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 #define ARFS_HASH_SHIFT BITS_PER_BYTE
40 #define ARFS_HASH_SIZE BIT(BITS_PER_BYTE)
41 
42 struct arfs_table {
43 	struct mlx5e_flow_table  ft;
44 	struct mlx5_flow_handle	 *default_rule;
45 	struct hlist_head	 rules_hash[ARFS_HASH_SIZE];
46 };
47 
48 enum arfs_type {
49 	ARFS_IPV4_TCP,
50 	ARFS_IPV6_TCP,
51 	ARFS_IPV4_UDP,
52 	ARFS_IPV6_UDP,
53 	ARFS_NUM_TYPES,
54 };
55 
56 struct mlx5e_arfs_tables {
57 	struct arfs_table arfs_tables[ARFS_NUM_TYPES];
58 	/* Protect aRFS rules list */
59 	spinlock_t                     arfs_lock;
60 	int                            last_filter_id;
61 	struct workqueue_struct        *wq;
62 };
63 
64 struct arfs_tuple {
65 	__be16 etype;
66 	u8     ip_proto;
67 	union {
68 		__be32 src_ipv4;
69 		struct in6_addr src_ipv6;
70 	};
71 	union {
72 		__be32 dst_ipv4;
73 		struct in6_addr dst_ipv6;
74 	};
75 	__be16 src_port;
76 	__be16 dst_port;
77 };
78 
79 struct arfs_rule {
80 	struct mlx5e_priv	*priv;
81 	struct work_struct      arfs_work;
82 	struct mlx5_flow_handle *rule;
83 	struct hlist_node	hlist;
84 	int			rxq;
85 	/* Flow ID passed to ndo_rx_flow_steer */
86 	int			flow_id;
87 	/* Filter ID returned by ndo_rx_flow_steer */
88 	int			filter_id;
89 	struct arfs_tuple	tuple;
90 };
91 
92 #define mlx5e_for_each_arfs_rule(hn, tmp, arfs_tables, i, j) \
93 	for (i = 0; i < ARFS_NUM_TYPES; i++) \
94 		mlx5e_for_each_hash_arfs_rule(hn, tmp, arfs_tables[i].rules_hash, j)
95 
96 #define mlx5e_for_each_hash_arfs_rule(hn, tmp, hash, j) \
97 	for (j = 0; j < ARFS_HASH_SIZE; j++) \
98 		hlist_for_each_entry_safe(hn, tmp, &hash[j], hlist)
99 
arfs_get_tt(enum arfs_type type)100 static enum mlx5_traffic_types arfs_get_tt(enum arfs_type type)
101 {
102 	switch (type) {
103 	case ARFS_IPV4_TCP:
104 		return MLX5_TT_IPV4_TCP;
105 	case ARFS_IPV4_UDP:
106 		return MLX5_TT_IPV4_UDP;
107 	case ARFS_IPV6_TCP:
108 		return MLX5_TT_IPV6_TCP;
109 	case ARFS_IPV6_UDP:
110 		return MLX5_TT_IPV6_UDP;
111 	default:
112 		return -EINVAL;
113 	}
114 }
115 
arfs_disable(struct mlx5e_flow_steering * fs)116 static int arfs_disable(struct mlx5e_flow_steering *fs)
117 {
118 	struct mlx5_ttc_table *ttc = mlx5e_fs_get_ttc(fs, false);
119 	int err, i;
120 
121 	for (i = 0; i < ARFS_NUM_TYPES; i++) {
122 		/* Modify ttc rules destination back to their default */
123 		err = mlx5_ttc_fwd_default_dest(ttc, arfs_get_tt(i));
124 		if (err) {
125 			fs_err(fs,
126 			       "%s: modify ttc[%d] default destination failed, err(%d)\n",
127 			       __func__, arfs_get_tt(i), err);
128 			return err;
129 		}
130 	}
131 	return 0;
132 }
133 
134 static void arfs_del_rules(struct mlx5e_flow_steering *fs);
135 
mlx5e_arfs_disable(struct mlx5e_flow_steering * fs)136 int mlx5e_arfs_disable(struct mlx5e_flow_steering *fs)
137 {
138 	/* Moving to switchdev mode, fs->arfs is freed by mlx5e_nic_profile
139 	 * cleanup_rx callback and it is not recreated when
140 	 * mlx5e_uplink_rep_profile is loaded as mlx5e_create_flow_steering()
141 	 * is not called by the uplink_rep profile init_rx callback. Thus, if
142 	 * ntuple is set, moving to switchdev flow will enter this function
143 	 * with fs->arfs nullified.
