1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
2 /* Copyright (c) 2016-2018 Mellanox Technologies. All rights reserved */
3 
4 #include <linux/kernel.h>
5 #include <linux/types.h>
6 #include <linux/rhashtable.h>
7 #include <linux/bitops.h>
8 #include <linux/in6.h>
9 #include <linux/notifier.h>
10 #include <linux/inetdevice.h>
11 #include <linux/netdevice.h>
12 #include <linux/if_bridge.h>
13 #include <linux/socket.h>
14 #include <linux/route.h>
15 #include <linux/gcd.h>
16 #include <linux/if_macvlan.h>
17 #include <linux/refcount.h>
18 #include <linux/jhash.h>
19 #include <linux/net_namespace.h>
20 #include <linux/mutex.h>
21 #include <net/netevent.h>
22 #include <net/neighbour.h>
23 #include <net/arp.h>
24 #include <net/inet_dscp.h>
25 #include <net/ip_fib.h>
26 #include <net/ip6_fib.h>
27 #include <net/nexthop.h>
28 #include <net/fib_rules.h>
29 #include <net/ip_tunnels.h>
30 #include <net/l3mdev.h>
31 #include <net/addrconf.h>
32 #include <net/ndisc.h>
33 #include <net/ipv6.h>
34 #include <net/fib_notifier.h>
35 #include <net/switchdev.h>
36 
37 #include "spectrum.h"
38 #include "core.h"
39 #include "reg.h"
40 #include "spectrum_cnt.h"
41 #include "spectrum_dpipe.h"
42 #include "spectrum_ipip.h"
43 #include "spectrum_mr.h"
44 #include "spectrum_mr_tcam.h"
45 #include "spectrum_router.h"
46 #include "spectrum_span.h"
47 
48 struct mlxsw_sp_fib;
49 struct mlxsw_sp_vr;
50 struct mlxsw_sp_lpm_tree;
51 struct mlxsw_sp_rif_ops;
52 
53 struct mlxsw_sp_rif {
54 	struct list_head nexthop_list;
55 	struct list_head neigh_list;
56 	struct net_device *dev; /* NULL for underlay RIF */
57 	struct mlxsw_sp_fid *fid;
58 	unsigned char addr[ETH_ALEN];
59 	int mtu;
60 	u16 rif_index;
61 	u8 mac_profile_id;
62 	u16 vr_id;
63 	const struct mlxsw_sp_rif_ops *ops;
64 	struct mlxsw_sp *mlxsw_sp;
65 
66 	unsigned int counter_ingress;
67 	bool counter_ingress_valid;
68 	unsigned int counter_egress;
69 	bool counter_egress_valid;
70 };
71 
72 struct mlxsw_sp_rif_params {
73 	struct net_device *dev;
74 	union {
75 		u16 system_port;
76 		u16 lag_id;
77 	};
78 	u16 vid;
79 	bool lag;
80 };
81 
82 struct mlxsw_sp_rif_subport {
83 	struct mlxsw_sp_rif common;
84 	refcount_t ref_count;
85 	union {
86 		u16 system_port;
87 		u16 lag_id;
88 	};
89 	u16 vid;
90 	bool lag;
91 };
92 
93 struct mlxsw_sp_rif_ipip_lb {
94 	struct mlxsw_sp_rif common;
95 	struct mlxsw_sp_rif_ipip_lb_config lb_config;
96 	u16 ul_vr_id; /* Reserved for Spectrum-2. */
97 	u16 ul_rif_id; /* Reserved for Spectrum. */
98 };
99 
100 struct mlxsw_sp_rif_params_ipip_lb {
101 	struct mlxsw_sp_rif_params common;
102 	struct mlxsw_sp_rif_ipip_lb_config lb_config;
103 };
104 
105 struct mlxsw_sp_rif_ops {
106 	enum mlxsw_sp_rif_type type;
107 	size_t rif_size;
108 
109 	void (*setup)(struct mlxsw_sp_rif *rif,
110 		      const struct mlxsw_sp_rif_params *params);
111 	int (*configure)(struct mlxsw_sp_rif *rif,
112 			 struct netlink_ext_ack *extack);
113 	void (*deconfigure)(struct mlxsw_sp_rif *rif);
114 	struct mlxsw_sp_fid * (*fid_get)(struct mlxsw_sp_rif *rif,
115 					 struct netlink_ext_ack *extack);
116 	void (*fdb_del)(struct mlxsw_sp_rif *rif, const char *mac);
117 };
118 
119 struct mlxsw_sp_rif_mac_profile {
120 	unsigned char mac_prefix[ETH_ALEN];
121 	refcount_t ref_count;
122 	u8 id;
123 };
124 
125 struct mlxsw_sp_router_ops {
126 	int (*init)(struct mlxsw_sp *mlxsw_sp);
127 	int (*ipips_init)(struct mlxsw_sp *mlxsw_sp);
128 };
129 
130 static struct mlxsw_sp_rif *
131 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
132 			 const struct net_device *dev);
133 static void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif);
134 static void mlxsw_sp_lpm_tree_hold(struct mlxsw_sp_lpm_tree *lpm_tree);
135 static void mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
136 				  struct mlxsw_sp_lpm_tree *lpm_tree);
137 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
138 				     const struct mlxsw_sp_fib *fib,
139 				     u8 tree_id);
140 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
141 				       const struct mlxsw_sp_fib *fib);
142 
143 static unsigned int *
144 mlxsw_sp_rif_p_counter_get(struct mlxsw_sp_rif *rif,
145 			   enum mlxsw_sp_rif_counter_dir dir)
146 {
147 	switch (dir) {
148 	case MLXSW_SP_RIF_COUNTER_EGRESS:
149 		return &rif->counter_egress;
150 	case MLXSW_SP_RIF_COUNTER_INGRESS:
151 		return &rif->counter_ingress;
152 	}
153 	return NULL;
154 }
155 
156 static bool
157 mlxsw_sp_rif_counter_valid_get(struct mlxsw_sp_rif *rif,
158 			       enum mlxsw_sp_rif_counter_dir dir)
159 {
160 	switch (dir) {
161 	case MLXSW_SP_RIF_COUNTER_EGRESS:
162 		return rif->counter_egress_valid;
163 	case MLXSW_SP_RIF_COUNTER_INGRESS:
164 		return rif->counter_ingress_valid;
165 	}
166 	return false;
167 }
168 
169 static void
170 mlxsw_sp_rif_counter_valid_set(struct mlxsw_sp_rif *rif,
171 			       enum mlxsw_sp_rif_counter_dir dir,
172 			       bool valid)
173 {
174 	switch (dir) {
175 	case MLXSW_SP_RIF_COUNTER_EGRESS:
176 		rif->counter_egress_valid = valid;
177 		break;
178 	case MLXSW_SP_RIF_COUNTER_INGRESS:
179 		rif->counter_ingress_valid = valid;
180 		break;
181 	}
182 }
183 
184 static int mlxsw_sp_rif_counter_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
185 				     unsigned int counter_index, bool enable,
186 				     enum mlxsw_sp_rif_counter_dir dir)
187 {
188 	char ritr_pl[MLXSW_REG_RITR_LEN];
189 	bool is_egress = false;
190 	int err;
191 
192 	if (dir == MLXSW_SP_RIF_COUNTER_EGRESS)
193 		is_egress = true;
194 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
195 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
196 	if (err)
197 		return err;
198 
199 	mlxsw_reg_ritr_counter_pack(ritr_pl, counter_index, enable,
200 				    is_egress);
201 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
202 }
203 
204 int mlxsw_sp_rif_counter_value_get(struct mlxsw_sp *mlxsw_sp,
205 				   struct mlxsw_sp_rif *rif,
206 				   enum mlxsw_sp_rif_counter_dir dir, u64 *cnt)
207 {
208 	char ricnt_pl[MLXSW_REG_RICNT_LEN];
209 	unsigned int *p_counter_index;
210 	bool valid;
211 	int err;
212 
213 	valid = mlxsw_sp_rif_counter_valid_get(rif, dir);
214 	if (!valid)
215 		return -EINVAL;
216 
217 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
218 	if (!p_counter_index)
219 		return -EINVAL;
220 	mlxsw_reg_ricnt_pack(ricnt_pl, *p_counter_index,
221 			     MLXSW_REG_RICNT_OPCODE_NOP);
222 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
223 	if (err)
224 		return err;
225 	*cnt = mlxsw_reg_ricnt_good_unicast_packets_get(ricnt_pl);
226 	return 0;
227 }
228 
229 struct mlxsw_sp_rif_counter_set_basic {
230 	u64 good_unicast_packets;
231 	u64 good_multicast_packets;
232 	u64 good_broadcast_packets;
233 	u64 good_unicast_bytes;
234 	u64 good_multicast_bytes;
235 	u64 good_broadcast_bytes;
236 	u64 error_packets;
237 	u64 discard_packets;
238 	u64 error_bytes;
239 	u64 discard_bytes;
240 };
241 
242 static int
243 mlxsw_sp_rif_counter_fetch_clear(struct mlxsw_sp_rif *rif,
244 				 enum mlxsw_sp_rif_counter_dir dir,
245 				 struct mlxsw_sp_rif_counter_set_basic *set)
246 {
247 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
248 	char ricnt_pl[MLXSW_REG_RICNT_LEN];
249 	unsigned int *p_counter_index;
250 	int err;
251 
252 	if (!mlxsw_sp_rif_counter_valid_get(rif, dir))
253 		return -EINVAL;
254 
255 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
256 	if (!p_counter_index)
257 		return -EINVAL;
258 
259 	mlxsw_reg_ricnt_pack(ricnt_pl, *p_counter_index,
260 			     MLXSW_REG_RICNT_OPCODE_CLEAR);
261 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
262 	if (err)
263 		return err;
264 
265 	if (!set)
266 		return 0;
267 
268 #define MLXSW_SP_RIF_COUNTER_EXTRACT(NAME)				\
269 		(set->NAME = mlxsw_reg_ricnt_ ## NAME ## _get(ricnt_pl))
270 
271 	MLXSW_SP_RIF_COUNTER_EXTRACT(good_unicast_packets);
272 	MLXSW_SP_RIF_COUNTER_EXTRACT(good_multicast_packets);
273 	MLXSW_SP_RIF_COUNTER_EXTRACT(good_broadcast_packets);
274 	MLXSW_SP_RIF_COUNTER_EXTRACT(good_unicast_bytes);
275 	MLXSW_SP_RIF_COUNTER_EXTRACT(good_multicast_bytes);
276 	MLXSW_SP_RIF_COUNTER_EXTRACT(good_broadcast_bytes);
277 	MLXSW_SP_RIF_COUNTER_EXTRACT(error_packets);
278 	MLXSW_SP_RIF_COUNTER_EXTRACT(discard_packets);
279 	MLXSW_SP_RIF_COUNTER_EXTRACT(error_bytes);
280 	MLXSW_SP_RIF_COUNTER_EXTRACT(discard_bytes);
281 
282 #undef MLXSW_SP_RIF_COUNTER_EXTRACT
283 
284 	return 0;
285 }
286 
287 static int mlxsw_sp_rif_counter_clear(struct mlxsw_sp *mlxsw_sp,
288 				      unsigned int counter_index)
289 {
290 	char ricnt_pl[MLXSW_REG_RICNT_LEN];
291 
292 	mlxsw_reg_ricnt_pack(ricnt_pl, counter_index,
293 			     MLXSW_REG_RICNT_OPCODE_CLEAR);
294 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
295 }
296 
297 int mlxsw_sp_rif_counter_alloc(struct mlxsw_sp_rif *rif,
298 			       enum mlxsw_sp_rif_counter_dir dir)
299 {
300 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
301 	unsigned int *p_counter_index;
302 	int err;
303 
304 	if (mlxsw_sp_rif_counter_valid_get(rif, dir))
305 		return 0;
306 
307 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
308 	if (!p_counter_index)
309 		return -EINVAL;
310 
311 	err = mlxsw_sp_counter_alloc(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
312 				     p_counter_index);
313 	if (err)
314 		return err;
315 
316 	err = mlxsw_sp_rif_counter_clear(mlxsw_sp, *p_counter_index);
317 	if (err)
318 		goto err_counter_clear;
319 
320 	err = mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
321 					*p_counter_index, true, dir);
322 	if (err)
323 		goto err_counter_edit;
324 	mlxsw_sp_rif_counter_valid_set(rif, dir, true);
325 	return 0;
326 
327 err_counter_edit:
328 err_counter_clear:
329 	mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
330 			      *p_counter_index);
331 	return err;
332 }
333 
334 void mlxsw_sp_rif_counter_free(struct mlxsw_sp_rif *rif,
335 			       enum mlxsw_sp_rif_counter_dir dir)
336 {
337 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
338 	unsigned int *p_counter_index;
339 
340 	if (!mlxsw_sp_rif_counter_valid_get(rif, dir))
341 		return;
342 
343 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
344 	if (WARN_ON(!p_counter_index))
345 		return;
346 	mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
347 				  *p_counter_index, false, dir);
348 	mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
349 			      *p_counter_index);
350 	mlxsw_sp_rif_counter_valid_set(rif, dir, false);
351 }
352 
353 static void mlxsw_sp_rif_counters_alloc(struct mlxsw_sp_rif *rif)
354 {
355 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
356 	struct devlink *devlink;
357 
358 	devlink = priv_to_devlink(mlxsw_sp->core);
359 	if (!devlink_dpipe_table_counter_enabled(devlink,
360 						 MLXSW_SP_DPIPE_TABLE_NAME_ERIF))
361 		return;
362 	mlxsw_sp_rif_counter_alloc(rif, MLXSW_SP_RIF_COUNTER_EGRESS);
363 }
364 
365 static void mlxsw_sp_rif_counters_free(struct mlxsw_sp_rif *rif)
366 {
367 	mlxsw_sp_rif_counter_free(rif, MLXSW_SP_RIF_COUNTER_EGRESS);
368 }
369 
370 #define MLXSW_SP_PREFIX_COUNT (sizeof(struct in6_addr) * BITS_PER_BYTE + 1)
371 
372 struct mlxsw_sp_prefix_usage {
373 	DECLARE_BITMAP(b, MLXSW_SP_PREFIX_COUNT);
374 };
375 
376 #define mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) \
377 	for_each_set_bit(prefix, (prefix_usage)->b, MLXSW_SP_PREFIX_COUNT)
378 
379 static bool
380 mlxsw_sp_prefix_usage_eq(struct mlxsw_sp_prefix_usage *prefix_usage1,
381 			 struct mlxsw_sp_prefix_usage *prefix_usage2)
382 {
383 	return !memcmp(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
384 }
385 
386 static void
387 mlxsw_sp_prefix_usage_cpy(struct mlxsw_sp_prefix_usage *prefix_usage1,
388 			  struct mlxsw_sp_prefix_usage *prefix_usage2)
389 {
390 	memcpy(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
391 }
392 
393 static void
394 mlxsw_sp_prefix_usage_set(struct mlxsw_sp_prefix_usage *prefix_usage,
395 			  unsigned char prefix_len)
396 {
397 	set_bit(prefix_len, prefix_usage->b);
398 }
399 
400 static void
401 mlxsw_sp_prefix_usage_clear(struct mlxsw_sp_prefix_usage *prefix_usage,
402 			    unsigned char prefix_len)
403 {
404 	clear_bit(prefix_len, prefix_usage->b);
405 }
406 
407 struct mlxsw_sp_fib_key {
408 	unsigned char addr[sizeof(struct in6_addr)];
409 	unsigned char prefix_len;
410 };
411 
412 enum mlxsw_sp_fib_entry_type {
413 	MLXSW_SP_FIB_ENTRY_TYPE_REMOTE,
414 	MLXSW_SP_FIB_ENTRY_TYPE_LOCAL,
415 	MLXSW_SP_FIB_ENTRY_TYPE_TRAP,
416 	MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE,
417 	MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE,
418 
419 	/* This is a special case of local delivery, where a packet should be
420 	 * decapsulated on reception. Note that there is no corresponding ENCAP,
421 	 * because that's a type of next hop, not of FIB entry. (There can be
422 	 * several next hops in a REMOTE entry, and some of them may be
423 	 * encapsulating entries.)
424 	 */
425 	MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP,
426 	MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP,
427 };
428 
429 struct mlxsw_sp_nexthop_group_info;
430 struct mlxsw_sp_nexthop_group;
431 struct mlxsw_sp_fib_entry;
432 
433 struct mlxsw_sp_fib_node {
434 	struct mlxsw_sp_fib_entry *fib_entry;
435 	struct list_head list;
436 	struct rhash_head ht_node;
437 	struct mlxsw_sp_fib *fib;
438 	struct mlxsw_sp_fib_key key;
439 };
440 
441 struct mlxsw_sp_fib_entry_decap {
442 	struct mlxsw_sp_ipip_entry *ipip_entry;
443 	u32 tunnel_index;
444 };
445 
446 struct mlxsw_sp_fib_entry {
447 	struct mlxsw_sp_fib_node *fib_node;
448 	enum mlxsw_sp_fib_entry_type type;
449 	struct list_head nexthop_group_node;
450 	struct mlxsw_sp_nexthop_group *nh_group;
451 	struct mlxsw_sp_fib_entry_decap decap; /* Valid for decap entries. */
452 };
453 
454 struct mlxsw_sp_fib4_entry {
455 	struct mlxsw_sp_fib_entry common;
456 	struct fib_info *fi;
457 	u32 tb_id;
458 	dscp_t dscp;
459 	u8 type;
460 };
461 
462 struct mlxsw_sp_fib6_entry {
463 	struct mlxsw_sp_fib_entry common;
464 	struct list_head rt6_list;
465 	unsigned int nrt6;
466 };
467 
468 struct mlxsw_sp_rt6 {
469 	struct list_head list;
470 	struct fib6_info *rt;
471 };
472 
473 struct mlxsw_sp_lpm_tree {
474 	u8 id; /* tree ID */
475 	unsigned int ref_count;
476 	enum mlxsw_sp_l3proto proto;
477 	unsigned long prefix_ref_count[MLXSW_SP_PREFIX_COUNT];
478 	struct mlxsw_sp_prefix_usage prefix_usage;
479 };
480 
481 struct mlxsw_sp_fib {
482 	struct rhashtable ht;
483 	struct list_head node_list;
484 	struct mlxsw_sp_vr *vr;
485 	struct mlxsw_sp_lpm_tree *lpm_tree;
486 	enum mlxsw_sp_l3proto proto;
487 };
488 
489 struct mlxsw_sp_vr {
490 	u16 id; /* virtual router ID */
491 	u32 tb_id; /* kernel fib table id */
492 	unsigned int rif_count;
493 	struct mlxsw_sp_fib *fib4;
494 	struct mlxsw_sp_fib *fib6;
495 	struct mlxsw_sp_mr_table *mr_table[MLXSW_SP_L3_PROTO_MAX];
496 	struct mlxsw_sp_rif *ul_rif;
497 	refcount_t ul_rif_refcnt;
498 };
499 
500 static const struct rhashtable_params mlxsw_sp_fib_ht_params;
501 
502 static struct mlxsw_sp_fib *mlxsw_sp_fib_create(struct mlxsw_sp *mlxsw_sp,
503 						struct mlxsw_sp_vr *vr,
504 						enum mlxsw_sp_l3proto proto)
505 {
506 	struct mlxsw_sp_lpm_tree *lpm_tree;
507 	struct mlxsw_sp_fib *fib;
508 	int err;
509 
510 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[proto];
511 	fib = kzalloc(sizeof(*fib), GFP_KERNEL);
512 	if (!fib)
513 		return ERR_PTR(-ENOMEM);
514 	err = rhashtable_init(&fib->ht, &mlxsw_sp_fib_ht_params);
515 	if (err)
516 		goto err_rhashtable_init;
517 	INIT_LIST_HEAD(&fib->node_list);
518 	fib->proto = proto;
519 	fib->vr = vr;
520 	fib->lpm_tree = lpm_tree;
521 	mlxsw_sp_lpm_tree_hold(lpm_tree);
522 	err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, lpm_tree->id);
523 	if (err)
524 		goto err_lpm_tree_bind;
525 	return fib;
526 
527 err_lpm_tree_bind:
528 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
529 err_rhashtable_init:
530 	kfree(fib);
531 	return ERR_PTR(err);
532 }
533 
534 static void mlxsw_sp_fib_destroy(struct mlxsw_sp *mlxsw_sp,
535 				 struct mlxsw_sp_fib *fib)
536 {
537 	mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, fib);
538 	mlxsw_sp_lpm_tree_put(mlxsw_sp, fib->lpm_tree);
539 	WARN_ON(!list_empty(&fib->node_list));
540 	rhashtable_destroy(&fib->ht);
541 	kfree(fib);
542 }
543 
544 static struct mlxsw_sp_lpm_tree *
545 mlxsw_sp_lpm_tree_find_unused(struct mlxsw_sp *mlxsw_sp)
546 {
547 	static struct mlxsw_sp_lpm_tree *lpm_tree;
548 	int i;
549 
550 	for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
551 		lpm_tree = &mlxsw_sp->router->lpm.trees[i];
552 		if (lpm_tree->ref_count == 0)
553 			return lpm_tree;
554 	}
555 	return NULL;
556 }
557 
558 static int mlxsw_sp_lpm_tree_alloc(struct mlxsw_sp *mlxsw_sp,
559 				   struct mlxsw_sp_lpm_tree *lpm_tree)
560 {
561 	char ralta_pl[MLXSW_REG_RALTA_LEN];
562 
563 	mlxsw_reg_ralta_pack(ralta_pl, true,
564 			     (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
565 			     lpm_tree->id);
566 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
567 }
568 
569 static void mlxsw_sp_lpm_tree_free(struct mlxsw_sp *mlxsw_sp,
570 				   struct mlxsw_sp_lpm_tree *lpm_tree)
571 {
572 	char ralta_pl[MLXSW_REG_RALTA_LEN];
573 
574 	mlxsw_reg_ralta_pack(ralta_pl, false,
575 			     (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
576 			     lpm_tree->id);
577 	mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
578 }
579 
580 static int
581 mlxsw_sp_lpm_tree_left_struct_set(struct mlxsw_sp *mlxsw_sp,
582 				  struct mlxsw_sp_prefix_usage *prefix_usage,
583 				  struct mlxsw_sp_lpm_tree *lpm_tree)
584 {
585 	char ralst_pl[MLXSW_REG_RALST_LEN];
586 	u8 root_bin = 0;
587 	u8 prefix;
588 	u8 last_prefix = MLXSW_REG_RALST_BIN_NO_CHILD;
589 
590 	mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage)
591 		root_bin = prefix;
592 
593 	mlxsw_reg_ralst_pack(ralst_pl, root_bin, lpm_tree->id);
594 	mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) {
595 		if (prefix == 0)
596 			continue;
597 		mlxsw_reg_ralst_bin_pack(ralst_pl, prefix, last_prefix,
598 					 MLXSW_REG_RALST_BIN_NO_CHILD);
599 		last_prefix = prefix;
600 	}
601 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
602 }
603 
604 static struct mlxsw_sp_lpm_tree *
605 mlxsw_sp_lpm_tree_create(struct mlxsw_sp *mlxsw_sp,
606 			 struct mlxsw_sp_prefix_usage *prefix_usage,
607 			 enum mlxsw_sp_l3proto proto)
608 {
609 	struct mlxsw_sp_lpm_tree *lpm_tree;
610 	int err;
611 
612 	lpm_tree = mlxsw_sp_lpm_tree_find_unused(mlxsw_sp);
613 	if (!lpm_tree)
614 		return ERR_PTR(-EBUSY);
615 	lpm_tree->proto = proto;
616 	err = mlxsw_sp_lpm_tree_alloc(mlxsw_sp, lpm_tree);
617 	if (err)
618 		return ERR_PTR(err);
619 
620 	err = mlxsw_sp_lpm_tree_left_struct_set(mlxsw_sp, prefix_usage,
621 						lpm_tree);
622 	if (err)
623 		goto err_left_struct_set;
624 	memcpy(&lpm_tree->prefix_usage, prefix_usage,
625 	       sizeof(lpm_tree->prefix_usage));
626 	memset(&lpm_tree->prefix_ref_count, 0,
627 	       sizeof(lpm_tree->prefix_ref_count));
628 	lpm_tree->ref_count = 1;
629 	return lpm_tree;
630 
631 err_left_struct_set:
632 	mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
633 	return ERR_PTR(err);
634 }
635 
636 static void mlxsw_sp_lpm_tree_destroy(struct mlxsw_sp *mlxsw_sp,
637 				      struct mlxsw_sp_lpm_tree *lpm_tree)
638 {
639 	mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
640 }
641 
642 static struct mlxsw_sp_lpm_tree *
643 mlxsw_sp_lpm_tree_get(struct mlxsw_sp *mlxsw_sp,
644 		      struct mlxsw_sp_prefix_usage *prefix_usage,
645 		      enum mlxsw_sp_l3proto proto)
646 {
647 	struct mlxsw_sp_lpm_tree *lpm_tree;
648 	int i;
649 
650 	for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
651 		lpm_tree = &mlxsw_sp->router->lpm.trees[i];
652 		if (lpm_tree->ref_count != 0 &&
653 		    lpm_tree->proto == proto &&
654 		    mlxsw_sp_prefix_usage_eq(&lpm_tree->prefix_usage,
655 					     prefix_usage)) {
656 			mlxsw_sp_lpm_tree_hold(lpm_tree);
657 			return lpm_tree;
658 		}
659 	}
660 	return mlxsw_sp_lpm_tree_create(mlxsw_sp, prefix_usage, proto);
661 }
662 
663 static void mlxsw_sp_lpm_tree_hold(struct mlxsw_sp_lpm_tree *lpm_tree)
664 {
665 	lpm_tree->ref_count++;
666 }
667 
668 static void mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
669 				  struct mlxsw_sp_lpm_tree *lpm_tree)
670 {
671 	if (--lpm_tree->ref_count == 0)
672 		mlxsw_sp_lpm_tree_destroy(mlxsw_sp, lpm_tree);
673 }
674 
675 #define MLXSW_SP_LPM_TREE_MIN 1 /* tree 0 is reserved */
676 
677 static int mlxsw_sp_lpm_init(struct mlxsw_sp *mlxsw_sp)
678 {
679 	struct mlxsw_sp_prefix_usage req_prefix_usage = {{ 0 } };
680 	struct mlxsw_sp_lpm_tree *lpm_tree;
681 	u64 max_trees;
682 	int err, i;
683 
684 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_LPM_TREES))
685 		return -EIO;
686 
687 	max_trees = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_LPM_TREES);
688 	mlxsw_sp->router->lpm.tree_count = max_trees - MLXSW_SP_LPM_TREE_MIN;
689 	mlxsw_sp->router->lpm.trees = kcalloc(mlxsw_sp->router->lpm.tree_count,
690 					     sizeof(struct mlxsw_sp_lpm_tree),
691 					     GFP_KERNEL);
692 	if (!mlxsw_sp->router->lpm.trees)
693 		return -ENOMEM;
694 
695 	for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
696 		lpm_tree = &mlxsw_sp->router->lpm.trees[i];
697 		lpm_tree->id = i + MLXSW_SP_LPM_TREE_MIN;
698 	}
699 
700 	lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
701 					 MLXSW_SP_L3_PROTO_IPV4);
702 	if (IS_ERR(lpm_tree)) {
703 		err = PTR_ERR(lpm_tree);
704 		goto err_ipv4_tree_get;
705 	}
706 	mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4] = lpm_tree;
707 
708 	lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
709 					 MLXSW_SP_L3_PROTO_IPV6);
710 	if (IS_ERR(lpm_tree)) {
711 		err = PTR_ERR(lpm_tree);
712 		goto err_ipv6_tree_get;
713 	}
714 	mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV6] = lpm_tree;
715 
716 	return 0;
717 
718 err_ipv6_tree_get:
719 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4];
720 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
721 err_ipv4_tree_get:
722 	kfree(mlxsw_sp->router->lpm.trees);
723 	return err;
724 }
725 
726 static void mlxsw_sp_lpm_fini(struct mlxsw_sp *mlxsw_sp)
727 {
728 	struct mlxsw_sp_lpm_tree *lpm_tree;
729 
730 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV6];
731 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
732 
733 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4];
734 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
735 
736 	kfree(mlxsw_sp->router->lpm.trees);
737 }
738 
739 static bool mlxsw_sp_vr_is_used(const struct mlxsw_sp_vr *vr)
740 {
741 	return !!vr->fib4 || !!vr->fib6 ||
742 	       !!vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] ||
743 	       !!vr->mr_table[MLXSW_SP_L3_PROTO_IPV6];
744 }
745 
746 static struct mlxsw_sp_vr *mlxsw_sp_vr_find_unused(struct mlxsw_sp *mlxsw_sp)
747 {
748 	struct mlxsw_sp_vr *vr;
749 	int i;
750 
751 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
752 		vr = &mlxsw_sp->router->vrs[i];
753 		if (!mlxsw_sp_vr_is_used(vr))
754 			return vr;
755 	}
756 	return NULL;
757 }
758 
759 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
760 				     const struct mlxsw_sp_fib *fib, u8 tree_id)
761 {
762 	char raltb_pl[MLXSW_REG_RALTB_LEN];
763 
764 	mlxsw_reg_raltb_pack(raltb_pl, fib->vr->id,
765 			     (enum mlxsw_reg_ralxx_protocol) fib->proto,
766 			     tree_id);
767 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
768 }
769 
770 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
771 				       const struct mlxsw_sp_fib *fib)
772 {
773 	char raltb_pl[MLXSW_REG_RALTB_LEN];
774 
775 	/* Bind to tree 0 which is default */
776 	mlxsw_reg_raltb_pack(raltb_pl, fib->vr->id,
777 			     (enum mlxsw_reg_ralxx_protocol) fib->proto, 0);
778 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
779 }
780 
781 static u32 mlxsw_sp_fix_tb_id(u32 tb_id)
782 {
783 	/* For our purpose, squash main, default and local tables into one */
784 	if (tb_id == RT_TABLE_LOCAL || tb_id == RT_TABLE_DEFAULT)
785 		tb_id = RT_TABLE_MAIN;
786 	return tb_id;
787 }
788 
789 static struct mlxsw_sp_vr *mlxsw_sp_vr_find(struct mlxsw_sp *mlxsw_sp,
790 					    u32 tb_id)
791 {
792 	struct mlxsw_sp_vr *vr;
793 	int i;
794 
795 	tb_id = mlxsw_sp_fix_tb_id(tb_id);
796 
797 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
798 		vr = &mlxsw_sp->router->vrs[i];
799 		if (mlxsw_sp_vr_is_used(vr) && vr->tb_id == tb_id)
800 			return vr;
801 	}
802 	return NULL;
803 }
804 
805 int mlxsw_sp_router_tb_id_vr_id(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
806 				u16 *vr_id)
807 {
808 	struct mlxsw_sp_vr *vr;
809 	int err = 0;
810 
811 	mutex_lock(&mlxsw_sp->router->lock);
812 	vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
813 	if (!vr) {
814 		err = -ESRCH;
815 		goto out;
816 	}
817 	*vr_id = vr->id;
818 out:
819 	mutex_unlock(&mlxsw_sp->router->lock);
820 	return err;
821 }
822 
823 static struct mlxsw_sp_fib *mlxsw_sp_vr_fib(const struct mlxsw_sp_vr *vr,
824 					    enum mlxsw_sp_l3proto proto)
825 {
826 	switch (proto) {
827 	case MLXSW_SP_L3_PROTO_IPV4:
828 		return vr->fib4;
829 	case MLXSW_SP_L3_PROTO_IPV6:
830 		return vr->fib6;
831 	}
832 	return NULL;
833 }
834 
835 static struct mlxsw_sp_vr *mlxsw_sp_vr_create(struct mlxsw_sp *mlxsw_sp,
836 					      u32 tb_id,
837 					      struct netlink_ext_ack *extack)
838 {
839 	struct mlxsw_sp_mr_table *mr4_table, *mr6_table;
840 	struct mlxsw_sp_fib *fib4;
841 	struct mlxsw_sp_fib *fib6;
842 	struct mlxsw_sp_vr *vr;
843 	int err;
844 
845 	vr = mlxsw_sp_vr_find_unused(mlxsw_sp);
846 	if (!vr) {
847 		NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported virtual routers");
848 		return ERR_PTR(-EBUSY);
849 	}
850 	fib4 = mlxsw_sp_fib_create(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV4);
851 	if (IS_ERR(fib4))
852 		return ERR_CAST(fib4);
853 	fib6 = mlxsw_sp_fib_create(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV6);
854 	if (IS_ERR(fib6)) {
855 		err = PTR_ERR(fib6);
856 		goto err_fib6_create;
857 	}
858 	mr4_table = mlxsw_sp_mr_table_create(mlxsw_sp, vr->id,
859 					     MLXSW_SP_L3_PROTO_IPV4);
860 	if (IS_ERR(mr4_table)) {
861 		err = PTR_ERR(mr4_table);
862 		goto err_mr4_table_create;
863 	}
864 	mr6_table = mlxsw_sp_mr_table_create(mlxsw_sp, vr->id,
865 					     MLXSW_SP_L3_PROTO_IPV6);
866 	if (IS_ERR(mr6_table)) {
867 		err = PTR_ERR(mr6_table);
868 		goto err_mr6_table_create;
869 	}
870 
871 	vr->fib4 = fib4;
872 	vr->fib6 = fib6;
873 	vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] = mr4_table;
874 	vr->mr_table[MLXSW_SP_L3_PROTO_IPV6] = mr6_table;
875 	vr->tb_id = tb_id;
876 	return vr;
877 
878 err_mr6_table_create:
879 	mlxsw_sp_mr_table_destroy(mr4_table);
880 err_mr4_table_create:
881 	mlxsw_sp_fib_destroy(mlxsw_sp, fib6);
882 err_fib6_create:
883 	mlxsw_sp_fib_destroy(mlxsw_sp, fib4);
884 	return ERR_PTR(err);
885 }
886 
887 static void mlxsw_sp_vr_destroy(struct mlxsw_sp *mlxsw_sp,
888 				struct mlxsw_sp_vr *vr)
889 {
890 	mlxsw_sp_mr_table_destroy(vr->mr_table[MLXSW_SP_L3_PROTO_IPV6]);
891 	vr->mr_table[MLXSW_SP_L3_PROTO_IPV6] = NULL;
892 	mlxsw_sp_mr_table_destroy(vr->mr_table[MLXSW_SP_L3_PROTO_IPV4]);
893 	vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] = NULL;
894 	mlxsw_sp_fib_destroy(mlxsw_sp, vr->fib6);
895 	vr->fib6 = NULL;
896 	mlxsw_sp_fib_destroy(mlxsw_sp, vr->fib4);
897 	vr->fib4 = NULL;
898 }
899 
900 static struct mlxsw_sp_vr *mlxsw_sp_vr_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
901 					   struct netlink_ext_ack *extack)
902 {
903 	struct mlxsw_sp_vr *vr;
904 
905 	tb_id = mlxsw_sp_fix_tb_id(tb_id);
906 	vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
907 	if (!vr)
908 		vr = mlxsw_sp_vr_create(mlxsw_sp, tb_id, extack);
909 	return vr;
910 }
911 
912 static void mlxsw_sp_vr_put(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr)
913 {
914 	if (!vr->rif_count && list_empty(&vr->fib4->node_list) &&
915 	    list_empty(&vr->fib6->node_list) &&
916 	    mlxsw_sp_mr_table_empty(vr->mr_table[MLXSW_SP_L3_PROTO_IPV4]) &&
917 	    mlxsw_sp_mr_table_empty(vr->mr_table[MLXSW_SP_L3_PROTO_IPV6]))
918 		mlxsw_sp_vr_destroy(mlxsw_sp, vr);
919 }
920 
921 static bool
922 mlxsw_sp_vr_lpm_tree_should_replace(struct mlxsw_sp_vr *vr,
923 				    enum mlxsw_sp_l3proto proto, u8 tree_id)
924 {
925 	struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
926 
927 	if (!mlxsw_sp_vr_is_used(vr))
928 		return false;
929 	if (fib->lpm_tree->id == tree_id)
930 		return true;
931 	return false;
932 }
933 
934 static int mlxsw_sp_vr_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
935 					struct mlxsw_sp_fib *fib,
936 					struct mlxsw_sp_lpm_tree *new_tree)
937 {
938 	struct mlxsw_sp_lpm_tree *old_tree = fib->lpm_tree;
939 	int err;
940 
941 	fib->lpm_tree = new_tree;
942 	mlxsw_sp_lpm_tree_hold(new_tree);
943 	err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, new_tree->id);
944 	if (err)
945 		goto err_tree_bind;
946 	mlxsw_sp_lpm_tree_put(mlxsw_sp, old_tree);
947 	return 0;
948 
949 err_tree_bind:
950 	mlxsw_sp_lpm_tree_put(mlxsw_sp, new_tree);
951 	fib->lpm_tree = old_tree;
952 	return err;
953 }
954 
955 static int mlxsw_sp_vrs_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
956 					 struct mlxsw_sp_fib *fib,
957 					 struct mlxsw_sp_lpm_tree *new_tree)
958 {
959 	enum mlxsw_sp_l3proto proto = fib->proto;
960 	struct mlxsw_sp_lpm_tree *old_tree;
961 	u8 old_id, new_id = new_tree->id;
962 	struct mlxsw_sp_vr *vr;
963 	int i, err;
964 
965 	old_tree = mlxsw_sp->router->lpm.proto_trees[proto];
966 	old_id = old_tree->id;
967 
968 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
969 		vr = &mlxsw_sp->router->vrs[i];
970 		if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, old_id))
971 			continue;
972 		err = mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
973 						   mlxsw_sp_vr_fib(vr, proto),
974 						   new_tree);
975 		if (err)
976 			goto err_tree_replace;
977 	}
978 
979 	memcpy(new_tree->prefix_ref_count, old_tree->prefix_ref_count,
980 	       sizeof(new_tree->prefix_ref_count));
981 	mlxsw_sp->router->lpm.proto_trees[proto] = new_tree;
982 	mlxsw_sp_lpm_tree_put(mlxsw_sp, old_tree);
983 
984 	return 0;
985 
986 err_tree_replace:
987 	for (i--; i >= 0; i--) {
988 		if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, new_id))
989 			continue;
990 		mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
991 					     mlxsw_sp_vr_fib(vr, proto),
992 					     old_tree);
993 	}
994 	return err;
995 }
996 
997 static int mlxsw_sp_vrs_init(struct mlxsw_sp *mlxsw_sp)
998 {
999 	struct mlxsw_sp_vr *vr;
1000 	u64 max_vrs;
1001 	int i;
1002 
1003 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_VRS))
1004 		return -EIO;
1005 
1006 	max_vrs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS);
1007 	mlxsw_sp->router->vrs = kcalloc(max_vrs, sizeof(struct mlxsw_sp_vr),
1008 					GFP_KERNEL);
1009 	if (!mlxsw_sp->router->vrs)
1010 		return -ENOMEM;
1011 
1012 	for (i = 0; i < max_vrs; i++) {
1013 		vr = &mlxsw_sp->router->vrs[i];
1014 		vr->id = i;
1015 	}
1016 
1017 	return 0;
1018 }
1019 
1020 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp);
1021 
1022 static void mlxsw_sp_vrs_fini(struct mlxsw_sp *mlxsw_sp)
1023 {
1024 	/* At this stage we're guaranteed not to have new incoming
1025 	 * FIB notifications and the work queue is free from FIBs
1026 	 * sitting on top of mlxsw netdevs. However, we can still
1027 	 * have other FIBs queued. Flush the queue before flushing
1028 	 * the device's tables. No need for locks, as we're the only
1029 	 * writer.
1030 	 */
1031 	mlxsw_core_flush_owq();
1032 	mlxsw_sp_router_fib_flush(mlxsw_sp);
1033 	kfree(mlxsw_sp->router->vrs);
1034 }
1035 
1036 u32 mlxsw_sp_ipip_dev_ul_tb_id(const struct net_device *ol_dev)
1037 {
1038 	struct net_device *d;
1039 	u32 tb_id;
1040 
1041 	rcu_read_lock();
1042 	d = mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1043 	if (d)
1044 		tb_id = l3mdev_fib_table(d) ? : RT_TABLE_MAIN;
1045 	else
1046 		tb_id = RT_TABLE_MAIN;
1047 	rcu_read_unlock();
1048 
1049 	return tb_id;
1050 }
1051 
1052 static struct mlxsw_sp_rif *
1053 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
1054 		    const struct mlxsw_sp_rif_params *params,
1055 		    struct netlink_ext_ack *extack);
1056 
1057 static struct mlxsw_sp_rif_ipip_lb *
1058 mlxsw_sp_ipip_ol_ipip_lb_create(struct mlxsw_sp *mlxsw_sp,
1059 				enum mlxsw_sp_ipip_type ipipt,
1060 				struct net_device *ol_dev,
1061 				struct netlink_ext_ack *extack)
1062 {
1063 	struct mlxsw_sp_rif_params_ipip_lb lb_params;
1064 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1065 	struct mlxsw_sp_rif *rif;
1066 
1067 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
1068 	lb_params = (struct mlxsw_sp_rif_params_ipip_lb) {
1069 		.common.dev = ol_dev,
1070 		.common.lag = false,
1071 		.lb_config = ipip_ops->ol_loopback_config(mlxsw_sp, ol_dev),
1072 	};
1073 
1074 	rif = mlxsw_sp_rif_create(mlxsw_sp, &lb_params.common, extack);
1075 	if (IS_ERR(rif))
1076 		return ERR_CAST(rif);
1077 	return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
1078 }
1079 
1080 static struct mlxsw_sp_ipip_entry *
1081 mlxsw_sp_ipip_entry_alloc(struct mlxsw_sp *mlxsw_sp,
1082 			  enum mlxsw_sp_ipip_type ipipt,
1083 			  struct net_device *ol_dev)
1084 {
1085 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1086 	struct mlxsw_sp_ipip_entry *ipip_entry;
1087 	struct mlxsw_sp_ipip_entry *ret = NULL;
1088 	int err;
1089 
1090 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
1091 	ipip_entry = kzalloc(sizeof(*ipip_entry), GFP_KERNEL);
1092 	if (!ipip_entry)
1093 		return ERR_PTR(-ENOMEM);
1094 
1095 	ipip_entry->ol_lb = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp, ipipt,
1096 							    ol_dev, NULL);
1097 	if (IS_ERR(ipip_entry->ol_lb)) {
1098 		ret = ERR_CAST(ipip_entry->ol_lb);
1099 		goto err_ol_ipip_lb_create;
1100 	}
1101 
1102 	ipip_entry->ipipt = ipipt;
1103 	ipip_entry->ol_dev = ol_dev;
1104 	ipip_entry->parms = ipip_ops->parms_init(ol_dev);
1105 
1106 	err = ipip_ops->rem_ip_addr_set(mlxsw_sp, ipip_entry);
1107 	if (err) {
1108 		ret = ERR_PTR(err);
1109 		goto err_rem_ip_addr_set;
1110 	}
1111 
1112 	return ipip_entry;
1113 
1114 err_rem_ip_addr_set:
1115 	mlxsw_sp_rif_destroy(&ipip_entry->ol_lb->common);
1116 err_ol_ipip_lb_create:
1117 	kfree(ipip_entry);
1118 	return ret;
1119 }
1120 
1121 static void mlxsw_sp_ipip_entry_dealloc(struct mlxsw_sp *mlxsw_sp,
1122 					struct mlxsw_sp_ipip_entry *ipip_entry)
1123 {
1124 	const struct mlxsw_sp_ipip_ops *ipip_ops =
1125 		mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1126 
1127 	ipip_ops->rem_ip_addr_unset(mlxsw_sp, ipip_entry);
1128 	mlxsw_sp_rif_destroy(&ipip_entry->ol_lb->common);
1129 	kfree(ipip_entry);
1130 }
1131 
1132 static bool
1133 mlxsw_sp_ipip_entry_saddr_matches(struct mlxsw_sp *mlxsw_sp,
1134 				  const enum mlxsw_sp_l3proto ul_proto,
1135 				  union mlxsw_sp_l3addr saddr,
1136 				  u32 ul_tb_id,
1137 				  struct mlxsw_sp_ipip_entry *ipip_entry)
1138 {
1139 	u32 tun_ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1140 	enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1141 	union mlxsw_sp_l3addr tun_saddr;
1142 
1143 	if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1144 		return false;
1145 
1146 	tun_saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ipip_entry->ol_dev);
1147 	return tun_ul_tb_id == ul_tb_id &&
1148 	       mlxsw_sp_l3addr_eq(&tun_saddr, &saddr);
1149 }
1150 
1151 static int mlxsw_sp_ipip_decap_parsing_depth_inc(struct mlxsw_sp *mlxsw_sp,
1152 						 enum mlxsw_sp_ipip_type ipipt)
1153 {
1154 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1155 
1156 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
1157 
1158 	/* Not all tunnels require to increase the default pasing depth
1159 	 * (96 bytes).
