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/ip_fib.h>
25 #include <net/ip6_fib.h>
26 #include <net/nexthop.h>
27 #include <net/fib_rules.h>
28 #include <net/ip_tunnels.h>
29 #include <net/l3mdev.h>
30 #include <net/addrconf.h>
31 #include <net/ndisc.h>
32 #include <net/ipv6.h>
33 #include <net/fib_notifier.h>
34 #include <net/switchdev.h>
35 
36 #include "spectrum.h"
37 #include "core.h"
38 #include "reg.h"
39 #include "spectrum_cnt.h"
40 #include "spectrum_dpipe.h"
41 #include "spectrum_ipip.h"
42 #include "spectrum_mr.h"
43 #include "spectrum_mr_tcam.h"
44 #include "spectrum_router.h"
45 #include "spectrum_span.h"
46 
47 struct mlxsw_sp_fib;
48 struct mlxsw_sp_vr;
49 struct mlxsw_sp_lpm_tree;
50 struct mlxsw_sp_rif_ops;
51 
52 struct mlxsw_sp_rif {
53 	struct list_head nexthop_list;
54 	struct list_head neigh_list;
55 	struct net_device *dev; /* NULL for underlay RIF */
56 	struct mlxsw_sp_fid *fid;
57 	unsigned char addr[ETH_ALEN];
58 	int mtu;
59 	u16 rif_index;
60 	u16 vr_id;
61 	const struct mlxsw_sp_rif_ops *ops;
62 	struct mlxsw_sp *mlxsw_sp;
63 
64 	unsigned int counter_ingress;
65 	bool counter_ingress_valid;
66 	unsigned int counter_egress;
67 	bool counter_egress_valid;
68 };
69 
70 struct mlxsw_sp_rif_params {
71 	struct net_device *dev;
72 	union {
73 		u16 system_port;
74 		u16 lag_id;
75 	};
76 	u16 vid;
77 	bool lag;
78 };
79 
80 struct mlxsw_sp_rif_subport {
81 	struct mlxsw_sp_rif common;
82 	refcount_t ref_count;
83 	union {
84 		u16 system_port;
85 		u16 lag_id;
86 	};
87 	u16 vid;
88 	bool lag;
89 };
90 
91 struct mlxsw_sp_rif_ipip_lb {
92 	struct mlxsw_sp_rif common;
93 	struct mlxsw_sp_rif_ipip_lb_config lb_config;
94 	u16 ul_vr_id; /* Reserved for Spectrum-2. */
95 	u16 ul_rif_id; /* Reserved for Spectrum. */
96 };
97 
98 struct mlxsw_sp_rif_params_ipip_lb {
99 	struct mlxsw_sp_rif_params common;
100 	struct mlxsw_sp_rif_ipip_lb_config lb_config;
101 };
102 
103 struct mlxsw_sp_rif_ops {
104 	enum mlxsw_sp_rif_type type;
105 	size_t rif_size;
106 
107 	void (*setup)(struct mlxsw_sp_rif *rif,
108 		      const struct mlxsw_sp_rif_params *params);
109 	int (*configure)(struct mlxsw_sp_rif *rif);
110 	void (*deconfigure)(struct mlxsw_sp_rif *rif);
111 	struct mlxsw_sp_fid * (*fid_get)(struct mlxsw_sp_rif *rif,
112 					 struct netlink_ext_ack *extack);
113 	void (*fdb_del)(struct mlxsw_sp_rif *rif, const char *mac);
114 };
115 
116 static struct mlxsw_sp_rif *
117 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
118 			 const struct net_device *dev);
119 static void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif);
120 static void mlxsw_sp_lpm_tree_hold(struct mlxsw_sp_lpm_tree *lpm_tree);
121 static void mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
122 				  struct mlxsw_sp_lpm_tree *lpm_tree);
123 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
124 				     const struct mlxsw_sp_fib *fib,
125 				     u8 tree_id);
126 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
127 				       const struct mlxsw_sp_fib *fib);
128 
129 static unsigned int *
130 mlxsw_sp_rif_p_counter_get(struct mlxsw_sp_rif *rif,
131 			   enum mlxsw_sp_rif_counter_dir dir)
132 {
133 	switch (dir) {
134 	case MLXSW_SP_RIF_COUNTER_EGRESS:
135 		return &rif->counter_egress;
136 	case MLXSW_SP_RIF_COUNTER_INGRESS:
137 		return &rif->counter_ingress;
138 	}
139 	return NULL;
140 }
141 
142 static bool
143 mlxsw_sp_rif_counter_valid_get(struct mlxsw_sp_rif *rif,
144 			       enum mlxsw_sp_rif_counter_dir dir)
145 {
146 	switch (dir) {
147 	case MLXSW_SP_RIF_COUNTER_EGRESS:
148 		return rif->counter_egress_valid;
149 	case MLXSW_SP_RIF_COUNTER_INGRESS:
150 		return rif->counter_ingress_valid;
151 	}
152 	return false;
153 }
154 
155 static void
156 mlxsw_sp_rif_counter_valid_set(struct mlxsw_sp_rif *rif,
157 			       enum mlxsw_sp_rif_counter_dir dir,
158 			       bool valid)
159 {
160 	switch (dir) {
161 	case MLXSW_SP_RIF_COUNTER_EGRESS:
162 		rif->counter_egress_valid = valid;
163 		break;
164 	case MLXSW_SP_RIF_COUNTER_INGRESS:
165 		rif->counter_ingress_valid = valid;
166 		break;
167 	}
168 }
169 
170 static int mlxsw_sp_rif_counter_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
171 				     unsigned int counter_index, bool enable,
172 				     enum mlxsw_sp_rif_counter_dir dir)
173 {
174 	char ritr_pl[MLXSW_REG_RITR_LEN];
175 	bool is_egress = false;
176 	int err;
177 
178 	if (dir == MLXSW_SP_RIF_COUNTER_EGRESS)
179 		is_egress = true;
180 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
181 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
182 	if (err)
183 		return err;
184 
185 	mlxsw_reg_ritr_counter_pack(ritr_pl, counter_index, enable,
186 				    is_egress);
187 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
188 }
189 
190 int mlxsw_sp_rif_counter_value_get(struct mlxsw_sp *mlxsw_sp,
191 				   struct mlxsw_sp_rif *rif,
192 				   enum mlxsw_sp_rif_counter_dir dir, u64 *cnt)
193 {
194 	char ricnt_pl[MLXSW_REG_RICNT_LEN];
195 	unsigned int *p_counter_index;
196 	bool valid;
197 	int err;
198 
199 	valid = mlxsw_sp_rif_counter_valid_get(rif, dir);
200 	if (!valid)
201 		return -EINVAL;
202 
203 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
204 	if (!p_counter_index)
205 		return -EINVAL;
206 	mlxsw_reg_ricnt_pack(ricnt_pl, *p_counter_index,
207 			     MLXSW_REG_RICNT_OPCODE_NOP);
208 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
209 	if (err)
210 		return err;
211 	*cnt = mlxsw_reg_ricnt_good_unicast_packets_get(ricnt_pl);
212 	return 0;
213 }
214 
215 static int mlxsw_sp_rif_counter_clear(struct mlxsw_sp *mlxsw_sp,
216 				      unsigned int counter_index)
217 {
218 	char ricnt_pl[MLXSW_REG_RICNT_LEN];
219 
220 	mlxsw_reg_ricnt_pack(ricnt_pl, counter_index,
221 			     MLXSW_REG_RICNT_OPCODE_CLEAR);
222 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
223 }
224 
225 int mlxsw_sp_rif_counter_alloc(struct mlxsw_sp *mlxsw_sp,
226 			       struct mlxsw_sp_rif *rif,
227 			       enum mlxsw_sp_rif_counter_dir dir)
228 {
229 	unsigned int *p_counter_index;
230 	int err;
231 
232 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
233 	if (!p_counter_index)
234 		return -EINVAL;
235 	err = mlxsw_sp_counter_alloc(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
236 				     p_counter_index);
237 	if (err)
238 		return err;
239 
240 	err = mlxsw_sp_rif_counter_clear(mlxsw_sp, *p_counter_index);
241 	if (err)
242 		goto err_counter_clear;
243 
244 	err = mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
245 					*p_counter_index, true, dir);
246 	if (err)
247 		goto err_counter_edit;
248 	mlxsw_sp_rif_counter_valid_set(rif, dir, true);
249 	return 0;
250 
251 err_counter_edit:
252 err_counter_clear:
253 	mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
254 			      *p_counter_index);
255 	return err;
256 }
257 
258 void mlxsw_sp_rif_counter_free(struct mlxsw_sp *mlxsw_sp,
259 			       struct mlxsw_sp_rif *rif,
260 			       enum mlxsw_sp_rif_counter_dir dir)
261 {
262 	unsigned int *p_counter_index;
263 
264 	if (!mlxsw_sp_rif_counter_valid_get(rif, dir))
265 		return;
266 
267 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
268 	if (WARN_ON(!p_counter_index))
269 		return;
270 	mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
271 				  *p_counter_index, false, dir);
272 	mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
273 			      *p_counter_index);
274 	mlxsw_sp_rif_counter_valid_set(rif, dir, false);
275 }
276 
277 static void mlxsw_sp_rif_counters_alloc(struct mlxsw_sp_rif *rif)
278 {
279 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
280 	struct devlink *devlink;
281 
282 	devlink = priv_to_devlink(mlxsw_sp->core);
283 	if (!devlink_dpipe_table_counter_enabled(devlink,
284 						 MLXSW_SP_DPIPE_TABLE_NAME_ERIF))
285 		return;
286 	mlxsw_sp_rif_counter_alloc(mlxsw_sp, rif, MLXSW_SP_RIF_COUNTER_EGRESS);
287 }
288 
289 static void mlxsw_sp_rif_counters_free(struct mlxsw_sp_rif *rif)
290 {
291 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
292 
293 	mlxsw_sp_rif_counter_free(mlxsw_sp, rif, MLXSW_SP_RIF_COUNTER_EGRESS);
294 }
295 
296 #define MLXSW_SP_PREFIX_COUNT (sizeof(struct in6_addr) * BITS_PER_BYTE + 1)
297 
298 struct mlxsw_sp_prefix_usage {
299 	DECLARE_BITMAP(b, MLXSW_SP_PREFIX_COUNT);
300 };
301 
302 #define mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) \
303 	for_each_set_bit(prefix, (prefix_usage)->b, MLXSW_SP_PREFIX_COUNT)
304 
305 static bool
306 mlxsw_sp_prefix_usage_eq(struct mlxsw_sp_prefix_usage *prefix_usage1,
307 			 struct mlxsw_sp_prefix_usage *prefix_usage2)
308 {
309 	return !memcmp(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
310 }
311 
312 static void
313 mlxsw_sp_prefix_usage_cpy(struct mlxsw_sp_prefix_usage *prefix_usage1,
314 			  struct mlxsw_sp_prefix_usage *prefix_usage2)
315 {
316 	memcpy(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
317 }
318 
319 static void
320 mlxsw_sp_prefix_usage_set(struct mlxsw_sp_prefix_usage *prefix_usage,
321 			  unsigned char prefix_len)
322 {
323 	set_bit(prefix_len, prefix_usage->b);
324 }
325 
326 static void
327 mlxsw_sp_prefix_usage_clear(struct mlxsw_sp_prefix_usage *prefix_usage,
328 			    unsigned char prefix_len)
329 {
330 	clear_bit(prefix_len, prefix_usage->b);
331 }
332 
333 struct mlxsw_sp_fib_key {
334 	unsigned char addr[sizeof(struct in6_addr)];
335 	unsigned char prefix_len;
336 };
337 
338 enum mlxsw_sp_fib_entry_type {
339 	MLXSW_SP_FIB_ENTRY_TYPE_REMOTE,
340 	MLXSW_SP_FIB_ENTRY_TYPE_LOCAL,
341 	MLXSW_SP_FIB_ENTRY_TYPE_TRAP,
342 	MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE,
343 	MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE,
344 
345 	/* This is a special case of local delivery, where a packet should be
346 	 * decapsulated on reception. Note that there is no corresponding ENCAP,
347 	 * because that's a type of next hop, not of FIB entry. (There can be
348 	 * several next hops in a REMOTE entry, and some of them may be
349 	 * encapsulating entries.)
350 	 */
351 	MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP,
352 	MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP,
353 };
354 
355 struct mlxsw_sp_nexthop_group;
356 struct mlxsw_sp_fib_entry;
357 
358 struct mlxsw_sp_fib_node {
359 	struct mlxsw_sp_fib_entry *fib_entry;
360 	struct list_head list;
361 	struct rhash_head ht_node;
362 	struct mlxsw_sp_fib *fib;
363 	struct mlxsw_sp_fib_key key;
364 };
365 
366 struct mlxsw_sp_fib_entry_decap {
367 	struct mlxsw_sp_ipip_entry *ipip_entry;
368 	u32 tunnel_index;
369 };
370 
371 struct mlxsw_sp_fib_entry {
372 	struct mlxsw_sp_fib_node *fib_node;
373 	enum mlxsw_sp_fib_entry_type type;
374 	struct list_head nexthop_group_node;
375 	struct mlxsw_sp_nexthop_group *nh_group;
376 	struct mlxsw_sp_fib_entry_decap decap; /* Valid for decap entries. */
377 };
378 
379 struct mlxsw_sp_fib4_entry {
380 	struct mlxsw_sp_fib_entry common;
381 	u32 tb_id;
382 	u32 prio;
383 	u8 tos;
384 	u8 type;
385 };
386 
387 struct mlxsw_sp_fib6_entry {
388 	struct mlxsw_sp_fib_entry common;
389 	struct list_head rt6_list;
390 	unsigned int nrt6;
391 };
392 
393 struct mlxsw_sp_rt6 {
394 	struct list_head list;
395 	struct fib6_info *rt;
396 };
397 
398 struct mlxsw_sp_lpm_tree {
399 	u8 id; /* tree ID */
400 	unsigned int ref_count;
401 	enum mlxsw_sp_l3proto proto;
402 	unsigned long prefix_ref_count[MLXSW_SP_PREFIX_COUNT];
403 	struct mlxsw_sp_prefix_usage prefix_usage;
404 };
405 
406 struct mlxsw_sp_fib {
407 	struct rhashtable ht;
408 	struct list_head node_list;
409 	struct mlxsw_sp_vr *vr;
410 	struct mlxsw_sp_lpm_tree *lpm_tree;
411 	enum mlxsw_sp_l3proto proto;
412 };
413 
414 struct mlxsw_sp_vr {
415 	u16 id; /* virtual router ID */
416 	u32 tb_id; /* kernel fib table id */
417 	unsigned int rif_count;
418 	struct mlxsw_sp_fib *fib4;
419 	struct mlxsw_sp_fib *fib6;
420 	struct mlxsw_sp_mr_table *mr_table[MLXSW_SP_L3_PROTO_MAX];
421 	struct mlxsw_sp_rif *ul_rif;
422 	refcount_t ul_rif_refcnt;
423 };
424 
425 static const struct rhashtable_params mlxsw_sp_fib_ht_params;
426 
427 static struct mlxsw_sp_fib *mlxsw_sp_fib_create(struct mlxsw_sp *mlxsw_sp,
428 						struct mlxsw_sp_vr *vr,
429 						enum mlxsw_sp_l3proto proto)
430 {
431 	struct mlxsw_sp_lpm_tree *lpm_tree;
432 	struct mlxsw_sp_fib *fib;
433 	int err;
434 
435 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[proto];
436 	fib = kzalloc(sizeof(*fib), GFP_KERNEL);
437 	if (!fib)
438 		return ERR_PTR(-ENOMEM);
439 	err = rhashtable_init(&fib->ht, &mlxsw_sp_fib_ht_params);
440 	if (err)
441 		goto err_rhashtable_init;
442 	INIT_LIST_HEAD(&fib->node_list);
443 	fib->proto = proto;
444 	fib->vr = vr;
445 	fib->lpm_tree = lpm_tree;
446 	mlxsw_sp_lpm_tree_hold(lpm_tree);
447 	err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, lpm_tree->id);
448 	if (err)
449 		goto err_lpm_tree_bind;
450 	return fib;
451 
452 err_lpm_tree_bind:
453 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
454 err_rhashtable_init:
455 	kfree(fib);
456 	return ERR_PTR(err);
457 }
458 
459 static void mlxsw_sp_fib_destroy(struct mlxsw_sp *mlxsw_sp,
460 				 struct mlxsw_sp_fib *fib)
461 {
462 	mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, fib);
463 	mlxsw_sp_lpm_tree_put(mlxsw_sp, fib->lpm_tree);
464 	WARN_ON(!list_empty(&fib->node_list));
465 	rhashtable_destroy(&fib->ht);
466 	kfree(fib);
467 }
468 
469 static struct mlxsw_sp_lpm_tree *
470 mlxsw_sp_lpm_tree_find_unused(struct mlxsw_sp *mlxsw_sp)
471 {
472 	static struct mlxsw_sp_lpm_tree *lpm_tree;
473 	int i;
474 
475 	for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
476 		lpm_tree = &mlxsw_sp->router->lpm.trees[i];
477 		if (lpm_tree->ref_count == 0)
478 			return lpm_tree;
479 	}
480 	return NULL;
481 }
482 
483 static int mlxsw_sp_lpm_tree_alloc(struct mlxsw_sp *mlxsw_sp,
484 				   struct mlxsw_sp_lpm_tree *lpm_tree)
485 {
486 	char ralta_pl[MLXSW_REG_RALTA_LEN];
487 
488 	mlxsw_reg_ralta_pack(ralta_pl, true,
489 			     (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
490 			     lpm_tree->id);
491 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
492 }
493 
494 static void mlxsw_sp_lpm_tree_free(struct mlxsw_sp *mlxsw_sp,
495 				   struct mlxsw_sp_lpm_tree *lpm_tree)
496 {
497 	char ralta_pl[MLXSW_REG_RALTA_LEN];
498 
499 	mlxsw_reg_ralta_pack(ralta_pl, false,
500 			     (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
501 			     lpm_tree->id);
502 	mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
503 }
504 
505 static int
506 mlxsw_sp_lpm_tree_left_struct_set(struct mlxsw_sp *mlxsw_sp,
507 				  struct mlxsw_sp_prefix_usage *prefix_usage,
508 				  struct mlxsw_sp_lpm_tree *lpm_tree)
509 {
510 	char ralst_pl[MLXSW_REG_RALST_LEN];
511 	u8 root_bin = 0;
512 	u8 prefix;
513 	u8 last_prefix = MLXSW_REG_RALST_BIN_NO_CHILD;
514 
515 	mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage)
516 		root_bin = prefix;
517 
518 	mlxsw_reg_ralst_pack(ralst_pl, root_bin, lpm_tree->id);
519 	mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) {
520 		if (prefix == 0)
521 			continue;
522 		mlxsw_reg_ralst_bin_pack(ralst_pl, prefix, last_prefix,
523 					 MLXSW_REG_RALST_BIN_NO_CHILD);
524 		last_prefix = prefix;
525 	}
526 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
527 }
528 
529 static struct mlxsw_sp_lpm_tree *
530 mlxsw_sp_lpm_tree_create(struct mlxsw_sp *mlxsw_sp,
531 			 struct mlxsw_sp_prefix_usage *prefix_usage,
532 			 enum mlxsw_sp_l3proto proto)
533 {
534 	struct mlxsw_sp_lpm_tree *lpm_tree;
535 	int err;
536 
537 	lpm_tree = mlxsw_sp_lpm_tree_find_unused(mlxsw_sp);
538 	if (!lpm_tree)
539 		return ERR_PTR(-EBUSY);
540 	lpm_tree->proto = proto;
541 	err = mlxsw_sp_lpm_tree_alloc(mlxsw_sp, lpm_tree);
542 	if (err)
543 		return ERR_PTR(err);
544 
545 	err = mlxsw_sp_lpm_tree_left_struct_set(mlxsw_sp, prefix_usage,
546 						lpm_tree);
547 	if (err)
548 		goto err_left_struct_set;
549 	memcpy(&lpm_tree->prefix_usage, prefix_usage,
550 	       sizeof(lpm_tree->prefix_usage));
551 	memset(&lpm_tree->prefix_ref_count, 0,
552 	       sizeof(lpm_tree->prefix_ref_count));
553 	lpm_tree->ref_count = 1;
554 	return lpm_tree;
555 
556 err_left_struct_set:
557 	mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
558 	return ERR_PTR(err);
559 }
560 
561 static void mlxsw_sp_lpm_tree_destroy(struct mlxsw_sp *mlxsw_sp,
562 				      struct mlxsw_sp_lpm_tree *lpm_tree)
563 {
564 	mlxsw_sp_lpm_tree_free(mlxsw_sp, lpm_tree);
565 }
566 
567 static struct mlxsw_sp_lpm_tree *
568 mlxsw_sp_lpm_tree_get(struct mlxsw_sp *mlxsw_sp,
569 		      struct mlxsw_sp_prefix_usage *prefix_usage,
570 		      enum mlxsw_sp_l3proto proto)
571 {
572 	struct mlxsw_sp_lpm_tree *lpm_tree;
573 	int i;
574 
575 	for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
576 		lpm_tree = &mlxsw_sp->router->lpm.trees[i];
577 		if (lpm_tree->ref_count != 0 &&
578 		    lpm_tree->proto == proto &&
579 		    mlxsw_sp_prefix_usage_eq(&lpm_tree->prefix_usage,
580 					     prefix_usage)) {
581 			mlxsw_sp_lpm_tree_hold(lpm_tree);
582 			return lpm_tree;
583 		}
584 	}
585 	return mlxsw_sp_lpm_tree_create(mlxsw_sp, prefix_usage, proto);
586 }
587 
588 static void mlxsw_sp_lpm_tree_hold(struct mlxsw_sp_lpm_tree *lpm_tree)
589 {
590 	lpm_tree->ref_count++;
591 }
592 
593 static void mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
594 				  struct mlxsw_sp_lpm_tree *lpm_tree)
595 {
596 	if (--lpm_tree->ref_count == 0)
597 		mlxsw_sp_lpm_tree_destroy(mlxsw_sp, lpm_tree);
598 }
599 
600 #define MLXSW_SP_LPM_TREE_MIN 1 /* tree 0 is reserved */
601 
602 static int mlxsw_sp_lpm_init(struct mlxsw_sp *mlxsw_sp)
603 {
604 	struct mlxsw_sp_prefix_usage req_prefix_usage = {{ 0 } };
605 	struct mlxsw_sp_lpm_tree *lpm_tree;
606 	u64 max_trees;
607 	int err, i;
608 
609 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_LPM_TREES))
610 		return -EIO;
611 
612 	max_trees = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_LPM_TREES);
613 	mlxsw_sp->router->lpm.tree_count = max_trees - MLXSW_SP_LPM_TREE_MIN;
614 	mlxsw_sp->router->lpm.trees = kcalloc(mlxsw_sp->router->lpm.tree_count,
615 					     sizeof(struct mlxsw_sp_lpm_tree),
616 					     GFP_KERNEL);
617 	if (!mlxsw_sp->router->lpm.trees)
618 		return -ENOMEM;
619 
620 	for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
621 		lpm_tree = &mlxsw_sp->router->lpm.trees[i];
622 		lpm_tree->id = i + MLXSW_SP_LPM_TREE_MIN;
623 	}
624 
625 	lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
626 					 MLXSW_SP_L3_PROTO_IPV4);
627 	if (IS_ERR(lpm_tree)) {
628 		err = PTR_ERR(lpm_tree);
629 		goto err_ipv4_tree_get;
630 	}
631 	mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4] = lpm_tree;
632 
633 	lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
634 					 MLXSW_SP_L3_PROTO_IPV6);
635 	if (IS_ERR(lpm_tree)) {
636 		err = PTR_ERR(lpm_tree);
637 		goto err_ipv6_tree_get;
638 	}
639 	mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV6] = lpm_tree;
640 
641 	return 0;
642 
643 err_ipv6_tree_get:
644 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4];
645 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
646 err_ipv4_tree_get:
647 	kfree(mlxsw_sp->router->lpm.trees);
648 	return err;
649 }
650 
651 static void mlxsw_sp_lpm_fini(struct mlxsw_sp *mlxsw_sp)
652 {
653 	struct mlxsw_sp_lpm_tree *lpm_tree;
654 
655 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV6];
656 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
657 
658 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4];
659 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
660 
661 	kfree(mlxsw_sp->router->lpm.trees);
662 }
663 
664 static bool mlxsw_sp_vr_is_used(const struct mlxsw_sp_vr *vr)
665 {
666 	return !!vr->fib4 || !!vr->fib6 ||
667 	       !!vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] ||
668 	       !!vr->mr_table[MLXSW_SP_L3_PROTO_IPV6];
669 }
670 
671 static struct mlxsw_sp_vr *mlxsw_sp_vr_find_unused(struct mlxsw_sp *mlxsw_sp)
672 {
673 	struct mlxsw_sp_vr *vr;
674 	int i;
675 
676 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
677 		vr = &mlxsw_sp->router->vrs[i];
678 		if (!mlxsw_sp_vr_is_used(vr))
679 			return vr;
680 	}
681 	return NULL;
682 }
683 
684 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
685 				     const struct mlxsw_sp_fib *fib, u8 tree_id)
686 {
687 	char raltb_pl[MLXSW_REG_RALTB_LEN];
688 
689 	mlxsw_reg_raltb_pack(raltb_pl, fib->vr->id,
690 			     (enum mlxsw_reg_ralxx_protocol) fib->proto,
691 			     tree_id);
692 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
693 }
694 
695 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
696 				       const struct mlxsw_sp_fib *fib)
697 {
698 	char raltb_pl[MLXSW_REG_RALTB_LEN];
699 
700 	/* Bind to tree 0 which is default */
701 	mlxsw_reg_raltb_pack(raltb_pl, fib->vr->id,
702 			     (enum mlxsw_reg_ralxx_protocol) fib->proto, 0);
703 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb), raltb_pl);
704 }
705 
706 static u32 mlxsw_sp_fix_tb_id(u32 tb_id)
707 {
708 	/* For our purpose, squash main, default and local tables into one */
709 	if (tb_id == RT_TABLE_LOCAL || tb_id == RT_TABLE_DEFAULT)
710 		tb_id = RT_TABLE_MAIN;
711 	return tb_id;
712 }
713 
714 static struct mlxsw_sp_vr *mlxsw_sp_vr_find(struct mlxsw_sp *mlxsw_sp,
715 					    u32 tb_id)
716 {
717 	struct mlxsw_sp_vr *vr;
718 	int i;
719 
720 	tb_id = mlxsw_sp_fix_tb_id(tb_id);
721 
722 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
723 		vr = &mlxsw_sp->router->vrs[i];
724 		if (mlxsw_sp_vr_is_used(vr) && vr->tb_id == tb_id)
725 			return vr;
726 	}
727 	return NULL;
728 }
729 
730 int mlxsw_sp_router_tb_id_vr_id(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
731 				u16 *vr_id)
732 {
733 	struct mlxsw_sp_vr *vr;
734 	int err = 0;
735 
736 	mutex_lock(&mlxsw_sp->router->lock);
737 	vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
738 	if (!vr) {
739 		err = -ESRCH;
740 		goto out;
741 	}
742 	*vr_id = vr->id;
743 out:
744 	mutex_unlock(&mlxsw_sp->router->lock);
745 	return err;
746 }
747 
748 static struct mlxsw_sp_fib *mlxsw_sp_vr_fib(const struct mlxsw_sp_vr *vr,
749 					    enum mlxsw_sp_l3proto proto)
750 {
751 	switch (proto) {
752 	case MLXSW_SP_L3_PROTO_IPV4:
753 		return vr->fib4;
754 	case MLXSW_SP_L3_PROTO_IPV6:
755 		return vr->fib6;
756 	}
757 	return NULL;
758 }
759 
760 static struct mlxsw_sp_vr *mlxsw_sp_vr_create(struct mlxsw_sp *mlxsw_sp,
761 					      u32 tb_id,
762 					      struct netlink_ext_ack *extack)
763 {
764 	struct mlxsw_sp_mr_table *mr4_table, *mr6_table;
765 	struct mlxsw_sp_fib *fib4;
766 	struct mlxsw_sp_fib *fib6;
767 	struct mlxsw_sp_vr *vr;
768 	int err;
769 
770 	vr = mlxsw_sp_vr_find_unused(mlxsw_sp);
771 	if (!vr) {
772 		NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported virtual routers");
773 		return ERR_PTR(-EBUSY);
774 	}
775 	fib4 = mlxsw_sp_fib_create(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV4);
776 	if (IS_ERR(fib4))
777 		return ERR_CAST(fib4);
778 	fib6 = mlxsw_sp_fib_create(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV6);
779 	if (IS_ERR(fib6)) {
780 		err = PTR_ERR(fib6);
781 		goto err_fib6_create;
782 	}
783 	mr4_table = mlxsw_sp_mr_table_create(mlxsw_sp, vr->id,
784 					     MLXSW_SP_L3_PROTO_IPV4);
785 	if (IS_ERR(mr4_table)) {
786 		err = PTR_ERR(mr4_table);
787 		goto err_mr4_table_create;
788 	}
789 	mr6_table = mlxsw_sp_mr_table_create(mlxsw_sp, vr->id,
790 					     MLXSW_SP_L3_PROTO_IPV6);
791 	if (IS_ERR(mr6_table)) {
792 		err = PTR_ERR(mr6_table);
793 		goto err_mr6_table_create;
794 	}
795 
796 	vr->fib4 = fib4;
797 	vr->fib6 = fib6;
798 	vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] = mr4_table;
799 	vr->mr_table[MLXSW_SP_L3_PROTO_IPV6] = mr6_table;
800 	vr->tb_id = tb_id;
801 	return vr;
802 
803 err_mr6_table_create:
804 	mlxsw_sp_mr_table_destroy(mr4_table);
805 err_mr4_table_create:
806 	mlxsw_sp_fib_destroy(mlxsw_sp, fib6);
807 err_fib6_create:
808 	mlxsw_sp_fib_destroy(mlxsw_sp, fib4);
809 	return ERR_PTR(err);
810 }
811 
812 static void mlxsw_sp_vr_destroy(struct mlxsw_sp *mlxsw_sp,
813 				struct mlxsw_sp_vr *vr)
814 {
815 	mlxsw_sp_mr_table_destroy(vr->mr_table[MLXSW_SP_L3_PROTO_IPV6]);
816 	vr->mr_table[MLXSW_SP_L3_PROTO_IPV6] = NULL;
817 	mlxsw_sp_mr_table_destroy(vr->mr_table[MLXSW_SP_L3_PROTO_IPV4]);
818 	vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] = NULL;
819 	mlxsw_sp_fib_destroy(mlxsw_sp, vr->fib6);
820 	vr->fib6 = NULL;
821 	mlxsw_sp_fib_destroy(mlxsw_sp, vr->fib4);
822 	vr->fib4 = NULL;
823 }
824 
825 static struct mlxsw_sp_vr *mlxsw_sp_vr_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
826 					   struct netlink_ext_ack *extack)
827 {
828 	struct mlxsw_sp_vr *vr;
829 
830 	tb_id = mlxsw_sp_fix_tb_id(tb_id);
831 	vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
832 	if (!vr)
833 		vr = mlxsw_sp_vr_create(mlxsw_sp, tb_id, extack);
834 	return vr;
835 }
836 
837 static void mlxsw_sp_vr_put(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr)
838 {
839 	if (!vr->rif_count && list_empty(&vr->fib4->node_list) &&
840 	    list_empty(&vr->fib6->node_list) &&
841 	    mlxsw_sp_mr_table_empty(vr->mr_table[MLXSW_SP_L3_PROTO_IPV4]) &&
842 	    mlxsw_sp_mr_table_empty(vr->mr_table[MLXSW_SP_L3_PROTO_IPV6]))
843 		mlxsw_sp_vr_destroy(mlxsw_sp, vr);
844 }
845 
846 static bool
847 mlxsw_sp_vr_lpm_tree_should_replace(struct mlxsw_sp_vr *vr,
848 				    enum mlxsw_sp_l3proto proto, u8 tree_id)
849 {
850 	struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
851 
852 	if (!mlxsw_sp_vr_is_used(vr))
853 		return false;
854 	if (fib->lpm_tree->id == tree_id)
855 		return true;
856 	return false;
857 }
858 
859 static int mlxsw_sp_vr_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
860 					struct mlxsw_sp_fib *fib,
861 					struct mlxsw_sp_lpm_tree *new_tree)
862 {
863 	struct mlxsw_sp_lpm_tree *old_tree = fib->lpm_tree;
864 	int err;
865 
866 	fib->lpm_tree = new_tree;
867 	mlxsw_sp_lpm_tree_hold(new_tree);
868 	err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, new_tree->id);
869 	if (err)
870 		goto err_tree_bind;
871 	mlxsw_sp_lpm_tree_put(mlxsw_sp, old_tree);
872 	return 0;
873 
874 err_tree_bind:
875 	mlxsw_sp_lpm_tree_put(mlxsw_sp, new_tree);
876 	fib->lpm_tree = old_tree;
877 	return err;
878 }
879 
880 static int mlxsw_sp_vrs_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
881 					 struct mlxsw_sp_fib *fib,
882 					 struct mlxsw_sp_lpm_tree *new_tree)
883 {
884 	enum mlxsw_sp_l3proto proto = fib->proto;
885 	struct mlxsw_sp_lpm_tree *old_tree;
886 	u8 old_id, new_id = new_tree->id;
887 	struct mlxsw_sp_vr *vr;
888 	int i, err;
889 
890 	old_tree = mlxsw_sp->router->lpm.proto_trees[proto];
891 	old_id = old_tree->id;
892 
893 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
894 		vr = &mlxsw_sp->router->vrs[i];
895 		if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, old_id))
896 			continue;
897 		err = mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
898 						   mlxsw_sp_vr_fib(vr, proto),
899 						   new_tree);
900 		if (err)
901 			goto err_tree_replace;
902 	}
903 
904 	memcpy(new_tree->prefix_ref_count, old_tree->prefix_ref_count,
905 	       sizeof(new_tree->prefix_ref_count));
906 	mlxsw_sp->router->lpm.proto_trees[proto] = new_tree;
907 	mlxsw_sp_lpm_tree_put(mlxsw_sp, old_tree);
908 
909 	return 0;
910 
911 err_tree_replace:
912 	for (i--; i >= 0; i--) {
913 		if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, new_id))
914 			continue;
915 		mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
916 					     mlxsw_sp_vr_fib(vr, proto),
917 					     old_tree);
918 	}
919 	return err;
920 }
921 
922 static int mlxsw_sp_vrs_init(struct mlxsw_sp *mlxsw_sp)
923 {
924 	struct mlxsw_sp_vr *vr;
925 	u64 max_vrs;
926 	int i;
927 
928 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_VRS))
929 		return -EIO;
930 
931 	max_vrs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS);
932 	mlxsw_sp->router->vrs = kcalloc(max_vrs, sizeof(struct mlxsw_sp_vr),
933 					GFP_KERNEL);
934 	if (!mlxsw_sp->router->vrs)
935 		return -ENOMEM;
936 
937 	for (i = 0; i < max_vrs; i++) {
938 		vr = &mlxsw_sp->router->vrs[i];
939 		vr->id = i;
940 	}
941 
942 	return 0;
943 }
944 
945 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp);
946 
947 static void mlxsw_sp_vrs_fini(struct mlxsw_sp *mlxsw_sp)
948 {
949 	/* At this stage we're guaranteed not to have new incoming
950 	 * FIB notifications and the work queue is free from FIBs
951 	 * sitting on top of mlxsw netdevs. However, we can still
952 	 * have other FIBs queued. Flush the queue before flushing
953 	 * the device's tables. No need for locks, as we're the only
954 	 * writer.
