1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
2 /* Copyright (c) 2018 Mellanox Technologies. All rights reserved */
3 
4 #include <linux/err.h>
5 #include <linux/gfp.h>
6 #include <linux/kernel.h>
7 #include <linux/list.h>
8 #include <linux/netlink.h>
9 #include <linux/rtnetlink.h>
10 #include <linux/slab.h>
11 #include <net/inet_ecn.h>
12 #include <net/ipv6.h>
13 
14 #include "reg.h"
15 #include "spectrum.h"
16 #include "spectrum_nve.h"
17 
18 const struct mlxsw_sp_nve_ops *mlxsw_sp1_nve_ops_arr[] = {
19 	[MLXSW_SP_NVE_TYPE_VXLAN]	= &mlxsw_sp1_nve_vxlan_ops,
20 };
21 
22 const struct mlxsw_sp_nve_ops *mlxsw_sp2_nve_ops_arr[] = {
23 	[MLXSW_SP_NVE_TYPE_VXLAN]	= &mlxsw_sp2_nve_vxlan_ops,
24 };
25 
26 struct mlxsw_sp_nve_mc_entry;
27 struct mlxsw_sp_nve_mc_record;
28 struct mlxsw_sp_nve_mc_list;
29 
30 struct mlxsw_sp_nve_mc_record_ops {
31 	enum mlxsw_reg_tnumt_record_type type;
32 	int (*entry_add)(struct mlxsw_sp_nve_mc_record *mc_record,
33 			 struct mlxsw_sp_nve_mc_entry *mc_entry,
34 			 const union mlxsw_sp_l3addr *addr);
35 	void (*entry_del)(const struct mlxsw_sp_nve_mc_record *mc_record,
36 			  const struct mlxsw_sp_nve_mc_entry *mc_entry);
37 	void (*entry_set)(const struct mlxsw_sp_nve_mc_record *mc_record,
38 			  const struct mlxsw_sp_nve_mc_entry *mc_entry,
39 			  char *tnumt_pl, unsigned int entry_index);
40 	bool (*entry_compare)(const struct mlxsw_sp_nve_mc_record *mc_record,
41 			      const struct mlxsw_sp_nve_mc_entry *mc_entry,
42 			      const union mlxsw_sp_l3addr *addr);
43 };
44 
45 struct mlxsw_sp_nve_mc_list_key {
46 	u16 fid_index;
47 };
48 
49 struct mlxsw_sp_nve_mc_ipv6_entry {
50 	struct in6_addr addr6;
51 	u32 addr6_kvdl_index;
52 };
53 
54 struct mlxsw_sp_nve_mc_entry {
55 	union {
56 		__be32 addr4;
57 		struct mlxsw_sp_nve_mc_ipv6_entry ipv6_entry;
58 	};
59 	u8 valid:1;
60 };
61 
62 struct mlxsw_sp_nve_mc_record {
63 	struct list_head list;
64 	enum mlxsw_sp_l3proto proto;
65 	unsigned int num_entries;
66 	struct mlxsw_sp *mlxsw_sp;
67 	struct mlxsw_sp_nve_mc_list *mc_list;
68 	const struct mlxsw_sp_nve_mc_record_ops *ops;
69 	u32 kvdl_index;
70 	struct mlxsw_sp_nve_mc_entry entries[];
71 };
72 
73 struct mlxsw_sp_nve_mc_list {
74 	struct list_head records_list;
75 	struct rhash_head ht_node;
76 	struct mlxsw_sp_nve_mc_list_key key;
77 };
78 
79 static const struct rhashtable_params mlxsw_sp_nve_mc_list_ht_params = {
80 	.key_len = sizeof(struct mlxsw_sp_nve_mc_list_key),
81 	.key_offset = offsetof(struct mlxsw_sp_nve_mc_list, key),
82 	.head_offset = offsetof(struct mlxsw_sp_nve_mc_list, ht_node),
83 };
84 
85 static int
86 mlxsw_sp_nve_mc_record_ipv4_entry_add(struct mlxsw_sp_nve_mc_record *mc_record,
87 				      struct mlxsw_sp_nve_mc_entry *mc_entry,
88 				      const union mlxsw_sp_l3addr *addr)
89 {
90 	mc_entry->addr4 = addr->addr4;
91 
92 	return 0;
93 }
94 
95 static void
96 mlxsw_sp_nve_mc_record_ipv4_entry_del(const struct mlxsw_sp_nve_mc_record *mc_record,
97 				      const struct mlxsw_sp_nve_mc_entry *mc_entry)
98 {
99 }
100 
101 static void
102 mlxsw_sp_nve_mc_record_ipv4_entry_set(const struct mlxsw_sp_nve_mc_record *mc_record,
103 				      const struct mlxsw_sp_nve_mc_entry *mc_entry,
104 				      char *tnumt_pl, unsigned int entry_index)
105 {
106 	u32 udip = be32_to_cpu(mc_entry->addr4);
107 
108 	mlxsw_reg_tnumt_udip_set(tnumt_pl, entry_index, udip);
109 }
110 
111 static bool
112 mlxsw_sp_nve_mc_record_ipv4_entry_compare(const struct mlxsw_sp_nve_mc_record *mc_record,
113 					  const struct mlxsw_sp_nve_mc_entry *mc_entry,
114 					  const union mlxsw_sp_l3addr *addr)
115 {
116 	return mc_entry->addr4 == addr->addr4;
117 }
118 
119 static const struct mlxsw_sp_nve_mc_record_ops
120 mlxsw_sp_nve_mc_record_ipv4_ops = {
121 	.type		= MLXSW_REG_TNUMT_RECORD_TYPE_IPV4,
122 	.entry_add	= &mlxsw_sp_nve_mc_record_ipv4_entry_add,
123 	.entry_del	= &mlxsw_sp_nve_mc_record_ipv4_entry_del,