1160 	 */
1161 	if (ipip_ops->inc_parsing_depth)
1162 		return mlxsw_sp_parsing_depth_inc(mlxsw_sp);
1163 
1164 	return 0;
1165 }
1166 
1167 static void mlxsw_sp_ipip_decap_parsing_depth_dec(struct mlxsw_sp *mlxsw_sp,
1168 						  enum mlxsw_sp_ipip_type ipipt)
1169 {
1170 	const struct mlxsw_sp_ipip_ops *ipip_ops =
1171 		mlxsw_sp->router->ipip_ops_arr[ipipt];
1172 
1173 	if (ipip_ops->inc_parsing_depth)
1174 		mlxsw_sp_parsing_depth_dec(mlxsw_sp);
1175 }
1176 
1177 static int
1178 mlxsw_sp_fib_entry_decap_init(struct mlxsw_sp *mlxsw_sp,
1179 			      struct mlxsw_sp_fib_entry *fib_entry,
1180 			      struct mlxsw_sp_ipip_entry *ipip_entry)
1181 {
1182 	u32 tunnel_index;
1183 	int err;
1184 
1185 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
1186 				  1, &tunnel_index);
1187 	if (err)
1188 		return err;
1189 
1190 	err = mlxsw_sp_ipip_decap_parsing_depth_inc(mlxsw_sp,
1191 						    ipip_entry->ipipt);
1192 	if (err)
1193 		goto err_parsing_depth_inc;
1194 
1195 	ipip_entry->decap_fib_entry = fib_entry;
1196 	fib_entry->decap.ipip_entry = ipip_entry;
1197 	fib_entry->decap.tunnel_index = tunnel_index;
1198 
1199 	return 0;
1200 
1201 err_parsing_depth_inc:
1202 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
1203 			   fib_entry->decap.tunnel_index);
1204 	return err;
1205 }
1206 
1207 static void mlxsw_sp_fib_entry_decap_fini(struct mlxsw_sp *mlxsw_sp,
1208 					  struct mlxsw_sp_fib_entry *fib_entry)
1209 {
1210 	enum mlxsw_sp_ipip_type ipipt = fib_entry->decap.ipip_entry->ipipt;
1211 
1212 	/* Unlink this node from the IPIP entry that it's the decap entry of. */
1213 	fib_entry->decap.ipip_entry->decap_fib_entry = NULL;
1214 	fib_entry->decap.ipip_entry = NULL;
1215 	mlxsw_sp_ipip_decap_parsing_depth_dec(mlxsw_sp, ipipt);
1216 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
1217 			   1, fib_entry->decap.tunnel_index);
1218 }
1219 
1220 static struct mlxsw_sp_fib_node *
1221 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
1222 			 size_t addr_len, unsigned char prefix_len);
1223 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1224 				     struct mlxsw_sp_fib_entry *fib_entry);
1225 
1226 static void
1227 mlxsw_sp_ipip_entry_demote_decap(struct mlxsw_sp *mlxsw_sp,
1228 				 struct mlxsw_sp_ipip_entry *ipip_entry)
1229 {
1230 	struct mlxsw_sp_fib_entry *fib_entry = ipip_entry->decap_fib_entry;
1231 
1232 	mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, fib_entry);
1233 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1234 
1235 	mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1236 }
1237 
1238 static void
1239 mlxsw_sp_ipip_entry_promote_decap(struct mlxsw_sp *mlxsw_sp,
1240 				  struct mlxsw_sp_ipip_entry *ipip_entry,
1241 				  struct mlxsw_sp_fib_entry *decap_fib_entry)
1242 {
1243 	if (mlxsw_sp_fib_entry_decap_init(mlxsw_sp, decap_fib_entry,
1244 					  ipip_entry))
1245 		return;
1246 	decap_fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
1247 
1248 	if (mlxsw_sp_fib_entry_update(mlxsw_sp, decap_fib_entry))
1249 		mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1250 }
1251 
1252 static struct mlxsw_sp_fib_entry *
1253 mlxsw_sp_router_ip2me_fib_entry_find(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
1254 				     enum mlxsw_sp_l3proto proto,
1255 				     const union mlxsw_sp_l3addr *addr,
1256 				     enum mlxsw_sp_fib_entry_type type)
1257 {
1258 	struct mlxsw_sp_fib_node *fib_node;
1259 	unsigned char addr_prefix_len;
1260 	struct mlxsw_sp_fib *fib;
1261 	struct mlxsw_sp_vr *vr;
1262 	const void *addrp;
1263 	size_t addr_len;
1264 	u32 addr4;
1265 
1266 	vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
1267 	if (!vr)
1268 		return NULL;
1269 	fib = mlxsw_sp_vr_fib(vr, proto);
1270 
1271 	switch (proto) {
1272 	case MLXSW_SP_L3_PROTO_IPV4:
1273 		addr4 = be32_to_cpu(addr->addr4);
1274 		addrp = &addr4;
1275 		addr_len = 4;
1276 		addr_prefix_len = 32;
1277 		break;
1278 	case MLXSW_SP_L3_PROTO_IPV6:
1279 		addrp = &addr->addr6;
1280 		addr_len = 16;
1281 		addr_prefix_len = 128;
1282 		break;
1283 	default:
1284 		WARN_ON(1);
1285 		return NULL;
1286 	}
1287 
1288 	fib_node = mlxsw_sp_fib_node_lookup(fib, addrp, addr_len,
1289 					    addr_prefix_len);
1290 	if (!fib_node || fib_node->fib_entry->type != type)
1291 		return NULL;
1292 
1293 	return fib_node->fib_entry;
1294 }
1295 
1296 /* Given an IPIP entry, find the corresponding decap route. */
1297 static struct mlxsw_sp_fib_entry *
1298 mlxsw_sp_ipip_entry_find_decap(struct mlxsw_sp *mlxsw_sp,
1299 			       struct mlxsw_sp_ipip_entry *ipip_entry)
1300 {
1301 	static struct mlxsw_sp_fib_node *fib_node;
1302 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1303 	unsigned char saddr_prefix_len;
1304 	union mlxsw_sp_l3addr saddr;
1305 	struct mlxsw_sp_fib *ul_fib;
1306 	struct mlxsw_sp_vr *ul_vr;
1307 	const void *saddrp;
1308 	size_t saddr_len;
1309 	u32 ul_tb_id;
1310 	u32 saddr4;
1311 
1312 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1313 
1314 	ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1315 	ul_vr = mlxsw_sp_vr_find(mlxsw_sp, ul_tb_id);
1316 	if (!ul_vr)
1317 		return NULL;
1318 
1319 	ul_fib = mlxsw_sp_vr_fib(ul_vr, ipip_ops->ul_proto);
1320 	saddr = mlxsw_sp_ipip_netdev_saddr(ipip_ops->ul_proto,
1321 					   ipip_entry->ol_dev);
1322 
1323 	switch (ipip_ops->ul_proto) {
1324 	case MLXSW_SP_L3_PROTO_IPV4:
1325 		saddr4 = be32_to_cpu(saddr.addr4);
1326 		saddrp = &saddr4;
1327 		saddr_len = 4;
1328 		saddr_prefix_len = 32;
1329 		break;
1330 	case MLXSW_SP_L3_PROTO_IPV6:
1331 		saddrp = &saddr.addr6;
1332 		saddr_len = 16;
1333 		saddr_prefix_len = 128;
1334 		break;
1335 	default:
1336 		WARN_ON(1);
1337 		return NULL;
1338 	}
1339 
1340 	fib_node = mlxsw_sp_fib_node_lookup(ul_fib, saddrp, saddr_len,
1341 					    saddr_prefix_len);
1342 	if (!fib_node ||
1343 	    fib_node->fib_entry->type != MLXSW_SP_FIB_ENTRY_TYPE_TRAP)
1344 		return NULL;
1345 
1346 	return fib_node->fib_entry;
1347 }
1348 
1349 static struct mlxsw_sp_ipip_entry *
1350 mlxsw_sp_ipip_entry_create(struct mlxsw_sp *mlxsw_sp,
1351 			   enum mlxsw_sp_ipip_type ipipt,
1352 			   struct net_device *ol_dev)
1353 {
1354 	struct mlxsw_sp_ipip_entry *ipip_entry;
1355 
1356 	ipip_entry = mlxsw_sp_ipip_entry_alloc(mlxsw_sp, ipipt, ol_dev);
1357 	if (IS_ERR(ipip_entry))
1358 		return ipip_entry;
1359 
1360 	list_add_tail(&ipip_entry->ipip_list_node,
1361 		      &mlxsw_sp->router->ipip_list);
1362 
1363 	return ipip_entry;
1364 }
1365 
1366 static void
1367 mlxsw_sp_ipip_entry_destroy(struct mlxsw_sp *mlxsw_sp,
1368 			    struct mlxsw_sp_ipip_entry *ipip_entry)
1369 {
1370 	list_del(&ipip_entry->ipip_list_node);
1371 	mlxsw_sp_ipip_entry_dealloc(mlxsw_sp, ipip_entry);
1372 }
1373 
1374 static bool
1375 mlxsw_sp_ipip_entry_matches_decap(struct mlxsw_sp *mlxsw_sp,
1376 				  const struct net_device *ul_dev,
1377 				  enum mlxsw_sp_l3proto ul_proto,
1378 				  union mlxsw_sp_l3addr ul_dip,
1379 				  struct mlxsw_sp_ipip_entry *ipip_entry)
1380 {
1381 	u32 ul_tb_id = l3mdev_fib_table(ul_dev) ? : RT_TABLE_MAIN;
1382 	enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1383 
1384 	if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1385 		return false;
1386 
1387 	return mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, ul_dip,
1388 						 ul_tb_id, ipip_entry);
1389 }
1390 
1391 /* Given decap parameters, find the corresponding IPIP entry. */
1392 static struct mlxsw_sp_ipip_entry *
1393 mlxsw_sp_ipip_entry_find_by_decap(struct mlxsw_sp *mlxsw_sp, int ul_dev_ifindex,
1394 				  enum mlxsw_sp_l3proto ul_proto,
1395 				  union mlxsw_sp_l3addr ul_dip)
1396 {
1397 	struct mlxsw_sp_ipip_entry *ipip_entry = NULL;
1398 	struct net_device *ul_dev;
1399 
1400 	rcu_read_lock();
1401 
1402 	ul_dev = dev_get_by_index_rcu(mlxsw_sp_net(mlxsw_sp), ul_dev_ifindex);
1403 	if (!ul_dev)
1404 		goto out_unlock;
1405 
1406 	list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1407 			    ipip_list_node)
1408 		if (mlxsw_sp_ipip_entry_matches_decap(mlxsw_sp, ul_dev,
1409 						      ul_proto, ul_dip,
1410 						      ipip_entry))
1411 			goto out_unlock;
1412 
1413 	rcu_read_unlock();
1414 
1415 	return NULL;
1416 
1417 out_unlock:
1418 	rcu_read_unlock();
1419 	return ipip_entry;
1420 }
1421 
1422 static bool mlxsw_sp_netdev_ipip_type(const struct mlxsw_sp *mlxsw_sp,
1423 				      const struct net_device *dev,
1424 				      enum mlxsw_sp_ipip_type *p_type)
1425 {
1426 	struct mlxsw_sp_router *router = mlxsw_sp->router;
1427 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1428 	enum mlxsw_sp_ipip_type ipipt;
1429 
1430 	for (ipipt = 0; ipipt < MLXSW_SP_IPIP_TYPE_MAX; ++ipipt) {
1431 		ipip_ops = router->ipip_ops_arr[ipipt];
1432 		if (dev->type == ipip_ops->dev_type) {
1433 			if (p_type)
1434 				*p_type = ipipt;
1435 			return true;
1436 		}
1437 	}
1438 	return false;
1439 }
1440 
1441 static bool mlxsw_sp_netdev_is_ipip_ol(const struct mlxsw_sp *mlxsw_sp,
1442 				       const struct net_device *dev)
1443 {
1444 	return mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL);
1445 }
1446 
1447 static struct mlxsw_sp_ipip_entry *
1448 mlxsw_sp_ipip_entry_find_by_ol_dev(struct mlxsw_sp *mlxsw_sp,
1449 				   const struct net_device *ol_dev)
1450 {
1451 	struct mlxsw_sp_ipip_entry *ipip_entry;
1452 
1453 	list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1454 			    ipip_list_node)
1455 		if (ipip_entry->ol_dev == ol_dev)
1456 			return ipip_entry;
1457 
1458 	return NULL;
1459 }
1460 
1461 static struct mlxsw_sp_ipip_entry *
1462 mlxsw_sp_ipip_entry_find_by_ul_dev(const struct mlxsw_sp *mlxsw_sp,
1463 				   const struct net_device *ul_dev,
1464 				   struct mlxsw_sp_ipip_entry *start)
1465 {
1466 	struct mlxsw_sp_ipip_entry *ipip_entry;
1467 
1468 	ipip_entry = list_prepare_entry(start, &mlxsw_sp->router->ipip_list,
1469 					ipip_list_node);
1470 	list_for_each_entry_continue(ipip_entry, &mlxsw_sp->router->ipip_list,
1471 				     ipip_list_node) {
1472 		struct net_device *ol_dev = ipip_entry->ol_dev;
1473 		struct net_device *ipip_ul_dev;
1474 
1475 		rcu_read_lock();
1476 		ipip_ul_dev = mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1477 		rcu_read_unlock();
1478 
1479 		if (ipip_ul_dev == ul_dev)
1480 			return ipip_entry;
1481 	}
1482 
1483 	return NULL;
1484 }
1485 
1486 static bool mlxsw_sp_netdev_is_ipip_ul(struct mlxsw_sp *mlxsw_sp,
1487 				       const struct net_device *dev)
1488 {
1489 	return mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp, dev, NULL);
1490 }
1491 
1492 static bool mlxsw_sp_netdevice_ipip_can_offload(struct mlxsw_sp *mlxsw_sp,
1493 						const struct net_device *ol_dev,
1494 						enum mlxsw_sp_ipip_type ipipt)
1495 {
1496 	const struct mlxsw_sp_ipip_ops *ops
1497 		= mlxsw_sp->router->ipip_ops_arr[ipipt];
1498 
1499 	return ops->can_offload(mlxsw_sp, ol_dev);
1500 }
1501 
1502 static int mlxsw_sp_netdevice_ipip_ol_reg_event(struct mlxsw_sp *mlxsw_sp,
1503 						struct net_device *ol_dev)
1504 {
1505 	enum mlxsw_sp_ipip_type ipipt = MLXSW_SP_IPIP_TYPE_MAX;
1506 	struct mlxsw_sp_ipip_entry *ipip_entry;
1507 	enum mlxsw_sp_l3proto ul_proto;
1508 	union mlxsw_sp_l3addr saddr;
1509 	u32 ul_tb_id;
1510 
1511 	mlxsw_sp_netdev_ipip_type(mlxsw_sp, ol_dev, &ipipt);
1512 	if (mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev, ipipt)) {
1513 		ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ol_dev);
1514 		ul_proto = mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto;
1515 		saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ol_dev);
1516 		if (!mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1517 							  saddr, ul_tb_id,
1518 							  NULL)) {
1519 			ipip_entry = mlxsw_sp_ipip_entry_create(mlxsw_sp, ipipt,
1520 								ol_dev);
1521 			if (IS_ERR(ipip_entry))
1522 				return PTR_ERR(ipip_entry);
1523 		}
1524 	}
1525 
1526 	return 0;
1527 }
1528 
1529 static void mlxsw_sp_netdevice_ipip_ol_unreg_event(struct mlxsw_sp *mlxsw_sp,
1530 						   struct net_device *ol_dev)
1531 {
1532 	struct mlxsw_sp_ipip_entry *ipip_entry;
1533 
1534 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1535 	if (ipip_entry)
1536 		mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1537 }
1538 
1539 static void
1540 mlxsw_sp_ipip_entry_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1541 				struct mlxsw_sp_ipip_entry *ipip_entry)
1542 {
1543 	struct mlxsw_sp_fib_entry *decap_fib_entry;
1544 
1545 	decap_fib_entry = mlxsw_sp_ipip_entry_find_decap(mlxsw_sp, ipip_entry);
1546 	if (decap_fib_entry)
1547 		mlxsw_sp_ipip_entry_promote_decap(mlxsw_sp, ipip_entry,
1548 						  decap_fib_entry);
1549 }
1550 
1551 static int
1552 mlxsw_sp_rif_ipip_lb_op(struct mlxsw_sp_rif_ipip_lb *lb_rif, u16 ul_vr_id,
1553 			u16 ul_rif_id, bool enable)
1554 {
1555 	struct mlxsw_sp_rif_ipip_lb_config lb_cf = lb_rif->lb_config;
1556 	enum mlxsw_reg_ritr_loopback_ipip_options ipip_options;
1557 	struct mlxsw_sp_rif *rif = &lb_rif->common;
1558 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
1559 	char ritr_pl[MLXSW_REG_RITR_LEN];
1560 	struct in6_addr *saddr6;
1561 	u32 saddr4;
1562 
1563 	ipip_options = MLXSW_REG_RITR_LOOPBACK_IPIP_OPTIONS_GRE_KEY_PRESET;
1564 	switch (lb_cf.ul_protocol) {
1565 	case MLXSW_SP_L3_PROTO_IPV4:
1566 		saddr4 = be32_to_cpu(lb_cf.saddr.addr4);
1567 		mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
1568 				    rif->rif_index, rif->vr_id, rif->dev->mtu);
1569 		mlxsw_reg_ritr_loopback_ipip4_pack(ritr_pl, lb_cf.lb_ipipt,
1570 						   ipip_options, ul_vr_id,
1571 						   ul_rif_id, saddr4,
1572 						   lb_cf.okey);
1573 		break;
1574 
1575 	case MLXSW_SP_L3_PROTO_IPV6:
1576 		saddr6 = &lb_cf.saddr.addr6;
1577 		mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
1578 				    rif->rif_index, rif->vr_id, rif->dev->mtu);
1579 		mlxsw_reg_ritr_loopback_ipip6_pack(ritr_pl, lb_cf.lb_ipipt,
1580 						   ipip_options, ul_vr_id,
1581 						   ul_rif_id, saddr6,
1582 						   lb_cf.okey);
1583 		break;
1584 	}
1585 
1586 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
1587 }
1588 
1589 static int mlxsw_sp_netdevice_ipip_ol_update_mtu(struct mlxsw_sp *mlxsw_sp,
1590 						 struct net_device *ol_dev)
1591 {
1592 	struct mlxsw_sp_ipip_entry *ipip_entry;
1593 	struct mlxsw_sp_rif_ipip_lb *lb_rif;
1594 	int err = 0;
1595 
1596 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1597 	if (ipip_entry) {
1598 		lb_rif = ipip_entry->ol_lb;
1599 		err = mlxsw_sp_rif_ipip_lb_op(lb_rif, lb_rif->ul_vr_id,
1600 					      lb_rif->ul_rif_id, true);
1601 		if (err)
1602 			goto out;
1603 		lb_rif->common.mtu = ol_dev->mtu;
1604 	}
1605 
1606 out:
1607 	return err;
1608 }
1609 
1610 static void mlxsw_sp_netdevice_ipip_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1611 						struct net_device *ol_dev)
1612 {
1613 	struct mlxsw_sp_ipip_entry *ipip_entry;
1614 
1615 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1616 	if (ipip_entry)
1617 		mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1618 }
1619 
1620 static void
1621 mlxsw_sp_ipip_entry_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1622 				  struct mlxsw_sp_ipip_entry *ipip_entry)
1623 {
1624 	if (ipip_entry->decap_fib_entry)
1625 		mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1626 }
1627 
1628 static void mlxsw_sp_netdevice_ipip_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1629 						  struct net_device *ol_dev)
1630 {
1631 	struct mlxsw_sp_ipip_entry *ipip_entry;
1632 
1633 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1634 	if (ipip_entry)
1635 		mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1636 }
1637 
1638 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
1639 					 struct mlxsw_sp_rif *old_rif,
1640 					 struct mlxsw_sp_rif *new_rif);
1641 static int
1642 mlxsw_sp_ipip_entry_ol_lb_update(struct mlxsw_sp *mlxsw_sp,
1643 				 struct mlxsw_sp_ipip_entry *ipip_entry,
1644 				 bool keep_encap,
1645 				 struct netlink_ext_ack *extack)
1646 {
1647 	struct mlxsw_sp_rif_ipip_lb *old_lb_rif = ipip_entry->ol_lb;
1648 	struct mlxsw_sp_rif_ipip_lb *new_lb_rif;
1649 
1650 	new_lb_rif = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp,
1651 						     ipip_entry->ipipt,
1652 						     ipip_entry->ol_dev,
1653 						     extack);
1654 	if (IS_ERR(new_lb_rif))
1655 		return PTR_ERR(new_lb_rif);
1656 	ipip_entry->ol_lb = new_lb_rif;
1657 
1658 	if (keep_encap)
1659 		mlxsw_sp_nexthop_rif_migrate(mlxsw_sp, &old_lb_rif->common,
1660 					     &new_lb_rif->common);
1661 
1662 	mlxsw_sp_rif_destroy(&old_lb_rif->common);
1663 
1664 	return 0;
1665 }
1666 
1667 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
1668 					struct mlxsw_sp_rif *rif);
1669 
1670 /**
1671  * __mlxsw_sp_ipip_entry_update_tunnel - Update offload related to IPIP entry.
1672  * @mlxsw_sp: mlxsw_sp.
1673  * @ipip_entry: IPIP entry.
1674  * @recreate_loopback: Recreates the associated loopback RIF.
1675  * @keep_encap: Updates next hops that use the tunnel netdevice. This is only
1676  *              relevant when recreate_loopback is true.
1677  * @update_nexthops: Updates next hops, keeping the current loopback RIF. This
1678  *                   is only relevant when recreate_loopback is false.
1679  * @extack: extack.
1680  *
1681  * Return: Non-zero value on failure.
1682  */
1683 int __mlxsw_sp_ipip_entry_update_tunnel(struct mlxsw_sp *mlxsw_sp,
1684 					struct mlxsw_sp_ipip_entry *ipip_entry,
1685 					bool recreate_loopback,
1686 					bool keep_encap,
1687 					bool update_nexthops,
1688 					struct netlink_ext_ack *extack)
1689 {
1690 	int err;
1691 
1692 	/* RIFs can't be edited, so to update loopback, we need to destroy and
1693 	 * recreate it. That creates a window of opportunity where RALUE and
1694 	 * RATR registers end up referencing a RIF that's already gone. RATRs
1695 	 * are handled in mlxsw_sp_ipip_entry_ol_lb_update(), and to take care
1696 	 * of RALUE, demote the decap route back.
1697 	 */
1698 	if (ipip_entry->decap_fib_entry)
1699 		mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1700 
1701 	if (recreate_loopback) {
1702 		err = mlxsw_sp_ipip_entry_ol_lb_update(mlxsw_sp, ipip_entry,
1703 						       keep_encap, extack);
1704 		if (err)
1705 			return err;
1706 	} else if (update_nexthops) {
1707 		mlxsw_sp_nexthop_rif_update(mlxsw_sp,
1708 					    &ipip_entry->ol_lb->common);
1709 	}
1710 
1711 	if (ipip_entry->ol_dev->flags & IFF_UP)
1712 		mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1713 
1714 	return 0;
1715 }
1716 
1717 static int mlxsw_sp_netdevice_ipip_ol_vrf_event(struct mlxsw_sp *mlxsw_sp,
1718 						struct net_device *ol_dev,
1719 						struct netlink_ext_ack *extack)
1720 {
1721 	struct mlxsw_sp_ipip_entry *ipip_entry =
1722 		mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1723 
1724 	if (!ipip_entry)
1725 		return 0;
1726 
1727 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1728 						   true, false, false, extack);
1729 }
1730 
1731 static int
1732 mlxsw_sp_netdevice_ipip_ul_vrf_event(struct mlxsw_sp *mlxsw_sp,
1733 				     struct mlxsw_sp_ipip_entry *ipip_entry,
1734 				     struct net_device *ul_dev,
1735 				     bool *demote_this,
1736 				     struct netlink_ext_ack *extack)
1737 {
1738 	u32 ul_tb_id = l3mdev_fib_table(ul_dev) ? : RT_TABLE_MAIN;
1739 	enum mlxsw_sp_l3proto ul_proto;
1740 	union mlxsw_sp_l3addr saddr;
1741 
1742 	/* Moving underlay to a different VRF might cause local address
1743 	 * conflict, and the conflicting tunnels need to be demoted.
1744 	 */
1745 	ul_proto = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt]->ul_proto;
1746 	saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ipip_entry->ol_dev);
1747 	if (mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1748 						 saddr, ul_tb_id,
1749 						 ipip_entry)) {
1750 		*demote_this = true;
1751 		return 0;
1752 	}
1753 
1754 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1755 						   true, true, false, extack);
1756 }
1757 
1758 static int
1759 mlxsw_sp_netdevice_ipip_ul_up_event(struct mlxsw_sp *mlxsw_sp,
1760 				    struct mlxsw_sp_ipip_entry *ipip_entry,
1761 				    struct net_device *ul_dev)
1762 {
1763 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1764 						   false, false, true, NULL);
1765 }
1766 
1767 static int
1768 mlxsw_sp_netdevice_ipip_ul_down_event(struct mlxsw_sp *mlxsw_sp,
1769 				      struct mlxsw_sp_ipip_entry *ipip_entry,
1770 				      struct net_device *ul_dev)
1771 {
1772 	/* A down underlay device causes encapsulated packets to not be
1773 	 * forwarded, but decap still works. So refresh next hops without
1774 	 * touching anything else.
1775 	 */
1776 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1777 						   false, false, true, NULL);
1778 }
1779 
1780 static int
1781 mlxsw_sp_netdevice_ipip_ol_change_event(struct mlxsw_sp *mlxsw_sp,
1782 					struct net_device *ol_dev,
1783 					struct netlink_ext_ack *extack)
1784 {
1785 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1786 	struct mlxsw_sp_ipip_entry *ipip_entry;
1787 	int err;
1788 
1789 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1790 	if (!ipip_entry)
1791 		/* A change might make a tunnel eligible for offloading, but
1792 		 * that is currently not implemented. What falls to slow path
1793 		 * stays there.
1794 		 */
1795 		return 0;
1796 
1797 	/* A change might make a tunnel not eligible for offloading. */
1798 	if (!mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev,
1799 						 ipip_entry->ipipt)) {
1800 		mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1801 		return 0;
1802 	}
1803 
1804 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1805 	err = ipip_ops->ol_netdev_change(mlxsw_sp, ipip_entry, extack);
1806 	return err;
1807 }
1808 
1809 void mlxsw_sp_ipip_entry_demote_tunnel(struct mlxsw_sp *mlxsw_sp,
1810 				       struct mlxsw_sp_ipip_entry *ipip_entry)
1811 {
1812 	struct net_device *ol_dev = ipip_entry->ol_dev;
1813 
1814 	if (ol_dev->flags & IFF_UP)
1815 		mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1816 	mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1817 }
1818 
1819 /* The configuration where several tunnels have the same local address in the
1820  * same underlay table needs special treatment in the HW. That is currently not
1821  * implemented in the driver. This function finds and demotes the first tunnel
1822  * with a given source address, except the one passed in in the argument
1823  * `except'.
1824  */
1825 bool
1826 mlxsw_sp_ipip_demote_tunnel_by_saddr(struct mlxsw_sp *mlxsw_sp,
1827 				     enum mlxsw_sp_l3proto ul_proto,
1828 				     union mlxsw_sp_l3addr saddr,
1829 				     u32 ul_tb_id,
1830 				     const struct mlxsw_sp_ipip_entry *except)
1831 {
1832 	struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1833 
1834 	list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1835 				 ipip_list_node) {
1836 		if (ipip_entry != except &&
1837 		    mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, saddr,
1838 						      ul_tb_id, ipip_entry)) {
1839 			mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1840 			return true;
1841 		}
1842 	}
1843 
1844 	return false;
1845 }
1846 
1847 static void mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(struct mlxsw_sp *mlxsw_sp,
1848 						     struct net_device *ul_dev)
1849 {
1850 	struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1851 
1852 	list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1853 				 ipip_list_node) {
1854 		struct net_device *ol_dev = ipip_entry->ol_dev;
1855 		struct net_device *ipip_ul_dev;
1856 
1857 		rcu_read_lock();
1858 		ipip_ul_dev = mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1859 		rcu_read_unlock();
1860 		if (ipip_ul_dev == ul_dev)
1861 			mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1862 	}
1863 }
1864 
1865 static int mlxsw_sp_netdevice_ipip_ol_event(struct mlxsw_sp *mlxsw_sp,
1866 					    struct net_device *ol_dev,
1867 					    unsigned long event,
1868 					    struct netdev_notifier_info *info)
1869 {
1870 	struct netdev_notifier_changeupper_info *chup;
1871 	struct netlink_ext_ack *extack;
1872 	int err = 0;
1873 
1874 	switch (event) {
1875 	case NETDEV_REGISTER:
1876 		err = mlxsw_sp_netdevice_ipip_ol_reg_event(mlxsw_sp, ol_dev);
1877 		break;
1878 	case NETDEV_UNREGISTER:
1879 		mlxsw_sp_netdevice_ipip_ol_unreg_event(mlxsw_sp, ol_dev);
1880 		break;
1881 	case NETDEV_UP:
1882 		mlxsw_sp_netdevice_ipip_ol_up_event(mlxsw_sp, ol_dev);
1883 		break;
1884 	case NETDEV_DOWN:
1885 		mlxsw_sp_netdevice_ipip_ol_down_event(mlxsw_sp, ol_dev);
1886 		break;
1887 	case NETDEV_CHANGEUPPER:
1888 		chup = container_of(info, typeof(*chup), info);
1889 		extack = info->extack;
1890 		if (netif_is_l3_master(chup->upper_dev))
1891 			err = mlxsw_sp_netdevice_ipip_ol_vrf_event(mlxsw_sp,
1892 								   ol_dev,
1893 								   extack);
1894 		break;
1895 	case NETDEV_CHANGE:
1896 		extack = info->extack;
1897 		err = mlxsw_sp_netdevice_ipip_ol_change_event(mlxsw_sp,
1898 							      ol_dev, extack);
1899 		break;
1900 	case NETDEV_CHANGEMTU:
1901 		err = mlxsw_sp_netdevice_ipip_ol_update_mtu(mlxsw_sp, ol_dev);
1902 		break;
1903 	}
1904 	return err;
1905 }
1906 
1907 static int
1908 __mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1909 				   struct mlxsw_sp_ipip_entry *ipip_entry,
1910 				   struct net_device *ul_dev,
1911 				   bool *demote_this,
1912 				   unsigned long event,
1913 				   struct netdev_notifier_info *info)
1914 {
1915 	struct netdev_notifier_changeupper_info *chup;
1916 	struct netlink_ext_ack *extack;
1917 
1918 	switch (event) {
1919 	case NETDEV_CHANGEUPPER:
1920 		chup = container_of(info, typeof(*chup), info);
1921 		extack = info->extack;
1922 		if (netif_is_l3_master(chup->upper_dev))
1923 			return mlxsw_sp_netdevice_ipip_ul_vrf_event(mlxsw_sp,
1924 								    ipip_entry,
1925 								    ul_dev,
1926 								    demote_this,
1927 								    extack);
1928 		break;
1929 
1930 	case NETDEV_UP:
1931 		return mlxsw_sp_netdevice_ipip_ul_up_event(mlxsw_sp, ipip_entry,
1932 							   ul_dev);
1933 	case NETDEV_DOWN:
1934 		return mlxsw_sp_netdevice_ipip_ul_down_event(mlxsw_sp,
1935 							     ipip_entry,
1936 							     ul_dev);
1937 	}
1938 	return 0;
1939 }
1940 
1941 static int
1942 mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1943 				 struct net_device *ul_dev,
1944 				 unsigned long event,
1945 				 struct netdev_notifier_info *info)
1946 {
1947 	struct mlxsw_sp_ipip_entry *ipip_entry = NULL;
1948 	int err;
1949 
1950 	while ((ipip_entry = mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp,
1951 								ul_dev,
1952 								ipip_entry))) {
1953 		struct mlxsw_sp_ipip_entry *prev;
1954 		bool demote_this = false;
1955 
1956 		err = __mlxsw_sp_netdevice_ipip_ul_event(mlxsw_sp, ipip_entry,
1957 							 ul_dev, &demote_this,
1958 							 event, info);
1959 		if (err) {
1960 			mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(mlxsw_sp,
1961 								 ul_dev);
1962 			return err;
1963 		}
1964 
1965 		if (demote_this) {
1966 			if (list_is_first(&ipip_entry->ipip_list_node,
1967 					  &mlxsw_sp->router->ipip_list))
1968 				prev = NULL;
1969 			else
1970 				/* This can't be cached from previous iteration,
1971 				 * because that entry could be gone now.
1972 				 */
1973 				prev = list_prev_entry(ipip_entry,
1974 						       ipip_list_node);
1975 			mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1976 			ipip_entry = prev;
1977 		}
1978 	}
1979 
1980 	return 0;
1981 }
1982 
1983 int mlxsw_sp_router_nve_promote_decap(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
1984 				      enum mlxsw_sp_l3proto ul_proto,
1985 				      const union mlxsw_sp_l3addr *ul_sip,
1986 				      u32 tunnel_index)
1987 {
1988 	enum mlxsw_sp_fib_entry_type type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1989 	struct mlxsw_sp_router *router = mlxsw_sp->router;
1990 	struct mlxsw_sp_fib_entry *fib_entry;
1991 	int err = 0;
1992 
1993 	mutex_lock(&mlxsw_sp->router->lock);
1994 
1995 	if (WARN_ON_ONCE(router->nve_decap_config.valid)) {
1996 		err = -EINVAL;
1997 		goto out;
1998 	}
1999 
2000 	router->nve_decap_config.ul_tb_id = ul_tb_id;
2001 	router->nve_decap_config.tunnel_index = tunnel_index;
2002 	router->nve_decap_config.ul_proto = ul_proto;
2003 	router->nve_decap_config.ul_sip = *ul_sip;
2004 	router->nve_decap_config.valid = true;
2005 
2006 	/* It is valid to create a tunnel with a local IP and only later
2007 	 * assign this IP address to a local interface
2008 	 */
2009 	fib_entry = mlxsw_sp_router_ip2me_fib_entry_find(mlxsw_sp, ul_tb_id,
2010 							 ul_proto, ul_sip,
2011 							 type);
2012 	if (!fib_entry)
2013 		goto out;
2014 
2015 	fib_entry->decap.tunnel_index = tunnel_index;
2016 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
2017 
2018 	err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
2019 	if (err)
2020 		goto err_fib_entry_update;
2021 
2022 	goto out;
2023 
2024 err_fib_entry_update:
2025 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
2026 	mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
2027 out:
2028 	mutex_unlock(&mlxsw_sp->router->lock);
2029 	return err;
2030 }
2031 
2032 void mlxsw_sp_router_nve_demote_decap(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
2033 				      enum mlxsw_sp_l3proto ul_proto,
2034 				      const union mlxsw_sp_l3addr *ul_sip)
2035 {
2036 	enum mlxsw_sp_fib_entry_type type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
2037 	struct mlxsw_sp_router *router = mlxsw_sp->router;
2038 	struct mlxsw_sp_fib_entry *fib_entry;
2039 
2040 	mutex_lock(&mlxsw_sp->router->lock);
2041 
2042 	if (WARN_ON_ONCE(!router->nve_decap_config.valid))
2043 		goto out;
2044 
2045 	router->nve_decap_config.valid = false;
2046 
2047 	fib_entry = mlxsw_sp_router_ip2me_fib_entry_find(mlxsw_sp, ul_tb_id,
2048 							 ul_proto, ul_sip,
2049 							 type);
2050 	if (!fib_entry)
2051 		goto out;
2052 
2053 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
2054 	mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
2055 out:
2056 	mutex_unlock(&mlxsw_sp->router->lock);
2057 }
2058 
2059 static bool mlxsw_sp_router_nve_is_decap(struct mlxsw_sp *mlxsw_sp,
2060 					 u32 ul_tb_id,
2061 					 enum mlxsw_sp_l3proto ul_proto,
2062 					 const union mlxsw_sp_l3addr *ul_sip)
2063 {
2064 	struct mlxsw_sp_router *router = mlxsw_sp->router;
2065 
2066 	return router->nve_decap_config.valid &&
2067 	       router->nve_decap_config.ul_tb_id == ul_tb_id &&
2068 	       router->nve_decap_config.ul_proto == ul_proto &&
2069 	       !memcmp(&router->nve_decap_config.ul_sip, ul_sip,
2070 		       sizeof(*ul_sip));
2071 }
2072 
2073 struct mlxsw_sp_neigh_key {
2074 	struct neighbour *n;
2075 };
2076 
2077 struct mlxsw_sp_neigh_entry {
2078 	struct list_head rif_list_node;
2079 	struct rhash_head ht_node;
2080 	struct mlxsw_sp_neigh_key key;
2081 	u16 rif;
2082 	bool connected;
2083 	unsigned char ha[ETH_ALEN];
2084 	struct list_head nexthop_list; /* list of nexthops using
2085 					* this neigh entry
2086 					*/
2087 	struct list_head nexthop_neighs_list_node;
2088 	unsigned int counter_index;
2089 	bool counter_valid;
2090 };
2091 
2092 static const struct rhashtable_params mlxsw_sp_neigh_ht_params = {
2093 	.key_offset = offsetof(struct mlxsw_sp_neigh_entry, key),
2094 	.head_offset = offsetof(struct mlxsw_sp_neigh_entry, ht_node),
2095 	.key_len = sizeof(struct mlxsw_sp_neigh_key),
2096 };
2097 
2098 struct mlxsw_sp_neigh_entry *
2099 mlxsw_sp_rif_neigh_next(struct mlxsw_sp_rif *rif,
2100 			struct mlxsw_sp_neigh_entry *neigh_entry)
2101 {
2102 	if (!neigh_entry) {
2103 		if (list_empty(&rif->neigh_list))
2104 			return NULL;
2105 		else
2106 			return list_first_entry(&rif->neigh_list,
2107 						typeof(*neigh_entry),
2108 						rif_list_node);
2109 	}
2110 	if (list_is_last(&neigh_entry->rif_list_node, &rif->neigh_list))
2111 		return NULL;
2112 	return list_next_entry(neigh_entry, rif_list_node);
2113 }
2114 
2115 int mlxsw_sp_neigh_entry_type(struct mlxsw_sp_neigh_entry *neigh_entry)
2116 {
2117 	return neigh_entry->key.n->tbl->family;
2118 }
2119 
2120 unsigned char *
2121 mlxsw_sp_neigh_entry_ha(struct mlxsw_sp_neigh_entry *neigh_entry)
2122 {
2123 	return neigh_entry->ha;
2124 }
2125 
2126 u32 mlxsw_sp_neigh4_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
2127 {
2128 	struct neighbour *n;
2129 
2130 	n = neigh_entry->key.n;
2131 	return ntohl(*((__be32 *) n->primary_key));
2132 }
2133 
2134 struct in6_addr *
2135 mlxsw_sp_neigh6_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
2136 {
2137 	struct neighbour *n;
2138 
2139 	n = neigh_entry->key.n;
2140 	return (struct in6_addr *) &n->primary_key;
2141 }
2142 
2143 int mlxsw_sp_neigh_counter_get(struct mlxsw_sp *mlxsw_sp,
2144 			       struct mlxsw_sp_neigh_entry *neigh_entry,
2145 			       u64 *p_counter)
2146 {
2147 	if (!neigh_entry->counter_valid)
2148 		return -EINVAL;
2149 
2150 	return mlxsw_sp_flow_counter_get(mlxsw_sp, neigh_entry->counter_index,
2151 					 p_counter, NULL);
2152 }
2153 
2154 static struct mlxsw_sp_neigh_entry *
2155 mlxsw_sp_neigh_entry_alloc(struct mlxsw_sp *mlxsw_sp, struct neighbour *n,
2156 			   u16 rif)
2157 {
2158 	struct mlxsw_sp_neigh_entry *neigh_entry;
2159 
2160 	neigh_entry = kzalloc(sizeof(*neigh_entry), GFP_KERNEL);
2161 	if (!neigh_entry)
2162 		return NULL;
2163 
2164 	neigh_entry->key.n = n;
2165 	neigh_entry->rif = rif;
2166 	INIT_LIST_HEAD(&neigh_entry->nexthop_list);
2167 
2168 	return neigh_entry;
2169 }
2170 
2171 static void mlxsw_sp_neigh_entry_free(struct mlxsw_sp_neigh_entry *neigh_entry)
2172 {
2173 	kfree(neigh_entry);
2174 }
2175 
2176 static int
2177 mlxsw_sp_neigh_entry_insert(struct mlxsw_sp *mlxsw_sp,
2178 			    struct mlxsw_sp_neigh_entry *neigh_entry)
2179 {
2180 	return rhashtable_insert_fast(&mlxsw_sp->router->neigh_ht,
2181 				      &neigh_entry->ht_node,
2182 				      mlxsw_sp_neigh_ht_params);
2183 }
2184 
2185 static void
2186 mlxsw_sp_neigh_entry_remove(struct mlxsw_sp *mlxsw_sp,
2187 			    struct mlxsw_sp_neigh_entry *neigh_entry)
2188 {
2189 	rhashtable_remove_fast(&mlxsw_sp->router->neigh_ht,
2190 			       &neigh_entry->ht_node,
2191 			       mlxsw_sp_neigh_ht_params);
2192 }
2193 
2194 static bool
2195 mlxsw_sp_neigh_counter_should_alloc(struct mlxsw_sp *mlxsw_sp,
2196 				    struct mlxsw_sp_neigh_entry *neigh_entry)
2197 {
2198 	struct devlink *devlink;
2199 	const char *table_name;
2200 
2201 	switch (mlxsw_sp_neigh_entry_type(neigh_entry)) {
2202 	case AF_INET:
2203 		table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST4;
2204 		break;
2205 	case AF_INET6:
2206 		table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST6;
2207 		break;
2208 	default:
2209 		WARN_ON(1);
2210 		return false;
2211 	}
2212 
2213 	devlink = priv_to_devlink(mlxsw_sp->core);
2214 	return devlink_dpipe_table_counter_enabled(devlink, table_name);
2215 }
2216 
2217 static void
2218 mlxsw_sp_neigh_counter_alloc(struct mlxsw_sp *mlxsw_sp,
2219 			     struct mlxsw_sp_neigh_entry *neigh_entry)
2220 {
2221 	if (!mlxsw_sp_neigh_counter_should_alloc(mlxsw_sp, neigh_entry))
2222 		return;
2223 
2224 	if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &neigh_entry->counter_index))
2225 		return;
2226 
2227 	neigh_entry->counter_valid = true;
2228 }
2229 
2230 static void
2231 mlxsw_sp_neigh_counter_free(struct mlxsw_sp *mlxsw_sp,
2232 			    struct mlxsw_sp_neigh_entry *neigh_entry)
2233 {
2234 	if (!neigh_entry->counter_valid)
2235 		return;
2236 	mlxsw_sp_flow_counter_free(mlxsw_sp,
2237 				   neigh_entry->counter_index);
2238 	neigh_entry->counter_valid = false;
2239 }
2240 
2241 static struct mlxsw_sp_neigh_entry *
2242 mlxsw_sp_neigh_entry_create(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
2243 {
2244 	struct mlxsw_sp_neigh_entry *neigh_entry;
2245 	struct mlxsw_sp_rif *rif;
2246 	int err;
2247 
2248 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, n->dev);
2249 	if (!rif)
2250 		return ERR_PTR(-EINVAL);
2251 
2252 	neigh_entry = mlxsw_sp_neigh_entry_alloc(mlxsw_sp, n, rif->rif_index);
2253 	if (!neigh_entry)
2254 		return ERR_PTR(-ENOMEM);
2255 
2256 	err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
2257 	if (err)
2258 		goto err_neigh_entry_insert;
2259 
2260 	mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
2261 	atomic_inc(&mlxsw_sp->router->neighs_update.neigh_count);
2262 	list_add(&neigh_entry->rif_list_node, &rif->neigh_list);
2263 
2264 	return neigh_entry;
2265 
2266 err_neigh_entry_insert:
2267 	mlxsw_sp_neigh_entry_free(neigh_entry);
2268 	return ERR_PTR(err);
2269 }
2270 
2271 static void
2272 mlxsw_sp_neigh_entry_destroy(struct mlxsw_sp *mlxsw_sp,
2273 			     struct mlxsw_sp_neigh_entry *neigh_entry)
2274 {
2275 	list_del(&neigh_entry->rif_list_node);
2276 	atomic_dec(&mlxsw_sp->router->neighs_update.neigh_count);
2277 	mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
2278 	mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
2279 	mlxsw_sp_neigh_entry_free(neigh_entry);
2280 }
2281 
2282 static struct mlxsw_sp_neigh_entry *
2283 mlxsw_sp_neigh_entry_lookup(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
2284 {
2285 	struct mlxsw_sp_neigh_key key;
2286 
2287 	key.n = n;
2288 	return rhashtable_lookup_fast(&mlxsw_sp->router->neigh_ht,
2289 				      &key, mlxsw_sp_neigh_ht_params);
2290 }
2291 
2292 static void
2293 mlxsw_sp_router_neighs_update_interval_init(struct mlxsw_sp *mlxsw_sp)
2294 {
2295 	unsigned long interval;
2296 
2297 #if IS_ENABLED(CONFIG_IPV6)
2298 	interval = min_t(unsigned long,
2299 			 NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME),
2300 			 NEIGH_VAR(&nd_tbl.parms, DELAY_PROBE_TIME));
2301 #else
2302 	interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME);
2303 #endif
2304 	mlxsw_sp->router->neighs_update.interval = jiffies_to_msecs(interval);
2305 }
2306 
2307 static void mlxsw_sp_router_neigh_ent_ipv4_process(struct mlxsw_sp *mlxsw_sp,
2308 						   char *rauhtd_pl,
2309 						   int ent_index)
2310 {
2311 	u64 max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
2312 	struct net_device *dev;
2313 	struct neighbour *n;
2314 	__be32 dipn;
2315 	u32 dip;
2316 	u16 rif;
2317 
2318 	mlxsw_reg_rauhtd_ent_ipv4_unpack(rauhtd_pl, ent_index, &rif, &dip);
2319 
2320 	if (WARN_ON_ONCE(rif >= max_rifs))
2321 		return;
2322 	if (!mlxsw_sp->router->rifs[rif]) {
2323 		dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
2324 		return;
2325 	}
2326 
2327 	dipn = htonl(dip);
2328 	dev = mlxsw_sp->router->rifs[rif]->dev;
2329 	n = neigh_lookup(&arp_tbl, &dipn, dev);
2330 	if (!n)
2331 		return;
2332 
2333 	netdev_dbg(dev, "Updating neighbour with IP=%pI4h\n", &dip);
2334 	neigh_event_send(n, NULL);
2335 	neigh_release(n);
2336 }
2337 
2338 #if IS_ENABLED(CONFIG_IPV6)
2339 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2340 						   char *rauhtd_pl,
2341 						   int rec_index)
2342 {
2343 	struct net_device *dev;
2344 	struct neighbour *n;
2345 	struct in6_addr dip;
2346 	u16 rif;
2347 
2348 	mlxsw_reg_rauhtd_ent_ipv6_unpack(rauhtd_pl, rec_index, &rif,
2349 					 (char *) &dip);
2350 
2351 	if (!mlxsw_sp->router->rifs[rif]) {
2352 		dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
2353 		return;
2354 	}
2355 
2356 	dev = mlxsw_sp->router->rifs[rif]->dev;
2357 	n = neigh_lookup(&nd_tbl, &dip, dev);
2358 	if (!n)
2359 		return;
2360 
2361 	netdev_dbg(dev, "Updating neighbour with IP=%pI6c\n", &dip);
2362 	neigh_event_send(n, NULL);
2363 	neigh_release(n);
2364 }
2365 #else
2366 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2367 						   char *rauhtd_pl,
2368 						   int rec_index)
2369 {
2370 }
2371 #endif
2372 
2373 static void mlxsw_sp_router_neigh_rec_ipv4_process(struct mlxsw_sp *mlxsw_sp,
2374 						   char *rauhtd_pl,
2375 						   int rec_index)
2376 {
2377 	u8 num_entries;
2378 	int i;
2379 
2380 	num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
2381 								rec_index);
2382 	/* Hardware starts counting at 0, so add 1. */
2383 	num_entries++;
2384 
2385 	/* Each record consists of several neighbour entries. */
2386 	for (i = 0; i < num_entries; i++) {
2387 		int ent_index;
2388 
2389 		ent_index = rec_index * MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC + i;
2390 		mlxsw_sp_router_neigh_ent_ipv4_process(mlxsw_sp, rauhtd_pl,
2391 						       ent_index);
2392 	}
2393 
2394 }
2395 
2396 static void mlxsw_sp_router_neigh_rec_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2397 						   char *rauhtd_pl,
2398 						   int rec_index)
2399 {
2400 	/* One record contains one entry. */
2401 	mlxsw_sp_router_neigh_ent_ipv6_process(mlxsw_sp, rauhtd_pl,
2402 					       rec_index);
2403 }
2404 
2405 static void mlxsw_sp_router_neigh_rec_process(struct mlxsw_sp *mlxsw_sp,
2406 					      char *rauhtd_pl, int rec_index)
2407 {
2408 	switch (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, rec_index)) {
2409 	case MLXSW_REG_RAUHTD_TYPE_IPV4:
2410 		mlxsw_sp_router_neigh_rec_ipv4_process(mlxsw_sp, rauhtd_pl,
2411 						       rec_index);
2412 		break;
2413 	case MLXSW_REG_RAUHTD_TYPE_IPV6:
2414 		mlxsw_sp_router_neigh_rec_ipv6_process(mlxsw_sp, rauhtd_pl,
2415 						       rec_index);
2416 		break;
2417 	}
2418 }
2419 
2420 static bool mlxsw_sp_router_rauhtd_is_full(char *rauhtd_pl)
2421 {
2422 	u8 num_rec, last_rec_index, num_entries;
2423 
2424 	num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2425 	last_rec_index = num_rec - 1;
2426 
2427 	if (num_rec < MLXSW_REG_RAUHTD_REC_MAX_NUM)
2428 		return false;
2429 	if (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, last_rec_index) ==
2430 	    MLXSW_REG_RAUHTD_TYPE_IPV6)
2431 		return true;
2432 
2433 	num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
2434 								last_rec_index);
2435 	if (++num_entries == MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC)
2436 		return true;
2437 	return false;
2438 }
2439 
2440 static int
2441 __mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp,
2442 				       char *rauhtd_pl,
2443 				       enum mlxsw_reg_rauhtd_type type)
2444 {
2445 	int i, num_rec;
2446 	int err;
2447 
2448 	/* Ensure the RIF we read from the device does not change mid-dump. */
2449 	mutex_lock(&mlxsw_sp->router->lock);
2450 	do {
2451 		mlxsw_reg_rauhtd_pack(rauhtd_pl, type);
2452 		err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(rauhtd),
2453 				      rauhtd_pl);
2454 		if (err) {
2455 			dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Failed to dump neighbour table\n");
2456 			break;
2457 		}
2458 		num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2459 		for (i = 0; i < num_rec; i++)
2460 			mlxsw_sp_router_neigh_rec_process(mlxsw_sp, rauhtd_pl,
2461 							  i);
2462 	} while (mlxsw_sp_router_rauhtd_is_full(rauhtd_pl));
2463 	mutex_unlock(&mlxsw_sp->router->lock);
2464 
2465 	return err;
2466 }
2467 
2468 static int mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp)
2469 {
2470 	enum mlxsw_reg_rauhtd_type type;
2471 	char *rauhtd_pl;
2472 	int err;
2473 
2474 	if (!atomic_read(&mlxsw_sp->router->neighs_update.neigh_count))
2475 		return 0;
2476 
2477 	rauhtd_pl = kmalloc(MLXSW_REG_RAUHTD_LEN, GFP_KERNEL);
2478 	if (!rauhtd_pl)
2479 		return -ENOMEM;
2480 
2481 	type = MLXSW_REG_RAUHTD_TYPE_IPV4;
2482 	err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2483 	if (err)
2484 		goto out;
2485 
2486 	type = MLXSW_REG_RAUHTD_TYPE_IPV6;
2487 	err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2488 out:
2489 	kfree(rauhtd_pl);
2490 	return err;
2491 }
2492 
2493 static void mlxsw_sp_router_neighs_update_nh(struct mlxsw_sp *mlxsw_sp)
2494 {
2495 	struct mlxsw_sp_neigh_entry *neigh_entry;
2496 
2497 	mutex_lock(&mlxsw_sp->router->lock);
2498 	list_for_each_entry(neigh_entry, &mlxsw_sp->router->nexthop_neighs_list,
2499 			    nexthop_neighs_list_node)
2500 		/* If this neigh have nexthops, make the kernel think this neigh
2501 		 * is active regardless of the traffic.
2502 		 */
2503 		neigh_event_send(neigh_entry->key.n, NULL);
2504 	mutex_unlock(&mlxsw_sp->router->lock);
2505 }
2506 
2507 static void
2508 mlxsw_sp_router_neighs_update_work_schedule(struct mlxsw_sp *mlxsw_sp)
2509 {
2510 	unsigned long interval = mlxsw_sp->router->neighs_update.interval;
2511 
2512 	mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw,
2513 			       msecs_to_jiffies(interval));
2514 }
2515 
2516 static void mlxsw_sp_router_neighs_update_work(struct work_struct *work)
2517 {
2518 	struct mlxsw_sp_router *router;
2519 	int err;
2520 
2521 	router = container_of(work, struct mlxsw_sp_router,
2522 			      neighs_update.dw.work);
2523 	err = mlxsw_sp_router_neighs_update_rauhtd(router->mlxsw_sp);
2524 	if (err)
2525 		dev_err(router->mlxsw_sp->bus_info->dev, "Could not update kernel for neigh activity");
2526 
2527 	mlxsw_sp_router_neighs_update_nh(router->mlxsw_sp);
2528 
2529 	mlxsw_sp_router_neighs_update_work_schedule(router->mlxsw_sp);
2530 }
2531 
2532 static void mlxsw_sp_router_probe_unresolved_nexthops(struct work_struct *work)
2533 {
2534 	struct mlxsw_sp_neigh_entry *neigh_entry;
2535 	struct mlxsw_sp_router *router;
2536 
2537 	router = container_of(work, struct mlxsw_sp_router,
2538 			      nexthop_probe_dw.work);
2539 	/* Iterate over nexthop neighbours, find those who are unresolved and
2540 	 * send arp on them. This solves the chicken-egg problem when
2541 	 * the nexthop wouldn't get offloaded until the neighbor is resolved
2542 	 * but it wouldn't get resolved ever in case traffic is flowing in HW
2543 	 * using different nexthop.
2544 	 */
2545 	mutex_lock(&router->lock);
2546 	list_for_each_entry(neigh_entry, &router->nexthop_neighs_list,
2547 			    nexthop_neighs_list_node)
2548 		if (!neigh_entry->connected)
2549 			neigh_event_send(neigh_entry->key.n, NULL);
2550 	mutex_unlock(&router->lock);
2551 
2552 	mlxsw_core_schedule_dw(&router->nexthop_probe_dw,
2553 			       MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL);
2554 }
2555 
2556 static void
2557 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
2558 			      struct mlxsw_sp_neigh_entry *neigh_entry,
2559 			      bool removing, bool dead);
2560 
2561 static enum mlxsw_reg_rauht_op mlxsw_sp_rauht_op(bool adding)
2562 {
2563 	return adding ? MLXSW_REG_RAUHT_OP_WRITE_ADD :
2564 			MLXSW_REG_RAUHT_OP_WRITE_DELETE;
2565 }
2566 
2567 static int
2568 mlxsw_sp_router_neigh_entry_op4(struct mlxsw_sp *mlxsw_sp,
2569 				struct mlxsw_sp_neigh_entry *neigh_entry,
2570 				enum mlxsw_reg_rauht_op op)
2571 {
2572 	struct neighbour *n = neigh_entry->key.n;
2573 	u32 dip = ntohl(*((__be32 *) n->primary_key));
2574 	char rauht_pl[MLXSW_REG_RAUHT_LEN];
2575 
2576 	mlxsw_reg_rauht_pack4(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2577 			      dip);
2578 	if (neigh_entry->counter_valid)
2579 		mlxsw_reg_rauht_pack_counter(rauht_pl,
2580 					     neigh_entry->counter_index);
2581 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2582 }
2583 
2584 static int
2585 mlxsw_sp_router_neigh_entry_op6(struct mlxsw_sp *mlxsw_sp,
2586 				struct mlxsw_sp_neigh_entry *neigh_entry,
2587 				enum mlxsw_reg_rauht_op op)
2588 {
2589 	struct neighbour *n = neigh_entry->key.n;
2590 	char rauht_pl[MLXSW_REG_RAUHT_LEN];
2591 	const char *dip = n->primary_key;
2592 
2593 	mlxsw_reg_rauht_pack6(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2594 			      dip);
2595 	if (neigh_entry->counter_valid)
2596 		mlxsw_reg_rauht_pack_counter(rauht_pl,
2597 					     neigh_entry->counter_index);
2598 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2599 }
2600 
2601 bool mlxsw_sp_neigh_ipv6_ignore(struct mlxsw_sp_neigh_entry *neigh_entry)
2602 {
2603 	struct neighbour *n = neigh_entry->key.n;
2604 
2605 	/* Packets with a link-local destination address are trapped
2606 	 * after LPM lookup and never reach the neighbour table, so
2607 	 * there is no need to program such neighbours to the device.
2608 	 */
2609 	if (ipv6_addr_type((struct in6_addr *) &n->primary_key) &
2610 	    IPV6_ADDR_LINKLOCAL)
2611 		return true;
2612 	return false;
2613 }
2614 
2615 static void
2616 mlxsw_sp_neigh_entry_update(struct mlxsw_sp *mlxsw_sp,
2617 			    struct mlxsw_sp_neigh_entry *neigh_entry,
2618 			    bool adding)
2619 {
2620 	enum mlxsw_reg_rauht_op op = mlxsw_sp_rauht_op(adding);
2621 	int err;
2622 
2623 	if (!adding && !neigh_entry->connected)
2624 		return;
2625 	neigh_entry->connected = adding;
2626 	if (neigh_entry->key.n->tbl->family == AF_INET) {
2627 		err = mlxsw_sp_router_neigh_entry_op4(mlxsw_sp, neigh_entry,
2628 						      op);
2629 		if (err)
2630 			return;
2631 	} else if (neigh_entry->key.n->tbl->family == AF_INET6) {
2632 		if (mlxsw_sp_neigh_ipv6_ignore(neigh_entry))
2633 			return;
2634 		err = mlxsw_sp_router_neigh_entry_op6(mlxsw_sp, neigh_entry,
2635 						      op);
2636 		if (err)
2637 			return;
2638 	} else {
2639 		WARN_ON_ONCE(1);
2640 		return;
2641 	}
2642 
2643 	if (adding)
2644 		neigh_entry->key.n->flags |= NTF_OFFLOADED;
2645 	else
2646 		neigh_entry->key.n->flags &= ~NTF_OFFLOADED;
2647 }
2648 
2649 void
2650 mlxsw_sp_neigh_entry_counter_update(struct mlxsw_sp *mlxsw_sp,
2651 				    struct mlxsw_sp_neigh_entry *neigh_entry,
2652 				    bool adding)
2653 {
2654 	if (adding)
2655 		mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
2656 	else
2657 		mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
2658 	mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, true);
2659 }
2660 
2661 struct mlxsw_sp_netevent_work {
2662 	struct work_struct work;
2663 	struct mlxsw_sp *mlxsw_sp;
2664 	struct neighbour *n;
2665 };
2666 
2667 static void mlxsw_sp_router_neigh_event_work(struct work_struct *work)
2668 {
2669 	struct mlxsw_sp_netevent_work *net_work =
2670 		container_of(work, struct mlxsw_sp_netevent_work, work);
2671 	struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2672 	struct mlxsw_sp_neigh_entry *neigh_entry;
2673 	struct neighbour *n = net_work->n;
2674 	unsigned char ha[ETH_ALEN];
2675 	bool entry_connected;
2676 	u8 nud_state, dead;
2677 
2678 	/* If these parameters are changed after we release the lock,
2679 	 * then we are guaranteed to receive another event letting us
2680 	 * know about it.
2681 	 */
2682 	read_lock_bh(&n->lock);
2683 	memcpy(ha, n->ha, ETH_ALEN);
2684 	nud_state = n->nud_state;
2685 	dead = n->dead;
2686 	read_unlock_bh(&n->lock);
2687 
2688 	mutex_lock(&mlxsw_sp->router->lock);
2689 	mlxsw_sp_span_respin(mlxsw_sp);
2690 
2691 	entry_connected = nud_state & NUD_VALID && !dead;
2692 	neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
2693 	if (!entry_connected && !neigh_entry)
2694 		goto out;
2695 	if (!neigh_entry) {
2696 		neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
2697 		if (IS_ERR(neigh_entry))
2698 			goto out;
2699 	}
2700 
2701 	if (neigh_entry->connected && entry_connected &&
2702 	    !memcmp(neigh_entry->ha, ha, ETH_ALEN))
2703 		goto out;
2704 
2705 	memcpy(neigh_entry->ha, ha, ETH_ALEN);
2706 	mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, entry_connected);
2707 	mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, !entry_connected,
2708 				      dead);
2709 
2710 	if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
2711 		mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2712 
2713 out:
2714 	mutex_unlock(&mlxsw_sp->router->lock);
2715 	neigh_release(n);
2716 	kfree(net_work);
2717 }
2718 
2719 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp);
2720 
2721 static void mlxsw_sp_router_mp_hash_event_work(struct work_struct *work)
2722 {
2723 	struct mlxsw_sp_netevent_work *net_work =
2724 		container_of(work, struct mlxsw_sp_netevent_work, work);
2725 	struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2726 
2727 	mlxsw_sp_mp_hash_init(mlxsw_sp);
2728 	kfree(net_work);
2729 }
2730 
2731 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp);
2732 
2733 static void mlxsw_sp_router_update_priority_work(struct work_struct *work)
2734 {
2735 	struct mlxsw_sp_netevent_work *net_work =
2736 		container_of(work, struct mlxsw_sp_netevent_work, work);
2737 	struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2738 
2739 	__mlxsw_sp_router_init(mlxsw_sp);
2740 	kfree(net_work);
2741 }
2742 
2743 static int mlxsw_sp_router_schedule_work(struct net *net,
2744 					 struct notifier_block *nb,
2745 					 void (*cb)(struct work_struct *))
2746 {
2747 	struct mlxsw_sp_netevent_work *net_work;
2748 	struct mlxsw_sp_router *router;
2749 
2750 	router = container_of(nb, struct mlxsw_sp_router, netevent_nb);
2751 	if (!net_eq(net, mlxsw_sp_net(router->mlxsw_sp)))
2752 		return NOTIFY_DONE;
2753 
2754 	net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2755 	if (!net_work)
2756 		return NOTIFY_BAD;
2757 
2758 	INIT_WORK(&net_work->work, cb);
2759 	net_work->mlxsw_sp = router->mlxsw_sp;
2760 	mlxsw_core_schedule_work(&net_work->work);
2761 	return NOTIFY_DONE;
2762 }
2763 
2764 static int mlxsw_sp_router_netevent_event(struct notifier_block *nb,
2765 					  unsigned long event, void *ptr)
2766 {
2767 	struct mlxsw_sp_netevent_work *net_work;
2768 	struct mlxsw_sp_port *mlxsw_sp_port;
2769 	struct mlxsw_sp *mlxsw_sp;
2770 	unsigned long interval;
2771 	struct neigh_parms *p;
2772 	struct neighbour *n;
2773 
2774 	switch (event) {
2775 	case NETEVENT_DELAY_PROBE_TIME_UPDATE:
2776 		p = ptr;
2777 
2778 		/* We don't care about changes in the default table. */
2779 		if (!p->dev || (p->tbl->family != AF_INET &&
2780 				p->tbl->family != AF_INET6))
2781 			return NOTIFY_DONE;
2782 
2783 		/* We are in atomic context and can't take RTNL mutex,
2784 		 * so use RCU variant to walk the device chain.