955 	 */
956 	mlxsw_core_flush_owq();
957 	mlxsw_sp_router_fib_flush(mlxsw_sp);
958 	kfree(mlxsw_sp->router->vrs);
959 }
960 
961 static struct net_device *
962 __mlxsw_sp_ipip_netdev_ul_dev_get(const struct net_device *ol_dev)
963 {
964 	struct ip_tunnel *tun = netdev_priv(ol_dev);
965 	struct net *net = dev_net(ol_dev);
966 
967 	return dev_get_by_index_rcu(net, tun->parms.link);
968 }
969 
970 u32 mlxsw_sp_ipip_dev_ul_tb_id(const struct net_device *ol_dev)
971 {
972 	struct net_device *d;
973 	u32 tb_id;
974 
975 	rcu_read_lock();
976 	d = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
977 	if (d)
978 		tb_id = l3mdev_fib_table(d) ? : RT_TABLE_MAIN;
979 	else
980 		tb_id = RT_TABLE_MAIN;
981 	rcu_read_unlock();
982 
983 	return tb_id;
984 }
985 
986 static struct mlxsw_sp_rif *
987 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
988 		    const struct mlxsw_sp_rif_params *params,
989 		    struct netlink_ext_ack *extack);
990 
991 static struct mlxsw_sp_rif_ipip_lb *
992 mlxsw_sp_ipip_ol_ipip_lb_create(struct mlxsw_sp *mlxsw_sp,
993 				enum mlxsw_sp_ipip_type ipipt,
994 				struct net_device *ol_dev,
995 				struct netlink_ext_ack *extack)
996 {
997 	struct mlxsw_sp_rif_params_ipip_lb lb_params;
998 	const struct mlxsw_sp_ipip_ops *ipip_ops;
999 	struct mlxsw_sp_rif *rif;
1000 
1001 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
1002 	lb_params = (struct mlxsw_sp_rif_params_ipip_lb) {
1003 		.common.dev = ol_dev,
1004 		.common.lag = false,
1005 		.lb_config = ipip_ops->ol_loopback_config(mlxsw_sp, ol_dev),
1006 	};
1007 
1008 	rif = mlxsw_sp_rif_create(mlxsw_sp, &lb_params.common, extack);
1009 	if (IS_ERR(rif))
1010 		return ERR_CAST(rif);
1011 	return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
1012 }
1013 
1014 static struct mlxsw_sp_ipip_entry *
1015 mlxsw_sp_ipip_entry_alloc(struct mlxsw_sp *mlxsw_sp,
1016 			  enum mlxsw_sp_ipip_type ipipt,
1017 			  struct net_device *ol_dev)
1018 {
1019 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1020 	struct mlxsw_sp_ipip_entry *ipip_entry;
1021 	struct mlxsw_sp_ipip_entry *ret = NULL;
1022 
1023 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
1024 	ipip_entry = kzalloc(sizeof(*ipip_entry), GFP_KERNEL);
1025 	if (!ipip_entry)
1026 		return ERR_PTR(-ENOMEM);
1027 
1028 	ipip_entry->ol_lb = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp, ipipt,
1029 							    ol_dev, NULL);
1030 	if (IS_ERR(ipip_entry->ol_lb)) {
1031 		ret = ERR_CAST(ipip_entry->ol_lb);
1032 		goto err_ol_ipip_lb_create;
1033 	}
1034 
1035 	ipip_entry->ipipt = ipipt;
1036 	ipip_entry->ol_dev = ol_dev;
1037 
1038 	switch (ipip_ops->ul_proto) {
1039 	case MLXSW_SP_L3_PROTO_IPV4:
1040 		ipip_entry->parms4 = mlxsw_sp_ipip_netdev_parms4(ol_dev);
1041 		break;
1042 	case MLXSW_SP_L3_PROTO_IPV6:
1043 		WARN_ON(1);
1044 		break;
1045 	}
1046 
1047 	return ipip_entry;
1048 
1049 err_ol_ipip_lb_create:
1050 	kfree(ipip_entry);
1051 	return ret;
1052 }
1053 
1054 static void
1055 mlxsw_sp_ipip_entry_dealloc(struct mlxsw_sp_ipip_entry *ipip_entry)
1056 {
1057 	mlxsw_sp_rif_destroy(&ipip_entry->ol_lb->common);
1058 	kfree(ipip_entry);
1059 }
1060 
1061 static bool
1062 mlxsw_sp_ipip_entry_saddr_matches(struct mlxsw_sp *mlxsw_sp,
1063 				  const enum mlxsw_sp_l3proto ul_proto,
1064 				  union mlxsw_sp_l3addr saddr,
1065 				  u32 ul_tb_id,
1066 				  struct mlxsw_sp_ipip_entry *ipip_entry)
1067 {
1068 	u32 tun_ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1069 	enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1070 	union mlxsw_sp_l3addr tun_saddr;
1071 
1072 	if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1073 		return false;
1074 
1075 	tun_saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ipip_entry->ol_dev);
1076 	return tun_ul_tb_id == ul_tb_id &&
1077 	       mlxsw_sp_l3addr_eq(&tun_saddr, &saddr);
1078 }
1079 
1080 static int
1081 mlxsw_sp_fib_entry_decap_init(struct mlxsw_sp *mlxsw_sp,
1082 			      struct mlxsw_sp_fib_entry *fib_entry,
1083 			      struct mlxsw_sp_ipip_entry *ipip_entry)
1084 {
1085 	u32 tunnel_index;
1086 	int err;
1087 
1088 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
1089 				  1, &tunnel_index);
1090 	if (err)
1091 		return err;
1092 
1093 	ipip_entry->decap_fib_entry = fib_entry;
1094 	fib_entry->decap.ipip_entry = ipip_entry;
1095 	fib_entry->decap.tunnel_index = tunnel_index;
1096 	return 0;
1097 }
1098 
1099 static void mlxsw_sp_fib_entry_decap_fini(struct mlxsw_sp *mlxsw_sp,
1100 					  struct mlxsw_sp_fib_entry *fib_entry)
1101 {
1102 	/* Unlink this node from the IPIP entry that it's the decap entry of. */
1103 	fib_entry->decap.ipip_entry->decap_fib_entry = NULL;
1104 	fib_entry->decap.ipip_entry = NULL;
1105 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
1106 			   1, fib_entry->decap.tunnel_index);
1107 }
1108 
1109 static struct mlxsw_sp_fib_node *
1110 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
1111 			 size_t addr_len, unsigned char prefix_len);
1112 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1113 				     struct mlxsw_sp_fib_entry *fib_entry);
1114 
1115 static void
1116 mlxsw_sp_ipip_entry_demote_decap(struct mlxsw_sp *mlxsw_sp,
1117 				 struct mlxsw_sp_ipip_entry *ipip_entry)
1118 {
1119 	struct mlxsw_sp_fib_entry *fib_entry = ipip_entry->decap_fib_entry;
1120 
1121 	mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, fib_entry);
1122 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1123 
1124 	mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1125 }
1126 
1127 static void
1128 mlxsw_sp_ipip_entry_promote_decap(struct mlxsw_sp *mlxsw_sp,
1129 				  struct mlxsw_sp_ipip_entry *ipip_entry,
1130 				  struct mlxsw_sp_fib_entry *decap_fib_entry)
1131 {
1132 	if (mlxsw_sp_fib_entry_decap_init(mlxsw_sp, decap_fib_entry,
1133 					  ipip_entry))
1134 		return;
1135 	decap_fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
1136 
1137 	if (mlxsw_sp_fib_entry_update(mlxsw_sp, decap_fib_entry))
1138 		mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1139 }
1140 
1141 static struct mlxsw_sp_fib_entry *
1142 mlxsw_sp_router_ip2me_fib_entry_find(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
1143 				     enum mlxsw_sp_l3proto proto,
1144 				     const union mlxsw_sp_l3addr *addr,
1145 				     enum mlxsw_sp_fib_entry_type type)
1146 {
1147 	struct mlxsw_sp_fib_node *fib_node;
1148 	unsigned char addr_prefix_len;
1149 	struct mlxsw_sp_fib *fib;
1150 	struct mlxsw_sp_vr *vr;
1151 	const void *addrp;
1152 	size_t addr_len;
1153 	u32 addr4;
1154 
1155 	vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
1156 	if (!vr)
1157 		return NULL;
1158 	fib = mlxsw_sp_vr_fib(vr, proto);
1159 
1160 	switch (proto) {
1161 	case MLXSW_SP_L3_PROTO_IPV4:
1162 		addr4 = be32_to_cpu(addr->addr4);
1163 		addrp = &addr4;
1164 		addr_len = 4;
1165 		addr_prefix_len = 32;
1166 		break;
1167 	case MLXSW_SP_L3_PROTO_IPV6: /* fall through */
1168 	default:
1169 		WARN_ON(1);
1170 		return NULL;
1171 	}
1172 
1173 	fib_node = mlxsw_sp_fib_node_lookup(fib, addrp, addr_len,
1174 					    addr_prefix_len);
1175 	if (!fib_node || fib_node->fib_entry->type != type)
1176 		return NULL;
1177 
1178 	return fib_node->fib_entry;
1179 }
1180 
1181 /* Given an IPIP entry, find the corresponding decap route. */
1182 static struct mlxsw_sp_fib_entry *
1183 mlxsw_sp_ipip_entry_find_decap(struct mlxsw_sp *mlxsw_sp,
1184 			       struct mlxsw_sp_ipip_entry *ipip_entry)
1185 {
1186 	static struct mlxsw_sp_fib_node *fib_node;
1187 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1188 	unsigned char saddr_prefix_len;
1189 	union mlxsw_sp_l3addr saddr;
1190 	struct mlxsw_sp_fib *ul_fib;
1191 	struct mlxsw_sp_vr *ul_vr;
1192 	const void *saddrp;
1193 	size_t saddr_len;
1194 	u32 ul_tb_id;
1195 	u32 saddr4;
1196 
1197 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1198 
1199 	ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1200 	ul_vr = mlxsw_sp_vr_find(mlxsw_sp, ul_tb_id);
1201 	if (!ul_vr)
1202 		return NULL;
1203 
1204 	ul_fib = mlxsw_sp_vr_fib(ul_vr, ipip_ops->ul_proto);
1205 	saddr = mlxsw_sp_ipip_netdev_saddr(ipip_ops->ul_proto,
1206 					   ipip_entry->ol_dev);
1207 
1208 	switch (ipip_ops->ul_proto) {
1209 	case MLXSW_SP_L3_PROTO_IPV4:
1210 		saddr4 = be32_to_cpu(saddr.addr4);
1211 		saddrp = &saddr4;
1212 		saddr_len = 4;
1213 		saddr_prefix_len = 32;
1214 		break;
1215 	default:
1216 		WARN_ON(1);
1217 		return NULL;
1218 	}
1219 
1220 	fib_node = mlxsw_sp_fib_node_lookup(ul_fib, saddrp, saddr_len,
1221 					    saddr_prefix_len);
1222 	if (!fib_node ||
1223 	    fib_node->fib_entry->type != MLXSW_SP_FIB_ENTRY_TYPE_TRAP)
1224 		return NULL;
1225 
1226 	return fib_node->fib_entry;
1227 }
1228 
1229 static struct mlxsw_sp_ipip_entry *
1230 mlxsw_sp_ipip_entry_create(struct mlxsw_sp *mlxsw_sp,
1231 			   enum mlxsw_sp_ipip_type ipipt,
1232 			   struct net_device *ol_dev)
1233 {
1234 	struct mlxsw_sp_ipip_entry *ipip_entry;
1235 
1236 	ipip_entry = mlxsw_sp_ipip_entry_alloc(mlxsw_sp, ipipt, ol_dev);
1237 	if (IS_ERR(ipip_entry))
1238 		return ipip_entry;
1239 
1240 	list_add_tail(&ipip_entry->ipip_list_node,
1241 		      &mlxsw_sp->router->ipip_list);
1242 
1243 	return ipip_entry;
1244 }
1245 
1246 static void
1247 mlxsw_sp_ipip_entry_destroy(struct mlxsw_sp *mlxsw_sp,
1248 			    struct mlxsw_sp_ipip_entry *ipip_entry)
1249 {
1250 	list_del(&ipip_entry->ipip_list_node);
1251 	mlxsw_sp_ipip_entry_dealloc(ipip_entry);
1252 }
1253 
1254 static bool
1255 mlxsw_sp_ipip_entry_matches_decap(struct mlxsw_sp *mlxsw_sp,
1256 				  const struct net_device *ul_dev,
1257 				  enum mlxsw_sp_l3proto ul_proto,
1258 				  union mlxsw_sp_l3addr ul_dip,
1259 				  struct mlxsw_sp_ipip_entry *ipip_entry)
1260 {
1261 	u32 ul_tb_id = l3mdev_fib_table(ul_dev) ? : RT_TABLE_MAIN;
1262 	enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1263 
1264 	if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1265 		return false;
1266 
1267 	return mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, ul_dip,
1268 						 ul_tb_id, ipip_entry);
1269 }
1270 
1271 /* Given decap parameters, find the corresponding IPIP entry. */
1272 static struct mlxsw_sp_ipip_entry *
1273 mlxsw_sp_ipip_entry_find_by_decap(struct mlxsw_sp *mlxsw_sp,
1274 				  const struct net_device *ul_dev,
1275 				  enum mlxsw_sp_l3proto ul_proto,
1276 				  union mlxsw_sp_l3addr ul_dip)
1277 {
1278 	struct mlxsw_sp_ipip_entry *ipip_entry;
1279 
1280 	list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1281 			    ipip_list_node)
1282 		if (mlxsw_sp_ipip_entry_matches_decap(mlxsw_sp, ul_dev,
1283 						      ul_proto, ul_dip,
1284 						      ipip_entry))
1285 			return ipip_entry;
1286 
1287 	return NULL;
1288 }
1289 
1290 static bool mlxsw_sp_netdev_ipip_type(const struct mlxsw_sp *mlxsw_sp,
1291 				      const struct net_device *dev,
1292 				      enum mlxsw_sp_ipip_type *p_type)
1293 {
1294 	struct mlxsw_sp_router *router = mlxsw_sp->router;
1295 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1296 	enum mlxsw_sp_ipip_type ipipt;
1297 
1298 	for (ipipt = 0; ipipt < MLXSW_SP_IPIP_TYPE_MAX; ++ipipt) {
1299 		ipip_ops = router->ipip_ops_arr[ipipt];
1300 		if (dev->type == ipip_ops->dev_type) {
1301 			if (p_type)
1302 				*p_type = ipipt;
1303 			return true;
1304 		}
1305 	}
1306 	return false;
1307 }
1308 
1309 bool mlxsw_sp_netdev_is_ipip_ol(const struct mlxsw_sp *mlxsw_sp,
1310 				const struct net_device *dev)
1311 {
1312 	return mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL);
1313 }
1314 
1315 static struct mlxsw_sp_ipip_entry *
1316 mlxsw_sp_ipip_entry_find_by_ol_dev(struct mlxsw_sp *mlxsw_sp,
1317 				   const struct net_device *ol_dev)
1318 {
1319 	struct mlxsw_sp_ipip_entry *ipip_entry;
1320 
1321 	list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1322 			    ipip_list_node)
1323 		if (ipip_entry->ol_dev == ol_dev)
1324 			return ipip_entry;
1325 
1326 	return NULL;
1327 }
1328 
1329 static struct mlxsw_sp_ipip_entry *
1330 mlxsw_sp_ipip_entry_find_by_ul_dev(const struct mlxsw_sp *mlxsw_sp,
1331 				   const struct net_device *ul_dev,
1332 				   struct mlxsw_sp_ipip_entry *start)
1333 {
1334 	struct mlxsw_sp_ipip_entry *ipip_entry;
1335 
1336 	ipip_entry = list_prepare_entry(start, &mlxsw_sp->router->ipip_list,
1337 					ipip_list_node);
1338 	list_for_each_entry_continue(ipip_entry, &mlxsw_sp->router->ipip_list,
1339 				     ipip_list_node) {
1340 		struct net_device *ol_dev = ipip_entry->ol_dev;
1341 		struct net_device *ipip_ul_dev;
1342 
1343 		rcu_read_lock();
1344 		ipip_ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1345 		rcu_read_unlock();
1346 
1347 		if (ipip_ul_dev == ul_dev)
1348 			return ipip_entry;
1349 	}
1350 
1351 	return NULL;
1352 }
1353 
1354 bool mlxsw_sp_netdev_is_ipip_ul(struct mlxsw_sp *mlxsw_sp,
1355 				const struct net_device *dev)
1356 {
1357 	bool is_ipip_ul;
1358 
1359 	mutex_lock(&mlxsw_sp->router->lock);
1360 	is_ipip_ul = mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp, dev, NULL);
1361 	mutex_unlock(&mlxsw_sp->router->lock);
1362 
1363 	return is_ipip_ul;
1364 }
1365 
1366 static bool mlxsw_sp_netdevice_ipip_can_offload(struct mlxsw_sp *mlxsw_sp,
1367 						const struct net_device *ol_dev,
1368 						enum mlxsw_sp_ipip_type ipipt)
1369 {
1370 	const struct mlxsw_sp_ipip_ops *ops
1371 		= mlxsw_sp->router->ipip_ops_arr[ipipt];
1372 
1373 	/* For deciding whether decap should be offloaded, we don't care about
1374 	 * overlay protocol, so ask whether either one is supported.
1375 	 */
1376 	return ops->can_offload(mlxsw_sp, ol_dev, MLXSW_SP_L3_PROTO_IPV4) ||
1377 	       ops->can_offload(mlxsw_sp, ol_dev, MLXSW_SP_L3_PROTO_IPV6);
1378 }
1379 
1380 static int mlxsw_sp_netdevice_ipip_ol_reg_event(struct mlxsw_sp *mlxsw_sp,
1381 						struct net_device *ol_dev)
1382 {
1383 	enum mlxsw_sp_ipip_type ipipt = MLXSW_SP_IPIP_TYPE_MAX;
1384 	struct mlxsw_sp_ipip_entry *ipip_entry;
1385 	enum mlxsw_sp_l3proto ul_proto;
1386 	union mlxsw_sp_l3addr saddr;
1387 	u32 ul_tb_id;
1388 
1389 	mlxsw_sp_netdev_ipip_type(mlxsw_sp, ol_dev, &ipipt);
1390 	if (mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev, ipipt)) {
1391 		ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ol_dev);
1392 		ul_proto = mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto;
1393 		saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ol_dev);
1394 		if (!mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1395 							  saddr, ul_tb_id,
1396 							  NULL)) {
1397 			ipip_entry = mlxsw_sp_ipip_entry_create(mlxsw_sp, ipipt,
1398 								ol_dev);
1399 			if (IS_ERR(ipip_entry))
1400 				return PTR_ERR(ipip_entry);
1401 		}
1402 	}
1403 
1404 	return 0;
1405 }
1406 
1407 static void mlxsw_sp_netdevice_ipip_ol_unreg_event(struct mlxsw_sp *mlxsw_sp,
1408 						   struct net_device *ol_dev)
1409 {
1410 	struct mlxsw_sp_ipip_entry *ipip_entry;
1411 
1412 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1413 	if (ipip_entry)
1414 		mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1415 }
1416 
1417 static void
1418 mlxsw_sp_ipip_entry_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1419 				struct mlxsw_sp_ipip_entry *ipip_entry)
1420 {
1421 	struct mlxsw_sp_fib_entry *decap_fib_entry;
1422 
1423 	decap_fib_entry = mlxsw_sp_ipip_entry_find_decap(mlxsw_sp, ipip_entry);
1424 	if (decap_fib_entry)
1425 		mlxsw_sp_ipip_entry_promote_decap(mlxsw_sp, ipip_entry,
1426 						  decap_fib_entry);
1427 }
1428 
1429 static int
1430 mlxsw_sp_rif_ipip_lb_op(struct mlxsw_sp_rif_ipip_lb *lb_rif, u16 ul_vr_id,
1431 			u16 ul_rif_id, bool enable)
1432 {
1433 	struct mlxsw_sp_rif_ipip_lb_config lb_cf = lb_rif->lb_config;
1434 	struct mlxsw_sp_rif *rif = &lb_rif->common;
1435 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
1436 	char ritr_pl[MLXSW_REG_RITR_LEN];
1437 	u32 saddr4;
1438 
1439 	switch (lb_cf.ul_protocol) {
1440 	case MLXSW_SP_L3_PROTO_IPV4:
1441 		saddr4 = be32_to_cpu(lb_cf.saddr.addr4);
1442 		mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
1443 				    rif->rif_index, rif->vr_id, rif->dev->mtu);
1444 		mlxsw_reg_ritr_loopback_ipip4_pack(ritr_pl, lb_cf.lb_ipipt,
1445 			    MLXSW_REG_RITR_LOOPBACK_IPIP_OPTIONS_GRE_KEY_PRESET,
1446 			    ul_vr_id, ul_rif_id, saddr4, lb_cf.okey);
1447 		break;
1448 
1449 	case MLXSW_SP_L3_PROTO_IPV6:
1450 		return -EAFNOSUPPORT;
1451 	}
1452 
1453 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
1454 }
1455 
1456 static int mlxsw_sp_netdevice_ipip_ol_update_mtu(struct mlxsw_sp *mlxsw_sp,
1457 						 struct net_device *ol_dev)
1458 {
1459 	struct mlxsw_sp_ipip_entry *ipip_entry;
1460 	struct mlxsw_sp_rif_ipip_lb *lb_rif;
1461 	int err = 0;
1462 
1463 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1464 	if (ipip_entry) {
1465 		lb_rif = ipip_entry->ol_lb;
1466 		err = mlxsw_sp_rif_ipip_lb_op(lb_rif, lb_rif->ul_vr_id,
1467 					      lb_rif->ul_rif_id, true);
1468 		if (err)
1469 			goto out;
1470 		lb_rif->common.mtu = ol_dev->mtu;
1471 	}
1472 
1473 out:
1474 	return err;
1475 }
1476 
1477 static void mlxsw_sp_netdevice_ipip_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1478 						struct net_device *ol_dev)
1479 {
1480 	struct mlxsw_sp_ipip_entry *ipip_entry;
1481 
1482 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1483 	if (ipip_entry)
1484 		mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1485 }
1486 
1487 static void
1488 mlxsw_sp_ipip_entry_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1489 				  struct mlxsw_sp_ipip_entry *ipip_entry)
1490 {
1491 	if (ipip_entry->decap_fib_entry)
1492 		mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1493 }
1494 
1495 static void mlxsw_sp_netdevice_ipip_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1496 						  struct net_device *ol_dev)
1497 {
1498 	struct mlxsw_sp_ipip_entry *ipip_entry;
1499 
1500 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1501 	if (ipip_entry)
1502 		mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1503 }
1504 
1505 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
1506 					 struct mlxsw_sp_rif *old_rif,
1507 					 struct mlxsw_sp_rif *new_rif);
1508 static int
1509 mlxsw_sp_ipip_entry_ol_lb_update(struct mlxsw_sp *mlxsw_sp,
1510 				 struct mlxsw_sp_ipip_entry *ipip_entry,
1511 				 bool keep_encap,
1512 				 struct netlink_ext_ack *extack)
1513 {
1514 	struct mlxsw_sp_rif_ipip_lb *old_lb_rif = ipip_entry->ol_lb;
1515 	struct mlxsw_sp_rif_ipip_lb *new_lb_rif;
1516 
1517 	new_lb_rif = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp,
1518 						     ipip_entry->ipipt,
1519 						     ipip_entry->ol_dev,
1520 						     extack);
1521 	if (IS_ERR(new_lb_rif))
1522 		return PTR_ERR(new_lb_rif);
1523 	ipip_entry->ol_lb = new_lb_rif;
1524 
1525 	if (keep_encap)
1526 		mlxsw_sp_nexthop_rif_migrate(mlxsw_sp, &old_lb_rif->common,
1527 					     &new_lb_rif->common);
1528 
1529 	mlxsw_sp_rif_destroy(&old_lb_rif->common);
1530 
1531 	return 0;
1532 }
1533 
1534 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
1535 					struct mlxsw_sp_rif *rif);
1536 
1537 /**
1538  * __mlxsw_sp_ipip_entry_update_tunnel - Update offload related to IPIP entry.
1539  * @mlxsw_sp: mlxsw_sp.
1540  * @ipip_entry: IPIP entry.
1541  * @recreate_loopback: Recreates the associated loopback RIF.
1542  * @keep_encap: Updates next hops that use the tunnel netdevice. This is only
1543  *              relevant when recreate_loopback is true.
1544  * @update_nexthops: Updates next hops, keeping the current loopback RIF. This
1545  *                   is only relevant when recreate_loopback is false.
1546  * @extack: extack.
1547  *
1548  * Return: Non-zero value on failure.
1549  */
1550 int __mlxsw_sp_ipip_entry_update_tunnel(struct mlxsw_sp *mlxsw_sp,
1551 					struct mlxsw_sp_ipip_entry *ipip_entry,
1552 					bool recreate_loopback,
1553 					bool keep_encap,
1554 					bool update_nexthops,
1555 					struct netlink_ext_ack *extack)
1556 {
1557 	int err;
1558 
1559 	/* RIFs can't be edited, so to update loopback, we need to destroy and
1560 	 * recreate it. That creates a window of opportunity where RALUE and
1561 	 * RATR registers end up referencing a RIF that's already gone. RATRs
1562 	 * are handled in mlxsw_sp_ipip_entry_ol_lb_update(), and to take care
1563 	 * of RALUE, demote the decap route back.
1564 	 */
1565 	if (ipip_entry->decap_fib_entry)
1566 		mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1567 
1568 	if (recreate_loopback) {
1569 		err = mlxsw_sp_ipip_entry_ol_lb_update(mlxsw_sp, ipip_entry,
1570 						       keep_encap, extack);
1571 		if (err)
1572 			return err;
1573 	} else if (update_nexthops) {
1574 		mlxsw_sp_nexthop_rif_update(mlxsw_sp,
1575 					    &ipip_entry->ol_lb->common);
1576 	}
1577 
1578 	if (ipip_entry->ol_dev->flags & IFF_UP)
1579 		mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1580 
1581 	return 0;
1582 }
1583 
1584 static int mlxsw_sp_netdevice_ipip_ol_vrf_event(struct mlxsw_sp *mlxsw_sp,
1585 						struct net_device *ol_dev,
1586 						struct netlink_ext_ack *extack)
1587 {
1588 	struct mlxsw_sp_ipip_entry *ipip_entry =
1589 		mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1590 
1591 	if (!ipip_entry)
1592 		return 0;
1593 
1594 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1595 						   true, false, false, extack);
1596 }
1597 
1598 static int
1599 mlxsw_sp_netdevice_ipip_ul_vrf_event(struct mlxsw_sp *mlxsw_sp,
1600 				     struct mlxsw_sp_ipip_entry *ipip_entry,
1601 				     struct net_device *ul_dev,
1602 				     bool *demote_this,
1603 				     struct netlink_ext_ack *extack)
1604 {
1605 	u32 ul_tb_id = l3mdev_fib_table(ul_dev) ? : RT_TABLE_MAIN;
1606 	enum mlxsw_sp_l3proto ul_proto;
1607 	union mlxsw_sp_l3addr saddr;
1608 
1609 	/* Moving underlay to a different VRF might cause local address
1610 	 * conflict, and the conflicting tunnels need to be demoted.
1611 	 */
1612 	ul_proto = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt]->ul_proto;
1613 	saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ipip_entry->ol_dev);
1614 	if (mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1615 						 saddr, ul_tb_id,
1616 						 ipip_entry)) {
1617 		*demote_this = true;
1618 		return 0;
1619 	}
1620 
1621 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1622 						   true, true, false, extack);
1623 }
1624 
1625 static int
1626 mlxsw_sp_netdevice_ipip_ul_up_event(struct mlxsw_sp *mlxsw_sp,
1627 				    struct mlxsw_sp_ipip_entry *ipip_entry,
1628 				    struct net_device *ul_dev)
1629 {
1630 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1631 						   false, false, true, NULL);
1632 }
1633 
1634 static int
1635 mlxsw_sp_netdevice_ipip_ul_down_event(struct mlxsw_sp *mlxsw_sp,
1636 				      struct mlxsw_sp_ipip_entry *ipip_entry,
1637 				      struct net_device *ul_dev)
1638 {
1639 	/* A down underlay device causes encapsulated packets to not be
1640 	 * forwarded, but decap still works. So refresh next hops without
1641 	 * touching anything else.
1642 	 */
1643 	return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1644 						   false, false, true, NULL);
1645 }
1646 
1647 static int
1648 mlxsw_sp_netdevice_ipip_ol_change_event(struct mlxsw_sp *mlxsw_sp,
1649 					struct net_device *ol_dev,
1650 					struct netlink_ext_ack *extack)
1651 {
1652 	const struct mlxsw_sp_ipip_ops *ipip_ops;
1653 	struct mlxsw_sp_ipip_entry *ipip_entry;
1654 	int err;
1655 
1656 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1657 	if (!ipip_entry)
1658 		/* A change might make a tunnel eligible for offloading, but
1659 		 * that is currently not implemented. What falls to slow path
1660 		 * stays there.
1661 		 */
1662 		return 0;
1663 
1664 	/* A change might make a tunnel not eligible for offloading. */
1665 	if (!mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev,
1666 						 ipip_entry->ipipt)) {
1667 		mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1668 		return 0;
1669 	}
1670 
1671 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1672 	err = ipip_ops->ol_netdev_change(mlxsw_sp, ipip_entry, extack);
1673 	return err;
1674 }
1675 
1676 void mlxsw_sp_ipip_entry_demote_tunnel(struct mlxsw_sp *mlxsw_sp,
1677 				       struct mlxsw_sp_ipip_entry *ipip_entry)
1678 {
1679 	struct net_device *ol_dev = ipip_entry->ol_dev;
1680 
1681 	if (ol_dev->flags & IFF_UP)
1682 		mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1683 	mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1684 }
1685 
1686 /* The configuration where several tunnels have the same local address in the
1687  * same underlay table needs special treatment in the HW. That is currently not
1688  * implemented in the driver. This function finds and demotes the first tunnel
1689  * with a given source address, except the one passed in in the argument
1690  * `except'.
1691  */
1692 bool
1693 mlxsw_sp_ipip_demote_tunnel_by_saddr(struct mlxsw_sp *mlxsw_sp,
1694 				     enum mlxsw_sp_l3proto ul_proto,
1695 				     union mlxsw_sp_l3addr saddr,
1696 				     u32 ul_tb_id,
1697 				     const struct mlxsw_sp_ipip_entry *except)
1698 {
1699 	struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1700 
1701 	list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1702 				 ipip_list_node) {
1703 		if (ipip_entry != except &&
1704 		    mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, saddr,
1705 						      ul_tb_id, ipip_entry)) {
1706 			mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1707 			return true;
1708 		}
1709 	}
1710 
1711 	return false;
1712 }
1713 
1714 static void mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(struct mlxsw_sp *mlxsw_sp,
1715 						     struct net_device *ul_dev)
1716 {
1717 	struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1718 
1719 	list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1720 				 ipip_list_node) {
1721 		struct net_device *ol_dev = ipip_entry->ol_dev;
1722 		struct net_device *ipip_ul_dev;
1723 
1724 		rcu_read_lock();
1725 		ipip_ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1726 		rcu_read_unlock();
1727 		if (ipip_ul_dev == ul_dev)
1728 			mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1729 	}
1730 }
1731 
1732 int mlxsw_sp_netdevice_ipip_ol_event(struct mlxsw_sp *mlxsw_sp,
1733 				     struct net_device *ol_dev,
1734 				     unsigned long event,
1735 				     struct netdev_notifier_info *info)
1736 {
1737 	struct netdev_notifier_changeupper_info *chup;
1738 	struct netlink_ext_ack *extack;
1739 	int err = 0;
1740 
1741 	mutex_lock(&mlxsw_sp->router->lock);
1742 	switch (event) {
1743 	case NETDEV_REGISTER:
1744 		err = mlxsw_sp_netdevice_ipip_ol_reg_event(mlxsw_sp, ol_dev);
1745 		break;
1746 	case NETDEV_UNREGISTER:
1747 		mlxsw_sp_netdevice_ipip_ol_unreg_event(mlxsw_sp, ol_dev);
1748 		break;
1749 	case NETDEV_UP:
1750 		mlxsw_sp_netdevice_ipip_ol_up_event(mlxsw_sp, ol_dev);
1751 		break;
1752 	case NETDEV_DOWN:
1753 		mlxsw_sp_netdevice_ipip_ol_down_event(mlxsw_sp, ol_dev);
1754 		break;
1755 	case NETDEV_CHANGEUPPER:
1756 		chup = container_of(info, typeof(*chup), info);
1757 		extack = info->extack;
1758 		if (netif_is_l3_master(chup->upper_dev))
1759 			err = mlxsw_sp_netdevice_ipip_ol_vrf_event(mlxsw_sp,
1760 								   ol_dev,
1761 								   extack);
1762 		break;
1763 	case NETDEV_CHANGE:
1764 		extack = info->extack;
1765 		err = mlxsw_sp_netdevice_ipip_ol_change_event(mlxsw_sp,
1766 							      ol_dev, extack);
1767 		break;
1768 	case NETDEV_CHANGEMTU:
1769 		err = mlxsw_sp_netdevice_ipip_ol_update_mtu(mlxsw_sp, ol_dev);
1770 		break;
1771 	}
1772 	mutex_unlock(&mlxsw_sp->router->lock);
1773 	return err;
1774 }
1775 
1776 static int
1777 __mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1778 				   struct mlxsw_sp_ipip_entry *ipip_entry,
1779 				   struct net_device *ul_dev,
1780 				   bool *demote_this,
1781 				   unsigned long event,
1782 				   struct netdev_notifier_info *info)
1783 {
1784 	struct netdev_notifier_changeupper_info *chup;
1785 	struct netlink_ext_ack *extack;
1786 
1787 	switch (event) {
1788 	case NETDEV_CHANGEUPPER:
1789 		chup = container_of(info, typeof(*chup), info);
1790 		extack = info->extack;
1791 		if (netif_is_l3_master(chup->upper_dev))
1792 			return mlxsw_sp_netdevice_ipip_ul_vrf_event(mlxsw_sp,
1793 								    ipip_entry,
1794 								    ul_dev,
1795 								    demote_this,
1796 								    extack);
1797 		break;
1798 
1799 	case NETDEV_UP:
1800 		return mlxsw_sp_netdevice_ipip_ul_up_event(mlxsw_sp, ipip_entry,
1801 							   ul_dev);
1802 	case NETDEV_DOWN:
1803 		return mlxsw_sp_netdevice_ipip_ul_down_event(mlxsw_sp,
1804 							     ipip_entry,
1805 							     ul_dev);
1806 	}
1807 	return 0;
1808 }
1809 
1810 int
1811 mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1812 				 struct net_device *ul_dev,
1813 				 unsigned long event,
1814 				 struct netdev_notifier_info *info)
1815 {
1816 	struct mlxsw_sp_ipip_entry *ipip_entry = NULL;
1817 	int err = 0;
1818 
1819 	mutex_lock(&mlxsw_sp->router->lock);
1820 	while ((ipip_entry = mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp,
1821 								ul_dev,
1822 								ipip_entry))) {
1823 		struct mlxsw_sp_ipip_entry *prev;
1824 		bool demote_this = false;
1825 
1826 		err = __mlxsw_sp_netdevice_ipip_ul_event(mlxsw_sp, ipip_entry,
1827 							 ul_dev, &demote_this,
1828 							 event, info);
1829 		if (err) {
1830 			mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(mlxsw_sp,
1831 								 ul_dev);
1832 			break;
1833 		}
1834 
1835 		if (demote_this) {
1836 			if (list_is_first(&ipip_entry->ipip_list_node,
1837 					  &mlxsw_sp->router->ipip_list))
1838 				prev = NULL;
1839 			else
1840 				/* This can't be cached from previous iteration,
1841 				 * because that entry could be gone now.
1842 				 */
1843 				prev = list_prev_entry(ipip_entry,
1844 						       ipip_list_node);
1845 			mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1846 			ipip_entry = prev;
1847 		}
1848 	}
1849 	mutex_unlock(&mlxsw_sp->router->lock);
1850 
1851 	return err;
1852 }
1853 
1854 int mlxsw_sp_router_nve_promote_decap(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
1855 				      enum mlxsw_sp_l3proto ul_proto,
1856 				      const union mlxsw_sp_l3addr *ul_sip,
1857 				      u32 tunnel_index)
1858 {
1859 	enum mlxsw_sp_fib_entry_type type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1860 	struct mlxsw_sp_router *router = mlxsw_sp->router;
1861 	struct mlxsw_sp_fib_entry *fib_entry;
1862 	int err = 0;
1863 
1864 	mutex_lock(&mlxsw_sp->router->lock);
1865 
1866 	if (WARN_ON_ONCE(router->nve_decap_config.valid)) {
1867 		err = -EINVAL;
1868 		goto out;
1869 	}
1870 
1871 	router->nve_decap_config.ul_tb_id = ul_tb_id;
1872 	router->nve_decap_config.tunnel_index = tunnel_index;
1873 	router->nve_decap_config.ul_proto = ul_proto;
1874 	router->nve_decap_config.ul_sip = *ul_sip;
1875 	router->nve_decap_config.valid = true;
1876 
1877 	/* It is valid to create a tunnel with a local IP and only later
1878 	 * assign this IP address to a local interface
1879 	 */
1880 	fib_entry = mlxsw_sp_router_ip2me_fib_entry_find(mlxsw_sp, ul_tb_id,
1881 							 ul_proto, ul_sip,
1882 							 type);
1883 	if (!fib_entry)
1884 		goto out;
1885 
1886 	fib_entry->decap.tunnel_index = tunnel_index;
1887 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
1888 
1889 	err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1890 	if (err)
1891 		goto err_fib_entry_update;
1892 
1893 	goto out;
1894 
1895 err_fib_entry_update:
1896 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1897 	mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1898 out:
1899 	mutex_unlock(&mlxsw_sp->router->lock);
1900 	return err;
1901 }
1902 
1903 void mlxsw_sp_router_nve_demote_decap(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
1904 				      enum mlxsw_sp_l3proto ul_proto,
1905 				      const union mlxsw_sp_l3addr *ul_sip)
1906 {
1907 	enum mlxsw_sp_fib_entry_type type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
1908 	struct mlxsw_sp_router *router = mlxsw_sp->router;
1909 	struct mlxsw_sp_fib_entry *fib_entry;
1910 
1911 	mutex_lock(&mlxsw_sp->router->lock);
1912 
1913 	if (WARN_ON_ONCE(!router->nve_decap_config.valid))
1914 		goto out;
1915 
1916 	router->nve_decap_config.valid = false;
1917 
1918 	fib_entry = mlxsw_sp_router_ip2me_fib_entry_find(mlxsw_sp, ul_tb_id,
1919 							 ul_proto, ul_sip,
1920 							 type);
1921 	if (!fib_entry)
1922 		goto out;
1923 
1924 	fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1925 	mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1926 out:
1927 	mutex_unlock(&mlxsw_sp->router->lock);
1928 }
1929 
1930 static bool mlxsw_sp_router_nve_is_decap(struct mlxsw_sp *mlxsw_sp,
1931 					 u32 ul_tb_id,
1932 					 enum mlxsw_sp_l3proto ul_proto,
1933 					 const union mlxsw_sp_l3addr *ul_sip)
1934 {
1935 	struct mlxsw_sp_router *router = mlxsw_sp->router;
1936 
1937 	return router->nve_decap_config.valid &&
1938 	       router->nve_decap_config.ul_tb_id == ul_tb_id &&
1939 	       router->nve_decap_config.ul_proto == ul_proto &&
1940 	       !memcmp(&router->nve_decap_config.ul_sip, ul_sip,
1941 		       sizeof(*ul_sip));
1942 }
1943 
1944 struct mlxsw_sp_neigh_key {
1945 	struct neighbour *n;
1946 };
1947 
1948 struct mlxsw_sp_neigh_entry {
1949 	struct list_head rif_list_node;
1950 	struct rhash_head ht_node;
1951 	struct mlxsw_sp_neigh_key key;
1952 	u16 rif;
1953 	bool connected;
1954 	unsigned char ha[ETH_ALEN];
1955 	struct list_head nexthop_list; /* list of nexthops using
1956 					* this neigh entry
1957 					*/
1958 	struct list_head nexthop_neighs_list_node;
1959 	unsigned int counter_index;
1960 	bool counter_valid;
1961 };
1962 
1963 static const struct rhashtable_params mlxsw_sp_neigh_ht_params = {
1964 	.key_offset = offsetof(struct mlxsw_sp_neigh_entry, key),
1965 	.head_offset = offsetof(struct mlxsw_sp_neigh_entry, ht_node),
1966 	.key_len = sizeof(struct mlxsw_sp_neigh_key),
1967 };
1968 
1969 struct mlxsw_sp_neigh_entry *
1970 mlxsw_sp_rif_neigh_next(struct mlxsw_sp_rif *rif,
1971 			struct mlxsw_sp_neigh_entry *neigh_entry)
1972 {
1973 	if (!neigh_entry) {
1974 		if (list_empty(&rif->neigh_list))
1975 			return NULL;
1976 		else
1977 			return list_first_entry(&rif->neigh_list,
1978 						typeof(*neigh_entry),
1979 						rif_list_node);
1980 	}
1981 	if (list_is_last(&neigh_entry->rif_list_node, &rif->neigh_list))
1982 		return NULL;
1983 	return list_next_entry(neigh_entry, rif_list_node);
1984 }
1985 
1986 int mlxsw_sp_neigh_entry_type(struct mlxsw_sp_neigh_entry *neigh_entry)
1987 {
1988 	return neigh_entry->key.n->tbl->family;
1989 }
1990 
1991 unsigned char *
1992 mlxsw_sp_neigh_entry_ha(struct mlxsw_sp_neigh_entry *neigh_entry)
1993 {
1994 	return neigh_entry->ha;
1995 }
1996 
1997 u32 mlxsw_sp_neigh4_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
1998 {
1999 	struct neighbour *n;
2000 
2001 	n = neigh_entry->key.n;
2002 	return ntohl(*((__be32 *) n->primary_key));
2003 }
2004 
2005 struct in6_addr *
2006 mlxsw_sp_neigh6_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
2007 {
2008 	struct neighbour *n;
2009 
2010 	n = neigh_entry->key.n;
2011 	return (struct in6_addr *) &n->primary_key;
2012 }
2013 
2014 int mlxsw_sp_neigh_counter_get(struct mlxsw_sp *mlxsw_sp,
2015 			       struct mlxsw_sp_neigh_entry *neigh_entry,
2016 			       u64 *p_counter)
2017 {
2018 	if (!neigh_entry->counter_valid)
2019 		return -EINVAL;
2020 
2021 	return mlxsw_sp_flow_counter_get(mlxsw_sp, neigh_entry->counter_index,
2022 					 p_counter, NULL);
2023 }
2024 
2025 static struct mlxsw_sp_neigh_entry *
2026 mlxsw_sp_neigh_entry_alloc(struct mlxsw_sp *mlxsw_sp, struct neighbour *n,
2027 			   u16 rif)
2028 {
2029 	struct mlxsw_sp_neigh_entry *neigh_entry;
2030 
2031 	neigh_entry = kzalloc(sizeof(*neigh_entry), GFP_KERNEL);
2032 	if (!neigh_entry)
2033 		return NULL;
2034 
2035 	neigh_entry->key.n = n;
2036 	neigh_entry->rif = rif;
2037 	INIT_LIST_HEAD(&neigh_entry->nexthop_list);
2038 
2039 	return neigh_entry;
2040 }
2041 
2042 static void mlxsw_sp_neigh_entry_free(struct mlxsw_sp_neigh_entry *neigh_entry)
2043 {
2044 	kfree(neigh_entry);
2045 }
2046 
2047 static int
2048 mlxsw_sp_neigh_entry_insert(struct mlxsw_sp *mlxsw_sp,
2049 			    struct mlxsw_sp_neigh_entry *neigh_entry)
2050 {
2051 	return rhashtable_insert_fast(&mlxsw_sp->router->neigh_ht,
2052 				      &neigh_entry->ht_node,
2053 				      mlxsw_sp_neigh_ht_params);
2054 }
2055 
2056 static void
2057 mlxsw_sp_neigh_entry_remove(struct mlxsw_sp *mlxsw_sp,
2058 			    struct mlxsw_sp_neigh_entry *neigh_entry)
2059 {
2060 	rhashtable_remove_fast(&mlxsw_sp->router->neigh_ht,
2061 			       &neigh_entry->ht_node,
2062 			       mlxsw_sp_neigh_ht_params);
2063 }
2064 
2065 static bool
2066 mlxsw_sp_neigh_counter_should_alloc(struct mlxsw_sp *mlxsw_sp,
2067 				    struct mlxsw_sp_neigh_entry *neigh_entry)
2068 {
2069 	struct devlink *devlink;
2070 	const char *table_name;
2071 
2072 	switch (mlxsw_sp_neigh_entry_type(neigh_entry)) {
2073 	case AF_INET:
2074 		table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST4;
2075 		break;
2076 	case AF_INET6:
2077 		table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST6;
2078 		break;
2079 	default:
2080 		WARN_ON(1);
2081 		return false;
2082 	}
2083 
2084 	devlink = priv_to_devlink(mlxsw_sp->core);
2085 	return devlink_dpipe_table_counter_enabled(devlink, table_name);
2086 }
2087 
2088 static void
2089 mlxsw_sp_neigh_counter_alloc(struct mlxsw_sp *mlxsw_sp,
2090 			     struct mlxsw_sp_neigh_entry *neigh_entry)
2091 {
2092 	if (!mlxsw_sp_neigh_counter_should_alloc(mlxsw_sp, neigh_entry))
2093 		return;
2094 
2095 	if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &neigh_entry->counter_index))
2096 		return;
2097 
2098 	neigh_entry->counter_valid = true;
2099 }
2100 
2101 static void
2102 mlxsw_sp_neigh_counter_free(struct mlxsw_sp *mlxsw_sp,
2103 			    struct mlxsw_sp_neigh_entry *neigh_entry)
2104 {
2105 	if (!neigh_entry->counter_valid)
2106 		return;
2107 	mlxsw_sp_flow_counter_free(mlxsw_sp,
2108 				   neigh_entry->counter_index);
2109 	neigh_entry->counter_valid = false;
2110 }
2111 
2112 static struct mlxsw_sp_neigh_entry *
2113 mlxsw_sp_neigh_entry_create(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
2114 {
2115 	struct mlxsw_sp_neigh_entry *neigh_entry;
2116 	struct mlxsw_sp_rif *rif;
2117 	int err;
2118 
2119 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, n->dev);
2120 	if (!rif)
2121 		return ERR_PTR(-EINVAL);
2122 
2123 	neigh_entry = mlxsw_sp_neigh_entry_alloc(mlxsw_sp, n, rif->rif_index);
2124 	if (!neigh_entry)
2125 		return ERR_PTR(-ENOMEM);
2126 
2127 	err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
2128 	if (err)
2129 		goto err_neigh_entry_insert;
2130 
2131 	mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
2132 	list_add(&neigh_entry->rif_list_node, &rif->neigh_list);
2133 
2134 	return neigh_entry;
2135 
2136 err_neigh_entry_insert:
2137 	mlxsw_sp_neigh_entry_free(neigh_entry);
2138 	return ERR_PTR(err);
2139 }
2140 
2141 static void
2142 mlxsw_sp_neigh_entry_destroy(struct mlxsw_sp *mlxsw_sp,
2143 			     struct mlxsw_sp_neigh_entry *neigh_entry)
2144 {
2145 	list_del(&neigh_entry->rif_list_node);
2146 	mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
2147 	mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
2148 	mlxsw_sp_neigh_entry_free(neigh_entry);
2149 }
2150 
2151 static struct mlxsw_sp_neigh_entry *
2152 mlxsw_sp_neigh_entry_lookup(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
2153 {
2154 	struct mlxsw_sp_neigh_key key;
2155 
2156 	key.n = n;
2157 	return rhashtable_lookup_fast(&mlxsw_sp->router->neigh_ht,
2158 				      &key, mlxsw_sp_neigh_ht_params);
2159 }
2160 
2161 static void
2162 mlxsw_sp_router_neighs_update_interval_init(struct mlxsw_sp *mlxsw_sp)
2163 {
2164 	unsigned long interval;
2165 
2166 #if IS_ENABLED(CONFIG_IPV6)
2167 	interval = min_t(unsigned long,
2168 			 NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME),
2169 			 NEIGH_VAR(&nd_tbl.parms, DELAY_PROBE_TIME));
2170 #else
2171 	interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME);
2172 #endif
2173 	mlxsw_sp->router->neighs_update.interval = jiffies_to_msecs(interval);
2174 }
2175 
2176 static void mlxsw_sp_router_neigh_ent_ipv4_process(struct mlxsw_sp *mlxsw_sp,
2177 						   char *rauhtd_pl,
2178 						   int ent_index)
2179 {
2180 	struct net_device *dev;
2181 	struct neighbour *n;
2182 	__be32 dipn;
2183 	u32 dip;
2184 	u16 rif;
2185 
2186 	mlxsw_reg_rauhtd_ent_ipv4_unpack(rauhtd_pl, ent_index, &rif, &dip);
2187 
2188 	if (!mlxsw_sp->router->rifs[rif]) {
2189 		dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
2190 		return;
2191 	}
2192 
2193 	dipn = htonl(dip);
2194 	dev = mlxsw_sp->router->rifs[rif]->dev;
2195 	n = neigh_lookup(&arp_tbl, &dipn, dev);
2196 	if (!n)
2197 		return;
2198 
2199 	netdev_dbg(dev, "Updating neighbour with IP=%pI4h\n", &dip);
2200 	neigh_event_send(n, NULL);
2201 	neigh_release(n);
2202 }
2203 
2204 #if IS_ENABLED(CONFIG_IPV6)
2205 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2206 						   char *rauhtd_pl,
2207 						   int rec_index)
2208 {
2209 	struct net_device *dev;
2210 	struct neighbour *n;
2211 	struct in6_addr dip;
2212 	u16 rif;
2213 
2214 	mlxsw_reg_rauhtd_ent_ipv6_unpack(rauhtd_pl, rec_index, &rif,
2215 					 (char *) &dip);
2216 
2217 	if (!mlxsw_sp->router->rifs[rif]) {
2218 		dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
2219 		return;
2220 	}
2221 
2222 	dev = mlxsw_sp->router->rifs[rif]->dev;
2223 	n = neigh_lookup(&nd_tbl, &dip, dev);
2224 	if (!n)
2225 		return;
2226 
2227 	netdev_dbg(dev, "Updating neighbour with IP=%pI6c\n", &dip);
2228 	neigh_event_send(n, NULL);
2229 	neigh_release(n);
2230 }
2231 #else
2232 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2233 						   char *rauhtd_pl,
2234 						   int rec_index)
2235 {
2236 }
2237 #endif
2238 
2239 static void mlxsw_sp_router_neigh_rec_ipv4_process(struct mlxsw_sp *mlxsw_sp,
2240 						   char *rauhtd_pl,
2241 						   int rec_index)
2242 {
2243 	u8 num_entries;
2244 	int i;
2245 
2246 	num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
2247 								rec_index);
2248 	/* Hardware starts counting at 0, so add 1. */
2249 	num_entries++;
2250 
2251 	/* Each record consists of several neighbour entries. */
2252 	for (i = 0; i < num_entries; i++) {
2253 		int ent_index;
2254 
2255 		ent_index = rec_index * MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC + i;
2256 		mlxsw_sp_router_neigh_ent_ipv4_process(mlxsw_sp, rauhtd_pl,
2257 						       ent_index);
2258 	}
2259 
2260 }
2261 
2262 static void mlxsw_sp_router_neigh_rec_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2263 						   char *rauhtd_pl,
2264 						   int rec_index)
2265 {
2266 	/* One record contains one entry. */
2267 	mlxsw_sp_router_neigh_ent_ipv6_process(mlxsw_sp, rauhtd_pl,
2268 					       rec_index);
2269 }
2270 
2271 static void mlxsw_sp_router_neigh_rec_process(struct mlxsw_sp *mlxsw_sp,
2272 					      char *rauhtd_pl, int rec_index)
2273 {
2274 	switch (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, rec_index)) {
2275 	case MLXSW_REG_RAUHTD_TYPE_IPV4:
2276 		mlxsw_sp_router_neigh_rec_ipv4_process(mlxsw_sp, rauhtd_pl,
2277 						       rec_index);
2278 		break;
2279 	case MLXSW_REG_RAUHTD_TYPE_IPV6:
2280 		mlxsw_sp_router_neigh_rec_ipv6_process(mlxsw_sp, rauhtd_pl,
2281 						       rec_index);
2282 		break;
2283 	}
2284 }
2285 
2286 static bool mlxsw_sp_router_rauhtd_is_full(char *rauhtd_pl)
2287 {
2288 	u8 num_rec, last_rec_index, num_entries;
2289 
2290 	num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2291 	last_rec_index = num_rec - 1;
2292 
2293 	if (num_rec < MLXSW_REG_RAUHTD_REC_MAX_NUM)
2294 		return false;
2295 	if (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, last_rec_index) ==
2296 	    MLXSW_REG_RAUHTD_TYPE_IPV6)
2297 		return true;
2298 
2299 	num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
2300 								last_rec_index);
2301 	if (++num_entries == MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC)
2302 		return true;
2303 	return false;
2304 }
2305 
2306 static int
2307 __mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp,
2308 				       char *rauhtd_pl,
2309 				       enum mlxsw_reg_rauhtd_type type)
2310 {
2311 	int i, num_rec;
2312 	int err;
2313 
2314 	/* Ensure the RIF we read from the device does not change mid-dump. */
2315 	mutex_lock(&mlxsw_sp->router->lock);
2316 	do {
2317 		mlxsw_reg_rauhtd_pack(rauhtd_pl, type);
2318 		err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(rauhtd),
2319 				      rauhtd_pl);
2320 		if (err) {
2321 			dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Failed to dump neighbour table\n");
2322 			break;
2323 		}
2324 		num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2325 		for (i = 0; i < num_rec; i++)
2326 			mlxsw_sp_router_neigh_rec_process(mlxsw_sp, rauhtd_pl,
2327 							  i);
2328 	} while (mlxsw_sp_router_rauhtd_is_full(rauhtd_pl));
2329 	mutex_unlock(&mlxsw_sp->router->lock);
2330 
2331 	return err;
2332 }
2333 
2334 static int mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp)
2335 {
2336 	enum mlxsw_reg_rauhtd_type type;
2337 	char *rauhtd_pl;
2338 	int err;
2339 
2340 	rauhtd_pl = kmalloc(MLXSW_REG_RAUHTD_LEN, GFP_KERNEL);
2341 	if (!rauhtd_pl)
2342 		return -ENOMEM;
2343 
2344 	type = MLXSW_REG_RAUHTD_TYPE_IPV4;
2345 	err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2346 	if (err)
2347 		goto out;
2348 
2349 	type = MLXSW_REG_RAUHTD_TYPE_IPV6;
2350 	err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2351 out:
2352 	kfree(rauhtd_pl);
2353 	return err;
2354 }
2355 
2356 static void mlxsw_sp_router_neighs_update_nh(struct mlxsw_sp *mlxsw_sp)
2357 {
2358 	struct mlxsw_sp_neigh_entry *neigh_entry;
2359 
2360 	mutex_lock(&mlxsw_sp->router->lock);
2361 	list_for_each_entry(neigh_entry, &mlxsw_sp->router->nexthop_neighs_list,
2362 			    nexthop_neighs_list_node)
2363 		/* If this neigh have nexthops, make the kernel think this neigh
2364 		 * is active regardless of the traffic.