144 	 */
145 	if (!mlx5e_fs_get_arfs(fs))
146 		return 0;
147 
148 	arfs_del_rules(fs);
149 
150 	return arfs_disable(fs);
151 }
152 
mlx5e_arfs_enable(struct mlx5e_flow_steering * fs)153 int mlx5e_arfs_enable(struct mlx5e_flow_steering *fs)
154 {
155 	struct mlx5_ttc_table *ttc = mlx5e_fs_get_ttc(fs, false);
156 	struct mlx5e_arfs_tables *arfs =  mlx5e_fs_get_arfs(fs);
157 	struct mlx5_flow_destination dest = {};
158 	int err, i;
159 
160 	dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
161 	for (i = 0; i < ARFS_NUM_TYPES; i++) {
162 		dest.ft = arfs->arfs_tables[i].ft.t;
163 		/* Modify ttc rules destination to point on the aRFS FTs */
164 		err = mlx5_ttc_fwd_dest(ttc, arfs_get_tt(i), &dest);
165 		if (err) {
166 			fs_err(fs, "%s: modify ttc[%d] dest to arfs, failed err(%d)\n",
167 			       __func__, arfs_get_tt(i), err);
168 			arfs_disable(fs);
169 			return err;
170 		}
171 	}
172 	return 0;
173 }
174 
arfs_destroy_table(struct arfs_table * arfs_t)175 static void arfs_destroy_table(struct arfs_table *arfs_t)
176 {
177 	mlx5_del_flow_rules(arfs_t->default_rule);
178 	mlx5e_destroy_flow_table(&arfs_t->ft);
179 }
180 
_mlx5e_cleanup_tables(struct mlx5e_flow_steering * fs)181 static void _mlx5e_cleanup_tables(struct mlx5e_flow_steering *fs)
182 {
183 	struct mlx5e_arfs_tables *arfs =  mlx5e_fs_get_arfs(fs);
184 	int i;
185 
186 	arfs_del_rules(fs);
187 	destroy_workqueue(arfs->wq);
188 	for (i = 0; i < ARFS_NUM_TYPES; i++) {
189 		if (!IS_ERR_OR_NULL(arfs->arfs_tables[i].ft.t))
190 			arfs_destroy_table(&arfs->arfs_tables[i]);
191 	}
192 }
193 
mlx5e_arfs_destroy_tables(struct mlx5e_flow_steering * fs,bool ntuple)194 void mlx5e_arfs_destroy_tables(struct mlx5e_flow_steering *fs, bool ntuple)
195 {
196 	struct mlx5e_arfs_tables *arfs =  mlx5e_fs_get_arfs(fs);
197 
198 	if (!ntuple)
199 		return;
200 
201 	_mlx5e_cleanup_tables(fs);
202 	mlx5e_fs_set_arfs(fs, NULL);
203 	kvfree(arfs);
204 }
205 
arfs_add_default_rule(struct mlx5e_flow_steering * fs,struct mlx5e_rx_res * rx_res,enum arfs_type type)206 static int arfs_add_default_rule(struct mlx5e_flow_steering *fs,
207 				 struct mlx5e_rx_res *rx_res,
208 				 enum arfs_type type)
209 {
210 	struct mlx5e_arfs_tables *arfs =  mlx5e_fs_get_arfs(fs);
211 	struct arfs_table *arfs_t = &arfs->arfs_tables[type];
212 	struct mlx5_flow_destination dest = {};
213 	MLX5_DECLARE_FLOW_ACT(flow_act);
214 	enum mlx5_traffic_types tt;
215 	int err = 0;
216 
217 	dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
218 	tt = arfs_get_tt(type);
219 	if (tt == -EINVAL) {
220 		fs_err(fs, "%s: bad arfs_type: %d\n", __func__, type);
221 		return -EINVAL;
222 	}
223 
224 	/* FIXME: Must use mlx5_ttc_get_default_dest(),
225 	 * but can't since TTC default is not setup yet !