2785 		 */
2786 		mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(p->dev);
2787 		if (!mlxsw_sp_port)
2788 			return NOTIFY_DONE;
2789 
2790 		mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2791 		interval = jiffies_to_msecs(NEIGH_VAR(p, DELAY_PROBE_TIME));
2792 		mlxsw_sp->router->neighs_update.interval = interval;
2793 
2794 		mlxsw_sp_port_dev_put(mlxsw_sp_port);
2795 		break;
2796 	case NETEVENT_NEIGH_UPDATE:
2797 		n = ptr;
2798 
2799 		if (n->tbl->family != AF_INET && n->tbl->family != AF_INET6)
2800 			return NOTIFY_DONE;
2801 
2802 		mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(n->dev);
2803 		if (!mlxsw_sp_port)
2804 			return NOTIFY_DONE;
2805 
2806 		net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2807 		if (!net_work) {
2808 			mlxsw_sp_port_dev_put(mlxsw_sp_port);
2809 			return NOTIFY_BAD;
2810 		}
2811 
2812 		INIT_WORK(&net_work->work, mlxsw_sp_router_neigh_event_work);
2813 		net_work->mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2814 		net_work->n = n;
2815 
2816 		/* Take a reference to ensure the neighbour won't be
2817 		 * destructed until we drop the reference in delayed
2818 		 * work.
2819 		 */
2820 		neigh_clone(n);
2821 		mlxsw_core_schedule_work(&net_work->work);
2822 		mlxsw_sp_port_dev_put(mlxsw_sp_port);
2823 		break;
2824 	case NETEVENT_IPV4_MPATH_HASH_UPDATE:
2825 	case NETEVENT_IPV6_MPATH_HASH_UPDATE:
2826 		return mlxsw_sp_router_schedule_work(ptr, nb,
2827 				mlxsw_sp_router_mp_hash_event_work);
2828 
2829 	case NETEVENT_IPV4_FWD_UPDATE_PRIORITY_UPDATE:
2830 		return mlxsw_sp_router_schedule_work(ptr, nb,
2831 				mlxsw_sp_router_update_priority_work);
2832 	}
2833 
2834 	return NOTIFY_DONE;
2835 }
2836 
2837 static int mlxsw_sp_neigh_init(struct mlxsw_sp *mlxsw_sp)
2838 {
2839 	int err;
2840 
2841 	err = rhashtable_init(&mlxsw_sp->router->neigh_ht,
2842 			      &mlxsw_sp_neigh_ht_params);
2843 	if (err)
2844 		return err;
2845 
2846 	/* Initialize the polling interval according to the default
2847 	 * table.
2848 	 */
2849 	mlxsw_sp_router_neighs_update_interval_init(mlxsw_sp);
2850 
2851 	/* Create the delayed works for the activity_update */
2852 	INIT_DELAYED_WORK(&mlxsw_sp->router->neighs_update.dw,
2853 			  mlxsw_sp_router_neighs_update_work);
2854 	INIT_DELAYED_WORK(&mlxsw_sp->router->nexthop_probe_dw,
2855 			  mlxsw_sp_router_probe_unresolved_nexthops);
2856 	atomic_set(&mlxsw_sp->router->neighs_update.neigh_count, 0);
2857 	mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw, 0);
2858 	mlxsw_core_schedule_dw(&mlxsw_sp->router->nexthop_probe_dw, 0);
2859 	return 0;
2860 }
2861 
2862 static void mlxsw_sp_neigh_fini(struct mlxsw_sp *mlxsw_sp)
2863 {
2864 	cancel_delayed_work_sync(&mlxsw_sp->router->neighs_update.dw);
2865 	cancel_delayed_work_sync(&mlxsw_sp->router->nexthop_probe_dw);
2866 	rhashtable_destroy(&mlxsw_sp->router->neigh_ht);
2867 }
2868 
2869 static void mlxsw_sp_neigh_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
2870 					 struct mlxsw_sp_rif *rif)
2871 {
2872 	struct mlxsw_sp_neigh_entry *neigh_entry, *tmp;
2873 
2874 	list_for_each_entry_safe(neigh_entry, tmp, &rif->neigh_list,
2875 				 rif_list_node) {
2876 		mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, false);
2877 		mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2878 	}
2879 }
2880 
2881 enum mlxsw_sp_nexthop_type {
2882 	MLXSW_SP_NEXTHOP_TYPE_ETH,
2883 	MLXSW_SP_NEXTHOP_TYPE_IPIP,
2884 };
2885 
2886 enum mlxsw_sp_nexthop_action {
2887 	/* Nexthop forwards packets to an egress RIF */
2888 	MLXSW_SP_NEXTHOP_ACTION_FORWARD,
2889 	/* Nexthop discards packets */
2890 	MLXSW_SP_NEXTHOP_ACTION_DISCARD,
2891 	/* Nexthop traps packets */
2892 	MLXSW_SP_NEXTHOP_ACTION_TRAP,
2893 };
2894 
2895 struct mlxsw_sp_nexthop_key {
2896 	struct fib_nh *fib_nh;
2897 };
2898 
2899 struct mlxsw_sp_nexthop {
2900 	struct list_head neigh_list_node; /* member of neigh entry list */
2901 	struct list_head rif_list_node;
2902 	struct list_head router_list_node;
2903 	struct mlxsw_sp_nexthop_group_info *nhgi; /* pointer back to the group
2904 						   * this nexthop belongs to
2905 						   */
2906 	struct rhash_head ht_node;
2907 	struct neigh_table *neigh_tbl;
2908 	struct mlxsw_sp_nexthop_key key;
2909 	unsigned char gw_addr[sizeof(struct in6_addr)];
2910 	int ifindex;
2911 	int nh_weight;
2912 	int norm_nh_weight;
2913 	int num_adj_entries;
2914 	struct mlxsw_sp_rif *rif;
2915 	u8 should_offload:1, /* set indicates this nexthop should be written
2916 			      * to the adjacency table.
2917 			      */
2918 	   offloaded:1, /* set indicates this nexthop was written to the
2919 			 * adjacency table.
2920 			 */
2921 	   update:1; /* set indicates this nexthop should be updated in the
2922 		      * adjacency table (f.e., its MAC changed).
2923 		      */
2924 	enum mlxsw_sp_nexthop_action action;
2925 	enum mlxsw_sp_nexthop_type type;
2926 	union {
2927 		struct mlxsw_sp_neigh_entry *neigh_entry;
2928 		struct mlxsw_sp_ipip_entry *ipip_entry;
2929 	};
2930 	unsigned int counter_index;
2931 	bool counter_valid;
2932 };
2933 
2934 enum mlxsw_sp_nexthop_group_type {
2935 	MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4,
2936 	MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6,
2937 	MLXSW_SP_NEXTHOP_GROUP_TYPE_OBJ,
2938 };
2939 
2940 struct mlxsw_sp_nexthop_group_info {
2941 	struct mlxsw_sp_nexthop_group *nh_grp;
2942 	u32 adj_index;
2943 	u16 ecmp_size;
2944 	u16 count;
2945 	int sum_norm_weight;
2946 	u8 adj_index_valid:1,
2947 	   gateway:1, /* routes using the group use a gateway */
2948 	   is_resilient:1;
2949 	struct list_head list; /* member in nh_res_grp_list */
2950 	struct mlxsw_sp_nexthop nexthops[0];
2951 #define nh_rif	nexthops[0].rif
2952 };
2953 
2954 struct mlxsw_sp_nexthop_group_vr_key {
2955 	u16 vr_id;
2956 	enum mlxsw_sp_l3proto proto;
2957 };
2958 
2959 struct mlxsw_sp_nexthop_group_vr_entry {
2960 	struct list_head list; /* member in vr_list */
2961 	struct rhash_head ht_node; /* member in vr_ht */
2962 	refcount_t ref_count;
2963 	struct mlxsw_sp_nexthop_group_vr_key key;
2964 };
2965 
2966 struct mlxsw_sp_nexthop_group {
2967 	struct rhash_head ht_node;
2968 	struct list_head fib_list; /* list of fib entries that use this group */
2969 	union {
2970 		struct {
2971 			struct fib_info *fi;
2972 		} ipv4;
2973 		struct {
2974 			u32 id;
2975 		} obj;
2976 	};
2977 	struct mlxsw_sp_nexthop_group_info *nhgi;
2978 	struct list_head vr_list;
2979 	struct rhashtable vr_ht;
2980 	enum mlxsw_sp_nexthop_group_type type;
2981 	bool can_destroy;
2982 };
2983 
2984 void mlxsw_sp_nexthop_counter_alloc(struct mlxsw_sp *mlxsw_sp,
2985 				    struct mlxsw_sp_nexthop *nh)
2986 {
2987 	struct devlink *devlink;
2988 
2989 	devlink = priv_to_devlink(mlxsw_sp->core);
2990 	if (!devlink_dpipe_table_counter_enabled(devlink,
2991 						 MLXSW_SP_DPIPE_TABLE_NAME_ADJ))
2992 		return;
2993 
2994 	if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &nh->counter_index))
2995 		return;
2996 
2997 	nh->counter_valid = true;
2998 }
2999 
3000 void mlxsw_sp_nexthop_counter_free(struct mlxsw_sp *mlxsw_sp,
3001 				   struct mlxsw_sp_nexthop *nh)
3002 {
3003 	if (!nh->counter_valid)
3004 		return;
3005 	mlxsw_sp_flow_counter_free(mlxsw_sp, nh->counter_index);
3006 	nh->counter_valid = false;
3007 }
3008 
3009 int mlxsw_sp_nexthop_counter_get(struct mlxsw_sp *mlxsw_sp,
3010 				 struct mlxsw_sp_nexthop *nh, u64 *p_counter)
3011 {
3012 	if (!nh->counter_valid)
3013 		return -EINVAL;
3014 
3015 	return mlxsw_sp_flow_counter_get(mlxsw_sp, nh->counter_index,
3016 					 p_counter, NULL);
3017 }
3018 
3019 struct mlxsw_sp_nexthop *mlxsw_sp_nexthop_next(struct mlxsw_sp_router *router,
3020 					       struct mlxsw_sp_nexthop *nh)
3021 {
3022 	if (!nh) {
3023 		if (list_empty(&router->nexthop_list))
3024 			return NULL;
3025 		else
3026 			return list_first_entry(&router->nexthop_list,
3027 						typeof(*nh), router_list_node);
3028 	}
3029 	if (list_is_last(&nh->router_list_node, &router->nexthop_list))
3030 		return NULL;
3031 	return list_next_entry(nh, router_list_node);
3032 }
3033 
3034 bool mlxsw_sp_nexthop_is_forward(const struct mlxsw_sp_nexthop *nh)
3035 {
3036 	return nh->offloaded && nh->action == MLXSW_SP_NEXTHOP_ACTION_FORWARD;
3037 }
3038 
3039 unsigned char *mlxsw_sp_nexthop_ha(struct mlxsw_sp_nexthop *nh)
3040 {
3041 	if (nh->type != MLXSW_SP_NEXTHOP_TYPE_ETH ||
3042 	    !mlxsw_sp_nexthop_is_forward(nh))
3043 		return NULL;
3044 	return nh->neigh_entry->ha;
3045 }
3046 
3047 int mlxsw_sp_nexthop_indexes(struct mlxsw_sp_nexthop *nh, u32 *p_adj_index,
3048 			     u32 *p_adj_size, u32 *p_adj_hash_index)
3049 {
3050 	struct mlxsw_sp_nexthop_group_info *nhgi = nh->nhgi;
3051 	u32 adj_hash_index = 0;
3052 	int i;
3053 
3054 	if (!nh->offloaded || !nhgi->adj_index_valid)
3055 		return -EINVAL;
3056 
3057 	*p_adj_index = nhgi->adj_index;
3058 	*p_adj_size = nhgi->ecmp_size;
3059 
3060 	for (i = 0; i < nhgi->count; i++) {
3061 		struct mlxsw_sp_nexthop *nh_iter = &nhgi->nexthops[i];
3062 
3063 		if (nh_iter == nh)
3064 			break;
3065 		if (nh_iter->offloaded)
3066 			adj_hash_index += nh_iter->num_adj_entries;
3067 	}
3068 
3069 	*p_adj_hash_index = adj_hash_index;
3070 	return 0;
3071 }
3072 
3073 struct mlxsw_sp_rif *mlxsw_sp_nexthop_rif(struct mlxsw_sp_nexthop *nh)
3074 {
3075 	return nh->rif;
3076 }
3077 
3078 bool mlxsw_sp_nexthop_group_has_ipip(struct mlxsw_sp_nexthop *nh)
3079 {
3080 	struct mlxsw_sp_nexthop_group_info *nhgi = nh->nhgi;
3081 	int i;
3082 
3083 	for (i = 0; i < nhgi->count; i++) {
3084 		struct mlxsw_sp_nexthop *nh_iter = &nhgi->nexthops[i];
3085 
3086 		if (nh_iter->type == MLXSW_SP_NEXTHOP_TYPE_IPIP)
3087 			return true;
3088 	}
3089 	return false;
3090 }
3091 
3092 static const struct rhashtable_params mlxsw_sp_nexthop_group_vr_ht_params = {
3093 	.key_offset = offsetof(struct mlxsw_sp_nexthop_group_vr_entry, key),
3094 	.head_offset = offsetof(struct mlxsw_sp_nexthop_group_vr_entry, ht_node),
3095 	.key_len = sizeof(struct mlxsw_sp_nexthop_group_vr_key),
3096 	.automatic_shrinking = true,
3097 };
3098 
3099 static struct mlxsw_sp_nexthop_group_vr_entry *
3100 mlxsw_sp_nexthop_group_vr_entry_lookup(struct mlxsw_sp_nexthop_group *nh_grp,
3101 				       const struct mlxsw_sp_fib *fib)
3102 {
3103 	struct mlxsw_sp_nexthop_group_vr_key key;
3104 
3105 	memset(&key, 0, sizeof(key));
3106 	key.vr_id = fib->vr->id;
3107 	key.proto = fib->proto;
3108 	return rhashtable_lookup_fast(&nh_grp->vr_ht, &key,
3109 				      mlxsw_sp_nexthop_group_vr_ht_params);
3110 }
3111 
3112 static int
3113 mlxsw_sp_nexthop_group_vr_entry_create(struct mlxsw_sp_nexthop_group *nh_grp,
3114 				       const struct mlxsw_sp_fib *fib)
3115 {
3116 	struct mlxsw_sp_nexthop_group_vr_entry *vr_entry;
3117 	int err;
3118 
3119 	vr_entry = kzalloc(sizeof(*vr_entry), GFP_KERNEL);
3120 	if (!vr_entry)
3121 		return -ENOMEM;
3122 
3123 	vr_entry->key.vr_id = fib->vr->id;
3124 	vr_entry->key.proto = fib->proto;
3125 	refcount_set(&vr_entry->ref_count, 1);
3126 
3127 	err = rhashtable_insert_fast(&nh_grp->vr_ht, &vr_entry->ht_node,
3128 				     mlxsw_sp_nexthop_group_vr_ht_params);
3129 	if (err)
3130 		goto err_hashtable_insert;
3131 
3132 	list_add(&vr_entry->list, &nh_grp->vr_list);
3133 
3134 	return 0;
3135 
3136 err_hashtable_insert:
3137 	kfree(vr_entry);
3138 	return err;
3139 }
3140 
3141 static void
3142 mlxsw_sp_nexthop_group_vr_entry_destroy(struct mlxsw_sp_nexthop_group *nh_grp,
3143 					struct mlxsw_sp_nexthop_group_vr_entry *vr_entry)
3144 {
3145 	list_del(&vr_entry->list);
3146 	rhashtable_remove_fast(&nh_grp->vr_ht, &vr_entry->ht_node,
3147 			       mlxsw_sp_nexthop_group_vr_ht_params);
3148 	kfree(vr_entry);
3149 }
3150 
3151 static int
3152 mlxsw_sp_nexthop_group_vr_link(struct mlxsw_sp_nexthop_group *nh_grp,
3153 			       const struct mlxsw_sp_fib *fib)
3154 {
3155 	struct mlxsw_sp_nexthop_group_vr_entry *vr_entry;
3156 
3157 	vr_entry = mlxsw_sp_nexthop_group_vr_entry_lookup(nh_grp, fib);
3158 	if (vr_entry) {
3159 		refcount_inc(&vr_entry->ref_count);
3160 		return 0;
3161 	}
3162 
3163 	return mlxsw_sp_nexthop_group_vr_entry_create(nh_grp, fib);
3164 }
3165 
3166 static void
3167 mlxsw_sp_nexthop_group_vr_unlink(struct mlxsw_sp_nexthop_group *nh_grp,
3168 				 const struct mlxsw_sp_fib *fib)
3169 {
3170 	struct mlxsw_sp_nexthop_group_vr_entry *vr_entry;
3171 
3172 	vr_entry = mlxsw_sp_nexthop_group_vr_entry_lookup(nh_grp, fib);
3173 	if (WARN_ON_ONCE(!vr_entry))
3174 		return;
3175 
3176 	if (!refcount_dec_and_test(&vr_entry->ref_count))
3177 		return;
3178 
3179 	mlxsw_sp_nexthop_group_vr_entry_destroy(nh_grp, vr_entry);
3180 }
3181 
3182 struct mlxsw_sp_nexthop_group_cmp_arg {
3183 	enum mlxsw_sp_nexthop_group_type type;
3184 	union {
3185 		struct fib_info *fi;
3186 		struct mlxsw_sp_fib6_entry *fib6_entry;
3187 		u32 id;
3188 	};
3189 };
3190 
3191 static bool
3192 mlxsw_sp_nexthop6_group_has_nexthop(const struct mlxsw_sp_nexthop_group *nh_grp,
3193 				    const struct in6_addr *gw, int ifindex,
3194 				    int weight)
3195 {
3196 	int i;
3197 
3198 	for (i = 0; i < nh_grp->nhgi->count; i++) {
3199 		const struct mlxsw_sp_nexthop *nh;
3200 
3201 		nh = &nh_grp->nhgi->nexthops[i];
3202 		if (nh->ifindex == ifindex && nh->nh_weight == weight &&
3203 		    ipv6_addr_equal(gw, (struct in6_addr *) nh->gw_addr))
3204 			return true;
3205 	}
3206 
3207 	return false;
3208 }
3209 
3210 static bool
3211 mlxsw_sp_nexthop6_group_cmp(const struct mlxsw_sp_nexthop_group *nh_grp,
3212 			    const struct mlxsw_sp_fib6_entry *fib6_entry)
3213 {
3214 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3215 
3216 	if (nh_grp->nhgi->count != fib6_entry->nrt6)
3217 		return false;
3218 
3219 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3220 		struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3221 		struct in6_addr *gw;
3222 		int ifindex, weight;
3223 
3224 		ifindex = fib6_nh->fib_nh_dev->ifindex;
3225 		weight = fib6_nh->fib_nh_weight;
3226 		gw = &fib6_nh->fib_nh_gw6;
3227 		if (!mlxsw_sp_nexthop6_group_has_nexthop(nh_grp, gw, ifindex,
3228 							 weight))
3229 			return false;
3230 	}
3231 
3232 	return true;
3233 }
3234 
3235 static int
3236 mlxsw_sp_nexthop_group_cmp(struct rhashtable_compare_arg *arg, const void *ptr)
3237 {
3238 	const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = arg->key;
3239 	const struct mlxsw_sp_nexthop_group *nh_grp = ptr;
3240 
3241 	if (nh_grp->type != cmp_arg->type)
3242 		return 1;
3243 
3244 	switch (cmp_arg->type) {
3245 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3246 		return cmp_arg->fi != nh_grp->ipv4.fi;
3247 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3248 		return !mlxsw_sp_nexthop6_group_cmp(nh_grp,
3249 						    cmp_arg->fib6_entry);
3250 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_OBJ:
3251 		return cmp_arg->id != nh_grp->obj.id;
3252 	default:
3253 		WARN_ON(1);
3254 		return 1;
3255 	}
3256 }
3257 
3258 static u32 mlxsw_sp_nexthop_group_hash_obj(const void *data, u32 len, u32 seed)
3259 {
3260 	const struct mlxsw_sp_nexthop_group *nh_grp = data;
3261 	const struct mlxsw_sp_nexthop *nh;
3262 	struct fib_info *fi;
3263 	unsigned int val;
3264 	int i;
3265 
3266 	switch (nh_grp->type) {
3267 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3268 		fi = nh_grp->ipv4.fi;
3269 		return jhash(&fi, sizeof(fi), seed);
3270 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3271 		val = nh_grp->nhgi->count;
3272 		for (i = 0; i < nh_grp->nhgi->count; i++) {
3273 			nh = &nh_grp->nhgi->nexthops[i];
3274 			val ^= jhash(&nh->ifindex, sizeof(nh->ifindex), seed);
3275 			val ^= jhash(&nh->gw_addr, sizeof(nh->gw_addr), seed);
3276 		}
3277 		return jhash(&val, sizeof(val), seed);
3278 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_OBJ:
3279 		return jhash(&nh_grp->obj.id, sizeof(nh_grp->obj.id), seed);
3280 	default:
3281 		WARN_ON(1);
3282 		return 0;
3283 	}
3284 }
3285 
3286 static u32
3287 mlxsw_sp_nexthop6_group_hash(struct mlxsw_sp_fib6_entry *fib6_entry, u32 seed)
3288 {
3289 	unsigned int val = fib6_entry->nrt6;
3290 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3291 
3292 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3293 		struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3294 		struct net_device *dev = fib6_nh->fib_nh_dev;
3295 		struct in6_addr *gw = &fib6_nh->fib_nh_gw6;
3296 
3297 		val ^= jhash(&dev->ifindex, sizeof(dev->ifindex), seed);
3298 		val ^= jhash(gw, sizeof(*gw), seed);
3299 	}
3300 
3301 	return jhash(&val, sizeof(val), seed);
3302 }
3303 
3304 static u32
3305 mlxsw_sp_nexthop_group_hash(const void *data, u32 len, u32 seed)
3306 {
3307 	const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = data;
3308 
3309 	switch (cmp_arg->type) {
3310 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3311 		return jhash(&cmp_arg->fi, sizeof(cmp_arg->fi), seed);
3312 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3313 		return mlxsw_sp_nexthop6_group_hash(cmp_arg->fib6_entry, seed);
3314 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_OBJ:
3315 		return jhash(&cmp_arg->id, sizeof(cmp_arg->id), seed);
3316 	default:
3317 		WARN_ON(1);
3318 		return 0;
3319 	}
3320 }
3321 
3322 static const struct rhashtable_params mlxsw_sp_nexthop_group_ht_params = {
3323 	.head_offset = offsetof(struct mlxsw_sp_nexthop_group, ht_node),
3324 	.hashfn	     = mlxsw_sp_nexthop_group_hash,
3325 	.obj_hashfn  = mlxsw_sp_nexthop_group_hash_obj,
3326 	.obj_cmpfn   = mlxsw_sp_nexthop_group_cmp,
3327 };
3328 
3329 static int mlxsw_sp_nexthop_group_insert(struct mlxsw_sp *mlxsw_sp,
3330 					 struct mlxsw_sp_nexthop_group *nh_grp)
3331 {
3332 	if (nh_grp->type == MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6 &&
3333 	    !nh_grp->nhgi->gateway)
3334 		return 0;
3335 
3336 	return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_group_ht,
3337 				      &nh_grp->ht_node,
3338 				      mlxsw_sp_nexthop_group_ht_params);
3339 }
3340 
3341 static void mlxsw_sp_nexthop_group_remove(struct mlxsw_sp *mlxsw_sp,
3342 					  struct mlxsw_sp_nexthop_group *nh_grp)
3343 {
3344 	if (nh_grp->type == MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6 &&
3345 	    !nh_grp->nhgi->gateway)
3346 		return;
3347 
3348 	rhashtable_remove_fast(&mlxsw_sp->router->nexthop_group_ht,
3349 			       &nh_grp->ht_node,
3350 			       mlxsw_sp_nexthop_group_ht_params);
3351 }
3352 
3353 static struct mlxsw_sp_nexthop_group *
3354 mlxsw_sp_nexthop4_group_lookup(struct mlxsw_sp *mlxsw_sp,
3355 			       struct fib_info *fi)
3356 {
3357 	struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
3358 
3359 	cmp_arg.type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4;
3360 	cmp_arg.fi = fi;
3361 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
3362 				      &cmp_arg,
3363 				      mlxsw_sp_nexthop_group_ht_params);
3364 }
3365 
3366 static struct mlxsw_sp_nexthop_group *
3367 mlxsw_sp_nexthop6_group_lookup(struct mlxsw_sp *mlxsw_sp,
3368 			       struct mlxsw_sp_fib6_entry *fib6_entry)
3369 {
3370 	struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
3371 
3372 	cmp_arg.type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6;
3373 	cmp_arg.fib6_entry = fib6_entry;
3374 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
3375 				      &cmp_arg,
3376 				      mlxsw_sp_nexthop_group_ht_params);
3377 }
3378 
3379 static const struct rhashtable_params mlxsw_sp_nexthop_ht_params = {
3380 	.key_offset = offsetof(struct mlxsw_sp_nexthop, key),
3381 	.head_offset = offsetof(struct mlxsw_sp_nexthop, ht_node),
3382 	.key_len = sizeof(struct mlxsw_sp_nexthop_key),
3383 };
3384 
3385 static int mlxsw_sp_nexthop_insert(struct mlxsw_sp *mlxsw_sp,
3386 				   struct mlxsw_sp_nexthop *nh)
3387 {
3388 	return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_ht,
3389 				      &nh->ht_node, mlxsw_sp_nexthop_ht_params);
3390 }
3391 
3392 static void mlxsw_sp_nexthop_remove(struct mlxsw_sp *mlxsw_sp,
3393 				    struct mlxsw_sp_nexthop *nh)
3394 {
3395 	rhashtable_remove_fast(&mlxsw_sp->router->nexthop_ht, &nh->ht_node,
3396 			       mlxsw_sp_nexthop_ht_params);
3397 }
3398 
3399 static struct mlxsw_sp_nexthop *
3400 mlxsw_sp_nexthop_lookup(struct mlxsw_sp *mlxsw_sp,
3401 			struct mlxsw_sp_nexthop_key key)
3402 {
3403 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_ht, &key,
3404 				      mlxsw_sp_nexthop_ht_params);
3405 }
3406 
3407 static int mlxsw_sp_adj_index_mass_update_vr(struct mlxsw_sp *mlxsw_sp,
3408 					     enum mlxsw_sp_l3proto proto,
3409 					     u16 vr_id,
3410 					     u32 adj_index, u16 ecmp_size,
3411 					     u32 new_adj_index,
3412 					     u16 new_ecmp_size)
3413 {
3414 	char raleu_pl[MLXSW_REG_RALEU_LEN];
3415 
3416 	mlxsw_reg_raleu_pack(raleu_pl,
3417 			     (enum mlxsw_reg_ralxx_protocol) proto, vr_id,
3418 			     adj_index, ecmp_size, new_adj_index,
3419 			     new_ecmp_size);
3420 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raleu), raleu_pl);
3421 }
3422 
3423 static int mlxsw_sp_adj_index_mass_update(struct mlxsw_sp *mlxsw_sp,
3424 					  struct mlxsw_sp_nexthop_group *nh_grp,
3425 					  u32 old_adj_index, u16 old_ecmp_size)
3426 {
3427 	struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
3428 	struct mlxsw_sp_nexthop_group_vr_entry *vr_entry;
3429 	int err;
3430 
3431 	list_for_each_entry(vr_entry, &nh_grp->vr_list, list) {
3432 		err = mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp,
3433 							vr_entry->key.proto,
3434 							vr_entry->key.vr_id,
3435 							old_adj_index,
3436 							old_ecmp_size,
3437 							nhgi->adj_index,
3438 							nhgi->ecmp_size);
3439 		if (err)
3440 			goto err_mass_update_vr;
3441 	}
3442 	return 0;
3443 
3444 err_mass_update_vr:
3445 	list_for_each_entry_continue_reverse(vr_entry, &nh_grp->vr_list, list)
3446 		mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp, vr_entry->key.proto,
3447 						  vr_entry->key.vr_id,
3448 						  nhgi->adj_index,
3449 						  nhgi->ecmp_size,
3450 						  old_adj_index, old_ecmp_size);
3451 	return err;
3452 }
3453 
3454 static int __mlxsw_sp_nexthop_eth_update(struct mlxsw_sp *mlxsw_sp,
3455 					 u32 adj_index,
3456 					 struct mlxsw_sp_nexthop *nh,
3457 					 bool force, char *ratr_pl)
3458 {
3459 	struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
3460 	enum mlxsw_reg_ratr_op op;
3461 	u16 rif_index;
3462 
3463 	rif_index = nh->rif ? nh->rif->rif_index :
3464 			      mlxsw_sp->router->lb_rif_index;
3465 	op = force ? MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY :
3466 		     MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY_ON_ACTIVITY;
3467 	mlxsw_reg_ratr_pack(ratr_pl, op, true, MLXSW_REG_RATR_TYPE_ETHERNET,
3468 			    adj_index, rif_index);
3469 	switch (nh->action) {
3470 	case MLXSW_SP_NEXTHOP_ACTION_FORWARD:
3471 		mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
3472 		break;
3473 	case MLXSW_SP_NEXTHOP_ACTION_DISCARD:
3474 		mlxsw_reg_ratr_trap_action_set(ratr_pl,
3475 					       MLXSW_REG_RATR_TRAP_ACTION_DISCARD_ERRORS);
3476 		break;
3477 	case MLXSW_SP_NEXTHOP_ACTION_TRAP:
3478 		mlxsw_reg_ratr_trap_action_set(ratr_pl,
3479 					       MLXSW_REG_RATR_TRAP_ACTION_TRAP);
3480 		mlxsw_reg_ratr_trap_id_set(ratr_pl, MLXSW_TRAP_ID_RTR_EGRESS0);
3481 		break;
3482 	default:
3483 		WARN_ON_ONCE(1);
3484 		return -EINVAL;
3485 	}
3486 	if (nh->counter_valid)
3487 		mlxsw_reg_ratr_counter_pack(ratr_pl, nh->counter_index, true);
3488 	else
3489 		mlxsw_reg_ratr_counter_pack(ratr_pl, 0, false);
3490 
3491 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
3492 }
3493 
3494 int mlxsw_sp_nexthop_eth_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
3495 				struct mlxsw_sp_nexthop *nh, bool force,
3496 				char *ratr_pl)
3497 {
3498 	int i;
3499 
3500 	for (i = 0; i < nh->num_adj_entries; i++) {
3501 		int err;
3502 
3503 		err = __mlxsw_sp_nexthop_eth_update(mlxsw_sp, adj_index + i,
3504 						    nh, force, ratr_pl);
3505 		if (err)
3506 			return err;
3507 	}
3508 
3509 	return 0;
3510 }
3511 
3512 static int __mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
3513 					  u32 adj_index,
3514 					  struct mlxsw_sp_nexthop *nh,
3515 					  bool force, char *ratr_pl)
3516 {
3517 	const struct mlxsw_sp_ipip_ops *ipip_ops;
3518 
3519 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[nh->ipip_entry->ipipt];
3520 	return ipip_ops->nexthop_update(mlxsw_sp, adj_index, nh->ipip_entry,
3521 					force, ratr_pl);
3522 }
3523 
3524 static int mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
3525 					u32 adj_index,
3526 					struct mlxsw_sp_nexthop *nh, bool force,
3527 					char *ratr_pl)
3528 {
3529 	int i;
3530 
3531 	for (i = 0; i < nh->num_adj_entries; i++) {
3532 		int err;
3533 
3534 		err = __mlxsw_sp_nexthop_ipip_update(mlxsw_sp, adj_index + i,
3535 						     nh, force, ratr_pl);
3536 		if (err)
3537 			return err;
3538 	}
3539 
3540 	return 0;
3541 }
3542 
3543 static int mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
3544 				   struct mlxsw_sp_nexthop *nh, bool force,
3545 				   char *ratr_pl)
3546 {
3547 	/* When action is discard or trap, the nexthop must be
3548 	 * programmed as an Ethernet nexthop.
3549 	 */
3550 	if (nh->type == MLXSW_SP_NEXTHOP_TYPE_ETH ||
3551 	    nh->action == MLXSW_SP_NEXTHOP_ACTION_DISCARD ||
3552 	    nh->action == MLXSW_SP_NEXTHOP_ACTION_TRAP)
3553 		return mlxsw_sp_nexthop_eth_update(mlxsw_sp, adj_index, nh,
3554 						   force, ratr_pl);
3555 	else
3556 		return mlxsw_sp_nexthop_ipip_update(mlxsw_sp, adj_index, nh,
3557 						    force, ratr_pl);
3558 }
3559 
3560 static int
3561 mlxsw_sp_nexthop_group_update(struct mlxsw_sp *mlxsw_sp,
3562 			      struct mlxsw_sp_nexthop_group_info *nhgi,
3563 			      bool reallocate)
3564 {
3565 	char ratr_pl[MLXSW_REG_RATR_LEN];
3566 	u32 adj_index = nhgi->adj_index; /* base */
3567 	struct mlxsw_sp_nexthop *nh;
3568 	int i;
3569 
3570 	for (i = 0; i < nhgi->count; i++) {
3571 		nh = &nhgi->nexthops[i];
3572 
3573 		if (!nh->should_offload) {
3574 			nh->offloaded = 0;
3575 			continue;
3576 		}
3577 
3578 		if (nh->update || reallocate) {
3579 			int err = 0;
3580 
3581 			err = mlxsw_sp_nexthop_update(mlxsw_sp, adj_index, nh,
3582 						      true, ratr_pl);
3583 			if (err)
3584 				return err;
3585 			nh->update = 0;
3586 			nh->offloaded = 1;
3587 		}
3588 		adj_index += nh->num_adj_entries;
3589 	}
3590 	return 0;
3591 }
3592 
3593 static int
3594 mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
3595 				    struct mlxsw_sp_nexthop_group *nh_grp)
3596 {
3597 	struct mlxsw_sp_fib_entry *fib_entry;
3598 	int err;
3599 
3600 	list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
3601 		err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
3602 		if (err)
3603 			return err;
3604 	}
3605 	return 0;
3606 }
3607 
3608 struct mlxsw_sp_adj_grp_size_range {
3609 	u16 start; /* Inclusive */
3610 	u16 end; /* Inclusive */
3611 };
3612 
3613 /* Ordered by range start value */
3614 static const struct mlxsw_sp_adj_grp_size_range
3615 mlxsw_sp1_adj_grp_size_ranges[] = {
3616 	{ .start = 1, .end = 64 },
3617 	{ .start = 512, .end = 512 },
3618 	{ .start = 1024, .end = 1024 },
3619 	{ .start = 2048, .end = 2048 },
3620 	{ .start = 4096, .end = 4096 },
3621 };
3622 
3623 /* Ordered by range start value */
3624 static const struct mlxsw_sp_adj_grp_size_range
3625 mlxsw_sp2_adj_grp_size_ranges[] = {
3626 	{ .start = 1, .end = 128 },
3627 	{ .start = 256, .end = 256 },
3628 	{ .start = 512, .end = 512 },
3629 	{ .start = 1024, .end = 1024 },
3630 	{ .start = 2048, .end = 2048 },
3631 	{ .start = 4096, .end = 4096 },
3632 };
3633 
3634 static void mlxsw_sp_adj_grp_size_round_up(const struct mlxsw_sp *mlxsw_sp,
3635 					   u16 *p_adj_grp_size)
3636 {
3637 	int i;
3638 
3639 	for (i = 0; i < mlxsw_sp->router->adj_grp_size_ranges_count; i++) {
3640 		const struct mlxsw_sp_adj_grp_size_range *size_range;
3641 
3642 		size_range = &mlxsw_sp->router->adj_grp_size_ranges[i];
3643 
3644 		if (*p_adj_grp_size >= size_range->start &&
3645 		    *p_adj_grp_size <= size_range->end)
3646 			return;
3647 
3648 		if (*p_adj_grp_size <= size_range->end) {
3649 			*p_adj_grp_size = size_range->end;
3650 			return;
3651 		}
3652 	}
3653 }
3654 
3655 static void mlxsw_sp_adj_grp_size_round_down(const struct mlxsw_sp *mlxsw_sp,
3656 					     u16 *p_adj_grp_size,
3657 					     unsigned int alloc_size)
3658 {
3659 	int i;
3660 
3661 	for (i = mlxsw_sp->router->adj_grp_size_ranges_count - 1; i >= 0; i--) {
3662 		const struct mlxsw_sp_adj_grp_size_range *size_range;
3663 
3664 		size_range = &mlxsw_sp->router->adj_grp_size_ranges[i];
3665 
3666 		if (alloc_size >= size_range->end) {
3667 			*p_adj_grp_size = size_range->end;
3668 			return;
3669 		}
3670 	}
3671 }
3672 
3673 static int mlxsw_sp_fix_adj_grp_size(struct mlxsw_sp *mlxsw_sp,
3674 				     u16 *p_adj_grp_size)
3675 {
3676 	unsigned int alloc_size;
3677 	int err;
3678 
3679 	/* Round up the requested group size to the next size supported
3680 	 * by the device and make sure the request can be satisfied.
3681 	 */
3682 	mlxsw_sp_adj_grp_size_round_up(mlxsw_sp, p_adj_grp_size);
3683 	err = mlxsw_sp_kvdl_alloc_count_query(mlxsw_sp,
3684 					      MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3685 					      *p_adj_grp_size, &alloc_size);
3686 	if (err)
3687 		return err;
3688 	/* It is possible the allocation results in more allocated
3689 	 * entries than requested. Try to use as much of them as
3690 	 * possible.
3691 	 */
3692 	mlxsw_sp_adj_grp_size_round_down(mlxsw_sp, p_adj_grp_size, alloc_size);
3693 
3694 	return 0;
3695 }
3696 
3697 static void
3698 mlxsw_sp_nexthop_group_normalize(struct mlxsw_sp_nexthop_group_info *nhgi)
3699 {
3700 	int i, g = 0, sum_norm_weight = 0;
3701 	struct mlxsw_sp_nexthop *nh;
3702 
3703 	for (i = 0; i < nhgi->count; i++) {
3704 		nh = &nhgi->nexthops[i];
3705 
3706 		if (!nh->should_offload)
3707 			continue;
3708 		if (g > 0)
3709 			g = gcd(nh->nh_weight, g);
3710 		else
3711 			g = nh->nh_weight;
3712 	}
3713 
3714 	for (i = 0; i < nhgi->count; i++) {
3715 		nh = &nhgi->nexthops[i];
3716 
3717 		if (!nh->should_offload)
3718 			continue;
3719 		nh->norm_nh_weight = nh->nh_weight / g;
3720 		sum_norm_weight += nh->norm_nh_weight;
3721 	}
3722 
3723 	nhgi->sum_norm_weight = sum_norm_weight;
3724 }
3725 
3726 static void
3727 mlxsw_sp_nexthop_group_rebalance(struct mlxsw_sp_nexthop_group_info *nhgi)
3728 {
3729 	int i, weight = 0, lower_bound = 0;
3730 	int total = nhgi->sum_norm_weight;
3731 	u16 ecmp_size = nhgi->ecmp_size;
3732 
3733 	for (i = 0; i < nhgi->count; i++) {
3734 		struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
3735 		int upper_bound;
3736 
3737 		if (!nh->should_offload)
3738 			continue;
3739 		weight += nh->norm_nh_weight;
3740 		upper_bound = DIV_ROUND_CLOSEST(ecmp_size * weight, total);
3741 		nh->num_adj_entries = upper_bound - lower_bound;
3742 		lower_bound = upper_bound;
3743 	}
3744 }
3745 
3746 static struct mlxsw_sp_nexthop *
3747 mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
3748 		     const struct mlxsw_sp_rt6 *mlxsw_sp_rt6);
3749 
3750 static void
3751 mlxsw_sp_nexthop4_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3752 					struct mlxsw_sp_nexthop_group *nh_grp)
3753 {
3754 	int i;
3755 
3756 	for (i = 0; i < nh_grp->nhgi->count; i++) {
3757 		struct mlxsw_sp_nexthop *nh = &nh_grp->nhgi->nexthops[i];
3758 
3759 		if (nh->offloaded)
3760 			nh->key.fib_nh->fib_nh_flags |= RTNH_F_OFFLOAD;
3761 		else
3762 			nh->key.fib_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
3763 	}
3764 }
3765 
3766 static void
3767 __mlxsw_sp_nexthop6_group_offload_refresh(struct mlxsw_sp_nexthop_group *nh_grp,
3768 					  struct mlxsw_sp_fib6_entry *fib6_entry)
3769 {
3770 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3771 
3772 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3773 		struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3774 		struct mlxsw_sp_nexthop *nh;
3775 
3776 		nh = mlxsw_sp_rt6_nexthop(nh_grp, mlxsw_sp_rt6);
3777 		if (nh && nh->offloaded)
3778 			fib6_nh->fib_nh_flags |= RTNH_F_OFFLOAD;
3779 		else
3780 			fib6_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
3781 	}
3782 }
3783 
3784 static void
3785 mlxsw_sp_nexthop6_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3786 					struct mlxsw_sp_nexthop_group *nh_grp)
3787 {
3788 	struct mlxsw_sp_fib6_entry *fib6_entry;
3789 
3790 	/* Unfortunately, in IPv6 the route and the nexthop are described by
3791 	 * the same struct, so we need to iterate over all the routes using the
3792 	 * nexthop group and set / clear the offload indication for them.
3793 	 */
3794 	list_for_each_entry(fib6_entry, &nh_grp->fib_list,
3795 			    common.nexthop_group_node)
3796 		__mlxsw_sp_nexthop6_group_offload_refresh(nh_grp, fib6_entry);
3797 }
3798 
3799 static void
3800 mlxsw_sp_nexthop_bucket_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3801 					const struct mlxsw_sp_nexthop *nh,
3802 					u16 bucket_index)
3803 {
3804 	struct mlxsw_sp_nexthop_group *nh_grp = nh->nhgi->nh_grp;
3805 	bool offload = false, trap = false;
3806 
3807 	if (nh->offloaded) {
3808 		if (nh->action == MLXSW_SP_NEXTHOP_ACTION_TRAP)
3809 			trap = true;
3810 		else
3811 			offload = true;
3812 	}
3813 	nexthop_bucket_set_hw_flags(mlxsw_sp_net(mlxsw_sp), nh_grp->obj.id,
3814 				    bucket_index, offload, trap);
3815 }
3816 
3817 static void
3818 mlxsw_sp_nexthop_obj_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3819 					   struct mlxsw_sp_nexthop_group *nh_grp)
3820 {
3821 	int i;
3822 
3823 	/* Do not update the flags if the nexthop group is being destroyed
3824 	 * since:
3825 	 * 1. The nexthop objects is being deleted, in which case the flags are
3826 	 * irrelevant.
3827 	 * 2. The nexthop group was replaced by a newer group, in which case
3828 	 * the flags of the nexthop object were already updated based on the
3829 	 * new group.
3830 	 */
3831 	if (nh_grp->can_destroy)
3832 		return;
3833 
3834 	nexthop_set_hw_flags(mlxsw_sp_net(mlxsw_sp), nh_grp->obj.id,
3835 			     nh_grp->nhgi->adj_index_valid, false);
3836 
3837 	/* Update flags of individual nexthop buckets in case of a resilient
3838 	 * nexthop group.
3839 	 */
3840 	if (!nh_grp->nhgi->is_resilient)
3841 		return;
3842 
3843 	for (i = 0; i < nh_grp->nhgi->count; i++) {
3844 		struct mlxsw_sp_nexthop *nh = &nh_grp->nhgi->nexthops[i];
3845 
3846 		mlxsw_sp_nexthop_bucket_offload_refresh(mlxsw_sp, nh, i);
3847 	}
3848 }
3849 
3850 static void
3851 mlxsw_sp_nexthop_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3852 				       struct mlxsw_sp_nexthop_group *nh_grp)
3853 {
3854 	switch (nh_grp->type) {
3855 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3856 		mlxsw_sp_nexthop4_group_offload_refresh(mlxsw_sp, nh_grp);
3857 		break;
3858 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3859 		mlxsw_sp_nexthop6_group_offload_refresh(mlxsw_sp, nh_grp);
3860 		break;
3861 	case MLXSW_SP_NEXTHOP_GROUP_TYPE_OBJ:
3862 		mlxsw_sp_nexthop_obj_group_offload_refresh(mlxsw_sp, nh_grp);
3863 		break;
3864 	}
3865 }
3866 
3867 static int
3868 mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
3869 			       struct mlxsw_sp_nexthop_group *nh_grp)
3870 {
3871 	struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
3872 	u16 ecmp_size, old_ecmp_size;
3873 	struct mlxsw_sp_nexthop *nh;
3874 	bool offload_change = false;
3875 	u32 adj_index;
3876 	bool old_adj_index_valid;
3877 	u32 old_adj_index;
3878 	int i, err2, err;
3879 
3880 	if (!nhgi->gateway)
3881 		return mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3882 
3883 	for (i = 0; i < nhgi->count; i++) {
3884 		nh = &nhgi->nexthops[i];
3885 
3886 		if (nh->should_offload != nh->offloaded) {
3887 			offload_change = true;
3888 			if (nh->should_offload)
3889 				nh->update = 1;
3890 		}
3891 	}
3892 	if (!offload_change) {
3893 		/* Nothing was added or removed, so no need to reallocate. Just
3894 		 * update MAC on existing adjacency indexes.
3895 		 */
3896 		err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nhgi, false);
3897 		if (err) {
3898 			dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3899 			goto set_trap;
3900 		}
3901 		/* Flags of individual nexthop buckets might need to be
3902 		 * updated.
3903 		 */
3904 		mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
3905 		return 0;
3906 	}
3907 	mlxsw_sp_nexthop_group_normalize(nhgi);
3908 	if (!nhgi->sum_norm_weight) {
3909 		/* No neigh of this group is connected so we just set
3910 		 * the trap and let everthing flow through kernel.
3911 		 */
3912 		err = 0;
3913 		goto set_trap;
3914 	}
3915 
3916 	ecmp_size = nhgi->sum_norm_weight;
3917 	err = mlxsw_sp_fix_adj_grp_size(mlxsw_sp, &ecmp_size);
3918 	if (err)
3919 		/* No valid allocation size available. */
3920 		goto set_trap;
3921 
3922 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3923 				  ecmp_size, &adj_index);
3924 	if (err) {
3925 		/* We ran out of KVD linear space, just set the
3926 		 * trap and let everything flow through kernel.
3927 		 */
3928 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to allocate KVD linear area for nexthop group.\n");
3929 		goto set_trap;
3930 	}
3931 	old_adj_index_valid = nhgi->adj_index_valid;
3932 	old_adj_index = nhgi->adj_index;
3933 	old_ecmp_size = nhgi->ecmp_size;
3934 	nhgi->adj_index_valid = 1;
3935 	nhgi->adj_index = adj_index;
3936 	nhgi->ecmp_size = ecmp_size;
3937 	mlxsw_sp_nexthop_group_rebalance(nhgi);
3938 	err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nhgi, true);
3939 	if (err) {
3940 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3941 		goto set_trap;
3942 	}
3943 
3944 	mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
3945 
3946 	if (!old_adj_index_valid) {
3947 		/* The trap was set for fib entries, so we have to call
3948 		 * fib entry update to unset it and use adjacency index.
3949 		 */
3950 		err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3951 		if (err) {
3952 			dev_warn(mlxsw_sp->bus_info->dev, "Failed to add adjacency index to fib entries.\n");
3953 			goto set_trap;
3954 		}
3955 		return 0;
3956 	}
3957 
3958 	err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, nh_grp,
3959 					     old_adj_index, old_ecmp_size);
3960 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3961 			   old_ecmp_size, old_adj_index);
3962 	if (err) {
3963 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to mass-update adjacency index for nexthop group.\n");
3964 		goto set_trap;
3965 	}
3966 
3967 	return 0;
3968 
3969 set_trap:
3970 	old_adj_index_valid = nhgi->adj_index_valid;
3971 	nhgi->adj_index_valid = 0;
3972 	for (i = 0; i < nhgi->count; i++) {
3973 		nh = &nhgi->nexthops[i];
3974 		nh->offloaded = 0;
3975 	}
3976 	err2 = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3977 	if (err2)
3978 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to set traps for fib entries.\n");
3979 	mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
3980 	if (old_adj_index_valid)
3981 		mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3982 				   nhgi->ecmp_size, nhgi->adj_index);
3983 	return err;
3984 }
3985 
3986 static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
3987 					    bool removing)
3988 {
3989 	if (!removing) {
3990 		nh->action = MLXSW_SP_NEXTHOP_ACTION_FORWARD;
3991 		nh->should_offload = 1;
3992 	} else if (nh->nhgi->is_resilient) {
3993 		nh->action = MLXSW_SP_NEXTHOP_ACTION_TRAP;
3994 		nh->should_offload = 1;
3995 	} else {
3996 		nh->should_offload = 0;
3997 	}
3998 	nh->update = 1;
3999 }
4000 
4001 static int
4002 mlxsw_sp_nexthop_dead_neigh_replace(struct mlxsw_sp *mlxsw_sp,
4003 				    struct mlxsw_sp_neigh_entry *neigh_entry)
4004 {
4005 	struct neighbour *n, *old_n = neigh_entry->key.n;
4006 	struct mlxsw_sp_nexthop *nh;
4007 	bool entry_connected;
4008 	u8 nud_state, dead;
4009 	int err;
4010 
4011 	nh = list_first_entry(&neigh_entry->nexthop_list,
4012 			      struct mlxsw_sp_nexthop, neigh_list_node);
4013 
4014 	n = neigh_lookup(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
4015 	if (!n) {
4016 		n = neigh_create(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
4017 		if (IS_ERR(n))
4018 			return PTR_ERR(n);
4019 		neigh_event_send(n, NULL);
4020 	}
4021 
4022 	mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
4023 	neigh_entry->key.n = n;
4024 	err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
4025 	if (err)
4026 		goto err_neigh_entry_insert;
4027 
4028 	read_lock_bh(&n->lock);
4029 	nud_state = n->nud_state;
4030 	dead = n->dead;
4031 	read_unlock_bh(&n->lock);
4032 	entry_connected = nud_state & NUD_VALID && !dead;
4033 
4034 	list_for_each_entry(nh, &neigh_entry->nexthop_list,
4035 			    neigh_list_node) {
4036 		neigh_release(old_n);
4037 		neigh_clone(n);
4038 		__mlxsw_sp_nexthop_neigh_update(nh, !entry_connected);
4039 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4040 	}
4041 
4042 	neigh_release(n);
4043 
4044 	return 0;
4045 
4046 err_neigh_entry_insert:
4047 	neigh_entry->key.n = old_n;
4048 	mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
4049 	neigh_release(n);
4050 	return err;
4051 }
4052 
4053 static void
4054 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
4055 			      struct mlxsw_sp_neigh_entry *neigh_entry,
4056 			      bool removing, bool dead)
4057 {
4058 	struct mlxsw_sp_nexthop *nh;
4059 
4060 	if (list_empty(&neigh_entry->nexthop_list))
4061 		return;
4062 
4063 	if (dead) {
4064 		int err;
4065 
4066 		err = mlxsw_sp_nexthop_dead_neigh_replace(mlxsw_sp,
4067 							  neigh_entry);
4068 		if (err)
4069 			dev_err(mlxsw_sp->bus_info->dev, "Failed to replace dead neigh\n");
4070 		return;
4071 	}
4072 
4073 	list_for_each_entry(nh, &neigh_entry->nexthop_list,
4074 			    neigh_list_node) {
4075 		__mlxsw_sp_nexthop_neigh_update(nh, removing);
4076 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4077 	}
4078 }
4079 
4080 static void mlxsw_sp_nexthop_rif_init(struct mlxsw_sp_nexthop *nh,
4081 				      struct mlxsw_sp_rif *rif)
4082 {
4083 	if (nh->rif)
4084 		return;
4085 
4086 	nh->rif = rif;
4087 	list_add(&nh->rif_list_node, &rif->nexthop_list);
4088 }
4089 
4090 static void mlxsw_sp_nexthop_rif_fini(struct mlxsw_sp_nexthop *nh)
4091 {
4092 	if (!nh->rif)
4093 		return;
4094 
4095 	list_del(&nh->rif_list_node);
4096 	nh->rif = NULL;
4097 }
4098 
4099 static int mlxsw_sp_nexthop_neigh_init(struct mlxsw_sp *mlxsw_sp,
4100 				       struct mlxsw_sp_nexthop *nh)
4101 {
4102 	struct mlxsw_sp_neigh_entry *neigh_entry;
4103 	struct neighbour *n;
4104 	u8 nud_state, dead;
4105 	int err;
4106 
4107 	if (!nh->nhgi->gateway || nh->neigh_entry)
4108 		return 0;
4109 
4110 	/* Take a reference of neigh here ensuring that neigh would
4111 	 * not be destructed before the nexthop entry is finished.
4112 	 * The reference is taken either in neigh_lookup() or
4113 	 * in neigh_create() in case n is not found.