2365 		 */
2366 		neigh_event_send(neigh_entry->key.n, NULL);
2367 	mutex_unlock(&mlxsw_sp->router->lock);
2368 }
2369 
2370 static void
2371 mlxsw_sp_router_neighs_update_work_schedule(struct mlxsw_sp *mlxsw_sp)
2372 {
2373 	unsigned long interval = mlxsw_sp->router->neighs_update.interval;
2374 
2375 	mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw,
2376 			       msecs_to_jiffies(interval));
2377 }
2378 
2379 static void mlxsw_sp_router_neighs_update_work(struct work_struct *work)
2380 {
2381 	struct mlxsw_sp_router *router;
2382 	int err;
2383 
2384 	router = container_of(work, struct mlxsw_sp_router,
2385 			      neighs_update.dw.work);
2386 	err = mlxsw_sp_router_neighs_update_rauhtd(router->mlxsw_sp);
2387 	if (err)
2388 		dev_err(router->mlxsw_sp->bus_info->dev, "Could not update kernel for neigh activity");
2389 
2390 	mlxsw_sp_router_neighs_update_nh(router->mlxsw_sp);
2391 
2392 	mlxsw_sp_router_neighs_update_work_schedule(router->mlxsw_sp);
2393 }
2394 
2395 static void mlxsw_sp_router_probe_unresolved_nexthops(struct work_struct *work)
2396 {
2397 	struct mlxsw_sp_neigh_entry *neigh_entry;
2398 	struct mlxsw_sp_router *router;
2399 
2400 	router = container_of(work, struct mlxsw_sp_router,
2401 			      nexthop_probe_dw.work);
2402 	/* Iterate over nexthop neighbours, find those who are unresolved and
2403 	 * send arp on them. This solves the chicken-egg problem when
2404 	 * the nexthop wouldn't get offloaded until the neighbor is resolved
2405 	 * but it wouldn't get resolved ever in case traffic is flowing in HW
2406 	 * using different nexthop.
2407 	 */
2408 	mutex_lock(&router->lock);
2409 	list_for_each_entry(neigh_entry, &router->nexthop_neighs_list,
2410 			    nexthop_neighs_list_node)
2411 		if (!neigh_entry->connected)
2412 			neigh_event_send(neigh_entry->key.n, NULL);
2413 	mutex_unlock(&router->lock);
2414 
2415 	mlxsw_core_schedule_dw(&router->nexthop_probe_dw,
2416 			       MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL);
2417 }
2418 
2419 static void
2420 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
2421 			      struct mlxsw_sp_neigh_entry *neigh_entry,
2422 			      bool removing, bool dead);
2423 
2424 static enum mlxsw_reg_rauht_op mlxsw_sp_rauht_op(bool adding)
2425 {
2426 	return adding ? MLXSW_REG_RAUHT_OP_WRITE_ADD :
2427 			MLXSW_REG_RAUHT_OP_WRITE_DELETE;
2428 }
2429 
2430 static int
2431 mlxsw_sp_router_neigh_entry_op4(struct mlxsw_sp *mlxsw_sp,
2432 				struct mlxsw_sp_neigh_entry *neigh_entry,
2433 				enum mlxsw_reg_rauht_op op)
2434 {
2435 	struct neighbour *n = neigh_entry->key.n;
2436 	u32 dip = ntohl(*((__be32 *) n->primary_key));
2437 	char rauht_pl[MLXSW_REG_RAUHT_LEN];
2438 
2439 	mlxsw_reg_rauht_pack4(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2440 			      dip);
2441 	if (neigh_entry->counter_valid)
2442 		mlxsw_reg_rauht_pack_counter(rauht_pl,
2443 					     neigh_entry->counter_index);
2444 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2445 }
2446 
2447 static int
2448 mlxsw_sp_router_neigh_entry_op6(struct mlxsw_sp *mlxsw_sp,
2449 				struct mlxsw_sp_neigh_entry *neigh_entry,
2450 				enum mlxsw_reg_rauht_op op)
2451 {
2452 	struct neighbour *n = neigh_entry->key.n;
2453 	char rauht_pl[MLXSW_REG_RAUHT_LEN];
2454 	const char *dip = n->primary_key;
2455 
2456 	mlxsw_reg_rauht_pack6(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2457 			      dip);
2458 	if (neigh_entry->counter_valid)
2459 		mlxsw_reg_rauht_pack_counter(rauht_pl,
2460 					     neigh_entry->counter_index);
2461 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2462 }
2463 
2464 bool mlxsw_sp_neigh_ipv6_ignore(struct mlxsw_sp_neigh_entry *neigh_entry)
2465 {
2466 	struct neighbour *n = neigh_entry->key.n;
2467 
2468 	/* Packets with a link-local destination address are trapped
2469 	 * after LPM lookup and never reach the neighbour table, so
2470 	 * there is no need to program such neighbours to the device.
2471 	 */
2472 	if (ipv6_addr_type((struct in6_addr *) &n->primary_key) &
2473 	    IPV6_ADDR_LINKLOCAL)
2474 		return true;
2475 	return false;
2476 }
2477 
2478 static void
2479 mlxsw_sp_neigh_entry_update(struct mlxsw_sp *mlxsw_sp,
2480 			    struct mlxsw_sp_neigh_entry *neigh_entry,
2481 			    bool adding)
2482 {
2483 	enum mlxsw_reg_rauht_op op = mlxsw_sp_rauht_op(adding);
2484 	int err;
2485 
2486 	if (!adding && !neigh_entry->connected)
2487 		return;
2488 	neigh_entry->connected = adding;
2489 	if (neigh_entry->key.n->tbl->family == AF_INET) {
2490 		err = mlxsw_sp_router_neigh_entry_op4(mlxsw_sp, neigh_entry,
2491 						      op);
2492 		if (err)
2493 			return;
2494 	} else if (neigh_entry->key.n->tbl->family == AF_INET6) {
2495 		if (mlxsw_sp_neigh_ipv6_ignore(neigh_entry))
2496 			return;
2497 		err = mlxsw_sp_router_neigh_entry_op6(mlxsw_sp, neigh_entry,
2498 						      op);
2499 		if (err)
2500 			return;
2501 	} else {
2502 		WARN_ON_ONCE(1);
2503 		return;
2504 	}
2505 
2506 	if (adding)
2507 		neigh_entry->key.n->flags |= NTF_OFFLOADED;
2508 	else
2509 		neigh_entry->key.n->flags &= ~NTF_OFFLOADED;
2510 }
2511 
2512 void
2513 mlxsw_sp_neigh_entry_counter_update(struct mlxsw_sp *mlxsw_sp,
2514 				    struct mlxsw_sp_neigh_entry *neigh_entry,
2515 				    bool adding)
2516 {
2517 	if (adding)
2518 		mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
2519 	else
2520 		mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
2521 	mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, true);
2522 }
2523 
2524 struct mlxsw_sp_netevent_work {
2525 	struct work_struct work;
2526 	struct mlxsw_sp *mlxsw_sp;
2527 	struct neighbour *n;
2528 };
2529 
2530 static void mlxsw_sp_router_neigh_event_work(struct work_struct *work)
2531 {
2532 	struct mlxsw_sp_netevent_work *net_work =
2533 		container_of(work, struct mlxsw_sp_netevent_work, work);
2534 	struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2535 	struct mlxsw_sp_neigh_entry *neigh_entry;
2536 	struct neighbour *n = net_work->n;
2537 	unsigned char ha[ETH_ALEN];
2538 	bool entry_connected;
2539 	u8 nud_state, dead;
2540 
2541 	/* If these parameters are changed after we release the lock,
2542 	 * then we are guaranteed to receive another event letting us
2543 	 * know about it.
2544 	 */
2545 	read_lock_bh(&n->lock);
2546 	memcpy(ha, n->ha, ETH_ALEN);
2547 	nud_state = n->nud_state;
2548 	dead = n->dead;
2549 	read_unlock_bh(&n->lock);
2550 
2551 	mutex_lock(&mlxsw_sp->router->lock);
2552 	mlxsw_sp_span_respin(mlxsw_sp);
2553 
2554 	entry_connected = nud_state & NUD_VALID && !dead;
2555 	neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
2556 	if (!entry_connected && !neigh_entry)
2557 		goto out;
2558 	if (!neigh_entry) {
2559 		neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
2560 		if (IS_ERR(neigh_entry))
2561 			goto out;
2562 	}
2563 
2564 	memcpy(neigh_entry->ha, ha, ETH_ALEN);
2565 	mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, entry_connected);
2566 	mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, !entry_connected,
2567 				      dead);
2568 
2569 	if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
2570 		mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2571 
2572 out:
2573 	mutex_unlock(&mlxsw_sp->router->lock);
2574 	neigh_release(n);
2575 	kfree(net_work);
2576 }
2577 
2578 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp);
2579 
2580 static void mlxsw_sp_router_mp_hash_event_work(struct work_struct *work)
2581 {
2582 	struct mlxsw_sp_netevent_work *net_work =
2583 		container_of(work, struct mlxsw_sp_netevent_work, work);
2584 	struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2585 
2586 	mlxsw_sp_mp_hash_init(mlxsw_sp);
2587 	kfree(net_work);
2588 }
2589 
2590 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp);
2591 
2592 static void mlxsw_sp_router_update_priority_work(struct work_struct *work)
2593 {
2594 	struct mlxsw_sp_netevent_work *net_work =
2595 		container_of(work, struct mlxsw_sp_netevent_work, work);
2596 	struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2597 
2598 	__mlxsw_sp_router_init(mlxsw_sp);
2599 	kfree(net_work);
2600 }
2601 
2602 static int mlxsw_sp_router_schedule_work(struct net *net,
2603 					 struct notifier_block *nb,
2604 					 void (*cb)(struct work_struct *))
2605 {
2606 	struct mlxsw_sp_netevent_work *net_work;
2607 	struct mlxsw_sp_router *router;
2608 
2609 	router = container_of(nb, struct mlxsw_sp_router, netevent_nb);
2610 	if (!net_eq(net, mlxsw_sp_net(router->mlxsw_sp)))
2611 		return NOTIFY_DONE;
2612 
2613 	net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2614 	if (!net_work)
2615 		return NOTIFY_BAD;
2616 
2617 	INIT_WORK(&net_work->work, cb);
2618 	net_work->mlxsw_sp = router->mlxsw_sp;
2619 	mlxsw_core_schedule_work(&net_work->work);
2620 	return NOTIFY_DONE;
2621 }
2622 
2623 static int mlxsw_sp_router_netevent_event(struct notifier_block *nb,
2624 					  unsigned long event, void *ptr)
2625 {
2626 	struct mlxsw_sp_netevent_work *net_work;
2627 	struct mlxsw_sp_port *mlxsw_sp_port;
2628 	struct mlxsw_sp *mlxsw_sp;
2629 	unsigned long interval;
2630 	struct neigh_parms *p;
2631 	struct neighbour *n;
2632 
2633 	switch (event) {
2634 	case NETEVENT_DELAY_PROBE_TIME_UPDATE:
2635 		p = ptr;
2636 
2637 		/* We don't care about changes in the default table. */
2638 		if (!p->dev || (p->tbl->family != AF_INET &&
2639 				p->tbl->family != AF_INET6))
2640 			return NOTIFY_DONE;
2641 
2642 		/* We are in atomic context and can't take RTNL mutex,
2643 		 * so use RCU variant to walk the device chain.
2644 		 */
2645 		mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(p->dev);
2646 		if (!mlxsw_sp_port)
2647 			return NOTIFY_DONE;
2648 
2649 		mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2650 		interval = jiffies_to_msecs(NEIGH_VAR(p, DELAY_PROBE_TIME));
2651 		mlxsw_sp->router->neighs_update.interval = interval;
2652 
2653 		mlxsw_sp_port_dev_put(mlxsw_sp_port);
2654 		break;
2655 	case NETEVENT_NEIGH_UPDATE:
2656 		n = ptr;
2657 
2658 		if (n->tbl->family != AF_INET && n->tbl->family != AF_INET6)
2659 			return NOTIFY_DONE;
2660 
2661 		mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(n->dev);
2662 		if (!mlxsw_sp_port)
2663 			return NOTIFY_DONE;
2664 
2665 		net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2666 		if (!net_work) {
2667 			mlxsw_sp_port_dev_put(mlxsw_sp_port);
2668 			return NOTIFY_BAD;
2669 		}
2670 
2671 		INIT_WORK(&net_work->work, mlxsw_sp_router_neigh_event_work);
2672 		net_work->mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2673 		net_work->n = n;
2674 
2675 		/* Take a reference to ensure the neighbour won't be
2676 		 * destructed until we drop the reference in delayed
2677 		 * work.
2678 		 */
2679 		neigh_clone(n);
2680 		mlxsw_core_schedule_work(&net_work->work);
2681 		mlxsw_sp_port_dev_put(mlxsw_sp_port);
2682 		break;
2683 	case NETEVENT_IPV4_MPATH_HASH_UPDATE:
2684 	case NETEVENT_IPV6_MPATH_HASH_UPDATE:
2685 		return mlxsw_sp_router_schedule_work(ptr, nb,
2686 				mlxsw_sp_router_mp_hash_event_work);
2687 
2688 	case NETEVENT_IPV4_FWD_UPDATE_PRIORITY_UPDATE:
2689 		return mlxsw_sp_router_schedule_work(ptr, nb,
2690 				mlxsw_sp_router_update_priority_work);
2691 	}
2692 
2693 	return NOTIFY_DONE;
2694 }
2695 
2696 static int mlxsw_sp_neigh_init(struct mlxsw_sp *mlxsw_sp)
2697 {
2698 	int err;
2699 
2700 	err = rhashtable_init(&mlxsw_sp->router->neigh_ht,
2701 			      &mlxsw_sp_neigh_ht_params);
2702 	if (err)
2703 		return err;
2704 
2705 	/* Initialize the polling interval according to the default
2706 	 * table.
2707 	 */
2708 	mlxsw_sp_router_neighs_update_interval_init(mlxsw_sp);
2709 
2710 	/* Create the delayed works for the activity_update */
2711 	INIT_DELAYED_WORK(&mlxsw_sp->router->neighs_update.dw,
2712 			  mlxsw_sp_router_neighs_update_work);
2713 	INIT_DELAYED_WORK(&mlxsw_sp->router->nexthop_probe_dw,
2714 			  mlxsw_sp_router_probe_unresolved_nexthops);
2715 	mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw, 0);
2716 	mlxsw_core_schedule_dw(&mlxsw_sp->router->nexthop_probe_dw, 0);
2717 	return 0;
2718 }
2719 
2720 static void mlxsw_sp_neigh_fini(struct mlxsw_sp *mlxsw_sp)
2721 {
2722 	cancel_delayed_work_sync(&mlxsw_sp->router->neighs_update.dw);
2723 	cancel_delayed_work_sync(&mlxsw_sp->router->nexthop_probe_dw);
2724 	rhashtable_destroy(&mlxsw_sp->router->neigh_ht);
2725 }
2726 
2727 static void mlxsw_sp_neigh_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
2728 					 struct mlxsw_sp_rif *rif)
2729 {
2730 	struct mlxsw_sp_neigh_entry *neigh_entry, *tmp;
2731 
2732 	list_for_each_entry_safe(neigh_entry, tmp, &rif->neigh_list,
2733 				 rif_list_node) {
2734 		mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, false);
2735 		mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2736 	}
2737 }
2738 
2739 enum mlxsw_sp_nexthop_type {
2740 	MLXSW_SP_NEXTHOP_TYPE_ETH,
2741 	MLXSW_SP_NEXTHOP_TYPE_IPIP,
2742 };
2743 
2744 struct mlxsw_sp_nexthop_key {
2745 	struct fib_nh *fib_nh;
2746 };
2747 
2748 struct mlxsw_sp_nexthop {
2749 	struct list_head neigh_list_node; /* member of neigh entry list */
2750 	struct list_head rif_list_node;
2751 	struct list_head router_list_node;
2752 	struct mlxsw_sp_nexthop_group *nh_grp; /* pointer back to the group
2753 						* this belongs to
2754 						*/
2755 	struct rhash_head ht_node;
2756 	struct mlxsw_sp_nexthop_key key;
2757 	unsigned char gw_addr[sizeof(struct in6_addr)];
2758 	int ifindex;
2759 	int nh_weight;
2760 	int norm_nh_weight;
2761 	int num_adj_entries;
2762 	struct mlxsw_sp_rif *rif;
2763 	u8 should_offload:1, /* set indicates this neigh is connected and
2764 			      * should be put to KVD linear area of this group.
2765 			      */
2766 	   offloaded:1, /* set in case the neigh is actually put into
2767 			 * KVD linear area of this group.
2768 			 */
2769 	   update:1; /* set indicates that MAC of this neigh should be
2770 		      * updated in HW
2771 		      */
2772 	enum mlxsw_sp_nexthop_type type;
2773 	union {
2774 		struct mlxsw_sp_neigh_entry *neigh_entry;
2775 		struct mlxsw_sp_ipip_entry *ipip_entry;
2776 	};
2777 	unsigned int counter_index;
2778 	bool counter_valid;
2779 };
2780 
2781 struct mlxsw_sp_nexthop_group {
2782 	void *priv;
2783 	struct rhash_head ht_node;
2784 	struct list_head fib_list; /* list of fib entries that use this group */
2785 	struct neigh_table *neigh_tbl;
2786 	u8 adj_index_valid:1,
2787 	   gateway:1; /* routes using the group use a gateway */
2788 	u32 adj_index;
2789 	u16 ecmp_size;
2790 	u16 count;
2791 	int sum_norm_weight;
2792 	struct mlxsw_sp_nexthop nexthops[0];
2793 #define nh_rif	nexthops[0].rif
2794 };
2795 
2796 void mlxsw_sp_nexthop_counter_alloc(struct mlxsw_sp *mlxsw_sp,
2797 				    struct mlxsw_sp_nexthop *nh)
2798 {
2799 	struct devlink *devlink;
2800 
2801 	devlink = priv_to_devlink(mlxsw_sp->core);
2802 	if (!devlink_dpipe_table_counter_enabled(devlink,
2803 						 MLXSW_SP_DPIPE_TABLE_NAME_ADJ))
2804 		return;
2805 
2806 	if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &nh->counter_index))
2807 		return;
2808 
2809 	nh->counter_valid = true;
2810 }
2811 
2812 void mlxsw_sp_nexthop_counter_free(struct mlxsw_sp *mlxsw_sp,
2813 				   struct mlxsw_sp_nexthop *nh)
2814 {
2815 	if (!nh->counter_valid)
2816 		return;
2817 	mlxsw_sp_flow_counter_free(mlxsw_sp, nh->counter_index);
2818 	nh->counter_valid = false;
2819 }
2820 
2821 int mlxsw_sp_nexthop_counter_get(struct mlxsw_sp *mlxsw_sp,
2822 				 struct mlxsw_sp_nexthop *nh, u64 *p_counter)
2823 {
2824 	if (!nh->counter_valid)
2825 		return -EINVAL;
2826 
2827 	return mlxsw_sp_flow_counter_get(mlxsw_sp, nh->counter_index,
2828 					 p_counter, NULL);
2829 }
2830 
2831 struct mlxsw_sp_nexthop *mlxsw_sp_nexthop_next(struct mlxsw_sp_router *router,
2832 					       struct mlxsw_sp_nexthop *nh)
2833 {
2834 	if (!nh) {
2835 		if (list_empty(&router->nexthop_list))
2836 			return NULL;
2837 		else
2838 			return list_first_entry(&router->nexthop_list,
2839 						typeof(*nh), router_list_node);
2840 	}
2841 	if (list_is_last(&nh->router_list_node, &router->nexthop_list))
2842 		return NULL;
2843 	return list_next_entry(nh, router_list_node);
2844 }
2845 
2846 bool mlxsw_sp_nexthop_offload(struct mlxsw_sp_nexthop *nh)
2847 {
2848 	return nh->offloaded;
2849 }
2850 
2851 unsigned char *mlxsw_sp_nexthop_ha(struct mlxsw_sp_nexthop *nh)
2852 {
2853 	if (!nh->offloaded)
2854 		return NULL;
2855 	return nh->neigh_entry->ha;
2856 }
2857 
2858 int mlxsw_sp_nexthop_indexes(struct mlxsw_sp_nexthop *nh, u32 *p_adj_index,
2859 			     u32 *p_adj_size, u32 *p_adj_hash_index)
2860 {
2861 	struct mlxsw_sp_nexthop_group *nh_grp = nh->nh_grp;
2862 	u32 adj_hash_index = 0;
2863 	int i;
2864 
2865 	if (!nh->offloaded || !nh_grp->adj_index_valid)
2866 		return -EINVAL;
2867 
2868 	*p_adj_index = nh_grp->adj_index;
2869 	*p_adj_size = nh_grp->ecmp_size;
2870 
2871 	for (i = 0; i < nh_grp->count; i++) {
2872 		struct mlxsw_sp_nexthop *nh_iter = &nh_grp->nexthops[i];
2873 
2874 		if (nh_iter == nh)
2875 			break;
2876 		if (nh_iter->offloaded)
2877 			adj_hash_index += nh_iter->num_adj_entries;
2878 	}
2879 
2880 	*p_adj_hash_index = adj_hash_index;
2881 	return 0;
2882 }
2883 
2884 struct mlxsw_sp_rif *mlxsw_sp_nexthop_rif(struct mlxsw_sp_nexthop *nh)
2885 {
2886 	return nh->rif;
2887 }
2888 
2889 bool mlxsw_sp_nexthop_group_has_ipip(struct mlxsw_sp_nexthop *nh)
2890 {
2891 	struct mlxsw_sp_nexthop_group *nh_grp = nh->nh_grp;
2892 	int i;
2893 
2894 	for (i = 0; i < nh_grp->count; i++) {
2895 		struct mlxsw_sp_nexthop *nh_iter = &nh_grp->nexthops[i];
2896 
2897 		if (nh_iter->type == MLXSW_SP_NEXTHOP_TYPE_IPIP)
2898 			return true;
2899 	}
2900 	return false;
2901 }
2902 
2903 static struct fib_info *
2904 mlxsw_sp_nexthop4_group_fi(const struct mlxsw_sp_nexthop_group *nh_grp)
2905 {
2906 	return nh_grp->priv;
2907 }
2908 
2909 struct mlxsw_sp_nexthop_group_cmp_arg {
2910 	enum mlxsw_sp_l3proto proto;
2911 	union {
2912 		struct fib_info *fi;
2913 		struct mlxsw_sp_fib6_entry *fib6_entry;
2914 	};
2915 };
2916 
2917 static bool
2918 mlxsw_sp_nexthop6_group_has_nexthop(const struct mlxsw_sp_nexthop_group *nh_grp,
2919 				    const struct in6_addr *gw, int ifindex,
2920 				    int weight)
2921 {
2922 	int i;
2923 
2924 	for (i = 0; i < nh_grp->count; i++) {
2925 		const struct mlxsw_sp_nexthop *nh;
2926 
2927 		nh = &nh_grp->nexthops[i];
2928 		if (nh->ifindex == ifindex && nh->nh_weight == weight &&
2929 		    ipv6_addr_equal(gw, (struct in6_addr *) nh->gw_addr))
2930 			return true;
2931 	}
2932 
2933 	return false;
2934 }
2935 
2936 static bool
2937 mlxsw_sp_nexthop6_group_cmp(const struct mlxsw_sp_nexthop_group *nh_grp,
2938 			    const struct mlxsw_sp_fib6_entry *fib6_entry)
2939 {
2940 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
2941 
2942 	if (nh_grp->count != fib6_entry->nrt6)
2943 		return false;
2944 
2945 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
2946 		struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
2947 		struct in6_addr *gw;
2948 		int ifindex, weight;
2949 
2950 		ifindex = fib6_nh->fib_nh_dev->ifindex;
2951 		weight = fib6_nh->fib_nh_weight;
2952 		gw = &fib6_nh->fib_nh_gw6;
2953 		if (!mlxsw_sp_nexthop6_group_has_nexthop(nh_grp, gw, ifindex,
2954 							 weight))
2955 			return false;
2956 	}
2957 
2958 	return true;
2959 }
2960 
2961 static int
2962 mlxsw_sp_nexthop_group_cmp(struct rhashtable_compare_arg *arg, const void *ptr)
2963 {
2964 	const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = arg->key;
2965 	const struct mlxsw_sp_nexthop_group *nh_grp = ptr;
2966 
2967 	switch (cmp_arg->proto) {
2968 	case MLXSW_SP_L3_PROTO_IPV4:
2969 		return cmp_arg->fi != mlxsw_sp_nexthop4_group_fi(nh_grp);
2970 	case MLXSW_SP_L3_PROTO_IPV6:
2971 		return !mlxsw_sp_nexthop6_group_cmp(nh_grp,
2972 						    cmp_arg->fib6_entry);
2973 	default:
2974 		WARN_ON(1);
2975 		return 1;
2976 	}
2977 }
2978 
2979 static int
2980 mlxsw_sp_nexthop_group_type(const struct mlxsw_sp_nexthop_group *nh_grp)
2981 {
2982 	return nh_grp->neigh_tbl->family;
2983 }
2984 
2985 static u32 mlxsw_sp_nexthop_group_hash_obj(const void *data, u32 len, u32 seed)
2986 {
2987 	const struct mlxsw_sp_nexthop_group *nh_grp = data;
2988 	const struct mlxsw_sp_nexthop *nh;
2989 	struct fib_info *fi;
2990 	unsigned int val;
2991 	int i;
2992 
2993 	switch (mlxsw_sp_nexthop_group_type(nh_grp)) {
2994 	case AF_INET:
2995 		fi = mlxsw_sp_nexthop4_group_fi(nh_grp);
2996 		return jhash(&fi, sizeof(fi), seed);
2997 	case AF_INET6:
2998 		val = nh_grp->count;
2999 		for (i = 0; i < nh_grp->count; i++) {
3000 			nh = &nh_grp->nexthops[i];
3001 			val ^= jhash(&nh->ifindex, sizeof(nh->ifindex), seed);
3002 		}
3003 		return jhash(&val, sizeof(val), seed);
3004 	default:
3005 		WARN_ON(1);
3006 		return 0;
3007 	}
3008 }
3009 
3010 static u32
3011 mlxsw_sp_nexthop6_group_hash(struct mlxsw_sp_fib6_entry *fib6_entry, u32 seed)
3012 {
3013 	unsigned int val = fib6_entry->nrt6;
3014 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3015 	struct net_device *dev;
3016 
3017 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3018 		dev = mlxsw_sp_rt6->rt->fib6_nh->fib_nh_dev;
3019 		val ^= jhash(&dev->ifindex, sizeof(dev->ifindex), seed);
3020 	}
3021 
3022 	return jhash(&val, sizeof(val), seed);
3023 }
3024 
3025 static u32
3026 mlxsw_sp_nexthop_group_hash(const void *data, u32 len, u32 seed)
3027 {
3028 	const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = data;
3029 
3030 	switch (cmp_arg->proto) {
3031 	case MLXSW_SP_L3_PROTO_IPV4:
3032 		return jhash(&cmp_arg->fi, sizeof(cmp_arg->fi), seed);
3033 	case MLXSW_SP_L3_PROTO_IPV6:
3034 		return mlxsw_sp_nexthop6_group_hash(cmp_arg->fib6_entry, seed);
3035 	default:
3036 		WARN_ON(1);
3037 		return 0;
3038 	}
3039 }
3040 
3041 static const struct rhashtable_params mlxsw_sp_nexthop_group_ht_params = {
3042 	.head_offset = offsetof(struct mlxsw_sp_nexthop_group, ht_node),
3043 	.hashfn	     = mlxsw_sp_nexthop_group_hash,
3044 	.obj_hashfn  = mlxsw_sp_nexthop_group_hash_obj,
3045 	.obj_cmpfn   = mlxsw_sp_nexthop_group_cmp,
3046 };
3047 
3048 static int mlxsw_sp_nexthop_group_insert(struct mlxsw_sp *mlxsw_sp,
3049 					 struct mlxsw_sp_nexthop_group *nh_grp)
3050 {
3051 	if (mlxsw_sp_nexthop_group_type(nh_grp) == AF_INET6 &&
3052 	    !nh_grp->gateway)
3053 		return 0;
3054 
3055 	return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_group_ht,
3056 				      &nh_grp->ht_node,
3057 				      mlxsw_sp_nexthop_group_ht_params);
3058 }
3059 
3060 static void mlxsw_sp_nexthop_group_remove(struct mlxsw_sp *mlxsw_sp,
3061 					  struct mlxsw_sp_nexthop_group *nh_grp)
3062 {
3063 	if (mlxsw_sp_nexthop_group_type(nh_grp) == AF_INET6 &&
3064 	    !nh_grp->gateway)
3065 		return;
3066 
3067 	rhashtable_remove_fast(&mlxsw_sp->router->nexthop_group_ht,
3068 			       &nh_grp->ht_node,
3069 			       mlxsw_sp_nexthop_group_ht_params);
3070 }
3071 
3072 static struct mlxsw_sp_nexthop_group *
3073 mlxsw_sp_nexthop4_group_lookup(struct mlxsw_sp *mlxsw_sp,
3074 			       struct fib_info *fi)
3075 {
3076 	struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
3077 
3078 	cmp_arg.proto = MLXSW_SP_L3_PROTO_IPV4;
3079 	cmp_arg.fi = fi;
3080 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
3081 				      &cmp_arg,
3082 				      mlxsw_sp_nexthop_group_ht_params);
3083 }
3084 
3085 static struct mlxsw_sp_nexthop_group *
3086 mlxsw_sp_nexthop6_group_lookup(struct mlxsw_sp *mlxsw_sp,
3087 			       struct mlxsw_sp_fib6_entry *fib6_entry)
3088 {
3089 	struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
3090 
3091 	cmp_arg.proto = MLXSW_SP_L3_PROTO_IPV6;
3092 	cmp_arg.fib6_entry = fib6_entry;
3093 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
3094 				      &cmp_arg,
3095 				      mlxsw_sp_nexthop_group_ht_params);
3096 }
3097 
3098 static const struct rhashtable_params mlxsw_sp_nexthop_ht_params = {
3099 	.key_offset = offsetof(struct mlxsw_sp_nexthop, key),
3100 	.head_offset = offsetof(struct mlxsw_sp_nexthop, ht_node),
3101 	.key_len = sizeof(struct mlxsw_sp_nexthop_key),
3102 };
3103 
3104 static int mlxsw_sp_nexthop_insert(struct mlxsw_sp *mlxsw_sp,
3105 				   struct mlxsw_sp_nexthop *nh)
3106 {
3107 	return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_ht,
3108 				      &nh->ht_node, mlxsw_sp_nexthop_ht_params);
3109 }
3110 
3111 static void mlxsw_sp_nexthop_remove(struct mlxsw_sp *mlxsw_sp,
3112 				    struct mlxsw_sp_nexthop *nh)
3113 {
3114 	rhashtable_remove_fast(&mlxsw_sp->router->nexthop_ht, &nh->ht_node,
3115 			       mlxsw_sp_nexthop_ht_params);
3116 }
3117 
3118 static struct mlxsw_sp_nexthop *
3119 mlxsw_sp_nexthop_lookup(struct mlxsw_sp *mlxsw_sp,
3120 			struct mlxsw_sp_nexthop_key key)
3121 {
3122 	return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_ht, &key,
3123 				      mlxsw_sp_nexthop_ht_params);
3124 }
3125 
3126 static int mlxsw_sp_adj_index_mass_update_vr(struct mlxsw_sp *mlxsw_sp,
3127 					     const struct mlxsw_sp_fib *fib,
3128 					     u32 adj_index, u16 ecmp_size,
3129 					     u32 new_adj_index,
3130 					     u16 new_ecmp_size)
3131 {
3132 	char raleu_pl[MLXSW_REG_RALEU_LEN];
3133 
3134 	mlxsw_reg_raleu_pack(raleu_pl,
3135 			     (enum mlxsw_reg_ralxx_protocol) fib->proto,
3136 			     fib->vr->id, adj_index, ecmp_size, new_adj_index,
3137 			     new_ecmp_size);
3138 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raleu), raleu_pl);
3139 }
3140 
3141 static int mlxsw_sp_adj_index_mass_update(struct mlxsw_sp *mlxsw_sp,
3142 					  struct mlxsw_sp_nexthop_group *nh_grp,
3143 					  u32 old_adj_index, u16 old_ecmp_size)
3144 {
3145 	struct mlxsw_sp_fib_entry *fib_entry;
3146 	struct mlxsw_sp_fib *fib = NULL;
3147 	int err;
3148 
3149 	list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
3150 		if (fib == fib_entry->fib_node->fib)
3151 			continue;
3152 		fib = fib_entry->fib_node->fib;
3153 		err = mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp, fib,
3154 							old_adj_index,
3155 							old_ecmp_size,
3156 							nh_grp->adj_index,
3157 							nh_grp->ecmp_size);
3158 		if (err)
3159 			return err;
3160 	}
3161 	return 0;
3162 }
3163 
3164 static int __mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
3165 				     struct mlxsw_sp_nexthop *nh)
3166 {
3167 	struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
3168 	char ratr_pl[MLXSW_REG_RATR_LEN];
3169 
3170 	mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
3171 			    true, MLXSW_REG_RATR_TYPE_ETHERNET,
3172 			    adj_index, neigh_entry->rif);
3173 	mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
3174 	if (nh->counter_valid)
3175 		mlxsw_reg_ratr_counter_pack(ratr_pl, nh->counter_index, true);
3176 	else
3177 		mlxsw_reg_ratr_counter_pack(ratr_pl, 0, false);
3178 
3179 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
3180 }
3181 
3182 int mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
3183 			    struct mlxsw_sp_nexthop *nh)
3184 {
3185 	int i;
3186 
3187 	for (i = 0; i < nh->num_adj_entries; i++) {
3188 		int err;
3189 
3190 		err = __mlxsw_sp_nexthop_update(mlxsw_sp, adj_index + i, nh);
3191 		if (err)
3192 			return err;
3193 	}
3194 
3195 	return 0;
3196 }
3197 
3198 static int __mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
3199 					  u32 adj_index,
3200 					  struct mlxsw_sp_nexthop *nh)
3201 {
3202 	const struct mlxsw_sp_ipip_ops *ipip_ops;
3203 
3204 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[nh->ipip_entry->ipipt];
3205 	return ipip_ops->nexthop_update(mlxsw_sp, adj_index, nh->ipip_entry);
3206 }
3207 
3208 static int mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
3209 					u32 adj_index,
3210 					struct mlxsw_sp_nexthop *nh)
3211 {
3212 	int i;
3213 
3214 	for (i = 0; i < nh->num_adj_entries; i++) {
3215 		int err;
3216 
3217 		err = __mlxsw_sp_nexthop_ipip_update(mlxsw_sp, adj_index + i,
3218 						     nh);
3219 		if (err)
3220 			return err;
3221 	}
3222 
3223 	return 0;
3224 }
3225 
3226 static int
3227 mlxsw_sp_nexthop_group_update(struct mlxsw_sp *mlxsw_sp,
3228 			      struct mlxsw_sp_nexthop_group *nh_grp,
3229 			      bool reallocate)
3230 {
3231 	u32 adj_index = nh_grp->adj_index; /* base */
3232 	struct mlxsw_sp_nexthop *nh;
3233 	int i;
3234 
3235 	for (i = 0; i < nh_grp->count; i++) {
3236 		nh = &nh_grp->nexthops[i];
3237 
3238 		if (!nh->should_offload) {
3239 			nh->offloaded = 0;
3240 			continue;
3241 		}
3242 
3243 		if (nh->update || reallocate) {
3244 			int err = 0;
3245 
3246 			switch (nh->type) {
3247 			case MLXSW_SP_NEXTHOP_TYPE_ETH:
3248 				err = mlxsw_sp_nexthop_update
3249 					    (mlxsw_sp, adj_index, nh);
3250 				break;
3251 			case MLXSW_SP_NEXTHOP_TYPE_IPIP:
3252 				err = mlxsw_sp_nexthop_ipip_update
3253 					    (mlxsw_sp, adj_index, nh);
3254 				break;
3255 			}
3256 			if (err)
3257 				return err;
3258 			nh->update = 0;
3259 			nh->offloaded = 1;
3260 		}
3261 		adj_index += nh->num_adj_entries;
3262 	}
3263 	return 0;
3264 }
3265 
3266 static int
3267 mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
3268 				    struct mlxsw_sp_nexthop_group *nh_grp)
3269 {
3270 	struct mlxsw_sp_fib_entry *fib_entry;
3271 	int err;
3272 
3273 	list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
3274 		err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
3275 		if (err)
3276 			return err;
3277 	}
3278 	return 0;
3279 }
3280 
3281 static void mlxsw_sp_adj_grp_size_round_up(u16 *p_adj_grp_size)
3282 {
3283 	/* Valid sizes for an adjacency group are:
3284 	 * 1-64, 512, 1024, 2048 and 4096.
