4834 	}
4835 
4836 	err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
4837 	if (err)
4838 		goto err_nexthop_group_insert;
4839 
4840 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4841 	return nh_grp;
4842 
4843 err_nexthop_group_insert:
4844 err_nexthop6_init:
4845 	for (i--; i >= 0; i--) {
4846 		nh = &nh_grp->nexthops[i];
4847 		mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
4848 	}
4849 	kfree(nh_grp);
4850 	return ERR_PTR(err);
4851 }
4852 
4853 static void
4854 mlxsw_sp_nexthop6_group_destroy(struct mlxsw_sp *mlxsw_sp,
4855 				struct mlxsw_sp_nexthop_group *nh_grp)
4856 {
4857 	struct mlxsw_sp_nexthop *nh;
4858 	int i = nh_grp->count;
4859 
4860 	mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
4861 	for (i--; i >= 0; i--) {
4862 		nh = &nh_grp->nexthops[i];
4863 		mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
4864 	}
4865 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4866 	WARN_ON(nh_grp->adj_index_valid);
4867 	kfree(nh_grp);
4868 }
4869 
4870 static int mlxsw_sp_nexthop6_group_get(struct mlxsw_sp *mlxsw_sp,
4871 				       struct mlxsw_sp_fib6_entry *fib6_entry)
4872 {
4873 	struct mlxsw_sp_nexthop_group *nh_grp;
4874 
4875 	nh_grp = mlxsw_sp_nexthop6_group_lookup(mlxsw_sp, fib6_entry);
4876 	if (!nh_grp) {
4877 		nh_grp = mlxsw_sp_nexthop6_group_create(mlxsw_sp, fib6_entry);
4878 		if (IS_ERR(nh_grp))
4879 			return PTR_ERR(nh_grp);
4880 	}
4881 
4882 	list_add_tail(&fib6_entry->common.nexthop_group_node,
4883 		      &nh_grp->fib_list);
4884 	fib6_entry->common.nh_group = nh_grp;
4885 
4886 	return 0;
4887 }
4888 
4889 static void mlxsw_sp_nexthop6_group_put(struct mlxsw_sp *mlxsw_sp,
4890 					struct mlxsw_sp_fib_entry *fib_entry)
4891 {
4892 	struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
4893 
4894 	list_del(&fib_entry->nexthop_group_node);
4895 	if (!list_empty(&nh_grp->fib_list))
4896 		return;
4897 	mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, nh_grp);
4898 }
4899 
4900 static int
4901 mlxsw_sp_nexthop6_group_update(struct mlxsw_sp *mlxsw_sp,
4902 			       struct mlxsw_sp_fib6_entry *fib6_entry)
4903 {
4904 	struct mlxsw_sp_nexthop_group *old_nh_grp = fib6_entry->common.nh_group;
4905 	int err;
4906 
4907 	fib6_entry->common.nh_group = NULL;
4908 	list_del(&fib6_entry->common.nexthop_group_node);
4909 
4910 	err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
4911 	if (err)
4912 		goto err_nexthop6_group_get;
4913 
4914 	/* In case this entry is offloaded, then the adjacency index
4915 	 * currently associated with it in the device's table is that
4916 	 * of the old group. Start using the new one instead.
4917 	 */
4918 	err = mlxsw_sp_fib_node_entry_add(mlxsw_sp, &fib6_entry->common);
4919 	if (err)
4920 		goto err_fib_node_entry_add;
4921 
4922 	if (list_empty(&old_nh_grp->fib_list))
4923 		mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, old_nh_grp);
4924 
4925 	return 0;
4926 
4927 err_fib_node_entry_add:
4928 	mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
4929 err_nexthop6_group_get:
4930 	list_add_tail(&fib6_entry->common.nexthop_group_node,
4931 		      &old_nh_grp->fib_list);
4932 	fib6_entry->common.nh_group = old_nh_grp;
4933 	return err;
4934 }
4935 
4936 static int
4937 mlxsw_sp_fib6_entry_nexthop_add(struct mlxsw_sp *mlxsw_sp,
4938 				struct mlxsw_sp_fib6_entry *fib6_entry,
4939 				struct rt6_info *rt)
4940 {
4941 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4942 	int err;
4943 
4944 	mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt);
4945 	if (IS_ERR(mlxsw_sp_rt6))
4946 		return PTR_ERR(mlxsw_sp_rt6);
4947 
4948 	list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
4949 	fib6_entry->nrt6++;
4950 
4951 	err = mlxsw_sp_nexthop6_group_update(mlxsw_sp, fib6_entry);
4952 	if (err)
4953 		goto err_nexthop6_group_update;
4954 
4955 	return 0;
4956 
4957 err_nexthop6_group_update:
4958 	fib6_entry->nrt6--;
4959 	list_del(&mlxsw_sp_rt6->list);
4960 	mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
4961 	return err;
4962 }
4963 
4964 static void
4965 mlxsw_sp_fib6_entry_nexthop_del(struct mlxsw_sp *mlxsw_sp,
4966 				struct mlxsw_sp_fib6_entry *fib6_entry,
4967 				struct rt6_info *rt)
4968 {
4969 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4970 
4971 	mlxsw_sp_rt6 = mlxsw_sp_fib6_entry_rt_find(fib6_entry, rt);
4972 	if (WARN_ON(!mlxsw_sp_rt6))
4973 		return;
4974 
4975 	fib6_entry->nrt6--;
4976 	list_del(&mlxsw_sp_rt6->list);
4977 	mlxsw_sp_nexthop6_group_update(mlxsw_sp, fib6_entry);
4978 	mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
4979 }
4980 
4981 static void mlxsw_sp_fib6_entry_type_set(struct mlxsw_sp *mlxsw_sp,
4982 					 struct mlxsw_sp_fib_entry *fib_entry,
4983 					 const struct rt6_info *rt)
4984 {
4985 	/* Packets hitting RTF_REJECT routes need to be discarded by the
4986 	 * stack. We can rely on their destination device not having a
4987 	 * RIF (it's the loopback device) and can thus use action type
4988 	 * local, which will cause them to be trapped with a lower
4989 	 * priority than packets that need to be locally received.
4990 	 */
4991 	if (rt->rt6i_flags & (RTF_LOCAL | RTF_ANYCAST))
4992 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
4993 	else if (rt->rt6i_flags & RTF_REJECT)
4994 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
4995 	else if (mlxsw_sp_rt6_is_gateway(mlxsw_sp, rt))
4996 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
4997 	else
4998 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
4999 }
5000 
5001 static void
5002 mlxsw_sp_fib6_entry_rt_destroy_all(struct mlxsw_sp_fib6_entry *fib6_entry)
5003 {
5004 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6, *tmp;
5005 
5006 	list_for_each_entry_safe(mlxsw_sp_rt6, tmp, &fib6_entry->rt6_list,
5007 				 list) {
5008 		fib6_entry->nrt6--;
5009 		list_del(&mlxsw_sp_rt6->list);
5010 		mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5011 	}
5012 }
5013 
5014 static struct mlxsw_sp_fib6_entry *
5015 mlxsw_sp_fib6_entry_create(struct mlxsw_sp *mlxsw_sp,
5016 			   struct mlxsw_sp_fib_node *fib_node,
5017 			   struct rt6_info *rt)
5018 {
5019 	struct mlxsw_sp_fib6_entry *fib6_entry;
5020 	struct mlxsw_sp_fib_entry *fib_entry;
5021 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5022 	int err;
5023 
5024 	fib6_entry = kzalloc(sizeof(*fib6_entry), GFP_KERNEL);
5025 	if (!fib6_entry)
5026 		return ERR_PTR(-ENOMEM);
5027 	fib_entry = &fib6_entry->common;
5028 
5029 	mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt);
5030 	if (IS_ERR(mlxsw_sp_rt6)) {
5031 		err = PTR_ERR(mlxsw_sp_rt6);
5032 		goto err_rt6_create;
5033 	}
5034 
5035 	mlxsw_sp_fib6_entry_type_set(mlxsw_sp, fib_entry, mlxsw_sp_rt6->rt);
5036 
5037 	INIT_LIST_HEAD(&fib6_entry->rt6_list);
5038 	list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
5039 	fib6_entry->nrt6 = 1;
5040 	err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
5041 	if (err)
5042 		goto err_nexthop6_group_get;
5043 
5044 	fib_entry->fib_node = fib_node;
5045 
5046 	return fib6_entry;
5047 
5048 err_nexthop6_group_get:
5049 	list_del(&mlxsw_sp_rt6->list);
5050 	mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5051 err_rt6_create:
5052 	kfree(fib6_entry);
5053 	return ERR_PTR(err);
5054 }
5055 
5056 static void mlxsw_sp_fib6_entry_destroy(struct mlxsw_sp *mlxsw_sp,
5057 					struct mlxsw_sp_fib6_entry *fib6_entry)
5058 {
5059 	mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
5060 	mlxsw_sp_fib6_entry_rt_destroy_all(fib6_entry);
5061 	WARN_ON(fib6_entry->nrt6);
5062 	kfree(fib6_entry);
5063 }
5064 
5065 static struct mlxsw_sp_fib6_entry *
5066 mlxsw_sp_fib6_node_entry_find(const struct mlxsw_sp_fib_node *fib_node,
5067 			      const struct rt6_info *nrt, bool replace)
5068 {
5069 	struct mlxsw_sp_fib6_entry *fib6_entry, *fallback = NULL;
5070 
5071 	list_for_each_entry(fib6_entry, &fib_node->entry_list, common.list) {
5072 		struct rt6_info *rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
5073 
5074 		if (rt->rt6i_table->tb6_id > nrt->rt6i_table->tb6_id)
5075 			continue;
5076 		if (rt->rt6i_table->tb6_id != nrt->rt6i_table->tb6_id)
5077 			break;
5078 		if (replace && rt->rt6i_metric == nrt->rt6i_metric) {
5079 			if (mlxsw_sp_fib6_rt_can_mp(rt) ==
5080 			    mlxsw_sp_fib6_rt_can_mp(nrt))
5081 				return fib6_entry;
5082 			if (mlxsw_sp_fib6_rt_can_mp(nrt))
5083 				fallback = fallback ?: fib6_entry;
5084 		}
5085 		if (rt->rt6i_metric > nrt->rt6i_metric)
5086 			return fallback ?: fib6_entry;
5087 	}
5088 
5089 	return fallback;
5090 }
5091 
5092 static int
5093 mlxsw_sp_fib6_node_list_insert(struct mlxsw_sp_fib6_entry *new6_entry,
5094 			       bool replace)
5095 {
5096 	struct mlxsw_sp_fib_node *fib_node = new6_entry->common.fib_node;
5097 	struct rt6_info *nrt = mlxsw_sp_fib6_entry_rt(new6_entry);
5098 	struct mlxsw_sp_fib6_entry *fib6_entry;
5099 
5100 	fib6_entry = mlxsw_sp_fib6_node_entry_find(fib_node, nrt, replace);
5101 
5102 	if (replace && WARN_ON(!fib6_entry))
5103 		return -EINVAL;
5104 
5105 	if (fib6_entry) {
5106 		list_add_tail(&new6_entry->common.list,
5107 			      &fib6_entry->common.list);
5108 	} else {
5109 		struct mlxsw_sp_fib6_entry *last;
5110 
5111 		list_for_each_entry(last, &fib_node->entry_list, common.list) {
5112 			struct rt6_info *rt = mlxsw_sp_fib6_entry_rt(last);
5113 
5114 			if (nrt->rt6i_table->tb6_id > rt->rt6i_table->tb6_id)