4114 	 */
4115 	n = neigh_lookup(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
4116 	if (!n) {
4117 		n = neigh_create(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
4118 		if (IS_ERR(n))
4119 			return PTR_ERR(n);
4120 		neigh_event_send(n, NULL);
4121 	}
4122 	neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
4123 	if (!neigh_entry) {
4124 		neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
4125 		if (IS_ERR(neigh_entry)) {
4126 			err = -EINVAL;
4127 			goto err_neigh_entry_create;
4128 		}
4129 	}
4130 
4131 	/* If that is the first nexthop connected to that neigh, add to
4132 	 * nexthop_neighs_list
4133 	 */
4134 	if (list_empty(&neigh_entry->nexthop_list))
4135 		list_add_tail(&neigh_entry->nexthop_neighs_list_node,
4136 			      &mlxsw_sp->router->nexthop_neighs_list);
4137 
4138 	nh->neigh_entry = neigh_entry;
4139 	list_add_tail(&nh->neigh_list_node, &neigh_entry->nexthop_list);
4140 	read_lock_bh(&n->lock);
4141 	nud_state = n->nud_state;
4142 	dead = n->dead;
4143 	read_unlock_bh(&n->lock);
4144 	__mlxsw_sp_nexthop_neigh_update(nh, !(nud_state & NUD_VALID && !dead));
4145 
4146 	return 0;
4147 
4148 err_neigh_entry_create:
4149 	neigh_release(n);
4150 	return err;
4151 }
4152 
4153 static void mlxsw_sp_nexthop_neigh_fini(struct mlxsw_sp *mlxsw_sp,
4154 					struct mlxsw_sp_nexthop *nh)
4155 {
4156 	struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
4157 	struct neighbour *n;
4158 
4159 	if (!neigh_entry)
4160 		return;
4161 	n = neigh_entry->key.n;
4162 
4163 	__mlxsw_sp_nexthop_neigh_update(nh, true);
4164 	list_del(&nh->neigh_list_node);
4165 	nh->neigh_entry = NULL;
4166 
4167 	/* If that is the last nexthop connected to that neigh, remove from
4168 	 * nexthop_neighs_list
4169 	 */
4170 	if (list_empty(&neigh_entry->nexthop_list))
4171 		list_del(&neigh_entry->nexthop_neighs_list_node);
4172 
4173 	if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
4174 		mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
4175 
4176 	neigh_release(n);
4177 }
4178 
4179 static bool mlxsw_sp_ipip_netdev_ul_up(struct net_device *ol_dev)
4180 {
4181 	struct net_device *ul_dev;
4182 	bool is_up;
4183 
4184 	rcu_read_lock();
4185 	ul_dev = mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
4186 	is_up = ul_dev ? (ul_dev->flags & IFF_UP) : true;
4187 	rcu_read_unlock();
4188 
4189 	return is_up;
4190 }
4191 
4192 static void mlxsw_sp_nexthop_ipip_init(struct mlxsw_sp *mlxsw_sp,
4193 				       struct mlxsw_sp_nexthop *nh,
4194 				       struct mlxsw_sp_ipip_entry *ipip_entry)
4195 {
4196 	bool removing;
4197 
4198 	if (!nh->nhgi->gateway || nh->ipip_entry)
4199 		return;
4200 
4201 	nh->ipip_entry = ipip_entry;
4202 	removing = !mlxsw_sp_ipip_netdev_ul_up(ipip_entry->ol_dev);
4203 	__mlxsw_sp_nexthop_neigh_update(nh, removing);
4204 	mlxsw_sp_nexthop_rif_init(nh, &ipip_entry->ol_lb->common);
4205 }
4206 
4207 static void mlxsw_sp_nexthop_ipip_fini(struct mlxsw_sp *mlxsw_sp,
4208 				       struct mlxsw_sp_nexthop *nh)
4209 {
4210 	struct mlxsw_sp_ipip_entry *ipip_entry = nh->ipip_entry;
4211 
4212 	if (!ipip_entry)
4213 		return;
4214 
4215 	__mlxsw_sp_nexthop_neigh_update(nh, true);
4216 	nh->ipip_entry = NULL;
4217 }
4218 
4219 static bool mlxsw_sp_nexthop4_ipip_type(const struct mlxsw_sp *mlxsw_sp,
4220 					const struct fib_nh *fib_nh,
4221 					enum mlxsw_sp_ipip_type *p_ipipt)
4222 {
4223 	struct net_device *dev = fib_nh->fib_nh_dev;
4224 
4225 	return dev &&
4226 	       fib_nh->nh_parent->fib_type == RTN_UNICAST &&
4227 	       mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, p_ipipt);
4228 }
4229 
4230 static int mlxsw_sp_nexthop_type_init(struct mlxsw_sp *mlxsw_sp,
4231 				      struct mlxsw_sp_nexthop *nh,
4232 				      const struct net_device *dev)
4233 {
4234 	const struct mlxsw_sp_ipip_ops *ipip_ops;
4235 	struct mlxsw_sp_ipip_entry *ipip_entry;
4236 	struct mlxsw_sp_rif *rif;
4237 	int err;
4238 
4239 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, dev);
4240 	if (ipip_entry) {
4241 		ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
4242 		if (ipip_ops->can_offload(mlxsw_sp, dev)) {
4243 			nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
4244 			mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, ipip_entry);
4245 			return 0;
4246 		}
4247 	}
4248 
4249 	nh->type = MLXSW_SP_NEXTHOP_TYPE_ETH;
4250 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
4251 	if (!rif)
4252 		return 0;
4253 
4254 	mlxsw_sp_nexthop_rif_init(nh, rif);
4255 	err = mlxsw_sp_nexthop_neigh_init(mlxsw_sp, nh);
4256 	if (err)
4257 		goto err_neigh_init;
4258 
4259 	return 0;
4260 
4261 err_neigh_init:
4262 	mlxsw_sp_nexthop_rif_fini(nh);
4263 	return err;
4264 }
4265 
4266 static void mlxsw_sp_nexthop_type_fini(struct mlxsw_sp *mlxsw_sp,
4267 				       struct mlxsw_sp_nexthop *nh)
4268 {
4269 	switch (nh->type) {
4270 	case MLXSW_SP_NEXTHOP_TYPE_ETH:
4271 		mlxsw_sp_nexthop_neigh_fini(mlxsw_sp, nh);
4272 		mlxsw_sp_nexthop_rif_fini(nh);
4273 		break;
4274 	case MLXSW_SP_NEXTHOP_TYPE_IPIP:
4275 		mlxsw_sp_nexthop_rif_fini(nh);
4276 		mlxsw_sp_nexthop_ipip_fini(mlxsw_sp, nh);
4277 		break;
4278 	}
4279 }
4280 
4281 static int mlxsw_sp_nexthop4_init(struct mlxsw_sp *mlxsw_sp,
4282 				  struct mlxsw_sp_nexthop_group *nh_grp,
4283 				  struct mlxsw_sp_nexthop *nh,
4284 				  struct fib_nh *fib_nh)
4285 {
4286 	struct net_device *dev = fib_nh->fib_nh_dev;
4287 	struct in_device *in_dev;
4288 	int err;
4289 
4290 	nh->nhgi = nh_grp->nhgi;
4291 	nh->key.fib_nh = fib_nh;
4292 #ifdef CONFIG_IP_ROUTE_MULTIPATH
4293 	nh->nh_weight = fib_nh->fib_nh_weight;
4294 #else
4295 	nh->nh_weight = 1;
4296 #endif
4297 	memcpy(&nh->gw_addr, &fib_nh->fib_nh_gw4, sizeof(fib_nh->fib_nh_gw4));
4298 	nh->neigh_tbl = &arp_tbl;
4299 	err = mlxsw_sp_nexthop_insert(mlxsw_sp, nh);
4300 	if (err)
4301 		return err;
4302 
4303 	mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
4304 	list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
4305 
4306 	if (!dev)
4307 		return 0;
4308 	nh->ifindex = dev->ifindex;
4309 
4310 	rcu_read_lock();
4311 	in_dev = __in_dev_get_rcu(dev);
4312 	if (in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
4313 	    fib_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
4314 		rcu_read_unlock();
4315 		return 0;
4316 	}
4317 	rcu_read_unlock();
4318 
4319 	err = mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, dev);
4320 	if (err)
4321 		goto err_nexthop_neigh_init;
4322 
4323 	return 0;
4324 
4325 err_nexthop_neigh_init:
4326 	list_del(&nh->router_list_node);
4327 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
4328 	mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
4329 	return err;
4330 }
4331 
4332 static void mlxsw_sp_nexthop4_fini(struct mlxsw_sp *mlxsw_sp,
4333 				   struct mlxsw_sp_nexthop *nh)
4334 {
4335 	mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4336 	list_del(&nh->router_list_node);
4337 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
4338 	mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
4339 }
4340 
4341 static void mlxsw_sp_nexthop4_event(struct mlxsw_sp *mlxsw_sp,
4342 				    unsigned long event, struct fib_nh *fib_nh)
4343 {
4344 	struct mlxsw_sp_nexthop_key key;
4345 	struct mlxsw_sp_nexthop *nh;
4346 
4347 	key.fib_nh = fib_nh;
4348 	nh = mlxsw_sp_nexthop_lookup(mlxsw_sp, key);
4349 	if (!nh)
4350 		return;
4351 
4352 	switch (event) {
4353 	case FIB_EVENT_NH_ADD:
4354 		mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, fib_nh->fib_nh_dev);
4355 		break;
4356 	case FIB_EVENT_NH_DEL:
4357 		mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4358 		break;
4359 	}
4360 
4361 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4362 }
4363 
4364 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
4365 					struct mlxsw_sp_rif *rif)
4366 {
4367 	struct mlxsw_sp_nexthop *nh;
4368 	bool removing;
4369 
4370 	list_for_each_entry(nh, &rif->nexthop_list, rif_list_node) {
4371 		switch (nh->type) {
4372 		case MLXSW_SP_NEXTHOP_TYPE_ETH:
4373 			removing = false;
4374 			break;
4375 		case MLXSW_SP_NEXTHOP_TYPE_IPIP:
4376 			removing = !mlxsw_sp_ipip_netdev_ul_up(rif->dev);
4377 			break;
4378 		default:
4379 			WARN_ON(1);
4380 			continue;
4381 		}
4382 
4383 		__mlxsw_sp_nexthop_neigh_update(nh, removing);
4384 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4385 	}
4386 }
4387 
4388 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
4389 					 struct mlxsw_sp_rif *old_rif,
4390 					 struct mlxsw_sp_rif *new_rif)
4391 {
4392 	struct mlxsw_sp_nexthop *nh;
4393 
4394 	list_splice_init(&old_rif->nexthop_list, &new_rif->nexthop_list);
4395 	list_for_each_entry(nh, &new_rif->nexthop_list, rif_list_node)
4396 		nh->rif = new_rif;
4397 	mlxsw_sp_nexthop_rif_update(mlxsw_sp, new_rif);
4398 }
4399 
4400 static void mlxsw_sp_nexthop_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
4401 					   struct mlxsw_sp_rif *rif)
4402 {
4403 	struct mlxsw_sp_nexthop *nh, *tmp;
4404 
4405 	list_for_each_entry_safe(nh, tmp, &rif->nexthop_list, rif_list_node) {
4406 		mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4407 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4408 	}
4409 }
4410 
4411 static int mlxsw_sp_adj_trap_entry_init(struct mlxsw_sp *mlxsw_sp)
4412 {
4413 	enum mlxsw_reg_ratr_trap_action trap_action;
4414 	char ratr_pl[MLXSW_REG_RATR_LEN];
4415 	int err;
4416 
4417 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
4418 				  &mlxsw_sp->router->adj_trap_index);
4419 	if (err)
4420 		return err;
4421 
4422 	trap_action = MLXSW_REG_RATR_TRAP_ACTION_TRAP;
4423 	mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY, true,
4424 			    MLXSW_REG_RATR_TYPE_ETHERNET,
4425 			    mlxsw_sp->router->adj_trap_index,
4426 			    mlxsw_sp->router->lb_rif_index);
4427 	mlxsw_reg_ratr_trap_action_set(ratr_pl, trap_action);
4428 	mlxsw_reg_ratr_trap_id_set(ratr_pl, MLXSW_TRAP_ID_RTR_EGRESS0);
4429 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
4430 	if (err)
4431 		goto err_ratr_write;
4432 
4433 	return 0;
4434 
4435 err_ratr_write:
4436 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
4437 			   mlxsw_sp->router->adj_trap_index);
4438 	return err;
4439 }
4440 
4441 static void mlxsw_sp_adj_trap_entry_fini(struct mlxsw_sp *mlxsw_sp)
4442 {
4443 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
4444 			   mlxsw_sp->router->adj_trap_index);
4445 }
4446 
4447 static int mlxsw_sp_nexthop_group_inc(struct mlxsw_sp *mlxsw_sp)
4448 {
4449 	int err;
4450 
4451 	if (refcount_inc_not_zero(&mlxsw_sp->router->num_groups))
4452 		return 0;
4453 
4454 	err = mlxsw_sp_adj_trap_entry_init(mlxsw_sp);
4455 	if (err)
4456 		return err;
4457 
4458 	refcount_set(&mlxsw_sp->router->num_groups, 1);
4459 
4460 	return 0;
4461 }
4462 
4463 static void mlxsw_sp_nexthop_group_dec(struct mlxsw_sp *mlxsw_sp)
4464 {
4465 	if (!refcount_dec_and_test(&mlxsw_sp->router->num_groups))
4466 		return;
4467 
4468 	mlxsw_sp_adj_trap_entry_fini(mlxsw_sp);
4469 }
4470 
4471 static void
4472 mlxsw_sp_nh_grp_activity_get(struct mlxsw_sp *mlxsw_sp,
4473 			     const struct mlxsw_sp_nexthop_group *nh_grp,
4474 			     unsigned long *activity)
4475 {
4476 	char *ratrad_pl;
4477 	int i, err;
4478 
4479 	ratrad_pl = kmalloc(MLXSW_REG_RATRAD_LEN, GFP_KERNEL);
4480 	if (!ratrad_pl)
4481 		return;
4482 
4483 	mlxsw_reg_ratrad_pack(ratrad_pl, nh_grp->nhgi->adj_index,
4484 			      nh_grp->nhgi->count);
4485 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ratrad), ratrad_pl);
4486 	if (err)
4487 		goto out;
4488 
4489 	for (i = 0; i < nh_grp->nhgi->count; i++) {
4490 		if (!mlxsw_reg_ratrad_activity_vector_get(ratrad_pl, i))
4491 			continue;
4492 		bitmap_set(activity, i, 1);
4493 	}
4494 
4495 out:
4496 	kfree(ratrad_pl);
4497 }
4498 
4499 #define MLXSW_SP_NH_GRP_ACTIVITY_UPDATE_INTERVAL 1000 /* ms */
4500 
4501 static void
4502 mlxsw_sp_nh_grp_activity_update(struct mlxsw_sp *mlxsw_sp,
4503 				const struct mlxsw_sp_nexthop_group *nh_grp)
4504 {
4505 	unsigned long *activity;
4506 
4507 	activity = bitmap_zalloc(nh_grp->nhgi->count, GFP_KERNEL);
4508 	if (!activity)
4509 		return;
4510 
4511 	mlxsw_sp_nh_grp_activity_get(mlxsw_sp, nh_grp, activity);
4512 	nexthop_res_grp_activity_update(mlxsw_sp_net(mlxsw_sp), nh_grp->obj.id,
4513 					nh_grp->nhgi->count, activity);
4514 
4515 	bitmap_free(activity);
4516 }
4517 
4518 static void
4519 mlxsw_sp_nh_grp_activity_work_schedule(struct mlxsw_sp *mlxsw_sp)
4520 {
4521 	unsigned int interval = MLXSW_SP_NH_GRP_ACTIVITY_UPDATE_INTERVAL;
4522 
4523 	mlxsw_core_schedule_dw(&mlxsw_sp->router->nh_grp_activity_dw,
4524 			       msecs_to_jiffies(interval));
4525 }
4526 
4527 static void mlxsw_sp_nh_grp_activity_work(struct work_struct *work)
4528 {
4529 	struct mlxsw_sp_nexthop_group_info *nhgi;
4530 	struct mlxsw_sp_router *router;
4531 	bool reschedule = false;
4532 
4533 	router = container_of(work, struct mlxsw_sp_router,
4534 			      nh_grp_activity_dw.work);
4535 
4536 	mutex_lock(&router->lock);
4537 
4538 	list_for_each_entry(nhgi, &router->nh_res_grp_list, list) {
4539 		mlxsw_sp_nh_grp_activity_update(router->mlxsw_sp, nhgi->nh_grp);
4540 		reschedule = true;
4541 	}
4542 
4543 	mutex_unlock(&router->lock);
4544 
4545 	if (!reschedule)
4546 		return;
4547 	mlxsw_sp_nh_grp_activity_work_schedule(router->mlxsw_sp);
4548 }
4549 
4550 static int
4551 mlxsw_sp_nexthop_obj_single_validate(struct mlxsw_sp *mlxsw_sp,
4552 				     const struct nh_notifier_single_info *nh,
4553 				     struct netlink_ext_ack *extack)
4554 {
4555 	int err = -EINVAL;
4556 
4557 	if (nh->is_fdb)
4558 		NL_SET_ERR_MSG_MOD(extack, "FDB nexthops are not supported");
4559 	else if (nh->has_encap)
4560 		NL_SET_ERR_MSG_MOD(extack, "Encapsulating nexthops are not supported");
4561 	else
4562 		err = 0;
4563 
4564 	return err;
4565 }
4566 
4567 static int
4568 mlxsw_sp_nexthop_obj_group_entry_validate(struct mlxsw_sp *mlxsw_sp,
4569 					  const struct nh_notifier_single_info *nh,
4570 					  struct netlink_ext_ack *extack)
4571 {
4572 	int err;
4573 
4574 	err = mlxsw_sp_nexthop_obj_single_validate(mlxsw_sp, nh, extack);
4575 	if (err)
4576 		return err;
4577 
4578 	/* Device only nexthops with an IPIP device are programmed as
4579 	 * encapsulating adjacency entries.
4580 	 */
4581 	if (!nh->gw_family && !nh->is_reject &&
4582 	    !mlxsw_sp_netdev_ipip_type(mlxsw_sp, nh->dev, NULL)) {
4583 		NL_SET_ERR_MSG_MOD(extack, "Nexthop group entry does not have a gateway");
4584 		return -EINVAL;
4585 	}
4586 
4587 	return 0;
4588 }
4589 
4590 static int
4591 mlxsw_sp_nexthop_obj_group_validate(struct mlxsw_sp *mlxsw_sp,
4592 				    const struct nh_notifier_grp_info *nh_grp,
4593 				    struct netlink_ext_ack *extack)
4594 {
4595 	int i;
4596 
4597 	if (nh_grp->is_fdb) {
4598 		NL_SET_ERR_MSG_MOD(extack, "FDB nexthop groups are not supported");
4599 		return -EINVAL;
4600 	}
4601 
4602 	for (i = 0; i < nh_grp->num_nh; i++) {
4603 		const struct nh_notifier_single_info *nh;
4604 		int err;
4605 
4606 		nh = &nh_grp->nh_entries[i].nh;
4607 		err = mlxsw_sp_nexthop_obj_group_entry_validate(mlxsw_sp, nh,
4608 								extack);
4609 		if (err)
4610 			return err;
4611 	}
4612 
4613 	return 0;
4614 }
4615 
4616 static int
4617 mlxsw_sp_nexthop_obj_res_group_size_validate(struct mlxsw_sp *mlxsw_sp,
4618 					     const struct nh_notifier_res_table_info *nh_res_table,
4619 					     struct netlink_ext_ack *extack)
4620 {
4621 	unsigned int alloc_size;
4622 	bool valid_size = false;
4623 	int err, i;
4624 
4625 	if (nh_res_table->num_nh_buckets < 32) {
4626 		NL_SET_ERR_MSG_MOD(extack, "Minimum number of buckets is 32");
4627 		return -EINVAL;
4628 	}
4629 
4630 	for (i = 0; i < mlxsw_sp->router->adj_grp_size_ranges_count; i++) {
4631 		const struct mlxsw_sp_adj_grp_size_range *size_range;
4632 
4633 		size_range = &mlxsw_sp->router->adj_grp_size_ranges[i];
4634 
4635 		if (nh_res_table->num_nh_buckets >= size_range->start &&
4636 		    nh_res_table->num_nh_buckets <= size_range->end) {
4637 			valid_size = true;
4638 			break;
4639 		}
4640 	}
4641 
4642 	if (!valid_size) {
4643 		NL_SET_ERR_MSG_MOD(extack, "Invalid number of buckets");
4644 		return -EINVAL;
4645 	}
4646 
4647 	err = mlxsw_sp_kvdl_alloc_count_query(mlxsw_sp,
4648 					      MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
4649 					      nh_res_table->num_nh_buckets,
4650 					      &alloc_size);
4651 	if (err || nh_res_table->num_nh_buckets != alloc_size) {
4652 		NL_SET_ERR_MSG_MOD(extack, "Number of buckets does not fit allocation size of any KVDL partition");
4653 		return -EINVAL;
4654 	}
4655 
4656 	return 0;
4657 }
4658 
4659 static int
4660 mlxsw_sp_nexthop_obj_res_group_validate(struct mlxsw_sp *mlxsw_sp,
4661 					const struct nh_notifier_res_table_info *nh_res_table,
4662 					struct netlink_ext_ack *extack)
4663 {
4664 	int err;
4665 	u16 i;
4666 
4667 	err = mlxsw_sp_nexthop_obj_res_group_size_validate(mlxsw_sp,
4668 							   nh_res_table,
4669 							   extack);
4670 	if (err)
4671 		return err;
4672 
4673 	for (i = 0; i < nh_res_table->num_nh_buckets; i++) {
4674 		const struct nh_notifier_single_info *nh;
4675 		int err;
4676 
4677 		nh = &nh_res_table->nhs[i];
4678 		err = mlxsw_sp_nexthop_obj_group_entry_validate(mlxsw_sp, nh,
4679 								extack);
4680 		if (err)
4681 			return err;
4682 	}
4683 
4684 	return 0;
4685 }
4686 
4687 static int mlxsw_sp_nexthop_obj_validate(struct mlxsw_sp *mlxsw_sp,
4688 					 unsigned long event,
4689 					 struct nh_notifier_info *info)
4690 {
4691 	struct nh_notifier_single_info *nh;
4692 
4693 	if (event != NEXTHOP_EVENT_REPLACE &&
4694 	    event != NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE &&
4695 	    event != NEXTHOP_EVENT_BUCKET_REPLACE)
4696 		return 0;
4697 
4698 	switch (info->type) {
4699 	case NH_NOTIFIER_INFO_TYPE_SINGLE:
4700 		return mlxsw_sp_nexthop_obj_single_validate(mlxsw_sp, info->nh,
4701 							    info->extack);
4702 	case NH_NOTIFIER_INFO_TYPE_GRP:
4703 		return mlxsw_sp_nexthop_obj_group_validate(mlxsw_sp,
4704 							   info->nh_grp,
4705 							   info->extack);
4706 	case NH_NOTIFIER_INFO_TYPE_RES_TABLE:
4707 		return mlxsw_sp_nexthop_obj_res_group_validate(mlxsw_sp,
4708 							       info->nh_res_table,
4709 							       info->extack);
4710 	case NH_NOTIFIER_INFO_TYPE_RES_BUCKET:
4711 		nh = &info->nh_res_bucket->new_nh;
4712 		return mlxsw_sp_nexthop_obj_group_entry_validate(mlxsw_sp, nh,
4713 								 info->extack);
4714 	default:
4715 		NL_SET_ERR_MSG_MOD(info->extack, "Unsupported nexthop type");
4716 		return -EOPNOTSUPP;
4717 	}
4718 }
4719 
4720 static bool mlxsw_sp_nexthop_obj_is_gateway(struct mlxsw_sp *mlxsw_sp,
4721 					    const struct nh_notifier_info *info)
4722 {
4723 	const struct net_device *dev;
4724 
4725 	switch (info->type) {
4726 	case NH_NOTIFIER_INFO_TYPE_SINGLE:
4727 		dev = info->nh->dev;
4728 		return info->nh->gw_family || info->nh->is_reject ||
4729 		       mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL);
4730 	case NH_NOTIFIER_INFO_TYPE_GRP:
4731 	case NH_NOTIFIER_INFO_TYPE_RES_TABLE:
4732 		/* Already validated earlier. */
4733 		return true;
4734 	default:
4735 		return false;
4736 	}
4737 }
4738 
4739 static void mlxsw_sp_nexthop_obj_blackhole_init(struct mlxsw_sp *mlxsw_sp,
4740 						struct mlxsw_sp_nexthop *nh)
4741 {
4742 	u16 lb_rif_index = mlxsw_sp->router->lb_rif_index;
4743 
4744 	nh->action = MLXSW_SP_NEXTHOP_ACTION_DISCARD;
4745 	nh->should_offload = 1;
4746 	/* While nexthops that discard packets do not forward packets
4747 	 * via an egress RIF, they still need to be programmed using a
4748 	 * valid RIF, so use the loopback RIF created during init.
4749 	 */
4750 	nh->rif = mlxsw_sp->router->rifs[lb_rif_index];
4751 }
4752 
4753 static void mlxsw_sp_nexthop_obj_blackhole_fini(struct mlxsw_sp *mlxsw_sp,
4754 						struct mlxsw_sp_nexthop *nh)
4755 {
4756 	nh->rif = NULL;
4757 	nh->should_offload = 0;
4758 }
4759 
4760 static int
4761 mlxsw_sp_nexthop_obj_init(struct mlxsw_sp *mlxsw_sp,
4762 			  struct mlxsw_sp_nexthop_group *nh_grp,
4763 			  struct mlxsw_sp_nexthop *nh,
4764 			  struct nh_notifier_single_info *nh_obj, int weight)
4765 {
4766 	struct net_device *dev = nh_obj->dev;
4767 	int err;
4768 
4769 	nh->nhgi = nh_grp->nhgi;
4770 	nh->nh_weight = weight;
4771 
4772 	switch (nh_obj->gw_family) {
4773 	case AF_INET:
4774 		memcpy(&nh->gw_addr, &nh_obj->ipv4, sizeof(nh_obj->ipv4));
4775 		nh->neigh_tbl = &arp_tbl;
4776 		break;
4777 	case AF_INET6:
4778 		memcpy(&nh->gw_addr, &nh_obj->ipv6, sizeof(nh_obj->ipv6));
4779 #if IS_ENABLED(CONFIG_IPV6)
4780 		nh->neigh_tbl = &nd_tbl;
4781 #endif
4782 		break;
4783 	}
4784 
4785 	mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
4786 	list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
4787 	nh->ifindex = dev->ifindex;
4788 
4789 	err = mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, dev);
4790 	if (err)
4791 		goto err_type_init;
4792 
4793 	if (nh_obj->is_reject)
4794 		mlxsw_sp_nexthop_obj_blackhole_init(mlxsw_sp, nh);
4795 
4796 	/* In a resilient nexthop group, all the nexthops must be written to
4797 	 * the adjacency table. Even if they do not have a valid neighbour or
4798 	 * RIF.
4799 	 */
4800 	if (nh_grp->nhgi->is_resilient && !nh->should_offload) {
4801 		nh->action = MLXSW_SP_NEXTHOP_ACTION_TRAP;
4802 		nh->should_offload = 1;
4803 	}
4804 
4805 	return 0;
4806 
4807 err_type_init:
4808 	list_del(&nh->router_list_node);
4809 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
4810 	return err;
4811 }
4812 
4813 static void mlxsw_sp_nexthop_obj_fini(struct mlxsw_sp *mlxsw_sp,
4814 				      struct mlxsw_sp_nexthop *nh)
4815 {
4816 	if (nh->action == MLXSW_SP_NEXTHOP_ACTION_DISCARD)
4817 		mlxsw_sp_nexthop_obj_blackhole_fini(mlxsw_sp, nh);
4818 	mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4819 	list_del(&nh->router_list_node);
4820 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
4821 	nh->should_offload = 0;
4822 }
4823 
4824 static int
4825 mlxsw_sp_nexthop_obj_group_info_init(struct mlxsw_sp *mlxsw_sp,
4826 				     struct mlxsw_sp_nexthop_group *nh_grp,
4827 				     struct nh_notifier_info *info)
4828 {
4829 	struct mlxsw_sp_nexthop_group_info *nhgi;
4830 	struct mlxsw_sp_nexthop *nh;
4831 	bool is_resilient = false;
4832 	unsigned int nhs;
4833 	int err, i;
4834 
4835 	switch (info->type) {
4836 	case NH_NOTIFIER_INFO_TYPE_SINGLE:
4837 		nhs = 1;
4838 		break;
4839 	case NH_NOTIFIER_INFO_TYPE_GRP:
4840 		nhs = info->nh_grp->num_nh;
4841 		break;
4842 	case NH_NOTIFIER_INFO_TYPE_RES_TABLE:
4843 		nhs = info->nh_res_table->num_nh_buckets;
4844 		is_resilient = true;
4845 		break;
4846 	default:
4847 		return -EINVAL;
4848 	}
4849 
4850 	nhgi = kzalloc(struct_size(nhgi, nexthops, nhs), GFP_KERNEL);
4851 	if (!nhgi)
4852 		return -ENOMEM;
4853 	nh_grp->nhgi = nhgi;
4854 	nhgi->nh_grp = nh_grp;
4855 	nhgi->gateway = mlxsw_sp_nexthop_obj_is_gateway(mlxsw_sp, info);
4856 	nhgi->is_resilient = is_resilient;
4857 	nhgi->count = nhs;
4858 	for (i = 0; i < nhgi->count; i++) {
4859 		struct nh_notifier_single_info *nh_obj;
4860 		int weight;
4861 
4862 		nh = &nhgi->nexthops[i];
4863 		switch (info->type) {
4864 		case NH_NOTIFIER_INFO_TYPE_SINGLE:
4865 			nh_obj = info->nh;
4866 			weight = 1;
4867 			break;
4868 		case NH_NOTIFIER_INFO_TYPE_GRP:
4869 			nh_obj = &info->nh_grp->nh_entries[i].nh;
4870 			weight = info->nh_grp->nh_entries[i].weight;
4871 			break;
4872 		case NH_NOTIFIER_INFO_TYPE_RES_TABLE:
4873 			nh_obj = &info->nh_res_table->nhs[i];
4874 			weight = 1;
4875 			break;
4876 		default:
4877 			err = -EINVAL;
4878 			goto err_nexthop_obj_init;
4879 		}
4880 		err = mlxsw_sp_nexthop_obj_init(mlxsw_sp, nh_grp, nh, nh_obj,
4881 						weight);
4882 		if (err)
4883 			goto err_nexthop_obj_init;
4884 	}
4885 	err = mlxsw_sp_nexthop_group_inc(mlxsw_sp);
4886 	if (err)
4887 		goto err_group_inc;
4888 	err = mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4889 	if (err) {
4890 		NL_SET_ERR_MSG_MOD(info->extack, "Failed to write adjacency entries to the device");
4891 		goto err_group_refresh;
4892 	}
4893 
4894 	/* Add resilient nexthop groups to a list so that the activity of their
4895 	 * nexthop buckets will be periodically queried and cleared.
4896 	 */
4897 	if (nhgi->is_resilient) {
4898 		if (list_empty(&mlxsw_sp->router->nh_res_grp_list))
4899 			mlxsw_sp_nh_grp_activity_work_schedule(mlxsw_sp);
4900 		list_add(&nhgi->list, &mlxsw_sp->router->nh_res_grp_list);
4901 	}
4902 
4903 	return 0;
4904 
4905 err_group_refresh:
4906 	mlxsw_sp_nexthop_group_dec(mlxsw_sp);
4907 err_group_inc:
4908 	i = nhgi->count;
4909 err_nexthop_obj_init:
4910 	for (i--; i >= 0; i--) {
4911 		nh = &nhgi->nexthops[i];
4912 		mlxsw_sp_nexthop_obj_fini(mlxsw_sp, nh);
4913 	}
4914 	kfree(nhgi);
4915 	return err;
4916 }
4917 
4918 static void
4919 mlxsw_sp_nexthop_obj_group_info_fini(struct mlxsw_sp *mlxsw_sp,
4920 				     struct mlxsw_sp_nexthop_group *nh_grp)
4921 {
4922 	struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
4923 	struct mlxsw_sp_router *router = mlxsw_sp->router;
4924 	int i;
4925 
4926 	if (nhgi->is_resilient) {
4927 		list_del(&nhgi->list);
4928 		if (list_empty(&mlxsw_sp->router->nh_res_grp_list))
4929 			cancel_delayed_work(&router->nh_grp_activity_dw);
4930 	}
4931 
4932 	mlxsw_sp_nexthop_group_dec(mlxsw_sp);
4933 	for (i = nhgi->count - 1; i >= 0; i--) {
4934 		struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
4935 
4936 		mlxsw_sp_nexthop_obj_fini(mlxsw_sp, nh);
4937 	}
4938 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4939 	WARN_ON_ONCE(nhgi->adj_index_valid);
4940 	kfree(nhgi);
4941 }
4942 
4943 static struct mlxsw_sp_nexthop_group *
4944 mlxsw_sp_nexthop_obj_group_create(struct mlxsw_sp *mlxsw_sp,
4945 				  struct nh_notifier_info *info)
4946 {
4947 	struct mlxsw_sp_nexthop_group *nh_grp;
4948 	int err;
4949 
4950 	nh_grp = kzalloc(sizeof(*nh_grp), GFP_KERNEL);
4951 	if (!nh_grp)
4952 		return ERR_PTR(-ENOMEM);
4953 	INIT_LIST_HEAD(&nh_grp->vr_list);
4954 	err = rhashtable_init(&nh_grp->vr_ht,
4955 			      &mlxsw_sp_nexthop_group_vr_ht_params);
4956 	if (err)
4957 		goto err_nexthop_group_vr_ht_init;
4958 	INIT_LIST_HEAD(&nh_grp->fib_list);
4959 	nh_grp->type = MLXSW_SP_NEXTHOP_GROUP_TYPE_OBJ;
4960 	nh_grp->obj.id = info->id;
4961 
4962 	err = mlxsw_sp_nexthop_obj_group_info_init(mlxsw_sp, nh_grp, info);
4963 	if (err)
4964 		goto err_nexthop_group_info_init;
4965 
4966 	nh_grp->can_destroy = false;
4967 
4968 	return nh_grp;
4969 
4970 err_nexthop_group_info_init:
4971 	rhashtable_destroy(&nh_grp->vr_ht);
4972 err_nexthop_group_vr_ht_init:
4973 	kfree(nh_grp);
4974 	return ERR_PTR(err);
4975 }
4976 
4977 static void
4978 mlxsw_sp_nexthop_obj_group_destroy(struct mlxsw_sp *mlxsw_sp,
4979 				   struct mlxsw_sp_nexthop_group *nh_grp)
4980 {
4981 	if (!nh_grp->can_destroy)
4982 		return;
4983 	mlxsw_sp_nexthop_obj_group_info_fini(mlxsw_sp, nh_grp);
4984 	WARN_ON_ONCE(!list_empty(&nh_grp->fib_list));
4985 	WARN_ON_ONCE(!list_empty(&nh_grp->vr_list));
4986 	rhashtable_destroy(&nh_grp->vr_ht);
4987 	kfree(nh_grp);
4988 }
4989 
4990 static struct mlxsw_sp_nexthop_group *
4991 mlxsw_sp_nexthop_obj_group_lookup(struct mlxsw_sp *mlxsw_sp, u32 id)
4992 {
4993 	struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
4994 
4995 	cmp_arg.type = MLXSW_SP_NEXTHOP_GROUP_TYPE_OBJ;
4996 	cmp_arg.id = id;
4997 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
4998 				      &cmp_arg,
4999 				      mlxsw_sp_nexthop_group_ht_params);
5000 }
5001 
5002 static int mlxsw_sp_nexthop_obj_group_add(struct mlxsw_sp *mlxsw_sp,
5003 					  struct mlxsw_sp_nexthop_group *nh_grp)
5004 {
5005 	return mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
5006 }
5007 
5008 static int
5009 mlxsw_sp_nexthop_obj_group_replace(struct mlxsw_sp *mlxsw_sp,
5010 				   struct mlxsw_sp_nexthop_group *nh_grp,
5011 				   struct mlxsw_sp_nexthop_group *old_nh_grp,
5012 				   struct netlink_ext_ack *extack)
5013 {
5014 	struct mlxsw_sp_nexthop_group_info *old_nhgi = old_nh_grp->nhgi;
5015 	struct mlxsw_sp_nexthop_group_info *new_nhgi = nh_grp->nhgi;
5016 	int err;
5017 
5018 	old_nh_grp->nhgi = new_nhgi;
5019 	new_nhgi->nh_grp = old_nh_grp;
5020 	nh_grp->nhgi = old_nhgi;
5021 	old_nhgi->nh_grp = nh_grp;
5022 
5023 	if (old_nhgi->adj_index_valid && new_nhgi->adj_index_valid) {
5024 		/* Both the old adjacency index and the new one are valid.
5025 		 * Routes are currently using the old one. Tell the device to
5026 		 * replace the old adjacency index with the new one.
5027 		 */
5028 		err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, old_nh_grp,
5029 						     old_nhgi->adj_index,
5030 						     old_nhgi->ecmp_size);
5031 		if (err) {
5032 			NL_SET_ERR_MSG_MOD(extack, "Failed to replace old adjacency index with new one");
5033 			goto err_out;
5034 		}
5035 	} else if (old_nhgi->adj_index_valid && !new_nhgi->adj_index_valid) {
5036 		/* The old adjacency index is valid, while the new one is not.
5037 		 * Iterate over all the routes using the group and change them
5038 		 * to trap packets to the CPU.
5039 		 */
5040 		err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, old_nh_grp);
5041 		if (err) {
5042 			NL_SET_ERR_MSG_MOD(extack, "Failed to update routes to trap packets");
5043 			goto err_out;
5044 		}
5045 	} else if (!old_nhgi->adj_index_valid && new_nhgi->adj_index_valid) {
5046 		/* The old adjacency index is invalid, while the new one is.
5047 		 * Iterate over all the routes using the group and change them
5048 		 * to forward packets using the new valid index.
5049 		 */
5050 		err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, old_nh_grp);
5051 		if (err) {
5052 			NL_SET_ERR_MSG_MOD(extack, "Failed to update routes to forward packets");
5053 			goto err_out;
5054 		}
5055 	}
5056 
5057 	/* Make sure the flags are set / cleared based on the new nexthop group
5058 	 * information.
5059 	 */
5060 	mlxsw_sp_nexthop_obj_group_offload_refresh(mlxsw_sp, old_nh_grp);
5061 
5062 	/* At this point 'nh_grp' is just a shell that is not used by anyone
5063 	 * and its nexthop group info is the old info that was just replaced
5064 	 * with the new one. Remove it.
5065 	 */
5066 	nh_grp->can_destroy = true;
5067 	mlxsw_sp_nexthop_obj_group_destroy(mlxsw_sp, nh_grp);
5068 
5069 	return 0;
5070 
5071 err_out:
5072 	old_nhgi->nh_grp = old_nh_grp;
5073 	nh_grp->nhgi = new_nhgi;
5074 	new_nhgi->nh_grp = nh_grp;
5075 	old_nh_grp->nhgi = old_nhgi;
5076 	return err;
5077 }
5078 
5079 static int mlxsw_sp_nexthop_obj_new(struct mlxsw_sp *mlxsw_sp,
5080 				    struct nh_notifier_info *info)
5081 {
5082 	struct mlxsw_sp_nexthop_group *nh_grp, *old_nh_grp;
5083 	struct netlink_ext_ack *extack = info->extack;
5084 	int err;
5085 
5086 	nh_grp = mlxsw_sp_nexthop_obj_group_create(mlxsw_sp, info);
5087 	if (IS_ERR(nh_grp))
5088 		return PTR_ERR(nh_grp);
5089 
5090 	old_nh_grp = mlxsw_sp_nexthop_obj_group_lookup(mlxsw_sp, info->id);
5091 	if (!old_nh_grp)
5092 		err = mlxsw_sp_nexthop_obj_group_add(mlxsw_sp, nh_grp);
5093 	else
5094 		err = mlxsw_sp_nexthop_obj_group_replace(mlxsw_sp, nh_grp,
5095 							 old_nh_grp, extack);
5096 
5097 	if (err) {
5098 		nh_grp->can_destroy = true;
5099 		mlxsw_sp_nexthop_obj_group_destroy(mlxsw_sp, nh_grp);
5100 	}
5101 
5102 	return err;
5103 }
5104 
5105 static void mlxsw_sp_nexthop_obj_del(struct mlxsw_sp *mlxsw_sp,
5106 				     struct nh_notifier_info *info)
5107 {
5108 	struct mlxsw_sp_nexthop_group *nh_grp;
5109 
5110 	nh_grp = mlxsw_sp_nexthop_obj_group_lookup(mlxsw_sp, info->id);
5111 	if (!nh_grp)
5112 		return;
5113 
5114 	nh_grp->can_destroy = true;
5115 	mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
5116 
5117 	/* If the group still has routes using it, then defer the delete
5118 	 * operation until the last route using it is deleted.
5119 	 */
5120 	if (!list_empty(&nh_grp->fib_list))
5121 		return;
5122 	mlxsw_sp_nexthop_obj_group_destroy(mlxsw_sp, nh_grp);
5123 }
5124 
5125 static int mlxsw_sp_nexthop_obj_bucket_query(struct mlxsw_sp *mlxsw_sp,
5126 					     u32 adj_index, char *ratr_pl)
5127 {
5128 	MLXSW_REG_ZERO(ratr, ratr_pl);
5129 	mlxsw_reg_ratr_op_set(ratr_pl, MLXSW_REG_RATR_OP_QUERY_READ);
5130 	mlxsw_reg_ratr_adjacency_index_low_set(ratr_pl, adj_index);
5131 	mlxsw_reg_ratr_adjacency_index_high_set(ratr_pl, adj_index >> 16);
5132 
5133 	return mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
5134 }
5135 
5136 static int mlxsw_sp_nexthop_obj_bucket_compare(char *ratr_pl, char *ratr_pl_new)
5137 {
5138 	/* Clear the opcode and activity on both the old and new payload as
5139 	 * they are irrelevant for the comparison.
5140 	 */
5141 	mlxsw_reg_ratr_op_set(ratr_pl, MLXSW_REG_RATR_OP_QUERY_READ);
5142 	mlxsw_reg_ratr_a_set(ratr_pl, 0);
5143 	mlxsw_reg_ratr_op_set(ratr_pl_new, MLXSW_REG_RATR_OP_QUERY_READ);
5144 	mlxsw_reg_ratr_a_set(ratr_pl_new, 0);
5145 
5146 	/* If the contents of the adjacency entry are consistent with the
5147 	 * replacement request, then replacement was successful.
5148 	 */
5149 	if (!memcmp(ratr_pl, ratr_pl_new, MLXSW_REG_RATR_LEN))
5150 		return 0;
5151 
5152 	return -EINVAL;
5153 }
5154 
5155 static int
5156 mlxsw_sp_nexthop_obj_bucket_adj_update(struct mlxsw_sp *mlxsw_sp,
5157 				       struct mlxsw_sp_nexthop *nh,
5158 				       struct nh_notifier_info *info)
5159 {
5160 	u16 bucket_index = info->nh_res_bucket->bucket_index;
5161 	struct netlink_ext_ack *extack = info->extack;
5162 	bool force = info->nh_res_bucket->force;
5163 	char ratr_pl_new[MLXSW_REG_RATR_LEN];
5164 	char ratr_pl[MLXSW_REG_RATR_LEN];
5165 	u32 adj_index;
5166 	int err;
5167 
5168 	/* No point in trying an atomic replacement if the idle timer interval
5169 	 * is smaller than the interval in which we query and clear activity.
5170 	 */
5171 	if (!force && info->nh_res_bucket->idle_timer_ms <
5172 	    MLXSW_SP_NH_GRP_ACTIVITY_UPDATE_INTERVAL)
5173 		force = true;
5174 
5175 	adj_index = nh->nhgi->adj_index + bucket_index;
5176 	err = mlxsw_sp_nexthop_update(mlxsw_sp, adj_index, nh, force, ratr_pl);
5177 	if (err) {
5178 		NL_SET_ERR_MSG_MOD(extack, "Failed to overwrite nexthop bucket");
5179 		return err;
5180 	}
5181 
5182 	if (!force) {
5183 		err = mlxsw_sp_nexthop_obj_bucket_query(mlxsw_sp, adj_index,
5184 							ratr_pl_new);
5185 		if (err) {
5186 			NL_SET_ERR_MSG_MOD(extack, "Failed to query nexthop bucket state after replacement. State might be inconsistent");
5187 			return err;
5188 		}
5189 
5190 		err = mlxsw_sp_nexthop_obj_bucket_compare(ratr_pl, ratr_pl_new);
5191 		if (err) {
5192 			NL_SET_ERR_MSG_MOD(extack, "Nexthop bucket was not replaced because it was active during replacement");
5193 			return err;
5194 		}
5195 	}
5196 
5197 	nh->update = 0;
5198 	nh->offloaded = 1;
5199 	mlxsw_sp_nexthop_bucket_offload_refresh(mlxsw_sp, nh, bucket_index);
5200 
5201 	return 0;
5202 }
5203 
5204 static int mlxsw_sp_nexthop_obj_bucket_replace(struct mlxsw_sp *mlxsw_sp,
5205 					       struct nh_notifier_info *info)
5206 {
5207 	u16 bucket_index = info->nh_res_bucket->bucket_index;
5208 	struct netlink_ext_ack *extack = info->extack;
5209 	struct mlxsw_sp_nexthop_group_info *nhgi;
5210 	struct nh_notifier_single_info *nh_obj;
5211 	struct mlxsw_sp_nexthop_group *nh_grp;
5212 	struct mlxsw_sp_nexthop *nh;
5213 	int err;
5214 
5215 	nh_grp = mlxsw_sp_nexthop_obj_group_lookup(mlxsw_sp, info->id);
5216 	if (!nh_grp) {
5217 		NL_SET_ERR_MSG_MOD(extack, "Nexthop group was not found");
5218 		return -EINVAL;
5219 	}
5220 
5221 	nhgi = nh_grp->nhgi;
5222 
5223 	if (bucket_index >= nhgi->count) {
5224 		NL_SET_ERR_MSG_MOD(extack, "Nexthop bucket index out of range");
5225 		return -EINVAL;
5226 	}
5227 
5228 	nh = &nhgi->nexthops[bucket_index];
5229 	mlxsw_sp_nexthop_obj_fini(mlxsw_sp, nh);
5230 
5231 	nh_obj = &info->nh_res_bucket->new_nh;
5232 	err = mlxsw_sp_nexthop_obj_init(mlxsw_sp, nh_grp, nh, nh_obj, 1);
5233 	if (err) {
5234 		NL_SET_ERR_MSG_MOD(extack, "Failed to initialize nexthop object for nexthop bucket replacement");
5235 		goto err_nexthop_obj_init;
5236 	}
5237 
5238 	err = mlxsw_sp_nexthop_obj_bucket_adj_update(mlxsw_sp, nh, info);
5239 	if (err)
5240 		goto err_nexthop_obj_bucket_adj_update;
5241 
5242 	return 0;
5243 
5244 err_nexthop_obj_bucket_adj_update:
5245 	mlxsw_sp_nexthop_obj_fini(mlxsw_sp, nh);
5246 err_nexthop_obj_init:
5247 	nh_obj = &info->nh_res_bucket->old_nh;
5248 	mlxsw_sp_nexthop_obj_init(mlxsw_sp, nh_grp, nh, nh_obj, 1);
5249 	/* The old adjacency entry was not overwritten */
5250 	nh->update = 0;
5251 	nh->offloaded = 1;
5252 	return err;
5253 }
5254 
5255 static int mlxsw_sp_nexthop_obj_event(struct notifier_block *nb,
5256 				      unsigned long event, void *ptr)
5257 {
5258 	struct nh_notifier_info *info = ptr;
5259 	struct mlxsw_sp_router *router;
5260 	int err = 0;
5261 
5262 	router = container_of(nb, struct mlxsw_sp_router, nexthop_nb);
5263 	err = mlxsw_sp_nexthop_obj_validate(router->mlxsw_sp, event, info);
5264 	if (err)
5265 		goto out;
5266 
5267 	mutex_lock(&router->lock);
5268 
5269 	switch (event) {
5270 	case NEXTHOP_EVENT_REPLACE:
5271 		err = mlxsw_sp_nexthop_obj_new(router->mlxsw_sp, info);
5272 		break;
5273 	case NEXTHOP_EVENT_DEL:
5274 		mlxsw_sp_nexthop_obj_del(router->mlxsw_sp, info);
5275 		break;
5276 	case NEXTHOP_EVENT_BUCKET_REPLACE:
5277 		err = mlxsw_sp_nexthop_obj_bucket_replace(router->mlxsw_sp,
5278 							  info);
5279 		break;
5280 	default:
5281 		break;
5282 	}
5283 
5284 	mutex_unlock(&router->lock);
5285 
5286 out:
5287 	return notifier_from_errno(err);
5288 }
5289 
5290 static bool mlxsw_sp_fi_is_gateway(const struct mlxsw_sp *mlxsw_sp,
5291 				   struct fib_info *fi)
5292 {
5293 	const struct fib_nh *nh = fib_info_nh(fi, 0);
5294 
5295 	return nh->fib_nh_scope == RT_SCOPE_LINK ||
5296 	       mlxsw_sp_nexthop4_ipip_type(mlxsw_sp, nh, NULL);
5297 }
5298 
5299 static int
5300 mlxsw_sp_nexthop4_group_info_init(struct mlxsw_sp *mlxsw_sp,
5301 				  struct mlxsw_sp_nexthop_group *nh_grp)
5302 {
5303 	unsigned int nhs = fib_info_num_path(nh_grp->ipv4.fi);
5304 	struct mlxsw_sp_nexthop_group_info *nhgi;
5305 	struct mlxsw_sp_nexthop *nh;
5306 	int err, i;
5307 
5308 	nhgi = kzalloc(struct_size(nhgi, nexthops, nhs), GFP_KERNEL);
5309 	if (!nhgi)
5310 		return -ENOMEM;
5311 	nh_grp->nhgi = nhgi;
5312 	nhgi->nh_grp = nh_grp;
5313 	nhgi->gateway = mlxsw_sp_fi_is_gateway(mlxsw_sp, nh_grp->ipv4.fi);
5314 	nhgi->count = nhs;
5315 	for (i = 0; i < nhgi->count; i++) {
5316 		struct fib_nh *fib_nh;
5317 
5318 		nh = &nhgi->nexthops[i];
5319 		fib_nh = fib_info_nh(nh_grp->ipv4.fi, i);
5320 		err = mlxsw_sp_nexthop4_init(mlxsw_sp, nh_grp, nh, fib_nh);
5321 		if (err)
5322 			goto err_nexthop4_init;
5323 	}
5324 	err = mlxsw_sp_nexthop_group_inc(mlxsw_sp);
5325 	if (err)
5326 		goto err_group_inc;
5327 	err = mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
5328 	if (err)
5329 		goto err_group_refresh;
5330 
5331 	return 0;
5332 
5333 err_group_refresh:
5334 	mlxsw_sp_nexthop_group_dec(mlxsw_sp);
5335 err_group_inc:
5336 	i = nhgi->count;
5337 err_nexthop4_init:
5338 	for (i--; i >= 0; i--) {
5339 		nh = &nhgi->nexthops[i];
5340 		mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
5341 	}
5342 	kfree(nhgi);
5343 	return err;
5344 }
5345 
5346 static void
5347 mlxsw_sp_nexthop4_group_info_fini(struct mlxsw_sp *mlxsw_sp,
5348 				  struct mlxsw_sp_nexthop_group *nh_grp)
5349 {
5350 	struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
5351 	int i;
5352 
5353 	mlxsw_sp_nexthop_group_dec(mlxsw_sp);
5354 	for (i = nhgi->count - 1; i >= 0; i--) {
5355 		struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
5356 
5357 		mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
5358 	}
5359 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
5360 	WARN_ON_ONCE(nhgi->adj_index_valid);
5361 	kfree(nhgi);
5362 }
5363 
5364 static struct mlxsw_sp_nexthop_group *
5365 mlxsw_sp_nexthop4_group_create(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
5366 {
5367 	struct mlxsw_sp_nexthop_group *nh_grp;
5368 	int err;
5369 
5370 	nh_grp = kzalloc(sizeof(*nh_grp), GFP_KERNEL);
5371 	if (!nh_grp)
5372 		return ERR_PTR(-ENOMEM);
5373 	INIT_LIST_HEAD(&nh_grp->vr_list);
5374 	err = rhashtable_init(&nh_grp->vr_ht,
5375 			      &mlxsw_sp_nexthop_group_vr_ht_params);
5376 	if (err)
5377 		goto err_nexthop_group_vr_ht_init;
5378 	INIT_LIST_HEAD(&nh_grp->fib_list);
5379 	nh_grp->type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4;
5380 	nh_grp->ipv4.fi = fi;
5381 	fib_info_hold(fi);
5382 
5383 	err = mlxsw_sp_nexthop4_group_info_init(mlxsw_sp, nh_grp);
5384 	if (err)
5385 		goto err_nexthop_group_info_init;
5386 
5387 	err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
5388 	if (err)
5389 		goto err_nexthop_group_insert;
5390 
5391 	nh_grp->can_destroy = true;
5392 
5393 	return nh_grp;
5394 
5395 err_nexthop_group_insert:
5396 	mlxsw_sp_nexthop4_group_info_fini(mlxsw_sp, nh_grp);
5397 err_nexthop_group_info_init:
5398 	fib_info_put(fi);
5399 	rhashtable_destroy(&nh_grp->vr_ht);
5400 err_nexthop_group_vr_ht_init:
5401 	kfree(nh_grp);
5402 	return ERR_PTR(err);
5403 }
5404 
5405 static void
5406 mlxsw_sp_nexthop4_group_destroy(struct mlxsw_sp *mlxsw_sp,
5407 				struct mlxsw_sp_nexthop_group *nh_grp)
5408 {
5409 	if (!nh_grp->can_destroy)
5410 		return;
5411 	mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
5412 	mlxsw_sp_nexthop4_group_info_fini(mlxsw_sp, nh_grp);
5413 	fib_info_put(nh_grp->ipv4.fi);
5414 	WARN_ON_ONCE(!list_empty(&nh_grp->vr_list));
5415 	rhashtable_destroy(&nh_grp->vr_ht);
5416 	kfree(nh_grp);
5417 }
5418 
5419 static int mlxsw_sp_nexthop4_group_get(struct mlxsw_sp *mlxsw_sp,
5420 				       struct mlxsw_sp_fib_entry *fib_entry,
5421 				       struct fib_info *fi)
5422 {
5423 	struct mlxsw_sp_nexthop_group *nh_grp;
5424 
5425 	if (fi->nh) {
5426 		nh_grp = mlxsw_sp_nexthop_obj_group_lookup(mlxsw_sp,
5427 							   fi->nh->id);
5428 		if (WARN_ON_ONCE(!nh_grp))
5429 			return -EINVAL;
5430 		goto out;
5431 	}
5432 
5433 	nh_grp = mlxsw_sp_nexthop4_group_lookup(mlxsw_sp, fi);
5434 	if (!nh_grp) {
5435 		nh_grp = mlxsw_sp_nexthop4_group_create(mlxsw_sp, fi);
5436 		if (IS_ERR(nh_grp))
5437 			return PTR_ERR(nh_grp);
5438 	}
5439 out:
5440 	list_add_tail(&fib_entry->nexthop_group_node, &nh_grp->fib_list);
5441 	fib_entry->nh_group = nh_grp;
5442 	return 0;
5443 }
5444 
5445 static void mlxsw_sp_nexthop4_group_put(struct mlxsw_sp *mlxsw_sp,
5446 					struct mlxsw_sp_fib_entry *fib_entry)
5447 {
5448 	struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
5449 
5450 	list_del(&fib_entry->nexthop_group_node);
5451 	if (!list_empty(&nh_grp->fib_list))
5452 		return;
5453 
5454 	if (nh_grp->type == MLXSW_SP_NEXTHOP_GROUP_TYPE_OBJ) {
5455 		mlxsw_sp_nexthop_obj_group_destroy(mlxsw_sp, nh_grp);
5456 		return;
5457 	}
5458 
5459 	mlxsw_sp_nexthop4_group_destroy(mlxsw_sp, nh_grp);
5460 }
5461 
5462 static bool
5463 mlxsw_sp_fib4_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
5464 {
5465 	struct mlxsw_sp_fib4_entry *fib4_entry;
5466 
5467 	fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
5468 				  common);
5469 	return !fib4_entry->dscp;
5470 }
5471 
5472 static bool
5473 mlxsw_sp_fib_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
5474 {
5475 	struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
5476 
5477 	switch (fib_entry->fib_node->fib->proto) {
5478 	case MLXSW_SP_L3_PROTO_IPV4:
5479 		if (!mlxsw_sp_fib4_entry_should_offload(fib_entry))
5480 			return false;
5481 		break;
5482 	case MLXSW_SP_L3_PROTO_IPV6:
5483 		break;
5484 	}
5485 
5486 	switch (fib_entry->type) {
5487 	case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
5488 		return !!nh_group->nhgi->adj_index_valid;
5489 	case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
5490 		return !!nh_group->nhgi->nh_rif;
5491 	case MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE:
5492 	case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
5493 	case MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP:
5494 		return true;
5495 	default:
5496 		return false;
5497 	}
5498 }
5499 
5500 static struct mlxsw_sp_nexthop *
5501 mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
5502 		     const struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
5503 {
5504 	int i;
5505 
5506 	for (i = 0; i < nh_grp->nhgi->count; i++) {
5507 		struct mlxsw_sp_nexthop *nh = &nh_grp->nhgi->nexthops[i];
5508 		struct fib6_info *rt = mlxsw_sp_rt6->rt;
5509 
5510 		if (nh->rif && nh->rif->dev == rt->fib6_nh->fib_nh_dev &&
5511 		    ipv6_addr_equal((const struct in6_addr *) &nh->gw_addr,
5512 				    &rt->fib6_nh->fib_nh_gw6))
5513 			return nh;
5514 	}
5515 
5516 	return NULL;
5517 }
5518 
5519 static void
5520 mlxsw_sp_fib4_offload_failed_flag_set(struct mlxsw_sp *mlxsw_sp,
5521 				      struct fib_entry_notifier_info *fen_info)
5522 {
5523 	u32 *p_dst = (u32 *) &fen_info->dst;
5524 	struct fib_rt_info fri;
5525 
5526 	fri.fi = fen_info->fi;
5527 	fri.tb_id = fen_info->tb_id;
5528 	fri.dst = cpu_to_be32(*p_dst);
5529 	fri.dst_len = fen_info->dst_len;
5530 	fri.dscp = fen_info->dscp;
5531 	fri.type = fen_info->type;
5532 	fri.offload = false;
5533 	fri.trap = false;
5534 	fri.offload_failed = true;
5535 	fib_alias_hw_flags_set(mlxsw_sp_net(mlxsw_sp), &fri);
5536 }
5537 
5538 static void
5539 mlxsw_sp_fib4_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
5540 				 struct mlxsw_sp_fib_entry *fib_entry)
5541 {
5542 	u32 *p_dst = (u32 *) fib_entry->fib_node->key.addr;
5543 	int dst_len = fib_entry->fib_node->key.prefix_len;
5544 	struct mlxsw_sp_fib4_entry *fib4_entry;
5545 	struct fib_rt_info fri;
5546 	bool should_offload;
5547 
5548 	should_offload = mlxsw_sp_fib_entry_should_offload(fib_entry);
5549 	fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
5550 				  common);
5551 	fri.fi = fib4_entry->fi;
5552 	fri.tb_id = fib4_entry->tb_id;
5553 	fri.dst = cpu_to_be32(*p_dst);
5554 	fri.dst_len = dst_len;
5555 	fri.dscp = fib4_entry->dscp;
5556 	fri.type = fib4_entry->type;
5557 	fri.offload = should_offload;
5558 	fri.trap = !should_offload;
5559 	fri.offload_failed = false;
5560 	fib_alias_hw_flags_set(mlxsw_sp_net(mlxsw_sp), &fri);
5561 }
5562 
5563 static void
5564 mlxsw_sp_fib4_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
5565 				   struct mlxsw_sp_fib_entry *fib_entry)
5566 {
5567 	u32 *p_dst = (u32 *) fib_entry->fib_node->key.addr;
5568 	int dst_len = fib_entry->fib_node->key.prefix_len;
5569 	struct mlxsw_sp_fib4_entry *fib4_entry;
5570 	struct fib_rt_info fri;
5571 
5572 	fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
5573 				  common);
5574 	fri.fi = fib4_entry->fi;
5575 	fri.tb_id = fib4_entry->tb_id;
5576 	fri.dst = cpu_to_be32(*p_dst);
5577 	fri.dst_len = dst_len;
5578 	fri.dscp = fib4_entry->dscp;
5579 	fri.type = fib4_entry->type;
5580 	fri.offload = false;
5581 	fri.trap = false;
5582 	fri.offload_failed = false;
5583 	fib_alias_hw_flags_set(mlxsw_sp_net(mlxsw_sp), &fri);
5584 }
5585 
5586 #if IS_ENABLED(CONFIG_IPV6)
5587 static void
5588 mlxsw_sp_fib6_offload_failed_flag_set(struct mlxsw_sp *mlxsw_sp,
5589 				      struct fib6_info **rt_arr,
5590 				      unsigned int nrt6)
5591 {
5592 	int i;
5593 
5594 	/* In IPv6 a multipath route is represented using multiple routes, so
5595 	 * we need to set the flags on all of them.