124 	.entry_set	= &mlxsw_sp_nve_mc_record_ipv4_entry_set,
125 	.entry_compare	= &mlxsw_sp_nve_mc_record_ipv4_entry_compare,
126 };
127 
128 static int
129 mlxsw_sp_nve_mc_record_ipv6_entry_add(struct mlxsw_sp_nve_mc_record *mc_record,
130 				      struct mlxsw_sp_nve_mc_entry *mc_entry,
131 				      const union mlxsw_sp_l3addr *addr)
132 {
133 	WARN_ON(1);
134 
135 	return -EINVAL;
136 }
137 
138 static void
139 mlxsw_sp_nve_mc_record_ipv6_entry_del(const struct mlxsw_sp_nve_mc_record *mc_record,
140 				      const struct mlxsw_sp_nve_mc_entry *mc_entry)
141 {
142 }
143 
144 static void
145 mlxsw_sp_nve_mc_record_ipv6_entry_set(const struct mlxsw_sp_nve_mc_record *mc_record,
146 				      const struct mlxsw_sp_nve_mc_entry *mc_entry,
147 				      char *tnumt_pl, unsigned int entry_index)
148 {
149 	u32 udip_ptr = mc_entry->ipv6_entry.addr6_kvdl_index;
150 
151 	mlxsw_reg_tnumt_udip_ptr_set(tnumt_pl, entry_index, udip_ptr);
152 }
153 
154 static bool
155 mlxsw_sp_nve_mc_record_ipv6_entry_compare(const struct mlxsw_sp_nve_mc_record *mc_record,
156 					  const struct mlxsw_sp_nve_mc_entry *mc_entry,
157 					  const union mlxsw_sp_l3addr *addr)
158 {
159 	return ipv6_addr_equal(&mc_entry->ipv6_entry.addr6, &addr->addr6);
160 }
161 
162 static const struct mlxsw_sp_nve_mc_record_ops
163 mlxsw_sp_nve_mc_record_ipv6_ops = {
164 	.type		= MLXSW_REG_TNUMT_RECORD_TYPE_IPV6,
165 	.entry_add	= &mlxsw_sp_nve_mc_record_ipv6_entry_add,
166 	.entry_del	= &mlxsw_sp_nve_mc_record_ipv6_entry_del,
167 	.entry_set	= &mlxsw_sp_nve_mc_record_ipv6_entry_set,
168 	.entry_compare	= &mlxsw_sp_nve_mc_record_ipv6_entry_compare,
169 };
170 
171 static const struct mlxsw_sp_nve_mc_record_ops *
172 mlxsw_sp_nve_mc_record_ops_arr[] = {
173 	[MLXSW_SP_L3_PROTO_IPV4] = &mlxsw_sp_nve_mc_record_ipv4_ops,
174 	[MLXSW_SP_L3_PROTO_IPV6] = &mlxsw_sp_nve_mc_record_ipv6_ops,
175 };
176 
177 int mlxsw_sp_nve_learned_ip_resolve(struct mlxsw_sp *mlxsw_sp, u32 uip,
178 				    enum mlxsw_sp_l3proto proto,
179 				    union mlxsw_sp_l3addr *addr)
180 {
181 	switch (proto) {
182 	case MLXSW_SP_L3_PROTO_IPV4:
183 		addr->addr4 = cpu_to_be32(uip);
184 		return 0;
185 	default:
186 		WARN_ON(1);
187 		return -EINVAL;
188 	}
189 }
190 
191 static struct mlxsw_sp_nve_mc_list *
192 mlxsw_sp_nve_mc_list_find(struct mlxsw_sp *mlxsw_sp,
193 			  const struct mlxsw_sp_nve_mc_list_key *key)
194 {
195 	struct mlxsw_sp_nve *nve = mlxsw_sp->nve;
196 
197 	return rhashtable_lookup_fast(&nve->mc_list_ht, key,
198 				      mlxsw_sp_nve_mc_list_ht_params);
199 }
200 
201 static struct mlxsw_sp_nve_mc_list *
202 mlxsw_sp_nve_mc_list_create(struct mlxsw_sp *mlxsw_sp,
203 			    const struct mlxsw_sp_nve_mc_list_key *key)
204 {
205 	struct mlxsw_sp_nve *nve = mlxsw_sp->nve;
206 	struct mlxsw_sp_nve_mc_list *mc_list;
207 	int err;
208 
209 	mc_list = kmalloc(sizeof(*mc_list), GFP_KERNEL);
210 	if (!mc_list)
211 		return ERR_PTR(-ENOMEM);
212 
213 	INIT_LIST_HEAD(&mc_list->records_list);
214 	mc_list->key = *key;
215 
216 	err = rhashtable_insert_fast(&nve->mc_list_ht, &mc_list->ht_node,
217 				     mlxsw_sp_nve_mc_list_ht_params);
218 	if (err)
219 		goto err_rhashtable_insert;
220 
221 	return mc_list;
222 
223 err_rhashtable_insert:
224 	kfree(mc_list);
225 	return ERR_PTR(err);
226 }
227 
228 static void mlxsw_sp_nve_mc_list_destroy(struct mlxsw_sp *mlxsw_sp,
229 					 struct mlxsw_sp_nve_mc_list *mc_list)
230 {
231 	struct mlxsw_sp_nve *nve = mlxsw_sp->nve;
232 
233 	rhashtable_remove_fast(&nve->mc_list_ht, &mc_list->ht_node,
234 			       mlxsw_sp_nve_mc_list_ht_params);
235 	WARN_ON(!list_empty(&mc_list->records_list));
236 	kfree(mc_list);
237 }
238 
239 static struct mlxsw_sp_nve_mc_list *
240 mlxsw_sp_nve_mc_list_get(struct mlxsw_sp *mlxsw_sp,
241 			 const struct mlxsw_sp_nve_mc_list_key *key)
242 {
243 	struct mlxsw_sp_nve_mc_list *mc_list;
244 
