1 /*
2  * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/hash.h>
34 #include <linux/mlx5/fs.h>
35 #include <linux/ip.h>
36 #include <linux/ipv6.h>
37 #include "en.h"
38 
39 struct arfs_tuple {
40 	__be16 etype;
41 	u8     ip_proto;
42 	union {
43 		__be32 src_ipv4;
44 		struct in6_addr src_ipv6;
45 	};
46 	union {
47 		__be32 dst_ipv4;
48 		struct in6_addr dst_ipv6;
49 	};
50 	__be16 src_port;
51 	__be16 dst_port;
52 };
53 
54 struct arfs_rule {
55 	struct mlx5e_priv	*priv;
56 	struct work_struct      arfs_work;
57 	struct mlx5_flow_handle *rule;
58 	struct hlist_node	hlist;
59 	int			rxq;
60 	/* Flow ID passed to ndo_rx_flow_steer */
61 	int			flow_id;
62 	/* Filter ID returned by ndo_rx_flow_steer */
63 	int			filter_id;
64 	struct arfs_tuple	tuple;
65 };
66 
67 #define mlx5e_for_each_arfs_rule(hn, tmp, arfs_tables, i, j) \
68 	for (i = 0; i < ARFS_NUM_TYPES; i++) \
69 		mlx5e_for_each_hash_arfs_rule(hn, tmp, arfs_tables[i].rules_hash, j)
70 
71 #define mlx5e_for_each_hash_arfs_rule(hn, tmp, hash, j) \
72 	for (j = 0; j < ARFS_HASH_SIZE; j++) \
73 		hlist_for_each_entry_safe(hn, tmp, &hash[j], hlist)
74 
arfs_get_tt(enum arfs_type type)75 static enum mlx5e_traffic_types arfs_get_tt(enum arfs_type type)
76 {
77 	switch (type) {
78 	case ARFS_IPV4_TCP:
79 		return MLX5E_TT_IPV4_TCP;
80 	case ARFS_IPV4_UDP:
81 		return MLX5E_TT_IPV4_UDP;
82 	case ARFS_IPV6_TCP:
83 		return MLX5E_TT_IPV6_TCP;
84 	case ARFS_IPV6_UDP:
85 		return MLX5E_TT_IPV6_UDP;
86 	default:
87 		return -EINVAL;
88 	}
89 }
90 
arfs_disable(struct mlx5e_priv * priv)91 static int arfs_disable(struct mlx5e_priv *priv)
92 {
93 	struct mlx5_flow_destination dest = {};
94 	struct mlx5e_tir *tir = priv->indir_tir;
95 	int err = 0;
96 	int tt;
97 	int i;
98 
99 	dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
100 	for (i = 0; i < ARFS_NUM_TYPES; i++) {
101 		dest.tir_num = tir[i].tirn;
102 		tt = arfs_get_tt(i);
103 		/* Modify ttc rules destination to bypass the aRFS tables*/
104 		err = mlx5_modify_rule_destination(priv->fs.ttc.rules[tt],
105 						   &dest, NULL);
106 		if (err) {
107 			netdev_err(priv->netdev,
108 				   "%s: modify ttc destination failed\n",
109 				   __func__);
110 			return err;
111 		}
112 	}
113 	return 0;
114 }
115 
116 static void arfs_del_rules(struct mlx5e_priv *priv);
117 
mlx5e_arfs_disable(struct mlx5e_priv * priv)118 int mlx5e_arfs_disable(struct mlx5e_priv *priv)
119 {
120 	arfs_del_rules(priv);
121 
122 	return arfs_disable(priv);
123 }
124 
mlx5e_arfs_enable(struct mlx5e_priv * priv)125 int mlx5e_arfs_enable(struct mlx5e_priv *priv)
126 {
127 	struct mlx5_flow_destination dest = {};
128 	int err = 0;
129 	int tt;
130 	int i;
131 
132 	dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
133 	for (i = 0; i < ARFS_NUM_TYPES; i++) {
134 		dest.ft = priv->fs.arfs.arfs_tables[i].ft.t;
135 		tt = arfs_get_tt(i);
136 		/* Modify ttc rules destination to point on the aRFS FTs */
137 		err = mlx5_modify_rule_destination(priv->fs.ttc.rules[tt],
138 						   &dest, NULL);
139 		if (err) {