226 	 */
227 	dest.tir_num = mlx5e_rx_res_get_tirn_rss(rx_res, tt);
228 	arfs_t->default_rule = mlx5_add_flow_rules(arfs_t->ft.t, NULL,
229 						   &flow_act,
230 						   &dest, 1);
231 	if (IS_ERR(arfs_t->default_rule)) {
232 		err = PTR_ERR(arfs_t->default_rule);
233 		arfs_t->default_rule = NULL;
234 		fs_err(fs, "%s: add rule failed, arfs type=%d\n", __func__, type);
235 	}
236 
237 	return err;
238 }
239 
240 #define MLX5E_ARFS_NUM_GROUPS	2
241 #define MLX5E_ARFS_GROUP1_SIZE	(BIT(16) - 1)
242 #define MLX5E_ARFS_GROUP2_SIZE	BIT(0)
243 #define MLX5E_ARFS_TABLE_SIZE	(MLX5E_ARFS_GROUP1_SIZE +\
244 				 MLX5E_ARFS_GROUP2_SIZE)
arfs_create_groups(struct mlx5e_flow_table * ft,enum arfs_type type)245 static int arfs_create_groups(struct mlx5e_flow_table *ft,
246 			      enum  arfs_type type)
247 {
248 	int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
249 	void *outer_headers_c;
250 	int ix = 0;
251 	u32 *in;
252 	int err;
253 	u8 *mc;
254 
255 	ft->g = kcalloc(MLX5E_ARFS_NUM_GROUPS,
256 			sizeof(*ft->g), GFP_KERNEL);
257 	if (!ft->g)
258 		return -ENOMEM;
259 
260 	in = kvzalloc(inlen, GFP_KERNEL);
261 	if (!in) {
262 		err = -ENOMEM;
263 		goto err_free_g;
264 	}
265 
266 	mc = MLX5_ADDR_OF(create_flow_group_in, in, match_criteria);
267 	outer_headers_c = MLX5_ADDR_OF(fte_match_param, mc,
268 				       outer_headers);
269 	MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, ethertype);
270 	switch (type) {
271 	case ARFS_IPV4_TCP:
272 	case ARFS_IPV6_TCP:
273 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_dport);
274 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_sport);
275 		break;
276 	case ARFS_IPV4_UDP:
277 	case ARFS_IPV6_UDP:
278 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_dport);
279 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_sport);
280 		break;
281 	default:
282 		err = -EINVAL;
283 		goto err_free_in;
284 	}
285 
286 	switch (type) {
287 	case ARFS_IPV4_TCP:
288 	case ARFS_IPV4_UDP:
289 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
290 				 src_ipv4_src_ipv6.ipv4_layout.ipv4);
291 		MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
292 				 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
293 		break;
294 	case ARFS_IPV6_TCP:
295 	case ARFS_IPV6_UDP:
296 		memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
297 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
298 		       0xff, 16);
299 		memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
300 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
301 		       0xff, 16);
302 		break;
303 	default:
304 		err = -EINVAL;
305 		goto err_free_in;
306 	}
307 
308 	MLX5_SET_CFG(in, match_criteria_enable, MLX5_MATCH_OUTER_HEADERS);
309 	MLX5_SET_CFG(in, start_flow_index, ix);
310 	ix += MLX5E_ARFS_GROUP1_SIZE;
311 	MLX5_SET_CFG(in, end_flow_index, ix - 1);
312 	ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
313 	if (IS_ERR(ft->g[ft->num_groups]))
314 		goto err_clean_group;
315 	ft->num_groups++;
316 
317 	memset(in, 0, inlen);
318 	MLX5_SET_CFG(in, start_flow_index, ix);
319 	ix += MLX5E_ARFS_GROUP2_SIZE;