245 	mc_list = mlxsw_sp_nve_mc_list_find(mlxsw_sp, key);
246 	if (mc_list)
247 		return mc_list;
248 
249 	return mlxsw_sp_nve_mc_list_create(mlxsw_sp, key);
250 }
251 
252 static void
253 mlxsw_sp_nve_mc_list_put(struct mlxsw_sp *mlxsw_sp,
254 			 struct mlxsw_sp_nve_mc_list *mc_list)
255 {
256 	if (!list_empty(&mc_list->records_list))
257 		return;
258 	mlxsw_sp_nve_mc_list_destroy(mlxsw_sp, mc_list);
259 }
260 
261 static struct mlxsw_sp_nve_mc_record *
262 mlxsw_sp_nve_mc_record_create(struct mlxsw_sp *mlxsw_sp,
263 			      struct mlxsw_sp_nve_mc_list *mc_list,
264 			      enum mlxsw_sp_l3proto proto)
265 {
266 	unsigned int num_max_entries = mlxsw_sp->nve->num_max_mc_entries[proto];
267 	struct mlxsw_sp_nve_mc_record *mc_record;
268 	int err;
269 
270 	mc_record = kzalloc(struct_size(mc_record, entries, num_max_entries),
271 			    GFP_KERNEL);
272 	if (!mc_record)
273 		return ERR_PTR(-ENOMEM);
274 
275 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_TNUMT, 1,
276 				  &mc_record->kvdl_index);
277 	if (err)
278 		goto err_kvdl_alloc;
279 
280 	mc_record->ops = mlxsw_sp_nve_mc_record_ops_arr[proto];
281 	mc_record->mlxsw_sp = mlxsw_sp;
282 	mc_record->mc_list = mc_list;
283 	mc_record->proto = proto;
284 	list_add_tail(&mc_record->list, &mc_list->records_list);
285 
286 	return mc_record;
287 
288 err_kvdl_alloc:
289 	kfree(mc_record);
290 	return ERR_PTR(err);
291 }
292 
293 static void
294 mlxsw_sp_nve_mc_record_destroy(struct mlxsw_sp_nve_mc_record *mc_record)
295 {
296 	struct mlxsw_sp *mlxsw_sp = mc_record->mlxsw_sp;
297 
298 	list_del(&mc_record->list);
299 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_TNUMT, 1,
300 			   mc_record->kvdl_index);
301 	WARN_ON(mc_record->num_entries);
302 	kfree(mc_record);
303 }
304 
305 static struct mlxsw_sp_nve_mc_record *
306 mlxsw_sp_nve_mc_record_get(struct mlxsw_sp *mlxsw_sp,
307 			   struct mlxsw_sp_nve_mc_list *mc_list,
308 			   enum mlxsw_sp_l3proto proto)
309 {
310 	struct mlxsw_sp_nve_mc_record *mc_record;
311 
312 	list_for_each_entry_reverse(mc_record, &mc_list->records_list, list) {
313 		unsigned int num_entries = mc_record->num_entries;
314 		struct mlxsw_sp_nve *nve = mlxsw_sp->nve;
315 
316 		if (mc_record->proto == proto &&
317 		    num_entries < nve->num_max_mc_entries[proto])
318 			return mc_record;
319 	}
320 
321 	return mlxsw_sp_nve_mc_record_create(mlxsw_sp, mc_list, proto);
322 }
323 
324 static void
325 mlxsw_sp_nve_mc_record_put(struct mlxsw_sp_nve_mc_record *mc_record)
326 {
327 	if (mc_record->num_entries != 0)
328 		return;
329 
330 	mlxsw_sp_nve_mc_record_destroy(mc_record);
331 }
332 
333 static struct mlxsw_sp_nve_mc_entry *
334 mlxsw_sp_nve_mc_free_entry_find(struct mlxsw_sp_nve_mc_record *mc_record)
335 {
336 	struct mlxsw_sp_nve *nve = mc_record->mlxsw_sp->nve;
337 	unsigned int num_max_entries;
338 	int i;
339 
340 	num_max_entries = nve->num_max_mc_entries[mc_record->proto];
341 	for (i = 0; i < num_max_entries; i++) {
342 		if (mc_record->entries[i].valid)
343 			continue;
344 		return &mc_record->entries[i];
345 	}
346 
347 	return NULL;
348 }
349 
350 static int
351 mlxsw_sp_nve_mc_record_refresh(struct mlxsw_sp_nve_mc_record *mc_record)
352 {
353 	enum mlxsw_reg_tnumt_record_type type = mc_record->ops->type;
354 	struct mlxsw_sp_nve_mc_list *mc_list = mc_record->mc_list;
355 	struct mlxsw_sp *mlxsw_sp = mc_record->mlxsw_sp;
356 	char tnumt_pl[MLXSW_REG_TNUMT_LEN];
357 	unsigned int num_max_entries;
358 	unsigned int num_entries = 0;
359 	u32 next_kvdl_index = 0;
360 	bool next_valid = false;
361 	int i;
362 
363 	if (!list_is_last(&mc_record->list, &mc_list->records_list)) {
364 		struct mlxsw_sp_nve_mc_record *next_record;
365 
366 		next_record = list_next_entry(mc_record, list);
367 		next_kvdl_index = next_record->kvdl_index;
368 		next_valid = true;
369 	}
370 
371 	mlxsw_reg_tnumt_pack(tnumt_pl, type, MLXSW_REG_TUNNEL_PORT_NVE,