140 			netdev_err(priv->netdev,
141 				   "%s: modify ttc destination failed err=%d\n",
142 				   __func__, err);
143 			arfs_disable(priv);
144 			return err;
145 		}
146 	}
147 	return 0;
148 }
149 
arfs_destroy_table(struct arfs_table * arfs_t)150 static void arfs_destroy_table(struct arfs_table *arfs_t)
151 {
152 	mlx5_del_flow_rules(arfs_t->default_rule);
153 	mlx5e_destroy_flow_table(&arfs_t->ft);
154 }
155 
mlx5e_arfs_destroy_tables(struct mlx5e_priv * priv)156 void mlx5e_arfs_destroy_tables(struct mlx5e_priv *priv)
157 {
158 	int i;
159 
160 	if (!(priv->netdev->hw_features & NETIF_F_NTUPLE))
161 		return;
162 
163 	arfs_del_rules(priv);
164 	destroy_workqueue(priv->fs.arfs.wq);
165 	for (i = 0; i < ARFS_NUM_TYPES; i++) {
166 		if (!IS_ERR_OR_NULL(priv->fs.arfs.arfs_tables[i].ft.t))
167 			arfs_destroy_table(&priv->fs.arfs.arfs_tables[i]);
168 	}
169 }
170 
arfs_add_default_rule(struct mlx5e_priv * priv,enum arfs_type type)171 static int arfs_add_default_rule(struct mlx5e_priv *priv,
172 				 enum arfs_type type)
173 {
174 	struct arfs_table *arfs_t = &priv->fs.arfs.arfs_tables[type];
175 	struct mlx5e_tir *tir = priv->indir_tir;
176 	struct mlx5_flow_destination dest = {};
177 	MLX5_DECLARE_FLOW_ACT(flow_act);
178 	struct mlx5_flow_spec *spec;
179 	enum mlx5e_traffic_types tt;
180 	int err = 0;
181 
182 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
183 	if (!spec) {
184 		err = -ENOMEM;
185 		goto out;
186 	}
187 
188 	dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
189 	tt = arfs_get_tt(type);
190 	if (tt == -EINVAL) {
191 		netdev_err(priv->netdev, "%s: bad arfs_type: %d\n",
192 			   __func__, type);
193 		err = -EINVAL;
194 		goto out;
195 	}
196 
197 	dest.tir_num = tir[tt].tirn;
198 
199 	arfs_t->default_rule = mlx5_add_flow_rules(arfs_t->ft.t, spec,
200 						   &flow_act,
201 						   &dest, 1);
202 	if (IS_ERR(arfs_t->default_rule)) {
203 		err = PTR_ERR(arfs_t->default_rule);
204 		arfs_t->default_rule = NULL;
205 		netdev_err(priv->netdev, "%s: add rule failed, arfs type=%d\n",
206 			   __func__, type);
207 	}
208 out:
209 	kvfree(spec);
210 	return err;
211 }
212 
213 #define MLX5E_ARFS_NUM_GROUPS	2
214 #define MLX5E_ARFS_GROUP1_SIZE	(BIT(16) - 1)
215 #define MLX5E_ARFS_GROUP2_SIZE	BIT(0)
216 #define MLX5E_ARFS_TABLE_SIZE	(MLX5E_ARFS_GROUP1_SIZE +\
217 				 MLX5E_ARFS_GROUP2_SIZE)
arfs_create_groups(struct mlx5e_flow_table * ft,enum arfs_type type)218 static int arfs_create_groups(struct mlx5e_flow_table *ft,
219 			      enum  arfs_type type)
220 {
221 	int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
222 	void *outer_headers_c;
223 	int ix = 0;
224 	u32 *in;
225 	int err;
226 	u8 *mc;
227 
228 	ft->g = kcalloc(MLX5E_ARFS_NUM_GROUPS,
229 			sizeof(*ft->g), GFP_KERNEL);
230 	in = kvzalloc(inlen, GFP_KERNEL);
231 	if  (!in || !ft->g) {
232 		kvfree(ft->g);
233 		kvfree(in);
234 		return -ENOMEM;
235 	}
236 
237 	mc = MLX5_ADDR_OF(create_flow_group_in, in, match_criteria);
238 	outer_headers_c = MLX5_ADDR_OF(fte_match_param, mc,