320 	MLX5_SET_CFG(in, end_flow_index, ix - 1);
321 	ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
322 	if (IS_ERR(ft->g[ft->num_groups]))
323 		goto err_clean_group;
324 	ft->num_groups++;
325 
326 	kvfree(in);
327 	return 0;
328 
329 err_clean_group:
330 	err = PTR_ERR(ft->g[ft->num_groups]);
331 	ft->g[ft->num_groups] = NULL;
332 err_free_in:
333 	kvfree(in);
334 err_free_g:
335 	kfree(ft->g);
336 	ft->g = NULL;
337 	return err;
338 }
339 
arfs_create_table(struct mlx5e_flow_steering * fs,struct mlx5e_rx_res * rx_res,enum arfs_type type)340 static int arfs_create_table(struct mlx5e_flow_steering *fs,
341 			     struct mlx5e_rx_res *rx_res,
342 			     enum arfs_type type)
343 {
344 	struct mlx5_flow_namespace *ns = mlx5e_fs_get_ns(fs, false);
345 	struct mlx5e_arfs_tables *arfs = mlx5e_fs_get_arfs(fs);
346 	struct mlx5e_flow_table *ft = &arfs->arfs_tables[type].ft;
347 	struct mlx5_flow_table_attr ft_attr = {};
348 	int err;
349 
350 	ft->num_groups = 0;
351 
352 	ft_attr.max_fte = MLX5E_ARFS_TABLE_SIZE;
353 	ft_attr.level = MLX5E_ARFS_FT_LEVEL;
354 	ft_attr.prio = MLX5E_NIC_PRIO;
355 
356 	ft->t = mlx5_create_flow_table(ns, &ft_attr);
357 	if (IS_ERR(ft->t)) {
358 		err = PTR_ERR(ft->t);
359 		ft->t = NULL;
360 		return err;
361 	}
362 
363 	err = arfs_create_groups(ft, type);
364 	if (err)
365 		goto err;
366 
367 	err = arfs_add_default_rule(fs, rx_res,  type);
368 	if (err)
369 		goto err;
370 
371 	return 0;
372 err:
373 	mlx5e_destroy_flow_table(ft);
374 	return err;
375 }
376 
mlx5e_arfs_create_tables(struct mlx5e_flow_steering * fs,struct mlx5e_rx_res * rx_res,bool ntuple)377 int mlx5e_arfs_create_tables(struct mlx5e_flow_steering *fs,
378 			     struct mlx5e_rx_res *rx_res, bool ntuple)
379 {
380 	struct mlx5e_arfs_tables *arfs;
381 	int err = -ENOMEM;
382 	int i;
383 
384 	if (!ntuple)
385 		return 0;
386 
387 	arfs = kvzalloc(sizeof(*arfs), GFP_KERNEL);
388 	if (!arfs)
389 		return -ENOMEM;
390 
391 	spin_lock_init(&arfs->arfs_lock);
392 	arfs->wq = create_singlethread_workqueue("mlx5e_arfs");
393 	if (!arfs->wq)
394 		goto err;
395 
396 	mlx5e_fs_set_arfs(fs, arfs);
397 
398 	for (i = 0; i < ARFS_NUM_TYPES; i++) {
399 		err = arfs_create_table(fs, rx_res, i);
400 		if (err)
401 			goto err_des;
402 	}
403 	return 0;
404 
405 err_des:
406 	_mlx5e_cleanup_tables(fs);
407 err:
408 	mlx5e_fs_set_arfs(fs, NULL);
409 	kvfree(arfs);
410 	return err;
411 }
412 
413 #define MLX5E_ARFS_EXPIRY_QUOTA 60
414 
arfs_may_expire_flow(struct mlx5e_priv * priv)415 static void arfs_may_expire_flow(struct mlx5e_priv *priv)
416 {
417 	struct mlx5e_arfs_tables *arfs = mlx5e_fs_get_arfs(priv->fs);
418 	struct arfs_rule *arfs_rule;
419 	struct hlist_node *htmp;
420 	HLIST_HEAD(del_list);
421 	int quota = 0;
422 	int i;
423 	int j;
424 
425 	spin_lock_bh(&arfs->arfs_lock);
426 	mlx5e_for_each_arfs_rule(arfs_rule, htmp, arfs->arfs_tables, i, j) {
427 		if (!work_pending(&arfs_rule->arfs_work) &&
428 		    rps_may_expire_flow(priv->netdev,
429 					arfs_rule->rxq, arfs_rule->flow_id,
430 					arfs_rule->filter_id)) {
431 			hlist_del_init(&arfs_rule->hlist);
432 			hlist_add_head(&arfs_rule->hlist, &del_list);