5596 	 */
5597 	for (i = 0; i < nrt6; i++)
5598 		fib6_info_hw_flags_set(mlxsw_sp_net(mlxsw_sp), rt_arr[i],
5599 				       false, false, true);
5600 }
5601 #else
5602 static void
5603 mlxsw_sp_fib6_offload_failed_flag_set(struct mlxsw_sp *mlxsw_sp,
5604 				      struct fib6_info **rt_arr,
5605 				      unsigned int nrt6)
5606 {
5607 }
5608 #endif
5609 
5610 #if IS_ENABLED(CONFIG_IPV6)
5611 static void
5612 mlxsw_sp_fib6_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
5613 				 struct mlxsw_sp_fib_entry *fib_entry)
5614 {
5615 	struct mlxsw_sp_fib6_entry *fib6_entry;
5616 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5617 	bool should_offload;
5618 
5619 	should_offload = mlxsw_sp_fib_entry_should_offload(fib_entry);
5620 
5621 	/* In IPv6 a multipath route is represented using multiple routes, so
5622 	 * we need to set the flags on all of them.
5623 	 */
5624 	fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
5625 				  common);
5626 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list)
5627 		fib6_info_hw_flags_set(mlxsw_sp_net(mlxsw_sp), mlxsw_sp_rt6->rt,
5628 				       should_offload, !should_offload, false);
5629 }
5630 #else
5631 static void
5632 mlxsw_sp_fib6_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
5633 				 struct mlxsw_sp_fib_entry *fib_entry)
5634 {
5635 }
5636 #endif
5637 
5638 #if IS_ENABLED(CONFIG_IPV6)
5639 static void
5640 mlxsw_sp_fib6_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
5641 				   struct mlxsw_sp_fib_entry *fib_entry)
5642 {
5643 	struct mlxsw_sp_fib6_entry *fib6_entry;
5644 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5645 
5646 	fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
5647 				  common);
5648 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list)
5649 		fib6_info_hw_flags_set(mlxsw_sp_net(mlxsw_sp), mlxsw_sp_rt6->rt,
5650 				       false, false, false);
5651 }
5652 #else
5653 static void
5654 mlxsw_sp_fib6_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
5655 				   struct mlxsw_sp_fib_entry *fib_entry)
5656 {
5657 }
5658 #endif
5659 
5660 static void
5661 mlxsw_sp_fib_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
5662 				struct mlxsw_sp_fib_entry *fib_entry)
5663 {
5664 	switch (fib_entry->fib_node->fib->proto) {
5665 	case MLXSW_SP_L3_PROTO_IPV4:
5666 		mlxsw_sp_fib4_entry_hw_flags_set(mlxsw_sp, fib_entry);
5667 		break;
5668 	case MLXSW_SP_L3_PROTO_IPV6:
5669 		mlxsw_sp_fib6_entry_hw_flags_set(mlxsw_sp, fib_entry);
5670 		break;
5671 	}
5672 }
5673 
5674 static void
5675 mlxsw_sp_fib_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
5676 				  struct mlxsw_sp_fib_entry *fib_entry)
5677 {
5678 	switch (fib_entry->fib_node->fib->proto) {
5679 	case MLXSW_SP_L3_PROTO_IPV4:
5680 		mlxsw_sp_fib4_entry_hw_flags_clear(mlxsw_sp, fib_entry);
5681 		break;
5682 	case MLXSW_SP_L3_PROTO_IPV6:
5683 		mlxsw_sp_fib6_entry_hw_flags_clear(mlxsw_sp, fib_entry);
5684 		break;
5685 	}
5686 }
5687 
5688 static void
5689 mlxsw_sp_fib_entry_hw_flags_refresh(struct mlxsw_sp *mlxsw_sp,
5690 				    struct mlxsw_sp_fib_entry *fib_entry,
5691 				    enum mlxsw_reg_ralue_op op)
5692 {
5693 	switch (op) {
5694 	case MLXSW_REG_RALUE_OP_WRITE_WRITE:
5695 		mlxsw_sp_fib_entry_hw_flags_set(mlxsw_sp, fib_entry);
5696 		break;
5697 	case MLXSW_REG_RALUE_OP_WRITE_DELETE:
5698 		mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, fib_entry);
5699 		break;
5700 	default:
5701 		break;
5702 	}
5703 }
5704 
5705 static void
5706 mlxsw_sp_fib_entry_ralue_pack(char *ralue_pl,
5707 			      const struct mlxsw_sp_fib_entry *fib_entry,
5708 			      enum mlxsw_reg_ralue_op op)
5709 {
5710 	struct mlxsw_sp_fib *fib = fib_entry->fib_node->fib;
5711 	enum mlxsw_reg_ralxx_protocol proto;
5712 	u32 *p_dip;
5713 
5714 	proto = (enum mlxsw_reg_ralxx_protocol) fib->proto;
5715 
5716 	switch (fib->proto) {
5717 	case MLXSW_SP_L3_PROTO_IPV4:
5718 		p_dip = (u32 *) fib_entry->fib_node->key.addr;
5719 		mlxsw_reg_ralue_pack4(ralue_pl, proto, op, fib->vr->id,
5720 				      fib_entry->fib_node->key.prefix_len,
5721 				      *p_dip);
5722 		break;
5723 	case MLXSW_SP_L3_PROTO_IPV6:
5724 		mlxsw_reg_ralue_pack6(ralue_pl, proto, op, fib->vr->id,
5725 				      fib_entry->fib_node->key.prefix_len,
5726 				      fib_entry->fib_node->key.addr);
5727 		break;
5728 	}
5729 }
5730 
5731 static int mlxsw_sp_fib_entry_op_remote(struct mlxsw_sp *mlxsw_sp,
5732 					struct mlxsw_sp_fib_entry *fib_entry,
5733 					enum mlxsw_reg_ralue_op op)
5734 {
5735 	struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
5736 	struct mlxsw_sp_nexthop_group_info *nhgi = nh_group->nhgi;
5737 	char ralue_pl[MLXSW_REG_RALUE_LEN];
5738 	enum mlxsw_reg_ralue_trap_action trap_action;
5739 	u16 trap_id = 0;
5740 	u32 adjacency_index = 0;
5741 	u16 ecmp_size = 0;
5742 
5743 	/* In case the nexthop group adjacency index is valid, use it
5744 	 * with provided ECMP size. Otherwise, setup trap and pass
5745 	 * traffic to kernel.
5746 	 */
5747 	if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
5748 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
5749 		adjacency_index = nhgi->adj_index;
5750 		ecmp_size = nhgi->ecmp_size;
5751 	} else if (!nhgi->adj_index_valid && nhgi->count && nhgi->nh_rif) {
5752 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
5753 		adjacency_index = mlxsw_sp->router->adj_trap_index;
5754 		ecmp_size = 1;
5755 	} else {
5756 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
5757 		trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
5758 	}
5759 
5760 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
5761 	mlxsw_reg_ralue_act_remote_pack(ralue_pl, trap_action, trap_id,
5762 					adjacency_index, ecmp_size);
5763 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
5764 }
5765 
5766 static int mlxsw_sp_fib_entry_op_local(struct mlxsw_sp *mlxsw_sp,
5767 				       struct mlxsw_sp_fib_entry *fib_entry,
5768 				       enum mlxsw_reg_ralue_op op)
5769 {
5770 	struct mlxsw_sp_rif *rif = fib_entry->nh_group->nhgi->nh_rif;
5771 	enum mlxsw_reg_ralue_trap_action trap_action;
5772 	char ralue_pl[MLXSW_REG_RALUE_LEN];
5773 	u16 trap_id = 0;
5774 	u16 rif_index = 0;
5775 
5776 	if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
5777 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
5778 		rif_index = rif->rif_index;
5779 	} else {
5780 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
5781 		trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
5782 	}
5783 
5784 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
5785 	mlxsw_reg_ralue_act_local_pack(ralue_pl, trap_action, trap_id,
5786 				       rif_index);
5787 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
5788 }
5789 
5790 static int mlxsw_sp_fib_entry_op_trap(struct mlxsw_sp *mlxsw_sp,
5791 				      struct mlxsw_sp_fib_entry *fib_entry,
5792 				      enum mlxsw_reg_ralue_op op)
5793 {
5794 	char ralue_pl[MLXSW_REG_RALUE_LEN];
5795 
5796 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
5797 	mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
5798 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
5799 }
5800 
5801 static int mlxsw_sp_fib_entry_op_blackhole(struct mlxsw_sp *mlxsw_sp,
5802 					   struct mlxsw_sp_fib_entry *fib_entry,
5803 					   enum mlxsw_reg_ralue_op op)
5804 {
5805 	enum mlxsw_reg_ralue_trap_action trap_action;
5806 	char ralue_pl[MLXSW_REG_RALUE_LEN];
5807 
5808 	trap_action = MLXSW_REG_RALUE_TRAP_ACTION_DISCARD_ERROR;
5809 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
5810 	mlxsw_reg_ralue_act_local_pack(ralue_pl, trap_action, 0, 0);
5811 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
5812 }
5813 
5814 static int
5815 mlxsw_sp_fib_entry_op_unreachable(struct mlxsw_sp *mlxsw_sp,
5816 				  struct mlxsw_sp_fib_entry *fib_entry,
5817 				  enum mlxsw_reg_ralue_op op)
5818 {
5819 	enum mlxsw_reg_ralue_trap_action trap_action;
5820 	char ralue_pl[MLXSW_REG_RALUE_LEN];
5821 	u16 trap_id;
5822 
5823 	trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
5824 	trap_id = MLXSW_TRAP_ID_RTR_INGRESS1;
5825 
5826 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
5827 	mlxsw_reg_ralue_act_local_pack(ralue_pl, trap_action, trap_id, 0);
5828 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
5829 }
5830 
5831 static int
5832 mlxsw_sp_fib_entry_op_ipip_decap(struct mlxsw_sp *mlxsw_sp,
5833 				 struct mlxsw_sp_fib_entry *fib_entry,
5834 				 enum mlxsw_reg_ralue_op op)
5835 {
5836 	struct mlxsw_sp_ipip_entry *ipip_entry = fib_entry->decap.ipip_entry;
5837 	const struct mlxsw_sp_ipip_ops *ipip_ops;
5838 	char ralue_pl[MLXSW_REG_RALUE_LEN];
5839 	int err;
5840 
5841 	if (WARN_ON(!ipip_entry))
5842 		return -EINVAL;
5843 
5844 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
5845 	err = ipip_ops->decap_config(mlxsw_sp, ipip_entry,
5846 				     fib_entry->decap.tunnel_index);
5847 	if (err)
5848 		return err;
5849 
5850 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
5851 	mlxsw_reg_ralue_act_ip2me_tun_pack(ralue_pl,
5852 					   fib_entry->decap.tunnel_index);
5853 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
5854 }
5855 
5856 static int mlxsw_sp_fib_entry_op_nve_decap(struct mlxsw_sp *mlxsw_sp,
5857 					   struct mlxsw_sp_fib_entry *fib_entry,
5858 					   enum mlxsw_reg_ralue_op op)
5859 {
5860 	char ralue_pl[MLXSW_REG_RALUE_LEN];
5861 
5862 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
5863 	mlxsw_reg_ralue_act_ip2me_tun_pack(ralue_pl,
5864 					   fib_entry->decap.tunnel_index);
5865 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
5866 }
5867 
5868 static int __mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
5869 				   struct mlxsw_sp_fib_entry *fib_entry,
5870 				   enum mlxsw_reg_ralue_op op)
5871 {
5872 	switch (fib_entry->type) {
5873 	case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
5874 		return mlxsw_sp_fib_entry_op_remote(mlxsw_sp, fib_entry, op);
5875 	case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
5876 		return mlxsw_sp_fib_entry_op_local(mlxsw_sp, fib_entry, op);
5877 	case MLXSW_SP_FIB_ENTRY_TYPE_TRAP:
5878 		return mlxsw_sp_fib_entry_op_trap(mlxsw_sp, fib_entry, op);
5879 	case MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE:
5880 		return mlxsw_sp_fib_entry_op_blackhole(mlxsw_sp, fib_entry, op);
5881 	case MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE:
5882 		return mlxsw_sp_fib_entry_op_unreachable(mlxsw_sp, fib_entry,
5883 							 op);
5884 	case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
5885 		return mlxsw_sp_fib_entry_op_ipip_decap(mlxsw_sp,
5886 							fib_entry, op);
5887 	case MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP:
5888 		return mlxsw_sp_fib_entry_op_nve_decap(mlxsw_sp, fib_entry, op);
5889 	}
5890 	return -EINVAL;
5891 }
5892 
5893 static int mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
5894 				 struct mlxsw_sp_fib_entry *fib_entry,
5895 				 enum mlxsw_reg_ralue_op op)
5896 {
5897 	int err = __mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry, op);
5898 
5899 	if (err)
5900 		return err;
5901 
5902 	mlxsw_sp_fib_entry_hw_flags_refresh(mlxsw_sp, fib_entry, op);
5903 
5904 	return err;
5905 }
5906 
5907 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
5908 				     struct mlxsw_sp_fib_entry *fib_entry)
5909 {
5910 	return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
5911 				     MLXSW_REG_RALUE_OP_WRITE_WRITE);
5912 }
5913 
5914 static int mlxsw_sp_fib_entry_del(struct mlxsw_sp *mlxsw_sp,
5915 				  struct mlxsw_sp_fib_entry *fib_entry)
5916 {
5917 	return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
5918 				     MLXSW_REG_RALUE_OP_WRITE_DELETE);
5919 }
5920 
5921 static int
5922 mlxsw_sp_fib4_entry_type_set(struct mlxsw_sp *mlxsw_sp,
5923 			     const struct fib_entry_notifier_info *fen_info,
5924 			     struct mlxsw_sp_fib_entry *fib_entry)
5925 {
5926 	struct mlxsw_sp_nexthop_group_info *nhgi = fib_entry->nh_group->nhgi;
5927 	union mlxsw_sp_l3addr dip = { .addr4 = htonl(fen_info->dst) };
5928 	struct mlxsw_sp_router *router = mlxsw_sp->router;
5929 	u32 tb_id = mlxsw_sp_fix_tb_id(fen_info->tb_id);
5930 	int ifindex = nhgi->nexthops[0].ifindex;
5931 	struct mlxsw_sp_ipip_entry *ipip_entry;
5932 
5933 	switch (fen_info->type) {
5934 	case RTN_LOCAL:
5935 		ipip_entry = mlxsw_sp_ipip_entry_find_by_decap(mlxsw_sp, ifindex,
5936 							       MLXSW_SP_L3_PROTO_IPV4, dip);
5937 		if (ipip_entry && ipip_entry->ol_dev->flags & IFF_UP) {
5938 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
5939 			return mlxsw_sp_fib_entry_decap_init(mlxsw_sp,
5940 							     fib_entry,
5941 							     ipip_entry);
5942 		}
5943 		if (mlxsw_sp_router_nve_is_decap(mlxsw_sp, tb_id,
5944 						 MLXSW_SP_L3_PROTO_IPV4,
5945 						 &dip)) {
5946 			u32 tunnel_index;
5947 
5948 			tunnel_index = router->nve_decap_config.tunnel_index;
5949 			fib_entry->decap.tunnel_index = tunnel_index;
5950 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
5951 			return 0;
5952 		}
5953 		fallthrough;
5954 	case RTN_BROADCAST:
5955 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
5956 		return 0;
5957 	case RTN_BLACKHOLE:
5958 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE;
5959 		return 0;
5960 	case RTN_UNREACHABLE:
5961 	case RTN_PROHIBIT:
5962 		/* Packets hitting these routes need to be trapped, but
5963 		 * can do so with a lower priority than packets directed
5964 		 * at the host, so use action type local instead of trap.
5965 		 */
5966 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE;
5967 		return 0;
5968 	case RTN_UNICAST:
5969 		if (nhgi->gateway)
5970 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
5971 		else
5972 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
5973 		return 0;
5974 	default:
5975 		return -EINVAL;
5976 	}
5977 }
5978 
5979 static void
5980 mlxsw_sp_fib_entry_type_unset(struct mlxsw_sp *mlxsw_sp,
5981 			      struct mlxsw_sp_fib_entry *fib_entry)
5982 {
5983 	switch (fib_entry->type) {
5984 	case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
5985 		mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, fib_entry);
5986 		break;
5987 	default:
5988 		break;
5989 	}
5990 }
5991 
5992 static void
5993 mlxsw_sp_fib4_entry_type_unset(struct mlxsw_sp *mlxsw_sp,
5994 			       struct mlxsw_sp_fib4_entry *fib4_entry)
5995 {
5996 	mlxsw_sp_fib_entry_type_unset(mlxsw_sp, &fib4_entry->common);
5997 }
5998 
5999 static struct mlxsw_sp_fib4_entry *
6000 mlxsw_sp_fib4_entry_create(struct mlxsw_sp *mlxsw_sp,
6001 			   struct mlxsw_sp_fib_node *fib_node,
6002 			   const struct fib_entry_notifier_info *fen_info)
6003 {
6004 	struct mlxsw_sp_fib4_entry *fib4_entry;
6005 	struct mlxsw_sp_fib_entry *fib_entry;
6006 	int err;
6007 
6008 	fib4_entry = kzalloc(sizeof(*fib4_entry), GFP_KERNEL);
6009 	if (!fib4_entry)
6010 		return ERR_PTR(-ENOMEM);
6011 	fib_entry = &fib4_entry->common;
6012 
6013 	err = mlxsw_sp_nexthop4_group_get(mlxsw_sp, fib_entry, fen_info->fi);
6014 	if (err)
6015 		goto err_nexthop4_group_get;
6016 
6017 	err = mlxsw_sp_nexthop_group_vr_link(fib_entry->nh_group,
6018 					     fib_node->fib);
6019 	if (err)
6020 		goto err_nexthop_group_vr_link;
6021 
6022 	err = mlxsw_sp_fib4_entry_type_set(mlxsw_sp, fen_info, fib_entry);
6023 	if (err)
6024 		goto err_fib4_entry_type_set;
6025 
6026 	fib4_entry->fi = fen_info->fi;
6027 	fib_info_hold(fib4_entry->fi);
6028 	fib4_entry->tb_id = fen_info->tb_id;
6029 	fib4_entry->type = fen_info->type;
6030 	fib4_entry->dscp = fen_info->dscp;
6031 
6032 	fib_entry->fib_node = fib_node;
6033 
6034 	return fib4_entry;
6035 
6036 err_fib4_entry_type_set:
6037 	mlxsw_sp_nexthop_group_vr_unlink(fib_entry->nh_group, fib_node->fib);
6038 err_nexthop_group_vr_link:
6039 	mlxsw_sp_nexthop4_group_put(mlxsw_sp, &fib4_entry->common);
6040 err_nexthop4_group_get:
6041 	kfree(fib4_entry);
6042 	return ERR_PTR(err);
6043 }
6044 
6045 static void mlxsw_sp_fib4_entry_destroy(struct mlxsw_sp *mlxsw_sp,
6046 					struct mlxsw_sp_fib4_entry *fib4_entry)
6047 {
6048 	struct mlxsw_sp_fib_node *fib_node = fib4_entry->common.fib_node;
6049 
6050 	fib_info_put(fib4_entry->fi);
6051 	mlxsw_sp_fib4_entry_type_unset(mlxsw_sp, fib4_entry);
6052 	mlxsw_sp_nexthop_group_vr_unlink(fib4_entry->common.nh_group,
6053 					 fib_node->fib);
6054 	mlxsw_sp_nexthop4_group_put(mlxsw_sp, &fib4_entry->common);
6055 	kfree(fib4_entry);
6056 }
6057 
6058 static struct mlxsw_sp_fib4_entry *
6059 mlxsw_sp_fib4_entry_lookup(struct mlxsw_sp *mlxsw_sp,
6060 			   const struct fib_entry_notifier_info *fen_info)
6061 {
6062 	struct mlxsw_sp_fib4_entry *fib4_entry;
6063 	struct mlxsw_sp_fib_node *fib_node;
6064 	struct mlxsw_sp_fib *fib;
6065 	struct mlxsw_sp_vr *vr;
6066 
6067 	vr = mlxsw_sp_vr_find(mlxsw_sp, fen_info->tb_id);
6068 	if (!vr)
6069 		return NULL;
6070 	fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV4);
6071 
6072 	fib_node = mlxsw_sp_fib_node_lookup(fib, &fen_info->dst,
6073 					    sizeof(fen_info->dst),
6074 					    fen_info->dst_len);
6075 	if (!fib_node)
6076 		return NULL;
6077 
6078 	fib4_entry = container_of(fib_node->fib_entry,
6079 				  struct mlxsw_sp_fib4_entry, common);
6080 	if (fib4_entry->tb_id == fen_info->tb_id &&
6081 	    fib4_entry->dscp == fen_info->dscp &&
6082 	    fib4_entry->type == fen_info->type &&
6083 	    fib4_entry->fi == fen_info->fi)
6084 		return fib4_entry;
6085 
6086 	return NULL;
6087 }
6088 
6089 static const struct rhashtable_params mlxsw_sp_fib_ht_params = {
6090 	.key_offset = offsetof(struct mlxsw_sp_fib_node, key),
6091 	.head_offset = offsetof(struct mlxsw_sp_fib_node, ht_node),
6092 	.key_len = sizeof(struct mlxsw_sp_fib_key),
6093 	.automatic_shrinking = true,
6094 };
6095 
6096 static int mlxsw_sp_fib_node_insert(struct mlxsw_sp_fib *fib,
6097 				    struct mlxsw_sp_fib_node *fib_node)
6098 {
6099 	return rhashtable_insert_fast(&fib->ht, &fib_node->ht_node,
6100 				      mlxsw_sp_fib_ht_params);
6101 }
6102 
6103 static void mlxsw_sp_fib_node_remove(struct mlxsw_sp_fib *fib,
6104 				     struct mlxsw_sp_fib_node *fib_node)
6105 {
6106 	rhashtable_remove_fast(&fib->ht, &fib_node->ht_node,
6107 			       mlxsw_sp_fib_ht_params);
6108 }
6109 
6110 static struct mlxsw_sp_fib_node *
6111 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
6112 			 size_t addr_len, unsigned char prefix_len)
6113 {
6114 	struct mlxsw_sp_fib_key key;
6115 
6116 	memset(&key, 0, sizeof(key));
6117 	memcpy(key.addr, addr, addr_len);
6118 	key.prefix_len = prefix_len;
6119 	return rhashtable_lookup_fast(&fib->ht, &key, mlxsw_sp_fib_ht_params);
6120 }
6121 
6122 static struct mlxsw_sp_fib_node *
6123 mlxsw_sp_fib_node_create(struct mlxsw_sp_fib *fib, const void *addr,
6124 			 size_t addr_len, unsigned char prefix_len)
6125 {
6126 	struct mlxsw_sp_fib_node *fib_node;
6127 
6128 	fib_node = kzalloc(sizeof(*fib_node), GFP_KERNEL);
6129 	if (!fib_node)
6130 		return NULL;
6131 
6132 	list_add(&fib_node->list, &fib->node_list);
6133 	memcpy(fib_node->key.addr, addr, addr_len);
6134 	fib_node->key.prefix_len = prefix_len;
6135 
6136 	return fib_node;
6137 }
6138 
6139 static void mlxsw_sp_fib_node_destroy(struct mlxsw_sp_fib_node *fib_node)
6140 {
6141 	list_del(&fib_node->list);
6142 	kfree(fib_node);
6143 }
6144 
6145 static int mlxsw_sp_fib_lpm_tree_link(struct mlxsw_sp *mlxsw_sp,
6146 				      struct mlxsw_sp_fib_node *fib_node)
6147 {
6148 	struct mlxsw_sp_prefix_usage req_prefix_usage;
6149 	struct mlxsw_sp_fib *fib = fib_node->fib;
6150 	struct mlxsw_sp_lpm_tree *lpm_tree;
6151 	int err;
6152 
6153 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[fib->proto];
6154 	if (lpm_tree->prefix_ref_count[fib_node->key.prefix_len] != 0)
6155 		goto out;
6156 
6157 	mlxsw_sp_prefix_usage_cpy(&req_prefix_usage, &lpm_tree->prefix_usage);
6158 	mlxsw_sp_prefix_usage_set(&req_prefix_usage, fib_node->key.prefix_len);
6159 	lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
6160 					 fib->proto);
6161 	if (IS_ERR(lpm_tree))
6162 		return PTR_ERR(lpm_tree);
6163 
6164 	err = mlxsw_sp_vrs_lpm_tree_replace(mlxsw_sp, fib, lpm_tree);
6165 	if (err)
6166 		goto err_lpm_tree_replace;
6167 
6168 out:
6169 	lpm_tree->prefix_ref_count[fib_node->key.prefix_len]++;
6170 	return 0;
6171 
6172 err_lpm_tree_replace:
6173 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
6174 	return err;
6175 }
6176 
6177 static void mlxsw_sp_fib_lpm_tree_unlink(struct mlxsw_sp *mlxsw_sp,
6178 					 struct mlxsw_sp_fib_node *fib_node)
6179 {
6180 	struct mlxsw_sp_lpm_tree *lpm_tree = fib_node->fib->lpm_tree;
6181 	struct mlxsw_sp_prefix_usage req_prefix_usage;
6182 	struct mlxsw_sp_fib *fib = fib_node->fib;
6183 	int err;
6184 
6185 	if (--lpm_tree->prefix_ref_count[fib_node->key.prefix_len] != 0)
6186 		return;
6187 	/* Try to construct a new LPM tree from the current prefix usage
6188 	 * minus the unused one. If we fail, continue using the old one.
6189 	 */
6190 	mlxsw_sp_prefix_usage_cpy(&req_prefix_usage, &lpm_tree->prefix_usage);
6191 	mlxsw_sp_prefix_usage_clear(&req_prefix_usage,
6192 				    fib_node->key.prefix_len);
6193 	lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
6194 					 fib->proto);
6195 	if (IS_ERR(lpm_tree))
6196 		return;
6197 
6198 	err = mlxsw_sp_vrs_lpm_tree_replace(mlxsw_sp, fib, lpm_tree);
6199 	if (err)
6200 		goto err_lpm_tree_replace;
6201 
6202 	return;
6203 
6204 err_lpm_tree_replace:
6205 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
6206 }
6207 
6208 static int mlxsw_sp_fib_node_init(struct mlxsw_sp *mlxsw_sp,
6209 				  struct mlxsw_sp_fib_node *fib_node,
6210 				  struct mlxsw_sp_fib *fib)
6211 {
6212 	int err;
6213 
6214 	err = mlxsw_sp_fib_node_insert(fib, fib_node);
6215 	if (err)
6216 		return err;
6217 	fib_node->fib = fib;
6218 
6219 	err = mlxsw_sp_fib_lpm_tree_link(mlxsw_sp, fib_node);
6220 	if (err)
6221 		goto err_fib_lpm_tree_link;
6222 
6223 	return 0;
6224 
6225 err_fib_lpm_tree_link:
6226 	fib_node->fib = NULL;
6227 	mlxsw_sp_fib_node_remove(fib, fib_node);
6228 	return err;
6229 }
6230 
6231 static void mlxsw_sp_fib_node_fini(struct mlxsw_sp *mlxsw_sp,
6232 				   struct mlxsw_sp_fib_node *fib_node)
6233 {
6234 	struct mlxsw_sp_fib *fib = fib_node->fib;
6235 
6236 	mlxsw_sp_fib_lpm_tree_unlink(mlxsw_sp, fib_node);
6237 	fib_node->fib = NULL;
6238 	mlxsw_sp_fib_node_remove(fib, fib_node);
6239 }
6240 
6241 static struct mlxsw_sp_fib_node *
6242 mlxsw_sp_fib_node_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id, const void *addr,
6243 		      size_t addr_len, unsigned char prefix_len,
6244 		      enum mlxsw_sp_l3proto proto)
6245 {
6246 	struct mlxsw_sp_fib_node *fib_node;
6247 	struct mlxsw_sp_fib *fib;
6248 	struct mlxsw_sp_vr *vr;
6249 	int err;
6250 
6251 	vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id, NULL);
6252 	if (IS_ERR(vr))
6253 		return ERR_CAST(vr);
6254 	fib = mlxsw_sp_vr_fib(vr, proto);
6255 
6256 	fib_node = mlxsw_sp_fib_node_lookup(fib, addr, addr_len, prefix_len);
6257 	if (fib_node)
6258 		return fib_node;
6259 
6260 	fib_node = mlxsw_sp_fib_node_create(fib, addr, addr_len, prefix_len);
6261 	if (!fib_node) {
6262 		err = -ENOMEM;
6263 		goto err_fib_node_create;
6264 	}
6265 
6266 	err = mlxsw_sp_fib_node_init(mlxsw_sp, fib_node, fib);
6267 	if (err)
6268 		goto err_fib_node_init;
6269 
6270 	return fib_node;
6271 
6272 err_fib_node_init:
6273 	mlxsw_sp_fib_node_destroy(fib_node);
6274 err_fib_node_create:
6275 	mlxsw_sp_vr_put(mlxsw_sp, vr);
6276 	return ERR_PTR(err);
6277 }
6278 
6279 static void mlxsw_sp_fib_node_put(struct mlxsw_sp *mlxsw_sp,
6280 				  struct mlxsw_sp_fib_node *fib_node)
6281 {
6282 	struct mlxsw_sp_vr *vr = fib_node->fib->vr;
6283 
6284 	if (fib_node->fib_entry)
6285 		return;
6286 	mlxsw_sp_fib_node_fini(mlxsw_sp, fib_node);
6287 	mlxsw_sp_fib_node_destroy(fib_node);
6288 	mlxsw_sp_vr_put(mlxsw_sp, vr);
6289 }
6290 
6291 static int mlxsw_sp_fib_node_entry_link(struct mlxsw_sp *mlxsw_sp,
6292 					struct mlxsw_sp_fib_entry *fib_entry)
6293 {
6294 	struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
6295 	int err;
6296 
6297 	fib_node->fib_entry = fib_entry;
6298 
6299 	err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
6300 	if (err)
6301 		goto err_fib_entry_update;
6302 
6303 	return 0;
6304 
6305 err_fib_entry_update:
6306 	fib_node->fib_entry = NULL;
6307 	return err;
6308 }
6309 
6310 static void
6311 mlxsw_sp_fib_node_entry_unlink(struct mlxsw_sp *mlxsw_sp,
6312 			       struct mlxsw_sp_fib_entry *fib_entry)
6313 {
6314 	struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
6315 
6316 	mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
6317 	fib_node->fib_entry = NULL;
6318 }
6319 
6320 static bool mlxsw_sp_fib4_allow_replace(struct mlxsw_sp_fib4_entry *fib4_entry)
6321 {
6322 	struct mlxsw_sp_fib_node *fib_node = fib4_entry->common.fib_node;
6323 	struct mlxsw_sp_fib4_entry *fib4_replaced;
6324 
6325 	if (!fib_node->fib_entry)
6326 		return true;
6327 
6328 	fib4_replaced = container_of(fib_node->fib_entry,
6329 				     struct mlxsw_sp_fib4_entry, common);
6330 	if (fib4_entry->tb_id == RT_TABLE_MAIN &&
6331 	    fib4_replaced->tb_id == RT_TABLE_LOCAL)
6332 		return false;
6333 
6334 	return true;
6335 }
6336 
6337 static int
6338 mlxsw_sp_router_fib4_replace(struct mlxsw_sp *mlxsw_sp,
6339 			     const struct fib_entry_notifier_info *fen_info)
6340 {
6341 	struct mlxsw_sp_fib4_entry *fib4_entry, *fib4_replaced;
6342 	struct mlxsw_sp_fib_entry *replaced;
6343 	struct mlxsw_sp_fib_node *fib_node;
6344 	int err;
6345 
6346 	if (fen_info->fi->nh &&
6347 	    !mlxsw_sp_nexthop_obj_group_lookup(mlxsw_sp, fen_info->fi->nh->id))
6348 		return 0;
6349 
6350 	fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, fen_info->tb_id,
6351 					 &fen_info->dst, sizeof(fen_info->dst),
6352 					 fen_info->dst_len,
6353 					 MLXSW_SP_L3_PROTO_IPV4);
6354 	if (IS_ERR(fib_node)) {
6355 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to get FIB node\n");
6356 		return PTR_ERR(fib_node);
6357 	}
6358 
6359 	fib4_entry = mlxsw_sp_fib4_entry_create(mlxsw_sp, fib_node, fen_info);
6360 	if (IS_ERR(fib4_entry)) {
6361 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to create FIB entry\n");
6362 		err = PTR_ERR(fib4_entry);
6363 		goto err_fib4_entry_create;
6364 	}
6365 
6366 	if (!mlxsw_sp_fib4_allow_replace(fib4_entry)) {
6367 		mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
6368 		mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
6369 		return 0;
6370 	}
6371 
6372 	replaced = fib_node->fib_entry;
6373 	err = mlxsw_sp_fib_node_entry_link(mlxsw_sp, &fib4_entry->common);
6374 	if (err) {
6375 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to link FIB entry to node\n");
6376 		goto err_fib_node_entry_link;
6377 	}
6378 
6379 	/* Nothing to replace */
6380 	if (!replaced)
6381 		return 0;
6382 
6383 	mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, replaced);
6384 	fib4_replaced = container_of(replaced, struct mlxsw_sp_fib4_entry,
6385 				     common);
6386 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_replaced);
6387 
6388 	return 0;
6389 
6390 err_fib_node_entry_link:
6391 	fib_node->fib_entry = replaced;
6392 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
6393 err_fib4_entry_create:
6394 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
6395 	return err;
6396 }
6397 
6398 static void mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
6399 				     struct fib_entry_notifier_info *fen_info)
6400 {
6401 	struct mlxsw_sp_fib4_entry *fib4_entry;
6402 	struct mlxsw_sp_fib_node *fib_node;
6403 
6404 	fib4_entry = mlxsw_sp_fib4_entry_lookup(mlxsw_sp, fen_info);
6405 	if (!fib4_entry)
6406 		return;
6407 	fib_node = fib4_entry->common.fib_node;
6408 
6409 	mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, &fib4_entry->common);
6410 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
6411 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
6412 }
6413 
6414 static bool mlxsw_sp_fib6_rt_should_ignore(const struct fib6_info *rt)
6415 {
6416 	/* Multicast routes aren't supported, so ignore them. Neighbour
6417 	 * Discovery packets are specifically trapped.
6418 	 */
6419 	if (ipv6_addr_type(&rt->fib6_dst.addr) & IPV6_ADDR_MULTICAST)
6420 		return true;
6421 
6422 	/* Cloned routes are irrelevant in the forwarding path. */
6423 	if (rt->fib6_flags & RTF_CACHE)
6424 		return true;
6425 
6426 	return false;
6427 }
6428 
6429 static struct mlxsw_sp_rt6 *mlxsw_sp_rt6_create(struct fib6_info *rt)
6430 {
6431 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
6432 
6433 	mlxsw_sp_rt6 = kzalloc(sizeof(*mlxsw_sp_rt6), GFP_KERNEL);
6434 	if (!mlxsw_sp_rt6)
6435 		return ERR_PTR(-ENOMEM);
6436 
6437 	/* In case of route replace, replaced route is deleted with
6438 	 * no notification. Take reference to prevent accessing freed
6439 	 * memory.
6440 	 */
6441 	mlxsw_sp_rt6->rt = rt;
6442 	fib6_info_hold(rt);
6443 
6444 	return mlxsw_sp_rt6;
6445 }
6446 
6447 #if IS_ENABLED(CONFIG_IPV6)
6448 static void mlxsw_sp_rt6_release(struct fib6_info *rt)
6449 {
6450 	fib6_info_release(rt);
6451 }
6452 #else
6453 static void mlxsw_sp_rt6_release(struct fib6_info *rt)
6454 {
6455 }
6456 #endif
6457 
6458 static void mlxsw_sp_rt6_destroy(struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
6459 {
6460 	struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
6461 
6462 	if (!mlxsw_sp_rt6->rt->nh)
6463 		fib6_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
6464 	mlxsw_sp_rt6_release(mlxsw_sp_rt6->rt);
6465 	kfree(mlxsw_sp_rt6);
6466 }
6467 
6468 static struct fib6_info *
6469 mlxsw_sp_fib6_entry_rt(const struct mlxsw_sp_fib6_entry *fib6_entry)
6470 {
6471 	return list_first_entry(&fib6_entry->rt6_list, struct mlxsw_sp_rt6,
6472 				list)->rt;
6473 }
6474 
6475 static struct mlxsw_sp_rt6 *
6476 mlxsw_sp_fib6_entry_rt_find(const struct mlxsw_sp_fib6_entry *fib6_entry,
6477 			    const struct fib6_info *rt)
6478 {
6479 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
6480 
6481 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
6482 		if (mlxsw_sp_rt6->rt == rt)
6483 			return mlxsw_sp_rt6;
6484 	}
6485 
6486 	return NULL;
6487 }
6488 
6489 static bool mlxsw_sp_nexthop6_ipip_type(const struct mlxsw_sp *mlxsw_sp,
6490 					const struct fib6_info *rt,
6491 					enum mlxsw_sp_ipip_type *ret)
6492 {
6493 	return rt->fib6_nh->fib_nh_dev &&
6494 	       mlxsw_sp_netdev_ipip_type(mlxsw_sp, rt->fib6_nh->fib_nh_dev, ret);
6495 }
6496 
6497 static int mlxsw_sp_nexthop6_init(struct mlxsw_sp *mlxsw_sp,
6498 				  struct mlxsw_sp_nexthop_group *nh_grp,
6499 				  struct mlxsw_sp_nexthop *nh,
6500 				  const struct fib6_info *rt)
6501 {
6502 	struct net_device *dev = rt->fib6_nh->fib_nh_dev;
6503 	int err;
6504 
6505 	nh->nhgi = nh_grp->nhgi;
6506 	nh->nh_weight = rt->fib6_nh->fib_nh_weight;
6507 	memcpy(&nh->gw_addr, &rt->fib6_nh->fib_nh_gw6, sizeof(nh->gw_addr));
6508 #if IS_ENABLED(CONFIG_IPV6)
6509 	nh->neigh_tbl = &nd_tbl;
6510 #endif
6511 	mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
6512 
6513 	list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
6514 
6515 	if (!dev)
6516 		return 0;
6517 	nh->ifindex = dev->ifindex;
6518 
6519 	err = mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, dev);
6520 	if (err)
6521 		goto err_nexthop_type_init;
6522 
6523 	return 0;
6524 
6525 err_nexthop_type_init:
6526 	list_del(&nh->router_list_node);
6527 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
6528 	return err;
6529 }
6530 
6531 static void mlxsw_sp_nexthop6_fini(struct mlxsw_sp *mlxsw_sp,
6532 				   struct mlxsw_sp_nexthop *nh)
6533 {
6534 	mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
6535 	list_del(&nh->router_list_node);
6536 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
6537 }
6538 
6539 static bool mlxsw_sp_rt6_is_gateway(const struct mlxsw_sp *mlxsw_sp,
6540 				    const struct fib6_info *rt)
6541 {
6542 	return rt->fib6_nh->fib_nh_gw_family ||
6543 	       mlxsw_sp_nexthop6_ipip_type(mlxsw_sp, rt, NULL);
6544 }
6545 
6546 static int
6547 mlxsw_sp_nexthop6_group_info_init(struct mlxsw_sp *mlxsw_sp,
6548 				  struct mlxsw_sp_nexthop_group *nh_grp,
6549 				  struct mlxsw_sp_fib6_entry *fib6_entry)
6550 {
6551 	struct mlxsw_sp_nexthop_group_info *nhgi;
6552 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
6553 	struct mlxsw_sp_nexthop *nh;
6554 	int err, i;
6555 
6556 	nhgi = kzalloc(struct_size(nhgi, nexthops, fib6_entry->nrt6),
6557 		       GFP_KERNEL);
6558 	if (!nhgi)
6559 		return -ENOMEM;
6560 	nh_grp->nhgi = nhgi;
6561 	nhgi->nh_grp = nh_grp;
6562 	mlxsw_sp_rt6 = list_first_entry(&fib6_entry->rt6_list,
6563 					struct mlxsw_sp_rt6, list);
6564 	nhgi->gateway = mlxsw_sp_rt6_is_gateway(mlxsw_sp, mlxsw_sp_rt6->rt);
6565 	nhgi->count = fib6_entry->nrt6;
6566 	for (i = 0; i < nhgi->count; i++) {
6567 		struct fib6_info *rt = mlxsw_sp_rt6->rt;
6568 
6569 		nh = &nhgi->nexthops[i];
6570 		err = mlxsw_sp_nexthop6_init(mlxsw_sp, nh_grp, nh, rt);
6571 		if (err)
6572 			goto err_nexthop6_init;
6573 		mlxsw_sp_rt6 = list_next_entry(mlxsw_sp_rt6, list);
6574 	}
6575 	nh_grp->nhgi = nhgi;
6576 	err = mlxsw_sp_nexthop_group_inc(mlxsw_sp);
6577 	if (err)
6578 		goto err_group_inc;
6579 	err = mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
6580 	if (err)
6581 		goto err_group_refresh;
6582 
6583 	return 0;
6584 
6585 err_group_refresh:
6586 	mlxsw_sp_nexthop_group_dec(mlxsw_sp);
6587 err_group_inc:
6588 	i = nhgi->count;
6589 err_nexthop6_init:
6590 	for (i--; i >= 0; i--) {
6591 		nh = &nhgi->nexthops[i];
6592 		mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
6593 	}
6594 	kfree(nhgi);
6595 	return err;
6596 }
6597 
6598 static void
6599 mlxsw_sp_nexthop6_group_info_fini(struct mlxsw_sp *mlxsw_sp,
6600 				  struct mlxsw_sp_nexthop_group *nh_grp)
6601 {
6602 	struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
6603 	int i;
6604 
6605 	mlxsw_sp_nexthop_group_dec(mlxsw_sp);
6606 	for (i = nhgi->count - 1; i >= 0; i--) {
6607 		struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
6608 
6609 		mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
6610 	}
6611 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
6612 	WARN_ON_ONCE(nhgi->adj_index_valid);
6613 	kfree(nhgi);
6614 }
6615 
6616 static struct mlxsw_sp_nexthop_group *
6617 mlxsw_sp_nexthop6_group_create(struct mlxsw_sp *mlxsw_sp,
6618 			       struct mlxsw_sp_fib6_entry *fib6_entry)
6619 {
6620 	struct mlxsw_sp_nexthop_group *nh_grp;
6621 	int err;
6622 
6623 	nh_grp = kzalloc(sizeof(*nh_grp), GFP_KERNEL);
6624 	if (!nh_grp)
6625 		return ERR_PTR(-ENOMEM);
6626 	INIT_LIST_HEAD(&nh_grp->vr_list);
6627 	err = rhashtable_init(&nh_grp->vr_ht,
6628 			      &mlxsw_sp_nexthop_group_vr_ht_params);
6629 	if (err)
6630 		goto err_nexthop_group_vr_ht_init;
6631 	INIT_LIST_HEAD(&nh_grp->fib_list);
6632 	nh_grp->type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6;
6633 
6634 	err = mlxsw_sp_nexthop6_group_info_init(mlxsw_sp, nh_grp, fib6_entry);
6635 	if (err)
6636 		goto err_nexthop_group_info_init;
6637 
6638 	err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
6639 	if (err)
6640 		goto err_nexthop_group_insert;
6641 
6642 	nh_grp->can_destroy = true;
6643 
6644 	return nh_grp;
6645 
6646 err_nexthop_group_insert:
6647 	mlxsw_sp_nexthop6_group_info_fini(mlxsw_sp, nh_grp);
6648 err_nexthop_group_info_init:
6649 	rhashtable_destroy(&nh_grp->vr_ht);
6650 err_nexthop_group_vr_ht_init:
6651 	kfree(nh_grp);
6652 	return ERR_PTR(err);
6653 }
6654 
6655 static void
6656 mlxsw_sp_nexthop6_group_destroy(struct mlxsw_sp *mlxsw_sp,
6657 				struct mlxsw_sp_nexthop_group *nh_grp)
6658 {
6659 	if (!nh_grp->can_destroy)
6660 		return;
6661 	mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
6662 	mlxsw_sp_nexthop6_group_info_fini(mlxsw_sp, nh_grp);
6663 	WARN_ON_ONCE(!list_empty(&nh_grp->vr_list));
6664 	rhashtable_destroy(&nh_grp->vr_ht);
6665 	kfree(nh_grp);
6666 }
6667 
6668 static int mlxsw_sp_nexthop6_group_get(struct mlxsw_sp *mlxsw_sp,
6669 				       struct mlxsw_sp_fib6_entry *fib6_entry)
6670 {
6671 	struct fib6_info *rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
6672 	struct mlxsw_sp_nexthop_group *nh_grp;
6673 
6674 	if (rt->nh) {
6675 		nh_grp = mlxsw_sp_nexthop_obj_group_lookup(mlxsw_sp,
6676 							   rt->nh->id);
6677 		if (WARN_ON_ONCE(!nh_grp))
6678 			return -EINVAL;
6679 		goto out;
6680 	}
6681 
6682 	nh_grp = mlxsw_sp_nexthop6_group_lookup(mlxsw_sp, fib6_entry);
6683 	if (!nh_grp) {
6684 		nh_grp = mlxsw_sp_nexthop6_group_create(mlxsw_sp, fib6_entry);
6685 		if (IS_ERR(nh_grp))
6686 			return PTR_ERR(nh_grp);
6687 	}
6688 
6689 	/* The route and the nexthop are described by the same struct, so we
6690 	 * need to the update the nexthop offload indication for the new route.
6691 	 */
6692 	__mlxsw_sp_nexthop6_group_offload_refresh(nh_grp, fib6_entry);
6693 
6694 out:
6695 	list_add_tail(&fib6_entry->common.nexthop_group_node,
6696 		      &nh_grp->fib_list);
6697 	fib6_entry->common.nh_group = nh_grp;
6698 
6699 	return 0;
6700 }
6701 
6702 static void mlxsw_sp_nexthop6_group_put(struct mlxsw_sp *mlxsw_sp,
6703 					struct mlxsw_sp_fib_entry *fib_entry)
6704 {
6705 	struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
6706 
6707 	list_del(&fib_entry->nexthop_group_node);
6708 	if (!list_empty(&nh_grp->fib_list))
6709 		return;
6710 
6711 	if (nh_grp->type == MLXSW_SP_NEXTHOP_GROUP_TYPE_OBJ) {
6712 		mlxsw_sp_nexthop_obj_group_destroy(mlxsw_sp, nh_grp);
6713 		return;
6714 	}
6715 
6716 	mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, nh_grp);
6717 }
6718 
6719 static int
6720 mlxsw_sp_nexthop6_group_update(struct mlxsw_sp *mlxsw_sp,
6721 			       struct mlxsw_sp_fib6_entry *fib6_entry)
6722 {
6723 	struct mlxsw_sp_nexthop_group *old_nh_grp = fib6_entry->common.nh_group;
6724 	struct mlxsw_sp_fib_node *fib_node = fib6_entry->common.fib_node;
6725 	int err;
6726 
6727 	mlxsw_sp_nexthop_group_vr_unlink(old_nh_grp, fib_node->fib);
6728 	fib6_entry->common.nh_group = NULL;
6729 	list_del(&fib6_entry->common.nexthop_group_node);
6730 
6731 	err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
6732 	if (err)
6733 		goto err_nexthop6_group_get;
6734 
6735 	err = mlxsw_sp_nexthop_group_vr_link(fib6_entry->common.nh_group,
6736 					     fib_node->fib);
6737 	if (err)
6738 		goto err_nexthop_group_vr_link;
6739 
6740 	/* In case this entry is offloaded, then the adjacency index
6741 	 * currently associated with it in the device's table is that
6742 	 * of the old group. Start using the new one instead.