5115 				break;
5116 			fib6_entry = last;
5117 		}
5118 
5119 		if (fib6_entry)
5120 			list_add(&new6_entry->common.list,
5121 				 &fib6_entry->common.list);
5122 		else
5123 			list_add(&new6_entry->common.list,
5124 				 &fib_node->entry_list);
5125 	}
5126 
5127 	return 0;
5128 }
5129 
5130 static void
5131 mlxsw_sp_fib6_node_list_remove(struct mlxsw_sp_fib6_entry *fib6_entry)
5132 {
5133 	list_del(&fib6_entry->common.list);
5134 }
5135 
5136 static int mlxsw_sp_fib6_node_entry_link(struct mlxsw_sp *mlxsw_sp,
5137 					 struct mlxsw_sp_fib6_entry *fib6_entry,
5138 					 bool replace)
5139 {
5140 	int err;
5141 
5142 	err = mlxsw_sp_fib6_node_list_insert(fib6_entry, replace);
5143 	if (err)
5144 		return err;
5145 
5146 	err = mlxsw_sp_fib_node_entry_add(mlxsw_sp, &fib6_entry->common);
5147 	if (err)
5148 		goto err_fib_node_entry_add;
5149 
5150 	return 0;
5151 
5152 err_fib_node_entry_add:
5153 	mlxsw_sp_fib6_node_list_remove(fib6_entry);
5154 	return err;
5155 }
5156 
5157 static void
5158 mlxsw_sp_fib6_node_entry_unlink(struct mlxsw_sp *mlxsw_sp,
5159 				struct mlxsw_sp_fib6_entry *fib6_entry)
5160 {
5161 	mlxsw_sp_fib_node_entry_del(mlxsw_sp, &fib6_entry->common);
5162 	mlxsw_sp_fib6_node_list_remove(fib6_entry);
5163 }
5164 
5165 static struct mlxsw_sp_fib6_entry *
5166 mlxsw_sp_fib6_entry_lookup(struct mlxsw_sp *mlxsw_sp,
5167 			   const struct rt6_info *rt)
5168 {
5169 	struct mlxsw_sp_fib6_entry *fib6_entry;
5170 	struct mlxsw_sp_fib_node *fib_node;
5171 	struct mlxsw_sp_fib *fib;
5172 	struct mlxsw_sp_vr *vr;
5173 
5174 	vr = mlxsw_sp_vr_find(mlxsw_sp, rt->rt6i_table->tb6_id);
5175 	if (!vr)
5176 		return NULL;
5177 	fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV6);
5178 
5179 	fib_node = mlxsw_sp_fib_node_lookup(fib, &rt->rt6i_dst.addr,
5180 					    sizeof(rt->rt6i_dst.addr),
5181 					    rt->rt6i_dst.plen);
5182 	if (!fib_node)
5183 		return NULL;
5184 
5185 	list_for_each_entry(fib6_entry, &fib_node->entry_list, common.list) {
5186 		struct rt6_info *iter_rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
5187 
5188 		if (rt->rt6i_table->tb6_id == iter_rt->rt6i_table->tb6_id &&
5189 		    rt->rt6i_metric == iter_rt->rt6i_metric &&
5190 		    mlxsw_sp_fib6_entry_rt_find(fib6_entry, rt))
5191 			return fib6_entry;
5192 	}
5193 
5194 	return NULL;
5195 }
5196 
5197 static void mlxsw_sp_fib6_entry_replace(struct mlxsw_sp *mlxsw_sp,
5198 					struct mlxsw_sp_fib6_entry *fib6_entry,
5199 					bool replace)
5200 {
5201 	struct mlxsw_sp_fib_node *fib_node = fib6_entry->common.fib_node;
5202 	struct mlxsw_sp_fib6_entry *replaced;
5203 
5204 	if (!replace)
5205 		return;
5206 
5207 	replaced = list_next_entry(fib6_entry, common.list);
5208 
5209 	mlxsw_sp_fib6_node_entry_unlink(mlxsw_sp, replaced);
5210 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, replaced);
5211 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5212 }
5213 
5214 static int mlxsw_sp_router_fib6_add(struct mlxsw_sp *mlxsw_sp,
5215 				    struct rt6_info *rt, bool replace)
5216 {
5217 	struct mlxsw_sp_fib6_entry *fib6_entry;
5218 	struct mlxsw_sp_fib_node *fib_node;
5219 	int err;
5220 
5221 	if (mlxsw_sp->router->aborted)
5222 		return 0;
5223 
5224 	if (rt->rt6i_src.plen)
5225 		return -EINVAL;
5226 
5227 	if (mlxsw_sp_fib6_rt_should_ignore(rt))
5228 		return 0;
5229 
5230 	fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->rt6i_table->tb6_id,
5231 					 &rt->rt6i_dst.addr,
5232 					 sizeof(rt->rt6i_dst.addr),
5233 					 rt->rt6i_dst.plen,
5234 					 MLXSW_SP_L3_PROTO_IPV6);
5235 	if (IS_ERR(fib_node))
5236 		return PTR_ERR(fib_node);
5237 
5238 	/* Before creating a new entry, try to append route to an existing
5239 	 * multipath entry.
5240 	 */
5241 	fib6_entry = mlxsw_sp_fib6_node_mp_entry_find(fib_node, rt, replace);
5242 	if (fib6_entry) {
5243 		err = mlxsw_sp_fib6_entry_nexthop_add(mlxsw_sp, fib6_entry, rt);
5244 		if (err)
5245 			goto err_fib6_entry_nexthop_add;
5246 		return 0;
5247 	}
5248 
5249 	fib6_entry = mlxsw_sp_fib6_entry_create(mlxsw_sp, fib_node, rt);
5250 	if (IS_ERR(fib6_entry)) {
5251 		err = PTR_ERR(fib6_entry);
5252 		goto err_fib6_entry_create;
5253 	}
5254 
5255 	err = mlxsw_sp_fib6_node_entry_link(mlxsw_sp, fib6_entry, replace);
5256 	if (err)
5257 		goto err_fib6_node_entry_link;
5258 
5259 	mlxsw_sp_fib6_entry_replace(mlxsw_sp, fib6_entry, replace);
5260 
5261 	return 0;
5262 
5263 err_fib6_node_entry_link:
5264 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5265 err_fib6_entry_create:
5266 err_fib6_entry_nexthop_add:
5267 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5268 	return err;
5269 }
5270 
5271 static void mlxsw_sp_router_fib6_del(struct mlxsw_sp *mlxsw_sp,
5272 				     struct rt6_info *rt)
5273 {
5274 	struct mlxsw_sp_fib6_entry *fib6_entry;
5275 	struct mlxsw_sp_fib_node *fib_node;
5276 
5277 	if (mlxsw_sp->router->aborted)
5278 		return;
5279 
5280 	if (mlxsw_sp_fib6_rt_should_ignore(rt))
5281 		return;
5282 
5283 	fib6_entry = mlxsw_sp_fib6_entry_lookup(mlxsw_sp, rt);
5284 	if (WARN_ON(!fib6_entry))
5285 		return;
5286 
5287 	/* If route is part of a multipath entry, but not the last one
5288 	 * removed, then only reduce its nexthop group.
5289 	 */
5290 	if (!list_is_singular(&fib6_entry->rt6_list)) {
5291 		mlxsw_sp_fib6_entry_nexthop_del(mlxsw_sp, fib6_entry, rt);
5292 		return;
5293 	}
5294 
5295 	fib_node = fib6_entry->common.fib_node;
5296 
5297 	mlxsw_sp_fib6_node_entry_unlink(mlxsw_sp, fib6_entry);
5298 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5299 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5300 }
5301 
5302 static int __mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp,
5303 					    enum mlxsw_reg_ralxx_protocol proto,
5304 					    u8 tree_id)
5305 {
5306 	char ralta_pl[MLXSW_REG_RALTA_LEN];
5307 	char ralst_pl[MLXSW_REG_RALST_LEN];
5308 	int i, err;
5309 
5310 	mlxsw_reg_ralta_pack(ralta_pl, true, proto, tree_id);
5311 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
5312 	if (err)
5313 		return err;
5314 
5315 	mlxsw_reg_ralst_pack(ralst_pl, 0xff, tree_id);
5316 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
5317 	if (err)
5318 		return err;
5319 
5320 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
5321 		struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
5322 		char raltb_pl[MLXSW_REG_RALTB_LEN];
5323 		char ralue_pl[MLXSW_REG_RALUE_LEN];
5324 
5325 		mlxsw_reg_raltb_pack(raltb_pl, vr->id, proto, tree_id);
5326 		err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb),
5327 				      raltb_pl);
5328 		if (err)
5329 			return err;
5330 
5331 		mlxsw_reg_ralue_pack(ralue_pl, proto,
5332 				     MLXSW_REG_RALUE_OP_WRITE_WRITE, vr->id, 0);
5333 		mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
5334 		err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue),
5335 				      ralue_pl);
5336 		if (err)
5337 			return err;
5338 	}
5339 
5340 	return 0;
5341 }
5342 
5343 static int mlxsw_sp_router_fibmr_add(struct mlxsw_sp *mlxsw_sp,
5344 				     struct mfc_entry_notifier_info *men_info,
5345 				     bool replace)
5346 {
5347 	struct mlxsw_sp_vr *vr;
5348 
5349 	if (mlxsw_sp->router->aborted)
5350 		return 0;
5351 
5352 	vr = mlxsw_sp_vr_get(mlxsw_sp, men_info->tb_id, NULL);
5353 	if (IS_ERR(vr))
5354 		return PTR_ERR(vr);
5355 
5356 	return mlxsw_sp_mr_route4_add(vr->mr4_table, men_info->mfc, replace);
5357 }
5358 
5359 static void mlxsw_sp_router_fibmr_del(struct mlxsw_sp *mlxsw_sp,
5360 				      struct mfc_entry_notifier_info *men_info)
5361 {
5362 	struct mlxsw_sp_vr *vr;
5363 
5364 	if (mlxsw_sp->router->aborted)
5365 		return;
5366 
5367 	vr = mlxsw_sp_vr_find(mlxsw_sp, men_info->tb_id);
5368 	if (WARN_ON(!vr))
5369 		return;
5370 
5371 	mlxsw_sp_mr_route4_del(vr->mr4_table, men_info->mfc);
5372 	mlxsw_sp_vr_put(vr);
5373 }
5374 
5375 static int
5376 mlxsw_sp_router_fibmr_vif_add(struct mlxsw_sp *mlxsw_sp,
5377 			      struct vif_entry_notifier_info *ven_info)
5378 {
5379 	struct mlxsw_sp_rif *rif;
5380 	struct mlxsw_sp_vr *vr;
5381 
5382 	if (mlxsw_sp->router->aborted)
5383 		return 0;
5384 
5385 	vr = mlxsw_sp_vr_get(mlxsw_sp, ven_info->tb_id, NULL);
5386 	if (IS_ERR(vr))
5387 		return PTR_ERR(vr);
5388 
5389 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, ven_info->dev);
5390 	return mlxsw_sp_mr_vif_add(vr->mr4_table, ven_info->dev,
5391 				   ven_info->vif_index,
5392 				   ven_info->vif_flags, rif);
5393 }
5394 
5395 static void
5396 mlxsw_sp_router_fibmr_vif_del(struct mlxsw_sp *mlxsw_sp,
5397 			      struct vif_entry_notifier_info *ven_info)
5398 {
5399 	struct mlxsw_sp_vr *vr;
5400 
5401 	if (mlxsw_sp->router->aborted)
5402 		return;
5403 
5404 	vr = mlxsw_sp_vr_find(mlxsw_sp, ven_info->tb_id);
5405 	if (WARN_ON(!vr))
5406 		return;
5407 
5408 	mlxsw_sp_mr_vif_del(vr->mr4_table, ven_info->vif_index);
5409 	mlxsw_sp_vr_put(vr);
5410 }
5411 
5412 static int mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp)
5413 {
5414 	enum mlxsw_reg_ralxx_protocol proto = MLXSW_REG_RALXX_PROTOCOL_IPV4;
5415 	int err;
5416 
5417 	err = __mlxsw_sp_router_set_abort_trap(mlxsw_sp, proto,
5418 					       MLXSW_SP_LPM_TREE_MIN);
5419 	if (err)
5420 		return err;
5421 
5422 	/* The multicast router code does not need an abort trap as by default,
5423 	 * packets that don't match any routes are trapped to the CPU.