433 			if (quota++ > MLX5E_ARFS_EXPIRY_QUOTA)
434 				break;
435 		}
436 	}
437 	spin_unlock_bh(&arfs->arfs_lock);
438 	hlist_for_each_entry_safe(arfs_rule, htmp, &del_list, hlist) {
439 		if (arfs_rule->rule) {
440 			mlx5_del_flow_rules(arfs_rule->rule);
441 			priv->channel_stats[arfs_rule->rxq]->rq.arfs_expired++;
442 		}
443 		hlist_del(&arfs_rule->hlist);
444 		kfree(arfs_rule);
445 	}
446 }
447 
arfs_del_rules(struct mlx5e_flow_steering * fs)448 static void arfs_del_rules(struct mlx5e_flow_steering *fs)
449 {
450 	struct mlx5e_arfs_tables *arfs = mlx5e_fs_get_arfs(fs);
451 	struct hlist_node *htmp;
452 	struct arfs_rule *rule;
453 	HLIST_HEAD(del_list);
454 	int i;
455 	int j;
456 
457 	spin_lock_bh(&arfs->arfs_lock);
458 	mlx5e_for_each_arfs_rule(rule, htmp, arfs->arfs_tables, i, j) {
459 		hlist_del_init(&rule->hlist);
460 		hlist_add_head(&rule->hlist, &del_list);
461 	}
462 	spin_unlock_bh(&arfs->arfs_lock);
463 
464 	hlist_for_each_entry_safe(rule, htmp, &del_list, hlist) {
465 		cancel_work_sync(&rule->arfs_work);
466 		if (rule->rule)
467 			mlx5_del_flow_rules(rule->rule);
468 		hlist_del(&rule->hlist);
469 		kfree(rule);
470 	}
471 }
472 
473 static struct hlist_head *
arfs_hash_bucket(struct arfs_table * arfs_t,__be16 src_port,__be16 dst_port)474 arfs_hash_bucket(struct arfs_table *arfs_t, __be16 src_port,
475 		 __be16 dst_port)
476 {
477 	unsigned long l;
478 	int bucket_idx;
479 
480 	l = (__force unsigned long)src_port |
481 	    ((__force unsigned long)dst_port << 2);
482 
483 	bucket_idx = hash_long(l, ARFS_HASH_SHIFT);
484 
485 	return &arfs_t->rules_hash[bucket_idx];
486 }
487 
arfs_get_table(struct mlx5e_arfs_tables * arfs,u8 ip_proto,__be16 etype)488 static struct arfs_table *arfs_get_table(struct mlx5e_arfs_tables *arfs,
489 					 u8 ip_proto, __be16 etype)
490 {
491 	if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_TCP)
492 		return &arfs->arfs_tables[ARFS_IPV4_TCP];
493 	if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_UDP)
494 		return &arfs->arfs_tables[ARFS_IPV4_UDP];
495 	if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_TCP)
496 		return &arfs->arfs_tables[ARFS_IPV6_TCP];
497 	if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_UDP)
498 		return &arfs->arfs_tables[ARFS_IPV6_UDP];
499 
500 	return NULL;
501 }
502 
arfs_add_rule(struct mlx5e_priv * priv,struct arfs_rule * arfs_rule)503 static struct mlx5_flow_handle *arfs_add_rule(struct mlx5e_priv *priv,
504 					      struct arfs_rule *arfs_rule)
505 {
506 	struct mlx5e_arfs_tables *arfs = mlx5e_fs_get_arfs(priv->fs);
507 	struct arfs_tuple *tuple = &arfs_rule->tuple;
508 	struct mlx5_flow_handle *rule = NULL;
509 	struct mlx5_flow_destination dest = {};
510 	MLX5_DECLARE_FLOW_ACT(flow_act);
511 	struct arfs_table *arfs_table;
512 	struct mlx5_flow_spec *spec;
513 	struct mlx5_flow_table *ft;
514 	int err = 0;
515 
516 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
517 	if (!spec) {
518 		priv->channel_stats[arfs_rule->rxq]->rq.arfs_err++;
519 		err = -ENOMEM;
520 		goto out;
521 	}
522 	spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