372 			     mc_record->kvdl_index, next_valid,
373 			     next_kvdl_index, mc_record->num_entries);
374 
375 	num_max_entries = mlxsw_sp->nve->num_max_mc_entries[mc_record->proto];
376 	for (i = 0; i < num_max_entries; i++) {
377 		struct mlxsw_sp_nve_mc_entry *mc_entry;
378 
379 		mc_entry = &mc_record->entries[i];
380 		if (!mc_entry->valid)
381 			continue;
382 		mc_record->ops->entry_set(mc_record, mc_entry, tnumt_pl,
383 					  num_entries++);
384 	}
385 
386 	WARN_ON(num_entries != mc_record->num_entries);
387 
388 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tnumt), tnumt_pl);
389 }
390 
391 static bool
392 mlxsw_sp_nve_mc_record_is_first(struct mlxsw_sp_nve_mc_record *mc_record)
393 {
394 	struct mlxsw_sp_nve_mc_list *mc_list = mc_record->mc_list;
395 	struct mlxsw_sp_nve_mc_record *first_record;
396 
397 	first_record = list_first_entry(&mc_list->records_list,
398 					struct mlxsw_sp_nve_mc_record, list);
399 
400 	return mc_record == first_record;
401 }
402 
403 static struct mlxsw_sp_nve_mc_entry *
404 mlxsw_sp_nve_mc_entry_find(struct mlxsw_sp_nve_mc_record *mc_record,
405 			   union mlxsw_sp_l3addr *addr)
406 {
407 	struct mlxsw_sp_nve *nve = mc_record->mlxsw_sp->nve;
408 	unsigned int num_max_entries;
409 	int i;
410 
411 	num_max_entries = nve->num_max_mc_entries[mc_record->proto];
412 	for (i = 0; i < num_max_entries; i++) {
413 		struct mlxsw_sp_nve_mc_entry *mc_entry;
414 
415 		mc_entry = &mc_record->entries[i];
416 		if (!mc_entry->valid)
417 			continue;
418 		if (mc_record->ops->entry_compare(mc_record, mc_entry, addr))
419 			return mc_entry;
420 	}
421 
422 	return NULL;
423 }
424 
425 static int
426 mlxsw_sp_nve_mc_record_ip_add(struct mlxsw_sp_nve_mc_record *mc_record,
427 			      union mlxsw_sp_l3addr *addr)
428 {
429 	struct mlxsw_sp_nve_mc_entry *mc_entry = NULL;
430 	int err;
431 
432 	mc_entry = mlxsw_sp_nve_mc_free_entry_find(mc_record);
433 	if (WARN_ON(!mc_entry))
434 		return -EINVAL;
435 
436 	err = mc_record->ops->entry_add(mc_record, mc_entry, addr);
437 	if (err)
438 		return err;
439 	mc_record->num_entries++;
440 	mc_entry->valid = true;
441 
442 	err = mlxsw_sp_nve_mc_record_refresh(mc_record);
443 	if (err)
444 		goto err_record_refresh;
445 
446 	/* If this is a new record and not the first one, then we need to
447 	 * update the next pointer of the previous entry
448 	 */
449 	if (mc_record->num_entries != 1 ||
450 	    mlxsw_sp_nve_mc_record_is_first(mc_record))
451 		return 0;
452 
453 	err = mlxsw_sp_nve_mc_record_refresh(list_prev_entry(mc_record, list));
454 	if (err)
455 		goto err_prev_record_refresh;
456 
457 	return 0;
458 
459 err_prev_record_refresh:
460 err_record_refresh:
461 	mc_entry->valid = false;
462 	mc_record->num_entries--;
463 	mc_record->ops->entry_del(mc_record, mc_entry);
464 	return err;
465 }
466 
467 static void
468 mlxsw_sp_nve_mc_record_entry_del(struct mlxsw_sp_nve_mc_record *mc_record,
469 				 struct mlxsw_sp_nve_mc_entry *mc_entry)
470 {
471 	struct mlxsw_sp_nve_mc_list *mc_list = mc_record->mc_list;
472 
473 	mc_entry->valid = false;
474 	mc_record->num_entries--;
475 
476 	/* When the record continues to exist we only need to invalidate
477 	 * the requested entry
478 	 */
479 	if (mc_record->num_entries != 0) {
480 		mlxsw_sp_nve_mc_record_refresh(mc_record);
481 		mc_record->ops->entry_del(mc_record, mc_entry);
482 		return;
483 	}
484 
485 	/* If the record needs to be deleted, but it is not the first,
486 	 * then we need to make sure that the previous record no longer
487 	 * points to it. Remove deleted record from the list to reflect
488 	 * that and then re-add it at the end, so that it could be
489 	 * properly removed by the record destruction code
490 	 */
491 	if (!mlxsw_sp_nve_mc_record_is_first(mc_record)) {
492 		struct mlxsw_sp_nve_mc_record *prev_record;
493 
494 		prev_record = list_prev_entry(mc_record, list);