5424 	 */
5425 
5426 	proto = MLXSW_REG_RALXX_PROTOCOL_IPV6;
5427 	return __mlxsw_sp_router_set_abort_trap(mlxsw_sp, proto,
5428 						MLXSW_SP_LPM_TREE_MIN + 1);
5429 }
5430 
5431 static void mlxsw_sp_fib4_node_flush(struct mlxsw_sp *mlxsw_sp,
5432 				     struct mlxsw_sp_fib_node *fib_node)
5433 {
5434 	struct mlxsw_sp_fib4_entry *fib4_entry, *tmp;
5435 
5436 	list_for_each_entry_safe(fib4_entry, tmp, &fib_node->entry_list,
5437 				 common.list) {
5438 		bool do_break = &tmp->common.list == &fib_node->entry_list;
5439 
5440 		mlxsw_sp_fib4_node_entry_unlink(mlxsw_sp, fib4_entry);
5441 		mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
5442 		mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5443 		/* Break when entry list is empty and node was freed.
5444 		 * Otherwise, we'll access freed memory in the next
5445 		 * iteration.
5446 		 */
5447 		if (do_break)
5448 			break;
5449 	}
5450 }
5451 
5452 static void mlxsw_sp_fib6_node_flush(struct mlxsw_sp *mlxsw_sp,
5453 				     struct mlxsw_sp_fib_node *fib_node)
5454 {
5455 	struct mlxsw_sp_fib6_entry *fib6_entry, *tmp;
5456 
5457 	list_for_each_entry_safe(fib6_entry, tmp, &fib_node->entry_list,
5458 				 common.list) {
5459 		bool do_break = &tmp->common.list == &fib_node->entry_list;
5460 
5461 		mlxsw_sp_fib6_node_entry_unlink(mlxsw_sp, fib6_entry);
5462 		mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5463 		mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5464 		if (do_break)
5465 			break;
5466 	}
5467 }
5468 
5469 static void mlxsw_sp_fib_node_flush(struct mlxsw_sp *mlxsw_sp,
5470 				    struct mlxsw_sp_fib_node *fib_node)
5471 {
5472 	switch (fib_node->fib->proto) {
5473 	case MLXSW_SP_L3_PROTO_IPV4:
5474 		mlxsw_sp_fib4_node_flush(mlxsw_sp, fib_node);
5475 		break;
5476 	case MLXSW_SP_L3_PROTO_IPV6:
5477 		mlxsw_sp_fib6_node_flush(mlxsw_sp, fib_node);
5478 		break;
5479 	}
5480 }
5481 
5482 static void mlxsw_sp_vr_fib_flush(struct mlxsw_sp *mlxsw_sp,
5483 				  struct mlxsw_sp_vr *vr,
5484 				  enum mlxsw_sp_l3proto proto)
5485 {
5486 	struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
5487 	struct mlxsw_sp_fib_node *fib_node, *tmp;
5488 
5489 	list_for_each_entry_safe(fib_node, tmp, &fib->node_list, list) {
5490 		bool do_break = &tmp->list == &fib->node_list;
5491 
5492 		mlxsw_sp_fib_node_flush(mlxsw_sp, fib_node);
5493 		if (do_break)
5494 			break;
5495 	}
5496 }
5497 
5498 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp)
5499 {
5500 	int i;
5501 
5502 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
5503 		struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
5504 
5505 		if (!mlxsw_sp_vr_is_used(vr))
5506 			continue;
5507 
5508 		mlxsw_sp_mr_table_flush(vr->mr4_table);
5509 		mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV4);
5510 
5511 		/* If virtual router was only used for IPv4, then it's no
5512 		 * longer used.
5513 		 */
5514 		if (!mlxsw_sp_vr_is_used(vr))
5515 			continue;
5516 		mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV6);
5517 	}
5518 }
5519 
5520 static void mlxsw_sp_router_fib_abort(struct mlxsw_sp *mlxsw_sp)
5521 {
5522 	int err;
5523 
5524 	if (mlxsw_sp->router->aborted)
5525 		return;
5526 	dev_warn(mlxsw_sp->bus_info->dev, "FIB abort triggered. Note that FIB entries are no longer being offloaded to this device.\n");
5527 	mlxsw_sp_router_fib_flush(mlxsw_sp);
5528 	mlxsw_sp->router->aborted = true;
5529 	err = mlxsw_sp_router_set_abort_trap(mlxsw_sp);
5530 	if (err)
5531 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to set abort trap.\n");
5532 }
5533 
5534 struct mlxsw_sp_fib_event_work {
5535 	struct work_struct work;
5536 	union {
5537 		struct fib6_entry_notifier_info fen6_info;
5538 		struct fib_entry_notifier_info fen_info;
5539 		struct fib_rule_notifier_info fr_info;
5540 		struct fib_nh_notifier_info fnh_info;
5541 		struct mfc_entry_notifier_info men_info;
5542 		struct vif_entry_notifier_info ven_info;
5543 	};
5544 	struct mlxsw_sp *mlxsw_sp;
5545 	unsigned long event;
5546 };
5547 
5548 static void mlxsw_sp_router_fib4_event_work(struct work_struct *work)
5549 {
5550 	struct mlxsw_sp_fib_event_work *fib_work =
5551 		container_of(work, struct mlxsw_sp_fib_event_work, work);
5552 	struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
5553 	bool replace, append;
5554 	int err;
5555 
5556 	/* Protect internal structures from changes */
5557 	rtnl_lock();
5558 	switch (fib_work->event) {
5559 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
5560 	case FIB_EVENT_ENTRY_APPEND: /* fall through */
5561 	case FIB_EVENT_ENTRY_ADD:
5562 		replace = fib_work->event == FIB_EVENT_ENTRY_REPLACE;
5563 		append = fib_work->event == FIB_EVENT_ENTRY_APPEND;
5564 		err = mlxsw_sp_router_fib4_add(mlxsw_sp, &fib_work->fen_info,
5565 					       replace, append);
5566 		if (err)
5567 			mlxsw_sp_router_fib_abort(mlxsw_sp);
5568 		fib_info_put(fib_work->fen_info.fi);
5569 		break;
5570 	case FIB_EVENT_ENTRY_DEL:
5571 		mlxsw_sp_router_fib4_del(mlxsw_sp, &fib_work->fen_info);
5572 		fib_info_put(fib_work->fen_info.fi);
5573 		break;
5574 	case FIB_EVENT_RULE_ADD:
5575 		/* if we get here, a rule was added that we do not support.
5576 		 * just do the fib_abort
5577 		 */
5578 		mlxsw_sp_router_fib_abort(mlxsw_sp);
5579 		break;
5580 	case FIB_EVENT_NH_ADD: /* fall through */
5581 	case FIB_EVENT_NH_DEL:
5582 		mlxsw_sp_nexthop4_event(mlxsw_sp, fib_work->event,
5583 					fib_work->fnh_info.fib_nh);
5584 		fib_info_put(fib_work->fnh_info.fib_nh->nh_parent);
5585 		break;
5586 	}
5587 	rtnl_unlock();
5588 	kfree(fib_work);
5589 }
5590 
5591 static void mlxsw_sp_router_fib6_event_work(struct work_struct *work)
5592 {
5593 	struct mlxsw_sp_fib_event_work *fib_work =
5594 		container_of(work, struct mlxsw_sp_fib_event_work, work);
5595 	struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
5596 	bool replace;
5597 	int err;
5598 
5599 	rtnl_lock();
5600 	switch (fib_work->event) {
5601 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
5602 	case FIB_EVENT_ENTRY_ADD:
5603 		replace = fib_work->event == FIB_EVENT_ENTRY_REPLACE;
5604 		err = mlxsw_sp_router_fib6_add(mlxsw_sp,
5605 					       fib_work->fen6_info.rt, replace);
5606 		if (err)
5607 			mlxsw_sp_router_fib_abort(mlxsw_sp);
5608 		mlxsw_sp_rt6_release(fib_work->fen6_info.rt);
5609 		break;
5610 	case FIB_EVENT_ENTRY_DEL:
5611 		mlxsw_sp_router_fib6_del(mlxsw_sp, fib_work->fen6_info.rt);
5612 		mlxsw_sp_rt6_release(fib_work->fen6_info.rt);
5613 		break;
5614 	case FIB_EVENT_RULE_ADD:
5615 		/* if we get here, a rule was added that we do not support.
5616 		 * just do the fib_abort
5617 		 */
5618 		mlxsw_sp_router_fib_abort(mlxsw_sp);
5619 		break;
5620 	}
5621 	rtnl_unlock();
5622 	kfree(fib_work);
5623 }
5624 
5625 static void mlxsw_sp_router_fibmr_event_work(struct work_struct *work)
5626 {
5627 	struct mlxsw_sp_fib_event_work *fib_work =
5628 		container_of(work, struct mlxsw_sp_fib_event_work, work);
5629 	struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
5630 	bool replace;
5631 	int err;
5632 
5633 	rtnl_lock();
5634 	switch (fib_work->event) {
5635 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
5636 	case FIB_EVENT_ENTRY_ADD:
5637 		replace = fib_work->event == FIB_EVENT_ENTRY_REPLACE;
5638 
5639 		err = mlxsw_sp_router_fibmr_add(mlxsw_sp, &fib_work->men_info,
5640 						replace);
5641 		if (err)
5642 			mlxsw_sp_router_fib_abort(mlxsw_sp);
5643 		ipmr_cache_put(fib_work->men_info.mfc);
5644 		break;
5645 	case FIB_EVENT_ENTRY_DEL:
5646 		mlxsw_sp_router_fibmr_del(mlxsw_sp, &fib_work->men_info);
5647 		ipmr_cache_put(fib_work->men_info.mfc);
5648 		break;
5649 	case FIB_EVENT_VIF_ADD:
5650 		err = mlxsw_sp_router_fibmr_vif_add(mlxsw_sp,
5651 						    &fib_work->ven_info);
5652 		if (err)
5653 			mlxsw_sp_router_fib_abort(mlxsw_sp);
5654 		dev_put(fib_work->ven_info.dev);
5655 		break;
5656 	case FIB_EVENT_VIF_DEL:
5657 		mlxsw_sp_router_fibmr_vif_del(mlxsw_sp,
5658 					      &fib_work->ven_info);
5659 		dev_put(fib_work->ven_info.dev);
5660 		break;
5661 	case FIB_EVENT_RULE_ADD:
5662 		/* if we get here, a rule was added that we do not support.
5663 		 * just do the fib_abort
5664 		 */
5665 		mlxsw_sp_router_fib_abort(mlxsw_sp);
5666 		break;
5667 	}
5668 	rtnl_unlock();
5669 	kfree(fib_work);
5670 }
5671 
5672 static void mlxsw_sp_router_fib4_event(struct mlxsw_sp_fib_event_work *fib_work,
5673 				       struct fib_notifier_info *info)
5674 {
5675 	struct fib_entry_notifier_info *fen_info;
5676 	struct fib_nh_notifier_info *fnh_info;
5677 
5678 	switch (fib_work->event) {
5679 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
5680 	case FIB_EVENT_ENTRY_APPEND: /* fall through */
5681 	case FIB_EVENT_ENTRY_ADD: /* fall through */
5682 	case FIB_EVENT_ENTRY_DEL:
5683 		fen_info = container_of(info, struct fib_entry_notifier_info,
5684 					info);
5685 		fib_work->fen_info = *fen_info;
5686 		/* Take reference on fib_info to prevent it from being
5687 		 * freed while work is queued. Release it afterwards.