6743 	 */
6744 	err = mlxsw_sp_fib_entry_update(mlxsw_sp, &fib6_entry->common);
6745 	if (err)
6746 		goto err_fib_entry_update;
6747 
6748 	if (list_empty(&old_nh_grp->fib_list))
6749 		mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, old_nh_grp);
6750 
6751 	return 0;
6752 
6753 err_fib_entry_update:
6754 	mlxsw_sp_nexthop_group_vr_unlink(fib6_entry->common.nh_group,
6755 					 fib_node->fib);
6756 err_nexthop_group_vr_link:
6757 	mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
6758 err_nexthop6_group_get:
6759 	list_add_tail(&fib6_entry->common.nexthop_group_node,
6760 		      &old_nh_grp->fib_list);
6761 	fib6_entry->common.nh_group = old_nh_grp;
6762 	mlxsw_sp_nexthop_group_vr_link(old_nh_grp, fib_node->fib);
6763 	return err;
6764 }
6765 
6766 static int
6767 mlxsw_sp_fib6_entry_nexthop_add(struct mlxsw_sp *mlxsw_sp,
6768 				struct mlxsw_sp_fib6_entry *fib6_entry,
6769 				struct fib6_info **rt_arr, unsigned int nrt6)
6770 {
6771 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
6772 	int err, i;
6773 
6774 	for (i = 0; i < nrt6; i++) {
6775 		mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt_arr[i]);
6776 		if (IS_ERR(mlxsw_sp_rt6)) {
6777 			err = PTR_ERR(mlxsw_sp_rt6);
6778 			goto err_rt6_unwind;
6779 		}
6780 
6781 		list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
6782 		fib6_entry->nrt6++;
6783 	}
6784 
6785 	err = mlxsw_sp_nexthop6_group_update(mlxsw_sp, fib6_entry);
6786 	if (err)
6787 		goto err_rt6_unwind;
6788 
6789 	return 0;
6790 
6791 err_rt6_unwind:
6792 	for (; i > 0; i--) {
6793 		fib6_entry->nrt6--;
6794 		mlxsw_sp_rt6 = list_last_entry(&fib6_entry->rt6_list,
6795 					       struct mlxsw_sp_rt6, list);
6796 		list_del(&mlxsw_sp_rt6->list);
6797 		mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
6798 	}
6799 	return err;
6800 }
6801 
6802 static void
6803 mlxsw_sp_fib6_entry_nexthop_del(struct mlxsw_sp *mlxsw_sp,
6804 				struct mlxsw_sp_fib6_entry *fib6_entry,
6805 				struct fib6_info **rt_arr, unsigned int nrt6)
6806 {
6807 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
6808 	int i;
6809 
6810 	for (i = 0; i < nrt6; i++) {
6811 		mlxsw_sp_rt6 = mlxsw_sp_fib6_entry_rt_find(fib6_entry,
6812 							   rt_arr[i]);
6813 		if (WARN_ON_ONCE(!mlxsw_sp_rt6))
6814 			continue;
6815 
6816 		fib6_entry->nrt6--;
6817 		list_del(&mlxsw_sp_rt6->list);
6818 		mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
6819 	}
6820 
6821 	mlxsw_sp_nexthop6_group_update(mlxsw_sp, fib6_entry);
6822 }
6823 
6824 static int
6825 mlxsw_sp_fib6_entry_type_set_local(struct mlxsw_sp *mlxsw_sp,
6826 				   struct mlxsw_sp_fib_entry *fib_entry,
6827 				   const struct fib6_info *rt)
6828 {
6829 	struct mlxsw_sp_nexthop_group_info *nhgi = fib_entry->nh_group->nhgi;
6830 	union mlxsw_sp_l3addr dip = { .addr6 = rt->fib6_dst.addr };
6831 	u32 tb_id = mlxsw_sp_fix_tb_id(rt->fib6_table->tb6_id);
6832 	struct mlxsw_sp_router *router = mlxsw_sp->router;
6833 	int ifindex = nhgi->nexthops[0].ifindex;
6834 	struct mlxsw_sp_ipip_entry *ipip_entry;
6835 
6836 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
6837 	ipip_entry = mlxsw_sp_ipip_entry_find_by_decap(mlxsw_sp, ifindex,
6838 						       MLXSW_SP_L3_PROTO_IPV6,
6839 						       dip);
6840 
6841 	if (ipip_entry && ipip_entry->ol_dev->flags & IFF_UP) {
6842 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
6843 		return mlxsw_sp_fib_entry_decap_init(mlxsw_sp, fib_entry,
6844 						     ipip_entry);
6845 	}
6846 	if (mlxsw_sp_router_nve_is_decap(mlxsw_sp, tb_id,
6847 					 MLXSW_SP_L3_PROTO_IPV6, &dip)) {
6848 		u32 tunnel_index;
6849 
6850 		tunnel_index = router->nve_decap_config.tunnel_index;
6851 		fib_entry->decap.tunnel_index = tunnel_index;
6852 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
6853 	}
6854 
6855 	return 0;
6856 }
6857 
6858 static int mlxsw_sp_fib6_entry_type_set(struct mlxsw_sp *mlxsw_sp,
6859 					struct mlxsw_sp_fib_entry *fib_entry,
6860 					const struct fib6_info *rt)
6861 {
6862 	if (rt->fib6_flags & RTF_LOCAL)
6863 		return mlxsw_sp_fib6_entry_type_set_local(mlxsw_sp, fib_entry,
6864 							  rt);
6865 	if (rt->fib6_flags & RTF_ANYCAST)
6866 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
6867 	else if (rt->fib6_type == RTN_BLACKHOLE)
6868 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE;
6869 	else if (rt->fib6_flags & RTF_REJECT)
6870 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE;
6871 	else if (fib_entry->nh_group->nhgi->gateway)
6872 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
6873 	else
6874 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
6875 
6876 	return 0;
6877 }
6878 
6879 static void
6880 mlxsw_sp_fib6_entry_rt_destroy_all(struct mlxsw_sp_fib6_entry *fib6_entry)
6881 {
6882 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6, *tmp;
6883 
6884 	list_for_each_entry_safe(mlxsw_sp_rt6, tmp, &fib6_entry->rt6_list,
6885 				 list) {
6886 		fib6_entry->nrt6--;
6887 		list_del(&mlxsw_sp_rt6->list);
6888 		mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
6889 	}
6890 }
6891 
6892 static struct mlxsw_sp_fib6_entry *
6893 mlxsw_sp_fib6_entry_create(struct mlxsw_sp *mlxsw_sp,
6894 			   struct mlxsw_sp_fib_node *fib_node,
6895 			   struct fib6_info **rt_arr, unsigned int nrt6)
6896 {
6897 	struct mlxsw_sp_fib6_entry *fib6_entry;
6898 	struct mlxsw_sp_fib_entry *fib_entry;
6899 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
6900 	int err, i;
6901 
6902 	fib6_entry = kzalloc(sizeof(*fib6_entry), GFP_KERNEL);
6903 	if (!fib6_entry)
6904 		return ERR_PTR(-ENOMEM);
6905 	fib_entry = &fib6_entry->common;
6906 
6907 	INIT_LIST_HEAD(&fib6_entry->rt6_list);
6908 
6909 	for (i = 0; i < nrt6; i++) {
6910 		mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt_arr[i]);
6911 		if (IS_ERR(mlxsw_sp_rt6)) {
6912 			err = PTR_ERR(mlxsw_sp_rt6);
6913 			goto err_rt6_unwind;
6914 		}
6915 		list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
6916 		fib6_entry->nrt6++;
6917 	}
6918 
6919 	err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
6920 	if (err)
6921 		goto err_rt6_unwind;
6922 
6923 	err = mlxsw_sp_nexthop_group_vr_link(fib_entry->nh_group,
6924 					     fib_node->fib);
6925 	if (err)
6926 		goto err_nexthop_group_vr_link;
6927 
6928 	err = mlxsw_sp_fib6_entry_type_set(mlxsw_sp, fib_entry, rt_arr[0]);
6929 	if (err)
6930 		goto err_fib6_entry_type_set;
6931 
6932 	fib_entry->fib_node = fib_node;
6933 
6934 	return fib6_entry;
6935 
6936 err_fib6_entry_type_set:
6937 	mlxsw_sp_nexthop_group_vr_unlink(fib_entry->nh_group, fib_node->fib);
6938 err_nexthop_group_vr_link:
6939 	mlxsw_sp_nexthop6_group_put(mlxsw_sp, fib_entry);
6940 err_rt6_unwind:
6941 	for (; i > 0; i--) {
6942 		fib6_entry->nrt6--;
6943 		mlxsw_sp_rt6 = list_last_entry(&fib6_entry->rt6_list,
6944 					       struct mlxsw_sp_rt6, list);
6945 		list_del(&mlxsw_sp_rt6->list);
6946 		mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
6947 	}
6948 	kfree(fib6_entry);
6949 	return ERR_PTR(err);
6950 }
6951 
6952 static void
6953 mlxsw_sp_fib6_entry_type_unset(struct mlxsw_sp *mlxsw_sp,
6954 			       struct mlxsw_sp_fib6_entry *fib6_entry)
6955 {
6956 	mlxsw_sp_fib_entry_type_unset(mlxsw_sp, &fib6_entry->common);
6957 }
6958 
6959 static void mlxsw_sp_fib6_entry_destroy(struct mlxsw_sp *mlxsw_sp,
6960 					struct mlxsw_sp_fib6_entry *fib6_entry)
6961 {
6962 	struct mlxsw_sp_fib_node *fib_node = fib6_entry->common.fib_node;
6963 
6964 	mlxsw_sp_fib6_entry_type_unset(mlxsw_sp, fib6_entry);
6965 	mlxsw_sp_nexthop_group_vr_unlink(fib6_entry->common.nh_group,
6966 					 fib_node->fib);
6967 	mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
6968 	mlxsw_sp_fib6_entry_rt_destroy_all(fib6_entry);
6969 	WARN_ON(fib6_entry->nrt6);
6970 	kfree(fib6_entry);
6971 }
6972 
6973 static struct mlxsw_sp_fib6_entry *
6974 mlxsw_sp_fib6_entry_lookup(struct mlxsw_sp *mlxsw_sp,
6975 			   const struct fib6_info *rt)
6976 {
6977 	struct mlxsw_sp_fib6_entry *fib6_entry;
6978 	struct mlxsw_sp_fib_node *fib_node;
6979 	struct mlxsw_sp_fib *fib;
6980 	struct fib6_info *cmp_rt;
6981 	struct mlxsw_sp_vr *vr;
6982 
6983 	vr = mlxsw_sp_vr_find(mlxsw_sp, rt->fib6_table->tb6_id);
6984 	if (!vr)
6985 		return NULL;
6986 	fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV6);
6987 
6988 	fib_node = mlxsw_sp_fib_node_lookup(fib, &rt->fib6_dst.addr,
6989 					    sizeof(rt->fib6_dst.addr),
6990 					    rt->fib6_dst.plen);
6991 	if (!fib_node)
6992 		return NULL;
6993 
6994 	fib6_entry = container_of(fib_node->fib_entry,
6995 				  struct mlxsw_sp_fib6_entry, common);
6996 	cmp_rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
6997 	if (rt->fib6_table->tb6_id == cmp_rt->fib6_table->tb6_id &&
6998 	    rt->fib6_metric == cmp_rt->fib6_metric &&
6999 	    mlxsw_sp_fib6_entry_rt_find(fib6_entry, rt))
7000 		return fib6_entry;
7001 
7002 	return NULL;
7003 }
7004 
7005 static bool mlxsw_sp_fib6_allow_replace(struct mlxsw_sp_fib6_entry *fib6_entry)
7006 {
7007 	struct mlxsw_sp_fib_node *fib_node = fib6_entry->common.fib_node;
7008 	struct mlxsw_sp_fib6_entry *fib6_replaced;
7009 	struct fib6_info *rt, *rt_replaced;
7010 
7011 	if (!fib_node->fib_entry)
7012 		return true;
7013 
7014 	fib6_replaced = container_of(fib_node->fib_entry,
7015 				     struct mlxsw_sp_fib6_entry,
7016 				     common);
7017 	rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
7018 	rt_replaced = mlxsw_sp_fib6_entry_rt(fib6_replaced);
7019 	if (rt->fib6_table->tb6_id == RT_TABLE_MAIN &&
7020 	    rt_replaced->fib6_table->tb6_id == RT_TABLE_LOCAL)
7021 		return false;
7022 
7023 	return true;
7024 }
7025 
7026 static int mlxsw_sp_router_fib6_replace(struct mlxsw_sp *mlxsw_sp,
7027 					struct fib6_info **rt_arr,
7028 					unsigned int nrt6)
7029 {
7030 	struct mlxsw_sp_fib6_entry *fib6_entry, *fib6_replaced;
7031 	struct mlxsw_sp_fib_entry *replaced;
7032 	struct mlxsw_sp_fib_node *fib_node;
7033 	struct fib6_info *rt = rt_arr[0];
7034 	int err;
7035 
7036 	if (rt->fib6_src.plen)
7037 		return -EINVAL;
7038 
7039 	if (mlxsw_sp_fib6_rt_should_ignore(rt))
7040 		return 0;
7041 
7042 	if (rt->nh && !mlxsw_sp_nexthop_obj_group_lookup(mlxsw_sp, rt->nh->id))
7043 		return 0;
7044 
7045 	fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->fib6_table->tb6_id,
7046 					 &rt->fib6_dst.addr,
7047 					 sizeof(rt->fib6_dst.addr),
7048 					 rt->fib6_dst.plen,
7049 					 MLXSW_SP_L3_PROTO_IPV6);
7050 	if (IS_ERR(fib_node))
7051 		return PTR_ERR(fib_node);
7052 
7053 	fib6_entry = mlxsw_sp_fib6_entry_create(mlxsw_sp, fib_node, rt_arr,
7054 						nrt6);
7055 	if (IS_ERR(fib6_entry)) {
7056 		err = PTR_ERR(fib6_entry);
7057 		goto err_fib6_entry_create;
7058 	}
7059 
7060 	if (!mlxsw_sp_fib6_allow_replace(fib6_entry)) {
7061 		mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
7062 		mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
7063 		return 0;
7064 	}
7065 
7066 	replaced = fib_node->fib_entry;
7067 	err = mlxsw_sp_fib_node_entry_link(mlxsw_sp, &fib6_entry->common);
7068 	if (err)
7069 		goto err_fib_node_entry_link;
7070 
7071 	/* Nothing to replace */
7072 	if (!replaced)
7073 		return 0;
7074 
7075 	mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, replaced);
7076 	fib6_replaced = container_of(replaced, struct mlxsw_sp_fib6_entry,
7077 				     common);
7078 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_replaced);
7079 
7080 	return 0;
7081 
7082 err_fib_node_entry_link:
7083 	fib_node->fib_entry = replaced;
7084 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
7085 err_fib6_entry_create:
7086 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
7087 	return err;
7088 }
7089 
7090 static int mlxsw_sp_router_fib6_append(struct mlxsw_sp *mlxsw_sp,
7091 				       struct fib6_info **rt_arr,
7092 				       unsigned int nrt6)
7093 {
7094 	struct mlxsw_sp_fib6_entry *fib6_entry;
7095 	struct mlxsw_sp_fib_node *fib_node;
7096 	struct fib6_info *rt = rt_arr[0];
7097 	int err;
7098 
7099 	if (rt->fib6_src.plen)
7100 		return -EINVAL;
7101 
7102 	if (mlxsw_sp_fib6_rt_should_ignore(rt))
7103 		return 0;
7104 
7105 	fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->fib6_table->tb6_id,
7106 					 &rt->fib6_dst.addr,
7107 					 sizeof(rt->fib6_dst.addr),
7108 					 rt->fib6_dst.plen,
7109 					 MLXSW_SP_L3_PROTO_IPV6);
7110 	if (IS_ERR(fib_node))
7111 		return PTR_ERR(fib_node);
7112 
7113 	if (WARN_ON_ONCE(!fib_node->fib_entry)) {
7114 		mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
7115 		return -EINVAL;
7116 	}
7117 
7118 	fib6_entry = container_of(fib_node->fib_entry,
7119 				  struct mlxsw_sp_fib6_entry, common);
7120 	err = mlxsw_sp_fib6_entry_nexthop_add(mlxsw_sp, fib6_entry, rt_arr,
7121 					      nrt6);
7122 	if (err)
7123 		goto err_fib6_entry_nexthop_add;
7124 
7125 	return 0;
7126 
7127 err_fib6_entry_nexthop_add:
7128 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
7129 	return err;
7130 }
7131 
7132 static void mlxsw_sp_router_fib6_del(struct mlxsw_sp *mlxsw_sp,
7133 				     struct fib6_info **rt_arr,
7134 				     unsigned int nrt6)
7135 {
7136 	struct mlxsw_sp_fib6_entry *fib6_entry;
7137 	struct mlxsw_sp_fib_node *fib_node;
7138 	struct fib6_info *rt = rt_arr[0];
7139 
7140 	if (mlxsw_sp_fib6_rt_should_ignore(rt))
7141 		return;
7142 
7143 	/* Multipath routes are first added to the FIB trie and only then
7144 	 * notified. If we vetoed the addition, we will get a delete
7145 	 * notification for a route we do not have. Therefore, do not warn if
7146 	 * route was not found.
7147 	 */
7148 	fib6_entry = mlxsw_sp_fib6_entry_lookup(mlxsw_sp, rt);
7149 	if (!fib6_entry)
7150 		return;
7151 
7152 	/* If not all the nexthops are deleted, then only reduce the nexthop
7153 	 * group.
7154 	 */
7155 	if (nrt6 != fib6_entry->nrt6) {
7156 		mlxsw_sp_fib6_entry_nexthop_del(mlxsw_sp, fib6_entry, rt_arr,
7157 						nrt6);
7158 		return;
7159 	}
7160 
7161 	fib_node = fib6_entry->common.fib_node;
7162 
7163 	mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, &fib6_entry->common);
7164 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
7165 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
7166 }
7167 
7168 static struct mlxsw_sp_mr_table *
7169 mlxsw_sp_router_fibmr_family_to_table(struct mlxsw_sp_vr *vr, int family)
7170 {
7171 	if (family == RTNL_FAMILY_IPMR)
7172 		return vr->mr_table[MLXSW_SP_L3_PROTO_IPV4];
7173 	else
7174 		return vr->mr_table[MLXSW_SP_L3_PROTO_IPV6];
7175 }
7176 
7177 static int mlxsw_sp_router_fibmr_add(struct mlxsw_sp *mlxsw_sp,
7178 				     struct mfc_entry_notifier_info *men_info,
7179 				     bool replace)
7180 {
7181 	struct mlxsw_sp_mr_table *mrt;
7182 	struct mlxsw_sp_vr *vr;
7183 
7184 	vr = mlxsw_sp_vr_get(mlxsw_sp, men_info->tb_id, NULL);
7185 	if (IS_ERR(vr))
7186 		return PTR_ERR(vr);
7187 
7188 	mrt = mlxsw_sp_router_fibmr_family_to_table(vr, men_info->info.family);
7189 	return mlxsw_sp_mr_route_add(mrt, men_info->mfc, replace);
7190 }
7191 
7192 static void mlxsw_sp_router_fibmr_del(struct mlxsw_sp *mlxsw_sp,
7193 				      struct mfc_entry_notifier_info *men_info)
7194 {
7195 	struct mlxsw_sp_mr_table *mrt;
7196 	struct mlxsw_sp_vr *vr;
7197 
7198 	vr = mlxsw_sp_vr_find(mlxsw_sp, men_info->tb_id);
7199 	if (WARN_ON(!vr))
7200 		return;
7201 
7202 	mrt = mlxsw_sp_router_fibmr_family_to_table(vr, men_info->info.family);
7203 	mlxsw_sp_mr_route_del(mrt, men_info->mfc);
7204 	mlxsw_sp_vr_put(mlxsw_sp, vr);
7205 }
7206 
7207 static int
7208 mlxsw_sp_router_fibmr_vif_add(struct mlxsw_sp *mlxsw_sp,
7209 			      struct vif_entry_notifier_info *ven_info)
7210 {
7211 	struct mlxsw_sp_mr_table *mrt;
7212 	struct mlxsw_sp_rif *rif;
7213 	struct mlxsw_sp_vr *vr;
7214 
7215 	vr = mlxsw_sp_vr_get(mlxsw_sp, ven_info->tb_id, NULL);
7216 	if (IS_ERR(vr))
7217 		return PTR_ERR(vr);
7218 
7219 	mrt = mlxsw_sp_router_fibmr_family_to_table(vr, ven_info->info.family);
7220 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, ven_info->dev);
7221 	return mlxsw_sp_mr_vif_add(mrt, ven_info->dev,
7222 				   ven_info->vif_index,
7223 				   ven_info->vif_flags, rif);
7224 }
7225 
7226 static void
7227 mlxsw_sp_router_fibmr_vif_del(struct mlxsw_sp *mlxsw_sp,
7228 			      struct vif_entry_notifier_info *ven_info)
7229 {
7230 	struct mlxsw_sp_mr_table *mrt;
7231 	struct mlxsw_sp_vr *vr;
7232 
7233 	vr = mlxsw_sp_vr_find(mlxsw_sp, ven_info->tb_id);
7234 	if (WARN_ON(!vr))
7235 		return;
7236 
7237 	mrt = mlxsw_sp_router_fibmr_family_to_table(vr, ven_info->info.family);
7238 	mlxsw_sp_mr_vif_del(mrt, ven_info->vif_index);
7239 	mlxsw_sp_vr_put(mlxsw_sp, vr);
7240 }
7241 
7242 static void mlxsw_sp_fib4_node_flush(struct mlxsw_sp *mlxsw_sp,
7243 				     struct mlxsw_sp_fib_node *fib_node)
7244 {
7245 	struct mlxsw_sp_fib4_entry *fib4_entry;
7246 
7247 	fib4_entry = container_of(fib_node->fib_entry,
7248 				  struct mlxsw_sp_fib4_entry, common);
7249 	mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, fib_node->fib_entry);
7250 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
7251 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
7252 }
7253 
7254 static void mlxsw_sp_fib6_node_flush(struct mlxsw_sp *mlxsw_sp,
7255 				     struct mlxsw_sp_fib_node *fib_node)
7256 {
7257 	struct mlxsw_sp_fib6_entry *fib6_entry;
7258 
7259 	fib6_entry = container_of(fib_node->fib_entry,
7260 				  struct mlxsw_sp_fib6_entry, common);
7261 	mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, fib_node->fib_entry);
7262 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
7263 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
7264 }
7265 
7266 static void mlxsw_sp_fib_node_flush(struct mlxsw_sp *mlxsw_sp,
7267 				    struct mlxsw_sp_fib_node *fib_node)
7268 {
7269 	switch (fib_node->fib->proto) {
7270 	case MLXSW_SP_L3_PROTO_IPV4:
7271 		mlxsw_sp_fib4_node_flush(mlxsw_sp, fib_node);
7272 		break;
7273 	case MLXSW_SP_L3_PROTO_IPV6:
7274 		mlxsw_sp_fib6_node_flush(mlxsw_sp, fib_node);
7275 		break;
7276 	}
7277 }
7278 
7279 static void mlxsw_sp_vr_fib_flush(struct mlxsw_sp *mlxsw_sp,
7280 				  struct mlxsw_sp_vr *vr,
7281 				  enum mlxsw_sp_l3proto proto)
7282 {
7283 	struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
7284 	struct mlxsw_sp_fib_node *fib_node, *tmp;
7285 
7286 	list_for_each_entry_safe(fib_node, tmp, &fib->node_list, list) {
7287 		bool do_break = &tmp->list == &fib->node_list;
7288 
7289 		mlxsw_sp_fib_node_flush(mlxsw_sp, fib_node);
7290 		if (do_break)
7291 			break;
7292 	}
7293 }
7294 
7295 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp)
7296 {
7297 	int i, j;
7298 
7299 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
7300 		struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
7301 
7302 		if (!mlxsw_sp_vr_is_used(vr))
7303 			continue;
7304 
7305 		for (j = 0; j < MLXSW_SP_L3_PROTO_MAX; j++)
7306 			mlxsw_sp_mr_table_flush(vr->mr_table[j]);
7307 		mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV4);
7308 
7309 		/* If virtual router was only used for IPv4, then it's no
7310 		 * longer used.
7311 		 */
7312 		if (!mlxsw_sp_vr_is_used(vr))
7313 			continue;
7314 		mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV6);
7315 	}
7316 }
7317 
7318 struct mlxsw_sp_fib6_event_work {
7319 	struct fib6_info **rt_arr;
7320 	unsigned int nrt6;
7321 };
7322 
7323 struct mlxsw_sp_fib_event_work {
7324 	struct work_struct work;
7325 	union {
7326 		struct mlxsw_sp_fib6_event_work fib6_work;
7327 		struct fib_entry_notifier_info fen_info;
7328 		struct fib_rule_notifier_info fr_info;
7329 		struct fib_nh_notifier_info fnh_info;
7330 		struct mfc_entry_notifier_info men_info;
7331 		struct vif_entry_notifier_info ven_info;
7332 	};
7333 	struct mlxsw_sp *mlxsw_sp;
7334 	unsigned long event;
7335 };
7336 
7337 static int
7338 mlxsw_sp_router_fib6_work_init(struct mlxsw_sp_fib6_event_work *fib6_work,
7339 			       struct fib6_entry_notifier_info *fen6_info)
7340 {
7341 	struct fib6_info *rt = fen6_info->rt;
7342 	struct fib6_info **rt_arr;
7343 	struct fib6_info *iter;
7344 	unsigned int nrt6;
7345 	int i = 0;
7346 
7347 	nrt6 = fen6_info->nsiblings + 1;
7348 
7349 	rt_arr = kcalloc(nrt6, sizeof(struct fib6_info *), GFP_ATOMIC);
7350 	if (!rt_arr)
7351 		return -ENOMEM;
7352 
7353 	fib6_work->rt_arr = rt_arr;
7354 	fib6_work->nrt6 = nrt6;
7355 
7356 	rt_arr[0] = rt;
7357 	fib6_info_hold(rt);
7358 
7359 	if (!fen6_info->nsiblings)
7360 		return 0;
7361 
7362 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
7363 		if (i == fen6_info->nsiblings)
7364 			break;
7365 
7366 		rt_arr[i + 1] = iter;
7367 		fib6_info_hold(iter);
7368 		i++;
7369 	}
7370 	WARN_ON_ONCE(i != fen6_info->nsiblings);
7371 
7372 	return 0;
7373 }
7374 
7375 static void
7376 mlxsw_sp_router_fib6_work_fini(struct mlxsw_sp_fib6_event_work *fib6_work)
7377 {
7378 	int i;
7379 
7380 	for (i = 0; i < fib6_work->nrt6; i++)
7381 		mlxsw_sp_rt6_release(fib6_work->rt_arr[i]);
7382 	kfree(fib6_work->rt_arr);
7383 }
7384 
7385 static void mlxsw_sp_router_fib4_event_work(struct work_struct *work)
7386 {
7387 	struct mlxsw_sp_fib_event_work *fib_work =
7388 		container_of(work, struct mlxsw_sp_fib_event_work, work);
7389 	struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
7390 	int err;
7391 
7392 	mutex_lock(&mlxsw_sp->router->lock);
7393 	mlxsw_sp_span_respin(mlxsw_sp);
7394 
7395 	switch (fib_work->event) {
7396 	case FIB_EVENT_ENTRY_REPLACE:
7397 		err = mlxsw_sp_router_fib4_replace(mlxsw_sp,
7398 						   &fib_work->fen_info);
7399 		if (err) {
7400 			dev_warn(mlxsw_sp->bus_info->dev, "FIB replace failed.\n");
7401 			mlxsw_sp_fib4_offload_failed_flag_set(mlxsw_sp,
7402 							      &fib_work->fen_info);
7403 		}
7404 		fib_info_put(fib_work->fen_info.fi);
7405 		break;
7406 	case FIB_EVENT_ENTRY_DEL:
7407 		mlxsw_sp_router_fib4_del(mlxsw_sp, &fib_work->fen_info);
7408 		fib_info_put(fib_work->fen_info.fi);
7409 		break;
7410 	case FIB_EVENT_NH_ADD:
7411 	case FIB_EVENT_NH_DEL:
7412 		mlxsw_sp_nexthop4_event(mlxsw_sp, fib_work->event,
7413 					fib_work->fnh_info.fib_nh);
7414 		fib_info_put(fib_work->fnh_info.fib_nh->nh_parent);
7415 		break;
7416 	}
7417 	mutex_unlock(&mlxsw_sp->router->lock);
7418 	kfree(fib_work);
7419 }
7420 
7421 static void mlxsw_sp_router_fib6_event_work(struct work_struct *work)
7422 {
7423 	struct mlxsw_sp_fib_event_work *fib_work =
7424 		    container_of(work, struct mlxsw_sp_fib_event_work, work);
7425 	struct mlxsw_sp_fib6_event_work *fib6_work = &fib_work->fib6_work;
7426 	struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
7427 	int err;
7428 
7429 	mutex_lock(&mlxsw_sp->router->lock);
7430 	mlxsw_sp_span_respin(mlxsw_sp);
7431 
7432 	switch (fib_work->event) {
7433 	case FIB_EVENT_ENTRY_REPLACE:
7434 		err = mlxsw_sp_router_fib6_replace(mlxsw_sp,
7435 						   fib6_work->rt_arr,
7436 						   fib6_work->nrt6);
7437 		if (err) {
7438 			dev_warn(mlxsw_sp->bus_info->dev, "FIB replace failed.\n");
7439 			mlxsw_sp_fib6_offload_failed_flag_set(mlxsw_sp,
7440 							      fib6_work->rt_arr,
7441 							      fib6_work->nrt6);
7442 		}
7443 		mlxsw_sp_router_fib6_work_fini(fib6_work);
7444 		break;
7445 	case FIB_EVENT_ENTRY_APPEND:
7446 		err = mlxsw_sp_router_fib6_append(mlxsw_sp,
7447 						  fib6_work->rt_arr,
7448 						  fib6_work->nrt6);
7449 		if (err) {
7450 			dev_warn(mlxsw_sp->bus_info->dev, "FIB append failed.\n");
7451 			mlxsw_sp_fib6_offload_failed_flag_set(mlxsw_sp,
7452 							      fib6_work->rt_arr,
7453 							      fib6_work->nrt6);
7454 		}
7455 		mlxsw_sp_router_fib6_work_fini(fib6_work);
7456 		break;
7457 	case FIB_EVENT_ENTRY_DEL:
7458 		mlxsw_sp_router_fib6_del(mlxsw_sp,
7459 					 fib6_work->rt_arr,
7460 					 fib6_work->nrt6);
7461 		mlxsw_sp_router_fib6_work_fini(fib6_work);
7462 		break;
7463 	}
7464 	mutex_unlock(&mlxsw_sp->router->lock);
7465 	kfree(fib_work);
7466 }
7467 
7468 static void mlxsw_sp_router_fibmr_event_work(struct work_struct *work)
7469 {
7470 	struct mlxsw_sp_fib_event_work *fib_work =
7471 		container_of(work, struct mlxsw_sp_fib_event_work, work);
7472 	struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
7473 	bool replace;
7474 	int err;
7475 
7476 	rtnl_lock();
7477 	mutex_lock(&mlxsw_sp->router->lock);
7478 	switch (fib_work->event) {
7479 	case FIB_EVENT_ENTRY_REPLACE:
7480 	case FIB_EVENT_ENTRY_ADD:
7481 		replace = fib_work->event == FIB_EVENT_ENTRY_REPLACE;
7482 
7483 		err = mlxsw_sp_router_fibmr_add(mlxsw_sp, &fib_work->men_info,
7484 						replace);
7485 		if (err)
7486 			dev_warn(mlxsw_sp->bus_info->dev, "MR entry add failed.\n");
7487 		mr_cache_put(fib_work->men_info.mfc);
7488 		break;
7489 	case FIB_EVENT_ENTRY_DEL:
7490 		mlxsw_sp_router_fibmr_del(mlxsw_sp, &fib_work->men_info);
7491 		mr_cache_put(fib_work->men_info.mfc);
7492 		break;
7493 	case FIB_EVENT_VIF_ADD:
7494 		err = mlxsw_sp_router_fibmr_vif_add(mlxsw_sp,
7495 						    &fib_work->ven_info);
7496 		if (err)
7497 			dev_warn(mlxsw_sp->bus_info->dev, "MR VIF add failed.\n");
7498 		dev_put(fib_work->ven_info.dev);
7499 		break;
7500 	case FIB_EVENT_VIF_DEL:
7501 		mlxsw_sp_router_fibmr_vif_del(mlxsw_sp,
7502 					      &fib_work->ven_info);
7503 		dev_put(fib_work->ven_info.dev);
7504 		break;
7505 	}
7506 	mutex_unlock(&mlxsw_sp->router->lock);
7507 	rtnl_unlock();
7508 	kfree(fib_work);
7509 }
7510 
7511 static void mlxsw_sp_router_fib4_event(struct mlxsw_sp_fib_event_work *fib_work,
7512 				       struct fib_notifier_info *info)
7513 {
7514 	struct fib_entry_notifier_info *fen_info;
7515 	struct fib_nh_notifier_info *fnh_info;
7516 
7517 	switch (fib_work->event) {
7518 	case FIB_EVENT_ENTRY_REPLACE:
7519 	case FIB_EVENT_ENTRY_DEL:
7520 		fen_info = container_of(info, struct fib_entry_notifier_info,
7521 					info);
7522 		fib_work->fen_info = *fen_info;
7523 		/* Take reference on fib_info to prevent it from being
7524 		 * freed while work is queued. Release it afterwards.
7525 		 */
7526 		fib_info_hold(fib_work->fen_info.fi);
7527 		break;
7528 	case FIB_EVENT_NH_ADD:
7529 	case FIB_EVENT_NH_DEL:
7530 		fnh_info = container_of(info, struct fib_nh_notifier_info,
7531 					info);
7532 		fib_work->fnh_info = *fnh_info;
7533 		fib_info_hold(fib_work->fnh_info.fib_nh->nh_parent);
7534 		break;
7535 	}
7536 }
7537 
7538 static int mlxsw_sp_router_fib6_event(struct mlxsw_sp_fib_event_work *fib_work,
7539 				      struct fib_notifier_info *info)
7540 {
7541 	struct fib6_entry_notifier_info *fen6_info;
7542 	int err;
7543 
7544 	switch (fib_work->event) {
7545 	case FIB_EVENT_ENTRY_REPLACE:
7546 	case FIB_EVENT_ENTRY_APPEND:
7547 	case FIB_EVENT_ENTRY_DEL:
7548 		fen6_info = container_of(info, struct fib6_entry_notifier_info,
7549 					 info);
7550 		err = mlxsw_sp_router_fib6_work_init(&fib_work->fib6_work,
7551 						     fen6_info);
7552 		if (err)
7553 			return err;
7554 		break;
7555 	}
7556 
7557 	return 0;
7558 }
7559 
7560 static void
7561 mlxsw_sp_router_fibmr_event(struct mlxsw_sp_fib_event_work *fib_work,
7562 			    struct fib_notifier_info *info)
7563 {
7564 	switch (fib_work->event) {
7565 	case FIB_EVENT_ENTRY_REPLACE:
7566 	case FIB_EVENT_ENTRY_ADD:
7567 	case FIB_EVENT_ENTRY_DEL:
7568 		memcpy(&fib_work->men_info, info, sizeof(fib_work->men_info));
7569 		mr_cache_hold(fib_work->men_info.mfc);
7570 		break;
7571 	case FIB_EVENT_VIF_ADD:
7572 	case FIB_EVENT_VIF_DEL:
7573 		memcpy(&fib_work->ven_info, info, sizeof(fib_work->ven_info));
7574 		dev_hold(fib_work->ven_info.dev);
7575 		break;
7576 	}
7577 }
7578 
7579 static int mlxsw_sp_router_fib_rule_event(unsigned long event,
7580 					  struct fib_notifier_info *info,
7581 					  struct mlxsw_sp *mlxsw_sp)
7582 {
7583 	struct netlink_ext_ack *extack = info->extack;
7584 	struct fib_rule_notifier_info *fr_info;
7585 	struct fib_rule *rule;
7586 	int err = 0;
7587 
7588 	/* nothing to do at the moment */
7589 	if (event == FIB_EVENT_RULE_DEL)
7590 		return 0;
7591 
7592 	fr_info = container_of(info, struct fib_rule_notifier_info, info);
7593 	rule = fr_info->rule;
7594 
7595 	/* Rule only affects locally generated traffic */
7596 	if (rule->iifindex == mlxsw_sp_net(mlxsw_sp)->loopback_dev->ifindex)
7597 		return 0;
7598 
7599 	switch (info->family) {
7600 	case AF_INET:
7601 		if (!fib4_rule_default(rule) && !rule->l3mdev)
7602 			err = -EOPNOTSUPP;
7603 		break;
7604 	case AF_INET6:
7605 		if (!fib6_rule_default(rule) && !rule->l3mdev)
7606 			err = -EOPNOTSUPP;
7607 		break;
7608 	case RTNL_FAMILY_IPMR:
7609 		if (!ipmr_rule_default(rule) && !rule->l3mdev)
7610 			err = -EOPNOTSUPP;
7611 		break;
7612 	case RTNL_FAMILY_IP6MR:
7613 		if (!ip6mr_rule_default(rule) && !rule->l3mdev)
7614 			err = -EOPNOTSUPP;
7615 		break;
7616 	}
7617 
7618 	if (err < 0)
7619 		NL_SET_ERR_MSG_MOD(extack, "FIB rules not supported");
7620 
7621 	return err;
7622 }
7623 
7624 /* Called with rcu_read_lock() */
7625 static int mlxsw_sp_router_fib_event(struct notifier_block *nb,
7626 				     unsigned long event, void *ptr)
7627 {
7628 	struct mlxsw_sp_fib_event_work *fib_work;
7629 	struct fib_notifier_info *info = ptr;
7630 	struct mlxsw_sp_router *router;
7631 	int err;
7632 
7633 	if ((info->family != AF_INET && info->family != AF_INET6 &&
7634 	     info->family != RTNL_FAMILY_IPMR &&
7635 	     info->family != RTNL_FAMILY_IP6MR))
7636 		return NOTIFY_DONE;
7637 
7638 	router = container_of(nb, struct mlxsw_sp_router, fib_nb);
7639 
7640 	switch (event) {
7641 	case FIB_EVENT_RULE_ADD:
7642 	case FIB_EVENT_RULE_DEL:
7643 		err = mlxsw_sp_router_fib_rule_event(event, info,
7644 						     router->mlxsw_sp);
7645 		return notifier_from_errno(err);
7646 	case FIB_EVENT_ENTRY_ADD:
7647 	case FIB_EVENT_ENTRY_REPLACE:
7648 	case FIB_EVENT_ENTRY_APPEND:
7649 		if (info->family == AF_INET) {
7650 			struct fib_entry_notifier_info *fen_info = ptr;
7651 
7652 			if (fen_info->fi->fib_nh_is_v6) {
7653 				NL_SET_ERR_MSG_MOD(info->extack, "IPv6 gateway with IPv4 route is not supported");
7654 				return notifier_from_errno(-EINVAL);
7655 			}
7656 		}
7657 		break;
7658 	}
7659 
7660 	fib_work = kzalloc(sizeof(*fib_work), GFP_ATOMIC);
7661 	if (!fib_work)
7662 		return NOTIFY_BAD;
7663 
7664 	fib_work->mlxsw_sp = router->mlxsw_sp;
7665 	fib_work->event = event;
7666 
7667 	switch (info->family) {
7668 	case AF_INET:
7669 		INIT_WORK(&fib_work->work, mlxsw_sp_router_fib4_event_work);
7670 		mlxsw_sp_router_fib4_event(fib_work, info);
7671 		break;
7672 	case AF_INET6:
7673 		INIT_WORK(&fib_work->work, mlxsw_sp_router_fib6_event_work);
7674 		err = mlxsw_sp_router_fib6_event(fib_work, info);
7675 		if (err)
7676 			goto err_fib_event;
7677 		break;
7678 	case RTNL_FAMILY_IP6MR:
7679 	case RTNL_FAMILY_IPMR:
7680 		INIT_WORK(&fib_work->work, mlxsw_sp_router_fibmr_event_work);
7681 		mlxsw_sp_router_fibmr_event(fib_work, info);
7682 		break;
7683 	}
7684 
7685 	mlxsw_core_schedule_work(&fib_work->work);
7686 
7687 	return NOTIFY_DONE;
7688 
7689 err_fib_event:
7690 	kfree(fib_work);
7691 	return NOTIFY_BAD;
7692 }
7693 
7694 static struct mlxsw_sp_rif *
7695 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
7696 			 const struct net_device *dev)
7697 {
7698 	int i;
7699 
7700 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
7701 		if (mlxsw_sp->router->rifs[i] &&
7702 		    mlxsw_sp->router->rifs[i]->dev == dev)
7703 			return mlxsw_sp->router->rifs[i];
7704 
7705 	return NULL;
7706 }
7707 
7708 bool mlxsw_sp_rif_exists(struct mlxsw_sp *mlxsw_sp,
7709 			 const struct net_device *dev)
7710 {
7711 	struct mlxsw_sp_rif *rif;
7712 
7713 	mutex_lock(&mlxsw_sp->router->lock);
7714 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7715 	mutex_unlock(&mlxsw_sp->router->lock);
7716 
7717 	return rif;
7718 }
7719 
7720 u16 mlxsw_sp_rif_vid(struct mlxsw_sp *mlxsw_sp, const struct net_device *dev)
7721 {
7722 	struct mlxsw_sp_rif *rif;
7723 	u16 vid = 0;
7724 
7725 	mutex_lock(&mlxsw_sp->router->lock);
7726 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7727 	if (!rif)
7728 		goto out;
7729 
7730 	/* We only return the VID for VLAN RIFs. Otherwise we return an
7731 	 * invalid value (0).
7732 	 */
7733 	if (rif->ops->type != MLXSW_SP_RIF_TYPE_VLAN)
7734 		goto out;
7735 
7736 	vid = mlxsw_sp_fid_8021q_vid(rif->fid);
7737 
7738 out:
7739 	mutex_unlock(&mlxsw_sp->router->lock);
7740 	return vid;
7741 }
7742 
7743 static int mlxsw_sp_router_rif_disable(struct mlxsw_sp *mlxsw_sp, u16 rif)
7744 {
7745 	char ritr_pl[MLXSW_REG_RITR_LEN];
7746 	int err;
7747 
7748 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif);
7749 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7750 	if (err)
7751 		return err;
7752 
7753 	mlxsw_reg_ritr_enable_set(ritr_pl, false);
7754 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7755 }
7756 
7757 static void mlxsw_sp_router_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
7758 					  struct mlxsw_sp_rif *rif)
7759 {
7760 	mlxsw_sp_router_rif_disable(mlxsw_sp, rif->rif_index);
7761 	mlxsw_sp_nexthop_rif_gone_sync(mlxsw_sp, rif);
7762 	mlxsw_sp_neigh_rif_gone_sync(mlxsw_sp, rif);
7763 }
7764 
7765 static bool
7766 mlxsw_sp_rif_should_config(struct mlxsw_sp_rif *rif, struct net_device *dev,
7767 			   unsigned long event)
7768 {
7769 	struct inet6_dev *inet6_dev;
7770 	bool addr_list_empty = true;
7771 	struct in_device *idev;
7772 
7773 	switch (event) {
7774 	case NETDEV_UP:
7775 		return rif == NULL;
7776 	case NETDEV_DOWN:
7777 		rcu_read_lock();
7778 		idev = __in_dev_get_rcu(dev);
7779 		if (idev && idev->ifa_list)
7780 			addr_list_empty = false;
7781 
7782 		inet6_dev = __in6_dev_get(dev);
7783 		if (addr_list_empty && inet6_dev &&
7784 		    !list_empty(&inet6_dev->addr_list))
7785 			addr_list_empty = false;
7786 		rcu_read_unlock();
7787 
7788 		/* macvlans do not have a RIF, but rather piggy back on the
7789 		 * RIF of their lower device.
7790 		 */
7791 		if (netif_is_macvlan(dev) && addr_list_empty)
7792 			return true;
7793 
7794 		if (rif && addr_list_empty &&
7795 		    !netif_is_l3_slave(rif->dev))
7796 			return true;
7797 		/* It is possible we already removed the RIF ourselves
7798 		 * if it was assigned to a netdev that is now a bridge
7799 		 * or LAG slave.
7800 		 */
7801 		return false;
7802 	}
7803 
7804 	return false;
7805 }
7806 
7807 static enum mlxsw_sp_rif_type
7808 mlxsw_sp_dev_rif_type(const struct mlxsw_sp *mlxsw_sp,
7809 		      const struct net_device *dev)
7810 {
7811 	enum mlxsw_sp_fid_type type;
7812 
7813 	if (mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL))
7814 		return MLXSW_SP_RIF_TYPE_IPIP_LB;
7815 
7816 	/* Otherwise RIF type is derived from the type of the underlying FID. */
7817 	if (is_vlan_dev(dev) && netif_is_bridge_master(vlan_dev_real_dev(dev)))
7818 		type = MLXSW_SP_FID_TYPE_8021Q;
7819 	else if (netif_is_bridge_master(dev) && br_vlan_enabled(dev))
7820 		type = MLXSW_SP_FID_TYPE_8021Q;
7821 	else if (netif_is_bridge_master(dev))
7822 		type = MLXSW_SP_FID_TYPE_8021D;
7823 	else
7824 		type = MLXSW_SP_FID_TYPE_RFID;
7825 
7826 	return mlxsw_sp_fid_type_rif_type(mlxsw_sp, type);
7827 }
7828 
7829 static int mlxsw_sp_rif_index_alloc(struct mlxsw_sp *mlxsw_sp, u16 *p_rif_index)
7830 {
7831 	int i;
7832 
7833 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++) {
7834 		if (!mlxsw_sp->router->rifs[i]) {
7835 			*p_rif_index = i;
7836 			return 0;
7837 		}
7838 	}
7839 
7840 	return -ENOBUFS;
7841 }
7842 
7843 static struct mlxsw_sp_rif *mlxsw_sp_rif_alloc(size_t rif_size, u16 rif_index,
7844 					       u16 vr_id,
7845 					       struct net_device *l3_dev)
7846 {
7847 	struct mlxsw_sp_rif *rif;
7848 
7849 	rif = kzalloc(rif_size, GFP_KERNEL);
7850 	if (!rif)
7851 		return NULL;
7852 
7853 	INIT_LIST_HEAD(&rif->nexthop_list);
7854 	INIT_LIST_HEAD(&rif->neigh_list);
7855 	if (l3_dev) {
7856 		ether_addr_copy(rif->addr, l3_dev->dev_addr);
7857 		rif->mtu = l3_dev->mtu;
7858 		rif->dev = l3_dev;
7859 	}
7860 	rif->vr_id = vr_id;
7861 	rif->rif_index = rif_index;
7862 
7863 	return rif;
7864 }
7865 
7866 struct mlxsw_sp_rif *mlxsw_sp_rif_by_index(const struct mlxsw_sp *mlxsw_sp,
7867 					   u16 rif_index)
7868 {
7869 	return mlxsw_sp->router->rifs[rif_index];
7870 }
7871 
7872 u16 mlxsw_sp_rif_index(const struct mlxsw_sp_rif *rif)
7873 {
7874 	return rif->rif_index;
7875 }
7876 
7877 u16 mlxsw_sp_ipip_lb_rif_index(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
7878 {
7879 	return lb_rif->common.rif_index;
7880 }
7881 
7882 u16 mlxsw_sp_ipip_lb_ul_vr_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
7883 {
7884 	u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(lb_rif->common.dev);
7885 	struct mlxsw_sp_vr *ul_vr;
7886 
7887 	ul_vr = mlxsw_sp_vr_get(lb_rif->common.mlxsw_sp, ul_tb_id, NULL);
7888 	if (WARN_ON(IS_ERR(ul_vr)))
7889 		return 0;
7890 
7891 	return ul_vr->id;
7892 }
7893 
7894 u16 mlxsw_sp_ipip_lb_ul_rif_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
7895 {
7896 	return lb_rif->ul_rif_id;
7897 }
7898 
7899 static bool
7900 mlxsw_sp_router_port_l3_stats_enabled(struct mlxsw_sp_rif *rif)
7901 {
7902 	return mlxsw_sp_rif_counter_valid_get(rif,
7903 					      MLXSW_SP_RIF_COUNTER_EGRESS) &&
7904 	       mlxsw_sp_rif_counter_valid_get(rif,
7905 					      MLXSW_SP_RIF_COUNTER_INGRESS);
7906 }
7907 
7908 static int
7909 mlxsw_sp_router_port_l3_stats_enable(struct mlxsw_sp_rif *rif)
7910 {
7911 	int err;
7912 
7913 	err = mlxsw_sp_rif_counter_alloc(rif, MLXSW_SP_RIF_COUNTER_INGRESS);
7914 	if (err)
7915 		return err;
7916 
7917 	/* Clear stale data. */
7918 	err = mlxsw_sp_rif_counter_fetch_clear(rif,
7919 					       MLXSW_SP_RIF_COUNTER_INGRESS,
7920 					       NULL);
7921 	if (err)
7922 		goto err_clear_ingress;
7923 
7924 	err = mlxsw_sp_rif_counter_alloc(rif, MLXSW_SP_RIF_COUNTER_EGRESS);
7925 	if (err)
7926 		goto err_alloc_egress;
7927 
7928 	/* Clear stale data. */
7929 	err = mlxsw_sp_rif_counter_fetch_clear(rif,
7930 					       MLXSW_SP_RIF_COUNTER_EGRESS,
7931 					       NULL);
7932 	if (err)
7933 		goto err_clear_egress;
7934 
7935 	return 0;
7936 
7937 err_clear_egress:
7938 	mlxsw_sp_rif_counter_free(rif, MLXSW_SP_RIF_COUNTER_EGRESS);
7939 err_alloc_egress:
7940 err_clear_ingress:
7941 	mlxsw_sp_rif_counter_free(rif, MLXSW_SP_RIF_COUNTER_INGRESS);
7942 	return err;
7943 }
7944 
7945 static void
7946 mlxsw_sp_router_port_l3_stats_disable(struct mlxsw_sp_rif *rif)
7947 {
7948 	mlxsw_sp_rif_counter_free(rif, MLXSW_SP_RIF_COUNTER_EGRESS);
7949 	mlxsw_sp_rif_counter_free(rif, MLXSW_SP_RIF_COUNTER_INGRESS);
7950 }
7951 
7952 static void
7953 mlxsw_sp_router_port_l3_stats_report_used(struct mlxsw_sp_rif *rif,
7954 					  struct netdev_notifier_offload_xstats_info *info)
7955 {
7956 	if (!mlxsw_sp_router_port_l3_stats_enabled(rif))
7957 		return;
7958 	netdev_offload_xstats_report_used(info->report_used);
7959 }
7960 
7961 static int
7962 mlxsw_sp_router_port_l3_stats_fetch(struct mlxsw_sp_rif *rif,
7963 				    struct rtnl_hw_stats64 *p_stats)
7964 {
7965 	struct mlxsw_sp_rif_counter_set_basic ingress;
7966 	struct mlxsw_sp_rif_counter_set_basic egress;
7967 	int err;
7968 
7969 	err = mlxsw_sp_rif_counter_fetch_clear(rif,
7970 					       MLXSW_SP_RIF_COUNTER_INGRESS,
7971 					       &ingress);
7972 	if (err)
7973 		return err;
7974 
7975 	err = mlxsw_sp_rif_counter_fetch_clear(rif,
7976 					       MLXSW_SP_RIF_COUNTER_EGRESS,
7977 					       &egress);
7978 	if (err)
7979 		return err;
7980 
7981 #define MLXSW_SP_ROUTER_ALL_GOOD(SET, SFX)		\
7982 		((SET.good_unicast_ ## SFX) +		\
7983 		 (SET.good_multicast_ ## SFX) +		\
7984 		 (SET.good_broadcast_ ## SFX))
7985 
7986 	p_stats->rx_packets = MLXSW_SP_ROUTER_ALL_GOOD(ingress, packets);
7987 	p_stats->tx_packets = MLXSW_SP_ROUTER_ALL_GOOD(egress, packets);
7988 	p_stats->rx_bytes = MLXSW_SP_ROUTER_ALL_GOOD(ingress, bytes);
7989 	p_stats->tx_bytes = MLXSW_SP_ROUTER_ALL_GOOD(egress, bytes);
7990 	p_stats->rx_errors = ingress.error_packets;
7991 	p_stats->tx_errors = egress.error_packets;
7992 	p_stats->rx_dropped = ingress.discard_packets;
7993 	p_stats->tx_dropped = egress.discard_packets;
7994 	p_stats->multicast = ingress.good_multicast_packets +
7995 			     ingress.good_broadcast_packets;
7996 
7997 #undef MLXSW_SP_ROUTER_ALL_GOOD
7998 
7999 	return 0;
8000 }
8001 
8002 static int
8003 mlxsw_sp_router_port_l3_stats_report_delta(struct mlxsw_sp_rif *rif,
8004 					   struct netdev_notifier_offload_xstats_info *info)
8005 {
8006 	struct rtnl_hw_stats64 stats = {};
8007 	int err;
8008 
8009 	if (!mlxsw_sp_router_port_l3_stats_enabled(rif))
8010 		return 0;
8011 
8012 	err = mlxsw_sp_router_port_l3_stats_fetch(rif, &stats);
8013 	if (err)
8014 		return err;
8015 
8016 	netdev_offload_xstats_report_delta(info->report_delta, &stats);
8017 	return 0;
8018 }
8019 
8020 struct mlxsw_sp_router_hwstats_notify_work {
8021 	struct work_struct work;
8022 	struct net_device *dev;
8023 };
8024 
8025 static void mlxsw_sp_router_hwstats_notify_work(struct work_struct *work)
8026 {
8027 	struct mlxsw_sp_router_hwstats_notify_work *hws_work =
8028 		container_of(work, struct mlxsw_sp_router_hwstats_notify_work,
8029 			     work);
8030 
8031 	rtnl_lock();
8032 	rtnl_offload_xstats_notify(hws_work->dev);
8033 	rtnl_unlock();
8034 	dev_put(hws_work->dev);
8035 	kfree(hws_work);
8036 }
8037 
8038 static void
8039 mlxsw_sp_router_hwstats_notify_schedule(struct net_device *dev)
8040 {
8041 	struct mlxsw_sp_router_hwstats_notify_work *hws_work;
8042 
8043 	/* To collect notification payload, the core ends up sending another
8044 	 * notifier block message, which would deadlock on the attempt to
8045 	 * acquire the router lock again. Just postpone the notification until
8046 	 * later.