495 		list_del(&mc_record->list);
496 		mlxsw_sp_nve_mc_record_refresh(prev_record);
497 		list_add_tail(&mc_record->list, &mc_list->records_list);
498 		mc_record->ops->entry_del(mc_record, mc_entry);
499 		return;
500 	}
501 
502 	/* If the first record needs to be deleted, but the list is not
503 	 * singular, then the second record needs to be written in the
504 	 * first record's address, as this address is stored as a property
505 	 * of the FID
506 	 */
507 	if (mlxsw_sp_nve_mc_record_is_first(mc_record) &&
508 	    !list_is_singular(&mc_list->records_list)) {
509 		struct mlxsw_sp_nve_mc_record *next_record;
510 
511 		next_record = list_next_entry(mc_record, list);
512 		swap(mc_record->kvdl_index, next_record->kvdl_index);
513 		mlxsw_sp_nve_mc_record_refresh(next_record);
514 		mc_record->ops->entry_del(mc_record, mc_entry);
515 		return;
516 	}
517 
518 	/* This is the last case where the last remaining record needs to
519 	 * be deleted. Simply delete the entry
520 	 */
521 	mc_record->ops->entry_del(mc_record, mc_entry);
522 }
523 
524 static struct mlxsw_sp_nve_mc_record *
525 mlxsw_sp_nve_mc_record_find(struct mlxsw_sp_nve_mc_list *mc_list,
526 			    enum mlxsw_sp_l3proto proto,
527 			    union mlxsw_sp_l3addr *addr,
528 			    struct mlxsw_sp_nve_mc_entry **mc_entry)
529 {
530 	struct mlxsw_sp_nve_mc_record *mc_record;
531 
532 	list_for_each_entry(mc_record, &mc_list->records_list, list) {
533 		if (mc_record->proto != proto)
534 			continue;
535 
536 		*mc_entry = mlxsw_sp_nve_mc_entry_find(mc_record, addr);
537 		if (*mc_entry)
538 			return mc_record;
539 	}
540 
541 	return NULL;
542 }
543 
544 static int mlxsw_sp_nve_mc_list_ip_add(struct mlxsw_sp *mlxsw_sp,
545 				       struct mlxsw_sp_nve_mc_list *mc_list,
546 				       enum mlxsw_sp_l3proto proto,
547 				       union mlxsw_sp_l3addr *addr)
548 {
549 	struct mlxsw_sp_nve_mc_record *mc_record;
550 	int err;
551 
552 	mc_record = mlxsw_sp_nve_mc_record_get(mlxsw_sp, mc_list, proto);
553 	if (IS_ERR(mc_record))
554 		return PTR_ERR(mc_record);
555 
556 	err = mlxsw_sp_nve_mc_record_ip_add(mc_record, addr);
557 	if (err)
558 		goto err_ip_add;
559 
560 	return 0;
561 
562 err_ip_add:
563 	mlxsw_sp_nve_mc_record_put(mc_record);
564 	return err;
565 }
566 
567 static void mlxsw_sp_nve_mc_list_ip_del(struct mlxsw_sp *mlxsw_sp,
568 					struct mlxsw_sp_nve_mc_list *mc_list,
569 					enum mlxsw_sp_l3proto proto,
570 					union mlxsw_sp_l3addr *addr)
571 {
572 	struct mlxsw_sp_nve_mc_record *mc_record;
573 	struct mlxsw_sp_nve_mc_entry *mc_entry;
574 
575 	mc_record = mlxsw_sp_nve_mc_record_find(mc_list, proto, addr,
576 						&mc_entry);
577 	if (!mc_record)
578 		return;
579 
580 	mlxsw_sp_nve_mc_record_entry_del(mc_record, mc_entry);
581 	mlxsw_sp_nve_mc_record_put(mc_record);
582 }
583 
584 static int
585 mlxsw_sp_nve_fid_flood_index_set(struct mlxsw_sp_fid *fid,
586 				 struct mlxsw_sp_nve_mc_list *mc_list)
587 {
588 	struct mlxsw_sp_nve_mc_record *mc_record;
589 
590 	/* The address of the first record in the list is a property of
591 	 * the FID and we never change it. It only needs to be set when
592 	 * a new list is created
593 	 */
594 	if (mlxsw_sp_fid_nve_flood_index_is_set(fid))
595 		return 0;
596 
597 	mc_record = list_first_entry(&mc_list->records_list,
598 				     struct mlxsw_sp_nve_mc_record, list);
599 
600 	return mlxsw_sp_fid_nve_flood_index_set(fid, mc_record->kvdl_index);
601 }
602 
603 static void
604 mlxsw_sp_nve_fid_flood_index_clear(struct mlxsw_sp_fid *fid,
605 				   struct mlxsw_sp_nve_mc_list *mc_list)
606 {
607 	struct mlxsw_sp_nve_mc_record *mc_record;
608 
609 	/* The address of the first record needs to be invalidated only when
610 	 * the last record is about to be removed
611 	 */
612 	if (!list_is_singular(&mc_list->records_list))
613 		return;
614 
615 	mc_record = list_first_entry(&mc_list->records_list,
616 				     struct mlxsw_sp_nve_mc_record, list);