239 				       outer_headers);
240 	MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, ethertype);
241 	switch (type) {
242 	case ARFS_IPV4_TCP:
243 	case ARFS_IPV6_TCP:
244 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_dport);
245 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_sport);
246 		break;
247 	case ARFS_IPV4_UDP:
248 	case ARFS_IPV6_UDP:
249 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_dport);
250 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_sport);
251 		break;
252 	default:
253 		err = -EINVAL;
254 		goto out;
255 	}
256 
257 	switch (type) {
258 	case ARFS_IPV4_TCP:
259 	case ARFS_IPV4_UDP:
260 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
261 				 src_ipv4_src_ipv6.ipv4_layout.ipv4);
262 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
263 				 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
264 		break;
265 	case ARFS_IPV6_TCP:
266 	case ARFS_IPV6_UDP:
267 		memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
268 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
269 		       0xff, 16);
270 		memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
271 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
272 		       0xff, 16);
273 		break;
274 	default:
275 		err = -EINVAL;
276 		goto out;
277 	}
278 
279 	MLX5_SET_CFG(in, match_criteria_enable, MLX5_MATCH_OUTER_HEADERS);
280 	MLX5_SET_CFG(in, start_flow_index, ix);
281 	ix += MLX5E_ARFS_GROUP1_SIZE;
282 	MLX5_SET_CFG(in, end_flow_index, ix - 1);
283 	ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
284 	if (IS_ERR(ft->g[ft->num_groups]))
285 		goto err;
286 	ft->num_groups++;
287 
288 	memset(in, 0, inlen);
289 	MLX5_SET_CFG(in, start_flow_index, ix);
290 	ix += MLX5E_ARFS_GROUP2_SIZE;
291 	MLX5_SET_CFG(in, end_flow_index, ix - 1);
292 	ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
293 	if (IS_ERR(ft->g[ft->num_groups]))
294 		goto err;
295 	ft->num_groups++;
296 
297 	kvfree(in);
298 	return 0;
299 
300 err:
301 	err = PTR_ERR(ft->g[ft->num_groups]);
302 	ft->g[ft->num_groups] = NULL;
303 out:
304 	kvfree(in);
305 
306 	return err;
307 }
308 
arfs_create_table(struct mlx5e_priv * priv,enum arfs_type type)309 static int arfs_create_table(struct mlx5e_priv *priv,
310 			     enum arfs_type type)
311 {
312 	struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
313 	struct mlx5e_flow_table *ft = &arfs->arfs_tables[type].ft;
314 	struct mlx5_flow_table_attr ft_attr = {};
315 	int err;
316 
317 	ft->num_groups = 0;
318 
319 	ft_attr.max_fte = MLX5E_ARFS_TABLE_SIZE;
320 	ft_attr.level = MLX5E_ARFS_FT_LEVEL;
321 	ft_attr.prio = MLX5E_NIC_PRIO;
322 
323 	ft->t = mlx5_create_flow_table(priv->fs.ns, &ft_attr);
324 	if (IS_ERR(ft->t)) {
325 		err = PTR_ERR(ft->t);
326 		ft->t = NULL;
327 		return err;
328 	}
329 
330 	err = arfs_create_groups(ft, type);
331 	if (err)
332 		goto err;
333 
334 	err = arfs_add_default_rule(priv, type);
335 	if (err)
336 		goto err;
337 
338 	return 0;
339 err:
340 	mlx5e_destroy_flow_table(ft);
341 	return err;
342 }