3285 	 */
3286 	if (*p_adj_grp_size <= 64)
3287 		return;
3288 	else if (*p_adj_grp_size <= 512)
3289 		*p_adj_grp_size = 512;
3290 	else if (*p_adj_grp_size <= 1024)
3291 		*p_adj_grp_size = 1024;
3292 	else if (*p_adj_grp_size <= 2048)
3293 		*p_adj_grp_size = 2048;
3294 	else
3295 		*p_adj_grp_size = 4096;
3296 }
3297 
3298 static void mlxsw_sp_adj_grp_size_round_down(u16 *p_adj_grp_size,
3299 					     unsigned int alloc_size)
3300 {
3301 	if (alloc_size >= 4096)
3302 		*p_adj_grp_size = 4096;
3303 	else if (alloc_size >= 2048)
3304 		*p_adj_grp_size = 2048;
3305 	else if (alloc_size >= 1024)
3306 		*p_adj_grp_size = 1024;
3307 	else if (alloc_size >= 512)
3308 		*p_adj_grp_size = 512;
3309 }
3310 
3311 static int mlxsw_sp_fix_adj_grp_size(struct mlxsw_sp *mlxsw_sp,
3312 				     u16 *p_adj_grp_size)
3313 {
3314 	unsigned int alloc_size;
3315 	int err;
3316 
3317 	/* Round up the requested group size to the next size supported
3318 	 * by the device and make sure the request can be satisfied.
3319 	 */
3320 	mlxsw_sp_adj_grp_size_round_up(p_adj_grp_size);
3321 	err = mlxsw_sp_kvdl_alloc_count_query(mlxsw_sp,
3322 					      MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3323 					      *p_adj_grp_size, &alloc_size);
3324 	if (err)
3325 		return err;
3326 	/* It is possible the allocation results in more allocated
3327 	 * entries than requested. Try to use as much of them as
3328 	 * possible.
3329 	 */
3330 	mlxsw_sp_adj_grp_size_round_down(p_adj_grp_size, alloc_size);
3331 
3332 	return 0;
3333 }
3334 
3335 static void
3336 mlxsw_sp_nexthop_group_normalize(struct mlxsw_sp_nexthop_group *nh_grp)
3337 {
3338 	int i, g = 0, sum_norm_weight = 0;
3339 	struct mlxsw_sp_nexthop *nh;
3340 
3341 	for (i = 0; i < nh_grp->count; i++) {
3342 		nh = &nh_grp->nexthops[i];
3343 
3344 		if (!nh->should_offload)
3345 			continue;
3346 		if (g > 0)
3347 			g = gcd(nh->nh_weight, g);
3348 		else
3349 			g = nh->nh_weight;
3350 	}
3351 
3352 	for (i = 0; i < nh_grp->count; i++) {
3353 		nh = &nh_grp->nexthops[i];
3354 
3355 		if (!nh->should_offload)
3356 			continue;
3357 		nh->norm_nh_weight = nh->nh_weight / g;
3358 		sum_norm_weight += nh->norm_nh_weight;
3359 	}
3360 
3361 	nh_grp->sum_norm_weight = sum_norm_weight;
3362 }
3363 
3364 static void
3365 mlxsw_sp_nexthop_group_rebalance(struct mlxsw_sp_nexthop_group *nh_grp)
3366 {
3367 	int total = nh_grp->sum_norm_weight;
3368 	u16 ecmp_size = nh_grp->ecmp_size;
3369 	int i, weight = 0, lower_bound = 0;
3370 
3371 	for (i = 0; i < nh_grp->count; i++) {
3372 		struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
3373 		int upper_bound;
3374 
3375 		if (!nh->should_offload)
3376 			continue;
3377 		weight += nh->norm_nh_weight;
3378 		upper_bound = DIV_ROUND_CLOSEST(ecmp_size * weight, total);
3379 		nh->num_adj_entries = upper_bound - lower_bound;
3380 		lower_bound = upper_bound;
3381 	}
3382 }
3383 
3384 static struct mlxsw_sp_nexthop *
3385 mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
3386 		     const struct mlxsw_sp_rt6 *mlxsw_sp_rt6);
3387 
3388 static void
3389 mlxsw_sp_nexthop4_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3390 					struct mlxsw_sp_nexthop_group *nh_grp)
3391 {
3392 	int i;
3393 
3394 	for (i = 0; i < nh_grp->count; i++) {
3395 		struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
3396 
3397 		if (nh->offloaded)
3398 			nh->key.fib_nh->fib_nh_flags |= RTNH_F_OFFLOAD;
3399 		else
3400 			nh->key.fib_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
3401 	}
3402 }
3403 
3404 static void
3405 __mlxsw_sp_nexthop6_group_offload_refresh(struct mlxsw_sp_nexthop_group *nh_grp,
3406 					  struct mlxsw_sp_fib6_entry *fib6_entry)
3407 {
3408 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3409 
3410 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3411 		struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3412 		struct mlxsw_sp_nexthop *nh;
3413 
3414 		nh = mlxsw_sp_rt6_nexthop(nh_grp, mlxsw_sp_rt6);
3415 		if (nh && nh->offloaded)
3416 			fib6_nh->fib_nh_flags |= RTNH_F_OFFLOAD;
3417 		else
3418 			fib6_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
3419 	}
3420 }
3421 
3422 static void
3423 mlxsw_sp_nexthop6_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3424 					struct mlxsw_sp_nexthop_group *nh_grp)
3425 {
3426 	struct mlxsw_sp_fib6_entry *fib6_entry;
3427 
3428 	/* Unfortunately, in IPv6 the route and the nexthop are described by
3429 	 * the same struct, so we need to iterate over all the routes using the
3430 	 * nexthop group and set / clear the offload indication for them.
3431 	 */
3432 	list_for_each_entry(fib6_entry, &nh_grp->fib_list,
3433 			    common.nexthop_group_node)
3434 		__mlxsw_sp_nexthop6_group_offload_refresh(nh_grp, fib6_entry);
3435 }
3436 
3437 static void
3438 mlxsw_sp_nexthop_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3439 				       struct mlxsw_sp_nexthop_group *nh_grp)
3440 {
3441 	switch (mlxsw_sp_nexthop_group_type(nh_grp)) {
3442 	case AF_INET:
3443 		mlxsw_sp_nexthop4_group_offload_refresh(mlxsw_sp, nh_grp);
3444 		break;
3445 	case AF_INET6:
3446 		mlxsw_sp_nexthop6_group_offload_refresh(mlxsw_sp, nh_grp);
3447 		break;
3448 	}
3449 }
3450 
3451 static void
3452 mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
3453 			       struct mlxsw_sp_nexthop_group *nh_grp)
3454 {
3455 	u16 ecmp_size, old_ecmp_size;
3456 	struct mlxsw_sp_nexthop *nh;
3457 	bool offload_change = false;
3458 	u32 adj_index;
3459 	bool old_adj_index_valid;
3460 	u32 old_adj_index;
3461 	int i;
3462 	int err;
3463 
3464 	if (!nh_grp->gateway) {
3465 		mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3466 		return;
3467 	}
3468 
3469 	for (i = 0; i < nh_grp->count; i++) {
3470 		nh = &nh_grp->nexthops[i];
3471 
3472 		if (nh->should_offload != nh->offloaded) {
3473 			offload_change = true;
3474 			if (nh->should_offload)
3475 				nh->update = 1;
3476 		}
3477 	}
3478 	if (!offload_change) {
3479 		/* Nothing was added or removed, so no need to reallocate. Just
3480 		 * update MAC on existing adjacency indexes.
3481 		 */
3482 		err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nh_grp, false);
3483 		if (err) {
3484 			dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3485 			goto set_trap;
3486 		}
3487 		return;
3488 	}
3489 	mlxsw_sp_nexthop_group_normalize(nh_grp);
3490 	if (!nh_grp->sum_norm_weight)
3491 		/* No neigh of this group is connected so we just set
3492 		 * the trap and let everthing flow through kernel.
3493 		 */
3494 		goto set_trap;
3495 
3496 	ecmp_size = nh_grp->sum_norm_weight;
3497 	err = mlxsw_sp_fix_adj_grp_size(mlxsw_sp, &ecmp_size);
3498 	if (err)
3499 		/* No valid allocation size available. */
3500 		goto set_trap;
3501 
3502 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3503 				  ecmp_size, &adj_index);
3504 	if (err) {
3505 		/* We ran out of KVD linear space, just set the
3506 		 * trap and let everything flow through kernel.
3507 		 */
3508 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to allocate KVD linear area for nexthop group.\n");
3509 		goto set_trap;
3510 	}
3511 	old_adj_index_valid = nh_grp->adj_index_valid;
3512 	old_adj_index = nh_grp->adj_index;
3513 	old_ecmp_size = nh_grp->ecmp_size;
3514 	nh_grp->adj_index_valid = 1;
3515 	nh_grp->adj_index = adj_index;
3516 	nh_grp->ecmp_size = ecmp_size;
3517 	mlxsw_sp_nexthop_group_rebalance(nh_grp);
3518 	err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nh_grp, true);
3519 	if (err) {
3520 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3521 		goto set_trap;
3522 	}
3523 
3524 	mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
3525 
3526 	if (!old_adj_index_valid) {
3527 		/* The trap was set for fib entries, so we have to call
3528 		 * fib entry update to unset it and use adjacency index.
3529 		 */
3530 		err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3531 		if (err) {
3532 			dev_warn(mlxsw_sp->bus_info->dev, "Failed to add adjacency index to fib entries.\n");
3533 			goto set_trap;
3534 		}
3535 		return;
3536 	}
3537 
3538 	err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, nh_grp,
3539 					     old_adj_index, old_ecmp_size);
3540 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3541 			   old_ecmp_size, old_adj_index);
3542 	if (err) {
3543 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to mass-update adjacency index for nexthop group.\n");
3544 		goto set_trap;
3545 	}
3546 
3547 	return;
3548 
3549 set_trap:
3550 	old_adj_index_valid = nh_grp->adj_index_valid;
3551 	nh_grp->adj_index_valid = 0;
3552 	for (i = 0; i < nh_grp->count; i++) {
3553 		nh = &nh_grp->nexthops[i];
3554 		nh->offloaded = 0;
3555 	}
3556 	err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3557 	if (err)
3558 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to set traps for fib entries.\n");
3559 	mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
3560 	if (old_adj_index_valid)
3561 		mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3562 				   nh_grp->ecmp_size, nh_grp->adj_index);
3563 }
3564 
3565 static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
3566 					    bool removing)
3567 {
3568 	if (!removing)
3569 		nh->should_offload = 1;
3570 	else
3571 		nh->should_offload = 0;
3572 	nh->update = 1;
3573 }
3574 
3575 static int
3576 mlxsw_sp_nexthop_dead_neigh_replace(struct mlxsw_sp *mlxsw_sp,
3577 				    struct mlxsw_sp_neigh_entry *neigh_entry)
3578 {
3579 	struct neighbour *n, *old_n = neigh_entry->key.n;
3580 	struct mlxsw_sp_nexthop *nh;
3581 	bool entry_connected;
3582 	u8 nud_state, dead;
3583 	int err;
3584 
3585 	nh = list_first_entry(&neigh_entry->nexthop_list,
3586 			      struct mlxsw_sp_nexthop, neigh_list_node);
3587 
3588 	n = neigh_lookup(nh->nh_grp->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3589 	if (!n) {
3590 		n = neigh_create(nh->nh_grp->neigh_tbl, &nh->gw_addr,
3591 				 nh->rif->dev);
3592 		if (IS_ERR(n))
3593 			return PTR_ERR(n);
3594 		neigh_event_send(n, NULL);
3595 	}
3596 
3597 	mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
3598 	neigh_entry->key.n = n;
3599 	err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
3600 	if (err)
3601 		goto err_neigh_entry_insert;
3602 
3603 	read_lock_bh(&n->lock);
3604 	nud_state = n->nud_state;
3605 	dead = n->dead;
3606 	read_unlock_bh(&n->lock);
3607 	entry_connected = nud_state & NUD_VALID && !dead;
3608 
3609 	list_for_each_entry(nh, &neigh_entry->nexthop_list,
3610 			    neigh_list_node) {
3611 		neigh_release(old_n);
3612 		neigh_clone(n);
3613 		__mlxsw_sp_nexthop_neigh_update(nh, !entry_connected);
3614 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
3615 	}
3616 
3617 	neigh_release(n);
3618 
3619 	return 0;
3620 
3621 err_neigh_entry_insert:
3622 	neigh_entry->key.n = old_n;
3623 	mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
3624 	neigh_release(n);
3625 	return err;
3626 }
3627 
3628 static void
3629 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
3630 			      struct mlxsw_sp_neigh_entry *neigh_entry,
3631 			      bool removing, bool dead)
3632 {
3633 	struct mlxsw_sp_nexthop *nh;
3634 
3635 	if (list_empty(&neigh_entry->nexthop_list))
3636 		return;
3637 
3638 	if (dead) {
3639 		int err;
3640 
3641 		err = mlxsw_sp_nexthop_dead_neigh_replace(mlxsw_sp,
3642 							  neigh_entry);
3643 		if (err)
3644 			dev_err(mlxsw_sp->bus_info->dev, "Failed to replace dead neigh\n");
3645 		return;
3646 	}
3647 
3648 	list_for_each_entry(nh, &neigh_entry->nexthop_list,
3649 			    neigh_list_node) {
3650 		__mlxsw_sp_nexthop_neigh_update(nh, removing);
3651 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
3652 	}
3653 }
3654 
3655 static void mlxsw_sp_nexthop_rif_init(struct mlxsw_sp_nexthop *nh,
3656 				      struct mlxsw_sp_rif *rif)
3657 {
3658 	if (nh->rif)
3659 		return;
3660 
3661 	nh->rif = rif;
3662 	list_add(&nh->rif_list_node, &rif->nexthop_list);
3663 }
3664 
3665 static void mlxsw_sp_nexthop_rif_fini(struct mlxsw_sp_nexthop *nh)
3666 {
3667 	if (!nh->rif)
3668 		return;
3669 
3670 	list_del(&nh->rif_list_node);
3671 	nh->rif = NULL;
3672 }
3673 
3674 static int mlxsw_sp_nexthop_neigh_init(struct mlxsw_sp *mlxsw_sp,
3675 				       struct mlxsw_sp_nexthop *nh)
3676 {
3677 	struct mlxsw_sp_neigh_entry *neigh_entry;
3678 	struct neighbour *n;
3679 	u8 nud_state, dead;
3680 	int err;
3681 
3682 	if (!nh->nh_grp->gateway || nh->neigh_entry)
3683 		return 0;
3684 
3685 	/* Take a reference of neigh here ensuring that neigh would
3686 	 * not be destructed before the nexthop entry is finished.
3687 	 * The reference is taken either in neigh_lookup() or
3688 	 * in neigh_create() in case n is not found.
3689 	 */
3690 	n = neigh_lookup(nh->nh_grp->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3691 	if (!n) {
3692 		n = neigh_create(nh->nh_grp->neigh_tbl, &nh->gw_addr,
3693 				 nh->rif->dev);
3694 		if (IS_ERR(n))
3695 			return PTR_ERR(n);
3696 		neigh_event_send(n, NULL);
3697 	}
3698 	neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
3699 	if (!neigh_entry) {
3700 		neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
3701 		if (IS_ERR(neigh_entry)) {
3702 			err = -EINVAL;
3703 			goto err_neigh_entry_create;
3704 		}
3705 	}
3706 
3707 	/* If that is the first nexthop connected to that neigh, add to
3708 	 * nexthop_neighs_list
3709 	 */
3710 	if (list_empty(&neigh_entry->nexthop_list))
3711 		list_add_tail(&neigh_entry->nexthop_neighs_list_node,
3712 			      &mlxsw_sp->router->nexthop_neighs_list);
3713 
3714 	nh->neigh_entry = neigh_entry;
3715 	list_add_tail(&nh->neigh_list_node, &neigh_entry->nexthop_list);
3716 	read_lock_bh(&n->lock);
3717 	nud_state = n->nud_state;
3718 	dead = n->dead;
3719 	read_unlock_bh(&n->lock);
3720 	__mlxsw_sp_nexthop_neigh_update(nh, !(nud_state & NUD_VALID && !dead));
3721 
3722 	return 0;
3723 
3724 err_neigh_entry_create:
3725 	neigh_release(n);
3726 	return err;
3727 }
3728 
3729 static void mlxsw_sp_nexthop_neigh_fini(struct mlxsw_sp *mlxsw_sp,
3730 					struct mlxsw_sp_nexthop *nh)
3731 {
3732 	struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
3733 	struct neighbour *n;
3734 
3735 	if (!neigh_entry)
3736 		return;
3737 	n = neigh_entry->key.n;
3738 
3739 	__mlxsw_sp_nexthop_neigh_update(nh, true);
3740 	list_del(&nh->neigh_list_node);
3741 	nh->neigh_entry = NULL;
3742 
3743 	/* If that is the last nexthop connected to that neigh, remove from
3744 	 * nexthop_neighs_list
3745 	 */
3746 	if (list_empty(&neigh_entry->nexthop_list))
3747 		list_del(&neigh_entry->nexthop_neighs_list_node);
3748 
3749 	if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
3750 		mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
3751 
3752 	neigh_release(n);
3753 }
3754 
3755 static bool mlxsw_sp_ipip_netdev_ul_up(struct net_device *ol_dev)
3756 {
3757 	struct net_device *ul_dev;
3758 	bool is_up;
3759 
3760 	rcu_read_lock();
3761 	ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
3762 	is_up = ul_dev ? (ul_dev->flags & IFF_UP) : true;
3763 	rcu_read_unlock();
3764 
3765 	return is_up;
3766 }
3767 
3768 static void mlxsw_sp_nexthop_ipip_init(struct mlxsw_sp *mlxsw_sp,
3769 				       struct mlxsw_sp_nexthop *nh,
3770 				       struct mlxsw_sp_ipip_entry *ipip_entry)
3771 {
3772 	bool removing;
3773 
3774 	if (!nh->nh_grp->gateway || nh->ipip_entry)
3775 		return;
3776 
3777 	nh->ipip_entry = ipip_entry;
3778 	removing = !mlxsw_sp_ipip_netdev_ul_up(ipip_entry->ol_dev);
3779 	__mlxsw_sp_nexthop_neigh_update(nh, removing);
3780 	mlxsw_sp_nexthop_rif_init(nh, &ipip_entry->ol_lb->common);
3781 }
3782 
3783 static void mlxsw_sp_nexthop_ipip_fini(struct mlxsw_sp *mlxsw_sp,
3784 				       struct mlxsw_sp_nexthop *nh)
3785 {
3786 	struct mlxsw_sp_ipip_entry *ipip_entry = nh->ipip_entry;
3787 
3788 	if (!ipip_entry)
3789 		return;
3790 
3791 	__mlxsw_sp_nexthop_neigh_update(nh, true);
3792 	nh->ipip_entry = NULL;
3793 }
3794 
3795 static bool mlxsw_sp_nexthop4_ipip_type(const struct mlxsw_sp *mlxsw_sp,
3796 					const struct fib_nh *fib_nh,
3797 					enum mlxsw_sp_ipip_type *p_ipipt)
3798 {
3799 	struct net_device *dev = fib_nh->fib_nh_dev;
3800 
3801 	return dev &&
3802 	       fib_nh->nh_parent->fib_type == RTN_UNICAST &&
3803 	       mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, p_ipipt);
3804 }
3805 
3806 static void mlxsw_sp_nexthop_type_fini(struct mlxsw_sp *mlxsw_sp,
3807 				       struct mlxsw_sp_nexthop *nh)
3808 {
3809 	switch (nh->type) {
3810 	case MLXSW_SP_NEXTHOP_TYPE_ETH:
3811 		mlxsw_sp_nexthop_neigh_fini(mlxsw_sp, nh);
3812 		mlxsw_sp_nexthop_rif_fini(nh);
3813 		break;
3814 	case MLXSW_SP_NEXTHOP_TYPE_IPIP:
3815 		mlxsw_sp_nexthop_rif_fini(nh);
3816 		mlxsw_sp_nexthop_ipip_fini(mlxsw_sp, nh);
3817 		break;
3818 	}
3819 }
3820 
3821 static int mlxsw_sp_nexthop4_type_init(struct mlxsw_sp *mlxsw_sp,
3822 				       struct mlxsw_sp_nexthop *nh,
3823 				       struct fib_nh *fib_nh)
3824 {
3825 	const struct mlxsw_sp_ipip_ops *ipip_ops;
3826 	struct net_device *dev = fib_nh->fib_nh_dev;
3827 	struct mlxsw_sp_ipip_entry *ipip_entry;
3828 	struct mlxsw_sp_rif *rif;
3829 	int err;
3830 
3831 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, dev);
3832 	if (ipip_entry) {
3833 		ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
3834 		if (ipip_ops->can_offload(mlxsw_sp, dev,
3835 					  MLXSW_SP_L3_PROTO_IPV4)) {
3836 			nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
3837 			mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, ipip_entry);
3838 			return 0;
3839 		}
3840 	}
3841 
3842 	nh->type = MLXSW_SP_NEXTHOP_TYPE_ETH;
3843 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
3844 	if (!rif)
3845 		return 0;
3846 
3847 	mlxsw_sp_nexthop_rif_init(nh, rif);
3848 	err = mlxsw_sp_nexthop_neigh_init(mlxsw_sp, nh);
3849 	if (err)
3850 		goto err_neigh_init;
3851 
3852 	return 0;
3853 
3854 err_neigh_init:
3855 	mlxsw_sp_nexthop_rif_fini(nh);
3856 	return err;
3857 }
3858 
3859 static void mlxsw_sp_nexthop4_type_fini(struct mlxsw_sp *mlxsw_sp,
3860 					struct mlxsw_sp_nexthop *nh)
3861 {
3862 	mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
3863 }
3864 
3865 static int mlxsw_sp_nexthop4_init(struct mlxsw_sp *mlxsw_sp,
3866 				  struct mlxsw_sp_nexthop_group *nh_grp,
3867 				  struct mlxsw_sp_nexthop *nh,
3868 				  struct fib_nh *fib_nh)
3869 {
3870 	struct net_device *dev = fib_nh->fib_nh_dev;
3871 	struct in_device *in_dev;
3872 	int err;
3873 
3874 	nh->nh_grp = nh_grp;
3875 	nh->key.fib_nh = fib_nh;
3876 #ifdef CONFIG_IP_ROUTE_MULTIPATH
3877 	nh->nh_weight = fib_nh->fib_nh_weight;
3878 #else
3879 	nh->nh_weight = 1;
3880 #endif
3881 	memcpy(&nh->gw_addr, &fib_nh->fib_nh_gw4, sizeof(fib_nh->fib_nh_gw4));
3882 	err = mlxsw_sp_nexthop_insert(mlxsw_sp, nh);
3883 	if (err)
3884 		return err;
3885 
3886 	mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
3887 	list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
3888 
3889 	if (!dev)
3890 		return 0;
3891 
3892 	rcu_read_lock();
3893 	in_dev = __in_dev_get_rcu(dev);
3894 	if (in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
3895 	    fib_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
3896 		rcu_read_unlock();
3897 		return 0;
3898 	}
3899 	rcu_read_unlock();
3900 
3901 	err = mlxsw_sp_nexthop4_type_init(mlxsw_sp, nh, fib_nh);
3902 	if (err)
3903 		goto err_nexthop_neigh_init;
3904 
3905 	return 0;
3906 
3907 err_nexthop_neigh_init:
3908 	mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
3909 	return err;
3910 }
3911 
3912 static void mlxsw_sp_nexthop4_fini(struct mlxsw_sp *mlxsw_sp,
3913 				   struct mlxsw_sp_nexthop *nh)
3914 {
3915 	mlxsw_sp_nexthop4_type_fini(mlxsw_sp, nh);
3916 	list_del(&nh->router_list_node);
3917 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
3918 	mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
3919 }
3920 
3921 static void mlxsw_sp_nexthop4_event(struct mlxsw_sp *mlxsw_sp,
3922 				    unsigned long event, struct fib_nh *fib_nh)
3923 {
3924 	struct mlxsw_sp_nexthop_key key;
3925 	struct mlxsw_sp_nexthop *nh;
3926 
3927 	if (mlxsw_sp->router->aborted)
3928 		return;
3929 
3930 	key.fib_nh = fib_nh;
3931 	nh = mlxsw_sp_nexthop_lookup(mlxsw_sp, key);
3932 	if (!nh)
3933 		return;
3934 
3935 	switch (event) {
3936 	case FIB_EVENT_NH_ADD:
3937 		mlxsw_sp_nexthop4_type_init(mlxsw_sp, nh, fib_nh);
3938 		break;
3939 	case FIB_EVENT_NH_DEL:
3940 		mlxsw_sp_nexthop4_type_fini(mlxsw_sp, nh);
3941 		break;
3942 	}
3943 
3944 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
3945 }
3946 
3947 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
3948 					struct mlxsw_sp_rif *rif)
3949 {
3950 	struct mlxsw_sp_nexthop *nh;
3951 	bool removing;
3952 
3953 	list_for_each_entry(nh, &rif->nexthop_list, rif_list_node) {
3954 		switch (nh->type) {
3955 		case MLXSW_SP_NEXTHOP_TYPE_ETH:
3956 			removing = false;
3957 			break;
3958 		case MLXSW_SP_NEXTHOP_TYPE_IPIP:
3959 			removing = !mlxsw_sp_ipip_netdev_ul_up(rif->dev);
3960 			break;
3961 		default:
3962 			WARN_ON(1);
3963 			continue;
3964 		}
3965 
3966 		__mlxsw_sp_nexthop_neigh_update(nh, removing);
3967 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
3968 	}
3969 }
3970 
3971 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
3972 					 struct mlxsw_sp_rif *old_rif,
3973 					 struct mlxsw_sp_rif *new_rif)
3974 {
3975 	struct mlxsw_sp_nexthop *nh;
3976 
3977 	list_splice_init(&old_rif->nexthop_list, &new_rif->nexthop_list);
3978 	list_for_each_entry(nh, &new_rif->nexthop_list, rif_list_node)
3979 		nh->rif = new_rif;
3980 	mlxsw_sp_nexthop_rif_update(mlxsw_sp, new_rif);
3981 }
3982 
3983 static void mlxsw_sp_nexthop_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
3984 					   struct mlxsw_sp_rif *rif)
3985 {
3986 	struct mlxsw_sp_nexthop *nh, *tmp;
3987 
3988 	list_for_each_entry_safe(nh, tmp, &rif->nexthop_list, rif_list_node) {
3989 		mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
3990 		mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nh_grp);
3991 	}
3992 }
3993 
3994 static bool mlxsw_sp_fi_is_gateway(const struct mlxsw_sp *mlxsw_sp,
3995 				   struct fib_info *fi)
3996 {
3997 	const struct fib_nh *nh = fib_info_nh(fi, 0);
3998 
3999 	return nh->fib_nh_scope == RT_SCOPE_LINK ||
4000 	       mlxsw_sp_nexthop4_ipip_type(mlxsw_sp, nh, NULL);
4001 }
4002 
4003 static struct mlxsw_sp_nexthop_group *
4004 mlxsw_sp_nexthop4_group_create(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
4005 {
4006 	unsigned int nhs = fib_info_num_path(fi);
4007 	struct mlxsw_sp_nexthop_group *nh_grp;
4008 	struct mlxsw_sp_nexthop *nh;
4009 	struct fib_nh *fib_nh;
4010 	int i;
4011 	int err;
4012 
4013 	nh_grp = kzalloc(struct_size(nh_grp, nexthops, nhs), GFP_KERNEL);
4014 	if (!nh_grp)
4015 		return ERR_PTR(-ENOMEM);
4016 	nh_grp->priv = fi;
4017 	INIT_LIST_HEAD(&nh_grp->fib_list);
4018 	nh_grp->neigh_tbl = &arp_tbl;
4019 
4020 	nh_grp->gateway = mlxsw_sp_fi_is_gateway(mlxsw_sp, fi);
4021 	nh_grp->count = nhs;
4022 	fib_info_hold(fi);
4023 	for (i = 0; i < nh_grp->count; i++) {
4024 		nh = &nh_grp->nexthops[i];
4025 		fib_nh = fib_info_nh(fi, i);
4026 		err = mlxsw_sp_nexthop4_init(mlxsw_sp, nh_grp, nh, fib_nh);
4027 		if (err)
4028 			goto err_nexthop4_init;
4029 	}
4030 	err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
4031 	if (err)
4032 		goto err_nexthop_group_insert;
4033 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4034 	return nh_grp;
4035 
4036 err_nexthop_group_insert:
4037 err_nexthop4_init:
4038 	for (i--; i >= 0; i--) {
4039 		nh = &nh_grp->nexthops[i];
4040 		mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
4041 	}
4042 	fib_info_put(fi);
4043 	kfree(nh_grp);
4044 	return ERR_PTR(err);
4045 }
4046 
4047 static void
4048 mlxsw_sp_nexthop4_group_destroy(struct mlxsw_sp *mlxsw_sp,
4049 				struct mlxsw_sp_nexthop_group *nh_grp)
4050 {
4051 	struct mlxsw_sp_nexthop *nh;
4052 	int i;
4053 
4054 	mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
4055 	for (i = 0; i < nh_grp->count; i++) {
4056 		nh = &nh_grp->nexthops[i];
4057 		mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
4058 	}
4059 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4060 	WARN_ON_ONCE(nh_grp->adj_index_valid);
4061 	fib_info_put(mlxsw_sp_nexthop4_group_fi(nh_grp));
4062 	kfree(nh_grp);
4063 }
4064 
4065 static int mlxsw_sp_nexthop4_group_get(struct mlxsw_sp *mlxsw_sp,
4066 				       struct mlxsw_sp_fib_entry *fib_entry,
4067 				       struct fib_info *fi)
4068 {
4069 	struct mlxsw_sp_nexthop_group *nh_grp;
4070 
4071 	nh_grp = mlxsw_sp_nexthop4_group_lookup(mlxsw_sp, fi);
4072 	if (!nh_grp) {
4073 		nh_grp = mlxsw_sp_nexthop4_group_create(mlxsw_sp, fi);
4074 		if (IS_ERR(nh_grp))
4075 			return PTR_ERR(nh_grp);
4076 	}
4077 	list_add_tail(&fib_entry->nexthop_group_node, &nh_grp->fib_list);
4078 	fib_entry->nh_group = nh_grp;
4079 	return 0;
4080 }
4081 
4082 static void mlxsw_sp_nexthop4_group_put(struct mlxsw_sp *mlxsw_sp,
4083 					struct mlxsw_sp_fib_entry *fib_entry)
4084 {
4085 	struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
4086 
4087 	list_del(&fib_entry->nexthop_group_node);
4088 	if (!list_empty(&nh_grp->fib_list))
4089 		return;
4090 	mlxsw_sp_nexthop4_group_destroy(mlxsw_sp, nh_grp);
4091 }
4092 
4093 static bool
4094 mlxsw_sp_fib4_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
4095 {
4096 	struct mlxsw_sp_fib4_entry *fib4_entry;
4097 
4098 	fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
4099 				  common);
4100 	return !fib4_entry->tos;
4101 }
4102 
4103 static bool
4104 mlxsw_sp_fib_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
4105 {
4106 	struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
4107 
4108 	switch (fib_entry->fib_node->fib->proto) {
4109 	case MLXSW_SP_L3_PROTO_IPV4:
4110 		if (!mlxsw_sp_fib4_entry_should_offload(fib_entry))
4111 			return false;
4112 		break;
4113 	case MLXSW_SP_L3_PROTO_IPV6:
4114 		break;
4115 	}
4116 
4117 	switch (fib_entry->type) {
4118 	case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
4119 		return !!nh_group->adj_index_valid;
4120 	case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
4121 		return !!nh_group->nh_rif;
4122 	case MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE:
4123 	case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
4124 	case MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP:
4125 		return true;
4126 	default:
4127 		return false;
4128 	}
4129 }
4130 
4131 static struct mlxsw_sp_nexthop *
4132 mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
4133 		     const struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
4134 {
4135 	int i;
4136 
4137 	for (i = 0; i < nh_grp->count; i++) {
4138 		struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
4139 		struct fib6_info *rt = mlxsw_sp_rt6->rt;
4140 
4141 		if (nh->rif && nh->rif->dev == rt->fib6_nh->fib_nh_dev &&
4142 		    ipv6_addr_equal((const struct in6_addr *) &nh->gw_addr,
4143 				    &rt->fib6_nh->fib_nh_gw6))
4144 			return nh;
4145 		continue;
4146 	}
4147 
4148 	return NULL;
4149 }
4150 
4151 static void
4152 mlxsw_sp_fib4_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
4153 				 struct mlxsw_sp_fib_entry *fib_entry)
4154 {
4155 	struct fib_info *fi = mlxsw_sp_nexthop4_group_fi(fib_entry->nh_group);
4156 	u32 *p_dst = (u32 *) fib_entry->fib_node->key.addr;
4157 	int dst_len = fib_entry->fib_node->key.prefix_len;
4158 	struct mlxsw_sp_fib4_entry *fib4_entry;
4159 	struct fib_rt_info fri;
4160 	bool should_offload;
4161 
4162 	should_offload = mlxsw_sp_fib_entry_should_offload(fib_entry);
4163 	fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
4164 				  common);
4165 	fri.fi = fi;
4166 	fri.tb_id = fib4_entry->tb_id;
4167 	fri.dst = cpu_to_be32(*p_dst);
4168 	fri.dst_len = dst_len;
4169 	fri.tos = fib4_entry->tos;
4170 	fri.type = fib4_entry->type;
4171 	fri.offload = should_offload;
4172 	fri.trap = !should_offload;
4173 	fib_alias_hw_flags_set(mlxsw_sp_net(mlxsw_sp), &fri);
4174 }
4175 
4176 static void
4177 mlxsw_sp_fib4_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
4178 				   struct mlxsw_sp_fib_entry *fib_entry)
4179 {
4180 	struct fib_info *fi = mlxsw_sp_nexthop4_group_fi(fib_entry->nh_group);
4181 	u32 *p_dst = (u32 *) fib_entry->fib_node->key.addr;
4182 	int dst_len = fib_entry->fib_node->key.prefix_len;
4183 	struct mlxsw_sp_fib4_entry *fib4_entry;
4184 	struct fib_rt_info fri;
4185 
4186 	fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
4187 				  common);
4188 	fri.fi = fi;
4189 	fri.tb_id = fib4_entry->tb_id;
4190 	fri.dst = cpu_to_be32(*p_dst);
4191 	fri.dst_len = dst_len;
4192 	fri.tos = fib4_entry->tos;
4193 	fri.type = fib4_entry->type;
4194 	fri.offload = false;
4195 	fri.trap = false;
4196 	fib_alias_hw_flags_set(mlxsw_sp_net(mlxsw_sp), &fri);
4197 }
4198 
4199 static void
4200 mlxsw_sp_fib6_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
4201 				 struct mlxsw_sp_fib_entry *fib_entry)
4202 {
4203 	struct mlxsw_sp_fib6_entry *fib6_entry;
4204 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4205 	bool should_offload;
4206 
4207 	should_offload = mlxsw_sp_fib_entry_should_offload(fib_entry);
4208 
4209 	/* In IPv6 a multipath route is represented using multiple routes, so
4210 	 * we need to set the flags on all of them.