523 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
524 			 outer_headers.ethertype);
525 	MLX5_SET(fte_match_param, spec->match_value, outer_headers.ethertype,
526 		 ntohs(tuple->etype));
527 	arfs_table = arfs_get_table(arfs, tuple->ip_proto, tuple->etype);
528 	if (!arfs_table) {
529 		WARN_ONCE(1, "arfs table does not exist for etype %u and ip_proto %u\n",
530 			  tuple->etype, tuple->ip_proto);
531 		err = -EINVAL;
532 		goto out;
533 	}
534 
535 	ft = arfs_table->ft.t;
536 	if (tuple->ip_proto == IPPROTO_TCP) {
537 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
538 				 outer_headers.tcp_dport);
539 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
540 				 outer_headers.tcp_sport);
541 		MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_dport,
542 			 ntohs(tuple->dst_port));
543 		MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_sport,
544 			 ntohs(tuple->src_port));
545 	} else {
546 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
547 				 outer_headers.udp_dport);
548 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
549 				 outer_headers.udp_sport);
550 		MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_dport,
551 			 ntohs(tuple->dst_port));
552 		MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_sport,
553 			 ntohs(tuple->src_port));
554 	}
555 	if (tuple->etype == htons(ETH_P_IP)) {
556 		memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
557 				    outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4),
558 		       &tuple->src_ipv4,
559 		       4);
560 		memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
561 				    outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
562 		       &tuple->dst_ipv4,
563 		       4);
564 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
565 				 outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4);
566 		MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
567 				 outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
568 	} else {
569 		memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
570 				    outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
571 		       &tuple->src_ipv6,
572 		       16);
573 		memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
574 				    outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
575 		       &tuple->dst_ipv6,
576 		       16);
577 		memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
578 				    outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
579 		       0xff,
580 		       16);
581 		memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
582 				    outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
583 		       0xff,
584 		       16);
585 	}
586 	dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
587 	dest.tir_num = mlx5e_rx_res_get_tirn_direct(priv->rx_res, arfs_rule->rxq);
588 	rule = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
589 	if (IS_ERR(rule)) {
590 		err = PTR_ERR(rule);
591 		priv->channel_stats[arfs_rule->rxq]->rq.arfs_err++;
592 		netdev_dbg(priv->netdev,
593 			   "%s: add rule(filter id=%d, rq idx=%d, ip proto=0x%x) failed,err=%d\n",