8047 	 */
8048 
8049 	hws_work = kzalloc(sizeof(*hws_work), GFP_KERNEL);
8050 	if (!hws_work)
8051 		return;
8052 
8053 	INIT_WORK(&hws_work->work, mlxsw_sp_router_hwstats_notify_work);
8054 	dev_hold(dev);
8055 	hws_work->dev = dev;
8056 	mlxsw_core_schedule_work(&hws_work->work);
8057 }
8058 
8059 int mlxsw_sp_rif_dev_ifindex(const struct mlxsw_sp_rif *rif)
8060 {
8061 	return rif->dev->ifindex;
8062 }
8063 
8064 const struct net_device *mlxsw_sp_rif_dev(const struct mlxsw_sp_rif *rif)
8065 {
8066 	return rif->dev;
8067 }
8068 
8069 static void mlxsw_sp_rif_push_l3_stats(struct mlxsw_sp_rif *rif)
8070 {
8071 	struct rtnl_hw_stats64 stats = {};
8072 
8073 	if (!mlxsw_sp_router_port_l3_stats_fetch(rif, &stats))
8074 		netdev_offload_xstats_push_delta(rif->dev,
8075 						 NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8076 						 &stats);
8077 }
8078 
8079 static struct mlxsw_sp_rif *
8080 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
8081 		    const struct mlxsw_sp_rif_params *params,
8082 		    struct netlink_ext_ack *extack)
8083 {
8084 	u32 tb_id = l3mdev_fib_table(params->dev);
8085 	const struct mlxsw_sp_rif_ops *ops;
8086 	struct mlxsw_sp_fid *fid = NULL;
8087 	enum mlxsw_sp_rif_type type;
8088 	struct mlxsw_sp_rif *rif;
8089 	struct mlxsw_sp_vr *vr;
8090 	u16 rif_index;
8091 	int i, err;
8092 
8093 	type = mlxsw_sp_dev_rif_type(mlxsw_sp, params->dev);
8094 	ops = mlxsw_sp->router->rif_ops_arr[type];
8095 
8096 	vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id ? : RT_TABLE_MAIN, extack);
8097 	if (IS_ERR(vr))
8098 		return ERR_CAST(vr);
8099 	vr->rif_count++;
8100 
8101 	err = mlxsw_sp_rif_index_alloc(mlxsw_sp, &rif_index);
8102 	if (err) {
8103 		NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported router interfaces");
8104 		goto err_rif_index_alloc;
8105 	}
8106 
8107 	rif = mlxsw_sp_rif_alloc(ops->rif_size, rif_index, vr->id, params->dev);
8108 	if (!rif) {
8109 		err = -ENOMEM;
8110 		goto err_rif_alloc;
8111 	}
8112 	dev_hold(rif->dev);
8113 	mlxsw_sp->router->rifs[rif_index] = rif;
8114 	rif->mlxsw_sp = mlxsw_sp;
8115 	rif->ops = ops;
8116 
8117 	if (ops->fid_get) {
8118 		fid = ops->fid_get(rif, extack);
8119 		if (IS_ERR(fid)) {
8120 			err = PTR_ERR(fid);
8121 			goto err_fid_get;
8122 		}
8123 		rif->fid = fid;
8124 	}
8125 
8126 	if (ops->setup)
8127 		ops->setup(rif, params);
8128 
8129 	err = ops->configure(rif, extack);
8130 	if (err)
8131 		goto err_configure;
8132 
8133 	for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++) {
8134 		err = mlxsw_sp_mr_rif_add(vr->mr_table[i], rif);
8135 		if (err)
8136 			goto err_mr_rif_add;
8137 	}
8138 
8139 	if (netdev_offload_xstats_enabled(rif->dev,
8140 					  NETDEV_OFFLOAD_XSTATS_TYPE_L3)) {
8141 		err = mlxsw_sp_router_port_l3_stats_enable(rif);
8142 		if (err)
8143 			goto err_stats_enable;
8144 		mlxsw_sp_router_hwstats_notify_schedule(rif->dev);
8145 	} else {
8146 		mlxsw_sp_rif_counters_alloc(rif);
8147 	}
8148 
8149 	atomic_inc(&mlxsw_sp->router->rifs_count);
8150 	return rif;
8151 
8152 err_stats_enable:
8153 err_mr_rif_add:
8154 	for (i--; i >= 0; i--)
8155 		mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
8156 	ops->deconfigure(rif);
8157 err_configure:
8158 	if (fid)
8159 		mlxsw_sp_fid_put(fid);
8160 err_fid_get:
8161 	mlxsw_sp->router->rifs[rif_index] = NULL;
8162 	dev_put(rif->dev);
8163 	kfree(rif);
8164 err_rif_alloc:
8165 err_rif_index_alloc:
8166 	vr->rif_count--;
8167 	mlxsw_sp_vr_put(mlxsw_sp, vr);
8168 	return ERR_PTR(err);
8169 }
8170 
8171 static void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif)
8172 {
8173 	const struct mlxsw_sp_rif_ops *ops = rif->ops;
8174 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
8175 	struct mlxsw_sp_fid *fid = rif->fid;
8176 	struct mlxsw_sp_vr *vr;
8177 	int i;
8178 
8179 	atomic_dec(&mlxsw_sp->router->rifs_count);
8180 	mlxsw_sp_router_rif_gone_sync(mlxsw_sp, rif);
8181 	vr = &mlxsw_sp->router->vrs[rif->vr_id];
8182 
8183 	if (netdev_offload_xstats_enabled(rif->dev,
8184 					  NETDEV_OFFLOAD_XSTATS_TYPE_L3)) {
8185 		mlxsw_sp_rif_push_l3_stats(rif);
8186 		mlxsw_sp_router_port_l3_stats_disable(rif);
8187 		mlxsw_sp_router_hwstats_notify_schedule(rif->dev);
8188 	} else {
8189 		mlxsw_sp_rif_counters_free(rif);
8190 	}
8191 
8192 	for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++)
8193 		mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
8194 	ops->deconfigure(rif);
8195 	if (fid)
8196 		/* Loopback RIFs are not associated with a FID. */
8197 		mlxsw_sp_fid_put(fid);
8198 	mlxsw_sp->router->rifs[rif->rif_index] = NULL;
8199 	dev_put(rif->dev);
8200 	kfree(rif);
8201 	vr->rif_count--;
8202 	mlxsw_sp_vr_put(mlxsw_sp, vr);
8203 }
8204 
8205 void mlxsw_sp_rif_destroy_by_dev(struct mlxsw_sp *mlxsw_sp,
8206 				 struct net_device *dev)
8207 {
8208 	struct mlxsw_sp_rif *rif;
8209 
8210 	mutex_lock(&mlxsw_sp->router->lock);
8211 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
8212 	if (!rif)
8213 		goto out;
8214 	mlxsw_sp_rif_destroy(rif);
8215 out:
8216 	mutex_unlock(&mlxsw_sp->router->lock);
8217 }
8218 
8219 static void
8220 mlxsw_sp_rif_subport_params_init(struct mlxsw_sp_rif_params *params,
8221 				 struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
8222 {
8223 	struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
8224 
8225 	params->vid = mlxsw_sp_port_vlan->vid;
8226 	params->lag = mlxsw_sp_port->lagged;
8227 	if (params->lag)
8228 		params->lag_id = mlxsw_sp_port->lag_id;
8229 	else
8230 		params->system_port = mlxsw_sp_port->local_port;
8231 }
8232 
8233 static struct mlxsw_sp_rif_subport *
8234 mlxsw_sp_rif_subport_rif(const struct mlxsw_sp_rif *rif)
8235 {
8236 	return container_of(rif, struct mlxsw_sp_rif_subport, common);
8237 }
8238 
8239 static struct mlxsw_sp_rif *
8240 mlxsw_sp_rif_subport_get(struct mlxsw_sp *mlxsw_sp,
8241 			 const struct mlxsw_sp_rif_params *params,
8242 			 struct netlink_ext_ack *extack)
8243 {
8244 	struct mlxsw_sp_rif_subport *rif_subport;
8245 	struct mlxsw_sp_rif *rif;
8246 
8247 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, params->dev);
8248 	if (!rif)
8249 		return mlxsw_sp_rif_create(mlxsw_sp, params, extack);
8250 
8251 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
8252 	refcount_inc(&rif_subport->ref_count);
8253 	return rif;
8254 }
8255 
8256 static void mlxsw_sp_rif_subport_put(struct mlxsw_sp_rif *rif)
8257 {
8258 	struct mlxsw_sp_rif_subport *rif_subport;
8259 
8260 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
8261 	if (!refcount_dec_and_test(&rif_subport->ref_count))
8262 		return;
8263 
8264 	mlxsw_sp_rif_destroy(rif);
8265 }
8266 
8267 static int mlxsw_sp_rif_mac_profile_index_alloc(struct mlxsw_sp *mlxsw_sp,
8268 						struct mlxsw_sp_rif_mac_profile *profile,
8269 						struct netlink_ext_ack *extack)
8270 {
8271 	u8 max_rif_mac_profiles = mlxsw_sp->router->max_rif_mac_profile;
8272 	struct mlxsw_sp_router *router = mlxsw_sp->router;
8273 	int id;
8274 
8275 	id = idr_alloc(&router->rif_mac_profiles_idr, profile, 0,
8276 		       max_rif_mac_profiles, GFP_KERNEL);
8277 
8278 	if (id >= 0) {
8279 		profile->id = id;
8280 		return 0;
8281 	}
8282 
8283 	if (id == -ENOSPC)
8284 		NL_SET_ERR_MSG_MOD(extack,
8285 				   "Exceeded number of supported router interface MAC profiles");
8286 
8287 	return id;
8288 }
8289 
8290 static struct mlxsw_sp_rif_mac_profile *
8291 mlxsw_sp_rif_mac_profile_index_free(struct mlxsw_sp *mlxsw_sp, u8 mac_profile)
8292 {
8293 	struct mlxsw_sp_rif_mac_profile *profile;
8294 
8295 	profile = idr_remove(&mlxsw_sp->router->rif_mac_profiles_idr,
8296 			     mac_profile);
8297 	WARN_ON(!profile);
8298 	return profile;
8299 }
8300 
8301 static struct mlxsw_sp_rif_mac_profile *
8302 mlxsw_sp_rif_mac_profile_alloc(const char *mac)
8303 {
8304 	struct mlxsw_sp_rif_mac_profile *profile;
8305 
8306 	profile = kzalloc(sizeof(*profile), GFP_KERNEL);
8307 	if (!profile)
8308 		return NULL;
8309 
8310 	ether_addr_copy(profile->mac_prefix, mac);
8311 	refcount_set(&profile->ref_count, 1);
8312 	return profile;
8313 }
8314 
8315 static struct mlxsw_sp_rif_mac_profile *
8316 mlxsw_sp_rif_mac_profile_find(const struct mlxsw_sp *mlxsw_sp, const char *mac)
8317 {
8318 	struct mlxsw_sp_router *router = mlxsw_sp->router;
8319 	struct mlxsw_sp_rif_mac_profile *profile;
8320 	int id;
8321 
8322 	idr_for_each_entry(&router->rif_mac_profiles_idr, profile, id) {
8323 		if (ether_addr_equal_masked(profile->mac_prefix, mac,
8324 					    mlxsw_sp->mac_mask))
8325 			return profile;
8326 	}
8327 
8328 	return NULL;
8329 }
8330 
8331 static u64 mlxsw_sp_rif_mac_profiles_occ_get(void *priv)
8332 {
8333 	const struct mlxsw_sp *mlxsw_sp = priv;
8334 
8335 	return atomic_read(&mlxsw_sp->router->rif_mac_profiles_count);
8336 }
8337 
8338 static u64 mlxsw_sp_rifs_occ_get(void *priv)
8339 {
8340 	const struct mlxsw_sp *mlxsw_sp = priv;
8341 
8342 	return atomic_read(&mlxsw_sp->router->rifs_count);
8343 }
8344 
8345 static struct mlxsw_sp_rif_mac_profile *
8346 mlxsw_sp_rif_mac_profile_create(struct mlxsw_sp *mlxsw_sp, const char *mac,
8347 				struct netlink_ext_ack *extack)
8348 {
8349 	struct mlxsw_sp_rif_mac_profile *profile;
8350 	int err;
8351 
8352 	profile = mlxsw_sp_rif_mac_profile_alloc(mac);
8353 	if (!profile)
8354 		return ERR_PTR(-ENOMEM);
8355 
8356 	err = mlxsw_sp_rif_mac_profile_index_alloc(mlxsw_sp, profile, extack);
8357 	if (err)
8358 		goto profile_index_alloc_err;
8359 
8360 	atomic_inc(&mlxsw_sp->router->rif_mac_profiles_count);
8361 	return profile;
8362 
8363 profile_index_alloc_err:
8364 	kfree(profile);
8365 	return ERR_PTR(err);
8366 }
8367 
8368 static void mlxsw_sp_rif_mac_profile_destroy(struct mlxsw_sp *mlxsw_sp,
8369 					     u8 mac_profile)
8370 {
8371 	struct mlxsw_sp_rif_mac_profile *profile;
8372 
8373 	atomic_dec(&mlxsw_sp->router->rif_mac_profiles_count);
8374 	profile = mlxsw_sp_rif_mac_profile_index_free(mlxsw_sp, mac_profile);
8375 	kfree(profile);
8376 }
8377 
8378 static int mlxsw_sp_rif_mac_profile_get(struct mlxsw_sp *mlxsw_sp,
8379 					const char *mac, u8 *p_mac_profile,
8380 					struct netlink_ext_ack *extack)
8381 {
8382 	struct mlxsw_sp_rif_mac_profile *profile;
8383 
8384 	profile = mlxsw_sp_rif_mac_profile_find(mlxsw_sp, mac);
8385 	if (profile) {
8386 		refcount_inc(&profile->ref_count);
8387 		goto out;
8388 	}
8389 
8390 	profile = mlxsw_sp_rif_mac_profile_create(mlxsw_sp, mac, extack);
8391 	if (IS_ERR(profile))
8392 		return PTR_ERR(profile);
8393 
8394 out:
8395 	*p_mac_profile = profile->id;
8396 	return 0;
8397 }
8398 
8399 static void mlxsw_sp_rif_mac_profile_put(struct mlxsw_sp *mlxsw_sp,
8400 					 u8 mac_profile)
8401 {
8402 	struct mlxsw_sp_rif_mac_profile *profile;
8403 
8404 	profile = idr_find(&mlxsw_sp->router->rif_mac_profiles_idr,
8405 			   mac_profile);
8406 	if (WARN_ON(!profile))
8407 		return;
8408 
8409 	if (!refcount_dec_and_test(&profile->ref_count))
8410 		return;
8411 
8412 	mlxsw_sp_rif_mac_profile_destroy(mlxsw_sp, mac_profile);
8413 }
8414 
8415 static bool mlxsw_sp_rif_mac_profile_is_shared(const struct mlxsw_sp_rif *rif)
8416 {
8417 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
8418 	struct mlxsw_sp_rif_mac_profile *profile;
8419 
8420 	profile = idr_find(&mlxsw_sp->router->rif_mac_profiles_idr,
8421 			   rif->mac_profile_id);
8422 	if (WARN_ON(!profile))
8423 		return false;
8424 
8425 	return refcount_read(&profile->ref_count) > 1;
8426 }
8427 
8428 static int mlxsw_sp_rif_mac_profile_edit(struct mlxsw_sp_rif *rif,
8429 					 const char *new_mac)
8430 {
8431 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
8432 	struct mlxsw_sp_rif_mac_profile *profile;
8433 
8434 	profile = idr_find(&mlxsw_sp->router->rif_mac_profiles_idr,
8435 			   rif->mac_profile_id);
8436 	if (WARN_ON(!profile))
8437 		return -EINVAL;
8438 
8439 	ether_addr_copy(profile->mac_prefix, new_mac);
8440 	return 0;
8441 }
8442 
8443 static int
8444 mlxsw_sp_rif_mac_profile_replace(struct mlxsw_sp *mlxsw_sp,
8445 				 struct mlxsw_sp_rif *rif,
8446 				 const char *new_mac,
8447 				 struct netlink_ext_ack *extack)
8448 {
8449 	u8 mac_profile;
8450 	int err;
8451 
8452 	if (!mlxsw_sp_rif_mac_profile_is_shared(rif) &&
8453 	    !mlxsw_sp_rif_mac_profile_find(mlxsw_sp, new_mac))
8454 		return mlxsw_sp_rif_mac_profile_edit(rif, new_mac);
8455 
8456 	err = mlxsw_sp_rif_mac_profile_get(mlxsw_sp, new_mac,
8457 					   &mac_profile, extack);
8458 	if (err)
8459 		return err;
8460 
8461 	mlxsw_sp_rif_mac_profile_put(mlxsw_sp, rif->mac_profile_id);
8462 	rif->mac_profile_id = mac_profile;
8463 	return 0;
8464 }
8465 
8466 static int
8467 __mlxsw_sp_port_vlan_router_join(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan,
8468 				 struct net_device *l3_dev,
8469 				 struct netlink_ext_ack *extack)
8470 {
8471 	struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
8472 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
8473 	struct mlxsw_sp_rif_params params = {
8474 		.dev = l3_dev,
8475 	};
8476 	u16 vid = mlxsw_sp_port_vlan->vid;
8477 	struct mlxsw_sp_rif *rif;
8478 	struct mlxsw_sp_fid *fid;
8479 	int err;
8480 
8481 	mlxsw_sp_rif_subport_params_init(&params, mlxsw_sp_port_vlan);
8482 	rif = mlxsw_sp_rif_subport_get(mlxsw_sp, &params, extack);
8483 	if (IS_ERR(rif))
8484 		return PTR_ERR(rif);
8485 
8486 	/* FID was already created, just take a reference */
8487 	fid = rif->ops->fid_get(rif, extack);
8488 	err = mlxsw_sp_fid_port_vid_map(fid, mlxsw_sp_port, vid);
8489 	if (err)
8490 		goto err_fid_port_vid_map;
8491 
8492 	err = mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, false);
8493 	if (err)
8494 		goto err_port_vid_learning_set;
8495 
8496 	err = mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid,
8497 					BR_STATE_FORWARDING);
8498 	if (err)
8499 		goto err_port_vid_stp_set;
8500 
8501 	mlxsw_sp_port_vlan->fid = fid;
8502 
8503 	return 0;
8504 
8505 err_port_vid_stp_set:
8506 	mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
8507 err_port_vid_learning_set:
8508 	mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
8509 err_fid_port_vid_map:
8510 	mlxsw_sp_fid_put(fid);
8511 	mlxsw_sp_rif_subport_put(rif);
8512 	return err;
8513 }
8514 
8515 static void
8516 __mlxsw_sp_port_vlan_router_leave(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
8517 {
8518 	struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
8519 	struct mlxsw_sp_fid *fid = mlxsw_sp_port_vlan->fid;
8520 	struct mlxsw_sp_rif *rif = mlxsw_sp_fid_rif(fid);
8521 	u16 vid = mlxsw_sp_port_vlan->vid;
8522 
8523 	if (WARN_ON(mlxsw_sp_fid_type(fid) != MLXSW_SP_FID_TYPE_RFID))
8524 		return;
8525 
8526 	mlxsw_sp_port_vlan->fid = NULL;
8527 	mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid, BR_STATE_BLOCKING);
8528 	mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
8529 	mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
8530 	mlxsw_sp_fid_put(fid);
8531 	mlxsw_sp_rif_subport_put(rif);
8532 }
8533 
8534 int
8535 mlxsw_sp_port_vlan_router_join(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan,
8536 			       struct net_device *l3_dev,
8537 			       struct netlink_ext_ack *extack)
8538 {
8539 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port_vlan->mlxsw_sp_port->mlxsw_sp;
8540 	struct mlxsw_sp_rif *rif;
8541 	int err = 0;
8542 
8543 	mutex_lock(&mlxsw_sp->router->lock);
8544 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
8545 	if (!rif)
8546 		goto out;
8547 
8548 	err = __mlxsw_sp_port_vlan_router_join(mlxsw_sp_port_vlan, l3_dev,
8549 					       extack);
8550 out:
8551 	mutex_unlock(&mlxsw_sp->router->lock);
8552 	return err;
8553 }
8554 
8555 void
8556 mlxsw_sp_port_vlan_router_leave(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
8557 {
8558 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port_vlan->mlxsw_sp_port->mlxsw_sp;
8559 
8560 	mutex_lock(&mlxsw_sp->router->lock);
8561 	__mlxsw_sp_port_vlan_router_leave(mlxsw_sp_port_vlan);
8562 	mutex_unlock(&mlxsw_sp->router->lock);
8563 }
8564 
8565 static int mlxsw_sp_inetaddr_port_vlan_event(struct net_device *l3_dev,
8566 					     struct net_device *port_dev,
8567 					     unsigned long event, u16 vid,
8568 					     struct netlink_ext_ack *extack)
8569 {
8570 	struct mlxsw_sp_port *mlxsw_sp_port = netdev_priv(port_dev);
8571 	struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan;
8572 
8573 	mlxsw_sp_port_vlan = mlxsw_sp_port_vlan_find_by_vid(mlxsw_sp_port, vid);
8574 	if (WARN_ON(!mlxsw_sp_port_vlan))
8575 		return -EINVAL;
8576 
8577 	switch (event) {
8578 	case NETDEV_UP:
8579 		return __mlxsw_sp_port_vlan_router_join(mlxsw_sp_port_vlan,
8580 							l3_dev, extack);
8581 	case NETDEV_DOWN:
8582 		__mlxsw_sp_port_vlan_router_leave(mlxsw_sp_port_vlan);
8583 		break;
8584 	}
8585 
8586 	return 0;
8587 }
8588 
8589 static int mlxsw_sp_inetaddr_port_event(struct net_device *port_dev,
8590 					unsigned long event,
8591 					struct netlink_ext_ack *extack)
8592 {
8593 	if (netif_is_bridge_port(port_dev) ||
8594 	    netif_is_lag_port(port_dev) ||
8595 	    netif_is_ovs_port(port_dev))
8596 		return 0;
8597 
8598 	return mlxsw_sp_inetaddr_port_vlan_event(port_dev, port_dev, event,
8599 						 MLXSW_SP_DEFAULT_VID, extack);
8600 }
8601 
8602 static int __mlxsw_sp_inetaddr_lag_event(struct net_device *l3_dev,
8603 					 struct net_device *lag_dev,
8604 					 unsigned long event, u16 vid,
8605 					 struct netlink_ext_ack *extack)
8606 {
8607 	struct net_device *port_dev;
8608 	struct list_head *iter;
8609 	int err;
8610 
8611 	netdev_for_each_lower_dev(lag_dev, port_dev, iter) {
8612 		if (mlxsw_sp_port_dev_check(port_dev)) {
8613 			err = mlxsw_sp_inetaddr_port_vlan_event(l3_dev,
8614 								port_dev,
8615 								event, vid,
8616 								extack);
8617 			if (err)
8618 				return err;
8619 		}
8620 	}
8621 
8622 	return 0;
8623 }
8624 
8625 static int mlxsw_sp_inetaddr_lag_event(struct net_device *lag_dev,
8626 				       unsigned long event,
8627 				       struct netlink_ext_ack *extack)
8628 {
8629 	if (netif_is_bridge_port(lag_dev))
8630 		return 0;
8631 
8632 	return __mlxsw_sp_inetaddr_lag_event(lag_dev, lag_dev, event,
8633 					     MLXSW_SP_DEFAULT_VID, extack);
8634 }
8635 
8636 static int mlxsw_sp_inetaddr_bridge_event(struct mlxsw_sp *mlxsw_sp,
8637 					  struct net_device *l3_dev,
8638 					  unsigned long event,
8639 					  struct netlink_ext_ack *extack)
8640 {
8641 	struct mlxsw_sp_rif_params params = {
8642 		.dev = l3_dev,
8643 	};
8644 	struct mlxsw_sp_rif *rif;
8645 
8646 	switch (event) {
8647 	case NETDEV_UP:
8648 		if (netif_is_bridge_master(l3_dev) && br_vlan_enabled(l3_dev)) {
8649 			u16 proto;
8650 
8651 			br_vlan_get_proto(l3_dev, &proto);
8652 			if (proto == ETH_P_8021AD) {
8653 				NL_SET_ERR_MSG_MOD(extack, "Adding an IP address to 802.1ad bridge is not supported");
8654 				return -EOPNOTSUPP;
8655 			}
8656 		}
8657 		rif = mlxsw_sp_rif_create(mlxsw_sp, &params, extack);
8658 		if (IS_ERR(rif))
8659 			return PTR_ERR(rif);
8660 		break;
8661 	case NETDEV_DOWN:
8662 		rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
8663 		mlxsw_sp_rif_destroy(rif);
8664 		break;
8665 	}
8666 
8667 	return 0;
8668 }
8669 
8670 static int mlxsw_sp_inetaddr_vlan_event(struct mlxsw_sp *mlxsw_sp,
8671 					struct net_device *vlan_dev,
8672 					unsigned long event,
8673 					struct netlink_ext_ack *extack)
8674 {
8675 	struct net_device *real_dev = vlan_dev_real_dev(vlan_dev);
8676 	u16 vid = vlan_dev_vlan_id(vlan_dev);
8677 
8678 	if (netif_is_bridge_port(vlan_dev))
8679 		return 0;
8680 
8681 	if (mlxsw_sp_port_dev_check(real_dev))
8682 		return mlxsw_sp_inetaddr_port_vlan_event(vlan_dev, real_dev,
8683 							 event, vid, extack);
8684 	else if (netif_is_lag_master(real_dev))
8685 		return __mlxsw_sp_inetaddr_lag_event(vlan_dev, real_dev, event,
8686 						     vid, extack);
8687 	else if (netif_is_bridge_master(real_dev) && br_vlan_enabled(real_dev))
8688 		return mlxsw_sp_inetaddr_bridge_event(mlxsw_sp, vlan_dev, event,
8689 						      extack);
8690 
8691 	return 0;
8692 }
8693 
8694 static bool mlxsw_sp_rif_macvlan_is_vrrp4(const u8 *mac)
8695 {
8696 	u8 vrrp4[ETH_ALEN] = { 0x00, 0x00, 0x5e, 0x00, 0x01, 0x00 };
8697 	u8 mask[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0x00 };
8698 
8699 	return ether_addr_equal_masked(mac, vrrp4, mask);
8700 }
8701 
8702 static bool mlxsw_sp_rif_macvlan_is_vrrp6(const u8 *mac)
8703 {
8704 	u8 vrrp6[ETH_ALEN] = { 0x00, 0x00, 0x5e, 0x00, 0x02, 0x00 };
8705 	u8 mask[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0x00 };
8706 
8707 	return ether_addr_equal_masked(mac, vrrp6, mask);
8708 }
8709 
8710 static int mlxsw_sp_rif_vrrp_op(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
8711 				const u8 *mac, bool adding)
8712 {
8713 	char ritr_pl[MLXSW_REG_RITR_LEN];
8714 	u8 vrrp_id = adding ? mac[5] : 0;
8715 	int err;
8716 
8717 	if (!mlxsw_sp_rif_macvlan_is_vrrp4(mac) &&
8718 	    !mlxsw_sp_rif_macvlan_is_vrrp6(mac))
8719 		return 0;
8720 
8721 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
8722 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
8723 	if (err)
8724 		return err;
8725 
8726 	if (mlxsw_sp_rif_macvlan_is_vrrp4(mac))
8727 		mlxsw_reg_ritr_if_vrrp_id_ipv4_set(ritr_pl, vrrp_id);
8728 	else
8729 		mlxsw_reg_ritr_if_vrrp_id_ipv6_set(ritr_pl, vrrp_id);
8730 
8731 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
8732 }
8733 
8734 static int mlxsw_sp_rif_macvlan_add(struct mlxsw_sp *mlxsw_sp,
8735 				    const struct net_device *macvlan_dev,
8736 				    struct netlink_ext_ack *extack)
8737 {
8738 	struct macvlan_dev *vlan = netdev_priv(macvlan_dev);
8739 	struct mlxsw_sp_rif *rif;
8740 	int err;
8741 
8742 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, vlan->lowerdev);
8743 	if (!rif) {
8744 		NL_SET_ERR_MSG_MOD(extack, "macvlan is only supported on top of router interfaces");
8745 		return -EOPNOTSUPP;
8746 	}
8747 
8748 	err = mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
8749 				  mlxsw_sp_fid_index(rif->fid), true);
8750 	if (err)
8751 		return err;
8752 
8753 	err = mlxsw_sp_rif_vrrp_op(mlxsw_sp, rif->rif_index,
8754 				   macvlan_dev->dev_addr, true);
8755 	if (err)
8756 		goto err_rif_vrrp_add;
8757 
8758 	/* Make sure the bridge driver does not have this MAC pointing at
8759 	 * some other port.
8760 	 */
8761 	if (rif->ops->fdb_del)
8762 		rif->ops->fdb_del(rif, macvlan_dev->dev_addr);
8763 
8764 	return 0;
8765 
8766 err_rif_vrrp_add:
8767 	mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
8768 			    mlxsw_sp_fid_index(rif->fid), false);
8769 	return err;
8770 }
8771 
8772 static void __mlxsw_sp_rif_macvlan_del(struct mlxsw_sp *mlxsw_sp,
8773 				       const struct net_device *macvlan_dev)
8774 {
8775 	struct macvlan_dev *vlan = netdev_priv(macvlan_dev);
8776 	struct mlxsw_sp_rif *rif;
8777 
8778 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, vlan->lowerdev);
8779 	/* If we do not have a RIF, then we already took care of
8780 	 * removing the macvlan's MAC during RIF deletion.
8781 	 */
8782 	if (!rif)
8783 		return;
8784 	mlxsw_sp_rif_vrrp_op(mlxsw_sp, rif->rif_index, macvlan_dev->dev_addr,
8785 			     false);
8786 	mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
8787 			    mlxsw_sp_fid_index(rif->fid), false);
8788 }
8789 
8790 void mlxsw_sp_rif_macvlan_del(struct mlxsw_sp *mlxsw_sp,
8791 			      const struct net_device *macvlan_dev)
8792 {
8793 	mutex_lock(&mlxsw_sp->router->lock);
8794 	__mlxsw_sp_rif_macvlan_del(mlxsw_sp, macvlan_dev);
8795 	mutex_unlock(&mlxsw_sp->router->lock);
8796 }
8797 
8798 static int mlxsw_sp_inetaddr_macvlan_event(struct mlxsw_sp *mlxsw_sp,
8799 					   struct net_device *macvlan_dev,
8800 					   unsigned long event,
8801 					   struct netlink_ext_ack *extack)
8802 {
8803 	switch (event) {
8804 	case NETDEV_UP:
8805 		return mlxsw_sp_rif_macvlan_add(mlxsw_sp, macvlan_dev, extack);
8806 	case NETDEV_DOWN:
8807 		__mlxsw_sp_rif_macvlan_del(mlxsw_sp, macvlan_dev);
8808 		break;
8809 	}
8810 
8811 	return 0;
8812 }
8813 
8814 static int __mlxsw_sp_inetaddr_event(struct mlxsw_sp *mlxsw_sp,
8815 				     struct net_device *dev,
8816 				     unsigned long event,
8817 				     struct netlink_ext_ack *extack)
8818 {
8819 	if (mlxsw_sp_port_dev_check(dev))
8820 		return mlxsw_sp_inetaddr_port_event(dev, event, extack);
8821 	else if (netif_is_lag_master(dev))
8822 		return mlxsw_sp_inetaddr_lag_event(dev, event, extack);
8823 	else if (netif_is_bridge_master(dev))
8824 		return mlxsw_sp_inetaddr_bridge_event(mlxsw_sp, dev, event,
8825 						      extack);
8826 	else if (is_vlan_dev(dev))
8827 		return mlxsw_sp_inetaddr_vlan_event(mlxsw_sp, dev, event,
8828 						    extack);
8829 	else if (netif_is_macvlan(dev))
8830 		return mlxsw_sp_inetaddr_macvlan_event(mlxsw_sp, dev, event,
8831 						       extack);
8832 	else
8833 		return 0;
8834 }
8835 
8836 static int mlxsw_sp_inetaddr_event(struct notifier_block *nb,
8837 				   unsigned long event, void *ptr)
8838 {
8839 	struct in_ifaddr *ifa = (struct in_ifaddr *) ptr;
8840 	struct net_device *dev = ifa->ifa_dev->dev;
8841 	struct mlxsw_sp_router *router;
8842 	struct mlxsw_sp_rif *rif;
8843 	int err = 0;
8844 
8845 	/* NETDEV_UP event is handled by mlxsw_sp_inetaddr_valid_event */
8846 	if (event == NETDEV_UP)
8847 		return NOTIFY_DONE;
8848 
8849 	router = container_of(nb, struct mlxsw_sp_router, inetaddr_nb);
8850 	mutex_lock(&router->lock);
8851 	rif = mlxsw_sp_rif_find_by_dev(router->mlxsw_sp, dev);
8852 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
8853 		goto out;
8854 
8855 	err = __mlxsw_sp_inetaddr_event(router->mlxsw_sp, dev, event, NULL);
8856 out:
8857 	mutex_unlock(&router->lock);
8858 	return notifier_from_errno(err);
8859 }
8860 
8861 int mlxsw_sp_inetaddr_valid_event(struct notifier_block *unused,
8862 				  unsigned long event, void *ptr)
8863 {
8864 	struct in_validator_info *ivi = (struct in_validator_info *) ptr;
8865 	struct net_device *dev = ivi->ivi_dev->dev;
8866 	struct mlxsw_sp *mlxsw_sp;
8867 	struct mlxsw_sp_rif *rif;
8868 	int err = 0;
8869 
8870 	mlxsw_sp = mlxsw_sp_lower_get(dev);
8871 	if (!mlxsw_sp)
8872 		return NOTIFY_DONE;
8873 
8874 	mutex_lock(&mlxsw_sp->router->lock);
8875 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
8876 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
8877 		goto out;
8878 
8879 	err = __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, ivi->extack);
8880 out:
8881 	mutex_unlock(&mlxsw_sp->router->lock);
8882 	return notifier_from_errno(err);
8883 }
8884 
8885 struct mlxsw_sp_inet6addr_event_work {
8886 	struct work_struct work;
8887 	struct mlxsw_sp *mlxsw_sp;
8888 	struct net_device *dev;
8889 	unsigned long event;
8890 };
8891 
8892 static void mlxsw_sp_inet6addr_event_work(struct work_struct *work)
8893 {
8894 	struct mlxsw_sp_inet6addr_event_work *inet6addr_work =
8895 		container_of(work, struct mlxsw_sp_inet6addr_event_work, work);
8896 	struct mlxsw_sp *mlxsw_sp = inet6addr_work->mlxsw_sp;
8897 	struct net_device *dev = inet6addr_work->dev;
8898 	unsigned long event = inet6addr_work->event;
8899 	struct mlxsw_sp_rif *rif;
8900 
8901 	rtnl_lock();
8902 	mutex_lock(&mlxsw_sp->router->lock);
8903 
8904 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
8905 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
8906 		goto out;
8907 
8908 	__mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, NULL);
8909 out:
8910 	mutex_unlock(&mlxsw_sp->router->lock);
8911 	rtnl_unlock();
8912 	dev_put(dev);
8913 	kfree(inet6addr_work);
8914 }
8915 
8916 /* Called with rcu_read_lock() */
8917 static int mlxsw_sp_inet6addr_event(struct notifier_block *nb,
8918 				    unsigned long event, void *ptr)
8919 {
8920 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *) ptr;
8921 	struct mlxsw_sp_inet6addr_event_work *inet6addr_work;
8922 	struct net_device *dev = if6->idev->dev;
8923 	struct mlxsw_sp_router *router;
8924 
8925 	/* NETDEV_UP event is handled by mlxsw_sp_inet6addr_valid_event */
8926 	if (event == NETDEV_UP)
8927 		return NOTIFY_DONE;
8928 
8929 	inet6addr_work = kzalloc(sizeof(*inet6addr_work), GFP_ATOMIC);
8930 	if (!inet6addr_work)
8931 		return NOTIFY_BAD;
8932 
8933 	router = container_of(nb, struct mlxsw_sp_router, inet6addr_nb);
8934 	INIT_WORK(&inet6addr_work->work, mlxsw_sp_inet6addr_event_work);
8935 	inet6addr_work->mlxsw_sp = router->mlxsw_sp;
8936 	inet6addr_work->dev = dev;
8937 	inet6addr_work->event = event;
8938 	dev_hold(dev);
8939 	mlxsw_core_schedule_work(&inet6addr_work->work);
8940 
8941 	return NOTIFY_DONE;
8942 }
8943 
8944 int mlxsw_sp_inet6addr_valid_event(struct notifier_block *unused,
8945 				   unsigned long event, void *ptr)
8946 {
8947 	struct in6_validator_info *i6vi = (struct in6_validator_info *) ptr;
8948 	struct net_device *dev = i6vi->i6vi_dev->dev;
8949 	struct mlxsw_sp *mlxsw_sp;
8950 	struct mlxsw_sp_rif *rif;
8951 	int err = 0;
8952 
8953 	mlxsw_sp = mlxsw_sp_lower_get(dev);
8954 	if (!mlxsw_sp)
8955 		return NOTIFY_DONE;
8956 
8957 	mutex_lock(&mlxsw_sp->router->lock);
8958 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
8959 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
8960 		goto out;
8961 
8962 	err = __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, i6vi->extack);
8963 out:
8964 	mutex_unlock(&mlxsw_sp->router->lock);
8965 	return notifier_from_errno(err);
8966 }
8967 
8968 static int mlxsw_sp_rif_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
8969 			     const char *mac, int mtu, u8 mac_profile)
8970 {
8971 	char ritr_pl[MLXSW_REG_RITR_LEN];
8972 	int err;
8973 
8974 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
8975 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
8976 	if (err)
8977 		return err;
8978 
8979 	mlxsw_reg_ritr_mtu_set(ritr_pl, mtu);
8980 	mlxsw_reg_ritr_if_mac_memcpy_to(ritr_pl, mac);
8981 	mlxsw_reg_ritr_if_mac_profile_id_set(ritr_pl, mac_profile);
8982 	mlxsw_reg_ritr_op_set(ritr_pl, MLXSW_REG_RITR_RIF_CREATE);
8983 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
8984 }
8985 
8986 static int
8987 mlxsw_sp_router_port_change_event(struct mlxsw_sp *mlxsw_sp,
8988 				  struct mlxsw_sp_rif *rif,
8989 				  struct netlink_ext_ack *extack)
8990 {
8991 	struct net_device *dev = rif->dev;
8992 	u8 old_mac_profile;
8993 	u16 fid_index;
8994 	int err;
8995 
8996 	fid_index = mlxsw_sp_fid_index(rif->fid);
8997 
8998 	err = mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, false);
8999 	if (err)
9000 		return err;
9001 
9002 	old_mac_profile = rif->mac_profile_id;
9003 	err = mlxsw_sp_rif_mac_profile_replace(mlxsw_sp, rif, dev->dev_addr,
9004 					       extack);
9005 	if (err)
9006 		goto err_rif_mac_profile_replace;
9007 
9008 	err = mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, dev->dev_addr,
9009 				dev->mtu, rif->mac_profile_id);
9010 	if (err)
9011 		goto err_rif_edit;
9012 
9013 	err = mlxsw_sp_rif_fdb_op(mlxsw_sp, dev->dev_addr, fid_index, true);
9014 	if (err)
9015 		goto err_rif_fdb_op;
9016 
9017 	if (rif->mtu != dev->mtu) {
9018 		struct mlxsw_sp_vr *vr;
9019 		int i;
9020 
9021 		/* The RIF is relevant only to its mr_table instance, as unlike
9022 		 * unicast routing, in multicast routing a RIF cannot be shared
9023 		 * between several multicast routing tables.
9024 		 */
9025 		vr = &mlxsw_sp->router->vrs[rif->vr_id];
9026 		for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++)
9027 			mlxsw_sp_mr_rif_mtu_update(vr->mr_table[i],
9028 						   rif, dev->mtu);
9029 	}
9030 
9031 	ether_addr_copy(rif->addr, dev->dev_addr);
9032 	rif->mtu = dev->mtu;
9033 
9034 	netdev_dbg(dev, "Updated RIF=%d\n", rif->rif_index);
9035 
9036 	return 0;
9037 
9038 err_rif_fdb_op:
9039 	mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, rif->addr, rif->mtu,
9040 			  old_mac_profile);
9041 err_rif_edit:
9042 	mlxsw_sp_rif_mac_profile_replace(mlxsw_sp, rif, rif->addr, extack);
9043 err_rif_mac_profile_replace:
9044 	mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, true);
9045 	return err;
9046 }
9047 
9048 static int mlxsw_sp_router_port_pre_changeaddr_event(struct mlxsw_sp_rif *rif,
9049 			    struct netdev_notifier_pre_changeaddr_info *info)
9050 {
9051 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9052 	struct mlxsw_sp_rif_mac_profile *profile;
9053 	struct netlink_ext_ack *extack;
9054 	u8 max_rif_mac_profiles;
9055 	u64 occ;
9056 
9057 	extack = netdev_notifier_info_to_extack(&info->info);
9058 
9059 	profile = mlxsw_sp_rif_mac_profile_find(mlxsw_sp, info->dev_addr);
9060 	if (profile)
9061 		return 0;
9062 
9063 	max_rif_mac_profiles = mlxsw_sp->router->max_rif_mac_profile;
9064 	occ = mlxsw_sp_rif_mac_profiles_occ_get(mlxsw_sp);
9065 	if (occ < max_rif_mac_profiles)
9066 		return 0;
9067 
9068 	if (!mlxsw_sp_rif_mac_profile_is_shared(rif))
9069 		return 0;
9070 
9071 	NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported router interface MAC profiles");
9072 	return -ENOBUFS;
9073 }
9074 
9075 static bool mlxsw_sp_is_offload_xstats_event(unsigned long event)
9076 {
9077 	switch (event) {
9078 	case NETDEV_OFFLOAD_XSTATS_ENABLE:
9079 	case NETDEV_OFFLOAD_XSTATS_DISABLE:
9080 	case NETDEV_OFFLOAD_XSTATS_REPORT_USED:
9081 	case NETDEV_OFFLOAD_XSTATS_REPORT_DELTA:
9082 		return true;
9083 	}
9084 
9085 	return false;
9086 }
9087 
9088 static int
9089 mlxsw_sp_router_port_offload_xstats_cmd(struct mlxsw_sp_rif *rif,
9090 					unsigned long event,
9091 					struct netdev_notifier_offload_xstats_info *info)
9092 {
9093 	switch (info->type) {
9094 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
9095 		break;
9096 	default:
9097 		return 0;
9098 	}
9099 
9100 	switch (event) {
9101 	case NETDEV_OFFLOAD_XSTATS_ENABLE:
9102 		return mlxsw_sp_router_port_l3_stats_enable(rif);
9103 	case NETDEV_OFFLOAD_XSTATS_DISABLE:
9104 		mlxsw_sp_router_port_l3_stats_disable(rif);
9105 		return 0;
9106 	case NETDEV_OFFLOAD_XSTATS_REPORT_USED:
9107 		mlxsw_sp_router_port_l3_stats_report_used(rif, info);
9108 		return 0;
9109 	case NETDEV_OFFLOAD_XSTATS_REPORT_DELTA:
9110 		return mlxsw_sp_router_port_l3_stats_report_delta(rif, info);
9111 	}
9112 
9113 	WARN_ON_ONCE(1);
9114 	return 0;
9115 }
9116 
9117 static int
9118 mlxsw_sp_netdevice_offload_xstats_cmd(struct mlxsw_sp *mlxsw_sp,
9119 				      struct net_device *dev,
9120 				      unsigned long event,
9121 				      struct netdev_notifier_offload_xstats_info *info)
9122 {
9123 	struct mlxsw_sp_rif *rif;
9124 
9125 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
9126 	if (!rif)
9127 		return 0;
9128 
9129 	return mlxsw_sp_router_port_offload_xstats_cmd(rif, event, info);
9130 }
9131 
9132 static bool mlxsw_sp_is_router_event(unsigned long event)
9133 {
9134 	switch (event) {
9135 	case NETDEV_PRE_CHANGEADDR:
9136 	case NETDEV_CHANGEADDR:
9137 	case NETDEV_CHANGEMTU:
9138 		return true;
9139 	default:
9140 		return false;
9141 	}
9142 }
9143 
9144 static int mlxsw_sp_netdevice_router_port_event(struct net_device *dev,
9145 						unsigned long event, void *ptr)
9146 {
9147 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
9148 	struct mlxsw_sp *mlxsw_sp;
9149 	struct mlxsw_sp_rif *rif;
9150 
9151 	mlxsw_sp = mlxsw_sp_lower_get(dev);
9152 	if (!mlxsw_sp)
9153 		return 0;
9154 
9155 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
9156 	if (!rif)
9157 		return 0;
9158 
9159 	switch (event) {
9160 	case NETDEV_CHANGEMTU:
9161 	case NETDEV_CHANGEADDR:
9162 		return mlxsw_sp_router_port_change_event(mlxsw_sp, rif, extack);
9163 	case NETDEV_PRE_CHANGEADDR:
9164 		return mlxsw_sp_router_port_pre_changeaddr_event(rif, ptr);
9165 	default:
9166 		WARN_ON_ONCE(1);
9167 		break;
9168 	}
9169 
9170 	return 0;
9171 }
9172 
9173 static int mlxsw_sp_port_vrf_join(struct mlxsw_sp *mlxsw_sp,
9174 				  struct net_device *l3_dev,
9175 				  struct netlink_ext_ack *extack)
9176 {
9177 	struct mlxsw_sp_rif *rif;
9178 
9179 	/* If netdev is already associated with a RIF, then we need to
9180 	 * destroy it and create a new one with the new virtual router ID.
9181 	 */
9182 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
9183 	if (rif)
9184 		__mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_DOWN,
9185 					  extack);
9186 
9187 	return __mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_UP, extack);
9188 }
9189 
9190 static void mlxsw_sp_port_vrf_leave(struct mlxsw_sp *mlxsw_sp,
9191 				    struct net_device *l3_dev)
9192 {
9193 	struct mlxsw_sp_rif *rif;
9194 
9195 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
9196 	if (!rif)
9197 		return;
9198 	__mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_DOWN, NULL);
9199 }
9200 
9201 static bool mlxsw_sp_is_vrf_event(unsigned long event, void *ptr)
9202 {
9203 	struct netdev_notifier_changeupper_info *info = ptr;
9204 
9205 	if (event != NETDEV_PRECHANGEUPPER && event != NETDEV_CHANGEUPPER)
9206 		return false;
9207 	return netif_is_l3_master(info->upper_dev);
9208 }
9209 
9210 static int
9211 mlxsw_sp_netdevice_vrf_event(struct net_device *l3_dev, unsigned long event,
9212 			     struct netdev_notifier_changeupper_info *info)
9213 {
9214 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_lower_get(l3_dev);
9215 	int err = 0;
9216 
9217 	/* We do not create a RIF for a macvlan, but only use it to
9218 	 * direct more MAC addresses to the router.