4211 	 */
4212 	fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
4213 				  common);
4214 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list)
4215 		fib6_info_hw_flags_set(mlxsw_sp_rt6->rt, should_offload,
4216 				       !should_offload);
4217 }
4218 
4219 static void
4220 mlxsw_sp_fib6_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
4221 				   struct mlxsw_sp_fib_entry *fib_entry)
4222 {
4223 	struct mlxsw_sp_fib6_entry *fib6_entry;
4224 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4225 
4226 	fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
4227 				  common);
4228 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list)
4229 		fib6_info_hw_flags_set(mlxsw_sp_rt6->rt, false, false);
4230 }
4231 
4232 static void
4233 mlxsw_sp_fib_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
4234 				struct mlxsw_sp_fib_entry *fib_entry)
4235 {
4236 	switch (fib_entry->fib_node->fib->proto) {
4237 	case MLXSW_SP_L3_PROTO_IPV4:
4238 		mlxsw_sp_fib4_entry_hw_flags_set(mlxsw_sp, fib_entry);
4239 		break;
4240 	case MLXSW_SP_L3_PROTO_IPV6:
4241 		mlxsw_sp_fib6_entry_hw_flags_set(mlxsw_sp, fib_entry);
4242 		break;
4243 	}
4244 }
4245 
4246 static void
4247 mlxsw_sp_fib_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
4248 				  struct mlxsw_sp_fib_entry *fib_entry)
4249 {
4250 	switch (fib_entry->fib_node->fib->proto) {
4251 	case MLXSW_SP_L3_PROTO_IPV4:
4252 		mlxsw_sp_fib4_entry_hw_flags_clear(mlxsw_sp, fib_entry);
4253 		break;
4254 	case MLXSW_SP_L3_PROTO_IPV6:
4255 		mlxsw_sp_fib6_entry_hw_flags_clear(mlxsw_sp, fib_entry);
4256 		break;
4257 	}
4258 }
4259 
4260 static void
4261 mlxsw_sp_fib_entry_hw_flags_refresh(struct mlxsw_sp *mlxsw_sp,
4262 				    struct mlxsw_sp_fib_entry *fib_entry,
4263 				    enum mlxsw_reg_ralue_op op)
4264 {
4265 	switch (op) {
4266 	case MLXSW_REG_RALUE_OP_WRITE_WRITE:
4267 		mlxsw_sp_fib_entry_hw_flags_set(mlxsw_sp, fib_entry);
4268 		break;
4269 	case MLXSW_REG_RALUE_OP_WRITE_DELETE:
4270 		mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, fib_entry);
4271 		break;
4272 	default:
4273 		break;
4274 	}
4275 }
4276 
4277 static void
4278 mlxsw_sp_fib_entry_ralue_pack(char *ralue_pl,
4279 			      const struct mlxsw_sp_fib_entry *fib_entry,
4280 			      enum mlxsw_reg_ralue_op op)
4281 {
4282 	struct mlxsw_sp_fib *fib = fib_entry->fib_node->fib;
4283 	enum mlxsw_reg_ralxx_protocol proto;
4284 	u32 *p_dip;
4285 
4286 	proto = (enum mlxsw_reg_ralxx_protocol) fib->proto;
4287 
4288 	switch (fib->proto) {
4289 	case MLXSW_SP_L3_PROTO_IPV4:
4290 		p_dip = (u32 *) fib_entry->fib_node->key.addr;
4291 		mlxsw_reg_ralue_pack4(ralue_pl, proto, op, fib->vr->id,
4292 				      fib_entry->fib_node->key.prefix_len,
4293 				      *p_dip);
4294 		break;
4295 	case MLXSW_SP_L3_PROTO_IPV6:
4296 		mlxsw_reg_ralue_pack6(ralue_pl, proto, op, fib->vr->id,
4297 				      fib_entry->fib_node->key.prefix_len,
4298 				      fib_entry->fib_node->key.addr);
4299 		break;
4300 	}
4301 }
4302 
4303 static int mlxsw_sp_adj_discard_write(struct mlxsw_sp *mlxsw_sp, u16 rif_index)
4304 {
4305 	enum mlxsw_reg_ratr_trap_action trap_action;
4306 	char ratr_pl[MLXSW_REG_RATR_LEN];
4307 	int err;
4308 
4309 	if (mlxsw_sp->router->adj_discard_index_valid)
4310 		return 0;
4311 
4312 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
4313 				  &mlxsw_sp->router->adj_discard_index);
4314 	if (err)
4315 		return err;
4316 
4317 	trap_action = MLXSW_REG_RATR_TRAP_ACTION_DISCARD_ERRORS;
4318 	mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY, true,
4319 			    MLXSW_REG_RATR_TYPE_ETHERNET,
4320 			    mlxsw_sp->router->adj_discard_index, rif_index);
4321 	mlxsw_reg_ratr_trap_action_set(ratr_pl, trap_action);
4322 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
4323 	if (err)
4324 		goto err_ratr_write;
4325 
4326 	mlxsw_sp->router->adj_discard_index_valid = true;
4327 
4328 	return 0;
4329 
4330 err_ratr_write:
4331 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
4332 			   mlxsw_sp->router->adj_discard_index);
4333 	return err;
4334 }
4335 
4336 static int mlxsw_sp_fib_entry_op_remote(struct mlxsw_sp *mlxsw_sp,
4337 					struct mlxsw_sp_fib_entry *fib_entry,
4338 					enum mlxsw_reg_ralue_op op)
4339 {
4340 	struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
4341 	char ralue_pl[MLXSW_REG_RALUE_LEN];
4342 	enum mlxsw_reg_ralue_trap_action trap_action;
4343 	u16 trap_id = 0;
4344 	u32 adjacency_index = 0;
4345 	u16 ecmp_size = 0;
4346 	int err;
4347 
4348 	/* In case the nexthop group adjacency index is valid, use it
4349 	 * with provided ECMP size. Otherwise, setup trap and pass
4350 	 * traffic to kernel.
4351 	 */
4352 	if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
4353 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
4354 		adjacency_index = fib_entry->nh_group->adj_index;
4355 		ecmp_size = fib_entry->nh_group->ecmp_size;
4356 	} else if (!nh_group->adj_index_valid && nh_group->count &&
4357 		   nh_group->nh_rif) {
4358 		err = mlxsw_sp_adj_discard_write(mlxsw_sp,
4359 						 nh_group->nh_rif->rif_index);
4360 		if (err)
4361 			return err;
4362 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
4363 		adjacency_index = mlxsw_sp->router->adj_discard_index;
4364 		ecmp_size = 1;
4365 	} else {
4366 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
4367 		trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
4368 	}
4369 
4370 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
4371 	mlxsw_reg_ralue_act_remote_pack(ralue_pl, trap_action, trap_id,
4372 					adjacency_index, ecmp_size);
4373 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
4374 }
4375 
4376 static int mlxsw_sp_fib_entry_op_local(struct mlxsw_sp *mlxsw_sp,
4377 				       struct mlxsw_sp_fib_entry *fib_entry,
4378 				       enum mlxsw_reg_ralue_op op)
4379 {
4380 	struct mlxsw_sp_rif *rif = fib_entry->nh_group->nh_rif;
4381 	enum mlxsw_reg_ralue_trap_action trap_action;
4382 	char ralue_pl[MLXSW_REG_RALUE_LEN];
4383 	u16 trap_id = 0;
4384 	u16 rif_index = 0;
4385 
4386 	if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
4387 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
4388 		rif_index = rif->rif_index;
4389 	} else {
4390 		trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
4391 		trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
4392 	}
4393 
4394 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
4395 	mlxsw_reg_ralue_act_local_pack(ralue_pl, trap_action, trap_id,
4396 				       rif_index);
4397 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
4398 }
4399 
4400 static int mlxsw_sp_fib_entry_op_trap(struct mlxsw_sp *mlxsw_sp,
4401 				      struct mlxsw_sp_fib_entry *fib_entry,
4402 				      enum mlxsw_reg_ralue_op op)
4403 {
4404 	char ralue_pl[MLXSW_REG_RALUE_LEN];
4405 
4406 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
4407 	mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
4408 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
4409 }
4410 
4411 static int mlxsw_sp_fib_entry_op_blackhole(struct mlxsw_sp *mlxsw_sp,
4412 					   struct mlxsw_sp_fib_entry *fib_entry,
4413 					   enum mlxsw_reg_ralue_op op)
4414 {
4415 	enum mlxsw_reg_ralue_trap_action trap_action;
4416 	char ralue_pl[MLXSW_REG_RALUE_LEN];
4417 
4418 	trap_action = MLXSW_REG_RALUE_TRAP_ACTION_DISCARD_ERROR;
4419 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
4420 	mlxsw_reg_ralue_act_local_pack(ralue_pl, trap_action, 0, 0);
4421 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
4422 }
4423 
4424 static int
4425 mlxsw_sp_fib_entry_op_unreachable(struct mlxsw_sp *mlxsw_sp,
4426 				  struct mlxsw_sp_fib_entry *fib_entry,
4427 				  enum mlxsw_reg_ralue_op op)
4428 {
4429 	enum mlxsw_reg_ralue_trap_action trap_action;
4430 	char ralue_pl[MLXSW_REG_RALUE_LEN];
4431 	u16 trap_id;
4432 
4433 	trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
4434 	trap_id = MLXSW_TRAP_ID_RTR_INGRESS1;
4435 
4436 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
4437 	mlxsw_reg_ralue_act_local_pack(ralue_pl, trap_action, trap_id, 0);
4438 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
4439 }
4440 
4441 static int
4442 mlxsw_sp_fib_entry_op_ipip_decap(struct mlxsw_sp *mlxsw_sp,
4443 				 struct mlxsw_sp_fib_entry *fib_entry,
4444 				 enum mlxsw_reg_ralue_op op)
4445 {
4446 	struct mlxsw_sp_ipip_entry *ipip_entry = fib_entry->decap.ipip_entry;
4447 	const struct mlxsw_sp_ipip_ops *ipip_ops;
4448 
4449 	if (WARN_ON(!ipip_entry))
4450 		return -EINVAL;
4451 
4452 	ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
4453 	return ipip_ops->fib_entry_op(mlxsw_sp, ipip_entry, op,
4454 				      fib_entry->decap.tunnel_index);
4455 }
4456 
4457 static int mlxsw_sp_fib_entry_op_nve_decap(struct mlxsw_sp *mlxsw_sp,
4458 					   struct mlxsw_sp_fib_entry *fib_entry,
4459 					   enum mlxsw_reg_ralue_op op)
4460 {
4461 	char ralue_pl[MLXSW_REG_RALUE_LEN];
4462 
4463 	mlxsw_sp_fib_entry_ralue_pack(ralue_pl, fib_entry, op);
4464 	mlxsw_reg_ralue_act_ip2me_tun_pack(ralue_pl,
4465 					   fib_entry->decap.tunnel_index);
4466 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
4467 }
4468 
4469 static int __mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
4470 				   struct mlxsw_sp_fib_entry *fib_entry,
4471 				   enum mlxsw_reg_ralue_op op)
4472 {
4473 	switch (fib_entry->type) {
4474 	case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
4475 		return mlxsw_sp_fib_entry_op_remote(mlxsw_sp, fib_entry, op);
4476 	case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
4477 		return mlxsw_sp_fib_entry_op_local(mlxsw_sp, fib_entry, op);
4478 	case MLXSW_SP_FIB_ENTRY_TYPE_TRAP:
4479 		return mlxsw_sp_fib_entry_op_trap(mlxsw_sp, fib_entry, op);
4480 	case MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE:
4481 		return mlxsw_sp_fib_entry_op_blackhole(mlxsw_sp, fib_entry, op);
4482 	case MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE:
4483 		return mlxsw_sp_fib_entry_op_unreachable(mlxsw_sp, fib_entry,
4484 							 op);
4485 	case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
4486 		return mlxsw_sp_fib_entry_op_ipip_decap(mlxsw_sp,
4487 							fib_entry, op);
4488 	case MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP:
4489 		return mlxsw_sp_fib_entry_op_nve_decap(mlxsw_sp, fib_entry, op);
4490 	}
4491 	return -EINVAL;
4492 }
4493 
4494 static int mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
4495 				 struct mlxsw_sp_fib_entry *fib_entry,
4496 				 enum mlxsw_reg_ralue_op op)
4497 {
4498 	int err = __mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry, op);
4499 
4500 	if (err)
4501 		return err;
4502 
4503 	mlxsw_sp_fib_entry_hw_flags_refresh(mlxsw_sp, fib_entry, op);
4504 
4505 	return err;
4506 }
4507 
4508 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
4509 				     struct mlxsw_sp_fib_entry *fib_entry)
4510 {
4511 	return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
4512 				     MLXSW_REG_RALUE_OP_WRITE_WRITE);
4513 }
4514 
4515 static int mlxsw_sp_fib_entry_del(struct mlxsw_sp *mlxsw_sp,
4516 				  struct mlxsw_sp_fib_entry *fib_entry)
4517 {
4518 	return mlxsw_sp_fib_entry_op(mlxsw_sp, fib_entry,
4519 				     MLXSW_REG_RALUE_OP_WRITE_DELETE);
4520 }
4521 
4522 static int
4523 mlxsw_sp_fib4_entry_type_set(struct mlxsw_sp *mlxsw_sp,
4524 			     const struct fib_entry_notifier_info *fen_info,
4525 			     struct mlxsw_sp_fib_entry *fib_entry)
4526 {
4527 	struct net_device *dev = fib_info_nh(fen_info->fi, 0)->fib_nh_dev;
4528 	union mlxsw_sp_l3addr dip = { .addr4 = htonl(fen_info->dst) };
4529 	struct mlxsw_sp_router *router = mlxsw_sp->router;
4530 	u32 tb_id = mlxsw_sp_fix_tb_id(fen_info->tb_id);
4531 	struct mlxsw_sp_ipip_entry *ipip_entry;
4532 	struct fib_info *fi = fen_info->fi;
4533 
4534 	switch (fen_info->type) {
4535 	case RTN_LOCAL:
4536 		ipip_entry = mlxsw_sp_ipip_entry_find_by_decap(mlxsw_sp, dev,
4537 						 MLXSW_SP_L3_PROTO_IPV4, dip);
4538 		if (ipip_entry && ipip_entry->ol_dev->flags & IFF_UP) {
4539 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
4540 			return mlxsw_sp_fib_entry_decap_init(mlxsw_sp,
4541 							     fib_entry,
4542 							     ipip_entry);
4543 		}
4544 		if (mlxsw_sp_router_nve_is_decap(mlxsw_sp, tb_id,
4545 						 MLXSW_SP_L3_PROTO_IPV4,
4546 						 &dip)) {
4547 			u32 tunnel_index;
4548 
4549 			tunnel_index = router->nve_decap_config.tunnel_index;
4550 			fib_entry->decap.tunnel_index = tunnel_index;
4551 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
4552 			return 0;
4553 		}
4554 		/* fall through */
4555 	case RTN_BROADCAST:
4556 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
4557 		return 0;
4558 	case RTN_BLACKHOLE:
4559 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE;
4560 		return 0;
4561 	case RTN_UNREACHABLE: /* fall through */
4562 	case RTN_PROHIBIT:
4563 		/* Packets hitting these routes need to be trapped, but
4564 		 * can do so with a lower priority than packets directed
4565 		 * at the host, so use action type local instead of trap.
4566 		 */
4567 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE;
4568 		return 0;
4569 	case RTN_UNICAST:
4570 		if (mlxsw_sp_fi_is_gateway(mlxsw_sp, fi))
4571 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
4572 		else
4573 			fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
4574 		return 0;
4575 	default:
4576 		return -EINVAL;
4577 	}
4578 }
4579 
4580 static void
4581 mlxsw_sp_fib4_entry_type_unset(struct mlxsw_sp *mlxsw_sp,
4582 			       struct mlxsw_sp_fib_entry *fib_entry)
4583 {
4584 	switch (fib_entry->type) {
4585 	case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
4586 		mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, fib_entry);
4587 		break;
4588 	default:
4589 		break;
4590 	}
4591 }
4592 
4593 static struct mlxsw_sp_fib4_entry *
4594 mlxsw_sp_fib4_entry_create(struct mlxsw_sp *mlxsw_sp,
4595 			   struct mlxsw_sp_fib_node *fib_node,
4596 			   const struct fib_entry_notifier_info *fen_info)
4597 {
4598 	struct mlxsw_sp_fib4_entry *fib4_entry;
4599 	struct mlxsw_sp_fib_entry *fib_entry;
4600 	int err;
4601 
4602 	fib4_entry = kzalloc(sizeof(*fib4_entry), GFP_KERNEL);
4603 	if (!fib4_entry)
4604 		return ERR_PTR(-ENOMEM);
4605 	fib_entry = &fib4_entry->common;
4606 
4607 	err = mlxsw_sp_fib4_entry_type_set(mlxsw_sp, fen_info, fib_entry);
4608 	if (err)
4609 		goto err_fib4_entry_type_set;
4610 
4611 	err = mlxsw_sp_nexthop4_group_get(mlxsw_sp, fib_entry, fen_info->fi);
4612 	if (err)
4613 		goto err_nexthop4_group_get;
4614 
4615 	fib4_entry->prio = fen_info->fi->fib_priority;
4616 	fib4_entry->tb_id = fen_info->tb_id;
4617 	fib4_entry->type = fen_info->type;
4618 	fib4_entry->tos = fen_info->tos;
4619 
4620 	fib_entry->fib_node = fib_node;
4621 
4622 	return fib4_entry;
4623 
4624 err_nexthop4_group_get:
4625 	mlxsw_sp_fib4_entry_type_unset(mlxsw_sp, fib_entry);
4626 err_fib4_entry_type_set:
4627 	kfree(fib4_entry);
4628 	return ERR_PTR(err);
4629 }
4630 
4631 static void mlxsw_sp_fib4_entry_destroy(struct mlxsw_sp *mlxsw_sp,
4632 					struct mlxsw_sp_fib4_entry *fib4_entry)
4633 {
4634 	mlxsw_sp_nexthop4_group_put(mlxsw_sp, &fib4_entry->common);
4635 	mlxsw_sp_fib4_entry_type_unset(mlxsw_sp, &fib4_entry->common);
4636 	kfree(fib4_entry);
4637 }
4638 
4639 static struct mlxsw_sp_fib4_entry *
4640 mlxsw_sp_fib4_entry_lookup(struct mlxsw_sp *mlxsw_sp,
4641 			   const struct fib_entry_notifier_info *fen_info)
4642 {
4643 	struct mlxsw_sp_fib4_entry *fib4_entry;
4644 	struct mlxsw_sp_fib_node *fib_node;
4645 	struct mlxsw_sp_fib *fib;
4646 	struct mlxsw_sp_vr *vr;
4647 
4648 	vr = mlxsw_sp_vr_find(mlxsw_sp, fen_info->tb_id);
4649 	if (!vr)
4650 		return NULL;
4651 	fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV4);
4652 
4653 	fib_node = mlxsw_sp_fib_node_lookup(fib, &fen_info->dst,
4654 					    sizeof(fen_info->dst),
4655 					    fen_info->dst_len);
4656 	if (!fib_node)
4657 		return NULL;
4658 
4659 	fib4_entry = container_of(fib_node->fib_entry,
4660 				  struct mlxsw_sp_fib4_entry, common);
4661 	if (fib4_entry->tb_id == fen_info->tb_id &&
4662 	    fib4_entry->tos == fen_info->tos &&
4663 	    fib4_entry->type == fen_info->type &&
4664 	    mlxsw_sp_nexthop4_group_fi(fib4_entry->common.nh_group) ==
4665 	    fen_info->fi)
4666 		return fib4_entry;
4667 
4668 	return NULL;
4669 }
4670 
4671 static const struct rhashtable_params mlxsw_sp_fib_ht_params = {
4672 	.key_offset = offsetof(struct mlxsw_sp_fib_node, key),
4673 	.head_offset = offsetof(struct mlxsw_sp_fib_node, ht_node),
4674 	.key_len = sizeof(struct mlxsw_sp_fib_key),
4675 	.automatic_shrinking = true,
4676 };
4677 
4678 static int mlxsw_sp_fib_node_insert(struct mlxsw_sp_fib *fib,
4679 				    struct mlxsw_sp_fib_node *fib_node)
4680 {
4681 	return rhashtable_insert_fast(&fib->ht, &fib_node->ht_node,
4682 				      mlxsw_sp_fib_ht_params);
4683 }
4684 
4685 static void mlxsw_sp_fib_node_remove(struct mlxsw_sp_fib *fib,
4686 				     struct mlxsw_sp_fib_node *fib_node)
4687 {
4688 	rhashtable_remove_fast(&fib->ht, &fib_node->ht_node,
4689 			       mlxsw_sp_fib_ht_params);
4690 }
4691 
4692 static struct mlxsw_sp_fib_node *
4693 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
4694 			 size_t addr_len, unsigned char prefix_len)
4695 {
4696 	struct mlxsw_sp_fib_key key;
4697 
4698 	memset(&key, 0, sizeof(key));
4699 	memcpy(key.addr, addr, addr_len);
4700 	key.prefix_len = prefix_len;
4701 	return rhashtable_lookup_fast(&fib->ht, &key, mlxsw_sp_fib_ht_params);
4702 }
4703 
4704 static struct mlxsw_sp_fib_node *
4705 mlxsw_sp_fib_node_create(struct mlxsw_sp_fib *fib, const void *addr,
4706 			 size_t addr_len, unsigned char prefix_len)
4707 {
4708 	struct mlxsw_sp_fib_node *fib_node;
4709 
4710 	fib_node = kzalloc(sizeof(*fib_node), GFP_KERNEL);
4711 	if (!fib_node)
4712 		return NULL;
4713 
4714 	list_add(&fib_node->list, &fib->node_list);
4715 	memcpy(fib_node->key.addr, addr, addr_len);
4716 	fib_node->key.prefix_len = prefix_len;
4717 
4718 	return fib_node;
4719 }
4720 
4721 static void mlxsw_sp_fib_node_destroy(struct mlxsw_sp_fib_node *fib_node)
4722 {
4723 	list_del(&fib_node->list);
4724 	kfree(fib_node);
4725 }
4726 
4727 static int mlxsw_sp_fib_lpm_tree_link(struct mlxsw_sp *mlxsw_sp,
4728 				      struct mlxsw_sp_fib_node *fib_node)
4729 {
4730 	struct mlxsw_sp_prefix_usage req_prefix_usage;
4731 	struct mlxsw_sp_fib *fib = fib_node->fib;
4732 	struct mlxsw_sp_lpm_tree *lpm_tree;
4733 	int err;
4734 
4735 	lpm_tree = mlxsw_sp->router->lpm.proto_trees[fib->proto];
4736 	if (lpm_tree->prefix_ref_count[fib_node->key.prefix_len] != 0)
4737 		goto out;
4738 
4739 	mlxsw_sp_prefix_usage_cpy(&req_prefix_usage, &lpm_tree->prefix_usage);
4740 	mlxsw_sp_prefix_usage_set(&req_prefix_usage, fib_node->key.prefix_len);
4741 	lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
4742 					 fib->proto);
4743 	if (IS_ERR(lpm_tree))
4744 		return PTR_ERR(lpm_tree);
4745 
4746 	err = mlxsw_sp_vrs_lpm_tree_replace(mlxsw_sp, fib, lpm_tree);
4747 	if (err)
4748 		goto err_lpm_tree_replace;
4749 
4750 out:
4751 	lpm_tree->prefix_ref_count[fib_node->key.prefix_len]++;
4752 	return 0;
4753 
4754 err_lpm_tree_replace:
4755 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
4756 	return err;
4757 }
4758 
4759 static void mlxsw_sp_fib_lpm_tree_unlink(struct mlxsw_sp *mlxsw_sp,
4760 					 struct mlxsw_sp_fib_node *fib_node)
4761 {
4762 	struct mlxsw_sp_lpm_tree *lpm_tree = fib_node->fib->lpm_tree;
4763 	struct mlxsw_sp_prefix_usage req_prefix_usage;
4764 	struct mlxsw_sp_fib *fib = fib_node->fib;
4765 	int err;
4766 
4767 	if (--lpm_tree->prefix_ref_count[fib_node->key.prefix_len] != 0)
4768 		return;
4769 	/* Try to construct a new LPM tree from the current prefix usage
4770 	 * minus the unused one. If we fail, continue using the old one.
4771 	 */
4772 	mlxsw_sp_prefix_usage_cpy(&req_prefix_usage, &lpm_tree->prefix_usage);
4773 	mlxsw_sp_prefix_usage_clear(&req_prefix_usage,
4774 				    fib_node->key.prefix_len);
4775 	lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
4776 					 fib->proto);
4777 	if (IS_ERR(lpm_tree))
4778 		return;
4779 
4780 	err = mlxsw_sp_vrs_lpm_tree_replace(mlxsw_sp, fib, lpm_tree);
4781 	if (err)
4782 		goto err_lpm_tree_replace;
4783 
4784 	return;
4785 
4786 err_lpm_tree_replace:
4787 	mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
4788 }
4789 
4790 static int mlxsw_sp_fib_node_init(struct mlxsw_sp *mlxsw_sp,
4791 				  struct mlxsw_sp_fib_node *fib_node,
4792 				  struct mlxsw_sp_fib *fib)
4793 {
4794 	int err;
4795 
4796 	err = mlxsw_sp_fib_node_insert(fib, fib_node);
4797 	if (err)
4798 		return err;
4799 	fib_node->fib = fib;
4800 
4801 	err = mlxsw_sp_fib_lpm_tree_link(mlxsw_sp, fib_node);
4802 	if (err)
4803 		goto err_fib_lpm_tree_link;
4804 
4805 	return 0;
4806 
4807 err_fib_lpm_tree_link:
4808 	fib_node->fib = NULL;
4809 	mlxsw_sp_fib_node_remove(fib, fib_node);
4810 	return err;
4811 }
4812 
4813 static void mlxsw_sp_fib_node_fini(struct mlxsw_sp *mlxsw_sp,
4814 				   struct mlxsw_sp_fib_node *fib_node)
4815 {
4816 	struct mlxsw_sp_fib *fib = fib_node->fib;
4817 
4818 	mlxsw_sp_fib_lpm_tree_unlink(mlxsw_sp, fib_node);
4819 	fib_node->fib = NULL;
4820 	mlxsw_sp_fib_node_remove(fib, fib_node);
4821 }
4822 
4823 static struct mlxsw_sp_fib_node *
4824 mlxsw_sp_fib_node_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id, const void *addr,
4825 		      size_t addr_len, unsigned char prefix_len,
4826 		      enum mlxsw_sp_l3proto proto)
4827 {
4828 	struct mlxsw_sp_fib_node *fib_node;
4829 	struct mlxsw_sp_fib *fib;
4830 	struct mlxsw_sp_vr *vr;
4831 	int err;
4832 
4833 	vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id, NULL);
4834 	if (IS_ERR(vr))
4835 		return ERR_CAST(vr);
4836 	fib = mlxsw_sp_vr_fib(vr, proto);
4837 
4838 	fib_node = mlxsw_sp_fib_node_lookup(fib, addr, addr_len, prefix_len);
4839 	if (fib_node)
4840 		return fib_node;
4841 
4842 	fib_node = mlxsw_sp_fib_node_create(fib, addr, addr_len, prefix_len);
4843 	if (!fib_node) {
4844 		err = -ENOMEM;
4845 		goto err_fib_node_create;
4846 	}
4847 
4848 	err = mlxsw_sp_fib_node_init(mlxsw_sp, fib_node, fib);
4849 	if (err)
4850 		goto err_fib_node_init;
4851 
4852 	return fib_node;
4853 
4854 err_fib_node_init:
4855 	mlxsw_sp_fib_node_destroy(fib_node);
4856 err_fib_node_create:
4857 	mlxsw_sp_vr_put(mlxsw_sp, vr);
4858 	return ERR_PTR(err);
4859 }
4860 
4861 static void mlxsw_sp_fib_node_put(struct mlxsw_sp *mlxsw_sp,
4862 				  struct mlxsw_sp_fib_node *fib_node)
4863 {
4864 	struct mlxsw_sp_vr *vr = fib_node->fib->vr;
4865 
4866 	if (fib_node->fib_entry)
4867 		return;
4868 	mlxsw_sp_fib_node_fini(mlxsw_sp, fib_node);
4869 	mlxsw_sp_fib_node_destroy(fib_node);
4870 	mlxsw_sp_vr_put(mlxsw_sp, vr);
4871 }
4872 
4873 static int mlxsw_sp_fib_node_entry_link(struct mlxsw_sp *mlxsw_sp,
4874 					struct mlxsw_sp_fib_entry *fib_entry)
4875 {
4876 	struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
4877 	int err;
4878 
4879 	fib_node->fib_entry = fib_entry;
4880 
4881 	err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
4882 	if (err)
4883 		goto err_fib_entry_update;
4884 
4885 	return 0;
4886 
4887 err_fib_entry_update:
4888 	fib_node->fib_entry = NULL;
4889 	return err;
4890 }
4891 
4892 static void
4893 mlxsw_sp_fib_node_entry_unlink(struct mlxsw_sp *mlxsw_sp,
4894 			       struct mlxsw_sp_fib_entry *fib_entry)
4895 {
4896 	struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
4897 
4898 	mlxsw_sp_fib_entry_del(mlxsw_sp, fib_entry);
4899 	fib_node->fib_entry = NULL;
4900 }
4901 
4902 static bool mlxsw_sp_fib4_allow_replace(struct mlxsw_sp_fib4_entry *fib4_entry)
4903 {
4904 	struct mlxsw_sp_fib_node *fib_node = fib4_entry->common.fib_node;
4905 	struct mlxsw_sp_fib4_entry *fib4_replaced;
4906 
4907 	if (!fib_node->fib_entry)
4908 		return true;
4909 
4910 	fib4_replaced = container_of(fib_node->fib_entry,
4911 				     struct mlxsw_sp_fib4_entry, common);
4912 	if (fib4_entry->tb_id == RT_TABLE_MAIN &&
4913 	    fib4_replaced->tb_id == RT_TABLE_LOCAL)
4914 		return false;
4915 
4916 	return true;
4917 }
4918 
4919 static int
4920 mlxsw_sp_router_fib4_replace(struct mlxsw_sp *mlxsw_sp,
4921 			     const struct fib_entry_notifier_info *fen_info)
4922 {
4923 	struct mlxsw_sp_fib4_entry *fib4_entry, *fib4_replaced;
4924 	struct mlxsw_sp_fib_entry *replaced;
4925 	struct mlxsw_sp_fib_node *fib_node;
4926 	int err;
4927 
4928 	if (mlxsw_sp->router->aborted)
4929 		return 0;
4930 
4931 	fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, fen_info->tb_id,
4932 					 &fen_info->dst, sizeof(fen_info->dst),
4933 					 fen_info->dst_len,
4934 					 MLXSW_SP_L3_PROTO_IPV4);
4935 	if (IS_ERR(fib_node)) {
4936 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to get FIB node\n");
4937 		return PTR_ERR(fib_node);
4938 	}
4939 
4940 	fib4_entry = mlxsw_sp_fib4_entry_create(mlxsw_sp, fib_node, fen_info);
4941 	if (IS_ERR(fib4_entry)) {
4942 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to create FIB entry\n");
4943 		err = PTR_ERR(fib4_entry);
4944 		goto err_fib4_entry_create;
4945 	}
4946 
4947 	if (!mlxsw_sp_fib4_allow_replace(fib4_entry)) {
4948 		mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
4949 		mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
4950 		return 0;
4951 	}
4952 
4953 	replaced = fib_node->fib_entry;
4954 	err = mlxsw_sp_fib_node_entry_link(mlxsw_sp, &fib4_entry->common);
4955 	if (err) {
4956 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to link FIB entry to node\n");
4957 		goto err_fib_node_entry_link;
4958 	}
4959 
4960 	/* Nothing to replace */
4961 	if (!replaced)
4962 		return 0;
4963 
4964 	mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, replaced);
4965 	fib4_replaced = container_of(replaced, struct mlxsw_sp_fib4_entry,
4966 				     common);
4967 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_replaced);
4968 
4969 	return 0;
4970 
4971 err_fib_node_entry_link:
4972 	fib_node->fib_entry = replaced;
4973 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
4974 err_fib4_entry_create:
4975 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
4976 	return err;
4977 }
4978 
4979 static void mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
4980 				     struct fib_entry_notifier_info *fen_info)
4981 {
4982 	struct mlxsw_sp_fib4_entry *fib4_entry;
4983 	struct mlxsw_sp_fib_node *fib_node;
4984 
4985 	if (mlxsw_sp->router->aborted)
4986 		return;
4987 
4988 	fib4_entry = mlxsw_sp_fib4_entry_lookup(mlxsw_sp, fen_info);
4989 	if (!fib4_entry)
4990 		return;
4991 	fib_node = fib4_entry->common.fib_node;
4992 
4993 	mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, &fib4_entry->common);
4994 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
4995 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
4996 }
4997 
4998 static bool mlxsw_sp_fib6_rt_should_ignore(const struct fib6_info *rt)
4999 {
5000 	/* Packets with link-local destination IP arriving to the router
5001 	 * are trapped to the CPU, so no need to program specific routes
5002 	 * for them.
5003 	 */
5004 	if (ipv6_addr_type(&rt->fib6_dst.addr) & IPV6_ADDR_LINKLOCAL)
5005 		return true;
5006 
5007 	/* Multicast routes aren't supported, so ignore them. Neighbour
5008 	 * Discovery packets are specifically trapped.
5009 	 */
5010 	if (ipv6_addr_type(&rt->fib6_dst.addr) & IPV6_ADDR_MULTICAST)
5011 		return true;
5012 
5013 	/* Cloned routes are irrelevant in the forwarding path. */
5014 	if (rt->fib6_flags & RTF_CACHE)
5015 		return true;
5016 
5017 	return false;
5018 }
5019 
5020 static struct mlxsw_sp_rt6 *mlxsw_sp_rt6_create(struct fib6_info *rt)
5021 {
5022 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5023 
5024 	mlxsw_sp_rt6 = kzalloc(sizeof(*mlxsw_sp_rt6), GFP_KERNEL);
5025 	if (!mlxsw_sp_rt6)
5026 		return ERR_PTR(-ENOMEM);
5027 
5028 	/* In case of route replace, replaced route is deleted with
5029 	 * no notification. Take reference to prevent accessing freed
5030 	 * memory.
5031 	 */
5032 	mlxsw_sp_rt6->rt = rt;
5033 	fib6_info_hold(rt);
5034 
5035 	return mlxsw_sp_rt6;
5036 }
5037 
5038 #if IS_ENABLED(CONFIG_IPV6)
5039 static void mlxsw_sp_rt6_release(struct fib6_info *rt)
5040 {
5041 	fib6_info_release(rt);
5042 }
5043 #else
5044 static void mlxsw_sp_rt6_release(struct fib6_info *rt)
5045 {
5046 }
5047 #endif
5048 
5049 static void mlxsw_sp_rt6_destroy(struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
5050 {
5051 	struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
5052 
5053 	fib6_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
5054 	mlxsw_sp_rt6_release(mlxsw_sp_rt6->rt);
5055 	kfree(mlxsw_sp_rt6);
5056 }
5057 
5058 static struct fib6_info *
5059 mlxsw_sp_fib6_entry_rt(const struct mlxsw_sp_fib6_entry *fib6_entry)
5060 {
5061 	return list_first_entry(&fib6_entry->rt6_list, struct mlxsw_sp_rt6,
5062 				list)->rt;
5063 }
5064 
5065 static struct mlxsw_sp_rt6 *
5066 mlxsw_sp_fib6_entry_rt_find(const struct mlxsw_sp_fib6_entry *fib6_entry,
5067 			    const struct fib6_info *rt)
5068 {
5069 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5070 
5071 	list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
5072 		if (mlxsw_sp_rt6->rt == rt)
5073 			return mlxsw_sp_rt6;
5074 	}
5075 
5076 	return NULL;
5077 }
5078 
5079 static bool mlxsw_sp_nexthop6_ipip_type(const struct mlxsw_sp *mlxsw_sp,
5080 					const struct fib6_info *rt,
5081 					enum mlxsw_sp_ipip_type *ret)
5082 {
5083 	return rt->fib6_nh->fib_nh_dev &&
5084 	       mlxsw_sp_netdev_ipip_type(mlxsw_sp, rt->fib6_nh->fib_nh_dev, ret);
5085 }
5086 
5087 static int mlxsw_sp_nexthop6_type_init(struct mlxsw_sp *mlxsw_sp,
5088 				       struct mlxsw_sp_nexthop_group *nh_grp,
5089 				       struct mlxsw_sp_nexthop *nh,
5090 				       const struct fib6_info *rt)
5091 {
5092 	const struct mlxsw_sp_ipip_ops *ipip_ops;
5093 	struct mlxsw_sp_ipip_entry *ipip_entry;
5094 	struct net_device *dev = rt->fib6_nh->fib_nh_dev;
5095 	struct mlxsw_sp_rif *rif;
5096 	int err;
5097 
5098 	ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, dev);
5099 	if (ipip_entry) {
5100 		ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
5101 		if (ipip_ops->can_offload(mlxsw_sp, dev,
5102 					  MLXSW_SP_L3_PROTO_IPV6)) {
5103 			nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
5104 			mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, ipip_entry);
5105 			return 0;
5106 		}
5107 	}
5108 
5109 	nh->type = MLXSW_SP_NEXTHOP_TYPE_ETH;
5110 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
5111 	if (!rif)
5112 		return 0;
5113 	mlxsw_sp_nexthop_rif_init(nh, rif);
5114 
5115 	err = mlxsw_sp_nexthop_neigh_init(mlxsw_sp, nh);
5116 	if (err)
5117 		goto err_nexthop_neigh_init;
5118 
5119 	return 0;
5120 
5121 err_nexthop_neigh_init:
5122 	mlxsw_sp_nexthop_rif_fini(nh);
5123 	return err;
5124 }
5125 
5126 static void mlxsw_sp_nexthop6_type_fini(struct mlxsw_sp *mlxsw_sp,
5127 					struct mlxsw_sp_nexthop *nh)
5128 {
5129 	mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
5130 }
5131 
5132 static int mlxsw_sp_nexthop6_init(struct mlxsw_sp *mlxsw_sp,
5133 				  struct mlxsw_sp_nexthop_group *nh_grp,
5134 				  struct mlxsw_sp_nexthop *nh,
5135 				  const struct fib6_info *rt)
5136 {
5137 	struct net_device *dev = rt->fib6_nh->fib_nh_dev;
5138 
5139 	nh->nh_grp = nh_grp;
5140 	nh->nh_weight = rt->fib6_nh->fib_nh_weight;
5141 	memcpy(&nh->gw_addr, &rt->fib6_nh->fib_nh_gw6, sizeof(nh->gw_addr));
5142 	mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
5143 
5144 	list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
5145 
5146 	if (!dev)
5147 		return 0;
5148 	nh->ifindex = dev->ifindex;
5149 
5150 	return mlxsw_sp_nexthop6_type_init(mlxsw_sp, nh_grp, nh, rt);
5151 }
5152 
5153 static void mlxsw_sp_nexthop6_fini(struct mlxsw_sp *mlxsw_sp,
5154 				   struct mlxsw_sp_nexthop *nh)
5155 {
5156 	mlxsw_sp_nexthop6_type_fini(mlxsw_sp, nh);
5157 	list_del(&nh->router_list_node);
5158 	mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
5159 }
5160 
5161 static bool mlxsw_sp_rt6_is_gateway(const struct mlxsw_sp *mlxsw_sp,
5162 				    const struct fib6_info *rt)
5163 {
5164 	return rt->fib6_nh->fib_nh_gw_family ||
5165 	       mlxsw_sp_nexthop6_ipip_type(mlxsw_sp, rt, NULL);
5166 }
5167 
5168 static struct mlxsw_sp_nexthop_group *
5169 mlxsw_sp_nexthop6_group_create(struct mlxsw_sp *mlxsw_sp,
5170 			       struct mlxsw_sp_fib6_entry *fib6_entry)
5171 {
5172 	struct mlxsw_sp_nexthop_group *nh_grp;
5173 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5174 	struct mlxsw_sp_nexthop *nh;
5175 	int i = 0;
5176 	int err;
5177 
5178 	nh_grp = kzalloc(struct_size(nh_grp, nexthops, fib6_entry->nrt6),
5179 			 GFP_KERNEL);
5180 	if (!nh_grp)
5181 		return ERR_PTR(-ENOMEM);
5182 	INIT_LIST_HEAD(&nh_grp->fib_list);
5183 #if IS_ENABLED(CONFIG_IPV6)
5184 	nh_grp->neigh_tbl = &nd_tbl;
5185 #endif
5186 	mlxsw_sp_rt6 = list_first_entry(&fib6_entry->rt6_list,
5187 					struct mlxsw_sp_rt6, list);
5188 	nh_grp->gateway = mlxsw_sp_rt6_is_gateway(mlxsw_sp, mlxsw_sp_rt6->rt);
5189 	nh_grp->count = fib6_entry->nrt6;
5190 	for (i = 0; i < nh_grp->count; i++) {
5191 		struct fib6_info *rt = mlxsw_sp_rt6->rt;
5192 
5193 		nh = &nh_grp->nexthops[i];
5194 		err = mlxsw_sp_nexthop6_init(mlxsw_sp, nh_grp, nh, rt);
5195 		if (err)
5196 			goto err_nexthop6_init;
5197 		mlxsw_sp_rt6 = list_next_entry(mlxsw_sp_rt6, list);
5198 	}
5199 
5200 	err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
5201 	if (err)
5202 		goto err_nexthop_group_insert;
5203 
5204 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
5205 	return nh_grp;
5206 
5207 err_nexthop_group_insert:
5208 err_nexthop6_init:
5209 	for (i--; i >= 0; i--) {
5210 		nh = &nh_grp->nexthops[i];
5211 		mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
5212 	}
5213 	kfree(nh_grp);
5214 	return ERR_PTR(err);
5215 }
5216 
5217 static void
5218 mlxsw_sp_nexthop6_group_destroy(struct mlxsw_sp *mlxsw_sp,
5219 				struct mlxsw_sp_nexthop_group *nh_grp)
5220 {
5221 	struct mlxsw_sp_nexthop *nh;
5222 	int i = nh_grp->count;
5223 
5224 	mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
5225 	for (i--; i >= 0; i--) {
5226 		nh = &nh_grp->nexthops[i];
5227 		mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
5228 	}
5229 	mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
5230 	WARN_ON(nh_grp->adj_index_valid);
5231 	kfree(nh_grp);
5232 }
5233 
5234 static int mlxsw_sp_nexthop6_group_get(struct mlxsw_sp *mlxsw_sp,
5235 				       struct mlxsw_sp_fib6_entry *fib6_entry)
5236 {
5237 	struct mlxsw_sp_nexthop_group *nh_grp;
5238 
5239 	nh_grp = mlxsw_sp_nexthop6_group_lookup(mlxsw_sp, fib6_entry);
5240 	if (!nh_grp) {
5241 		nh_grp = mlxsw_sp_nexthop6_group_create(mlxsw_sp, fib6_entry);
5242 		if (IS_ERR(nh_grp))
5243 			return PTR_ERR(nh_grp);
5244 	}
5245 
5246 	list_add_tail(&fib6_entry->common.nexthop_group_node,
5247 		      &nh_grp->fib_list);
5248 	fib6_entry->common.nh_group = nh_grp;
5249 
5250 	/* The route and the nexthop are described by the same struct, so we
5251 	 * need to the update the nexthop offload indication for the new route.
