2 * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #ifdef CONFIG_RFS_ACCEL
35 #include <linux/hash.h>
36 #include <linux/mlx5/fs.h>
38 #include <linux/ipv6.h>
46 struct in6_addr src_ipv6;
50 struct in6_addr dst_ipv6;
57 struct mlx5e_priv *priv;
58 struct work_struct arfs_work;
59 struct mlx5_flow_handle *rule;
60 struct hlist_node hlist;
62 /* Flow ID passed to ndo_rx_flow_steer */
64 /* Filter ID returned by ndo_rx_flow_steer */
66 struct arfs_tuple tuple;
69 #define mlx5e_for_each_arfs_rule(hn, tmp, arfs_tables, i, j) \
70 for (i = 0; i < ARFS_NUM_TYPES; i++) \
71 mlx5e_for_each_hash_arfs_rule(hn, tmp, arfs_tables[i].rules_hash, j)
73 #define mlx5e_for_each_hash_arfs_rule(hn, tmp, hash, j) \
74 for (j = 0; j < ARFS_HASH_SIZE; j++) \
75 hlist_for_each_entry_safe(hn, tmp, &hash[j], hlist)
77 static enum mlx5e_traffic_types arfs_get_tt(enum arfs_type type)
81 return MLX5E_TT_IPV4_TCP;
83 return MLX5E_TT_IPV4_UDP;
85 return MLX5E_TT_IPV6_TCP;
87 return MLX5E_TT_IPV6_UDP;
93 static int arfs_disable(struct mlx5e_priv *priv)
95 struct mlx5_flow_destination dest = {};
96 struct mlx5e_tir *tir = priv->indir_tir;
101 dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
102 for (i = 0; i < ARFS_NUM_TYPES; i++) {
103 dest.tir_num = tir[i].tirn;
105 /* Modify ttc rules destination to bypass the aRFS tables*/
106 err = mlx5_modify_rule_destination(priv->fs.ttc.rules[tt],
109 netdev_err(priv->netdev,
110 "%s: modify ttc destination failed\n",
118 static void arfs_del_rules(struct mlx5e_priv *priv);
120 int mlx5e_arfs_disable(struct mlx5e_priv *priv)
122 arfs_del_rules(priv);
124 return arfs_disable(priv);
127 int mlx5e_arfs_enable(struct mlx5e_priv *priv)
129 struct mlx5_flow_destination dest = {};
134 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
135 for (i = 0; i < ARFS_NUM_TYPES; i++) {
136 dest.ft = priv->fs.arfs.arfs_tables[i].ft.t;
138 /* Modify ttc rules destination to point on the aRFS FTs */
139 err = mlx5_modify_rule_destination(priv->fs.ttc.rules[tt],
142 netdev_err(priv->netdev,
143 "%s: modify ttc destination failed err=%d\n",
152 static void arfs_destroy_table(struct arfs_table *arfs_t)
154 mlx5_del_flow_rules(arfs_t->default_rule);
155 mlx5e_destroy_flow_table(&arfs_t->ft);
158 void mlx5e_arfs_destroy_tables(struct mlx5e_priv *priv)
162 if (!(priv->netdev->hw_features & NETIF_F_NTUPLE))
165 arfs_del_rules(priv);
166 destroy_workqueue(priv->fs.arfs.wq);
167 for (i = 0; i < ARFS_NUM_TYPES; i++) {
168 if (!IS_ERR_OR_NULL(priv->fs.arfs.arfs_tables[i].ft.t))
169 arfs_destroy_table(&priv->fs.arfs.arfs_tables[i]);
173 static int arfs_add_default_rule(struct mlx5e_priv *priv,
176 struct arfs_table *arfs_t = &priv->fs.arfs.arfs_tables[type];
177 struct mlx5e_tir *tir = priv->indir_tir;
178 struct mlx5_flow_destination dest = {};
179 MLX5_DECLARE_FLOW_ACT(flow_act);
180 struct mlx5_flow_spec *spec;
181 enum mlx5e_traffic_types tt;
184 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
190 dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
191 tt = arfs_get_tt(type);
193 netdev_err(priv->netdev, "%s: bad arfs_type: %d\n",
199 dest.tir_num = tir[tt].tirn;
201 arfs_t->default_rule = mlx5_add_flow_rules(arfs_t->ft.t, spec,
204 if (IS_ERR(arfs_t->default_rule)) {
205 err = PTR_ERR(arfs_t->default_rule);
206 arfs_t->default_rule = NULL;
207 netdev_err(priv->netdev, "%s: add rule failed, arfs type=%d\n",
215 #define MLX5E_ARFS_NUM_GROUPS 2
216 #define MLX5E_ARFS_GROUP1_SIZE BIT(12)
217 #define MLX5E_ARFS_GROUP2_SIZE BIT(0)
218 #define MLX5E_ARFS_TABLE_SIZE (MLX5E_ARFS_GROUP1_SIZE +\
219 MLX5E_ARFS_GROUP2_SIZE)
220 static int arfs_create_groups(struct mlx5e_flow_table *ft,
223 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
224 void *outer_headers_c;
230 ft->g = kcalloc(MLX5E_ARFS_NUM_GROUPS,
231 sizeof(*ft->g), GFP_KERNEL);
232 in = kvzalloc(inlen, GFP_KERNEL);
239 mc = MLX5_ADDR_OF(create_flow_group_in, in, match_criteria);
240 outer_headers_c = MLX5_ADDR_OF(fte_match_param, mc,