5688 		 */
5689 		fib_info_hold(fib_work->fen_info.fi);
5690 		break;
5691 	case FIB_EVENT_NH_ADD: /* fall through */
5692 	case FIB_EVENT_NH_DEL:
5693 		fnh_info = container_of(info, struct fib_nh_notifier_info,
5694 					info);
5695 		fib_work->fnh_info = *fnh_info;
5696 		fib_info_hold(fib_work->fnh_info.fib_nh->nh_parent);
5697 		break;
5698 	}
5699 }
5700 
5701 static void mlxsw_sp_router_fib6_event(struct mlxsw_sp_fib_event_work *fib_work,
5702 				       struct fib_notifier_info *info)
5703 {
5704 	struct fib6_entry_notifier_info *fen6_info;
5705 
5706 	switch (fib_work->event) {
5707 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
5708 	case FIB_EVENT_ENTRY_ADD: /* fall through */
5709 	case FIB_EVENT_ENTRY_DEL:
5710 		fen6_info = container_of(info, struct fib6_entry_notifier_info,
5711 					 info);
5712 		fib_work->fen6_info = *fen6_info;
5713 		rt6_hold(fib_work->fen6_info.rt);
5714 		break;
5715 	}
5716 }
5717 
5718 static void
5719 mlxsw_sp_router_fibmr_event(struct mlxsw_sp_fib_event_work *fib_work,
5720 			    struct fib_notifier_info *info)
5721 {
5722 	switch (fib_work->event) {
5723 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
5724 	case FIB_EVENT_ENTRY_ADD: /* fall through */
5725 	case FIB_EVENT_ENTRY_DEL:
5726 		memcpy(&fib_work->men_info, info, sizeof(fib_work->men_info));
5727 		ipmr_cache_hold(fib_work->men_info.mfc);
5728 		break;
5729 	case FIB_EVENT_VIF_ADD: /* fall through */
5730 	case FIB_EVENT_VIF_DEL:
5731 		memcpy(&fib_work->ven_info, info, sizeof(fib_work->ven_info));
5732 		dev_hold(fib_work->ven_info.dev);
5733 		break;
5734 	}
5735 }
5736 
5737 static int mlxsw_sp_router_fib_rule_event(unsigned long event,
5738 					  struct fib_notifier_info *info,
5739 					  struct mlxsw_sp *mlxsw_sp)
5740 {
5741 	struct netlink_ext_ack *extack = info->extack;
5742 	struct fib_rule_notifier_info *fr_info;
5743 	struct fib_rule *rule;
5744 	int err = 0;
5745 
5746 	/* nothing to do at the moment */
5747 	if (event == FIB_EVENT_RULE_DEL)
5748 		return 0;
5749 
5750 	if (mlxsw_sp->router->aborted)
5751 		return 0;
5752 
5753 	fr_info = container_of(info, struct fib_rule_notifier_info, info);
5754 	rule = fr_info->rule;
5755 
5756 	switch (info->family) {
5757 	case AF_INET:
5758 		if (!fib4_rule_default(rule) && !rule->l3mdev)
5759 			err = -1;
5760 		break;
5761 	case AF_INET6:
5762 		if (!fib6_rule_default(rule) && !rule->l3mdev)
5763 			err = -1;
5764 		break;
5765 	case RTNL_FAMILY_IPMR:
5766 		if (!ipmr_rule_default(rule) && !rule->l3mdev)
5767 			err = -1;
5768 		break;
5769 	}
5770 
5771 	if (err < 0)
5772 		NL_SET_ERR_MSG(extack, "spectrum: FIB rules not supported. Aborting offload");
5773 
5774 	return err;
5775 }
5776 
5777 /* Called with rcu_read_lock() */
5778 static int mlxsw_sp_router_fib_event(struct notifier_block *nb,
5779 				     unsigned long event, void *ptr)
5780 {
5781 	struct mlxsw_sp_fib_event_work *fib_work;
5782 	struct fib_notifier_info *info = ptr;
5783 	struct mlxsw_sp_router *router;
5784 	int err;
5785 
5786 	if (!net_eq(info->net, &init_net) ||
5787 	    (info->family != AF_INET && info->family != AF_INET6 &&
5788 	     info->family != RTNL_FAMILY_IPMR))
5789 		return NOTIFY_DONE;
5790 
5791 	router = container_of(nb, struct mlxsw_sp_router, fib_nb);
5792 
5793 	switch (event) {
5794 	case FIB_EVENT_RULE_ADD: /* fall through */
5795 	case FIB_EVENT_RULE_DEL:
5796 		err = mlxsw_sp_router_fib_rule_event(event, info,
5797 						     router->mlxsw_sp);
5798 		if (!err)
5799 			return NOTIFY_DONE;
5800 	}
5801 
5802 	fib_work = kzalloc(sizeof(*fib_work), GFP_ATOMIC);
5803 	if (WARN_ON(!fib_work))
5804 		return NOTIFY_BAD;
5805 
5806 	fib_work->mlxsw_sp = router->mlxsw_sp;
5807 	fib_work->event = event;
5808 
5809 	switch (info->family) {
5810 	case AF_INET:
5811 		INIT_WORK(&fib_work->work, mlxsw_sp_router_fib4_event_work);
5812 		mlxsw_sp_router_fib4_event(fib_work, info);
5813 		break;
5814 	case AF_INET6:
5815 		INIT_WORK(&fib_work->work, mlxsw_sp_router_fib6_event_work);
5816 		mlxsw_sp_router_fib6_event(fib_work, info);
5817 		break;
5818 	case RTNL_FAMILY_IPMR:
5819 		INIT_WORK(&fib_work->work, mlxsw_sp_router_fibmr_event_work);
5820 		mlxsw_sp_router_fibmr_event(fib_work, info);
5821 		break;
5822 	}
5823 
5824 	mlxsw_core_schedule_work(&fib_work->work);
5825 
5826 	return NOTIFY_DONE;
5827 }
5828 
5829 static struct mlxsw_sp_rif *
5830 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
5831 			 const struct net_device *dev)
5832 {
5833 	int i;
5834 
5835 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
5836 		if (mlxsw_sp->router->rifs[i] &&
5837 		    mlxsw_sp->router->rifs[i]->dev == dev)
5838 			return mlxsw_sp->router->rifs[i];
5839 
5840 	return NULL;
5841 }
5842 
5843 static int mlxsw_sp_router_rif_disable(struct mlxsw_sp *mlxsw_sp, u16 rif)
5844 {
5845 	char ritr_pl[MLXSW_REG_RITR_LEN];
5846 	int err;
5847 
5848 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif);
5849 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
5850 	if (WARN_ON_ONCE(err))
5851 		return err;
5852 
5853 	mlxsw_reg_ritr_enable_set(ritr_pl, false);
5854 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
5855 }
5856 
5857 static void mlxsw_sp_router_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
5858 					  struct mlxsw_sp_rif *rif)
5859 {
5860 	mlxsw_sp_router_rif_disable(mlxsw_sp, rif->rif_index);
5861 	mlxsw_sp_nexthop_rif_gone_sync(mlxsw_sp, rif);
5862 	mlxsw_sp_neigh_rif_gone_sync(mlxsw_sp, rif);
5863 }
5864 
5865 static bool
5866 mlxsw_sp_rif_should_config(struct mlxsw_sp_rif *rif, struct net_device *dev,
5867 			   unsigned long event)
5868 {
5869 	struct inet6_dev *inet6_dev;
5870 	bool addr_list_empty = true;
5871 	struct in_device *idev;
5872 
5873 	switch (event) {
5874 	case NETDEV_UP:
5875 		return rif == NULL;
5876 	case NETDEV_DOWN:
5877 		idev = __in_dev_get_rtnl(dev);
5878 		if (idev && idev->ifa_list)
5879 			addr_list_empty = false;
5880 
5881 		inet6_dev = __in6_dev_get(dev);
5882 		if (addr_list_empty && inet6_dev &&
5883 		    !list_empty(&inet6_dev->addr_list))
5884 			addr_list_empty = false;
5885 
5886 		if (rif && addr_list_empty &&
5887 		    !netif_is_l3_slave(rif->dev))
5888 			return true;
5889 		/* It is possible we already removed the RIF ourselves
5890 		 * if it was assigned to a netdev that is now a bridge
5891 		 * or LAG slave.
5892 		 */
5893 		return false;
5894 	}
5895 
5896 	return false;
5897 }
5898 
5899 static enum mlxsw_sp_rif_type
5900 mlxsw_sp_dev_rif_type(const struct mlxsw_sp *mlxsw_sp,
5901 		      const struct net_device *dev)
5902 {
5903 	enum mlxsw_sp_fid_type type;
5904 
5905 	if (mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL))
5906 		return MLXSW_SP_RIF_TYPE_IPIP_LB;
5907 
5908 	/* Otherwise RIF type is derived from the type of the underlying FID. */
5909 	if (is_vlan_dev(dev) && netif_is_bridge_master(vlan_dev_real_dev(dev)))
5910 		type = MLXSW_SP_FID_TYPE_8021Q;
5911 	else if (netif_is_bridge_master(dev) && br_vlan_enabled(dev))
5912 		type = MLXSW_SP_FID_TYPE_8021Q;
5913 	else if (netif_is_bridge_master(dev))
5914 		type = MLXSW_SP_FID_TYPE_8021D;
5915 	else
5916 		type = MLXSW_SP_FID_TYPE_RFID;
5917 
5918 	return mlxsw_sp_fid_type_rif_type(mlxsw_sp, type);
5919 }
5920 
5921 static int mlxsw_sp_rif_index_alloc(struct mlxsw_sp *mlxsw_sp, u16 *p_rif_index)
5922 {
5923 	int i;
5924 
5925 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++) {
5926 		if (!mlxsw_sp->router->rifs[i]) {
5927 			*p_rif_index = i;
5928 			return 0;
5929 		}
5930 	}
5931 
5932 	return -ENOBUFS;
5933 }
5934 
5935 static struct mlxsw_sp_rif *mlxsw_sp_rif_alloc(size_t rif_size, u16 rif_index,
5936 					       u16 vr_id,
5937 					       struct net_device *l3_dev)
5938 {
5939 	struct mlxsw_sp_rif *rif;
5940 
5941 	rif = kzalloc(rif_size, GFP_KERNEL);
5942 	if (!rif)
5943 		return NULL;
5944 
5945 	INIT_LIST_HEAD(&rif->nexthop_list);
5946 	INIT_LIST_HEAD(&rif->neigh_list);
5947 	ether_addr_copy(rif->addr, l3_dev->dev_addr);
5948 	rif->mtu = l3_dev->mtu;
5949 	rif->vr_id = vr_id;
5950 	rif->dev = l3_dev;
5951 	rif->rif_index = rif_index;
5952 
5953 	return rif;
5954 }
5955 
5956 struct mlxsw_sp_rif *mlxsw_sp_rif_by_index(const struct mlxsw_sp *mlxsw_sp,
5957 					   u16 rif_index)
5958 {
5959 	return mlxsw_sp->router->rifs[rif_index];
5960 }
5961 
5962 u16 mlxsw_sp_rif_index(const struct mlxsw_sp_rif *rif)
5963 {
5964 	return rif->rif_index;
5965 }
5966 