343 
mlx5e_arfs_create_tables(struct mlx5e_priv * priv)344 int mlx5e_arfs_create_tables(struct mlx5e_priv *priv)
345 {
346 	int err = 0;
347 	int i;
348 
349 	if (!(priv->netdev->hw_features & NETIF_F_NTUPLE))
350 		return 0;
351 
352 	spin_lock_init(&priv->fs.arfs.arfs_lock);
353 	INIT_LIST_HEAD(&priv->fs.arfs.rules);
354 	priv->fs.arfs.wq = create_singlethread_workqueue("mlx5e_arfs");
355 	if (!priv->fs.arfs.wq)
356 		return -ENOMEM;
357 
358 	for (i = 0; i < ARFS_NUM_TYPES; i++) {
359 		err = arfs_create_table(priv, i);
360 		if (err)
361 			goto err;
362 	}
363 	return 0;
364 err:
365 	mlx5e_arfs_destroy_tables(priv);
366 	return err;
367 }
368 
369 #define MLX5E_ARFS_EXPIRY_QUOTA 60
370 
arfs_may_expire_flow(struct mlx5e_priv * priv)371 static void arfs_may_expire_flow(struct mlx5e_priv *priv)
372 {
373 	struct arfs_rule *arfs_rule;
374 	struct hlist_node *htmp;
375 	int quota = 0;
376 	int i;
377 	int j;
378 
379 	HLIST_HEAD(del_list);
380 	spin_lock_bh(&priv->fs.arfs.arfs_lock);
381 	mlx5e_for_each_arfs_rule(arfs_rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
382 		if (!work_pending(&arfs_rule->arfs_work) &&
383 		    rps_may_expire_flow(priv->netdev,
384 					arfs_rule->rxq, arfs_rule->flow_id,
385 					arfs_rule->filter_id)) {
386 			hlist_del_init(&arfs_rule->hlist);
387 			hlist_add_head(&arfs_rule->hlist, &del_list);
388 			if (quota++ > MLX5E_ARFS_EXPIRY_QUOTA)
389 				break;
390 		}
391 	}
392 	spin_unlock_bh(&priv->fs.arfs.arfs_lock);
393 	hlist_for_each_entry_safe(arfs_rule, htmp, &del_list, hlist) {
394 		if (arfs_rule->rule)
395 			mlx5_del_flow_rules(arfs_rule->rule);
396 		hlist_del(&arfs_rule->hlist);
397 		kfree(arfs_rule);
398 	}
399 }
400 
arfs_del_rules(struct mlx5e_priv * priv)401 static void arfs_del_rules(struct mlx5e_priv *priv)
402 {
403 	struct hlist_node *htmp;
404 	struct arfs_rule *rule;
405 	int i;
406 	int j;
407 
408 	HLIST_HEAD(del_list);
409 	spin_lock_bh(&priv->fs.arfs.arfs_lock);
410 	mlx5e_for_each_arfs_rule(rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
411 		hlist_del_init(&rule->hlist);
412 		hlist_add_head(&rule->hlist, &del_list);
413 	}
414 	spin_unlock_bh(&priv->fs.arfs.arfs_lock);
415 
416 	hlist_for_each_entry_safe(rule, htmp, &del_list, hlist) {
417 		cancel_work_sync(&rule->arfs_work);
418 		if (rule->rule)
419 			mlx5_del_flow_rules(rule->rule);
420 		hlist_del(&rule->hlist);
421 		kfree(rule);
422 	}
423 }
424 
425 static struct hlist_head *
arfs_hash_bucket(struct arfs_table * arfs_t,__be16 src_port,__be16 dst_port)426 arfs_hash_bucket(struct arfs_table *arfs_t, __be16 src_port,
427 		 __be16 dst_port)
428 {
429 	unsigned long l;
430 	int bucket_idx;
431 
432 	l = (__force unsigned long)src_port |
433 	    ((__force unsigned long)dst_port << 2);
434 
435 	bucket_idx = hash_long(l, ARFS_HASH_SHIFT);
436 
437 	return &arfs_t->rules_hash[bucket_idx];
438 }
439 
arfs_get_ip_proto(const struct sk_buff * skb)440 static u8 arfs_get_ip_proto(const struct sk_buff *skb)
441 {
442 	return (skb->protocol == htons(ETH_P_IP)) ?