9219 	 */
9220 	if (!mlxsw_sp || netif_is_macvlan(l3_dev))
9221 		return 0;
9222 
9223 	switch (event) {
9224 	case NETDEV_PRECHANGEUPPER:
9225 		break;
9226 	case NETDEV_CHANGEUPPER:
9227 		if (info->linking) {
9228 			struct netlink_ext_ack *extack;
9229 
9230 			extack = netdev_notifier_info_to_extack(&info->info);
9231 			err = mlxsw_sp_port_vrf_join(mlxsw_sp, l3_dev, extack);
9232 		} else {
9233 			mlxsw_sp_port_vrf_leave(mlxsw_sp, l3_dev);
9234 		}
9235 		break;
9236 	}
9237 
9238 	return err;
9239 }
9240 
9241 static int mlxsw_sp_router_netdevice_event(struct notifier_block *nb,
9242 					   unsigned long event, void *ptr)
9243 {
9244 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
9245 	struct mlxsw_sp_router *router;
9246 	struct mlxsw_sp *mlxsw_sp;
9247 	int err = 0;
9248 
9249 	router = container_of(nb, struct mlxsw_sp_router, netdevice_nb);
9250 	mlxsw_sp = router->mlxsw_sp;
9251 
9252 	mutex_lock(&mlxsw_sp->router->lock);
9253 
9254 	if (mlxsw_sp_is_offload_xstats_event(event))
9255 		err = mlxsw_sp_netdevice_offload_xstats_cmd(mlxsw_sp, dev,
9256 							    event, ptr);
9257 	else if (mlxsw_sp_netdev_is_ipip_ol(mlxsw_sp, dev))
9258 		err = mlxsw_sp_netdevice_ipip_ol_event(mlxsw_sp, dev,
9259 						       event, ptr);
9260 	else if (mlxsw_sp_netdev_is_ipip_ul(mlxsw_sp, dev))
9261 		err = mlxsw_sp_netdevice_ipip_ul_event(mlxsw_sp, dev,
9262 						       event, ptr);
9263 	else if (mlxsw_sp_is_router_event(event))
9264 		err = mlxsw_sp_netdevice_router_port_event(dev, event, ptr);
9265 	else if (mlxsw_sp_is_vrf_event(event, ptr))
9266 		err = mlxsw_sp_netdevice_vrf_event(dev, event, ptr);
9267 
9268 	mutex_unlock(&mlxsw_sp->router->lock);
9269 
9270 	return notifier_from_errno(err);
9271 }
9272 
9273 static int __mlxsw_sp_rif_macvlan_flush(struct net_device *dev,
9274 					struct netdev_nested_priv *priv)
9275 {
9276 	struct mlxsw_sp_rif *rif = (struct mlxsw_sp_rif *)priv->data;
9277 
9278 	if (!netif_is_macvlan(dev))
9279 		return 0;
9280 
9281 	return mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, dev->dev_addr,
9282 				   mlxsw_sp_fid_index(rif->fid), false);
9283 }
9284 
9285 static int mlxsw_sp_rif_macvlan_flush(struct mlxsw_sp_rif *rif)
9286 {
9287 	struct netdev_nested_priv priv = {
9288 		.data = (void *)rif,
9289 	};
9290 
9291 	if (!netif_is_macvlan_port(rif->dev))
9292 		return 0;
9293 
9294 	netdev_warn(rif->dev, "Router interface is deleted. Upper macvlans will not work\n");
9295 	return netdev_walk_all_upper_dev_rcu(rif->dev,
9296 					     __mlxsw_sp_rif_macvlan_flush, &priv);
9297 }
9298 
9299 static void mlxsw_sp_rif_subport_setup(struct mlxsw_sp_rif *rif,
9300 				       const struct mlxsw_sp_rif_params *params)
9301 {
9302 	struct mlxsw_sp_rif_subport *rif_subport;
9303 
9304 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
9305 	refcount_set(&rif_subport->ref_count, 1);
9306 	rif_subport->vid = params->vid;
9307 	rif_subport->lag = params->lag;
9308 	if (params->lag)
9309 		rif_subport->lag_id = params->lag_id;
9310 	else
9311 		rif_subport->system_port = params->system_port;
9312 }
9313 
9314 static int mlxsw_sp_rif_subport_op(struct mlxsw_sp_rif *rif, bool enable)
9315 {
9316 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9317 	struct mlxsw_sp_rif_subport *rif_subport;
9318 	char ritr_pl[MLXSW_REG_RITR_LEN];
9319 	u16 efid;
9320 
9321 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
9322 	mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_SP_IF,
9323 			    rif->rif_index, rif->vr_id, rif->dev->mtu);
9324 	mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
9325 	mlxsw_reg_ritr_if_mac_profile_id_set(ritr_pl, rif->mac_profile_id);
9326 	efid = mlxsw_sp_fid_index(rif->fid);
9327 	mlxsw_reg_ritr_sp_if_pack(ritr_pl, rif_subport->lag,
9328 				  rif_subport->lag ? rif_subport->lag_id :
9329 						     rif_subport->system_port,
9330 				  efid, 0);
9331 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
9332 }
9333 
9334 static int mlxsw_sp_rif_subport_configure(struct mlxsw_sp_rif *rif,
9335 					  struct netlink_ext_ack *extack)
9336 {
9337 	u8 mac_profile;
9338 	int err;
9339 
9340 	err = mlxsw_sp_rif_mac_profile_get(rif->mlxsw_sp, rif->addr,
9341 					   &mac_profile, extack);
9342 	if (err)
9343 		return err;
9344 	rif->mac_profile_id = mac_profile;
9345 
9346 	err = mlxsw_sp_rif_subport_op(rif, true);
9347 	if (err)
9348 		goto err_rif_subport_op;
9349 
9350 	err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
9351 				  mlxsw_sp_fid_index(rif->fid), true);
9352 	if (err)
9353 		goto err_rif_fdb_op;
9354 
9355 	err = mlxsw_sp_fid_rif_set(rif->fid, rif);
9356 	if (err)
9357 		goto err_fid_rif_set;
9358 
9359 	return 0;
9360 
9361 err_fid_rif_set:
9362 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
9363 			    mlxsw_sp_fid_index(rif->fid), false);
9364 err_rif_fdb_op:
9365 	mlxsw_sp_rif_subport_op(rif, false);
9366 err_rif_subport_op:
9367 	mlxsw_sp_rif_mac_profile_put(rif->mlxsw_sp, mac_profile);
9368 	return err;
9369 }
9370 
9371 static void mlxsw_sp_rif_subport_deconfigure(struct mlxsw_sp_rif *rif)
9372 {
9373 	struct mlxsw_sp_fid *fid = rif->fid;
9374 
9375 	mlxsw_sp_fid_rif_unset(fid);
9376 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
9377 			    mlxsw_sp_fid_index(fid), false);
9378 	mlxsw_sp_rif_macvlan_flush(rif);
9379 	mlxsw_sp_rif_subport_op(rif, false);
9380 	mlxsw_sp_rif_mac_profile_put(rif->mlxsw_sp, rif->mac_profile_id);
9381 }
9382 
9383 static struct mlxsw_sp_fid *
9384 mlxsw_sp_rif_subport_fid_get(struct mlxsw_sp_rif *rif,
9385 			     struct netlink_ext_ack *extack)
9386 {
9387 	return mlxsw_sp_fid_rfid_get(rif->mlxsw_sp, rif->rif_index);
9388 }
9389 
9390 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_subport_ops = {
9391 	.type			= MLXSW_SP_RIF_TYPE_SUBPORT,
9392 	.rif_size		= sizeof(struct mlxsw_sp_rif_subport),
9393 	.setup			= mlxsw_sp_rif_subport_setup,
9394 	.configure		= mlxsw_sp_rif_subport_configure,
9395 	.deconfigure		= mlxsw_sp_rif_subport_deconfigure,
9396 	.fid_get		= mlxsw_sp_rif_subport_fid_get,
9397 };
9398 
9399 static int mlxsw_sp_rif_fid_op(struct mlxsw_sp_rif *rif, u16 fid, bool enable)
9400 {
9401 	enum mlxsw_reg_ritr_if_type type = MLXSW_REG_RITR_FID_IF;
9402 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9403 	char ritr_pl[MLXSW_REG_RITR_LEN];
9404 
9405 	mlxsw_reg_ritr_pack(ritr_pl, enable, type, rif->rif_index, rif->vr_id,
9406 			    rif->dev->mtu);
9407 	mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
9408 	mlxsw_reg_ritr_if_mac_profile_id_set(ritr_pl, rif->mac_profile_id);
9409 	mlxsw_reg_ritr_fid_if_fid_set(ritr_pl, fid);
9410 
9411 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
9412 }
9413 
9414 u16 mlxsw_sp_router_port(const struct mlxsw_sp *mlxsw_sp)
9415 {
9416 	return mlxsw_core_max_ports(mlxsw_sp->core) + 1;
9417 }
9418 
9419 static int mlxsw_sp_rif_fid_configure(struct mlxsw_sp_rif *rif,
9420 				      struct netlink_ext_ack *extack)
9421 {
9422 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9423 	u16 fid_index = mlxsw_sp_fid_index(rif->fid);
9424 	u8 mac_profile;
9425 	int err;
9426 
9427 	err = mlxsw_sp_rif_mac_profile_get(mlxsw_sp, rif->addr,
9428 					   &mac_profile, extack);
9429 	if (err)
9430 		return err;
9431 	rif->mac_profile_id = mac_profile;
9432 
9433 	err = mlxsw_sp_rif_fid_op(rif, fid_index, true);
9434 	if (err)
9435 		goto err_rif_fid_op;
9436 
9437 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
9438 				     mlxsw_sp_router_port(mlxsw_sp), true);
9439 	if (err)
9440 		goto err_fid_mc_flood_set;
9441 
9442 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
9443 				     mlxsw_sp_router_port(mlxsw_sp), true);
9444 	if (err)
9445 		goto err_fid_bc_flood_set;
9446 
9447 	err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
9448 				  mlxsw_sp_fid_index(rif->fid), true);
9449 	if (err)
9450 		goto err_rif_fdb_op;
9451 
9452 	err = mlxsw_sp_fid_rif_set(rif->fid, rif);
9453 	if (err)
9454 		goto err_fid_rif_set;
9455 
9456 	return 0;
9457 
9458 err_fid_rif_set:
9459 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
9460 			    mlxsw_sp_fid_index(rif->fid), false);
9461 err_rif_fdb_op:
9462 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
9463 			       mlxsw_sp_router_port(mlxsw_sp), false);
9464 err_fid_bc_flood_set:
9465 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
9466 			       mlxsw_sp_router_port(mlxsw_sp), false);
9467 err_fid_mc_flood_set:
9468 	mlxsw_sp_rif_fid_op(rif, fid_index, false);
9469 err_rif_fid_op:
9470 	mlxsw_sp_rif_mac_profile_put(mlxsw_sp, mac_profile);
9471 	return err;
9472 }
9473 
9474 static void mlxsw_sp_rif_fid_deconfigure(struct mlxsw_sp_rif *rif)
9475 {
9476 	u16 fid_index = mlxsw_sp_fid_index(rif->fid);
9477 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9478 	struct mlxsw_sp_fid *fid = rif->fid;
9479 
9480 	mlxsw_sp_fid_rif_unset(fid);
9481 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
9482 			    mlxsw_sp_fid_index(fid), false);
9483 	mlxsw_sp_rif_macvlan_flush(rif);
9484 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
9485 			       mlxsw_sp_router_port(mlxsw_sp), false);
9486 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
9487 			       mlxsw_sp_router_port(mlxsw_sp), false);
9488 	mlxsw_sp_rif_fid_op(rif, fid_index, false);
9489 	mlxsw_sp_rif_mac_profile_put(rif->mlxsw_sp, rif->mac_profile_id);
9490 }
9491 
9492 static struct mlxsw_sp_fid *
9493 mlxsw_sp_rif_fid_fid_get(struct mlxsw_sp_rif *rif,
9494 			 struct netlink_ext_ack *extack)
9495 {
9496 	return mlxsw_sp_fid_8021d_get(rif->mlxsw_sp, rif->dev->ifindex);
9497 }
9498 
9499 static void mlxsw_sp_rif_fid_fdb_del(struct mlxsw_sp_rif *rif, const char *mac)
9500 {
9501 	struct switchdev_notifier_fdb_info info = {};
9502 	struct net_device *dev;
9503 
9504 	dev = br_fdb_find_port(rif->dev, mac, 0);
9505 	if (!dev)
9506 		return;
9507 
9508 	info.addr = mac;
9509 	info.vid = 0;
9510 	call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_BRIDGE, dev, &info.info,
9511 				 NULL);
9512 }
9513 
9514 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_fid_ops = {
9515 	.type			= MLXSW_SP_RIF_TYPE_FID,
9516 	.rif_size		= sizeof(struct mlxsw_sp_rif),
9517 	.configure		= mlxsw_sp_rif_fid_configure,
9518 	.deconfigure		= mlxsw_sp_rif_fid_deconfigure,
9519 	.fid_get		= mlxsw_sp_rif_fid_fid_get,
9520 	.fdb_del		= mlxsw_sp_rif_fid_fdb_del,
9521 };
9522 
9523 static struct mlxsw_sp_fid *
9524 mlxsw_sp_rif_vlan_fid_get(struct mlxsw_sp_rif *rif,
9525 			  struct netlink_ext_ack *extack)
9526 {
9527 	struct net_device *br_dev;
9528 	u16 vid;
9529 	int err;
9530 
9531 	if (is_vlan_dev(rif->dev)) {
9532 		vid = vlan_dev_vlan_id(rif->dev);
9533 		br_dev = vlan_dev_real_dev(rif->dev);
9534 		if (WARN_ON(!netif_is_bridge_master(br_dev)))
9535 			return ERR_PTR(-EINVAL);
9536 	} else {
9537 		err = br_vlan_get_pvid(rif->dev, &vid);
9538 		if (err < 0 || !vid) {
9539 			NL_SET_ERR_MSG_MOD(extack, "Couldn't determine bridge PVID");
9540 			return ERR_PTR(-EINVAL);
9541 		}
9542 	}
9543 
9544 	return mlxsw_sp_fid_8021q_get(rif->mlxsw_sp, vid);
9545 }
9546 
9547 static void mlxsw_sp_rif_vlan_fdb_del(struct mlxsw_sp_rif *rif, const char *mac)
9548 {
9549 	struct switchdev_notifier_fdb_info info = {};
9550 	u16 vid = mlxsw_sp_fid_8021q_vid(rif->fid);
9551 	struct net_device *br_dev;
9552 	struct net_device *dev;
9553 
9554 	br_dev = is_vlan_dev(rif->dev) ? vlan_dev_real_dev(rif->dev) : rif->dev;
9555 	dev = br_fdb_find_port(br_dev, mac, vid);
9556 	if (!dev)
9557 		return;
9558 
9559 	info.addr = mac;
9560 	info.vid = vid;
9561 	call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_BRIDGE, dev, &info.info,
9562 				 NULL);
9563 }
9564 
9565 static int mlxsw_sp_rif_vlan_op(struct mlxsw_sp_rif *rif, u16 vid, u16 efid,
9566 				bool enable)
9567 {
9568 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9569 	char ritr_pl[MLXSW_REG_RITR_LEN];
9570 
9571 	mlxsw_reg_ritr_vlan_if_pack(ritr_pl, enable, rif->rif_index, rif->vr_id,
9572 				    rif->dev->mtu, rif->dev->dev_addr,
9573 				    rif->mac_profile_id, vid, efid);
9574 
9575 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
9576 }
9577 
9578 static int mlxsw_sp_rif_vlan_configure(struct mlxsw_sp_rif *rif, u16 efid,
9579 				       struct netlink_ext_ack *extack)
9580 {
9581 	u16 vid = mlxsw_sp_fid_8021q_vid(rif->fid);
9582 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9583 	u8 mac_profile;
9584 	int err;
9585 
9586 	err = mlxsw_sp_rif_mac_profile_get(mlxsw_sp, rif->addr,
9587 					   &mac_profile, extack);
9588 	if (err)
9589 		return err;
9590 	rif->mac_profile_id = mac_profile;
9591 
9592 	err = mlxsw_sp_rif_vlan_op(rif, vid, efid, true);
9593 	if (err)
9594 		goto err_rif_vlan_fid_op;
9595 
9596 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
9597 				     mlxsw_sp_router_port(mlxsw_sp), true);
9598 	if (err)
9599 		goto err_fid_mc_flood_set;
9600 
9601 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
9602 				     mlxsw_sp_router_port(mlxsw_sp), true);
9603 	if (err)
9604 		goto err_fid_bc_flood_set;
9605 
9606 	err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
9607 				  mlxsw_sp_fid_index(rif->fid), true);
9608 	if (err)
9609 		goto err_rif_fdb_op;
9610 
9611 	err = mlxsw_sp_fid_rif_set(rif->fid, rif);
9612 	if (err)
9613 		goto err_fid_rif_set;
9614 
9615 	return 0;
9616 
9617 err_fid_rif_set:
9618 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
9619 			    mlxsw_sp_fid_index(rif->fid), false);
9620 err_rif_fdb_op:
9621 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
9622 			       mlxsw_sp_router_port(mlxsw_sp), false);
9623 err_fid_bc_flood_set:
9624 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
9625 			       mlxsw_sp_router_port(mlxsw_sp), false);
9626 err_fid_mc_flood_set:
9627 	mlxsw_sp_rif_vlan_op(rif, vid, 0, false);
9628 err_rif_vlan_fid_op:
9629 	mlxsw_sp_rif_mac_profile_put(mlxsw_sp, mac_profile);
9630 	return err;
9631 }
9632 
9633 static void mlxsw_sp_rif_vlan_deconfigure(struct mlxsw_sp_rif *rif)
9634 {
9635 	u16 vid = mlxsw_sp_fid_8021q_vid(rif->fid);
9636 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9637 
9638 	mlxsw_sp_fid_rif_unset(rif->fid);
9639 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
9640 			    mlxsw_sp_fid_index(rif->fid), false);
9641 	mlxsw_sp_rif_macvlan_flush(rif);
9642 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
9643 			       mlxsw_sp_router_port(mlxsw_sp), false);
9644 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
9645 			       mlxsw_sp_router_port(mlxsw_sp), false);
9646 	mlxsw_sp_rif_vlan_op(rif, vid, 0, false);
9647 	mlxsw_sp_rif_mac_profile_put(rif->mlxsw_sp, rif->mac_profile_id);
9648 }
9649 
9650 static int mlxsw_sp1_rif_vlan_configure(struct mlxsw_sp_rif *rif,
9651 					struct netlink_ext_ack *extack)
9652 {
9653 	return mlxsw_sp_rif_vlan_configure(rif, 0, extack);
9654 }
9655 
9656 static const struct mlxsw_sp_rif_ops mlxsw_sp1_rif_vlan_ops = {
9657 	.type			= MLXSW_SP_RIF_TYPE_VLAN,
9658 	.rif_size		= sizeof(struct mlxsw_sp_rif),
9659 	.configure		= mlxsw_sp1_rif_vlan_configure,
9660 	.deconfigure		= mlxsw_sp_rif_vlan_deconfigure,
9661 	.fid_get		= mlxsw_sp_rif_vlan_fid_get,
9662 	.fdb_del		= mlxsw_sp_rif_vlan_fdb_del,
9663 };
9664 
9665 static int mlxsw_sp2_rif_vlan_configure(struct mlxsw_sp_rif *rif,
9666 					struct netlink_ext_ack *extack)
9667 {
9668 	u16 efid = mlxsw_sp_fid_index(rif->fid);
9669 
9670 	return mlxsw_sp_rif_vlan_configure(rif, efid, extack);
9671 }
9672 
9673 static const struct mlxsw_sp_rif_ops mlxsw_sp2_rif_vlan_ops = {
9674 	.type			= MLXSW_SP_RIF_TYPE_VLAN,
9675 	.rif_size		= sizeof(struct mlxsw_sp_rif),
9676 	.configure		= mlxsw_sp2_rif_vlan_configure,
9677 	.deconfigure		= mlxsw_sp_rif_vlan_deconfigure,
9678 	.fid_get		= mlxsw_sp_rif_vlan_fid_get,
9679 	.fdb_del		= mlxsw_sp_rif_vlan_fdb_del,
9680 };
9681 
9682 static struct mlxsw_sp_rif_ipip_lb *
9683 mlxsw_sp_rif_ipip_lb_rif(struct mlxsw_sp_rif *rif)
9684 {
9685 	return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
9686 }
9687 
9688 static void
9689 mlxsw_sp_rif_ipip_lb_setup(struct mlxsw_sp_rif *rif,
9690 			   const struct mlxsw_sp_rif_params *params)
9691 {
9692 	struct mlxsw_sp_rif_params_ipip_lb *params_lb;
9693 	struct mlxsw_sp_rif_ipip_lb *rif_lb;
9694 
9695 	params_lb = container_of(params, struct mlxsw_sp_rif_params_ipip_lb,
9696 				 common);
9697 	rif_lb = mlxsw_sp_rif_ipip_lb_rif(rif);
9698 	rif_lb->lb_config = params_lb->lb_config;
9699 }
9700 
9701 static int
9702 mlxsw_sp1_rif_ipip_lb_configure(struct mlxsw_sp_rif *rif,
9703 				struct netlink_ext_ack *extack)
9704 {
9705 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
9706 	u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(rif->dev);
9707 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9708 	struct mlxsw_sp_vr *ul_vr;
9709 	int err;
9710 
9711 	ul_vr = mlxsw_sp_vr_get(mlxsw_sp, ul_tb_id, NULL);
9712 	if (IS_ERR(ul_vr))
9713 		return PTR_ERR(ul_vr);
9714 
9715 	err = mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr->id, 0, true);
9716 	if (err)
9717 		goto err_loopback_op;
9718 
9719 	lb_rif->ul_vr_id = ul_vr->id;
9720 	lb_rif->ul_rif_id = 0;
9721 	++ul_vr->rif_count;
9722 	return 0;
9723 
9724 err_loopback_op:
9725 	mlxsw_sp_vr_put(mlxsw_sp, ul_vr);
9726 	return err;
9727 }
9728 
9729 static void mlxsw_sp1_rif_ipip_lb_deconfigure(struct mlxsw_sp_rif *rif)
9730 {
9731 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
9732 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9733 	struct mlxsw_sp_vr *ul_vr;
9734 
9735 	ul_vr = &mlxsw_sp->router->vrs[lb_rif->ul_vr_id];
9736 	mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr->id, 0, false);
9737 
9738 	--ul_vr->rif_count;
9739 	mlxsw_sp_vr_put(mlxsw_sp, ul_vr);
9740 }
9741 
9742 static const struct mlxsw_sp_rif_ops mlxsw_sp1_rif_ipip_lb_ops = {
9743 	.type			= MLXSW_SP_RIF_TYPE_IPIP_LB,
9744 	.rif_size		= sizeof(struct mlxsw_sp_rif_ipip_lb),
9745 	.setup                  = mlxsw_sp_rif_ipip_lb_setup,
9746 	.configure		= mlxsw_sp1_rif_ipip_lb_configure,
9747 	.deconfigure		= mlxsw_sp1_rif_ipip_lb_deconfigure,
9748 };
9749 
9750 static const struct mlxsw_sp_rif_ops *mlxsw_sp1_rif_ops_arr[] = {
9751 	[MLXSW_SP_RIF_TYPE_SUBPORT]	= &mlxsw_sp_rif_subport_ops,
9752 	[MLXSW_SP_RIF_TYPE_VLAN]	= &mlxsw_sp1_rif_vlan_ops,
9753 	[MLXSW_SP_RIF_TYPE_FID]		= &mlxsw_sp_rif_fid_ops,
9754 	[MLXSW_SP_RIF_TYPE_IPIP_LB]	= &mlxsw_sp1_rif_ipip_lb_ops,
9755 };
9756 
9757 static int
9758 mlxsw_sp_rif_ipip_lb_ul_rif_op(struct mlxsw_sp_rif *ul_rif, bool enable)
9759 {
9760 	struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
9761 	char ritr_pl[MLXSW_REG_RITR_LEN];
9762 
9763 	mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
9764 			    ul_rif->rif_index, ul_rif->vr_id, IP_MAX_MTU);
9765 	mlxsw_reg_ritr_loopback_protocol_set(ritr_pl,
9766 					     MLXSW_REG_RITR_LOOPBACK_GENERIC);
9767 
9768 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
9769 }
9770 
9771 static struct mlxsw_sp_rif *
9772 mlxsw_sp_ul_rif_create(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr,
9773 		       struct netlink_ext_ack *extack)
9774 {
9775 	struct mlxsw_sp_rif *ul_rif;
9776 	u16 rif_index;
9777 	int err;
9778 
9779 	err = mlxsw_sp_rif_index_alloc(mlxsw_sp, &rif_index);
9780 	if (err) {
9781 		NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported router interfaces");
9782 		return ERR_PTR(err);
9783 	}
9784 
9785 	ul_rif = mlxsw_sp_rif_alloc(sizeof(*ul_rif), rif_index, vr->id, NULL);
9786 	if (!ul_rif)
9787 		return ERR_PTR(-ENOMEM);
9788 
9789 	mlxsw_sp->router->rifs[rif_index] = ul_rif;
9790 	ul_rif->mlxsw_sp = mlxsw_sp;
9791 	err = mlxsw_sp_rif_ipip_lb_ul_rif_op(ul_rif, true);
9792 	if (err)
9793 		goto ul_rif_op_err;
9794 
9795 	atomic_inc(&mlxsw_sp->router->rifs_count);
9796 	return ul_rif;
9797 
9798 ul_rif_op_err:
9799 	mlxsw_sp->router->rifs[rif_index] = NULL;
9800 	kfree(ul_rif);
9801 	return ERR_PTR(err);
9802 }
9803 
9804 static void mlxsw_sp_ul_rif_destroy(struct mlxsw_sp_rif *ul_rif)
9805 {
9806 	struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
9807 
9808 	atomic_dec(&mlxsw_sp->router->rifs_count);
9809 	mlxsw_sp_rif_ipip_lb_ul_rif_op(ul_rif, false);
9810 	mlxsw_sp->router->rifs[ul_rif->rif_index] = NULL;
9811 	kfree(ul_rif);
9812 }
9813 
9814 static struct mlxsw_sp_rif *
9815 mlxsw_sp_ul_rif_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
9816 		    struct netlink_ext_ack *extack)
9817 {
9818 	struct mlxsw_sp_vr *vr;
9819 	int err;
9820 
9821 	vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id, extack);
9822 	if (IS_ERR(vr))
9823 		return ERR_CAST(vr);
9824 
9825 	if (refcount_inc_not_zero(&vr->ul_rif_refcnt))
9826 		return vr->ul_rif;
9827 
9828 	vr->ul_rif = mlxsw_sp_ul_rif_create(mlxsw_sp, vr, extack);
9829 	if (IS_ERR(vr->ul_rif)) {
9830 		err = PTR_ERR(vr->ul_rif);
9831 		goto err_ul_rif_create;
9832 	}
9833 
9834 	vr->rif_count++;
9835 	refcount_set(&vr->ul_rif_refcnt, 1);
9836 
9837 	return vr->ul_rif;
9838 
9839 err_ul_rif_create:
9840 	mlxsw_sp_vr_put(mlxsw_sp, vr);
9841 	return ERR_PTR(err);
9842 }
9843 
9844 static void mlxsw_sp_ul_rif_put(struct mlxsw_sp_rif *ul_rif)
9845 {
9846 	struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
9847 	struct mlxsw_sp_vr *vr;
9848 
9849 	vr = &mlxsw_sp->router->vrs[ul_rif->vr_id];
9850 
9851 	if (!refcount_dec_and_test(&vr->ul_rif_refcnt))
9852 		return;
9853 
9854 	vr->rif_count--;
9855 	mlxsw_sp_ul_rif_destroy(ul_rif);
9856 	mlxsw_sp_vr_put(mlxsw_sp, vr);
9857 }
9858 
9859 int mlxsw_sp_router_ul_rif_get(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
9860 			       u16 *ul_rif_index)
9861 {
9862 	struct mlxsw_sp_rif *ul_rif;
9863 	int err = 0;
9864 
9865 	mutex_lock(&mlxsw_sp->router->lock);
9866 	ul_rif = mlxsw_sp_ul_rif_get(mlxsw_sp, ul_tb_id, NULL);
9867 	if (IS_ERR(ul_rif)) {
9868 		err = PTR_ERR(ul_rif);
9869 		goto out;
9870 	}
9871 	*ul_rif_index = ul_rif->rif_index;
9872 out:
9873 	mutex_unlock(&mlxsw_sp->router->lock);
9874 	return err;
9875 }
9876 
9877 void mlxsw_sp_router_ul_rif_put(struct mlxsw_sp *mlxsw_sp, u16 ul_rif_index)
9878 {
9879 	struct mlxsw_sp_rif *ul_rif;
9880 
9881 	mutex_lock(&mlxsw_sp->router->lock);
9882 	ul_rif = mlxsw_sp->router->rifs[ul_rif_index];
9883 	if (WARN_ON(!ul_rif))
9884 		goto out;
9885 
9886 	mlxsw_sp_ul_rif_put(ul_rif);
9887 out:
9888 	mutex_unlock(&mlxsw_sp->router->lock);
9889 }
9890 
9891 static int
9892 mlxsw_sp2_rif_ipip_lb_configure(struct mlxsw_sp_rif *rif,
9893 				struct netlink_ext_ack *extack)
9894 {
9895 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
9896 	u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(rif->dev);
9897 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9898 	struct mlxsw_sp_rif *ul_rif;
9899 	int err;
9900 
9901 	ul_rif = mlxsw_sp_ul_rif_get(mlxsw_sp, ul_tb_id, NULL);
9902 	if (IS_ERR(ul_rif))
9903 		return PTR_ERR(ul_rif);
9904 
9905 	err = mlxsw_sp_rif_ipip_lb_op(lb_rif, 0, ul_rif->rif_index, true);
9906 	if (err)
9907 		goto err_loopback_op;
9908 
9909 	lb_rif->ul_vr_id = 0;
9910 	lb_rif->ul_rif_id = ul_rif->rif_index;
9911 
9912 	return 0;
9913 
9914 err_loopback_op:
9915 	mlxsw_sp_ul_rif_put(ul_rif);
9916 	return err;
9917 }
9918 
9919 static void mlxsw_sp2_rif_ipip_lb_deconfigure(struct mlxsw_sp_rif *rif)
9920 {
9921 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
9922 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
9923 	struct mlxsw_sp_rif *ul_rif;
9924 
9925 	ul_rif = mlxsw_sp_rif_by_index(mlxsw_sp, lb_rif->ul_rif_id);
9926 	mlxsw_sp_rif_ipip_lb_op(lb_rif, 0, lb_rif->ul_rif_id, false);
9927 	mlxsw_sp_ul_rif_put(ul_rif);
9928 }
9929 
9930 static const struct mlxsw_sp_rif_ops mlxsw_sp2_rif_ipip_lb_ops = {
9931 	.type			= MLXSW_SP_RIF_TYPE_IPIP_LB,
9932 	.rif_size		= sizeof(struct mlxsw_sp_rif_ipip_lb),
9933 	.setup                  = mlxsw_sp_rif_ipip_lb_setup,
9934 	.configure		= mlxsw_sp2_rif_ipip_lb_configure,
9935 	.deconfigure		= mlxsw_sp2_rif_ipip_lb_deconfigure,
9936 };
9937 
9938 static const struct mlxsw_sp_rif_ops *mlxsw_sp2_rif_ops_arr[] = {
9939 	[MLXSW_SP_RIF_TYPE_SUBPORT]	= &mlxsw_sp_rif_subport_ops,
9940 	[MLXSW_SP_RIF_TYPE_VLAN]	= &mlxsw_sp2_rif_vlan_ops,
9941 	[MLXSW_SP_RIF_TYPE_FID]		= &mlxsw_sp_rif_fid_ops,
9942 	[MLXSW_SP_RIF_TYPE_IPIP_LB]	= &mlxsw_sp2_rif_ipip_lb_ops,
9943 };
9944 
9945 static int mlxsw_sp_rifs_init(struct mlxsw_sp *mlxsw_sp)
9946 {
9947 	u64 max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
9948 	struct devlink *devlink = priv_to_devlink(mlxsw_sp->core);
9949 	struct mlxsw_core *core = mlxsw_sp->core;
9950 
9951 	if (!MLXSW_CORE_RES_VALID(core, MAX_RIF_MAC_PROFILES))
9952 		return -EIO;
9953 	mlxsw_sp->router->max_rif_mac_profile =
9954 		MLXSW_CORE_RES_GET(core, MAX_RIF_MAC_PROFILES);
9955 
9956 	mlxsw_sp->router->rifs = kcalloc(max_rifs,
9957 					 sizeof(struct mlxsw_sp_rif *),
9958 					 GFP_KERNEL);
9959 	if (!mlxsw_sp->router->rifs)
9960 		return -ENOMEM;
9961 
9962 	idr_init(&mlxsw_sp->router->rif_mac_profiles_idr);
9963 	atomic_set(&mlxsw_sp->router->rif_mac_profiles_count, 0);
9964 	atomic_set(&mlxsw_sp->router->rifs_count, 0);
9965 	devl_resource_occ_get_register(devlink,
9966 				       MLXSW_SP_RESOURCE_RIF_MAC_PROFILES,
9967 				       mlxsw_sp_rif_mac_profiles_occ_get,
9968 				       mlxsw_sp);
9969 	devl_resource_occ_get_register(devlink,
9970 				       MLXSW_SP_RESOURCE_RIFS,
9971 				       mlxsw_sp_rifs_occ_get,
9972 				       mlxsw_sp);
9973 
9974 	return 0;
9975 }
9976 
9977 static void mlxsw_sp_rifs_fini(struct mlxsw_sp *mlxsw_sp)
9978 {
9979 	struct devlink *devlink = priv_to_devlink(mlxsw_sp->core);
9980 	int i;
9981 
9982 	WARN_ON_ONCE(atomic_read(&mlxsw_sp->router->rifs_count));
9983 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
9984 		WARN_ON_ONCE(mlxsw_sp->router->rifs[i]);
9985 
9986 	devl_resource_occ_get_unregister(devlink, MLXSW_SP_RESOURCE_RIFS);
9987 	devl_resource_occ_get_unregister(devlink,
9988 					 MLXSW_SP_RESOURCE_RIF_MAC_PROFILES);
9989 	WARN_ON(!idr_is_empty(&mlxsw_sp->router->rif_mac_profiles_idr));
9990 	idr_destroy(&mlxsw_sp->router->rif_mac_profiles_idr);
9991 	kfree(mlxsw_sp->router->rifs);
9992 }
9993 
9994 static int
9995 mlxsw_sp_ipip_config_tigcr(struct mlxsw_sp *mlxsw_sp)
9996 {
9997 	char tigcr_pl[MLXSW_REG_TIGCR_LEN];
9998 
9999 	mlxsw_reg_tigcr_pack(tigcr_pl, true, 0);
10000 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tigcr), tigcr_pl);
10001 }
10002 
10003 static int mlxsw_sp_ipips_init(struct mlxsw_sp *mlxsw_sp)
10004 {
10005 	int err;
10006 
10007 	INIT_LIST_HEAD(&mlxsw_sp->router->ipip_list);
10008 
10009 	err = mlxsw_sp_ipip_ecn_encap_init(mlxsw_sp);
10010 	if (err)
10011 		return err;
10012 	err = mlxsw_sp_ipip_ecn_decap_init(mlxsw_sp);
10013 	if (err)
10014 		return err;
10015 
10016 	return mlxsw_sp_ipip_config_tigcr(mlxsw_sp);
10017 }
10018 
10019 static int mlxsw_sp1_ipips_init(struct mlxsw_sp *mlxsw_sp)
10020 {
10021 	mlxsw_sp->router->ipip_ops_arr = mlxsw_sp1_ipip_ops_arr;
10022 	return mlxsw_sp_ipips_init(mlxsw_sp);
10023 }
10024 
10025 static int mlxsw_sp2_ipips_init(struct mlxsw_sp *mlxsw_sp)
10026 {
10027 	mlxsw_sp->router->ipip_ops_arr = mlxsw_sp2_ipip_ops_arr;
10028 	return mlxsw_sp_ipips_init(mlxsw_sp);
10029 }
10030 
10031 static void mlxsw_sp_ipips_fini(struct mlxsw_sp *mlxsw_sp)
10032 {
10033 	WARN_ON(!list_empty(&mlxsw_sp->router->ipip_list));
10034 }
10035 
10036 static void mlxsw_sp_router_fib_dump_flush(struct notifier_block *nb)
10037 {
10038 	struct mlxsw_sp_router *router;
10039 
10040 	/* Flush pending FIB notifications and then flush the device's
10041 	 * table before requesting another dump. The FIB notification
10042 	 * block is unregistered, so no need to take RTNL.
10043 	 */
10044 	mlxsw_core_flush_owq();
10045 	router = container_of(nb, struct mlxsw_sp_router, fib_nb);
10046 	mlxsw_sp_router_fib_flush(router->mlxsw_sp);
10047 }
10048 
10049 #ifdef CONFIG_IP_ROUTE_MULTIPATH
10050 struct mlxsw_sp_mp_hash_config {
10051 	DECLARE_BITMAP(headers, __MLXSW_REG_RECR2_HEADER_CNT);
10052 	DECLARE_BITMAP(fields, __MLXSW_REG_RECR2_FIELD_CNT);
10053 	DECLARE_BITMAP(inner_headers, __MLXSW_REG_RECR2_HEADER_CNT);
10054 	DECLARE_BITMAP(inner_fields, __MLXSW_REG_RECR2_INNER_FIELD_CNT);
10055 	bool inc_parsing_depth;
10056 };
10057 
10058 #define MLXSW_SP_MP_HASH_HEADER_SET(_headers, _header) \
10059 	bitmap_set(_headers, MLXSW_REG_RECR2_##_header, 1)
10060 
10061 #define MLXSW_SP_MP_HASH_FIELD_SET(_fields, _field) \
10062 	bitmap_set(_fields, MLXSW_REG_RECR2_##_field, 1)
10063 
10064 #define MLXSW_SP_MP_HASH_FIELD_RANGE_SET(_fields, _field, _nr) \
10065 	bitmap_set(_fields, MLXSW_REG_RECR2_##_field, _nr)
10066 
10067 static void mlxsw_sp_mp_hash_inner_l3(struct mlxsw_sp_mp_hash_config *config)
10068 {
10069 	unsigned long *inner_headers = config->inner_headers;
10070 	unsigned long *inner_fields = config->inner_fields;
10071 
10072 	/* IPv4 inner */
10073 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, IPV4_EN_NOT_TCP_NOT_UDP);
10074 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, IPV4_EN_TCP_UDP);
10075 	MLXSW_SP_MP_HASH_FIELD_RANGE_SET(inner_fields, INNER_IPV4_SIP0, 4);
10076 	MLXSW_SP_MP_HASH_FIELD_RANGE_SET(inner_fields, INNER_IPV4_DIP0, 4);
10077 	/* IPv6 inner */
10078 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, IPV6_EN_NOT_TCP_NOT_UDP);
10079 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, IPV6_EN_TCP_UDP);
10080 	MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_IPV6_SIP0_7);
10081 	MLXSW_SP_MP_HASH_FIELD_RANGE_SET(inner_fields, INNER_IPV6_SIP8, 8);
10082 	MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_IPV6_DIP0_7);
10083 	MLXSW_SP_MP_HASH_FIELD_RANGE_SET(inner_fields, INNER_IPV6_DIP8, 8);
10084 	MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_IPV6_NEXT_HEADER);
10085 	MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_IPV6_FLOW_LABEL);
10086 }
10087 
10088 static void mlxsw_sp_mp4_hash_outer_addr(struct mlxsw_sp_mp_hash_config *config)
10089 {
10090 	unsigned long *headers = config->headers;
10091 	unsigned long *fields = config->fields;
10092 
10093 	MLXSW_SP_MP_HASH_HEADER_SET(headers, IPV4_EN_NOT_TCP_NOT_UDP);
10094 	MLXSW_SP_MP_HASH_HEADER_SET(headers, IPV4_EN_TCP_UDP);
10095 	MLXSW_SP_MP_HASH_FIELD_RANGE_SET(fields, IPV4_SIP0, 4);
10096 	MLXSW_SP_MP_HASH_FIELD_RANGE_SET(fields, IPV4_DIP0, 4);
10097 }
10098 
10099 static void
10100 mlxsw_sp_mp_hash_inner_custom(struct mlxsw_sp_mp_hash_config *config,
10101 			      u32 hash_fields)
10102 {
10103 	unsigned long *inner_headers = config->inner_headers;
10104 	unsigned long *inner_fields = config->inner_fields;
10105 
10106 	/* IPv4 Inner */
10107 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, IPV4_EN_NOT_TCP_NOT_UDP);
10108 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, IPV4_EN_TCP_UDP);
10109 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
10110 		MLXSW_SP_MP_HASH_FIELD_RANGE_SET(inner_fields, INNER_IPV4_SIP0, 4);
10111 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
10112 		MLXSW_SP_MP_HASH_FIELD_RANGE_SET(inner_fields, INNER_IPV4_DIP0, 4);
10113 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
10114 		MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_IPV4_PROTOCOL);
10115 	/* IPv6 inner */
10116 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, IPV6_EN_NOT_TCP_NOT_UDP);
10117 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, IPV6_EN_TCP_UDP);
10118 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) {
10119 		MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_IPV6_SIP0_7);
10120 		MLXSW_SP_MP_HASH_FIELD_RANGE_SET(inner_fields, INNER_IPV6_SIP8, 8);
10121 	}
10122 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) {
10123 		MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_IPV6_DIP0_7);
10124 		MLXSW_SP_MP_HASH_FIELD_RANGE_SET(inner_fields, INNER_IPV6_DIP8, 8);
10125 	}
10126 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
10127 		MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_IPV6_NEXT_HEADER);
10128 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
10129 		MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_IPV6_FLOW_LABEL);
10130 	/* L4 inner */
10131 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, TCP_UDP_EN_IPV4);
10132 	MLXSW_SP_MP_HASH_HEADER_SET(inner_headers, TCP_UDP_EN_IPV6);
10133 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
10134 		MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_TCP_UDP_SPORT);
10135 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
10136 		MLXSW_SP_MP_HASH_FIELD_SET(inner_fields, INNER_TCP_UDP_DPORT);
10137 }
10138 
10139 static void mlxsw_sp_mp4_hash_init(struct mlxsw_sp *mlxsw_sp,
10140 				   struct mlxsw_sp_mp_hash_config *config)
10141 {
10142 	struct net *net = mlxsw_sp_net(mlxsw_sp);
10143 	unsigned long *headers = config->headers;
10144 	unsigned long *fields = config->fields;
10145 	u32 hash_fields;
10146 
10147 	switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
10148 	case 0:
10149 		mlxsw_sp_mp4_hash_outer_addr(config);
10150 		break;
10151 	case 1:
10152 		mlxsw_sp_mp4_hash_outer_addr(config);
10153 		MLXSW_SP_MP_HASH_HEADER_SET(headers, TCP_UDP_EN_IPV4);
10154 		MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV4_PROTOCOL);
10155 		MLXSW_SP_MP_HASH_FIELD_SET(fields, TCP_UDP_SPORT);
10156 		MLXSW_SP_MP_HASH_FIELD_SET(fields, TCP_UDP_DPORT);
10157 		break;
10158 	case 2:
10159 		/* Outer */
10160 		mlxsw_sp_mp4_hash_outer_addr(config);
10161 		/* Inner */
10162 		mlxsw_sp_mp_hash_inner_l3(config);
10163 		break;
10164 	case 3:
10165 		hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
10166 		/* Outer */
10167 		MLXSW_SP_MP_HASH_HEADER_SET(headers, IPV4_EN_NOT_TCP_NOT_UDP);
10168 		MLXSW_SP_MP_HASH_HEADER_SET(headers, IPV4_EN_TCP_UDP);
10169 		MLXSW_SP_MP_HASH_HEADER_SET(headers, TCP_UDP_EN_IPV4);
10170 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
10171 			MLXSW_SP_MP_HASH_FIELD_RANGE_SET(fields, IPV4_SIP0, 4);
10172 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
10173 			MLXSW_SP_MP_HASH_FIELD_RANGE_SET(fields, IPV4_DIP0, 4);
10174 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
10175 			MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV4_PROTOCOL);
10176 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
10177 			MLXSW_SP_MP_HASH_FIELD_SET(fields, TCP_UDP_SPORT);
10178 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
10179 			MLXSW_SP_MP_HASH_FIELD_SET(fields, TCP_UDP_DPORT);
10180 		/* Inner */
10181 		mlxsw_sp_mp_hash_inner_custom(config, hash_fields);
10182 		break;
10183 	}
10184 }
10185 
10186 static void mlxsw_sp_mp6_hash_outer_addr(struct mlxsw_sp_mp_hash_config *config)
10187 {
10188 	unsigned long *headers = config->headers;
10189 	unsigned long *fields = config->fields;
10190 
10191 	MLXSW_SP_MP_HASH_HEADER_SET(headers, IPV6_EN_NOT_TCP_NOT_UDP);
10192 	MLXSW_SP_MP_HASH_HEADER_SET(headers, IPV6_EN_TCP_UDP);
10193 	MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_SIP0_7);
10194 	MLXSW_SP_MP_HASH_FIELD_RANGE_SET(fields, IPV6_SIP8, 8);
10195 	MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_DIP0_7);
10196 	MLXSW_SP_MP_HASH_FIELD_RANGE_SET(fields, IPV6_DIP8, 8);
10197 }
10198 
10199 static void mlxsw_sp_mp6_hash_init(struct mlxsw_sp *mlxsw_sp,
10200 				   struct mlxsw_sp_mp_hash_config *config)
10201 {
10202 	u32 hash_fields = ip6_multipath_hash_fields(mlxsw_sp_net(mlxsw_sp));
10203 	unsigned long *headers = config->headers;
10204 	unsigned long *fields = config->fields;
10205 
10206 	switch (ip6_multipath_hash_policy(mlxsw_sp_net(mlxsw_sp))) {
10207 	case 0:
10208 		mlxsw_sp_mp6_hash_outer_addr(config);
10209 		MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_NEXT_HEADER);
10210 		MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_FLOW_LABEL);
10211 		break;
10212 	case 1:
10213 		mlxsw_sp_mp6_hash_outer_addr(config);
10214 		MLXSW_SP_MP_HASH_HEADER_SET(headers, TCP_UDP_EN_IPV6);
10215 		MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_NEXT_HEADER);
10216 		MLXSW_SP_MP_HASH_FIELD_SET(fields, TCP_UDP_SPORT);
10217 		MLXSW_SP_MP_HASH_FIELD_SET(fields, TCP_UDP_DPORT);
10218 		break;
10219 	case 2:
10220 		/* Outer */
10221 		mlxsw_sp_mp6_hash_outer_addr(config);
10222 		MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_NEXT_HEADER);
10223 		MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_FLOW_LABEL);
10224 		/* Inner */
10225 		mlxsw_sp_mp_hash_inner_l3(config);
10226 		config->inc_parsing_depth = true;
10227 		break;
10228 	case 3:
10229 		/* Outer */
10230 		MLXSW_SP_MP_HASH_HEADER_SET(headers, IPV6_EN_NOT_TCP_NOT_UDP);
10231 		MLXSW_SP_MP_HASH_HEADER_SET(headers, IPV6_EN_TCP_UDP);
10232 		MLXSW_SP_MP_HASH_HEADER_SET(headers, TCP_UDP_EN_IPV6);
10233 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) {
10234 			MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_SIP0_7);
10235 			MLXSW_SP_MP_HASH_FIELD_RANGE_SET(fields, IPV6_SIP8, 8);
10236 		}
10237 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) {
10238 			MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_DIP0_7);
10239 			MLXSW_SP_MP_HASH_FIELD_RANGE_SET(fields, IPV6_DIP8, 8);
10240 		}
10241 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
10242 			MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_NEXT_HEADER);
10243 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
10244 			MLXSW_SP_MP_HASH_FIELD_SET(fields, IPV6_FLOW_LABEL);
10245 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
10246 			MLXSW_SP_MP_HASH_FIELD_SET(fields, TCP_UDP_SPORT);
10247 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
10248 			MLXSW_SP_MP_HASH_FIELD_SET(fields, TCP_UDP_DPORT);
10249 		/* Inner */
10250 		mlxsw_sp_mp_hash_inner_custom(config, hash_fields);
10251 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK)
10252 			config->inc_parsing_depth = true;
10253 		break;
10254 	}
10255 }
10256 
10257 static int mlxsw_sp_mp_hash_parsing_depth_adjust(struct mlxsw_sp *mlxsw_sp,
10258 						 bool old_inc_parsing_depth,
10259 						 bool new_inc_parsing_depth)
10260 {
10261 	int err;
10262 
10263 	if (!old_inc_parsing_depth && new_inc_parsing_depth) {
10264 		err = mlxsw_sp_parsing_depth_inc(mlxsw_sp);
10265 		if (err)
10266 			return err;
10267 		mlxsw_sp->router->inc_parsing_depth = true;
10268 	} else if (old_inc_parsing_depth && !new_inc_parsing_depth) {
10269 		mlxsw_sp_parsing_depth_dec(mlxsw_sp);
10270 		mlxsw_sp->router->inc_parsing_depth = false;
10271 	}
10272 
10273 	return 0;
10274 }
10275 
10276 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
10277 {
10278 	bool old_inc_parsing_depth, new_inc_parsing_depth;
10279 	struct mlxsw_sp_mp_hash_config config = {};
10280 	char recr2_pl[MLXSW_REG_RECR2_LEN];
10281 	unsigned long bit;
10282 	u32 seed;
10283 	int err;
10284 
10285 	seed = jhash(mlxsw_sp->base_mac, sizeof(mlxsw_sp->base_mac), 0);
10286 	mlxsw_reg_recr2_pack(recr2_pl, seed);
10287 	mlxsw_sp_mp4_hash_init(mlxsw_sp, &config);
10288 	mlxsw_sp_mp6_hash_init(mlxsw_sp, &config);
10289 
10290 	old_inc_parsing_depth = mlxsw_sp->router->inc_parsing_depth;
10291 	new_inc_parsing_depth = config.inc_parsing_depth;
10292 	err = mlxsw_sp_mp_hash_parsing_depth_adjust(mlxsw_sp,
10293 						    old_inc_parsing_depth,
10294 						    new_inc_parsing_depth);
10295 	if (err)
10296 		return err;
10297 
10298 	for_each_set_bit(bit, config.headers, __MLXSW_REG_RECR2_HEADER_CNT)
10299 		mlxsw_reg_recr2_outer_header_enables_set(recr2_pl, bit, 1);
10300 	for_each_set_bit(bit, config.fields, __MLXSW_REG_RECR2_FIELD_CNT)
10301 		mlxsw_reg_recr2_outer_header_fields_enable_set(recr2_pl, bit, 1);
10302 	for_each_set_bit(bit, config.inner_headers, __MLXSW_REG_RECR2_HEADER_CNT)
10303 		mlxsw_reg_recr2_inner_header_enables_set(recr2_pl, bit, 1);
10304 	for_each_set_bit(bit, config.inner_fields, __MLXSW_REG_RECR2_INNER_FIELD_CNT)
10305 		mlxsw_reg_recr2_inner_header_fields_enable_set(recr2_pl, bit, 1);
10306 
10307 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(recr2), recr2_pl);
10308 	if (err)
10309 		goto err_reg_write;
10310 
10311 	return 0;
10312 
10313 err_reg_write:
10314 	mlxsw_sp_mp_hash_parsing_depth_adjust(mlxsw_sp, new_inc_parsing_depth,
10315 					      old_inc_parsing_depth);
10316 	return err;
10317 }
10318 #else
10319 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
10320 {
10321 	return 0;
10322 }
10323 #endif
10324 
10325 static int mlxsw_sp_dscp_init(struct mlxsw_sp *mlxsw_sp)
10326 {
10327 	char rdpm_pl[MLXSW_REG_RDPM_LEN];
10328 	unsigned int i;
10329 
10330 	MLXSW_REG_ZERO(rdpm, rdpm_pl);
10331 
10332 	/* HW is determining switch priority based on DSCP-bits, but the
10333 	 * kernel is still doing that based on the ToS. Since there's a
10334 	 * mismatch in bits we need to make sure to translate the right
10335 	 * value ToS would observe, skipping the 2 least-significant ECN bits.
10336 	 */
10337 	for (i = 0; i < MLXSW_REG_RDPM_DSCP_ENTRY_REC_MAX_COUNT; i++)
10338 		mlxsw_reg_rdpm_pack(rdpm_pl, i, rt_tos2priority(i << 2));
10339 
10340 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rdpm), rdpm_pl);
10341 }
10342 
10343 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
10344 {
10345 	struct net *net = mlxsw_sp_net(mlxsw_sp);
10346 	bool usp = net->ipv4.sysctl_ip_fwd_update_priority;
10347 	char rgcr_pl[MLXSW_REG_RGCR_LEN];
10348 	u64 max_rifs;
10349 
10350 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_RIFS))
10351 		return -EIO;
10352 	max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
10353 
10354 	mlxsw_reg_rgcr_pack(rgcr_pl, true, true);
10355 	mlxsw_reg_rgcr_max_router_interfaces_set(rgcr_pl, max_rifs);
10356 	mlxsw_reg_rgcr_usp_set(rgcr_pl, usp);
10357 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
10358 }
10359 
10360 static void __mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
10361 {
10362 	char rgcr_pl[MLXSW_REG_RGCR_LEN];
10363 
10364 	mlxsw_reg_rgcr_pack(rgcr_pl, false, false);
10365 	mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
10366 }
10367 
10368 static int mlxsw_sp_lb_rif_init(struct mlxsw_sp *mlxsw_sp)
10369 {
10370 	u16 lb_rif_index;
10371 	int err;
10372 
10373 	/* Create a generic loopback RIF associated with the main table
10374 	 * (default VRF). Any table can be used, but the main table exists
10375 	 * anyway, so we do not waste resources.
10376 	 */
10377 	err = mlxsw_sp_router_ul_rif_get(mlxsw_sp, RT_TABLE_MAIN,
10378 					 &lb_rif_index);
10379 	if (err)
10380 		return err;
10381 
10382 	mlxsw_sp->router->lb_rif_index = lb_rif_index;
10383 
10384 	return 0;
10385 }
10386 
10387 static void mlxsw_sp_lb_rif_fini(struct mlxsw_sp *mlxsw_sp)
10388 {
10389 	mlxsw_sp_router_ul_rif_put(mlxsw_sp, mlxsw_sp->router->lb_rif_index);
10390 }
10391 
10392 static int mlxsw_sp1_router_init(struct mlxsw_sp *mlxsw_sp)
10393 {
10394 	size_t size_ranges_count = ARRAY_SIZE(mlxsw_sp1_adj_grp_size_ranges);
10395 
10396 	mlxsw_sp->router->rif_ops_arr = mlxsw_sp1_rif_ops_arr;
10397 	mlxsw_sp->router->adj_grp_size_ranges = mlxsw_sp1_adj_grp_size_ranges;
10398 	mlxsw_sp->router->adj_grp_size_ranges_count = size_ranges_count;
10399 
10400 	return 0;
10401 }
10402 
10403 const struct mlxsw_sp_router_ops mlxsw_sp1_router_ops = {
10404 	.init = mlxsw_sp1_router_init,
10405 	.ipips_init = mlxsw_sp1_ipips_init,
10406 };
10407 
10408 static int mlxsw_sp2_router_init(struct mlxsw_sp *mlxsw_sp)
10409 {
10410 	size_t size_ranges_count = ARRAY_SIZE(mlxsw_sp2_adj_grp_size_ranges);
10411 
10412 	mlxsw_sp->router->rif_ops_arr = mlxsw_sp2_rif_ops_arr;
10413 	mlxsw_sp->router->adj_grp_size_ranges = mlxsw_sp2_adj_grp_size_ranges;
10414 	mlxsw_sp->router->adj_grp_size_ranges_count = size_ranges_count;
10415 
10416 	return 0;
10417 }
10418 
10419 const struct mlxsw_sp_router_ops mlxsw_sp2_router_ops = {
10420 	.init = mlxsw_sp2_router_init,
10421 	.ipips_init = mlxsw_sp2_ipips_init,
10422 };
10423 
10424 int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp,
10425 			 struct netlink_ext_ack *extack)
10426 {
10427 	struct mlxsw_sp_router *router;
10428 	int err;
10429 
10430 	router = kzalloc(sizeof(*mlxsw_sp->router), GFP_KERNEL);
10431 	if (!router)
10432 		return -ENOMEM;
10433 	mutex_init(&router->lock);
10434 	mlxsw_sp->router = router;
10435 	router->mlxsw_sp = mlxsw_sp;
10436 
10437 	err = mlxsw_sp->router_ops->init(mlxsw_sp);
10438 	if (err)
10439 		goto err_router_ops_init;
10440 
10441 	INIT_LIST_HEAD(&mlxsw_sp->router->nh_res_grp_list);
10442 	INIT_DELAYED_WORK(&mlxsw_sp->router->nh_grp_activity_dw,
10443 			  mlxsw_sp_nh_grp_activity_work);
10444 	INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_neighs_list);
10445 	err = __mlxsw_sp_router_init(mlxsw_sp);
10446 	if (err)
10447 		goto err_router_init;
10448 
10449 	err = mlxsw_sp_rifs_init(mlxsw_sp);
10450 	if (err)
10451 		goto err_rifs_init;
10452 
10453 	err = mlxsw_sp->router_ops->ipips_init(mlxsw_sp);
10454 	if (err)
10455 		goto err_ipips_init;
10456 
10457 	err = rhashtable_init(&mlxsw_sp->router->nexthop_ht,
10458 			      &mlxsw_sp_nexthop_ht_params);
10459 	if (err)
10460 		goto err_nexthop_ht_init;
10461 
10462 	err = rhashtable_init(&mlxsw_sp->router->nexthop_group_ht,
10463 			      &mlxsw_sp_nexthop_group_ht_params);
10464 	if (err)
10465 		goto err_nexthop_group_ht_init;
10466 
10467 	INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_list);
10468 	err = mlxsw_sp_lpm_init(mlxsw_sp);
10469 	if (err)
10470 		goto err_lpm_init;
10471 
10472 	err = mlxsw_sp_mr_init(mlxsw_sp, &mlxsw_sp_mr_tcam_ops);
10473 	if (err)
10474 		goto err_mr_init;
10475 
10476 	err = mlxsw_sp_vrs_init(mlxsw_sp);
10477 	if (err)
10478 		goto err_vrs_init;
10479 
10480 	err = mlxsw_sp_lb_rif_init(mlxsw_sp);
10481 	if (err)
10482 		goto err_lb_rif_init;
10483 
10484 	err = mlxsw_sp_neigh_init(mlxsw_sp);
10485 	if (err)
10486 		goto err_neigh_init;
10487 
10488 	err = mlxsw_sp_mp_hash_init(mlxsw_sp);
10489 	if (err)
10490 		goto err_mp_hash_init;
10491 
10492 	err = mlxsw_sp_dscp_init(mlxsw_sp);
10493 	if (err)
10494 		goto err_dscp_init;
10495 
10496 	router->inetaddr_nb.notifier_call = mlxsw_sp_inetaddr_event;
10497 	err = register_inetaddr_notifier(&router->inetaddr_nb);
10498 	if (err)
10499 		goto err_register_inetaddr_notifier;
10500 
10501 	router->inet6addr_nb.notifier_call = mlxsw_sp_inet6addr_event;
10502 	err = register_inet6addr_notifier(&router->inet6addr_nb);
10503 	if (err)
10504 		goto err_register_inet6addr_notifier;
10505 
10506 	mlxsw_sp->router->netevent_nb.notifier_call =
10507 		mlxsw_sp_router_netevent_event;
10508 	err = register_netevent_notifier(&mlxsw_sp->router->netevent_nb);
10509 	if (err)
10510 		goto err_register_netevent_notifier;
10511 
10512 	mlxsw_sp->router->nexthop_nb.notifier_call =
10513 		mlxsw_sp_nexthop_obj_event;
10514 	err = register_nexthop_notifier(mlxsw_sp_net(mlxsw_sp),
10515 					&mlxsw_sp->router->nexthop_nb,
10516 					extack);
10517 	if (err)
10518 		goto err_register_nexthop_notifier;
10519 
10520 	mlxsw_sp->router->fib_nb.notifier_call = mlxsw_sp_router_fib_event;
10521 	err = register_fib_notifier(mlxsw_sp_net(mlxsw_sp),
10522 				    &mlxsw_sp->router->fib_nb,
10523 				    mlxsw_sp_router_fib_dump_flush, extack);
10524 	if (err)
10525 		goto err_register_fib_notifier;
10526 
10527 	mlxsw_sp->router->netdevice_nb.notifier_call =
10528 		mlxsw_sp_router_netdevice_event;
10529 	err = register_netdevice_notifier_net(mlxsw_sp_net(mlxsw_sp),
10530 					      &mlxsw_sp->router->netdevice_nb);
10531 	if (err)
10532 		goto err_register_netdev_notifier;
10533 
10534 	return 0;
10535 
10536 err_register_netdev_notifier:
10537 	unregister_fib_notifier(mlxsw_sp_net(mlxsw_sp),
10538 				&mlxsw_sp->router->fib_nb);
10539 err_register_fib_notifier:
10540 	unregister_nexthop_notifier(mlxsw_sp_net(mlxsw_sp),
10541 				    &mlxsw_sp->router->nexthop_nb);
10542 err_register_nexthop_notifier:
10543 	unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
10544 err_register_netevent_notifier:
10545 	unregister_inet6addr_notifier(&router->inet6addr_nb);
10546 err_register_inet6addr_notifier:
10547 	unregister_inetaddr_notifier(&router->inetaddr_nb);
10548 err_register_inetaddr_notifier:
10549 	mlxsw_core_flush_owq();
10550 err_dscp_init:
10551 err_mp_hash_init:
10552 	mlxsw_sp_neigh_fini(mlxsw_sp);
10553 err_neigh_init:
10554 	mlxsw_sp_lb_rif_fini(mlxsw_sp);
10555 err_lb_rif_init:
10556 	mlxsw_sp_vrs_fini(mlxsw_sp);
10557 err_vrs_init:
10558 	mlxsw_sp_mr_fini(mlxsw_sp);
10559 err_mr_init:
10560 	mlxsw_sp_lpm_fini(mlxsw_sp);
10561 err_lpm_init:
10562 	rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
10563 err_nexthop_group_ht_init:
10564 	rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
10565 err_nexthop_ht_init:
10566 	mlxsw_sp_ipips_fini(mlxsw_sp);
10567 err_ipips_init:
10568 	mlxsw_sp_rifs_fini(mlxsw_sp);
10569 err_rifs_init:
10570 	__mlxsw_sp_router_fini(mlxsw_sp);
10571 err_router_init:
10572 	cancel_delayed_work_sync(&mlxsw_sp->router->nh_grp_activity_dw);
10573 err_router_ops_init:
10574 	mutex_destroy(&mlxsw_sp->router->lock);
10575 	kfree(mlxsw_sp->router);
10576 	return err;
10577 }
10578 
10579 void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
10580 {
10581 	unregister_netdevice_notifier_net(mlxsw_sp_net(mlxsw_sp),
10582 					  &mlxsw_sp->router->netdevice_nb);
10583 	unregister_fib_notifier(mlxsw_sp_net(mlxsw_sp),
10584 				&mlxsw_sp->router->fib_nb);
10585 	unregister_nexthop_notifier(mlxsw_sp_net(mlxsw_sp),
10586 				    &mlxsw_sp->router->nexthop_nb);
10587 	unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
10588 	unregister_inet6addr_notifier(&mlxsw_sp->router->inet6addr_nb);
10589 	unregister_inetaddr_notifier(&mlxsw_sp->router->inetaddr_nb);
10590 	mlxsw_core_flush_owq();
10591 	mlxsw_sp_neigh_fini(mlxsw_sp);
10592 	mlxsw_sp_lb_rif_fini(mlxsw_sp);
10593 	mlxsw_sp_vrs_fini(mlxsw_sp);
10594 	mlxsw_sp_mr_fini(mlxsw_sp);
10595 	mlxsw_sp_lpm_fini(mlxsw_sp);
10596 	rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
10597 	rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
10598 	mlxsw_sp_ipips_fini(mlxsw_sp);
10599 	mlxsw_sp_rifs_fini(mlxsw_sp);
10600 	__mlxsw_sp_router_fini(mlxsw_sp);
10601 	cancel_delayed_work_sync(&mlxsw_sp->router->nh_grp_activity_dw);
10602 	mutex_destroy(&mlxsw_sp->router->lock);
10603 	kfree(mlxsw_sp->router);
10604 }
10605