617 	if (mc_record->num_entries != 1)
618 		return;
619 
620 	return mlxsw_sp_fid_nve_flood_index_clear(fid);
621 }
622 
623 int mlxsw_sp_nve_flood_ip_add(struct mlxsw_sp *mlxsw_sp,
624 			      struct mlxsw_sp_fid *fid,
625 			      enum mlxsw_sp_l3proto proto,
626 			      union mlxsw_sp_l3addr *addr)
627 {
628 	struct mlxsw_sp_nve_mc_list_key key = { 0 };
629 	struct mlxsw_sp_nve_mc_list *mc_list;
630 	int err;
631 
632 	key.fid_index = mlxsw_sp_fid_index(fid);
633 	mc_list = mlxsw_sp_nve_mc_list_get(mlxsw_sp, &key);
634 	if (IS_ERR(mc_list))
635 		return PTR_ERR(mc_list);
636 
637 	err = mlxsw_sp_nve_mc_list_ip_add(mlxsw_sp, mc_list, proto, addr);
638 	if (err)
639 		goto err_add_ip;
640 
641 	err = mlxsw_sp_nve_fid_flood_index_set(fid, mc_list);
642 	if (err)
643 		goto err_fid_flood_index_set;
644 
645 	return 0;
646 
647 err_fid_flood_index_set:
648 	mlxsw_sp_nve_mc_list_ip_del(mlxsw_sp, mc_list, proto, addr);
649 err_add_ip:
650 	mlxsw_sp_nve_mc_list_put(mlxsw_sp, mc_list);
651 	return err;
652 }
653 
654 void mlxsw_sp_nve_flood_ip_del(struct mlxsw_sp *mlxsw_sp,
655 			       struct mlxsw_sp_fid *fid,
656 			       enum mlxsw_sp_l3proto proto,
657 			       union mlxsw_sp_l3addr *addr)
658 {
659 	struct mlxsw_sp_nve_mc_list_key key = { 0 };
660 	struct mlxsw_sp_nve_mc_list *mc_list;
661 
662 	key.fid_index = mlxsw_sp_fid_index(fid);
663 	mc_list = mlxsw_sp_nve_mc_list_find(mlxsw_sp, &key);
664 	if (!mc_list)
665 		return;
666 
667 	mlxsw_sp_nve_fid_flood_index_clear(fid, mc_list);
668 	mlxsw_sp_nve_mc_list_ip_del(mlxsw_sp, mc_list, proto, addr);
669 	mlxsw_sp_nve_mc_list_put(mlxsw_sp, mc_list);
670 }
671 
672 static void
673 mlxsw_sp_nve_mc_record_delete(struct mlxsw_sp_nve_mc_record *mc_record)
674 {
675 	struct mlxsw_sp_nve *nve = mc_record->mlxsw_sp->nve;
676 	unsigned int num_max_entries;
677 	int i;
678 
679 	num_max_entries = nve->num_max_mc_entries[mc_record->proto];
680 	for (i = 0; i < num_max_entries; i++) {
681 		struct mlxsw_sp_nve_mc_entry *mc_entry = &mc_record->entries[i];
682 
683 		if (!mc_entry->valid)
684 			continue;
685 		mlxsw_sp_nve_mc_record_entry_del(mc_record, mc_entry);
686 	}
687 
688 	WARN_ON(mc_record->num_entries);
689 	mlxsw_sp_nve_mc_record_put(mc_record);
690 }
691 
692 static void mlxsw_sp_nve_flood_ip_flush(struct mlxsw_sp *mlxsw_sp,
693 					struct mlxsw_sp_fid *fid)
694 {
695 	struct mlxsw_sp_nve_mc_record *mc_record, *tmp;
696 	struct mlxsw_sp_nve_mc_list_key key = { 0 };
697 	struct mlxsw_sp_nve_mc_list *mc_list;
698 
699 	if (!mlxsw_sp_fid_nve_flood_index_is_set(fid))
700 		return;
701 
702 	mlxsw_sp_fid_nve_flood_index_clear(fid);
703 
704 	key.fid_index = mlxsw_sp_fid_index(fid);
705 	mc_list = mlxsw_sp_nve_mc_list_find(mlxsw_sp, &key);
706 	if (WARN_ON(!mc_list))
707 		return;
708 
709 	list_for_each_entry_safe(mc_record, tmp, &mc_list->records_list, list)
710 		mlxsw_sp_nve_mc_record_delete(mc_record);
711 
712 	WARN_ON(!list_empty(&mc_list->records_list));
713 	mlxsw_sp_nve_mc_list_put(mlxsw_sp, mc_list);
714 }
715 
716 static int mlxsw_sp_nve_tunnel_init(struct mlxsw_sp *mlxsw_sp,
717 				    struct mlxsw_sp_nve_config *config)
718 {
719 	struct mlxsw_sp_nve *nve = mlxsw_sp->nve;
720 	const struct mlxsw_sp_nve_ops *ops;
721 	int err;
722 
723 	if (nve->num_nve_tunnels++ != 0)
724 		return 0;
725 
726 	nve->config = *config;
727 
728 	err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
729 				  &nve->tunnel_index);
730 	if (err)
731 		goto err_kvdl_alloc;
732 
733 	ops = nve->nve_ops_arr[config->type];
734 	err = ops->init(nve, config);
735 	if (err)
736 		goto err_ops_init;
737 
738 	return 0;
739 
740 err_ops_init:
741 	mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
742 			   nve->tunnel_index);
743 err_kvdl_alloc:
744 	memset(&nve->config, 0, sizeof(nve->config));
745 	nve->num_nve_tunnels--;
746 	return err;
747 }
748 