5967 u16 mlxsw_sp_ipip_lb_rif_index(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
5968 {
5969 	return lb_rif->common.rif_index;
5970 }
5971 
5972 u16 mlxsw_sp_ipip_lb_ul_vr_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
5973 {
5974 	return lb_rif->ul_vr_id;
5975 }
5976 
5977 int mlxsw_sp_rif_dev_ifindex(const struct mlxsw_sp_rif *rif)
5978 {
5979 	return rif->dev->ifindex;
5980 }
5981 
5982 const struct net_device *mlxsw_sp_rif_dev(const struct mlxsw_sp_rif *rif)
5983 {
5984 	return rif->dev;
5985 }
5986 
5987 static struct mlxsw_sp_rif *
5988 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
5989 		    const struct mlxsw_sp_rif_params *params,
5990 		    struct netlink_ext_ack *extack)
5991 {
5992 	u32 tb_id = l3mdev_fib_table(params->dev);
5993 	const struct mlxsw_sp_rif_ops *ops;
5994 	struct mlxsw_sp_fid *fid = NULL;
5995 	enum mlxsw_sp_rif_type type;
5996 	struct mlxsw_sp_rif *rif;
5997 	struct mlxsw_sp_vr *vr;
5998 	u16 rif_index;
5999 	int err;
6000 
6001 	type = mlxsw_sp_dev_rif_type(mlxsw_sp, params->dev);
6002 	ops = mlxsw_sp->router->rif_ops_arr[type];
6003 
6004 	vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id ? : RT_TABLE_MAIN, extack);
6005 	if (IS_ERR(vr))
6006 		return ERR_CAST(vr);
6007 	vr->rif_count++;
6008 
6009 	err = mlxsw_sp_rif_index_alloc(mlxsw_sp, &rif_index);
6010 	if (err) {
6011 		NL_SET_ERR_MSG(extack, "spectrum: Exceeded number of supported router interfaces");
6012 		goto err_rif_index_alloc;
6013 	}
6014 
6015 	rif = mlxsw_sp_rif_alloc(ops->rif_size, rif_index, vr->id, params->dev);
6016 	if (!rif) {
6017 		err = -ENOMEM;
6018 		goto err_rif_alloc;
6019 	}
6020 	rif->mlxsw_sp = mlxsw_sp;
6021 	rif->ops = ops;
6022 
6023 	if (ops->fid_get) {
6024 		fid = ops->fid_get(rif);
6025 		if (IS_ERR(fid)) {
6026 			err = PTR_ERR(fid);
6027 			goto err_fid_get;
6028 		}
6029 		rif->fid = fid;
6030 	}
6031 
6032 	if (ops->setup)
6033 		ops->setup(rif, params);
6034 
6035 	err = ops->configure(rif);
6036 	if (err)
6037 		goto err_configure;
6038 
6039 	err = mlxsw_sp_mr_rif_add(vr->mr4_table, rif);
6040 	if (err)
6041 		goto err_mr_rif_add;
6042 
6043 	mlxsw_sp_rif_counters_alloc(rif);
6044 	mlxsw_sp->router->rifs[rif_index] = rif;
6045 
6046 	return rif;
6047 
6048 err_mr_rif_add:
6049 	ops->deconfigure(rif);
6050 err_configure:
6051 	if (fid)
6052 		mlxsw_sp_fid_put(fid);
6053 err_fid_get:
6054 	kfree(rif);
6055 err_rif_alloc:
6056 err_rif_index_alloc:
6057 	vr->rif_count--;
6058 	mlxsw_sp_vr_put(vr);
6059 	return ERR_PTR(err);
6060 }
6061 
6062 void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif)
6063 {
6064 	const struct mlxsw_sp_rif_ops *ops = rif->ops;
6065 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6066 	struct mlxsw_sp_fid *fid = rif->fid;
6067 	struct mlxsw_sp_vr *vr;
6068 
6069 	mlxsw_sp_router_rif_gone_sync(mlxsw_sp, rif);
6070 	vr = &mlxsw_sp->router->vrs[rif->vr_id];
6071 
6072 	mlxsw_sp->router->rifs[rif->rif_index] = NULL;
6073 	mlxsw_sp_rif_counters_free(rif);
6074 	mlxsw_sp_mr_rif_del(vr->mr4_table, rif);
6075 	ops->deconfigure(rif);
6076 	if (fid)
6077 		/* Loopback RIFs are not associated with a FID. */
6078 		mlxsw_sp_fid_put(fid);
6079 	kfree(rif);
6080 	vr->rif_count--;
6081 	mlxsw_sp_vr_put(vr);
6082 }
6083 
6084 static void
6085 mlxsw_sp_rif_subport_params_init(struct mlxsw_sp_rif_params *params,
6086 				 struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
6087 {
6088 	struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
6089 
6090 	params->vid = mlxsw_sp_port_vlan->vid;
6091 	params->lag = mlxsw_sp_port->lagged;
6092 	if (params->lag)
6093 		params->lag_id = mlxsw_sp_port->lag_id;
6094 	else
6095 		params->system_port = mlxsw_sp_port->local_port;
6096 }
6097 
6098 static int
6099 mlxsw_sp_port_vlan_router_join(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan,
6100 			       struct net_device *l3_dev,
6101 			       struct netlink_ext_ack *extack)
6102 {
6103 	struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
6104 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
6105 	u16 vid = mlxsw_sp_port_vlan->vid;
6106 	struct mlxsw_sp_rif *rif;
6107 	struct mlxsw_sp_fid *fid;
6108 	int err;
6109 
6110 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
6111 	if (!rif) {
6112 		struct mlxsw_sp_rif_params params = {
6113 			.dev = l3_dev,
6114 		};
6115 
6116 		mlxsw_sp_rif_subport_params_init(&params, mlxsw_sp_port_vlan);
6117 		rif = mlxsw_sp_rif_create(mlxsw_sp, &params, extack);
6118 		if (IS_ERR(rif))
6119 			return PTR_ERR(rif);
6120 	}
6121 
6122 	/* FID was already created, just take a reference */
6123 	fid = rif->ops->fid_get(rif);
6124 	err = mlxsw_sp_fid_port_vid_map(fid, mlxsw_sp_port, vid);
6125 	if (err)
6126 		goto err_fid_port_vid_map;
6127 
6128 	err = mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, false);
6129 	if (err)
6130 		goto err_port_vid_learning_set;
6131 
6132 	err = mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid,
6133 					BR_STATE_FORWARDING);
6134 	if (err)
6135 		goto err_port_vid_stp_set;
6136 
6137 	mlxsw_sp_port_vlan->fid = fid;
6138 
6139 	return 0;
6140 
6141 err_port_vid_stp_set:
6142 	mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
6143 err_port_vid_learning_set:
6144 	mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
6145 err_fid_port_vid_map:
6146 	mlxsw_sp_fid_put(fid);
6147 	return err;
6148 }
6149 
6150 void
6151 mlxsw_sp_port_vlan_router_leave(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
6152 {
6153 	struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
6154 	struct mlxsw_sp_fid *fid = mlxsw_sp_port_vlan->fid;
6155 	u16 vid = mlxsw_sp_port_vlan->vid;
6156 
6157 	if (WARN_ON(mlxsw_sp_fid_type(fid) != MLXSW_SP_FID_TYPE_RFID))
6158 		return;
6159 
6160 	mlxsw_sp_port_vlan->fid = NULL;
6161 	mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid, BR_STATE_BLOCKING);
6162 	mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
6163 	mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
6164 	/* If router port holds the last reference on the rFID, then the
6165 	 * associated Sub-port RIF will be destroyed.
6166 	 */
6167 	mlxsw_sp_fid_put(fid);
6168 }
6169 
6170 static int mlxsw_sp_inetaddr_port_vlan_event(struct net_device *l3_dev,
6171 					     struct net_device *port_dev,
6172 					     unsigned long event, u16 vid,
6173 					     struct netlink_ext_ack *extack)
6174 {
6175 	struct mlxsw_sp_port *mlxsw_sp_port = netdev_priv(port_dev);
6176 	struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan;
6177 
6178 	mlxsw_sp_port_vlan = mlxsw_sp_port_vlan_find_by_vid(mlxsw_sp_port, vid);
6179 	if (WARN_ON(!mlxsw_sp_port_vlan))
6180 		return -EINVAL;
6181 
6182 	switch (event) {
6183 	case NETDEV_UP:
6184 		return mlxsw_sp_port_vlan_router_join(mlxsw_sp_port_vlan,
6185 						      l3_dev, extack);
6186 	case NETDEV_DOWN:
6187 		mlxsw_sp_port_vlan_router_leave(mlxsw_sp_port_vlan);
6188 		break;
6189 	}
6190 
6191 	return 0;
6192 }
6193 
6194 static int mlxsw_sp_inetaddr_port_event(struct net_device *port_dev,
6195 					unsigned long event,
6196 					struct netlink_ext_ack *extack)
6197 {
6198 	if (netif_is_bridge_port(port_dev) ||
6199 	    netif_is_lag_port(port_dev) ||
6200 	    netif_is_ovs_port(port_dev))
6201 		return 0;
6202 
6203 	return mlxsw_sp_inetaddr_port_vlan_event(port_dev, port_dev, event, 1,
6204 						 extack);
6205 }
6206 
6207 static int __mlxsw_sp_inetaddr_lag_event(struct net_device *l3_dev,
6208 					 struct net_device *lag_dev,
6209 					 unsigned long event, u16 vid,
6210 					 struct netlink_ext_ack *extack)
6211 {
6212 	struct net_device *port_dev;
6213 	struct list_head *iter;
6214 	int err;
6215 
6216 	netdev_for_each_lower_dev(lag_dev, port_dev, iter) {
6217 		if (mlxsw_sp_port_dev_check(port_dev)) {
6218 			err = mlxsw_sp_inetaddr_port_vlan_event(l3_dev,
6219 								port_dev,
6220 								event, vid,
6221 								extack);
6222 			if (err)
6223 				return err;
6224 		}
6225 	}
6226 
6227 	return 0;
6228 }
6229 
6230 static int mlxsw_sp_inetaddr_lag_event(struct net_device *lag_dev,
6231 				       unsigned long event,
6232 				       struct netlink_ext_ack *extack)
6233 {
6234 	if (netif_is_bridge_port(lag_dev))
6235 		return 0;
6236 
6237 	return __mlxsw_sp_inetaddr_lag_event(lag_dev, lag_dev, event, 1,
6238 					     extack);
6239 }
6240 
6241 static int mlxsw_sp_inetaddr_bridge_event(struct net_device *l3_dev,
6242 					  unsigned long event,
6243 					  struct netlink_ext_ack *extack)
6244 {