443 		ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
444 }
445 
arfs_get_table(struct mlx5e_arfs_tables * arfs,u8 ip_proto,__be16 etype)446 static struct arfs_table *arfs_get_table(struct mlx5e_arfs_tables *arfs,
447 					 u8 ip_proto, __be16 etype)
448 {
449 	if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_TCP)
450 		return &arfs->arfs_tables[ARFS_IPV4_TCP];
451 	if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_UDP)
452 		return &arfs->arfs_tables[ARFS_IPV4_UDP];
453 	if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_TCP)
454 		return &arfs->arfs_tables[ARFS_IPV6_TCP];
455 	if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_UDP)
456 		return &arfs->arfs_tables[ARFS_IPV6_UDP];
457 
458 	return NULL;
459 }
460 
arfs_add_rule(struct mlx5e_priv * priv,struct arfs_rule * arfs_rule)461 static struct mlx5_flow_handle *arfs_add_rule(struct mlx5e_priv *priv,
462 					      struct arfs_rule *arfs_rule)
463 {
464 	struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
465 	struct arfs_tuple *tuple = &arfs_rule->tuple;
466 	struct mlx5_flow_handle *rule = NULL;
467 	struct mlx5_flow_destination dest = {};
468 	MLX5_DECLARE_FLOW_ACT(flow_act);
469 	struct arfs_table *arfs_table;
470 	struct mlx5_flow_spec *spec;
471 	struct mlx5_flow_table *ft;
472 	int err = 0;
473 
474 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
475 	if (!spec) {
476 		err = -ENOMEM;
477 		goto out;
478 	}
479 	spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
480 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
481 			 outer_headers.ethertype);
482 	MLX5_SET(fte_match_param, spec->match_value, outer_headers.ethertype,
483 		 ntohs(tuple->etype));
484 	arfs_table = arfs_get_table(arfs, tuple->ip_proto, tuple->etype);
485 	if (!arfs_table) {
486 		err = -EINVAL;
487 		goto out;
488 	}
489 
490 	ft = arfs_table->ft.t;
491 	if (tuple->ip_proto == IPPROTO_TCP) {
492 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
493 				 outer_headers.tcp_dport);
494 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
495 				 outer_headers.tcp_sport);
496 		MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_dport,
497 			 ntohs(tuple->dst_port));
498 		MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_sport,
499 			 ntohs(tuple->src_port));
500 	} else {
501 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
502 				 outer_headers.udp_dport);
503 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
504 				 outer_headers.udp_sport);
505 		MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_dport,
506 			 ntohs(tuple->dst_port));
507 		MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_sport,
508 			 ntohs(tuple->src_port));
509 	}
510 	if (tuple->etype == htons(ETH_P_IP)) {
511 		memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
512 				    outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4),
513 		       &tuple->src_ipv4,
514 		       4);
515 		memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
516 				    outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
517 		       &tuple->dst_ipv4,
518 		       4);
519 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
520 				 outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4);
521 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
522 				 outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
523 	} else {
524 		memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
525 				    outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
526 		       &tuple->src_ipv6,
527 		       16);
528 		memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
529 				    outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