5252 	 */
5253 	__mlxsw_sp_nexthop6_group_offload_refresh(nh_grp, fib6_entry);
5254 
5255 	return 0;
5256 }
5257 
5258 static void mlxsw_sp_nexthop6_group_put(struct mlxsw_sp *mlxsw_sp,
5259 					struct mlxsw_sp_fib_entry *fib_entry)
5260 {
5261 	struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
5262 
5263 	list_del(&fib_entry->nexthop_group_node);
5264 	if (!list_empty(&nh_grp->fib_list))
5265 		return;
5266 	mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, nh_grp);
5267 }
5268 
5269 static int
5270 mlxsw_sp_nexthop6_group_update(struct mlxsw_sp *mlxsw_sp,
5271 			       struct mlxsw_sp_fib6_entry *fib6_entry)
5272 {
5273 	struct mlxsw_sp_nexthop_group *old_nh_grp = fib6_entry->common.nh_group;
5274 	int err;
5275 
5276 	fib6_entry->common.nh_group = NULL;
5277 	list_del(&fib6_entry->common.nexthop_group_node);
5278 
5279 	err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
5280 	if (err)
5281 		goto err_nexthop6_group_get;
5282 
5283 	/* In case this entry is offloaded, then the adjacency index
5284 	 * currently associated with it in the device's table is that
5285 	 * of the old group. Start using the new one instead.
5286 	 */
5287 	err = mlxsw_sp_fib_entry_update(mlxsw_sp, &fib6_entry->common);
5288 	if (err)
5289 		goto err_fib_entry_update;
5290 
5291 	if (list_empty(&old_nh_grp->fib_list))
5292 		mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, old_nh_grp);
5293 
5294 	return 0;
5295 
5296 err_fib_entry_update:
5297 	mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
5298 err_nexthop6_group_get:
5299 	list_add_tail(&fib6_entry->common.nexthop_group_node,
5300 		      &old_nh_grp->fib_list);
5301 	fib6_entry->common.nh_group = old_nh_grp;
5302 	return err;
5303 }
5304 
5305 static int
5306 mlxsw_sp_fib6_entry_nexthop_add(struct mlxsw_sp *mlxsw_sp,
5307 				struct mlxsw_sp_fib6_entry *fib6_entry,
5308 				struct fib6_info **rt_arr, unsigned int nrt6)
5309 {
5310 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5311 	int err, i;
5312 
5313 	for (i = 0; i < nrt6; i++) {
5314 		mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt_arr[i]);
5315 		if (IS_ERR(mlxsw_sp_rt6)) {
5316 			err = PTR_ERR(mlxsw_sp_rt6);
5317 			goto err_rt6_create;
5318 		}
5319 
5320 		list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
5321 		fib6_entry->nrt6++;
5322 	}
5323 
5324 	err = mlxsw_sp_nexthop6_group_update(mlxsw_sp, fib6_entry);
5325 	if (err)
5326 		goto err_nexthop6_group_update;
5327 
5328 	return 0;
5329 
5330 err_nexthop6_group_update:
5331 	i = nrt6;
5332 err_rt6_create:
5333 	for (i--; i >= 0; i--) {
5334 		fib6_entry->nrt6--;
5335 		mlxsw_sp_rt6 = list_last_entry(&fib6_entry->rt6_list,
5336 					       struct mlxsw_sp_rt6, list);
5337 		list_del(&mlxsw_sp_rt6->list);
5338 		mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5339 	}
5340 	return err;
5341 }
5342 
5343 static void
5344 mlxsw_sp_fib6_entry_nexthop_del(struct mlxsw_sp *mlxsw_sp,
5345 				struct mlxsw_sp_fib6_entry *fib6_entry,
5346 				struct fib6_info **rt_arr, unsigned int nrt6)
5347 {
5348 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5349 	int i;
5350 
5351 	for (i = 0; i < nrt6; i++) {
5352 		mlxsw_sp_rt6 = mlxsw_sp_fib6_entry_rt_find(fib6_entry,
5353 							   rt_arr[i]);
5354 		if (WARN_ON_ONCE(!mlxsw_sp_rt6))
5355 			continue;
5356 
5357 		fib6_entry->nrt6--;
5358 		list_del(&mlxsw_sp_rt6->list);
5359 		mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5360 	}
5361 
5362 	mlxsw_sp_nexthop6_group_update(mlxsw_sp, fib6_entry);
5363 }
5364 
5365 static void mlxsw_sp_fib6_entry_type_set(struct mlxsw_sp *mlxsw_sp,
5366 					 struct mlxsw_sp_fib_entry *fib_entry,
5367 					 const struct fib6_info *rt)
5368 {
5369 	/* Packets hitting RTF_REJECT routes need to be discarded by the
5370 	 * stack. We can rely on their destination device not having a
5371 	 * RIF (it's the loopback device) and can thus use action type
5372 	 * local, which will cause them to be trapped with a lower
5373 	 * priority than packets that need to be locally received.
5374 	 */
5375 	if (rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
5376 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
5377 	else if (rt->fib6_type == RTN_BLACKHOLE)
5378 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE;
5379 	else if (rt->fib6_flags & RTF_REJECT)
5380 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE;
5381 	else if (mlxsw_sp_rt6_is_gateway(mlxsw_sp, rt))
5382 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
5383 	else
5384 		fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
5385 }
5386 
5387 static void
5388 mlxsw_sp_fib6_entry_rt_destroy_all(struct mlxsw_sp_fib6_entry *fib6_entry)
5389 {
5390 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6, *tmp;
5391 
5392 	list_for_each_entry_safe(mlxsw_sp_rt6, tmp, &fib6_entry->rt6_list,
5393 				 list) {
5394 		fib6_entry->nrt6--;
5395 		list_del(&mlxsw_sp_rt6->list);
5396 		mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5397 	}
5398 }
5399 
5400 static struct mlxsw_sp_fib6_entry *
5401 mlxsw_sp_fib6_entry_create(struct mlxsw_sp *mlxsw_sp,
5402 			   struct mlxsw_sp_fib_node *fib_node,
5403 			   struct fib6_info **rt_arr, unsigned int nrt6)
5404 {
5405 	struct mlxsw_sp_fib6_entry *fib6_entry;
5406 	struct mlxsw_sp_fib_entry *fib_entry;
5407 	struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5408 	int err, i;
5409 
5410 	fib6_entry = kzalloc(sizeof(*fib6_entry), GFP_KERNEL);
5411 	if (!fib6_entry)
5412 		return ERR_PTR(-ENOMEM);
5413 	fib_entry = &fib6_entry->common;
5414 
5415 	INIT_LIST_HEAD(&fib6_entry->rt6_list);
5416 
5417 	for (i = 0; i < nrt6; i++) {
5418 		mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt_arr[i]);
5419 		if (IS_ERR(mlxsw_sp_rt6)) {
5420 			err = PTR_ERR(mlxsw_sp_rt6);
5421 			goto err_rt6_create;
5422 		}
5423 		list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
5424 		fib6_entry->nrt6++;
5425 	}
5426 
5427 	mlxsw_sp_fib6_entry_type_set(mlxsw_sp, fib_entry, rt_arr[0]);
5428 
5429 	err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
5430 	if (err)
5431 		goto err_nexthop6_group_get;
5432 
5433 	fib_entry->fib_node = fib_node;
5434 
5435 	return fib6_entry;
5436 
5437 err_nexthop6_group_get:
5438 	i = nrt6;
5439 err_rt6_create:
5440 	for (i--; i >= 0; i--) {
5441 		fib6_entry->nrt6--;
5442 		mlxsw_sp_rt6 = list_last_entry(&fib6_entry->rt6_list,
5443 					       struct mlxsw_sp_rt6, list);
5444 		list_del(&mlxsw_sp_rt6->list);
5445 		mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5446 	}
5447 	kfree(fib6_entry);
5448 	return ERR_PTR(err);
5449 }
5450 
5451 static void mlxsw_sp_fib6_entry_destroy(struct mlxsw_sp *mlxsw_sp,
5452 					struct mlxsw_sp_fib6_entry *fib6_entry)
5453 {
5454 	mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
5455 	mlxsw_sp_fib6_entry_rt_destroy_all(fib6_entry);
5456 	WARN_ON(fib6_entry->nrt6);
5457 	kfree(fib6_entry);
5458 }
5459 
5460 static struct mlxsw_sp_fib6_entry *
5461 mlxsw_sp_fib6_entry_lookup(struct mlxsw_sp *mlxsw_sp,
5462 			   const struct fib6_info *rt)
5463 {
5464 	struct mlxsw_sp_fib6_entry *fib6_entry;
5465 	struct mlxsw_sp_fib_node *fib_node;
5466 	struct mlxsw_sp_fib *fib;
5467 	struct fib6_info *cmp_rt;
5468 	struct mlxsw_sp_vr *vr;
5469 
5470 	vr = mlxsw_sp_vr_find(mlxsw_sp, rt->fib6_table->tb6_id);
5471 	if (!vr)
5472 		return NULL;
5473 	fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV6);
5474 
5475 	fib_node = mlxsw_sp_fib_node_lookup(fib, &rt->fib6_dst.addr,
5476 					    sizeof(rt->fib6_dst.addr),
5477 					    rt->fib6_dst.plen);
5478 	if (!fib_node)
5479 		return NULL;
5480 
5481 	fib6_entry = container_of(fib_node->fib_entry,
5482 				  struct mlxsw_sp_fib6_entry, common);
5483 	cmp_rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
5484 	if (rt->fib6_table->tb6_id == cmp_rt->fib6_table->tb6_id &&
5485 	    rt->fib6_metric == cmp_rt->fib6_metric &&
5486 	    mlxsw_sp_fib6_entry_rt_find(fib6_entry, rt))
5487 		return fib6_entry;
5488 
5489 	return NULL;
5490 }
5491 
5492 static bool mlxsw_sp_fib6_allow_replace(struct mlxsw_sp_fib6_entry *fib6_entry)
5493 {
5494 	struct mlxsw_sp_fib_node *fib_node = fib6_entry->common.fib_node;
5495 	struct mlxsw_sp_fib6_entry *fib6_replaced;
5496 	struct fib6_info *rt, *rt_replaced;
5497 
5498 	if (!fib_node->fib_entry)
5499 		return true;
5500 
5501 	fib6_replaced = container_of(fib_node->fib_entry,
5502 				     struct mlxsw_sp_fib6_entry,
5503 				     common);
5504 	rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
5505 	rt_replaced = mlxsw_sp_fib6_entry_rt(fib6_replaced);
5506 	if (rt->fib6_table->tb6_id == RT_TABLE_MAIN &&
5507 	    rt_replaced->fib6_table->tb6_id == RT_TABLE_LOCAL)
5508 		return false;
5509 
5510 	return true;
5511 }
5512 
5513 static int mlxsw_sp_router_fib6_replace(struct mlxsw_sp *mlxsw_sp,
5514 					struct fib6_info **rt_arr,
5515 					unsigned int nrt6)
5516 {
5517 	struct mlxsw_sp_fib6_entry *fib6_entry, *fib6_replaced;
5518 	struct mlxsw_sp_fib_entry *replaced;
5519 	struct mlxsw_sp_fib_node *fib_node;
5520 	struct fib6_info *rt = rt_arr[0];
5521 	int err;
5522 
5523 	if (mlxsw_sp->router->aborted)
5524 		return 0;
5525 
5526 	if (rt->fib6_src.plen)
5527 		return -EINVAL;
5528 
5529 	if (mlxsw_sp_fib6_rt_should_ignore(rt))
5530 		return 0;
5531 
5532 	fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->fib6_table->tb6_id,
5533 					 &rt->fib6_dst.addr,
5534 					 sizeof(rt->fib6_dst.addr),
5535 					 rt->fib6_dst.plen,
5536 					 MLXSW_SP_L3_PROTO_IPV6);
5537 	if (IS_ERR(fib_node))
5538 		return PTR_ERR(fib_node);
5539 
5540 	fib6_entry = mlxsw_sp_fib6_entry_create(mlxsw_sp, fib_node, rt_arr,
5541 						nrt6);
5542 	if (IS_ERR(fib6_entry)) {
5543 		err = PTR_ERR(fib6_entry);
5544 		goto err_fib6_entry_create;
5545 	}
5546 
5547 	if (!mlxsw_sp_fib6_allow_replace(fib6_entry)) {
5548 		mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5549 		mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5550 		return 0;
5551 	}
5552 
5553 	replaced = fib_node->fib_entry;
5554 	err = mlxsw_sp_fib_node_entry_link(mlxsw_sp, &fib6_entry->common);
5555 	if (err)
5556 		goto err_fib_node_entry_link;
5557 
5558 	/* Nothing to replace */
5559 	if (!replaced)
5560 		return 0;
5561 
5562 	mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, replaced);
5563 	fib6_replaced = container_of(replaced, struct mlxsw_sp_fib6_entry,
5564 				     common);
5565 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_replaced);
5566 
5567 	return 0;
5568 
5569 err_fib_node_entry_link:
5570 	fib_node->fib_entry = replaced;
5571 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5572 err_fib6_entry_create:
5573 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5574 	return err;
5575 }
5576 
5577 static int mlxsw_sp_router_fib6_append(struct mlxsw_sp *mlxsw_sp,
5578 				       struct fib6_info **rt_arr,
5579 				       unsigned int nrt6)
5580 {
5581 	struct mlxsw_sp_fib6_entry *fib6_entry;
5582 	struct mlxsw_sp_fib_node *fib_node;
5583 	struct fib6_info *rt = rt_arr[0];
5584 	int err;
5585 
5586 	if (mlxsw_sp->router->aborted)
5587 		return 0;
5588 
5589 	if (rt->fib6_src.plen)
5590 		return -EINVAL;
5591 
5592 	if (mlxsw_sp_fib6_rt_should_ignore(rt))
5593 		return 0;
5594 
5595 	fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->fib6_table->tb6_id,
5596 					 &rt->fib6_dst.addr,
5597 					 sizeof(rt->fib6_dst.addr),
5598 					 rt->fib6_dst.plen,
5599 					 MLXSW_SP_L3_PROTO_IPV6);
5600 	if (IS_ERR(fib_node))
5601 		return PTR_ERR(fib_node);
5602 
5603 	if (WARN_ON_ONCE(!fib_node->fib_entry)) {
5604 		mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5605 		return -EINVAL;
5606 	}
5607 
5608 	fib6_entry = container_of(fib_node->fib_entry,
5609 				  struct mlxsw_sp_fib6_entry, common);
5610 	err = mlxsw_sp_fib6_entry_nexthop_add(mlxsw_sp, fib6_entry, rt_arr,
5611 					      nrt6);
5612 	if (err)
5613 		goto err_fib6_entry_nexthop_add;
5614 
5615 	return 0;
5616 
5617 err_fib6_entry_nexthop_add:
5618 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5619 	return err;
5620 }
5621 
5622 static void mlxsw_sp_router_fib6_del(struct mlxsw_sp *mlxsw_sp,
5623 				     struct fib6_info **rt_arr,
5624 				     unsigned int nrt6)
5625 {
5626 	struct mlxsw_sp_fib6_entry *fib6_entry;
5627 	struct mlxsw_sp_fib_node *fib_node;
5628 	struct fib6_info *rt = rt_arr[0];
5629 
5630 	if (mlxsw_sp->router->aborted)
5631 		return;
5632 
5633 	if (mlxsw_sp_fib6_rt_should_ignore(rt))
5634 		return;
5635 
5636 	/* Multipath routes are first added to the FIB trie and only then
5637 	 * notified. If we vetoed the addition, we will get a delete
5638 	 * notification for a route we do not have. Therefore, do not warn if
5639 	 * route was not found.
5640 	 */
5641 	fib6_entry = mlxsw_sp_fib6_entry_lookup(mlxsw_sp, rt);
5642 	if (!fib6_entry)
5643 		return;
5644 
5645 	/* If not all the nexthops are deleted, then only reduce the nexthop
5646 	 * group.
5647 	 */
5648 	if (nrt6 != fib6_entry->nrt6) {
5649 		mlxsw_sp_fib6_entry_nexthop_del(mlxsw_sp, fib6_entry, rt_arr,
5650 						nrt6);
5651 		return;
5652 	}
5653 
5654 	fib_node = fib6_entry->common.fib_node;
5655 
5656 	mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, &fib6_entry->common);
5657 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5658 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5659 }
5660 
5661 static int __mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp,
5662 					    enum mlxsw_reg_ralxx_protocol proto,
5663 					    u8 tree_id)
5664 {
5665 	char ralta_pl[MLXSW_REG_RALTA_LEN];
5666 	char ralst_pl[MLXSW_REG_RALST_LEN];
5667 	int i, err;
5668 
5669 	mlxsw_reg_ralta_pack(ralta_pl, true, proto, tree_id);
5670 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta), ralta_pl);
5671 	if (err)
5672 		return err;
5673 
5674 	mlxsw_reg_ralst_pack(ralst_pl, 0xff, tree_id);
5675 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst), ralst_pl);
5676 	if (err)
5677 		return err;
5678 
5679 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
5680 		struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
5681 		char raltb_pl[MLXSW_REG_RALTB_LEN];
5682 		char ralue_pl[MLXSW_REG_RALUE_LEN];
5683 
5684 		mlxsw_reg_raltb_pack(raltb_pl, vr->id, proto, tree_id);
5685 		err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb),
5686 				      raltb_pl);
5687 		if (err)
5688 			return err;
5689 
5690 		mlxsw_reg_ralue_pack(ralue_pl, proto,
5691 				     MLXSW_REG_RALUE_OP_WRITE_WRITE, vr->id, 0);
5692 		mlxsw_reg_ralue_act_ip2me_pack(ralue_pl);
5693 		err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue),
5694 				      ralue_pl);
5695 		if (err)
5696 			return err;
5697 	}
5698 
5699 	return 0;
5700 }
5701 
5702 static struct mlxsw_sp_mr_table *
5703 mlxsw_sp_router_fibmr_family_to_table(struct mlxsw_sp_vr *vr, int family)
5704 {
5705 	if (family == RTNL_FAMILY_IPMR)
5706 		return vr->mr_table[MLXSW_SP_L3_PROTO_IPV4];
5707 	else
5708 		return vr->mr_table[MLXSW_SP_L3_PROTO_IPV6];
5709 }
5710 
5711 static int mlxsw_sp_router_fibmr_add(struct mlxsw_sp *mlxsw_sp,
5712 				     struct mfc_entry_notifier_info *men_info,
5713 				     bool replace)
5714 {
5715 	struct mlxsw_sp_mr_table *mrt;
5716 	struct mlxsw_sp_vr *vr;
5717 
5718 	if (mlxsw_sp->router->aborted)
5719 		return 0;
5720 
5721 	vr = mlxsw_sp_vr_get(mlxsw_sp, men_info->tb_id, NULL);
5722 	if (IS_ERR(vr))
5723 		return PTR_ERR(vr);
5724 
5725 	mrt = mlxsw_sp_router_fibmr_family_to_table(vr, men_info->info.family);
5726 	return mlxsw_sp_mr_route_add(mrt, men_info->mfc, replace);
5727 }
5728 
5729 static void mlxsw_sp_router_fibmr_del(struct mlxsw_sp *mlxsw_sp,
5730 				      struct mfc_entry_notifier_info *men_info)
5731 {
5732 	struct mlxsw_sp_mr_table *mrt;
5733 	struct mlxsw_sp_vr *vr;
5734 
5735 	if (mlxsw_sp->router->aborted)
5736 		return;
5737 
5738 	vr = mlxsw_sp_vr_find(mlxsw_sp, men_info->tb_id);
5739 	if (WARN_ON(!vr))
5740 		return;
5741 
5742 	mrt = mlxsw_sp_router_fibmr_family_to_table(vr, men_info->info.family);
5743 	mlxsw_sp_mr_route_del(mrt, men_info->mfc);
5744 	mlxsw_sp_vr_put(mlxsw_sp, vr);
5745 }
5746 
5747 static int
5748 mlxsw_sp_router_fibmr_vif_add(struct mlxsw_sp *mlxsw_sp,
5749 			      struct vif_entry_notifier_info *ven_info)
5750 {
5751 	struct mlxsw_sp_mr_table *mrt;
5752 	struct mlxsw_sp_rif *rif;
5753 	struct mlxsw_sp_vr *vr;
5754 
5755 	if (mlxsw_sp->router->aborted)
5756 		return 0;
5757 
5758 	vr = mlxsw_sp_vr_get(mlxsw_sp, ven_info->tb_id, NULL);
5759 	if (IS_ERR(vr))
5760 		return PTR_ERR(vr);
5761 
5762 	mrt = mlxsw_sp_router_fibmr_family_to_table(vr, ven_info->info.family);
5763 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, ven_info->dev);
5764 	return mlxsw_sp_mr_vif_add(mrt, ven_info->dev,
5765 				   ven_info->vif_index,
5766 				   ven_info->vif_flags, rif);
5767 }
5768 
5769 static void
5770 mlxsw_sp_router_fibmr_vif_del(struct mlxsw_sp *mlxsw_sp,
5771 			      struct vif_entry_notifier_info *ven_info)
5772 {
5773 	struct mlxsw_sp_mr_table *mrt;
5774 	struct mlxsw_sp_vr *vr;
5775 
5776 	if (mlxsw_sp->router->aborted)
5777 		return;
5778 
5779 	vr = mlxsw_sp_vr_find(mlxsw_sp, ven_info->tb_id);
5780 	if (WARN_ON(!vr))
5781 		return;
5782 
5783 	mrt = mlxsw_sp_router_fibmr_family_to_table(vr, ven_info->info.family);
5784 	mlxsw_sp_mr_vif_del(mrt, ven_info->vif_index);
5785 	mlxsw_sp_vr_put(mlxsw_sp, vr);
5786 }
5787 
5788 static int mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp)
5789 {
5790 	enum mlxsw_reg_ralxx_protocol proto = MLXSW_REG_RALXX_PROTOCOL_IPV4;
5791 	int err;
5792 
5793 	err = __mlxsw_sp_router_set_abort_trap(mlxsw_sp, proto,
5794 					       MLXSW_SP_LPM_TREE_MIN);
5795 	if (err)
5796 		return err;
5797 
5798 	/* The multicast router code does not need an abort trap as by default,
5799 	 * packets that don't match any routes are trapped to the CPU.
5800 	 */
5801 
5802 	proto = MLXSW_REG_RALXX_PROTOCOL_IPV6;
5803 	return __mlxsw_sp_router_set_abort_trap(mlxsw_sp, proto,
5804 						MLXSW_SP_LPM_TREE_MIN + 1);
5805 }
5806 
5807 static void mlxsw_sp_fib4_node_flush(struct mlxsw_sp *mlxsw_sp,
5808 				     struct mlxsw_sp_fib_node *fib_node)
5809 {
5810 	struct mlxsw_sp_fib4_entry *fib4_entry;
5811 
5812 	fib4_entry = container_of(fib_node->fib_entry,
5813 				  struct mlxsw_sp_fib4_entry, common);
5814 	mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, fib_node->fib_entry);
5815 	mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
5816 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5817 }
5818 
5819 static void mlxsw_sp_fib6_node_flush(struct mlxsw_sp *mlxsw_sp,
5820 				     struct mlxsw_sp_fib_node *fib_node)
5821 {
5822 	struct mlxsw_sp_fib6_entry *fib6_entry;
5823 
5824 	fib6_entry = container_of(fib_node->fib_entry,
5825 				  struct mlxsw_sp_fib6_entry, common);
5826 	mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, fib_node->fib_entry);
5827 	mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5828 	mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5829 }
5830 
5831 static void mlxsw_sp_fib_node_flush(struct mlxsw_sp *mlxsw_sp,
5832 				    struct mlxsw_sp_fib_node *fib_node)
5833 {
5834 	switch (fib_node->fib->proto) {
5835 	case MLXSW_SP_L3_PROTO_IPV4:
5836 		mlxsw_sp_fib4_node_flush(mlxsw_sp, fib_node);
5837 		break;
5838 	case MLXSW_SP_L3_PROTO_IPV6:
5839 		mlxsw_sp_fib6_node_flush(mlxsw_sp, fib_node);
5840 		break;
5841 	}
5842 }
5843 
5844 static void mlxsw_sp_vr_fib_flush(struct mlxsw_sp *mlxsw_sp,
5845 				  struct mlxsw_sp_vr *vr,
5846 				  enum mlxsw_sp_l3proto proto)
5847 {
5848 	struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
5849 	struct mlxsw_sp_fib_node *fib_node, *tmp;
5850 
5851 	list_for_each_entry_safe(fib_node, tmp, &fib->node_list, list) {
5852 		bool do_break = &tmp->list == &fib->node_list;
5853 
5854 		mlxsw_sp_fib_node_flush(mlxsw_sp, fib_node);
5855 		if (do_break)
5856 			break;
5857 	}
5858 }
5859 
5860 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp)
5861 {
5862 	int i, j;
5863 
5864 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
5865 		struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
5866 
5867 		if (!mlxsw_sp_vr_is_used(vr))
5868 			continue;
5869 
5870 		for (j = 0; j < MLXSW_SP_L3_PROTO_MAX; j++)
5871 			mlxsw_sp_mr_table_flush(vr->mr_table[j]);
5872 		mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV4);
5873 
5874 		/* If virtual router was only used for IPv4, then it's no
5875 		 * longer used.
5876 		 */
5877 		if (!mlxsw_sp_vr_is_used(vr))
5878 			continue;
5879 		mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV6);
5880 	}
5881 
5882 	/* After flushing all the routes, it is not possible anyone is still
5883 	 * using the adjacency index that is discarding packets, so free it in
5884 	 * case it was allocated.
5885 	 */
5886 	if (!mlxsw_sp->router->adj_discard_index_valid)
5887 		return;
5888 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
5889 			   mlxsw_sp->router->adj_discard_index);
5890 	mlxsw_sp->router->adj_discard_index_valid = false;
5891 }
5892 
5893 static void mlxsw_sp_router_fib_abort(struct mlxsw_sp *mlxsw_sp)
5894 {
5895 	int err;
5896 
5897 	if (mlxsw_sp->router->aborted)
5898 		return;
5899 	dev_warn(mlxsw_sp->bus_info->dev, "FIB abort triggered. Note that FIB entries are no longer being offloaded to this device.\n");
5900 	mlxsw_sp_router_fib_flush(mlxsw_sp);
5901 	mlxsw_sp->router->aborted = true;
5902 	err = mlxsw_sp_router_set_abort_trap(mlxsw_sp);
5903 	if (err)
5904 		dev_warn(mlxsw_sp->bus_info->dev, "Failed to set abort trap.\n");
5905 }
5906 
5907 struct mlxsw_sp_fib6_event_work {
5908 	struct fib6_info **rt_arr;
5909 	unsigned int nrt6;
5910 };
5911 
5912 struct mlxsw_sp_fib_event_work {
5913 	struct work_struct work;
5914 	union {
5915 		struct mlxsw_sp_fib6_event_work fib6_work;
5916 		struct fib_entry_notifier_info fen_info;
5917 		struct fib_rule_notifier_info fr_info;
5918 		struct fib_nh_notifier_info fnh_info;
5919 		struct mfc_entry_notifier_info men_info;
5920 		struct vif_entry_notifier_info ven_info;
5921 	};
5922 	struct mlxsw_sp *mlxsw_sp;
5923 	unsigned long event;
5924 };
5925 
5926 static int
5927 mlxsw_sp_router_fib6_work_init(struct mlxsw_sp_fib6_event_work *fib6_work,
5928 			       struct fib6_entry_notifier_info *fen6_info)
5929 {
5930 	struct fib6_info *rt = fen6_info->rt;
5931 	struct fib6_info **rt_arr;
5932 	struct fib6_info *iter;
5933 	unsigned int nrt6;
5934 	int i = 0;
5935 
5936 	nrt6 = fen6_info->nsiblings + 1;
5937 
5938 	rt_arr = kcalloc(nrt6, sizeof(struct fib6_info *), GFP_ATOMIC);
5939 	if (!rt_arr)
5940 		return -ENOMEM;
5941 
5942 	fib6_work->rt_arr = rt_arr;
5943 	fib6_work->nrt6 = nrt6;
5944 
5945 	rt_arr[0] = rt;
5946 	fib6_info_hold(rt);
5947 
5948 	if (!fen6_info->nsiblings)
5949 		return 0;
5950 
5951 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
5952 		if (i == fen6_info->nsiblings)
5953 			break;
5954 
5955 		rt_arr[i + 1] = iter;
5956 		fib6_info_hold(iter);
5957 		i++;
5958 	}
5959 	WARN_ON_ONCE(i != fen6_info->nsiblings);
5960 
5961 	return 0;
5962 }
5963 
5964 static void
5965 mlxsw_sp_router_fib6_work_fini(struct mlxsw_sp_fib6_event_work *fib6_work)
5966 {
5967 	int i;
5968 
5969 	for (i = 0; i < fib6_work->nrt6; i++)
5970 		mlxsw_sp_rt6_release(fib6_work->rt_arr[i]);
5971 	kfree(fib6_work->rt_arr);
5972 }
5973 
5974 static void mlxsw_sp_router_fib4_event_work(struct work_struct *work)
5975 {
5976 	struct mlxsw_sp_fib_event_work *fib_work =
5977 		container_of(work, struct mlxsw_sp_fib_event_work, work);
5978 	struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
5979 	int err;
5980 
5981 	mutex_lock(&mlxsw_sp->router->lock);
5982 	mlxsw_sp_span_respin(mlxsw_sp);
5983 
5984 	switch (fib_work->event) {
5985 	case FIB_EVENT_ENTRY_REPLACE:
5986 		err = mlxsw_sp_router_fib4_replace(mlxsw_sp,
5987 						   &fib_work->fen_info);
5988 		if (err)
5989 			mlxsw_sp_router_fib_abort(mlxsw_sp);
5990 		fib_info_put(fib_work->fen_info.fi);
5991 		break;
5992 	case FIB_EVENT_ENTRY_DEL:
5993 		mlxsw_sp_router_fib4_del(mlxsw_sp, &fib_work->fen_info);
5994 		fib_info_put(fib_work->fen_info.fi);
5995 		break;
5996 	case FIB_EVENT_NH_ADD: /* fall through */
5997 	case FIB_EVENT_NH_DEL:
5998 		mlxsw_sp_nexthop4_event(mlxsw_sp, fib_work->event,
5999 					fib_work->fnh_info.fib_nh);
6000 		fib_info_put(fib_work->fnh_info.fib_nh->nh_parent);
6001 		break;
6002 	}
6003 	mutex_unlock(&mlxsw_sp->router->lock);
6004 	kfree(fib_work);
6005 }
6006 
6007 static void mlxsw_sp_router_fib6_event_work(struct work_struct *work)
6008 {
6009 	struct mlxsw_sp_fib_event_work *fib_work =
6010 		container_of(work, struct mlxsw_sp_fib_event_work, work);
6011 	struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
6012 	int err;
6013 
6014 	mutex_lock(&mlxsw_sp->router->lock);
6015 	mlxsw_sp_span_respin(mlxsw_sp);
6016 
6017 	switch (fib_work->event) {
6018 	case FIB_EVENT_ENTRY_REPLACE:
6019 		err = mlxsw_sp_router_fib6_replace(mlxsw_sp,
6020 						   fib_work->fib6_work.rt_arr,
6021 						   fib_work->fib6_work.nrt6);
6022 		if (err)
6023 			mlxsw_sp_router_fib_abort(mlxsw_sp);
6024 		mlxsw_sp_router_fib6_work_fini(&fib_work->fib6_work);
6025 		break;
6026 	case FIB_EVENT_ENTRY_APPEND:
6027 		err = mlxsw_sp_router_fib6_append(mlxsw_sp,
6028 						  fib_work->fib6_work.rt_arr,
6029 						  fib_work->fib6_work.nrt6);
6030 		if (err)
6031 			mlxsw_sp_router_fib_abort(mlxsw_sp);
6032 		mlxsw_sp_router_fib6_work_fini(&fib_work->fib6_work);
6033 		break;
6034 	case FIB_EVENT_ENTRY_DEL:
6035 		mlxsw_sp_router_fib6_del(mlxsw_sp,
6036 					 fib_work->fib6_work.rt_arr,
6037 					 fib_work->fib6_work.nrt6);
6038 		mlxsw_sp_router_fib6_work_fini(&fib_work->fib6_work);
6039 		break;
6040 	}
6041 	mutex_unlock(&mlxsw_sp->router->lock);
6042 	kfree(fib_work);
6043 }
6044 
6045 static void mlxsw_sp_router_fibmr_event_work(struct work_struct *work)
6046 {
6047 	struct mlxsw_sp_fib_event_work *fib_work =
6048 		container_of(work, struct mlxsw_sp_fib_event_work, work);
6049 	struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
6050 	bool replace;
6051 	int err;
6052 
6053 	rtnl_lock();
6054 	mutex_lock(&mlxsw_sp->router->lock);
6055 	switch (fib_work->event) {
6056 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
6057 	case FIB_EVENT_ENTRY_ADD:
6058 		replace = fib_work->event == FIB_EVENT_ENTRY_REPLACE;
6059 
6060 		err = mlxsw_sp_router_fibmr_add(mlxsw_sp, &fib_work->men_info,
6061 						replace);
6062 		if (err)
6063 			mlxsw_sp_router_fib_abort(mlxsw_sp);
6064 		mr_cache_put(fib_work->men_info.mfc);
6065 		break;
6066 	case FIB_EVENT_ENTRY_DEL:
6067 		mlxsw_sp_router_fibmr_del(mlxsw_sp, &fib_work->men_info);
6068 		mr_cache_put(fib_work->men_info.mfc);
6069 		break;
6070 	case FIB_EVENT_VIF_ADD:
6071 		err = mlxsw_sp_router_fibmr_vif_add(mlxsw_sp,
6072 						    &fib_work->ven_info);
6073 		if (err)
6074 			mlxsw_sp_router_fib_abort(mlxsw_sp);
6075 		dev_put(fib_work->ven_info.dev);
6076 		break;
6077 	case FIB_EVENT_VIF_DEL:
6078 		mlxsw_sp_router_fibmr_vif_del(mlxsw_sp,
6079 					      &fib_work->ven_info);
6080 		dev_put(fib_work->ven_info.dev);
6081 		break;
6082 	}
6083 	mutex_unlock(&mlxsw_sp->router->lock);
6084 	rtnl_unlock();
6085 	kfree(fib_work);
6086 }
6087 
6088 static void mlxsw_sp_router_fib4_event(struct mlxsw_sp_fib_event_work *fib_work,
6089 				       struct fib_notifier_info *info)
6090 {
6091 	struct fib_entry_notifier_info *fen_info;
6092 	struct fib_nh_notifier_info *fnh_info;
6093 
6094 	switch (fib_work->event) {
6095 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
6096 	case FIB_EVENT_ENTRY_DEL:
6097 		fen_info = container_of(info, struct fib_entry_notifier_info,
6098 					info);
6099 		fib_work->fen_info = *fen_info;
6100 		/* Take reference on fib_info to prevent it from being
6101 		 * freed while work is queued. Release it afterwards.