594 			   __func__, arfs_rule->filter_id, arfs_rule->rxq,
595 			   tuple->ip_proto, err);
596 	}
597 
598 out:
599 	kvfree(spec);
600 	return err ? ERR_PTR(err) : rule;
601 }
602 
arfs_modify_rule_rq(struct mlx5e_priv * priv,struct mlx5_flow_handle * rule,u16 rxq)603 static void arfs_modify_rule_rq(struct mlx5e_priv *priv,
604 				struct mlx5_flow_handle *rule, u16 rxq)
605 {
606 	struct mlx5_flow_destination dst = {};
607 	int err = 0;
608 
609 	dst.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
610 	dst.tir_num = mlx5e_rx_res_get_tirn_direct(priv->rx_res, rxq);
611 	err =  mlx5_modify_rule_destination(rule, &dst, NULL);
612 	if (err) {
613 		priv->channel_stats[rxq]->rq.arfs_err++;
614 		netdev_warn(priv->netdev,
615 			    "Failed to modify aRFS rule destination to rq=%d\n", rxq);
616 	}
617 }
618 
arfs_handle_work(struct work_struct * work)619 static void arfs_handle_work(struct work_struct *work)
620 {
621 	struct arfs_rule *arfs_rule = container_of(work,
622 						   struct arfs_rule,
623 						   arfs_work);
624 	struct mlx5e_priv *priv = arfs_rule->priv;
625 	struct mlx5e_arfs_tables *arfs;
626 	struct mlx5_flow_handle *rule;
627 
628 	arfs = mlx5e_fs_get_arfs(priv->fs);
629 	mutex_lock(&priv->state_lock);
630 	if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
631 		spin_lock_bh(&arfs->arfs_lock);
632 		hlist_del(&arfs_rule->hlist);
633 		spin_unlock_bh(&arfs->arfs_lock);
634 
635 		mutex_unlock(&priv->state_lock);
636 		kfree(arfs_rule);
637 		goto out;
638 	}
639 	mutex_unlock(&priv->state_lock);
640 
641 	if (!arfs_rule->rule) {
642 		rule = arfs_add_rule(priv, arfs_rule);
643 		if (IS_ERR(rule))
644 			goto out;
645 		arfs_rule->rule = rule;
646 		priv->channel_stats[arfs_rule->rxq]->rq.arfs_add++;
647 	} else {
648 		arfs_modify_rule_rq(priv, arfs_rule->rule,
649 				    arfs_rule->rxq);
650 	}
651 out:
652 	arfs_may_expire_flow(priv);
653 }
654 
arfs_alloc_rule(struct mlx5e_priv * priv,struct arfs_table * arfs_t,const struct flow_keys * fk,u16 rxq,u32 flow_id)655 static struct arfs_rule *arfs_alloc_rule(struct mlx5e_priv *priv,
656 					 struct arfs_table *arfs_t,
657 					 const struct flow_keys *fk,
658 					 u16 rxq, u32 flow_id)
659 {
660 	struct mlx5e_arfs_tables *arfs =  mlx5e_fs_get_arfs(priv->fs);
661 	struct arfs_rule *rule;
662 	struct arfs_tuple *tuple;
663 
664 	rule = kzalloc(sizeof(*rule), GFP_ATOMIC);
665 	if (!rule) {
666 		priv->channel_stats[rxq]->rq.arfs_err++;
667 		return NULL;
668 	}
669 
670 	rule->priv = priv;
671 	rule->rxq = rxq;
672 	INIT_WORK(&rule->arfs_work, arfs_handle_work);
673 
674 	tuple = &rule->tuple;
675 	tuple->etype = fk->basic.n_proto;
676 	tuple->ip_proto = fk->basic.ip_proto;
677 	if (tuple->etype == htons(ETH_P_IP)) {
678 		tuple->src_ipv4 = fk->addrs.v4addrs.src;
679 		tuple->dst_ipv4 = fk->addrs.v4addrs.dst;
680 	} else {
681 		memcpy(&tuple->src_ipv6, &fk->addrs.v6addrs.src,
682 		       sizeof(struct in6_addr));
683 		memcpy(&tuple->dst_ipv6, &fk->addrs.v6addrs.dst,
684 		       sizeof(struct in6_addr));
685 	}
686 	tuple->src_port = fk->ports.src;