6245 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_lower_get(l3_dev);
6246 	struct mlxsw_sp_rif_params params = {
6247 		.dev = l3_dev,
6248 	};
6249 	struct mlxsw_sp_rif *rif;
6250 
6251 	switch (event) {
6252 	case NETDEV_UP:
6253 		rif = mlxsw_sp_rif_create(mlxsw_sp, &params, extack);
6254 		if (IS_ERR(rif))
6255 			return PTR_ERR(rif);
6256 		break;
6257 	case NETDEV_DOWN:
6258 		rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
6259 		mlxsw_sp_rif_destroy(rif);
6260 		break;
6261 	}
6262 
6263 	return 0;
6264 }
6265 
6266 static int mlxsw_sp_inetaddr_vlan_event(struct net_device *vlan_dev,
6267 					unsigned long event,
6268 					struct netlink_ext_ack *extack)
6269 {
6270 	struct net_device *real_dev = vlan_dev_real_dev(vlan_dev);
6271 	u16 vid = vlan_dev_vlan_id(vlan_dev);
6272 
6273 	if (netif_is_bridge_port(vlan_dev))
6274 		return 0;
6275 
6276 	if (mlxsw_sp_port_dev_check(real_dev))
6277 		return mlxsw_sp_inetaddr_port_vlan_event(vlan_dev, real_dev,
6278 							 event, vid, extack);
6279 	else if (netif_is_lag_master(real_dev))
6280 		return __mlxsw_sp_inetaddr_lag_event(vlan_dev, real_dev, event,
6281 						     vid, extack);
6282 	else if (netif_is_bridge_master(real_dev) && br_vlan_enabled(real_dev))
6283 		return mlxsw_sp_inetaddr_bridge_event(vlan_dev, event, extack);
6284 
6285 	return 0;
6286 }
6287 
6288 static int __mlxsw_sp_inetaddr_event(struct net_device *dev,
6289 				     unsigned long event,
6290 				     struct netlink_ext_ack *extack)
6291 {
6292 	if (mlxsw_sp_port_dev_check(dev))
6293 		return mlxsw_sp_inetaddr_port_event(dev, event, extack);
6294 	else if (netif_is_lag_master(dev))
6295 		return mlxsw_sp_inetaddr_lag_event(dev, event, extack);
6296 	else if (netif_is_bridge_master(dev))
6297 		return mlxsw_sp_inetaddr_bridge_event(dev, event, extack);
6298 	else if (is_vlan_dev(dev))
6299 		return mlxsw_sp_inetaddr_vlan_event(dev, event, extack);
6300 	else
6301 		return 0;
6302 }
6303 
6304 int mlxsw_sp_inetaddr_event(struct notifier_block *unused,
6305 			    unsigned long event, void *ptr)
6306 {
6307 	struct in_ifaddr *ifa = (struct in_ifaddr *) ptr;
6308 	struct net_device *dev = ifa->ifa_dev->dev;
6309 	struct mlxsw_sp *mlxsw_sp;
6310 	struct mlxsw_sp_rif *rif;
6311 	int err = 0;
6312 
6313 	/* NETDEV_UP event is handled by mlxsw_sp_inetaddr_valid_event */
6314 	if (event == NETDEV_UP)
6315 		goto out;
6316 
6317 	mlxsw_sp = mlxsw_sp_lower_get(dev);
6318 	if (!mlxsw_sp)
6319 		goto out;
6320 
6321 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6322 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
6323 		goto out;
6324 
6325 	err = __mlxsw_sp_inetaddr_event(dev, event, NULL);
6326 out:
6327 	return notifier_from_errno(err);
6328 }
6329 
6330 int mlxsw_sp_inetaddr_valid_event(struct notifier_block *unused,
6331 				  unsigned long event, void *ptr)
6332 {
6333 	struct in_validator_info *ivi = (struct in_validator_info *) ptr;
6334 	struct net_device *dev = ivi->ivi_dev->dev;
6335 	struct mlxsw_sp *mlxsw_sp;
6336 	struct mlxsw_sp_rif *rif;
6337 	int err = 0;
6338 
6339 	mlxsw_sp = mlxsw_sp_lower_get(dev);
6340 	if (!mlxsw_sp)
6341 		goto out;
6342 
6343 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6344 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
6345 		goto out;
6346 
6347 	err = __mlxsw_sp_inetaddr_event(dev, event, ivi->extack);
6348 out:
6349 	return notifier_from_errno(err);
6350 }
6351 
6352 struct mlxsw_sp_inet6addr_event_work {
6353 	struct work_struct work;
6354 	struct net_device *dev;
6355 	unsigned long event;
6356 };
6357 
6358 static void mlxsw_sp_inet6addr_event_work(struct work_struct *work)
6359 {
6360 	struct mlxsw_sp_inet6addr_event_work *inet6addr_work =
6361 		container_of(work, struct mlxsw_sp_inet6addr_event_work, work);
6362 	struct net_device *dev = inet6addr_work->dev;
6363 	unsigned long event = inet6addr_work->event;
6364 	struct mlxsw_sp *mlxsw_sp;
6365 	struct mlxsw_sp_rif *rif;
6366 
6367 	rtnl_lock();
6368 	mlxsw_sp = mlxsw_sp_lower_get(dev);
6369 	if (!mlxsw_sp)
6370 		goto out;
6371 
6372 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6373 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
6374 		goto out;
6375 
6376 	__mlxsw_sp_inetaddr_event(dev, event, NULL);
6377 out:
6378 	rtnl_unlock();
6379 	dev_put(dev);
6380 	kfree(inet6addr_work);
6381 }
6382 
6383 /* Called with rcu_read_lock() */
6384 int mlxsw_sp_inet6addr_event(struct notifier_block *unused,
6385 			     unsigned long event, void *ptr)
6386 {
6387 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *) ptr;
6388 	struct mlxsw_sp_inet6addr_event_work *inet6addr_work;
6389 	struct net_device *dev = if6->idev->dev;
6390 
6391 	/* NETDEV_UP event is handled by mlxsw_sp_inet6addr_valid_event */
6392 	if (event == NETDEV_UP)
6393 		return NOTIFY_DONE;
6394 
6395 	if (!mlxsw_sp_port_dev_lower_find_rcu(dev))
6396 		return NOTIFY_DONE;
6397 
6398 	inet6addr_work = kzalloc(sizeof(*inet6addr_work), GFP_ATOMIC);
6399 	if (!inet6addr_work)
6400 		return NOTIFY_BAD;
6401 
6402 	INIT_WORK(&inet6addr_work->work, mlxsw_sp_inet6addr_event_work);
6403 	inet6addr_work->dev = dev;
6404 	inet6addr_work->event = event;
6405 	dev_hold(dev);
6406 	mlxsw_core_schedule_work(&inet6addr_work->work);
6407 
6408 	return NOTIFY_DONE;
6409 }
6410 
6411 int mlxsw_sp_inet6addr_valid_event(struct notifier_block *unused,
6412 				   unsigned long event, void *ptr)
6413 {
6414 	struct in6_validator_info *i6vi = (struct in6_validator_info *) ptr;
6415 	struct net_device *dev = i6vi->i6vi_dev->dev;
6416 	struct mlxsw_sp *mlxsw_sp;
6417 	struct mlxsw_sp_rif *rif;
6418 	int err = 0;
6419 
6420 	mlxsw_sp = mlxsw_sp_lower_get(dev);
6421 	if (!mlxsw_sp)
6422 		goto out;
6423 
6424 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6425 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
6426 		goto out;
6427 
6428 	err = __mlxsw_sp_inetaddr_event(dev, event, i6vi->extack);
6429 out:
6430 	return notifier_from_errno(err);
6431 }
6432 
6433 static int mlxsw_sp_rif_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
6434 			     const char *mac, int mtu)
6435 {
6436 	char ritr_pl[MLXSW_REG_RITR_LEN];
6437 	int err;
6438 
6439 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
6440 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6441 	if (err)
6442 		return err;
6443 
6444 	mlxsw_reg_ritr_mtu_set(ritr_pl, mtu);
6445 	mlxsw_reg_ritr_if_mac_memcpy_to(ritr_pl, mac);
6446 	mlxsw_reg_ritr_op_set(ritr_pl, MLXSW_REG_RITR_RIF_CREATE);
6447 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6448 }
6449 
6450 int mlxsw_sp_netdevice_router_port_event(struct net_device *dev)
6451 {
6452 	struct mlxsw_sp *mlxsw_sp;
6453 	struct mlxsw_sp_rif *rif;
6454 	u16 fid_index;
6455 	int err;
6456 
6457 	mlxsw_sp = mlxsw_sp_lower_get(dev);
6458 	if (!mlxsw_sp)
6459 		return 0;
6460 
6461 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6462 	if (!rif)
6463 		return 0;
6464 	fid_index = mlxsw_sp_fid_index(rif->fid);
6465 
6466 	err = mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, false);
6467 	if (err)
6468 		return err;
6469 
6470 	err = mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, dev->dev_addr,
6471 				dev->mtu);
6472 	if (err)
6473 		goto err_rif_edit;
6474 
6475 	err = mlxsw_sp_rif_fdb_op(mlxsw_sp, dev->dev_addr, fid_index, true);
6476 	if (err)
6477 		goto err_rif_fdb_op;
6478 
6479 	if (rif->mtu != dev->mtu) {
6480 		struct mlxsw_sp_vr *vr;
6481 
6482 		/* The RIF is relevant only to its mr_table instance, as unlike
6483 		 * unicast routing, in multicast routing a RIF cannot be shared
6484 		 * between several multicast routing tables.
6485 		 */
6486 		vr = &mlxsw_sp->router->vrs[rif->vr_id];
6487 		mlxsw_sp_mr_rif_mtu_update(vr->mr4_table, rif, dev->mtu);
6488 	}
6489 
6490 	ether_addr_copy(rif->addr, dev->dev_addr);
6491 	rif->mtu = dev->mtu;
6492 
6493 	netdev_dbg(dev, "Updated RIF=%d\n", rif->rif_index);
6494 
6495 	return 0;
6496 
6497 err_rif_fdb_op:
6498 	mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, rif->addr, rif->mtu);
6499 err_rif_edit:
6500 	mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, true);
6501 	return err;
6502 }
6503 
6504 static int mlxsw_sp_port_vrf_join(struct mlxsw_sp *mlxsw_sp,
6505 				  struct net_device *l3_dev,
6506 				  struct netlink_ext_ack *extack)
6507 {
6508 	struct mlxsw_sp_rif *rif;
6509 
6510 	/* If netdev is already associated with a RIF, then we need to
6511 	 * destroy it and create a new one with the new virtual router ID.