530 		       &tuple->dst_ipv6,
531 		       16);
532 		memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
533 				    outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
534 		       0xff,
535 		       16);
536 		memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
537 				    outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
538 		       0xff,
539 		       16);
540 	}
541 	dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
542 	dest.tir_num = priv->direct_tir[arfs_rule->rxq].tirn;
543 	rule = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
544 	if (IS_ERR(rule)) {
545 		err = PTR_ERR(rule);
546 		netdev_err(priv->netdev, "%s: add rule(filter id=%d, rq idx=%d) failed, err=%d\n",
547 			   __func__, arfs_rule->filter_id, arfs_rule->rxq, err);
548 	}
549 
550 out:
551 	kvfree(spec);
552 	return err ? ERR_PTR(err) : rule;
553 }
554 
arfs_modify_rule_rq(struct mlx5e_priv * priv,struct mlx5_flow_handle * rule,u16 rxq)555 static void arfs_modify_rule_rq(struct mlx5e_priv *priv,
556 				struct mlx5_flow_handle *rule, u16 rxq)
557 {
558 	struct mlx5_flow_destination dst = {};
559 	int err = 0;
560 
561 	dst.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
562 	dst.tir_num = priv->direct_tir[rxq].tirn;
563 	err =  mlx5_modify_rule_destination(rule, &dst, NULL);
564 	if (err)
565 		netdev_warn(priv->netdev,
566 			    "Failed to modify aRFS rule destination to rq=%d\n", rxq);
567 }
568 
arfs_handle_work(struct work_struct * work)569 static void arfs_handle_work(struct work_struct *work)
570 {
571 	struct arfs_rule *arfs_rule = container_of(work,
572 						   struct arfs_rule,
573 						   arfs_work);
574 	struct mlx5e_priv *priv = arfs_rule->priv;
575 	struct mlx5_flow_handle *rule;
576 
577 	mutex_lock(&priv->state_lock);
578 	if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
579 		spin_lock_bh(&priv->fs.arfs.arfs_lock);
580 		hlist_del(&arfs_rule->hlist);
581 		spin_unlock_bh(&priv->fs.arfs.arfs_lock);
582 
583 		mutex_unlock(&priv->state_lock);
584 		kfree(arfs_rule);
585 		goto out;
586 	}
587 	mutex_unlock(&priv->state_lock);
588 
589 	if (!arfs_rule->rule) {
590 		rule = arfs_add_rule(priv, arfs_rule);
591 		if (IS_ERR(rule))
592 			goto out;
593 		arfs_rule->rule = rule;
594 	} else {
595 		arfs_modify_rule_rq(priv, arfs_rule->rule,
596 				    arfs_rule->rxq);
597 	}
598 out:
599 	arfs_may_expire_flow(priv);
600 }
601 
602 /* return L4 destination port from ip4/6 packets */
arfs_get_dst_port(const struct sk_buff * skb)603 static __be16 arfs_get_dst_port(const struct sk_buff *skb)
604 {
605 	char *transport_header;
606 
607 	transport_header = skb_transport_header(skb);
608 	if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
609 		return ((struct tcphdr *)transport_header)->dest;
610 	return ((struct udphdr *)transport_header)->dest;
611 }
612 
613 /* return L4 source port from ip4/6 packets */
arfs_get_src_port(const struct sk_buff * skb)614 static __be16 arfs_get_src_port(const struct sk_buff *skb)
615 {
616 	char *transport_header;
617 
618 	transport_header = skb_transport_header(skb);
619 	if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
620 		return ((struct tcphdr *)transport_header)->source;
621 	return ((struct udphdr *)transport_header)->source;
622 }
623 
arfs_alloc_rule(struct mlx5e_priv * priv,struct arfs_table * arfs_t,const struct sk_buff * skb,u16 rxq,u32 flow_id)624 static struct arfs_rule *arfs_alloc_rule(struct mlx5e_priv *priv,
625 					 struct arfs_table *arfs_t,
626 					 const struct sk_buff *skb,
627 					 u16 rxq, u32 flow_id)
628 {
629 	struct arfs_rule *rule;
630 	struct arfs_tuple *tuple;
631 