6102 		 */
6103 		fib_info_hold(fib_work->fen_info.fi);
6104 		break;
6105 	case FIB_EVENT_NH_ADD: /* fall through */
6106 	case FIB_EVENT_NH_DEL:
6107 		fnh_info = container_of(info, struct fib_nh_notifier_info,
6108 					info);
6109 		fib_work->fnh_info = *fnh_info;
6110 		fib_info_hold(fib_work->fnh_info.fib_nh->nh_parent);
6111 		break;
6112 	}
6113 }
6114 
6115 static int mlxsw_sp_router_fib6_event(struct mlxsw_sp_fib_event_work *fib_work,
6116 				      struct fib_notifier_info *info)
6117 {
6118 	struct fib6_entry_notifier_info *fen6_info;
6119 	int err;
6120 
6121 	switch (fib_work->event) {
6122 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
6123 	case FIB_EVENT_ENTRY_APPEND: /* fall through */
6124 	case FIB_EVENT_ENTRY_DEL:
6125 		fen6_info = container_of(info, struct fib6_entry_notifier_info,
6126 					 info);
6127 		err = mlxsw_sp_router_fib6_work_init(&fib_work->fib6_work,
6128 						     fen6_info);
6129 		if (err)
6130 			return err;
6131 		break;
6132 	}
6133 
6134 	return 0;
6135 }
6136 
6137 static void
6138 mlxsw_sp_router_fibmr_event(struct mlxsw_sp_fib_event_work *fib_work,
6139 			    struct fib_notifier_info *info)
6140 {
6141 	switch (fib_work->event) {
6142 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
6143 	case FIB_EVENT_ENTRY_ADD: /* fall through */
6144 	case FIB_EVENT_ENTRY_DEL:
6145 		memcpy(&fib_work->men_info, info, sizeof(fib_work->men_info));
6146 		mr_cache_hold(fib_work->men_info.mfc);
6147 		break;
6148 	case FIB_EVENT_VIF_ADD: /* fall through */
6149 	case FIB_EVENT_VIF_DEL:
6150 		memcpy(&fib_work->ven_info, info, sizeof(fib_work->ven_info));
6151 		dev_hold(fib_work->ven_info.dev);
6152 		break;
6153 	}
6154 }
6155 
6156 static int mlxsw_sp_router_fib_rule_event(unsigned long event,
6157 					  struct fib_notifier_info *info,
6158 					  struct mlxsw_sp *mlxsw_sp)
6159 {
6160 	struct netlink_ext_ack *extack = info->extack;
6161 	struct fib_rule_notifier_info *fr_info;
6162 	struct fib_rule *rule;
6163 	int err = 0;
6164 
6165 	/* nothing to do at the moment */
6166 	if (event == FIB_EVENT_RULE_DEL)
6167 		return 0;
6168 
6169 	if (mlxsw_sp->router->aborted)
6170 		return 0;
6171 
6172 	fr_info = container_of(info, struct fib_rule_notifier_info, info);
6173 	rule = fr_info->rule;
6174 
6175 	/* Rule only affects locally generated traffic */
6176 	if (rule->iifindex == mlxsw_sp_net(mlxsw_sp)->loopback_dev->ifindex)
6177 		return 0;
6178 
6179 	switch (info->family) {
6180 	case AF_INET:
6181 		if (!fib4_rule_default(rule) && !rule->l3mdev)
6182 			err = -EOPNOTSUPP;
6183 		break;
6184 	case AF_INET6:
6185 		if (!fib6_rule_default(rule) && !rule->l3mdev)
6186 			err = -EOPNOTSUPP;
6187 		break;
6188 	case RTNL_FAMILY_IPMR:
6189 		if (!ipmr_rule_default(rule) && !rule->l3mdev)
6190 			err = -EOPNOTSUPP;
6191 		break;
6192 	case RTNL_FAMILY_IP6MR:
6193 		if (!ip6mr_rule_default(rule) && !rule->l3mdev)
6194 			err = -EOPNOTSUPP;
6195 		break;
6196 	}
6197 
6198 	if (err < 0)
6199 		NL_SET_ERR_MSG_MOD(extack, "FIB rules not supported");
6200 
6201 	return err;
6202 }
6203 
6204 /* Called with rcu_read_lock() */
6205 static int mlxsw_sp_router_fib_event(struct notifier_block *nb,
6206 				     unsigned long event, void *ptr)
6207 {
6208 	struct mlxsw_sp_fib_event_work *fib_work;
6209 	struct fib_notifier_info *info = ptr;
6210 	struct mlxsw_sp_router *router;
6211 	int err;
6212 
6213 	if ((info->family != AF_INET && info->family != AF_INET6 &&
6214 	     info->family != RTNL_FAMILY_IPMR &&
6215 	     info->family != RTNL_FAMILY_IP6MR))
6216 		return NOTIFY_DONE;
6217 
6218 	router = container_of(nb, struct mlxsw_sp_router, fib_nb);
6219 
6220 	switch (event) {
6221 	case FIB_EVENT_RULE_ADD: /* fall through */
6222 	case FIB_EVENT_RULE_DEL:
6223 		err = mlxsw_sp_router_fib_rule_event(event, info,
6224 						     router->mlxsw_sp);
6225 		return notifier_from_errno(err);
6226 	case FIB_EVENT_ENTRY_ADD: /* fall through */
6227 	case FIB_EVENT_ENTRY_REPLACE: /* fall through */
6228 	case FIB_EVENT_ENTRY_APPEND:
6229 		if (router->aborted) {
6230 			NL_SET_ERR_MSG_MOD(info->extack, "FIB offload was aborted. Not configuring route");
6231 			return notifier_from_errno(-EINVAL);
6232 		}
6233 		if (info->family == AF_INET) {
6234 			struct fib_entry_notifier_info *fen_info = ptr;
6235 
6236 			if (fen_info->fi->fib_nh_is_v6) {
6237 				NL_SET_ERR_MSG_MOD(info->extack, "IPv6 gateway with IPv4 route is not supported");
6238 				return notifier_from_errno(-EINVAL);
6239 			}
6240 			if (fen_info->fi->nh) {
6241 				NL_SET_ERR_MSG_MOD(info->extack, "IPv4 route with nexthop objects is not supported");
6242 				return notifier_from_errno(-EINVAL);
6243 			}
6244 		} else if (info->family == AF_INET6) {
6245 			struct fib6_entry_notifier_info *fen6_info;
6246 
6247 			fen6_info = container_of(info,
6248 						 struct fib6_entry_notifier_info,
6249 						 info);
6250 			if (fen6_info->rt->nh) {
6251 				NL_SET_ERR_MSG_MOD(info->extack, "IPv6 route with nexthop objects is not supported");
6252 				return notifier_from_errno(-EINVAL);
6253 			}
6254 		}
6255 		break;
6256 	}
6257 
6258 	fib_work = kzalloc(sizeof(*fib_work), GFP_ATOMIC);
6259 	if (WARN_ON(!fib_work))
6260 		return NOTIFY_BAD;
6261 
6262 	fib_work->mlxsw_sp = router->mlxsw_sp;
6263 	fib_work->event = event;
6264 
6265 	switch (info->family) {
6266 	case AF_INET:
6267 		INIT_WORK(&fib_work->work, mlxsw_sp_router_fib4_event_work);
6268 		mlxsw_sp_router_fib4_event(fib_work, info);
6269 		break;
6270 	case AF_INET6:
6271 		INIT_WORK(&fib_work->work, mlxsw_sp_router_fib6_event_work);
6272 		err = mlxsw_sp_router_fib6_event(fib_work, info);
6273 		if (err)
6274 			goto err_fib_event;
6275 		break;
6276 	case RTNL_FAMILY_IP6MR:
6277 	case RTNL_FAMILY_IPMR:
6278 		INIT_WORK(&fib_work->work, mlxsw_sp_router_fibmr_event_work);
6279 		mlxsw_sp_router_fibmr_event(fib_work, info);
6280 		break;
6281 	}
6282 
6283 	mlxsw_core_schedule_work(&fib_work->work);
6284 
6285 	return NOTIFY_DONE;
6286 
6287 err_fib_event:
6288 	kfree(fib_work);
6289 	return NOTIFY_BAD;
6290 }
6291 
6292 static struct mlxsw_sp_rif *
6293 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
6294 			 const struct net_device *dev)
6295 {
6296 	int i;
6297 
6298 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
6299 		if (mlxsw_sp->router->rifs[i] &&
6300 		    mlxsw_sp->router->rifs[i]->dev == dev)
6301 			return mlxsw_sp->router->rifs[i];
6302 
6303 	return NULL;
6304 }
6305 
6306 bool mlxsw_sp_rif_exists(struct mlxsw_sp *mlxsw_sp,
6307 			 const struct net_device *dev)
6308 {
6309 	struct mlxsw_sp_rif *rif;
6310 
6311 	mutex_lock(&mlxsw_sp->router->lock);
6312 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6313 	mutex_unlock(&mlxsw_sp->router->lock);
6314 
6315 	return rif;
6316 }
6317 
6318 u16 mlxsw_sp_rif_vid(struct mlxsw_sp *mlxsw_sp, const struct net_device *dev)
6319 {
6320 	struct mlxsw_sp_rif *rif;
6321 	u16 vid = 0;
6322 
6323 	mutex_lock(&mlxsw_sp->router->lock);
6324 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6325 	if (!rif)
6326 		goto out;
6327 
6328 	/* We only return the VID for VLAN RIFs. Otherwise we return an
6329 	 * invalid value (0).
6330 	 */
6331 	if (rif->ops->type != MLXSW_SP_RIF_TYPE_VLAN)
6332 		goto out;
6333 
6334 	vid = mlxsw_sp_fid_8021q_vid(rif->fid);
6335 
6336 out:
6337 	mutex_unlock(&mlxsw_sp->router->lock);
6338 	return vid;
6339 }
6340 
6341 static int mlxsw_sp_router_rif_disable(struct mlxsw_sp *mlxsw_sp, u16 rif)
6342 {
6343 	char ritr_pl[MLXSW_REG_RITR_LEN];
6344 	int err;
6345 
6346 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif);
6347 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6348 	if (err)
6349 		return err;
6350 
6351 	mlxsw_reg_ritr_enable_set(ritr_pl, false);
6352 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6353 }
6354 
6355 static void mlxsw_sp_router_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
6356 					  struct mlxsw_sp_rif *rif)
6357 {
6358 	mlxsw_sp_router_rif_disable(mlxsw_sp, rif->rif_index);
6359 	mlxsw_sp_nexthop_rif_gone_sync(mlxsw_sp, rif);
6360 	mlxsw_sp_neigh_rif_gone_sync(mlxsw_sp, rif);
6361 }
6362 
6363 static bool
6364 mlxsw_sp_rif_should_config(struct mlxsw_sp_rif *rif, struct net_device *dev,
6365 			   unsigned long event)
6366 {
6367 	struct inet6_dev *inet6_dev;
6368 	bool addr_list_empty = true;
6369 	struct in_device *idev;
6370 
6371 	switch (event) {
6372 	case NETDEV_UP:
6373 		return rif == NULL;
6374 	case NETDEV_DOWN:
6375 		rcu_read_lock();
6376 		idev = __in_dev_get_rcu(dev);
6377 		if (idev && idev->ifa_list)
6378 			addr_list_empty = false;
6379 
6380 		inet6_dev = __in6_dev_get(dev);
6381 		if (addr_list_empty && inet6_dev &&
6382 		    !list_empty(&inet6_dev->addr_list))
6383 			addr_list_empty = false;
6384 		rcu_read_unlock();
6385 
6386 		/* macvlans do not have a RIF, but rather piggy back on the
6387 		 * RIF of their lower device.
6388 		 */
6389 		if (netif_is_macvlan(dev) && addr_list_empty)
6390 			return true;
6391 
6392 		if (rif && addr_list_empty &&
6393 		    !netif_is_l3_slave(rif->dev))
6394 			return true;
6395 		/* It is possible we already removed the RIF ourselves
6396 		 * if it was assigned to a netdev that is now a bridge
6397 		 * or LAG slave.
6398 		 */
6399 		return false;
6400 	}
6401 
6402 	return false;
6403 }
6404 
6405 static enum mlxsw_sp_rif_type
6406 mlxsw_sp_dev_rif_type(const struct mlxsw_sp *mlxsw_sp,
6407 		      const struct net_device *dev)
6408 {
6409 	enum mlxsw_sp_fid_type type;
6410 
6411 	if (mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL))
6412 		return MLXSW_SP_RIF_TYPE_IPIP_LB;
6413 
6414 	/* Otherwise RIF type is derived from the type of the underlying FID. */
6415 	if (is_vlan_dev(dev) && netif_is_bridge_master(vlan_dev_real_dev(dev)))
6416 		type = MLXSW_SP_FID_TYPE_8021Q;
6417 	else if (netif_is_bridge_master(dev) && br_vlan_enabled(dev))
6418 		type = MLXSW_SP_FID_TYPE_8021Q;
6419 	else if (netif_is_bridge_master(dev))
6420 		type = MLXSW_SP_FID_TYPE_8021D;
6421 	else
6422 		type = MLXSW_SP_FID_TYPE_RFID;
6423 
6424 	return mlxsw_sp_fid_type_rif_type(mlxsw_sp, type);
6425 }
6426 
6427 static int mlxsw_sp_rif_index_alloc(struct mlxsw_sp *mlxsw_sp, u16 *p_rif_index)
6428 {
6429 	int i;
6430 
6431 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++) {
6432 		if (!mlxsw_sp->router->rifs[i]) {
6433 			*p_rif_index = i;
6434 			return 0;
6435 		}
6436 	}
6437 
6438 	return -ENOBUFS;
6439 }
6440 
6441 static struct mlxsw_sp_rif *mlxsw_sp_rif_alloc(size_t rif_size, u16 rif_index,
6442 					       u16 vr_id,
6443 					       struct net_device *l3_dev)
6444 {
6445 	struct mlxsw_sp_rif *rif;
6446 
6447 	rif = kzalloc(rif_size, GFP_KERNEL);
6448 	if (!rif)
6449 		return NULL;
6450 
6451 	INIT_LIST_HEAD(&rif->nexthop_list);
6452 	INIT_LIST_HEAD(&rif->neigh_list);
6453 	if (l3_dev) {
6454 		ether_addr_copy(rif->addr, l3_dev->dev_addr);
6455 		rif->mtu = l3_dev->mtu;
6456 		rif->dev = l3_dev;
6457 	}
6458 	rif->vr_id = vr_id;
6459 	rif->rif_index = rif_index;
6460 
6461 	return rif;
6462 }
6463 
6464 struct mlxsw_sp_rif *mlxsw_sp_rif_by_index(const struct mlxsw_sp *mlxsw_sp,
6465 					   u16 rif_index)
6466 {
6467 	return mlxsw_sp->router->rifs[rif_index];
6468 }
6469 
6470 u16 mlxsw_sp_rif_index(const struct mlxsw_sp_rif *rif)
6471 {
6472 	return rif->rif_index;
6473 }
6474 
6475 u16 mlxsw_sp_ipip_lb_rif_index(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
6476 {
6477 	return lb_rif->common.rif_index;
6478 }
6479 
6480 u16 mlxsw_sp_ipip_lb_ul_vr_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
6481 {
6482 	u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(lb_rif->common.dev);
6483 	struct mlxsw_sp_vr *ul_vr;
6484 
6485 	ul_vr = mlxsw_sp_vr_get(lb_rif->common.mlxsw_sp, ul_tb_id, NULL);
6486 	if (WARN_ON(IS_ERR(ul_vr)))
6487 		return 0;
6488 
6489 	return ul_vr->id;
6490 }
6491 
6492 u16 mlxsw_sp_ipip_lb_ul_rif_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
6493 {
6494 	return lb_rif->ul_rif_id;
6495 }
6496 
6497 int mlxsw_sp_rif_dev_ifindex(const struct mlxsw_sp_rif *rif)
6498 {
6499 	return rif->dev->ifindex;
6500 }
6501 
6502 const struct net_device *mlxsw_sp_rif_dev(const struct mlxsw_sp_rif *rif)
6503 {
6504 	return rif->dev;
6505 }
6506 
6507 static struct mlxsw_sp_rif *
6508 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
6509 		    const struct mlxsw_sp_rif_params *params,
6510 		    struct netlink_ext_ack *extack)
6511 {
6512 	u32 tb_id = l3mdev_fib_table(params->dev);
6513 	const struct mlxsw_sp_rif_ops *ops;
6514 	struct mlxsw_sp_fid *fid = NULL;
6515 	enum mlxsw_sp_rif_type type;
6516 	struct mlxsw_sp_rif *rif;
6517 	struct mlxsw_sp_vr *vr;
6518 	u16 rif_index;
6519 	int i, err;
6520 
6521 	type = mlxsw_sp_dev_rif_type(mlxsw_sp, params->dev);
6522 	ops = mlxsw_sp->rif_ops_arr[type];
6523 
6524 	vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id ? : RT_TABLE_MAIN, extack);
6525 	if (IS_ERR(vr))
6526 		return ERR_CAST(vr);
6527 	vr->rif_count++;
6528 
6529 	err = mlxsw_sp_rif_index_alloc(mlxsw_sp, &rif_index);
6530 	if (err) {
6531 		NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported router interfaces");
6532 		goto err_rif_index_alloc;
6533 	}
6534 
6535 	rif = mlxsw_sp_rif_alloc(ops->rif_size, rif_index, vr->id, params->dev);
6536 	if (!rif) {
6537 		err = -ENOMEM;
6538 		goto err_rif_alloc;
6539 	}
6540 	dev_hold(rif->dev);
6541 	mlxsw_sp->router->rifs[rif_index] = rif;
6542 	rif->mlxsw_sp = mlxsw_sp;
6543 	rif->ops = ops;
6544 
6545 	if (ops->fid_get) {
6546 		fid = ops->fid_get(rif, extack);
6547 		if (IS_ERR(fid)) {
6548 			err = PTR_ERR(fid);
6549 			goto err_fid_get;
6550 		}
6551 		rif->fid = fid;
6552 	}
6553 
6554 	if (ops->setup)
6555 		ops->setup(rif, params);
6556 
6557 	err = ops->configure(rif);
6558 	if (err)
6559 		goto err_configure;
6560 
6561 	for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++) {
6562 		err = mlxsw_sp_mr_rif_add(vr->mr_table[i], rif);
6563 		if (err)
6564 			goto err_mr_rif_add;
6565 	}
6566 
6567 	mlxsw_sp_rif_counters_alloc(rif);
6568 
6569 	return rif;
6570 
6571 err_mr_rif_add:
6572 	for (i--; i >= 0; i--)
6573 		mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
6574 	ops->deconfigure(rif);
6575 err_configure:
6576 	if (fid)
6577 		mlxsw_sp_fid_put(fid);
6578 err_fid_get:
6579 	mlxsw_sp->router->rifs[rif_index] = NULL;
6580 	dev_put(rif->dev);
6581 	kfree(rif);
6582 err_rif_alloc:
6583 err_rif_index_alloc:
6584 	vr->rif_count--;
6585 	mlxsw_sp_vr_put(mlxsw_sp, vr);
6586 	return ERR_PTR(err);
6587 }
6588 
6589 static void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif)
6590 {
6591 	const struct mlxsw_sp_rif_ops *ops = rif->ops;
6592 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6593 	struct mlxsw_sp_fid *fid = rif->fid;
6594 	struct mlxsw_sp_vr *vr;
6595 	int i;
6596 
6597 	mlxsw_sp_router_rif_gone_sync(mlxsw_sp, rif);
6598 	vr = &mlxsw_sp->router->vrs[rif->vr_id];
6599 
6600 	mlxsw_sp_rif_counters_free(rif);
6601 	for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++)
6602 		mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
6603 	ops->deconfigure(rif);
6604 	if (fid)
6605 		/* Loopback RIFs are not associated with a FID. */
6606 		mlxsw_sp_fid_put(fid);
6607 	mlxsw_sp->router->rifs[rif->rif_index] = NULL;
6608 	dev_put(rif->dev);
6609 	kfree(rif);
6610 	vr->rif_count--;
6611 	mlxsw_sp_vr_put(mlxsw_sp, vr);
6612 }
6613 
6614 void mlxsw_sp_rif_destroy_by_dev(struct mlxsw_sp *mlxsw_sp,
6615 				 struct net_device *dev)
6616 {
6617 	struct mlxsw_sp_rif *rif;
6618 
6619 	mutex_lock(&mlxsw_sp->router->lock);
6620 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6621 	if (!rif)
6622 		goto out;
6623 	mlxsw_sp_rif_destroy(rif);
6624 out:
6625 	mutex_unlock(&mlxsw_sp->router->lock);
6626 }
6627 
6628 static void
6629 mlxsw_sp_rif_subport_params_init(struct mlxsw_sp_rif_params *params,
6630 				 struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
6631 {
6632 	struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
6633 
6634 	params->vid = mlxsw_sp_port_vlan->vid;
6635 	params->lag = mlxsw_sp_port->lagged;
6636 	if (params->lag)
6637 		params->lag_id = mlxsw_sp_port->lag_id;
6638 	else
6639 		params->system_port = mlxsw_sp_port->local_port;
6640 }
6641 
6642 static struct mlxsw_sp_rif_subport *
6643 mlxsw_sp_rif_subport_rif(const struct mlxsw_sp_rif *rif)
6644 {
6645 	return container_of(rif, struct mlxsw_sp_rif_subport, common);
6646 }
6647 
6648 static struct mlxsw_sp_rif *
6649 mlxsw_sp_rif_subport_get(struct mlxsw_sp *mlxsw_sp,
6650 			 const struct mlxsw_sp_rif_params *params,
6651 			 struct netlink_ext_ack *extack)
6652 {
6653 	struct mlxsw_sp_rif_subport *rif_subport;
6654 	struct mlxsw_sp_rif *rif;
6655 
6656 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, params->dev);
6657 	if (!rif)
6658 		return mlxsw_sp_rif_create(mlxsw_sp, params, extack);
6659 
6660 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
6661 	refcount_inc(&rif_subport->ref_count);
6662 	return rif;
6663 }
6664 
6665 static void mlxsw_sp_rif_subport_put(struct mlxsw_sp_rif *rif)
6666 {
6667 	struct mlxsw_sp_rif_subport *rif_subport;
6668 
6669 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
6670 	if (!refcount_dec_and_test(&rif_subport->ref_count))
6671 		return;
6672 
6673 	mlxsw_sp_rif_destroy(rif);
6674 }
6675 
6676 static int
6677 mlxsw_sp_port_vlan_router_join(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan,
6678 			       struct net_device *l3_dev,
6679 			       struct netlink_ext_ack *extack)
6680 {
6681 	struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
6682 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
6683 	struct mlxsw_sp_rif_params params = {
6684 		.dev = l3_dev,
6685 	};
6686 	u16 vid = mlxsw_sp_port_vlan->vid;
6687 	struct mlxsw_sp_rif *rif;
6688 	struct mlxsw_sp_fid *fid;
6689 	int err;
6690 
6691 	mlxsw_sp_rif_subport_params_init(&params, mlxsw_sp_port_vlan);
6692 	rif = mlxsw_sp_rif_subport_get(mlxsw_sp, &params, extack);
6693 	if (IS_ERR(rif))
6694 		return PTR_ERR(rif);
6695 
6696 	/* FID was already created, just take a reference */
6697 	fid = rif->ops->fid_get(rif, extack);
6698 	err = mlxsw_sp_fid_port_vid_map(fid, mlxsw_sp_port, vid);
6699 	if (err)
6700 		goto err_fid_port_vid_map;
6701 
6702 	err = mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, false);
6703 	if (err)
6704 		goto err_port_vid_learning_set;
6705 
6706 	err = mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid,
6707 					BR_STATE_FORWARDING);
6708 	if (err)
6709 		goto err_port_vid_stp_set;
6710 
6711 	mlxsw_sp_port_vlan->fid = fid;
6712 
6713 	return 0;
6714 
6715 err_port_vid_stp_set:
6716 	mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
6717 err_port_vid_learning_set:
6718 	mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
6719 err_fid_port_vid_map:
6720 	mlxsw_sp_fid_put(fid);
6721 	mlxsw_sp_rif_subport_put(rif);
6722 	return err;
6723 }
6724 
6725 static void
6726 __mlxsw_sp_port_vlan_router_leave(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
6727 {
6728 	struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
6729 	struct mlxsw_sp_fid *fid = mlxsw_sp_port_vlan->fid;
6730 	struct mlxsw_sp_rif *rif = mlxsw_sp_fid_rif(fid);
6731 	u16 vid = mlxsw_sp_port_vlan->vid;
6732 
6733 	if (WARN_ON(mlxsw_sp_fid_type(fid) != MLXSW_SP_FID_TYPE_RFID))
6734 		return;
6735 
6736 	mlxsw_sp_port_vlan->fid = NULL;
6737 	mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid, BR_STATE_BLOCKING);
6738 	mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
6739 	mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
6740 	mlxsw_sp_fid_put(fid);
6741 	mlxsw_sp_rif_subport_put(rif);
6742 }
6743 
6744 void
6745 mlxsw_sp_port_vlan_router_leave(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
6746 {
6747 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port_vlan->mlxsw_sp_port->mlxsw_sp;
6748 
6749 	mutex_lock(&mlxsw_sp->router->lock);
6750 	__mlxsw_sp_port_vlan_router_leave(mlxsw_sp_port_vlan);
6751 	mutex_unlock(&mlxsw_sp->router->lock);
6752 }
6753 
6754 static int mlxsw_sp_inetaddr_port_vlan_event(struct net_device *l3_dev,
6755 					     struct net_device *port_dev,
6756 					     unsigned long event, u16 vid,
6757 					     struct netlink_ext_ack *extack)
6758 {
6759 	struct mlxsw_sp_port *mlxsw_sp_port = netdev_priv(port_dev);
6760 	struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan;
6761 
6762 	mlxsw_sp_port_vlan = mlxsw_sp_port_vlan_find_by_vid(mlxsw_sp_port, vid);
6763 	if (WARN_ON(!mlxsw_sp_port_vlan))
6764 		return -EINVAL;
6765 
6766 	switch (event) {
6767 	case NETDEV_UP:
6768 		return mlxsw_sp_port_vlan_router_join(mlxsw_sp_port_vlan,
6769 						      l3_dev, extack);
6770 	case NETDEV_DOWN:
6771 		__mlxsw_sp_port_vlan_router_leave(mlxsw_sp_port_vlan);
6772 		break;
6773 	}
6774 
6775 	return 0;
6776 }
6777 
6778 static int mlxsw_sp_inetaddr_port_event(struct net_device *port_dev,
6779 					unsigned long event,
6780 					struct netlink_ext_ack *extack)
6781 {
6782 	if (netif_is_bridge_port(port_dev) ||
6783 	    netif_is_lag_port(port_dev) ||
6784 	    netif_is_ovs_port(port_dev))
6785 		return 0;
6786 
6787 	return mlxsw_sp_inetaddr_port_vlan_event(port_dev, port_dev, event,
6788 						 MLXSW_SP_DEFAULT_VID, extack);
6789 }
6790 
6791 static int __mlxsw_sp_inetaddr_lag_event(struct net_device *l3_dev,
6792 					 struct net_device *lag_dev,
6793 					 unsigned long event, u16 vid,
6794 					 struct netlink_ext_ack *extack)
6795 {
6796 	struct net_device *port_dev;
6797 	struct list_head *iter;
6798 	int err;
6799 
6800 	netdev_for_each_lower_dev(lag_dev, port_dev, iter) {
6801 		if (mlxsw_sp_port_dev_check(port_dev)) {
6802 			err = mlxsw_sp_inetaddr_port_vlan_event(l3_dev,
6803 								port_dev,
6804 								event, vid,
6805 								extack);
6806 			if (err)
6807 				return err;
6808 		}
6809 	}
6810 
6811 	return 0;
6812 }
6813 
6814 static int mlxsw_sp_inetaddr_lag_event(struct net_device *lag_dev,
6815 				       unsigned long event,
6816 				       struct netlink_ext_ack *extack)
6817 {
6818 	if (netif_is_bridge_port(lag_dev))
6819 		return 0;
6820 
6821 	return __mlxsw_sp_inetaddr_lag_event(lag_dev, lag_dev, event,
6822 					     MLXSW_SP_DEFAULT_VID, extack);
6823 }
6824 
6825 static int mlxsw_sp_inetaddr_bridge_event(struct mlxsw_sp *mlxsw_sp,
6826 					  struct net_device *l3_dev,
6827 					  unsigned long event,
6828 					  struct netlink_ext_ack *extack)
6829 {
6830 	struct mlxsw_sp_rif_params params = {
6831 		.dev = l3_dev,
6832 	};
6833 	struct mlxsw_sp_rif *rif;
6834 
6835 	switch (event) {
6836 	case NETDEV_UP:
6837 		rif = mlxsw_sp_rif_create(mlxsw_sp, &params, extack);
6838 		if (IS_ERR(rif))
6839 			return PTR_ERR(rif);
6840 		break;
6841 	case NETDEV_DOWN:
6842 		rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
6843 		mlxsw_sp_rif_destroy(rif);
6844 		break;
6845 	}
6846 
6847 	return 0;
6848 }
6849 
6850 static int mlxsw_sp_inetaddr_vlan_event(struct mlxsw_sp *mlxsw_sp,
6851 					struct net_device *vlan_dev,
6852 					unsigned long event,
6853 					struct netlink_ext_ack *extack)
6854 {
6855 	struct net_device *real_dev = vlan_dev_real_dev(vlan_dev);
6856 	u16 vid = vlan_dev_vlan_id(vlan_dev);
6857 
6858 	if (netif_is_bridge_port(vlan_dev))
6859 		return 0;
6860 
6861 	if (mlxsw_sp_port_dev_check(real_dev))
6862 		return mlxsw_sp_inetaddr_port_vlan_event(vlan_dev, real_dev,
6863 							 event, vid, extack);
6864 	else if (netif_is_lag_master(real_dev))
6865 		return __mlxsw_sp_inetaddr_lag_event(vlan_dev, real_dev, event,
6866 						     vid, extack);
6867 	else if (netif_is_bridge_master(real_dev) && br_vlan_enabled(real_dev))
6868 		return mlxsw_sp_inetaddr_bridge_event(mlxsw_sp, vlan_dev, event,
6869 						      extack);
6870 
6871 	return 0;
6872 }
6873 
6874 static bool mlxsw_sp_rif_macvlan_is_vrrp4(const u8 *mac)
6875 {
6876 	u8 vrrp4[ETH_ALEN] = { 0x00, 0x00, 0x5e, 0x00, 0x01, 0x00 };
6877 	u8 mask[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0x00 };
6878 
6879 	return ether_addr_equal_masked(mac, vrrp4, mask);
6880 }
6881 
6882 static bool mlxsw_sp_rif_macvlan_is_vrrp6(const u8 *mac)
6883 {
6884 	u8 vrrp6[ETH_ALEN] = { 0x00, 0x00, 0x5e, 0x00, 0x02, 0x00 };
6885 	u8 mask[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0x00 };
6886 
6887 	return ether_addr_equal_masked(mac, vrrp6, mask);
6888 }
6889 
6890 static int mlxsw_sp_rif_vrrp_op(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
6891 				const u8 *mac, bool adding)
6892 {
6893 	char ritr_pl[MLXSW_REG_RITR_LEN];
6894 	u8 vrrp_id = adding ? mac[5] : 0;
6895 	int err;
6896 
6897 	if (!mlxsw_sp_rif_macvlan_is_vrrp4(mac) &&
6898 	    !mlxsw_sp_rif_macvlan_is_vrrp6(mac))
6899 		return 0;
6900 
6901 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
6902 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6903 	if (err)
6904 		return err;
6905 
6906 	if (mlxsw_sp_rif_macvlan_is_vrrp4(mac))
6907 		mlxsw_reg_ritr_if_vrrp_id_ipv4_set(ritr_pl, vrrp_id);
6908 	else
6909 		mlxsw_reg_ritr_if_vrrp_id_ipv6_set(ritr_pl, vrrp_id);
6910 
6911 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6912 }
6913 
6914 static int mlxsw_sp_rif_macvlan_add(struct mlxsw_sp *mlxsw_sp,
6915 				    const struct net_device *macvlan_dev,
6916 				    struct netlink_ext_ack *extack)
6917 {
6918 	struct macvlan_dev *vlan = netdev_priv(macvlan_dev);
6919 	struct mlxsw_sp_rif *rif;
6920 	int err;
6921 
6922 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, vlan->lowerdev);
6923 	if (!rif) {
6924 		NL_SET_ERR_MSG_MOD(extack, "macvlan is only supported on top of router interfaces");
6925 		return -EOPNOTSUPP;
6926 	}
6927 
6928 	err = mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
6929 				  mlxsw_sp_fid_index(rif->fid), true);
6930 	if (err)
6931 		return err;
6932 
6933 	err = mlxsw_sp_rif_vrrp_op(mlxsw_sp, rif->rif_index,
6934 				   macvlan_dev->dev_addr, true);
6935 	if (err)
6936 		goto err_rif_vrrp_add;
6937 
6938 	/* Make sure the bridge driver does not have this MAC pointing at
6939 	 * some other port.
6940 	 */
6941 	if (rif->ops->fdb_del)
6942 		rif->ops->fdb_del(rif, macvlan_dev->dev_addr);
6943 
6944 	return 0;
6945 
6946 err_rif_vrrp_add:
6947 	mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
6948 			    mlxsw_sp_fid_index(rif->fid), false);
6949 	return err;
6950 }
6951 
6952 static void __mlxsw_sp_rif_macvlan_del(struct mlxsw_sp *mlxsw_sp,
6953 				       const struct net_device *macvlan_dev)
6954 {
6955 	struct macvlan_dev *vlan = netdev_priv(macvlan_dev);
6956 	struct mlxsw_sp_rif *rif;
6957 
6958 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, vlan->lowerdev);
6959 	/* If we do not have a RIF, then we already took care of
6960 	 * removing the macvlan's MAC during RIF deletion.
6961 	 */
6962 	if (!rif)
6963 		return;
6964 	mlxsw_sp_rif_vrrp_op(mlxsw_sp, rif->rif_index, macvlan_dev->dev_addr,
6965 			     false);
6966 	mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
6967 			    mlxsw_sp_fid_index(rif->fid), false);
6968 }
6969 
6970 void mlxsw_sp_rif_macvlan_del(struct mlxsw_sp *mlxsw_sp,
6971 			      const struct net_device *macvlan_dev)
6972 {
6973 	mutex_lock(&mlxsw_sp->router->lock);
6974 	__mlxsw_sp_rif_macvlan_del(mlxsw_sp, macvlan_dev);
6975 	mutex_unlock(&mlxsw_sp->router->lock);
6976 }
6977 
6978 static int mlxsw_sp_inetaddr_macvlan_event(struct mlxsw_sp *mlxsw_sp,
6979 					   struct net_device *macvlan_dev,
6980 					   unsigned long event,
6981 					   struct netlink_ext_ack *extack)
6982 {
6983 	switch (event) {
6984 	case NETDEV_UP:
6985 		return mlxsw_sp_rif_macvlan_add(mlxsw_sp, macvlan_dev, extack);
6986 	case NETDEV_DOWN:
6987 		__mlxsw_sp_rif_macvlan_del(mlxsw_sp, macvlan_dev);
6988 		break;
6989 	}
6990 
6991 	return 0;
6992 }
6993 
6994 static int mlxsw_sp_router_port_check_rif_addr(struct mlxsw_sp *mlxsw_sp,
6995 					       struct net_device *dev,
6996 					       const unsigned char *dev_addr,
6997 					       struct netlink_ext_ack *extack)
6998 {
6999 	struct mlxsw_sp_rif *rif;
7000 	int i;
7001 
7002 	/* A RIF is not created for macvlan netdevs. Their MAC is used to
7003 	 * populate the FDB
7004 	 */
7005 	if (netif_is_macvlan(dev) || netif_is_l3_master(dev))
7006 		return 0;
7007 
7008 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++) {
7009 		rif = mlxsw_sp->router->rifs[i];
7010 		if (rif && rif->ops &&
7011 		    rif->ops->type == MLXSW_SP_RIF_TYPE_IPIP_LB)
7012 			continue;
7013 		if (rif && rif->dev && rif->dev != dev &&
7014 		    !ether_addr_equal_masked(rif->dev->dev_addr, dev_addr,
7015 					     mlxsw_sp->mac_mask)) {
7016 			NL_SET_ERR_MSG_MOD(extack, "All router interface MAC addresses must have the same prefix");
7017 			return -EINVAL;
7018 		}
7019 	}
7020 
7021 	return 0;
7022 }
7023 
7024 static int __mlxsw_sp_inetaddr_event(struct mlxsw_sp *mlxsw_sp,
7025 				     struct net_device *dev,
7026 				     unsigned long event,
7027 				     struct netlink_ext_ack *extack)
7028 {
7029 	if (mlxsw_sp_port_dev_check(dev))
7030 		return mlxsw_sp_inetaddr_port_event(dev, event, extack);
7031 	else if (netif_is_lag_master(dev))
7032 		return mlxsw_sp_inetaddr_lag_event(dev, event, extack);
7033 	else if (netif_is_bridge_master(dev))
7034 		return mlxsw_sp_inetaddr_bridge_event(mlxsw_sp, dev, event,
7035 						      extack);
7036 	else if (is_vlan_dev(dev))
7037 		return mlxsw_sp_inetaddr_vlan_event(mlxsw_sp, dev, event,
7038 						    extack);
7039 	else if (netif_is_macvlan(dev))
7040 		return mlxsw_sp_inetaddr_macvlan_event(mlxsw_sp, dev, event,
7041 						       extack);
7042 	else
7043 		return 0;
7044 }
7045 
7046 static int mlxsw_sp_inetaddr_event(struct notifier_block *nb,
7047 				   unsigned long event, void *ptr)
7048 {
7049 	struct in_ifaddr *ifa = (struct in_ifaddr *) ptr;
7050 	struct net_device *dev = ifa->ifa_dev->dev;
7051 	struct mlxsw_sp_router *router;
7052 	struct mlxsw_sp_rif *rif;
7053 	int err = 0;
7054 
7055 	/* NETDEV_UP event is handled by mlxsw_sp_inetaddr_valid_event */
7056 	if (event == NETDEV_UP)
7057 		return NOTIFY_DONE;
7058 
7059 	router = container_of(nb, struct mlxsw_sp_router, inetaddr_nb);
7060 	mutex_lock(&router->lock);
7061 	rif = mlxsw_sp_rif_find_by_dev(router->mlxsw_sp, dev);
7062 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
7063 		goto out;
7064 
7065 	err = __mlxsw_sp_inetaddr_event(router->mlxsw_sp, dev, event, NULL);
7066 out:
7067 	mutex_unlock(&router->lock);
7068 	return notifier_from_errno(err);
7069 }
7070 
7071 int mlxsw_sp_inetaddr_valid_event(struct notifier_block *unused,
7072 				  unsigned long event, void *ptr)
7073 {
7074 	struct in_validator_info *ivi = (struct in_validator_info *) ptr;
7075 	struct net_device *dev = ivi->ivi_dev->dev;
7076 	struct mlxsw_sp *mlxsw_sp;
7077 	struct mlxsw_sp_rif *rif;
7078 	int err = 0;
7079 
7080 	mlxsw_sp = mlxsw_sp_lower_get(dev);
7081 	if (!mlxsw_sp)
7082 		return NOTIFY_DONE;
7083 
7084 	mutex_lock(&mlxsw_sp->router->lock);
7085 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7086 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
7087 		goto out;
7088 
7089 	err = mlxsw_sp_router_port_check_rif_addr(mlxsw_sp, dev, dev->dev_addr,
7090 						  ivi->extack);
7091 	if (err)
7092 		goto out;
7093 
7094 	err = __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, ivi->extack);
7095 out:
7096 	mutex_unlock(&mlxsw_sp->router->lock);
7097 	return notifier_from_errno(err);
7098 }
7099 
7100 struct mlxsw_sp_inet6addr_event_work {
7101 	struct work_struct work;
7102 	struct mlxsw_sp *mlxsw_sp;
7103 	struct net_device *dev;
7104 	unsigned long event;
7105 };
7106 
7107 static void mlxsw_sp_inet6addr_event_work(struct work_struct *work)
7108 {
7109 	struct mlxsw_sp_inet6addr_event_work *inet6addr_work =
7110 		container_of(work, struct mlxsw_sp_inet6addr_event_work, work);
7111 	struct mlxsw_sp *mlxsw_sp = inet6addr_work->mlxsw_sp;
7112 	struct net_device *dev = inet6addr_work->dev;
7113 	unsigned long event = inet6addr_work->event;
7114 	struct mlxsw_sp_rif *rif;
7115 
7116 	rtnl_lock();
7117 	mutex_lock(&mlxsw_sp->router->lock);
7118 
7119 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7120 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
7121 		goto out;
7122 
7123 	__mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, NULL);
7124 out:
7125 	mutex_unlock(&mlxsw_sp->router->lock);
7126 	rtnl_unlock();
7127 	dev_put(dev);
7128 	kfree(inet6addr_work);
7129 }
7130 
7131 /* Called with rcu_read_lock() */
7132 static int mlxsw_sp_inet6addr_event(struct notifier_block *nb,
7133 				    unsigned long event, void *ptr)
7134 {
7135 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *) ptr;
7136 	struct mlxsw_sp_inet6addr_event_work *inet6addr_work;
7137 	struct net_device *dev = if6->idev->dev;
7138 	struct mlxsw_sp_router *router;
7139 
7140 	/* NETDEV_UP event is handled by mlxsw_sp_inet6addr_valid_event */
7141 	if (event == NETDEV_UP)
7142 		return NOTIFY_DONE;
7143 
7144 	inet6addr_work = kzalloc(sizeof(*inet6addr_work), GFP_ATOMIC);
7145 	if (!inet6addr_work)
7146 		return NOTIFY_BAD;
7147 
7148 	router = container_of(nb, struct mlxsw_sp_router, inet6addr_nb);
7149 	INIT_WORK(&inet6addr_work->work, mlxsw_sp_inet6addr_event_work);
7150 	inet6addr_work->mlxsw_sp = router->mlxsw_sp;
7151 	inet6addr_work->dev = dev;
7152 	inet6addr_work->event = event;
7153 	dev_hold(dev);
7154 	mlxsw_core_schedule_work(&inet6addr_work->work);
7155 
7156 	return NOTIFY_DONE;
7157 }
7158 
7159 int mlxsw_sp_inet6addr_valid_event(struct notifier_block *unused,
7160 				   unsigned long event, void *ptr)
7161 {
7162 	struct in6_validator_info *i6vi = (struct in6_validator_info *) ptr;
7163 	struct net_device *dev = i6vi->i6vi_dev->dev;
7164 	struct mlxsw_sp *mlxsw_sp;
7165 	struct mlxsw_sp_rif *rif;
7166 	int err = 0;
7167 
7168 	mlxsw_sp = mlxsw_sp_lower_get(dev);
7169 	if (!mlxsw_sp)
7170 		return NOTIFY_DONE;
7171 
7172 	mutex_lock(&mlxsw_sp->router->lock);
7173 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7174 	if (!mlxsw_sp_rif_should_config(rif, dev, event))
7175 		goto out;
7176 
7177 	err = mlxsw_sp_router_port_check_rif_addr(mlxsw_sp, dev, dev->dev_addr,
7178 						  i6vi->extack);
7179 	if (err)
7180 		goto out;
7181 
7182 	err = __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, i6vi->extack);
7183 out:
7184 	mutex_unlock(&mlxsw_sp->router->lock);
7185 	return notifier_from_errno(err);
7186 }
7187 
7188 static int mlxsw_sp_rif_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
7189 			     const char *mac, int mtu)
7190 {
7191 	char ritr_pl[MLXSW_REG_RITR_LEN];
7192 	int err;
7193 
7194 	mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
7195 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7196 	if (err)
7197 		return err;
7198 
7199 	mlxsw_reg_ritr_mtu_set(ritr_pl, mtu);
7200 	mlxsw_reg_ritr_if_mac_memcpy_to(ritr_pl, mac);
7201 	mlxsw_reg_ritr_op_set(ritr_pl, MLXSW_REG_RITR_RIF_CREATE);
7202 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7203 }
7204 
7205 static int
7206 mlxsw_sp_router_port_change_event(struct mlxsw_sp *mlxsw_sp,
7207 				  struct mlxsw_sp_rif *rif)
7208 {
7209 	struct net_device *dev = rif->dev;
7210 	u16 fid_index;
7211 	int err;
7212 
7213 	fid_index = mlxsw_sp_fid_index(rif->fid);
7214 
7215 	err = mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, false);
7216 	if (err)
7217 		return err;
7218 
7219 	err = mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, dev->dev_addr,
7220 				dev->mtu);
7221 	if (err)
7222 		goto err_rif_edit;
7223 
7224 	err = mlxsw_sp_rif_fdb_op(mlxsw_sp, dev->dev_addr, fid_index, true);
7225 	if (err)
7226 		goto err_rif_fdb_op;
7227 
7228 	if (rif->mtu != dev->mtu) {
7229 		struct mlxsw_sp_vr *vr;
7230 		int i;
7231 
7232 		/* The RIF is relevant only to its mr_table instance, as unlike
7233 		 * unicast routing, in multicast routing a RIF cannot be shared
7234 		 * between several multicast routing tables.