6512 	 */
6513 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
6514 	if (rif)
6515 		__mlxsw_sp_inetaddr_event(l3_dev, NETDEV_DOWN, extack);
6516 
6517 	return __mlxsw_sp_inetaddr_event(l3_dev, NETDEV_UP, extack);
6518 }
6519 
6520 static void mlxsw_sp_port_vrf_leave(struct mlxsw_sp *mlxsw_sp,
6521 				    struct net_device *l3_dev)
6522 {
6523 	struct mlxsw_sp_rif *rif;
6524 
6525 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
6526 	if (!rif)
6527 		return;
6528 	__mlxsw_sp_inetaddr_event(l3_dev, NETDEV_DOWN, NULL);
6529 }
6530 
6531 int mlxsw_sp_netdevice_vrf_event(struct net_device *l3_dev, unsigned long event,
6532 				 struct netdev_notifier_changeupper_info *info)
6533 {
6534 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_lower_get(l3_dev);
6535 	int err = 0;
6536 
6537 	if (!mlxsw_sp)
6538 		return 0;
6539 
6540 	switch (event) {
6541 	case NETDEV_PRECHANGEUPPER:
6542 		return 0;
6543 	case NETDEV_CHANGEUPPER:
6544 		if (info->linking) {
6545 			struct netlink_ext_ack *extack;
6546 
6547 			extack = netdev_notifier_info_to_extack(&info->info);
6548 			err = mlxsw_sp_port_vrf_join(mlxsw_sp, l3_dev, extack);
6549 		} else {
6550 			mlxsw_sp_port_vrf_leave(mlxsw_sp, l3_dev);
6551 		}
6552 		break;
6553 	}
6554 
6555 	return err;
6556 }
6557 
6558 static struct mlxsw_sp_rif_subport *
6559 mlxsw_sp_rif_subport_rif(const struct mlxsw_sp_rif *rif)
6560 {
6561 	return container_of(rif, struct mlxsw_sp_rif_subport, common);
6562 }
6563 
6564 static void mlxsw_sp_rif_subport_setup(struct mlxsw_sp_rif *rif,
6565 				       const struct mlxsw_sp_rif_params *params)
6566 {
6567 	struct mlxsw_sp_rif_subport *rif_subport;
6568 
6569 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
6570 	rif_subport->vid = params->vid;
6571 	rif_subport->lag = params->lag;
6572 	if (params->lag)
6573 		rif_subport->lag_id = params->lag_id;
6574 	else
6575 		rif_subport->system_port = params->system_port;
6576 }
6577 
6578 static int mlxsw_sp_rif_subport_op(struct mlxsw_sp_rif *rif, bool enable)
6579 {
6580 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6581 	struct mlxsw_sp_rif_subport *rif_subport;
6582 	char ritr_pl[MLXSW_REG_RITR_LEN];
6583 
6584 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
6585 	mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_SP_IF,
6586 			    rif->rif_index, rif->vr_id, rif->dev->mtu);
6587 	mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
6588 	mlxsw_reg_ritr_sp_if_pack(ritr_pl, rif_subport->lag,
6589 				  rif_subport->lag ? rif_subport->lag_id :
6590 						     rif_subport->system_port,
6591 				  rif_subport->vid);
6592 
6593 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6594 }
6595 
6596 static int mlxsw_sp_rif_subport_configure(struct mlxsw_sp_rif *rif)
6597 {
6598 	int err;
6599 
6600 	err = mlxsw_sp_rif_subport_op(rif, true);
6601 	if (err)
6602 		return err;
6603 
6604 	err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
6605 				  mlxsw_sp_fid_index(rif->fid), true);
6606 	if (err)
6607 		goto err_rif_fdb_op;
6608 
6609 	mlxsw_sp_fid_rif_set(rif->fid, rif);
6610 	return 0;
6611 
6612 err_rif_fdb_op:
6613 	mlxsw_sp_rif_subport_op(rif, false);
6614 	return err;
6615 }
6616 
6617 static void mlxsw_sp_rif_subport_deconfigure(struct mlxsw_sp_rif *rif)
6618 {
6619 	struct mlxsw_sp_fid *fid = rif->fid;
6620 
6621 	mlxsw_sp_fid_rif_set(fid, NULL);
6622 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
6623 			    mlxsw_sp_fid_index(fid), false);
6624 	mlxsw_sp_rif_subport_op(rif, false);
6625 }
6626 
6627 static struct mlxsw_sp_fid *
6628 mlxsw_sp_rif_subport_fid_get(struct mlxsw_sp_rif *rif)
6629 {
6630 	return mlxsw_sp_fid_rfid_get(rif->mlxsw_sp, rif->rif_index);
6631 }
6632 
6633 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_subport_ops = {
6634 	.type			= MLXSW_SP_RIF_TYPE_SUBPORT,
6635 	.rif_size		= sizeof(struct mlxsw_sp_rif_subport),
6636 	.setup			= mlxsw_sp_rif_subport_setup,
6637 	.configure		= mlxsw_sp_rif_subport_configure,
6638 	.deconfigure		= mlxsw_sp_rif_subport_deconfigure,
6639 	.fid_get		= mlxsw_sp_rif_subport_fid_get,
6640 };
6641 
6642 static int mlxsw_sp_rif_vlan_fid_op(struct mlxsw_sp_rif *rif,
6643 				    enum mlxsw_reg_ritr_if_type type,
6644 				    u16 vid_fid, bool enable)
6645 {
6646 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6647 	char ritr_pl[MLXSW_REG_RITR_LEN];
6648 
6649 	mlxsw_reg_ritr_pack(ritr_pl, enable, type, rif->rif_index, rif->vr_id,
6650 			    rif->dev->mtu);
6651 	mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
6652 	mlxsw_reg_ritr_fid_set(ritr_pl, type, vid_fid);
6653 
6654 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6655 }
6656 
6657 u8 mlxsw_sp_router_port(const struct mlxsw_sp *mlxsw_sp)
6658 {
6659 	return mlxsw_core_max_ports(mlxsw_sp->core) + 1;
6660 }
6661 
6662 static int mlxsw_sp_rif_vlan_configure(struct mlxsw_sp_rif *rif)
6663 {
6664 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6665 	u16 vid = mlxsw_sp_fid_8021q_vid(rif->fid);
6666 	int err;
6667 
6668 	err = mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_VLAN_IF, vid, true);
6669 	if (err)
6670 		return err;
6671 
6672 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
6673 				     mlxsw_sp_router_port(mlxsw_sp), true);
6674 	if (err)
6675 		goto err_fid_mc_flood_set;
6676 
6677 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
6678 				     mlxsw_sp_router_port(mlxsw_sp), true);
6679 	if (err)
6680 		goto err_fid_bc_flood_set;
6681 
6682 	err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
6683 				  mlxsw_sp_fid_index(rif->fid), true);
6684 	if (err)
6685 		goto err_rif_fdb_op;
6686 
6687 	mlxsw_sp_fid_rif_set(rif->fid, rif);
6688 	return 0;
6689 
6690 err_rif_fdb_op:
6691 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
6692 			       mlxsw_sp_router_port(mlxsw_sp), false);
6693 err_fid_bc_flood_set:
6694 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
6695 			       mlxsw_sp_router_port(mlxsw_sp), false);
6696 err_fid_mc_flood_set:
6697 	mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_VLAN_IF, vid, false);
6698 	return err;
6699 }
6700 
6701 static void mlxsw_sp_rif_vlan_deconfigure(struct mlxsw_sp_rif *rif)
6702 {
6703 	u16 vid = mlxsw_sp_fid_8021q_vid(rif->fid);
6704 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6705 	struct mlxsw_sp_fid *fid = rif->fid;
6706 
6707 	mlxsw_sp_fid_rif_set(fid, NULL);
6708 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
6709 			    mlxsw_sp_fid_index(fid), false);
6710 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
6711 			       mlxsw_sp_router_port(mlxsw_sp), false);
6712 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
6713 			       mlxsw_sp_router_port(mlxsw_sp), false);
6714 	mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_VLAN_IF, vid, false);
6715 }
6716 
6717 static struct mlxsw_sp_fid *
6718 mlxsw_sp_rif_vlan_fid_get(struct mlxsw_sp_rif *rif)
6719 {
6720 	u16 vid = is_vlan_dev(rif->dev) ? vlan_dev_vlan_id(rif->dev) : 1;
6721 
6722 	return mlxsw_sp_fid_8021q_get(rif->mlxsw_sp, vid);
6723 }
6724 
6725 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_vlan_ops = {
6726 	.type			= MLXSW_SP_RIF_TYPE_VLAN,
6727 	.rif_size		= sizeof(struct mlxsw_sp_rif),
6728 	.configure		= mlxsw_sp_rif_vlan_configure,
6729 	.deconfigure		= mlxsw_sp_rif_vlan_deconfigure,
6730 	.fid_get		= mlxsw_sp_rif_vlan_fid_get,
6731 };
6732 
6733 static int mlxsw_sp_rif_fid_configure(struct mlxsw_sp_rif *rif)
6734 {
6735 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6736 	u16 fid_index = mlxsw_sp_fid_index(rif->fid);
6737 	int err;
6738 
6739 	err = mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index,
6740 				       true);
6741 	if (err)
6742 		return err;
6743 
6744 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
6745 				     mlxsw_sp_router_port(mlxsw_sp), true);
6746 	if (err)
6747 		goto err_fid_mc_flood_set;
6748 
6749 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
6750 				     mlxsw_sp_router_port(mlxsw_sp), true);
6751 	if (err)
6752 		goto err_fid_bc_flood_set;
6753 
6754 	err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
6755 				  mlxsw_sp_fid_index(rif->fid), true);
6756 	if (err)
6757 		goto err_rif_fdb_op;
6758 
6759 	mlxsw_sp_fid_rif_set(rif->fid, rif);
6760 	return 0;
6761 
6762 err_rif_fdb_op:
6763 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
6764 			       mlxsw_sp_router_port(mlxsw_sp), false);
6765 err_fid_bc_flood_set:
6766 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
6767 			       mlxsw_sp_router_port(mlxsw_sp), false);
6768 err_fid_mc_flood_set:
6769 	mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index, false);
6770 	return err;
6771 }
6772 
6773 static void mlxsw_sp_rif_fid_deconfigure(struct mlxsw_sp_rif *rif)
6774 {
6775 	u16 fid_index = mlxsw_sp_fid_index(rif->fid);
6776 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6777 	struct mlxsw_sp_fid *fid = rif->fid;
6778 
6779 	mlxsw_sp_fid_rif_set(fid, NULL);
6780 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
6781 			    mlxsw_sp_fid_index(fid), false);
6782 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
6783 			       mlxsw_sp_router_port(mlxsw_sp), false);
6784 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
6785 			       mlxsw_sp_router_port(mlxsw_sp), false);
6786 	mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index, false);
6787 }
6788 
6789 static struct mlxsw_sp_fid *
6790 mlxsw_sp_rif_fid_fid_get(struct mlxsw_sp_rif *rif)
6791 {
6792 	return mlxsw_sp_fid_8021d_get(rif->mlxsw_sp, rif->dev->ifindex);
6793 }
6794 
6795 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_fid_ops = {
6796 	.type			= MLXSW_SP_RIF_TYPE_FID,
6797 	.rif_size		= sizeof(struct mlxsw_sp_rif),
6798 	.configure		= mlxsw_sp_rif_fid_configure,
6799 	.deconfigure		= mlxsw_sp_rif_fid_deconfigure,
6800 	.fid_get		= mlxsw_sp_rif_fid_fid_get,
6801 };
6802 
6803 static struct mlxsw_sp_rif_ipip_lb *
6804 mlxsw_sp_rif_ipip_lb_rif(struct mlxsw_sp_rif *rif)
6805 {
6806 	return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
6807 }
6808 
6809 static void
6810 mlxsw_sp_rif_ipip_lb_setup(struct mlxsw_sp_rif *rif,
6811 			   const struct mlxsw_sp_rif_params *params)
6812 {
6813 	struct mlxsw_sp_rif_params_ipip_lb *params_lb;
6814 	struct mlxsw_sp_rif_ipip_lb *rif_lb;
6815 
6816 	params_lb = container_of(params, struct mlxsw_sp_rif_params_ipip_lb,
6817 				 common);
6818 	rif_lb = mlxsw_sp_rif_ipip_lb_rif(rif);
6819 	rif_lb->lb_config = params_lb->lb_config;
6820 }
6821 
6822 static int
6823 mlxsw_sp_rif_ipip_lb_op(struct mlxsw_sp_rif_ipip_lb *lb_rif,
6824 			struct mlxsw_sp_vr *ul_vr, bool enable)
6825 {
6826 	struct mlxsw_sp_rif_ipip_lb_config lb_cf = lb_rif->lb_config;
6827 	struct mlxsw_sp_rif *rif = &lb_rif->common;