749 static void mlxsw_sp_nve_tunnel_fini(struct mlxsw_sp *mlxsw_sp)
750 {
751 	struct mlxsw_sp_nve *nve = mlxsw_sp->nve;
752 	const struct mlxsw_sp_nve_ops *ops;
753 
754 	ops = nve->nve_ops_arr[nve->config.type];
755 
756 	if (mlxsw_sp->nve->num_nve_tunnels == 1) {
757 		ops->fini(nve);
758 		mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
759 				   nve->tunnel_index);
760 		memset(&nve->config, 0, sizeof(nve->config));
761 	}
762 	nve->num_nve_tunnels--;
763 }
764 
765 static void mlxsw_sp_nve_fdb_flush_by_fid(struct mlxsw_sp *mlxsw_sp,
766 					  u16 fid_index)
767 {
768 	char sfdf_pl[MLXSW_REG_SFDF_LEN];
769 
770 	mlxsw_reg_sfdf_pack(sfdf_pl, MLXSW_REG_SFDF_FLUSH_PER_NVE_AND_FID);
771 	mlxsw_reg_sfdf_fid_set(sfdf_pl, fid_index);
772 	mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(sfdf), sfdf_pl);
773 }
774 
775 static void mlxsw_sp_nve_fdb_clear_offload(struct mlxsw_sp *mlxsw_sp,
776 					   const struct mlxsw_sp_fid *fid,
777 					   const struct net_device *nve_dev,
778 					   __be32 vni)
779 {
780 	const struct mlxsw_sp_nve_ops *ops;
781 	enum mlxsw_sp_nve_type type;
782 
783 	if (WARN_ON(mlxsw_sp_fid_nve_type(fid, &type)))
784 		return;
785 
786 	ops = mlxsw_sp->nve->nve_ops_arr[type];
787 	ops->fdb_clear_offload(nve_dev, vni);
788 }
789 
790 int mlxsw_sp_nve_fid_enable(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_fid *fid,
791 			    struct mlxsw_sp_nve_params *params,
792 			    struct netlink_ext_ack *extack)
793 {
794 	struct mlxsw_sp_nve *nve = mlxsw_sp->nve;
795 	const struct mlxsw_sp_nve_ops *ops;
796 	struct mlxsw_sp_nve_config config;
797 	int err;
798 
799 	ops = nve->nve_ops_arr[params->type];
800 
801 	if (!ops->can_offload(nve, params, extack))
802 		return -EINVAL;
803 
804 	memset(&config, 0, sizeof(config));
805 	ops->nve_config(nve, params, &config);
806 	if (nve->num_nve_tunnels &&
807 	    memcmp(&config, &nve->config, sizeof(config))) {
808 		NL_SET_ERR_MSG_MOD(extack, "Conflicting NVE tunnels configuration");
809 		return -EINVAL;
810 	}
811 
812 	err = mlxsw_sp_nve_tunnel_init(mlxsw_sp, &config);
813 	if (err) {
814 		NL_SET_ERR_MSG_MOD(extack, "Failed to initialize NVE tunnel");
815 		return err;
816 	}
817 
818 	err = mlxsw_sp_fid_vni_set(fid, params->type, params->vni,
819 				   params->dev->ifindex);
820 	if (err) {
821 		NL_SET_ERR_MSG_MOD(extack, "Failed to set VNI on FID");
822 		goto err_fid_vni_set;
823 	}
824 
825 	err = ops->fdb_replay(params->dev, params->vni, extack);
826 	if (err)
827 		goto err_fdb_replay;
828 
829 	return 0;
830 
831 err_fdb_replay:
832 	mlxsw_sp_fid_vni_clear(fid);
833 err_fid_vni_set:
834 	mlxsw_sp_nve_tunnel_fini(mlxsw_sp);
835 	return err;
836 }
837 
838 void mlxsw_sp_nve_fid_disable(struct mlxsw_sp *mlxsw_sp,
839 			      struct mlxsw_sp_fid *fid)
840 {
841 	u16 fid_index = mlxsw_sp_fid_index(fid);
842 	struct net_device *nve_dev;
843 	int nve_ifindex;
844 	__be32 vni;
845 
846 	mlxsw_sp_nve_flood_ip_flush(mlxsw_sp, fid);
847 	mlxsw_sp_nve_fdb_flush_by_fid(mlxsw_sp, fid_index);
848 
849 	if (WARN_ON(mlxsw_sp_fid_nve_ifindex(fid, &nve_ifindex) ||
850 		    mlxsw_sp_fid_vni(fid, &vni)))
851 		goto out;
852 
853 	nve_dev = dev_get_by_index(mlxsw_sp_net(mlxsw_sp), nve_ifindex);
854 	if (!nve_dev)
855 		goto out;
856 
857 	mlxsw_sp_nve_fdb_clear_offload(mlxsw_sp, fid, nve_dev, vni);
858 	mlxsw_sp_fid_fdb_clear_offload(fid, nve_dev);
859 
860 	dev_put(nve_dev);
861 
862 out:
863 	mlxsw_sp_fid_vni_clear(fid);
864 	mlxsw_sp_nve_tunnel_fini(mlxsw_sp);
865 }
866 
867 int mlxsw_sp_port_nve_init(struct mlxsw_sp_port *mlxsw_sp_port)
868 {
869 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
870 	char tnqdr_pl[MLXSW_REG_TNQDR_LEN];
871 
872 	mlxsw_reg_tnqdr_pack(tnqdr_pl, mlxsw_sp_port->local_port);
873 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tnqdr), tnqdr_pl);
874 }
875 
876 void mlxsw_sp_port_nve_fini(struct mlxsw_sp_port *mlxsw_sp_port)
877 {
878 }