7235 		 */
7236 		vr = &mlxsw_sp->router->vrs[rif->vr_id];
7237 		for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++)
7238 			mlxsw_sp_mr_rif_mtu_update(vr->mr_table[i],
7239 						   rif, dev->mtu);
7240 	}
7241 
7242 	ether_addr_copy(rif->addr, dev->dev_addr);
7243 	rif->mtu = dev->mtu;
7244 
7245 	netdev_dbg(dev, "Updated RIF=%d\n", rif->rif_index);
7246 
7247 	return 0;
7248 
7249 err_rif_fdb_op:
7250 	mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, rif->addr, rif->mtu);
7251 err_rif_edit:
7252 	mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, true);
7253 	return err;
7254 }
7255 
7256 static int mlxsw_sp_router_port_pre_changeaddr_event(struct mlxsw_sp_rif *rif,
7257 			    struct netdev_notifier_pre_changeaddr_info *info)
7258 {
7259 	struct netlink_ext_ack *extack;
7260 
7261 	extack = netdev_notifier_info_to_extack(&info->info);
7262 	return mlxsw_sp_router_port_check_rif_addr(rif->mlxsw_sp, rif->dev,
7263 						   info->dev_addr, extack);
7264 }
7265 
7266 int mlxsw_sp_netdevice_router_port_event(struct net_device *dev,
7267 					 unsigned long event, void *ptr)
7268 {
7269 	struct mlxsw_sp *mlxsw_sp;
7270 	struct mlxsw_sp_rif *rif;
7271 	int err = 0;
7272 
7273 	mlxsw_sp = mlxsw_sp_lower_get(dev);
7274 	if (!mlxsw_sp)
7275 		return 0;
7276 
7277 	mutex_lock(&mlxsw_sp->router->lock);
7278 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7279 	if (!rif)
7280 		goto out;
7281 
7282 	switch (event) {
7283 	case NETDEV_CHANGEMTU: /* fall through */
7284 	case NETDEV_CHANGEADDR:
7285 		err = mlxsw_sp_router_port_change_event(mlxsw_sp, rif);
7286 		break;
7287 	case NETDEV_PRE_CHANGEADDR:
7288 		err = mlxsw_sp_router_port_pre_changeaddr_event(rif, ptr);
7289 		break;
7290 	}
7291 
7292 out:
7293 	mutex_unlock(&mlxsw_sp->router->lock);
7294 	return err;
7295 }
7296 
7297 static int mlxsw_sp_port_vrf_join(struct mlxsw_sp *mlxsw_sp,
7298 				  struct net_device *l3_dev,
7299 				  struct netlink_ext_ack *extack)
7300 {
7301 	struct mlxsw_sp_rif *rif;
7302 
7303 	/* If netdev is already associated with a RIF, then we need to
7304 	 * destroy it and create a new one with the new virtual router ID.
7305 	 */
7306 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
7307 	if (rif)
7308 		__mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_DOWN,
7309 					  extack);
7310 
7311 	return __mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_UP, extack);
7312 }
7313 
7314 static void mlxsw_sp_port_vrf_leave(struct mlxsw_sp *mlxsw_sp,
7315 				    struct net_device *l3_dev)
7316 {
7317 	struct mlxsw_sp_rif *rif;
7318 
7319 	rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
7320 	if (!rif)
7321 		return;
7322 	__mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_DOWN, NULL);
7323 }
7324 
7325 int mlxsw_sp_netdevice_vrf_event(struct net_device *l3_dev, unsigned long event,
7326 				 struct netdev_notifier_changeupper_info *info)
7327 {
7328 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_lower_get(l3_dev);
7329 	int err = 0;
7330 
7331 	/* We do not create a RIF for a macvlan, but only use it to
7332 	 * direct more MAC addresses to the router.
7333 	 */
7334 	if (!mlxsw_sp || netif_is_macvlan(l3_dev))
7335 		return 0;
7336 
7337 	mutex_lock(&mlxsw_sp->router->lock);
7338 	switch (event) {
7339 	case NETDEV_PRECHANGEUPPER:
7340 		break;
7341 	case NETDEV_CHANGEUPPER:
7342 		if (info->linking) {
7343 			struct netlink_ext_ack *extack;
7344 
7345 			extack = netdev_notifier_info_to_extack(&info->info);
7346 			err = mlxsw_sp_port_vrf_join(mlxsw_sp, l3_dev, extack);
7347 		} else {
7348 			mlxsw_sp_port_vrf_leave(mlxsw_sp, l3_dev);
7349 		}
7350 		break;
7351 	}
7352 	mutex_unlock(&mlxsw_sp->router->lock);
7353 
7354 	return err;
7355 }
7356 
7357 static int __mlxsw_sp_rif_macvlan_flush(struct net_device *dev, void *data)
7358 {
7359 	struct mlxsw_sp_rif *rif = data;
7360 
7361 	if (!netif_is_macvlan(dev))
7362 		return 0;
7363 
7364 	return mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, dev->dev_addr,
7365 				   mlxsw_sp_fid_index(rif->fid), false);
7366 }
7367 
7368 static int mlxsw_sp_rif_macvlan_flush(struct mlxsw_sp_rif *rif)
7369 {
7370 	if (!netif_is_macvlan_port(rif->dev))
7371 		return 0;
7372 
7373 	netdev_warn(rif->dev, "Router interface is deleted. Upper macvlans will not work\n");
7374 	return netdev_walk_all_upper_dev_rcu(rif->dev,
7375 					     __mlxsw_sp_rif_macvlan_flush, rif);
7376 }
7377 
7378 static void mlxsw_sp_rif_subport_setup(struct mlxsw_sp_rif *rif,
7379 				       const struct mlxsw_sp_rif_params *params)
7380 {
7381 	struct mlxsw_sp_rif_subport *rif_subport;
7382 
7383 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
7384 	refcount_set(&rif_subport->ref_count, 1);
7385 	rif_subport->vid = params->vid;
7386 	rif_subport->lag = params->lag;
7387 	if (params->lag)
7388 		rif_subport->lag_id = params->lag_id;
7389 	else
7390 		rif_subport->system_port = params->system_port;
7391 }
7392 
7393 static int mlxsw_sp_rif_subport_op(struct mlxsw_sp_rif *rif, bool enable)
7394 {
7395 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7396 	struct mlxsw_sp_rif_subport *rif_subport;
7397 	char ritr_pl[MLXSW_REG_RITR_LEN];
7398 
7399 	rif_subport = mlxsw_sp_rif_subport_rif(rif);
7400 	mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_SP_IF,
7401 			    rif->rif_index, rif->vr_id, rif->dev->mtu);
7402 	mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
7403 	mlxsw_reg_ritr_sp_if_pack(ritr_pl, rif_subport->lag,
7404 				  rif_subport->lag ? rif_subport->lag_id :
7405 						     rif_subport->system_port,
7406 				  rif_subport->vid);
7407 
7408 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7409 }
7410 
7411 static int mlxsw_sp_rif_subport_configure(struct mlxsw_sp_rif *rif)
7412 {
7413 	int err;
7414 
7415 	err = mlxsw_sp_rif_subport_op(rif, true);
7416 	if (err)
7417 		return err;
7418 
7419 	err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7420 				  mlxsw_sp_fid_index(rif->fid), true);
7421 	if (err)
7422 		goto err_rif_fdb_op;
7423 
7424 	mlxsw_sp_fid_rif_set(rif->fid, rif);
7425 	return 0;
7426 
7427 err_rif_fdb_op:
7428 	mlxsw_sp_rif_subport_op(rif, false);
7429 	return err;
7430 }
7431 
7432 static void mlxsw_sp_rif_subport_deconfigure(struct mlxsw_sp_rif *rif)
7433 {
7434 	struct mlxsw_sp_fid *fid = rif->fid;
7435 
7436 	mlxsw_sp_fid_rif_set(fid, NULL);
7437 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7438 			    mlxsw_sp_fid_index(fid), false);
7439 	mlxsw_sp_rif_macvlan_flush(rif);
7440 	mlxsw_sp_rif_subport_op(rif, false);
7441 }
7442 
7443 static struct mlxsw_sp_fid *
7444 mlxsw_sp_rif_subport_fid_get(struct mlxsw_sp_rif *rif,
7445 			     struct netlink_ext_ack *extack)
7446 {
7447 	return mlxsw_sp_fid_rfid_get(rif->mlxsw_sp, rif->rif_index);
7448 }
7449 
7450 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_subport_ops = {
7451 	.type			= MLXSW_SP_RIF_TYPE_SUBPORT,
7452 	.rif_size		= sizeof(struct mlxsw_sp_rif_subport),
7453 	.setup			= mlxsw_sp_rif_subport_setup,
7454 	.configure		= mlxsw_sp_rif_subport_configure,
7455 	.deconfigure		= mlxsw_sp_rif_subport_deconfigure,
7456 	.fid_get		= mlxsw_sp_rif_subport_fid_get,
7457 };
7458 
7459 static int mlxsw_sp_rif_vlan_fid_op(struct mlxsw_sp_rif *rif,
7460 				    enum mlxsw_reg_ritr_if_type type,
7461 				    u16 vid_fid, bool enable)
7462 {
7463 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7464 	char ritr_pl[MLXSW_REG_RITR_LEN];
7465 
7466 	mlxsw_reg_ritr_pack(ritr_pl, enable, type, rif->rif_index, rif->vr_id,
7467 			    rif->dev->mtu);
7468 	mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
7469 	mlxsw_reg_ritr_fid_set(ritr_pl, type, vid_fid);
7470 
7471 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7472 }
7473 
7474 u8 mlxsw_sp_router_port(const struct mlxsw_sp *mlxsw_sp)
7475 {
7476 	return mlxsw_core_max_ports(mlxsw_sp->core) + 1;
7477 }
7478 
7479 static int mlxsw_sp_rif_fid_configure(struct mlxsw_sp_rif *rif)
7480 {
7481 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7482 	u16 fid_index = mlxsw_sp_fid_index(rif->fid);
7483 	int err;
7484 
7485 	err = mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index,
7486 				       true);
7487 	if (err)
7488 		return err;
7489 
7490 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
7491 				     mlxsw_sp_router_port(mlxsw_sp), true);
7492 	if (err)
7493 		goto err_fid_mc_flood_set;
7494 
7495 	err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
7496 				     mlxsw_sp_router_port(mlxsw_sp), true);
7497 	if (err)
7498 		goto err_fid_bc_flood_set;
7499 
7500 	err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7501 				  mlxsw_sp_fid_index(rif->fid), true);
7502 	if (err)
7503 		goto err_rif_fdb_op;
7504 
7505 	mlxsw_sp_fid_rif_set(rif->fid, rif);
7506 	return 0;
7507 
7508 err_rif_fdb_op:
7509 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
7510 			       mlxsw_sp_router_port(mlxsw_sp), false);
7511 err_fid_bc_flood_set:
7512 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
7513 			       mlxsw_sp_router_port(mlxsw_sp), false);
7514 err_fid_mc_flood_set:
7515 	mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index, false);
7516 	return err;
7517 }
7518 
7519 static void mlxsw_sp_rif_fid_deconfigure(struct mlxsw_sp_rif *rif)
7520 {
7521 	u16 fid_index = mlxsw_sp_fid_index(rif->fid);
7522 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7523 	struct mlxsw_sp_fid *fid = rif->fid;
7524 
7525 	mlxsw_sp_fid_rif_set(fid, NULL);
7526 	mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7527 			    mlxsw_sp_fid_index(fid), false);
7528 	mlxsw_sp_rif_macvlan_flush(rif);
7529 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
7530 			       mlxsw_sp_router_port(mlxsw_sp), false);
7531 	mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
7532 			       mlxsw_sp_router_port(mlxsw_sp), false);
7533 	mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index, false);
7534 }
7535 
7536 static struct mlxsw_sp_fid *
7537 mlxsw_sp_rif_fid_fid_get(struct mlxsw_sp_rif *rif,
7538 			 struct netlink_ext_ack *extack)
7539 {
7540 	return mlxsw_sp_fid_8021d_get(rif->mlxsw_sp, rif->dev->ifindex);
7541 }
7542 
7543 static void mlxsw_sp_rif_fid_fdb_del(struct mlxsw_sp_rif *rif, const char *mac)
7544 {
7545 	struct switchdev_notifier_fdb_info info;
7546 	struct net_device *dev;
7547 
7548 	dev = br_fdb_find_port(rif->dev, mac, 0);
7549 	if (!dev)
7550 		return;
7551 
7552 	info.addr = mac;
7553 	info.vid = 0;
7554 	call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_BRIDGE, dev, &info.info,
7555 				 NULL);
7556 }
7557 
7558 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_fid_ops = {
7559 	.type			= MLXSW_SP_RIF_TYPE_FID,
7560 	.rif_size		= sizeof(struct mlxsw_sp_rif),
7561 	.configure		= mlxsw_sp_rif_fid_configure,
7562 	.deconfigure		= mlxsw_sp_rif_fid_deconfigure,
7563 	.fid_get		= mlxsw_sp_rif_fid_fid_get,
7564 	.fdb_del		= mlxsw_sp_rif_fid_fdb_del,
7565 };
7566 
7567 static struct mlxsw_sp_fid *
7568 mlxsw_sp_rif_vlan_fid_get(struct mlxsw_sp_rif *rif,
7569 			  struct netlink_ext_ack *extack)
7570 {
7571 	struct net_device *br_dev = rif->dev;
7572 	u16 vid;
7573 	int err;
7574 
7575 	if (is_vlan_dev(rif->dev)) {
7576 		vid = vlan_dev_vlan_id(rif->dev);
7577 		br_dev = vlan_dev_real_dev(rif->dev);
7578 		if (WARN_ON(!netif_is_bridge_master(br_dev)))
7579 			return ERR_PTR(-EINVAL);
7580 	} else {
7581 		err = br_vlan_get_pvid(rif->dev, &vid);
7582 		if (err < 0 || !vid) {
7583 			NL_SET_ERR_MSG_MOD(extack, "Couldn't determine bridge PVID");
7584 			return ERR_PTR(-EINVAL);
7585 		}
7586 	}
7587 
7588 	return mlxsw_sp_fid_8021q_get(rif->mlxsw_sp, vid);
7589 }
7590 
7591 static void mlxsw_sp_rif_vlan_fdb_del(struct mlxsw_sp_rif *rif, const char *mac)
7592 {
7593 	u16 vid = mlxsw_sp_fid_8021q_vid(rif->fid);
7594 	struct switchdev_notifier_fdb_info info;
7595 	struct net_device *br_dev;
7596 	struct net_device *dev;
7597 
7598 	br_dev = is_vlan_dev(rif->dev) ? vlan_dev_real_dev(rif->dev) : rif->dev;
7599 	dev = br_fdb_find_port(br_dev, mac, vid);
7600 	if (!dev)
7601 		return;
7602 
7603 	info.addr = mac;
7604 	info.vid = vid;
7605 	call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_BRIDGE, dev, &info.info,
7606 				 NULL);
7607 }
7608 
7609 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_vlan_emu_ops = {
7610 	.type			= MLXSW_SP_RIF_TYPE_VLAN,
7611 	.rif_size		= sizeof(struct mlxsw_sp_rif),
7612 	.configure		= mlxsw_sp_rif_fid_configure,
7613 	.deconfigure		= mlxsw_sp_rif_fid_deconfigure,
7614 	.fid_get		= mlxsw_sp_rif_vlan_fid_get,
7615 	.fdb_del		= mlxsw_sp_rif_vlan_fdb_del,
7616 };
7617 
7618 static struct mlxsw_sp_rif_ipip_lb *
7619 mlxsw_sp_rif_ipip_lb_rif(struct mlxsw_sp_rif *rif)
7620 {
7621 	return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
7622 }
7623 
7624 static void
7625 mlxsw_sp_rif_ipip_lb_setup(struct mlxsw_sp_rif *rif,
7626 			   const struct mlxsw_sp_rif_params *params)
7627 {
7628 	struct mlxsw_sp_rif_params_ipip_lb *params_lb;
7629 	struct mlxsw_sp_rif_ipip_lb *rif_lb;
7630 
7631 	params_lb = container_of(params, struct mlxsw_sp_rif_params_ipip_lb,
7632 				 common);
7633 	rif_lb = mlxsw_sp_rif_ipip_lb_rif(rif);
7634 	rif_lb->lb_config = params_lb->lb_config;
7635 }
7636 
7637 static int
7638 mlxsw_sp1_rif_ipip_lb_configure(struct mlxsw_sp_rif *rif)
7639 {
7640 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
7641 	u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(rif->dev);
7642 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7643 	struct mlxsw_sp_vr *ul_vr;
7644 	int err;
7645 
7646 	ul_vr = mlxsw_sp_vr_get(mlxsw_sp, ul_tb_id, NULL);
7647 	if (IS_ERR(ul_vr))
7648 		return PTR_ERR(ul_vr);
7649 
7650 	err = mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr->id, 0, true);
7651 	if (err)
7652 		goto err_loopback_op;
7653 
7654 	lb_rif->ul_vr_id = ul_vr->id;
7655 	lb_rif->ul_rif_id = 0;
7656 	++ul_vr->rif_count;
7657 	return 0;
7658 
7659 err_loopback_op:
7660 	mlxsw_sp_vr_put(mlxsw_sp, ul_vr);
7661 	return err;
7662 }
7663 
7664 static void mlxsw_sp1_rif_ipip_lb_deconfigure(struct mlxsw_sp_rif *rif)
7665 {
7666 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
7667 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7668 	struct mlxsw_sp_vr *ul_vr;
7669 
7670 	ul_vr = &mlxsw_sp->router->vrs[lb_rif->ul_vr_id];
7671 	mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr->id, 0, false);
7672 
7673 	--ul_vr->rif_count;
7674 	mlxsw_sp_vr_put(mlxsw_sp, ul_vr);
7675 }
7676 
7677 static const struct mlxsw_sp_rif_ops mlxsw_sp1_rif_ipip_lb_ops = {
7678 	.type			= MLXSW_SP_RIF_TYPE_IPIP_LB,
7679 	.rif_size		= sizeof(struct mlxsw_sp_rif_ipip_lb),
7680 	.setup                  = mlxsw_sp_rif_ipip_lb_setup,
7681 	.configure		= mlxsw_sp1_rif_ipip_lb_configure,
7682 	.deconfigure		= mlxsw_sp1_rif_ipip_lb_deconfigure,
7683 };
7684 
7685 const struct mlxsw_sp_rif_ops *mlxsw_sp1_rif_ops_arr[] = {
7686 	[MLXSW_SP_RIF_TYPE_SUBPORT]	= &mlxsw_sp_rif_subport_ops,
7687 	[MLXSW_SP_RIF_TYPE_VLAN]	= &mlxsw_sp_rif_vlan_emu_ops,
7688 	[MLXSW_SP_RIF_TYPE_FID]		= &mlxsw_sp_rif_fid_ops,
7689 	[MLXSW_SP_RIF_TYPE_IPIP_LB]	= &mlxsw_sp1_rif_ipip_lb_ops,
7690 };
7691 
7692 static int
7693 mlxsw_sp_rif_ipip_lb_ul_rif_op(struct mlxsw_sp_rif *ul_rif, bool enable)
7694 {
7695 	struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
7696 	char ritr_pl[MLXSW_REG_RITR_LEN];
7697 
7698 	mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
7699 			    ul_rif->rif_index, ul_rif->vr_id, IP_MAX_MTU);
7700 	mlxsw_reg_ritr_loopback_protocol_set(ritr_pl,
7701 					     MLXSW_REG_RITR_LOOPBACK_GENERIC);
7702 
7703 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7704 }
7705 
7706 static struct mlxsw_sp_rif *
7707 mlxsw_sp_ul_rif_create(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr,
7708 		       struct netlink_ext_ack *extack)
7709 {
7710 	struct mlxsw_sp_rif *ul_rif;
7711 	u16 rif_index;
7712 	int err;
7713 
7714 	err = mlxsw_sp_rif_index_alloc(mlxsw_sp, &rif_index);
7715 	if (err) {
7716 		NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported router interfaces");
7717 		return ERR_PTR(err);
7718 	}
7719 
7720 	ul_rif = mlxsw_sp_rif_alloc(sizeof(*ul_rif), rif_index, vr->id, NULL);
7721 	if (!ul_rif)
7722 		return ERR_PTR(-ENOMEM);
7723 
7724 	mlxsw_sp->router->rifs[rif_index] = ul_rif;
7725 	ul_rif->mlxsw_sp = mlxsw_sp;
7726 	err = mlxsw_sp_rif_ipip_lb_ul_rif_op(ul_rif, true);
7727 	if (err)
7728 		goto ul_rif_op_err;
7729 
7730 	return ul_rif;
7731 
7732 ul_rif_op_err:
7733 	mlxsw_sp->router->rifs[rif_index] = NULL;
7734 	kfree(ul_rif);
7735 	return ERR_PTR(err);
7736 }
7737 
7738 static void mlxsw_sp_ul_rif_destroy(struct mlxsw_sp_rif *ul_rif)
7739 {
7740 	struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
7741 
7742 	mlxsw_sp_rif_ipip_lb_ul_rif_op(ul_rif, false);
7743 	mlxsw_sp->router->rifs[ul_rif->rif_index] = NULL;
7744 	kfree(ul_rif);
7745 }
7746 
7747 static struct mlxsw_sp_rif *
7748 mlxsw_sp_ul_rif_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
7749 		    struct netlink_ext_ack *extack)
7750 {
7751 	struct mlxsw_sp_vr *vr;
7752 	int err;
7753 
7754 	vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id, extack);
7755 	if (IS_ERR(vr))
7756 		return ERR_CAST(vr);
7757 
7758 	if (refcount_inc_not_zero(&vr->ul_rif_refcnt))
7759 		return vr->ul_rif;
7760 
7761 	vr->ul_rif = mlxsw_sp_ul_rif_create(mlxsw_sp, vr, extack);
7762 	if (IS_ERR(vr->ul_rif)) {
7763 		err = PTR_ERR(vr->ul_rif);
7764 		goto err_ul_rif_create;
7765 	}
7766 
7767 	vr->rif_count++;
7768 	refcount_set(&vr->ul_rif_refcnt, 1);
7769 
7770 	return vr->ul_rif;
7771 
7772 err_ul_rif_create:
7773 	mlxsw_sp_vr_put(mlxsw_sp, vr);
7774 	return ERR_PTR(err);
7775 }
7776 
7777 static void mlxsw_sp_ul_rif_put(struct mlxsw_sp_rif *ul_rif)
7778 {
7779 	struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
7780 	struct mlxsw_sp_vr *vr;
7781 
7782 	vr = &mlxsw_sp->router->vrs[ul_rif->vr_id];
7783 
7784 	if (!refcount_dec_and_test(&vr->ul_rif_refcnt))
7785 		return;
7786 
7787 	vr->rif_count--;
7788 	mlxsw_sp_ul_rif_destroy(ul_rif);
7789 	mlxsw_sp_vr_put(mlxsw_sp, vr);
7790 }
7791 
7792 int mlxsw_sp_router_ul_rif_get(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
7793 			       u16 *ul_rif_index)
7794 {
7795 	struct mlxsw_sp_rif *ul_rif;
7796 	int err = 0;
7797 
7798 	mutex_lock(&mlxsw_sp->router->lock);
7799 	ul_rif = mlxsw_sp_ul_rif_get(mlxsw_sp, ul_tb_id, NULL);
7800 	if (IS_ERR(ul_rif)) {
7801 		err = PTR_ERR(ul_rif);
7802 		goto out;
7803 	}
7804 	*ul_rif_index = ul_rif->rif_index;
7805 out:
7806 	mutex_unlock(&mlxsw_sp->router->lock);
7807 	return err;
7808 }
7809 
7810 void mlxsw_sp_router_ul_rif_put(struct mlxsw_sp *mlxsw_sp, u16 ul_rif_index)
7811 {
7812 	struct mlxsw_sp_rif *ul_rif;
7813 
7814 	mutex_lock(&mlxsw_sp->router->lock);
7815 	ul_rif = mlxsw_sp->router->rifs[ul_rif_index];
7816 	if (WARN_ON(!ul_rif))
7817 		goto out;
7818 
7819 	mlxsw_sp_ul_rif_put(ul_rif);
7820 out:
7821 	mutex_unlock(&mlxsw_sp->router->lock);
7822 }
7823 
7824 static int
7825 mlxsw_sp2_rif_ipip_lb_configure(struct mlxsw_sp_rif *rif)
7826 {
7827 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
7828 	u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(rif->dev);
7829 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7830 	struct mlxsw_sp_rif *ul_rif;
7831 	int err;
7832 
7833 	ul_rif = mlxsw_sp_ul_rif_get(mlxsw_sp, ul_tb_id, NULL);
7834 	if (IS_ERR(ul_rif))
7835 		return PTR_ERR(ul_rif);
7836 
7837 	err = mlxsw_sp_rif_ipip_lb_op(lb_rif, 0, ul_rif->rif_index, true);
7838 	if (err)
7839 		goto err_loopback_op;
7840 
7841 	lb_rif->ul_vr_id = 0;
7842 	lb_rif->ul_rif_id = ul_rif->rif_index;
7843 
7844 	return 0;
7845 
7846 err_loopback_op:
7847 	mlxsw_sp_ul_rif_put(ul_rif);
7848 	return err;
7849 }
7850 
7851 static void mlxsw_sp2_rif_ipip_lb_deconfigure(struct mlxsw_sp_rif *rif)
7852 {
7853 	struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
7854 	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7855 	struct mlxsw_sp_rif *ul_rif;
7856 
7857 	ul_rif = mlxsw_sp_rif_by_index(mlxsw_sp, lb_rif->ul_rif_id);
7858 	mlxsw_sp_rif_ipip_lb_op(lb_rif, 0, lb_rif->ul_rif_id, false);
7859 	mlxsw_sp_ul_rif_put(ul_rif);
7860 }
7861 
7862 static const struct mlxsw_sp_rif_ops mlxsw_sp2_rif_ipip_lb_ops = {
7863 	.type			= MLXSW_SP_RIF_TYPE_IPIP_LB,
7864 	.rif_size		= sizeof(struct mlxsw_sp_rif_ipip_lb),
7865 	.setup                  = mlxsw_sp_rif_ipip_lb_setup,
7866 	.configure		= mlxsw_sp2_rif_ipip_lb_configure,
7867 	.deconfigure		= mlxsw_sp2_rif_ipip_lb_deconfigure,
7868 };
7869 
7870 const struct mlxsw_sp_rif_ops *mlxsw_sp2_rif_ops_arr[] = {
7871 	[MLXSW_SP_RIF_TYPE_SUBPORT]	= &mlxsw_sp_rif_subport_ops,
7872 	[MLXSW_SP_RIF_TYPE_VLAN]	= &mlxsw_sp_rif_vlan_emu_ops,
7873 	[MLXSW_SP_RIF_TYPE_FID]		= &mlxsw_sp_rif_fid_ops,
7874 	[MLXSW_SP_RIF_TYPE_IPIP_LB]	= &mlxsw_sp2_rif_ipip_lb_ops,
7875 };
7876 
7877 static int mlxsw_sp_rifs_init(struct mlxsw_sp *mlxsw_sp)
7878 {
7879 	u64 max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
7880 
7881 	mlxsw_sp->router->rifs = kcalloc(max_rifs,
7882 					 sizeof(struct mlxsw_sp_rif *),
7883 					 GFP_KERNEL);
7884 	if (!mlxsw_sp->router->rifs)
7885 		return -ENOMEM;
7886 
7887 	return 0;
7888 }
7889 
7890 static void mlxsw_sp_rifs_fini(struct mlxsw_sp *mlxsw_sp)
7891 {
7892 	int i;
7893 
7894 	for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
7895 		WARN_ON_ONCE(mlxsw_sp->router->rifs[i]);
7896 
7897 	kfree(mlxsw_sp->router->rifs);
7898 }
7899 
7900 static int
7901 mlxsw_sp_ipip_config_tigcr(struct mlxsw_sp *mlxsw_sp)
7902 {
7903 	char tigcr_pl[MLXSW_REG_TIGCR_LEN];
7904 
7905 	mlxsw_reg_tigcr_pack(tigcr_pl, true, 0);
7906 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tigcr), tigcr_pl);
7907 }
7908 
7909 static int mlxsw_sp_ipips_init(struct mlxsw_sp *mlxsw_sp)
7910 {
7911 	int err;
7912 
7913 	mlxsw_sp->router->ipip_ops_arr = mlxsw_sp_ipip_ops_arr;
7914 	INIT_LIST_HEAD(&mlxsw_sp->router->ipip_list);
7915 
7916 	err = mlxsw_sp_ipip_ecn_encap_init(mlxsw_sp);
7917 	if (err)
7918 		return err;
7919 	err = mlxsw_sp_ipip_ecn_decap_init(mlxsw_sp);
7920 	if (err)
7921 		return err;
7922 
7923 	return mlxsw_sp_ipip_config_tigcr(mlxsw_sp);
7924 }
7925 
7926 static void mlxsw_sp_ipips_fini(struct mlxsw_sp *mlxsw_sp)
7927 {
7928 	WARN_ON(!list_empty(&mlxsw_sp->router->ipip_list));
7929 }
7930 
7931 static void mlxsw_sp_router_fib_dump_flush(struct notifier_block *nb)
7932 {
7933 	struct mlxsw_sp_router *router;
7934 
7935 	/* Flush pending FIB notifications and then flush the device's
7936 	 * table before requesting another dump. The FIB notification
7937 	 * block is unregistered, so no need to take RTNL.
7938 	 */
7939 	mlxsw_core_flush_owq();
7940 	router = container_of(nb, struct mlxsw_sp_router, fib_nb);
7941 	mlxsw_sp_router_fib_flush(router->mlxsw_sp);
7942 }
7943 
7944 #ifdef CONFIG_IP_ROUTE_MULTIPATH
7945 static void mlxsw_sp_mp_hash_header_set(char *recr2_pl, int header)
7946 {
7947 	mlxsw_reg_recr2_outer_header_enables_set(recr2_pl, header, true);
7948 }
7949 
7950 static void mlxsw_sp_mp_hash_field_set(char *recr2_pl, int field)
7951 {
7952 	mlxsw_reg_recr2_outer_header_fields_enable_set(recr2_pl, field, true);
7953 }
7954 
7955 static void mlxsw_sp_mp4_hash_init(struct mlxsw_sp *mlxsw_sp, char *recr2_pl)
7956 {
7957 	struct net *net = mlxsw_sp_net(mlxsw_sp);
7958 	bool only_l3 = !net->ipv4.sysctl_fib_multipath_hash_policy;
7959 
7960 	mlxsw_sp_mp_hash_header_set(recr2_pl,
7961 				    MLXSW_REG_RECR2_IPV4_EN_NOT_TCP_NOT_UDP);
7962 	mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_IPV4_EN_TCP_UDP);
7963 	mlxsw_reg_recr2_ipv4_sip_enable(recr2_pl);
7964 	mlxsw_reg_recr2_ipv4_dip_enable(recr2_pl);
7965 	if (only_l3)
7966 		return;
7967 	mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_EN_IPV4);
7968 	mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_IPV4_PROTOCOL);
7969 	mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_SPORT);
7970 	mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_DPORT);
7971 }
7972 
7973 static void mlxsw_sp_mp6_hash_init(struct mlxsw_sp *mlxsw_sp, char *recr2_pl)
7974 {
7975 	bool only_l3 = !ip6_multipath_hash_policy(mlxsw_sp_net(mlxsw_sp));
7976 
7977 	mlxsw_sp_mp_hash_header_set(recr2_pl,
7978 				    MLXSW_REG_RECR2_IPV6_EN_NOT_TCP_NOT_UDP);
7979 	mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_IPV6_EN_TCP_UDP);
7980 	mlxsw_reg_recr2_ipv6_sip_enable(recr2_pl);
7981 	mlxsw_reg_recr2_ipv6_dip_enable(recr2_pl);
7982 	mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_IPV6_NEXT_HEADER);
7983 	if (only_l3) {
7984 		mlxsw_sp_mp_hash_field_set(recr2_pl,
7985 					   MLXSW_REG_RECR2_IPV6_FLOW_LABEL);
7986 	} else {
7987 		mlxsw_sp_mp_hash_header_set(recr2_pl,
7988 					    MLXSW_REG_RECR2_TCP_UDP_EN_IPV6);
7989 		mlxsw_sp_mp_hash_field_set(recr2_pl,
7990 					   MLXSW_REG_RECR2_TCP_UDP_SPORT);
7991 		mlxsw_sp_mp_hash_field_set(recr2_pl,
7992 					   MLXSW_REG_RECR2_TCP_UDP_DPORT);
7993 	}
7994 }
7995 
7996 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
7997 {
7998 	char recr2_pl[MLXSW_REG_RECR2_LEN];
7999 	u32 seed;
8000 
8001 	seed = jhash(mlxsw_sp->base_mac, sizeof(mlxsw_sp->base_mac), 0);
8002 	mlxsw_reg_recr2_pack(recr2_pl, seed);
8003 	mlxsw_sp_mp4_hash_init(mlxsw_sp, recr2_pl);
8004 	mlxsw_sp_mp6_hash_init(mlxsw_sp, recr2_pl);
8005 
8006 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(recr2), recr2_pl);
8007 }
8008 #else
8009 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
8010 {
8011 	return 0;
8012 }
8013 #endif
8014 
8015 static int mlxsw_sp_dscp_init(struct mlxsw_sp *mlxsw_sp)
8016 {
8017 	char rdpm_pl[MLXSW_REG_RDPM_LEN];
8018 	unsigned int i;
8019 
8020 	MLXSW_REG_ZERO(rdpm, rdpm_pl);
8021 
8022 	/* HW is determining switch priority based on DSCP-bits, but the
8023 	 * kernel is still doing that based on the ToS. Since there's a
8024 	 * mismatch in bits we need to make sure to translate the right
8025 	 * value ToS would observe, skipping the 2 least-significant ECN bits.
8026 	 */
8027 	for (i = 0; i < MLXSW_REG_RDPM_DSCP_ENTRY_REC_MAX_COUNT; i++)
8028 		mlxsw_reg_rdpm_pack(rdpm_pl, i, rt_tos2priority(i << 2));
8029 
8030 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rdpm), rdpm_pl);
8031 }
8032 
8033 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
8034 {
8035 	struct net *net = mlxsw_sp_net(mlxsw_sp);
8036 	bool usp = net->ipv4.sysctl_ip_fwd_update_priority;
8037 	char rgcr_pl[MLXSW_REG_RGCR_LEN];
8038 	u64 max_rifs;
8039 	int err;
8040 
8041 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_RIFS))
8042 		return -EIO;
8043 	max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
8044 
8045 	mlxsw_reg_rgcr_pack(rgcr_pl, true, true);
8046 	mlxsw_reg_rgcr_max_router_interfaces_set(rgcr_pl, max_rifs);
8047 	mlxsw_reg_rgcr_usp_set(rgcr_pl, usp);
8048 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
8049 	if (err)
8050 		return err;
8051 	return 0;
8052 }
8053 
8054 static void __mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
8055 {
8056 	char rgcr_pl[MLXSW_REG_RGCR_LEN];
8057 
8058 	mlxsw_reg_rgcr_pack(rgcr_pl, false, false);
8059 	mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
8060 }
8061 
8062 int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp,
8063 			 struct netlink_ext_ack *extack)
8064 {
8065 	struct mlxsw_sp_router *router;
8066 	int err;
8067 
8068 	router = kzalloc(sizeof(*mlxsw_sp->router), GFP_KERNEL);
8069 	if (!router)
8070 		return -ENOMEM;
8071 	mutex_init(&router->lock);
8072 	mlxsw_sp->router = router;
8073 	router->mlxsw_sp = mlxsw_sp;
8074 
8075 	router->inetaddr_nb.notifier_call = mlxsw_sp_inetaddr_event;
8076 	err = register_inetaddr_notifier(&router->inetaddr_nb);
8077 	if (err)
8078 		goto err_register_inetaddr_notifier;
8079 
8080 	router->inet6addr_nb.notifier_call = mlxsw_sp_inet6addr_event;
8081 	err = register_inet6addr_notifier(&router->inet6addr_nb);
8082 	if (err)
8083 		goto err_register_inet6addr_notifier;
8084 
8085 	INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_neighs_list);
8086 	err = __mlxsw_sp_router_init(mlxsw_sp);
8087 	if (err)
8088 		goto err_router_init;
8089 
8090 	err = mlxsw_sp_rifs_init(mlxsw_sp);
8091 	if (err)
8092 		goto err_rifs_init;
8093 
8094 	err = mlxsw_sp_ipips_init(mlxsw_sp);
8095 	if (err)
8096 		goto err_ipips_init;
8097 
8098 	err = rhashtable_init(&mlxsw_sp->router->nexthop_ht,
8099 			      &mlxsw_sp_nexthop_ht_params);
8100 	if (err)
8101 		goto err_nexthop_ht_init;
8102 
8103 	err = rhashtable_init(&mlxsw_sp->router->nexthop_group_ht,
8104 			      &mlxsw_sp_nexthop_group_ht_params);
8105 	if (err)
8106 		goto err_nexthop_group_ht_init;
8107 
8108 	INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_list);
8109 	err = mlxsw_sp_lpm_init(mlxsw_sp);
8110 	if (err)
8111 		goto err_lpm_init;
8112 
8113 	err = mlxsw_sp_mr_init(mlxsw_sp, &mlxsw_sp_mr_tcam_ops);
8114 	if (err)
8115 		goto err_mr_init;
8116 
8117 	err = mlxsw_sp_vrs_init(mlxsw_sp);
8118 	if (err)
8119 		goto err_vrs_init;
8120 
8121 	err = mlxsw_sp_neigh_init(mlxsw_sp);
8122 	if (err)
8123 		goto err_neigh_init;
8124 
8125 	mlxsw_sp->router->netevent_nb.notifier_call =
8126 		mlxsw_sp_router_netevent_event;
8127 	err = register_netevent_notifier(&mlxsw_sp->router->netevent_nb);
8128 	if (err)
8129 		goto err_register_netevent_notifier;
8130 
8131 	err = mlxsw_sp_mp_hash_init(mlxsw_sp);
8132 	if (err)
8133 		goto err_mp_hash_init;
8134 
8135 	err = mlxsw_sp_dscp_init(mlxsw_sp);
8136 	if (err)
8137 		goto err_dscp_init;
8138 
8139 	mlxsw_sp->router->fib_nb.notifier_call = mlxsw_sp_router_fib_event;
8140 	err = register_fib_notifier(mlxsw_sp_net(mlxsw_sp),
8141 				    &mlxsw_sp->router->fib_nb,
8142 				    mlxsw_sp_router_fib_dump_flush, extack);
8143 	if (err)
8144 		goto err_register_fib_notifier;
8145 
8146 	return 0;
8147 
8148 err_register_fib_notifier:
8149 err_dscp_init:
8150 err_mp_hash_init:
8151 	unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
8152 err_register_netevent_notifier:
8153 	mlxsw_sp_neigh_fini(mlxsw_sp);
8154 err_neigh_init:
8155 	mlxsw_sp_vrs_fini(mlxsw_sp);
8156 err_vrs_init:
8157 	mlxsw_sp_mr_fini(mlxsw_sp);
8158 err_mr_init:
8159 	mlxsw_sp_lpm_fini(mlxsw_sp);
8160 err_lpm_init:
8161 	rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
8162 err_nexthop_group_ht_init:
8163 	rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
8164 err_nexthop_ht_init:
8165 	mlxsw_sp_ipips_fini(mlxsw_sp);
8166 err_ipips_init:
8167 	mlxsw_sp_rifs_fini(mlxsw_sp);
8168 err_rifs_init:
8169 	__mlxsw_sp_router_fini(mlxsw_sp);
8170 err_router_init:
8171 	unregister_inet6addr_notifier(&router->inet6addr_nb);
8172 err_register_inet6addr_notifier:
8173 	unregister_inetaddr_notifier(&router->inetaddr_nb);
8174 err_register_inetaddr_notifier:
8175 	mutex_destroy(&mlxsw_sp->router->lock);
8176 	kfree(mlxsw_sp->router);
8177 	return err;
8178 }
8179 
8180 void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
8181 {
8182 	unregister_fib_notifier(mlxsw_sp_net(mlxsw_sp),
8183 				&mlxsw_sp->router->fib_nb);
8184 	unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
8185 	mlxsw_sp_neigh_fini(mlxsw_sp);
8186 	mlxsw_sp_vrs_fini(mlxsw_sp);
8187 	mlxsw_sp_mr_fini(mlxsw_sp);
8188 	mlxsw_sp_lpm_fini(mlxsw_sp);
8189 	rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
8190 	rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
8191 	mlxsw_sp_ipips_fini(mlxsw_sp);
8192 	mlxsw_sp_rifs_fini(mlxsw_sp);
8193 	__mlxsw_sp_router_fini(mlxsw_sp);
8194 	unregister_inet6addr_notifier(&mlxsw_sp->router->inet6addr_nb);
8195 	unregister_inetaddr_notifier(&mlxsw_sp->router->inetaddr_nb);
8196 	mutex_destroy(&mlxsw_sp->router->lock);
8197 	kfree(mlxsw_sp->router);
8198 }
8199