687 	tuple->dst_port = fk->ports.dst;
688 
689 	rule->flow_id = flow_id;
690 	rule->filter_id = arfs->last_filter_id++ % RPS_NO_FILTER;
691 
692 	hlist_add_head(&rule->hlist,
693 		       arfs_hash_bucket(arfs_t, tuple->src_port,
694 					tuple->dst_port));
695 	return rule;
696 }
697 
arfs_cmp(const struct arfs_tuple * tuple,const struct flow_keys * fk)698 static bool arfs_cmp(const struct arfs_tuple *tuple, const struct flow_keys *fk)
699 {
700 	if (tuple->src_port != fk->ports.src || tuple->dst_port != fk->ports.dst)
701 		return false;
702 	if (tuple->etype != fk->basic.n_proto)
703 		return false;
704 	if (tuple->etype == htons(ETH_P_IP))
705 		return tuple->src_ipv4 == fk->addrs.v4addrs.src &&
706 		       tuple->dst_ipv4 == fk->addrs.v4addrs.dst;
707 	if (tuple->etype == htons(ETH_P_IPV6))
708 		return !memcmp(&tuple->src_ipv6, &fk->addrs.v6addrs.src,
709 			       sizeof(struct in6_addr)) &&
710 		       !memcmp(&tuple->dst_ipv6, &fk->addrs.v6addrs.dst,
711 			       sizeof(struct in6_addr));
712 	return false;
713 }
714 
arfs_find_rule(struct arfs_table * arfs_t,const struct flow_keys * fk)715 static struct arfs_rule *arfs_find_rule(struct arfs_table *arfs_t,
716 					const struct flow_keys *fk)
717 {
718 	struct arfs_rule *arfs_rule;
719 	struct hlist_head *head;
720 
721 	head = arfs_hash_bucket(arfs_t, fk->ports.src, fk->ports.dst);
722 	hlist_for_each_entry(arfs_rule, head, hlist) {
723 		if (arfs_cmp(&arfs_rule->tuple, fk))
724 			return arfs_rule;
725 	}
726 
727 	return NULL;
728 }
729 
mlx5e_rx_flow_steer(struct net_device * dev,const struct sk_buff * skb,u16 rxq_index,u32 flow_id)730 int mlx5e_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
731 			u16 rxq_index, u32 flow_id)
732 {
733 	struct mlx5e_priv *priv = netdev_priv(dev);
734 	struct mlx5e_arfs_tables *arfs;
735 	struct arfs_rule *arfs_rule;
736 	struct arfs_table *arfs_t;
737 	struct flow_keys fk;
738 
739 	arfs =  mlx5e_fs_get_arfs(priv->fs);
740 	if (!skb_flow_dissect_flow_keys(skb, &fk, 0))
741 		return -EPROTONOSUPPORT;
742 
743 	if (fk.basic.n_proto != htons(ETH_P_IP) &&
744 	    fk.basic.n_proto != htons(ETH_P_IPV6))
745 		return -EPROTONOSUPPORT;
746 
747 	if (skb->encapsulation)
748 		return -EPROTONOSUPPORT;
749 
750 	arfs_t = arfs_get_table(arfs, fk.basic.ip_proto, fk.basic.n_proto);
751 	if (!arfs_t)
752 		return -EPROTONOSUPPORT;
753 
754 	spin_lock_bh(&arfs->arfs_lock);
755 	arfs_rule = arfs_find_rule(arfs_t, &fk);
756 	if (arfs_rule) {
757 		if (arfs_rule->rxq == rxq_index || work_busy(&arfs_rule->arfs_work)) {
758 			spin_unlock_bh(&arfs->arfs_lock);
759 			return arfs_rule->filter_id;
760 		}
761 
762 		priv->channel_stats[rxq_index]->rq.arfs_request_in++;
763 		priv->channel_stats[arfs_rule->rxq]->rq.arfs_request_out++;
764 		arfs_rule->rxq = rxq_index;
765 	} else {
766 		arfs_rule = arfs_alloc_rule(priv, arfs_t, &fk, rxq_index, flow_id);
767 		if (!arfs_rule) {
768 			spin_unlock_bh(&arfs->arfs_lock);
769 			return -ENOMEM;
770 		}
771 	}
772 	queue_work(arfs->wq, &arfs_rule->arfs_work);
773 	spin_unlock_bh(&arfs->arfs_lock);
774 	return arfs_rule->filter_id;
775 }
776 
777