6828 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6829 	char ritr_pl[MLXSW_REG_RITR_LEN];
6830 	u32 saddr4;
6831 
6832 	switch (lb_cf.ul_protocol) {
6833 	case MLXSW_SP_L3_PROTO_IPV4:
6834 		saddr4 = be32_to_cpu(lb_cf.saddr.addr4);
6835 		mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
6836 				    rif->rif_index, rif->vr_id, rif->dev->mtu);
6837 		mlxsw_reg_ritr_loopback_ipip4_pack(ritr_pl, lb_cf.lb_ipipt,
6838 			    MLXSW_REG_RITR_LOOPBACK_IPIP_OPTIONS_GRE_KEY_PRESET,
6839 			    ul_vr->id, saddr4, lb_cf.okey);
6840 		break;
6841 
6842 	case MLXSW_SP_L3_PROTO_IPV6:
6843 		return -EAFNOSUPPORT;
6844 	}
6845 
6846 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6847 }
6848 
6849 static int
6850 mlxsw_sp_rif_ipip_lb_configure(struct mlxsw_sp_rif *rif)
6851 {
6852 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
6853 	u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(rif->dev);
6854 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6855 	struct mlxsw_sp_vr *ul_vr;
6856 	int err;
6857 
6858 	ul_vr = mlxsw_sp_vr_get(mlxsw_sp, ul_tb_id, NULL);
6859 	if (IS_ERR(ul_vr))
6860 		return PTR_ERR(ul_vr);
6861 
6862 	err = mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr, true);
6863 	if (err)
6864 		goto err_loopback_op;
6865 
6866 	lb_rif->ul_vr_id = ul_vr->id;
6867 	++ul_vr->rif_count;
6868 	return 0;
6869 
6870 err_loopback_op:
6871 	mlxsw_sp_vr_put(ul_vr);
6872 	return err;
6873 }
6874 
6875 static void mlxsw_sp_rif_ipip_lb_deconfigure(struct mlxsw_sp_rif *rif)
6876 {
6877 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
6878 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6879 	struct mlxsw_sp_vr *ul_vr;
6880 
6881 	ul_vr = &mlxsw_sp->router->vrs[lb_rif->ul_vr_id];
6882 	mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr, false);
6883 
6884 	--ul_vr->rif_count;
6885 	mlxsw_sp_vr_put(ul_vr);
6886 }
6887 
6888 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_ipip_lb_ops = {
6889 	.type			= MLXSW_SP_RIF_TYPE_IPIP_LB,
6890 	.rif_size		= sizeof(struct mlxsw_sp_rif_ipip_lb),
6891 	.setup                  = mlxsw_sp_rif_ipip_lb_setup,
6892 	.configure		= mlxsw_sp_rif_ipip_lb_configure,
6893 	.deconfigure		= mlxsw_sp_rif_ipip_lb_deconfigure,
6894 };
6895 
6896 static const struct mlxsw_sp_rif_ops *mlxsw_sp_rif_ops_arr[] = {
6897 	[MLXSW_SP_RIF_TYPE_SUBPORT]	= &mlxsw_sp_rif_subport_ops,
6898 	[MLXSW_SP_RIF_TYPE_VLAN]	= &mlxsw_sp_rif_vlan_ops,
6899 	[MLXSW_SP_RIF_TYPE_FID]		= &mlxsw_sp_rif_fid_ops,
6900 	[MLXSW_SP_RIF_TYPE_IPIP_LB]	= &mlxsw_sp_rif_ipip_lb_ops,
6901 };
6902 
6903 static int mlxsw_sp_rifs_init(struct mlxsw_sp *mlxsw_sp)
6904 {
6905 	u64 max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
6906 
6907 	mlxsw_sp->router->rifs = kcalloc(max_rifs,
6908 					 sizeof(struct mlxsw_sp_rif *),
6909 					 GFP_KERNEL);
6910 	if (!mlxsw_sp->router->rifs)
6911 		return -ENOMEM;
6912 
6913 	mlxsw_sp->router->rif_ops_arr = mlxsw_sp_rif_ops_arr;
6914 
6915 	return 0;
6916 }
6917 
6918 static void mlxsw_sp_rifs_fini(struct mlxsw_sp *mlxsw_sp)
6919 {
6920 	int i;
6921 
6922 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
6923 		WARN_ON_ONCE(mlxsw_sp->router->rifs[i]);
6924 
6925 	kfree(mlxsw_sp->router->rifs);
6926 }
6927 
6928 static int
6929 mlxsw_sp_ipip_config_tigcr(struct mlxsw_sp *mlxsw_sp)
6930 {
6931 	char tigcr_pl[MLXSW_REG_TIGCR_LEN];
6932 
6933 	mlxsw_reg_tigcr_pack(tigcr_pl, true, 0);
6934 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tigcr), tigcr_pl);
6935 }
6936 
6937 static int mlxsw_sp_ipips_init(struct mlxsw_sp *mlxsw_sp)
6938 {
6939 	mlxsw_sp->router->ipip_ops_arr = mlxsw_sp_ipip_ops_arr;
6940 	INIT_LIST_HEAD(&mlxsw_sp->router->ipip_list);
6941 	return mlxsw_sp_ipip_config_tigcr(mlxsw_sp);
6942 }
6943 
6944 static void mlxsw_sp_ipips_fini(struct mlxsw_sp *mlxsw_sp)
6945 {
6946 	WARN_ON(!list_empty(&mlxsw_sp->router->ipip_list));
6947 }
6948 
6949 static void mlxsw_sp_router_fib_dump_flush(struct notifier_block *nb)
6950 {
6951 	struct mlxsw_sp_router *router;
6952 
6953 	/* Flush pending FIB notifications and then flush the device's
6954 	 * table before requesting another dump. The FIB notification
6955 	 * block is unregistered, so no need to take RTNL.
6956 	 */
6957 	mlxsw_core_flush_owq();
6958 	router = container_of(nb, struct mlxsw_sp_router, fib_nb);
6959 	mlxsw_sp_router_fib_flush(router->mlxsw_sp);
6960 }
6961 
6962 #ifdef CONFIG_IP_ROUTE_MULTIPATH
6963 static void mlxsw_sp_mp_hash_header_set(char *recr2_pl, int header)
6964 {
6965 	mlxsw_reg_recr2_outer_header_enables_set(recr2_pl, header, true);
6966 }
6967 
6968 static void mlxsw_sp_mp_hash_field_set(char *recr2_pl, int field)
6969 {
6970 	mlxsw_reg_recr2_outer_header_fields_enable_set(recr2_pl, field, true);
6971 }
6972 
6973 static void mlxsw_sp_mp4_hash_init(char *recr2_pl)
6974 {
6975 	bool only_l3 = !init_net.ipv4.sysctl_fib_multipath_hash_policy;
6976 
6977 	mlxsw_sp_mp_hash_header_set(recr2_pl,
6978 				    MLXSW_REG_RECR2_IPV4_EN_NOT_TCP_NOT_UDP);
6979 	mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_IPV4_EN_TCP_UDP);
6980 	mlxsw_reg_recr2_ipv4_sip_enable(recr2_pl);
6981 	mlxsw_reg_recr2_ipv4_dip_enable(recr2_pl);
6982 	if (only_l3)
6983 		return;
6984 	mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_EN_IPV4);
6985 	mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_IPV4_PROTOCOL);
6986 	mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_SPORT);
6987 	mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_DPORT);
6988 }
6989 
6990 static void mlxsw_sp_mp6_hash_init(char *recr2_pl)
6991 {
6992 	mlxsw_sp_mp_hash_header_set(recr2_pl,
6993 				    MLXSW_REG_RECR2_IPV6_EN_NOT_TCP_NOT_UDP);
6994 	mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_IPV6_EN_TCP_UDP);
6995 	mlxsw_reg_recr2_ipv6_sip_enable(recr2_pl);
6996 	mlxsw_reg_recr2_ipv6_dip_enable(recr2_pl);
6997 	mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_IPV6_FLOW_LABEL);
6998 	mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_IPV6_NEXT_HEADER);
6999 }
7000 
7001 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
7002 {
7003 	char recr2_pl[MLXSW_REG_RECR2_LEN];
7004 	u32 seed;
7005 
7006 	get_random_bytes(&seed, sizeof(seed));
7007 	mlxsw_reg_recr2_pack(recr2_pl, seed);
7008 	mlxsw_sp_mp4_hash_init(recr2_pl);
7009 	mlxsw_sp_mp6_hash_init(recr2_pl);
7010 
7011 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(recr2), recr2_pl);
7012 }
7013 #else
7014 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
7015 {
7016 	return 0;
7017 }
7018 #endif
7019 
7020 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
7021 {
7022 	char rgcr_pl[MLXSW_REG_RGCR_LEN];
7023 	u64 max_rifs;
7024 	int err;
7025 
7026 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_RIFS))
7027 		return -EIO;
7028 	max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
7029 
7030 	mlxsw_reg_rgcr_pack(rgcr_pl, true, true);
7031 	mlxsw_reg_rgcr_max_router_interfaces_set(rgcr_pl, max_rifs);
7032 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
7033 	if (err)
7034 		return err;
7035 	return 0;
7036 }
7037 
7038 static void __mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
7039 {
7040 	char rgcr_pl[MLXSW_REG_RGCR_LEN];
7041 
7042 	mlxsw_reg_rgcr_pack(rgcr_pl, false, false);
7043 	mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
7044 }
7045 
7046 int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
7047 {
7048 	struct mlxsw_sp_router *router;
7049 	int err;
7050 
7051 	router = kzalloc(sizeof(*mlxsw_sp->router), GFP_KERNEL);
7052 	if (!router)
7053 		return -ENOMEM;
7054 	mlxsw_sp->router = router;
7055 	router->mlxsw_sp = mlxsw_sp;
7056 
7057 	INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_neighs_list);
7058 	err = __mlxsw_sp_router_init(mlxsw_sp);
7059 	if (err)
7060 		goto err_router_init;
7061 
7062 	err = mlxsw_sp_rifs_init(mlxsw_sp);
7063 	if (err)
7064 		goto err_rifs_init;
7065 
7066 	err = mlxsw_sp_ipips_init(mlxsw_sp);
7067 	if (err)
7068 		goto err_ipips_init;
7069 
7070 	err = rhashtable_init(&mlxsw_sp->router->nexthop_ht,
7071 			      &mlxsw_sp_nexthop_ht_params);
7072 	if (err)
7073 		goto err_nexthop_ht_init;
7074 
7075 	err = rhashtable_init(&mlxsw_sp->router->nexthop_group_ht,
7076 			      &mlxsw_sp_nexthop_group_ht_params);
7077 	if (err)
7078 		goto err_nexthop_group_ht_init;
7079 
7080 	INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_list);
7081 	err = mlxsw_sp_lpm_init(mlxsw_sp);
7082 	if (err)
7083 		goto err_lpm_init;
7084 
7085 	err = mlxsw_sp_mr_init(mlxsw_sp, &mlxsw_sp_mr_tcam_ops);
7086 	if (err)
7087 		goto err_mr_init;
7088 
7089 	err = mlxsw_sp_vrs_init(mlxsw_sp);
7090 	if (err)
7091 		goto err_vrs_init;
7092 
7093 	err = mlxsw_sp_neigh_init(mlxsw_sp);
7094 	if (err)
7095 		goto err_neigh_init;
7096 
7097 	mlxsw_sp->router->netevent_nb.notifier_call =
7098 		mlxsw_sp_router_netevent_event;
7099 	err = register_netevent_notifier(&mlxsw_sp->router->netevent_nb);
7100 	if (err)
7101 		goto err_register_netevent_notifier;
7102 
7103 	err = mlxsw_sp_mp_hash_init(mlxsw_sp);
7104 	if (err)
7105 		goto err_mp_hash_init;
7106 
7107 	mlxsw_sp->router->fib_nb.notifier_call = mlxsw_sp_router_fib_event;
7108 	err = register_fib_notifier(&mlxsw_sp->router->fib_nb,
7109 				    mlxsw_sp_router_fib_dump_flush);
7110 	if (err)
7111 		goto err_register_fib_notifier;
7112 
7113 	return 0;
7114 
7115 err_register_fib_notifier:
7116 err_mp_hash_init:
7117 	unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
7118 err_register_netevent_notifier:
7119 	mlxsw_sp_neigh_fini(mlxsw_sp);
7120 err_neigh_init:
7121 	mlxsw_sp_vrs_fini(mlxsw_sp);
7122 err_vrs_init:
7123 	mlxsw_sp_mr_fini(mlxsw_sp);
7124 err_mr_init:
7125 	mlxsw_sp_lpm_fini(mlxsw_sp);
7126 err_lpm_init:
7127 	rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
7128 err_nexthop_group_ht_init:
7129 	rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
7130 err_nexthop_ht_init:
7131 	mlxsw_sp_ipips_fini(mlxsw_sp);
7132 err_ipips_init:
7133 	mlxsw_sp_rifs_fini(mlxsw_sp);
7134 err_rifs_init:
7135 	__mlxsw_sp_router_fini(mlxsw_sp);
7136 err_router_init:
7137 	kfree(mlxsw_sp->router);
7138 	return err;
7139 }
7140 
7141 void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
7142 {
7143 	unregister_fib_notifier(&mlxsw_sp->router->fib_nb);
7144 	unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
7145 	mlxsw_sp_neigh_fini(mlxsw_sp);
7146 	mlxsw_sp_vrs_fini(mlxsw_sp);
7147 	mlxsw_sp_mr_fini(mlxsw_sp);
7148 	mlxsw_sp_lpm_fini(mlxsw_sp);
7149 	rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
7150 	rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
7151 	mlxsw_sp_ipips_fini(mlxsw_sp);
7152 	mlxsw_sp_rifs_fini(mlxsw_sp);
7153 	__mlxsw_sp_router_fini(mlxsw_sp);
7154 	kfree(mlxsw_sp->router);
7155 }
7156