879 
880 static int mlxsw_sp_nve_qos_init(struct mlxsw_sp *mlxsw_sp)
881 {
882 	char tnqcr_pl[MLXSW_REG_TNQCR_LEN];
883 
884 	mlxsw_reg_tnqcr_pack(tnqcr_pl);
885 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tnqcr), tnqcr_pl);
886 }
887 
888 static int mlxsw_sp_nve_ecn_encap_init(struct mlxsw_sp *mlxsw_sp)
889 {
890 	int i;
891 
892 	/* Iterate over inner ECN values */
893 	for (i = INET_ECN_NOT_ECT; i <= INET_ECN_CE; i++) {
894 		u8 outer_ecn = INET_ECN_encapsulate(0, i);
895 		char tneem_pl[MLXSW_REG_TNEEM_LEN];
896 		int err;
897 
898 		mlxsw_reg_tneem_pack(tneem_pl, i, outer_ecn);
899 		err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tneem),
900 				      tneem_pl);
901 		if (err)
902 			return err;
903 	}
904 
905 	return 0;
906 }
907 
908 static int __mlxsw_sp_nve_ecn_decap_init(struct mlxsw_sp *mlxsw_sp,
909 					 u8 inner_ecn, u8 outer_ecn)
910 {
911 	char tndem_pl[MLXSW_REG_TNDEM_LEN];
912 	bool trap_en, set_ce = false;
913 	u8 new_inner_ecn;
914 
915 	trap_en = !!__INET_ECN_decapsulate(outer_ecn, inner_ecn, &set_ce);
916 	new_inner_ecn = set_ce ? INET_ECN_CE : inner_ecn;
917 
918 	mlxsw_reg_tndem_pack(tndem_pl, outer_ecn, inner_ecn, new_inner_ecn,
919 			     trap_en, trap_en ? MLXSW_TRAP_ID_DECAP_ECN0 : 0);
920 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tndem), tndem_pl);
921 }
922 
923 static int mlxsw_sp_nve_ecn_decap_init(struct mlxsw_sp *mlxsw_sp)
924 {
925 	int i;
926 
927 	/* Iterate over inner ECN values */
928 	for (i = INET_ECN_NOT_ECT; i <= INET_ECN_CE; i++) {
929 		int j;
930 
931 		/* Iterate over outer ECN values */
932 		for (j = INET_ECN_NOT_ECT; j <= INET_ECN_CE; j++) {
933 			int err;
934 
935 			err = __mlxsw_sp_nve_ecn_decap_init(mlxsw_sp, i, j);
936 			if (err)
937 				return err;
938 		}
939 	}
940 
941 	return 0;
942 }
943 
944 static int mlxsw_sp_nve_ecn_init(struct mlxsw_sp *mlxsw_sp)
945 {
946 	int err;
947 
948 	err = mlxsw_sp_nve_ecn_encap_init(mlxsw_sp);
949 	if (err)
950 		return err;
951 
952 	return mlxsw_sp_nve_ecn_decap_init(mlxsw_sp);
953 }
954 
955 static int mlxsw_sp_nve_resources_query(struct mlxsw_sp *mlxsw_sp)
956 {
957 	unsigned int max;
958 
959 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_NVE_MC_ENTRIES_IPV4) ||
960 	    !MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_NVE_MC_ENTRIES_IPV6))
961 		return -EIO;
962 	max = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_NVE_MC_ENTRIES_IPV4);
963 	mlxsw_sp->nve->num_max_mc_entries[MLXSW_SP_L3_PROTO_IPV4] = max;
964 	max = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_NVE_MC_ENTRIES_IPV6);
965 	mlxsw_sp->nve->num_max_mc_entries[MLXSW_SP_L3_PROTO_IPV6] = max;
966 
967 	return 0;
968 }
969 
970 int mlxsw_sp_nve_init(struct mlxsw_sp *mlxsw_sp)
971 {
972 	struct mlxsw_sp_nve *nve;
973 	int err;
974 
975 	nve = kzalloc(sizeof(*mlxsw_sp->nve), GFP_KERNEL);
976 	if (!nve)
977 		return -ENOMEM;
978 	mlxsw_sp->nve = nve;
979 	nve->mlxsw_sp = mlxsw_sp;
980 	nve->nve_ops_arr = mlxsw_sp->nve_ops_arr;
981 
982 	err = rhashtable_init(&nve->mc_list_ht,
983 			      &mlxsw_sp_nve_mc_list_ht_params);
984 	if (err)
985 		goto err_rhashtable_init;
986 
987 	err = mlxsw_sp_nve_qos_init(mlxsw_sp);
988 	if (err)
989 		goto err_nve_qos_init;
990 
991 	err = mlxsw_sp_nve_ecn_init(mlxsw_sp);
992 	if (err)
993 		goto err_nve_ecn_init;
994 
995 	err = mlxsw_sp_nve_resources_query(mlxsw_sp);
996 	if (err)
997 		goto err_nve_resources_query;
998 
999 	return 0;
1000 
1001 err_nve_resources_query:
1002 err_nve_ecn_init:
1003 err_nve_qos_init:
1004 	rhashtable_destroy(&nve->mc_list_ht);
1005 err_rhashtable_init:
1006 	mlxsw_sp->nve = NULL;
1007 	kfree(nve);
1008 	return err;
1009 }
1010 
1011 void mlxsw_sp_nve_fini(struct mlxsw_sp *mlxsw_sp)
1012 {
1013 	WARN_ON(mlxsw_sp->nve->num_nve_tunnels);
1014 	rhashtable_destroy(&mlxsw_sp->nve->mc_list_ht);
1015 	kfree(mlxsw_sp->nve);
1016 	mlxsw_sp->nve = NULL;
1017 }
1018