632 	rule = kzalloc(sizeof(*rule), GFP_ATOMIC);
633 	if (!rule)
634 		return NULL;
635 
636 	rule->priv = priv;
637 	rule->rxq = rxq;
638 	INIT_WORK(&rule->arfs_work, arfs_handle_work);
639 
640 	tuple = &rule->tuple;
641 	tuple->etype = skb->protocol;
642 	if (tuple->etype == htons(ETH_P_IP)) {
643 		tuple->src_ipv4 = ip_hdr(skb)->saddr;
644 		tuple->dst_ipv4 = ip_hdr(skb)->daddr;
645 	} else {
646 		memcpy(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
647 		       sizeof(struct in6_addr));
648 		memcpy(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
649 		       sizeof(struct in6_addr));
650 	}
651 	tuple->ip_proto = arfs_get_ip_proto(skb);
652 	tuple->src_port = arfs_get_src_port(skb);
653 	tuple->dst_port = arfs_get_dst_port(skb);
654 
655 	rule->flow_id = flow_id;
656 	rule->filter_id = priv->fs.arfs.last_filter_id++ % RPS_NO_FILTER;
657 
658 	hlist_add_head(&rule->hlist,
659 		       arfs_hash_bucket(arfs_t, tuple->src_port,
660 					tuple->dst_port));
661 	return rule;
662 }
663 
arfs_cmp_ips(struct arfs_tuple * tuple,const struct sk_buff * skb)664 static bool arfs_cmp_ips(struct arfs_tuple *tuple,
665 			 const struct sk_buff *skb)
666 {
667 	if (tuple->etype == htons(ETH_P_IP) &&
668 	    tuple->src_ipv4 == ip_hdr(skb)->saddr &&
669 	    tuple->dst_ipv4 == ip_hdr(skb)->daddr)
670 		return true;
671 	if (tuple->etype == htons(ETH_P_IPV6) &&
672 	    (!memcmp(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
673 		     sizeof(struct in6_addr))) &&
674 	    (!memcmp(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
675 		     sizeof(struct in6_addr))))
676 		return true;
677 	return false;
678 }
679 
arfs_find_rule(struct arfs_table * arfs_t,const struct sk_buff * skb)680 static struct arfs_rule *arfs_find_rule(struct arfs_table *arfs_t,
681 					const struct sk_buff *skb)
682 {
683 	struct arfs_rule *arfs_rule;
684 	struct hlist_head *head;
685 	__be16 src_port = arfs_get_src_port(skb);
686 	__be16 dst_port = arfs_get_dst_port(skb);
687 
688 	head = arfs_hash_bucket(arfs_t, src_port, dst_port);
689 	hlist_for_each_entry(arfs_rule, head, hlist) {
690 		if (arfs_rule->tuple.src_port == src_port &&
691 		    arfs_rule->tuple.dst_port == dst_port &&
692 		    arfs_cmp_ips(&arfs_rule->tuple, skb)) {
693 			return arfs_rule;
694 		}
695 	}
696 
697 	return NULL;
698 }
699 
mlx5e_rx_flow_steer(struct net_device * dev,const struct sk_buff * skb,u16 rxq_index,u32 flow_id)700 int mlx5e_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
701 			u16 rxq_index, u32 flow_id)
702 {
703 	struct mlx5e_priv *priv = netdev_priv(dev);
704 	struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
705 	struct arfs_table *arfs_t;
706 	struct arfs_rule *arfs_rule;
707 
708 	if (skb->protocol != htons(ETH_P_IP) &&
709 	    skb->protocol != htons(ETH_P_IPV6))
710 		return -EPROTONOSUPPORT;
711 
712 	if (skb->encapsulation)
713 		return -EPROTONOSUPPORT;
714 
715 	arfs_t = arfs_get_table(arfs, arfs_get_ip_proto(skb), skb->protocol);
716 	if (!arfs_t)
717 		return -EPROTONOSUPPORT;
718 
719 	spin_lock_bh(&arfs->arfs_lock);
720 	arfs_rule = arfs_find_rule(arfs_t, skb);
721 	if (arfs_rule) {
722 		if (arfs_rule->rxq == rxq_index) {
723 			spin_unlock_bh(&arfs->arfs_lock);
724 			return arfs_rule->filter_id;
725 		}
726 		arfs_rule->rxq = rxq_index;
727 	} else {
728 		arfs_rule = arfs_alloc_rule(priv, arfs_t, skb,
729 					    rxq_index, flow_id);
730 		if (!arfs_rule) {
731 			spin_unlock_bh(&arfs->arfs_lock);
732 			return -ENOMEM;
733 		}
734 	}
735 	queue_work(priv->fs.arfs.wq, &arfs_rule->arfs_work);
736 	spin_unlock_bh(&arfs->arfs_lock);
737 	return arfs_rule->filter_id;
738 }
739 
740