242 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, ethertype);
246 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_dport);
247 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, tcp_sport);
251 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_dport);
252 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c, udp_sport);
262 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
263 src_ipv4_src_ipv6.ipv4_layout.ipv4);
264 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, outer_headers_c,
265 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
269 memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
270 src_ipv4_src_ipv6.ipv6_layout.ipv6),
272 memset(MLX5_ADDR_OF(fte_match_set_lyr_2_4, outer_headers_c,
273 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
281 MLX5_SET_CFG(in, match_criteria_enable, MLX5_MATCH_OUTER_HEADERS);
282 MLX5_SET_CFG(in, start_flow_index, ix);
283 ix += MLX5E_ARFS_GROUP1_SIZE;
284 MLX5_SET_CFG(in, end_flow_index, ix - 1);
285 ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
286 if (IS_ERR(ft->g[ft->num_groups]))
290 memset(in, 0, inlen);
291 MLX5_SET_CFG(in, start_flow_index, ix);
292 ix += MLX5E_ARFS_GROUP2_SIZE;
293 MLX5_SET_CFG(in, end_flow_index, ix - 1);
294 ft->g[ft->num_groups] = mlx5_create_flow_group(ft->t, in);
295 if (IS_ERR(ft->g[ft->num_groups]))
303 err = PTR_ERR(ft->g[ft->num_groups]);
304 ft->g[ft->num_groups] = NULL;
311 static int arfs_create_table(struct mlx5e_priv *priv,
314 struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
315 struct mlx5e_flow_table *ft = &arfs->arfs_tables[type].ft;
316 struct mlx5_flow_table_attr ft_attr = {};
321 ft_attr.max_fte = MLX5E_ARFS_TABLE_SIZE;
322 ft_attr.level = MLX5E_ARFS_FT_LEVEL;
323 ft_attr.prio = MLX5E_NIC_PRIO;
325 ft->t = mlx5_create_flow_table(priv->fs.ns, &ft_attr);
327 err = PTR_ERR(ft->t);
332 err = arfs_create_groups(ft, type);
336 err = arfs_add_default_rule(priv, type);
342 mlx5e_destroy_flow_table(ft);
346 int mlx5e_arfs_create_tables(struct mlx5e_priv *priv)
351 if (!(priv->netdev->hw_features & NETIF_F_NTUPLE))
354 spin_lock_init(&priv->fs.arfs.arfs_lock);
355 INIT_LIST_HEAD(&priv->fs.arfs.rules);
356 priv->fs.arfs.wq = create_singlethread_workqueue("mlx5e_arfs");
357 if (!priv->fs.arfs.wq)
360 for (i = 0; i < ARFS_NUM_TYPES; i++) {
361 err = arfs_create_table(priv, i);
367 mlx5e_arfs_destroy_tables(priv);
371 #define MLX5E_ARFS_EXPIRY_QUOTA 60
373 static void arfs_may_expire_flow(struct mlx5e_priv *priv)
375 struct arfs_rule *arfs_rule;
376 struct hlist_node *htmp;
381 HLIST_HEAD(del_list);
382 spin_lock_bh(&priv->fs.arfs.arfs_lock);
383 mlx5e_for_each_arfs_rule(arfs_rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
384 if (quota++ > MLX5E_ARFS_EXPIRY_QUOTA)
386 if (!work_pending(&arfs_rule->arfs_work) &&
387 rps_may_expire_flow(priv->netdev,
388 arfs_rule->rxq, arfs_rule->flow_id,
389 arfs_rule->filter_id)) {
390 hlist_del_init(&arfs_rule->hlist);
391 hlist_add_head(&arfs_rule->hlist, &del_list);
394 spin_unlock_bh(&priv->fs.arfs.arfs_lock);
395 hlist_for_each_entry_safe(arfs_rule, htmp, &del_list, hlist) {
397 mlx5_del_flow_rules(arfs_rule->rule);
398 hlist_del(&arfs_rule->hlist);
403 static void arfs_del_rules(struct mlx5e_priv *priv)
405 struct hlist_node *htmp;
406 struct arfs_rule *rule;
410 HLIST_HEAD(del_list);
411 spin_lock_bh(&priv->fs.arfs.arfs_lock);
412 mlx5e_for_each_arfs_rule(rule, htmp, priv->fs.arfs.arfs_tables, i, j) {
413 hlist_del_init(&rule->hlist);
414 hlist_add_head(&rule->hlist, &del_list);
416 spin_unlock_bh(&priv->fs.arfs.arfs_lock);
418 hlist_for_each_entry_safe(rule, htmp, &del_list, hlist) {
419 cancel_work_sync(&rule->arfs_work);
421 mlx5_del_flow_rules(rule->rule);
422 hlist_del(&rule->hlist);
427 static struct hlist_head *
428 arfs_hash_bucket(struct arfs_table *arfs_t, __be16 src_port,
434 l = (__force unsigned long)src_port |
435 ((__force unsigned long)dst_port << 2);
437 bucket_idx = hash_long(l, ARFS_HASH_SHIFT);
439 return &arfs_t->rules_hash[bucket_idx];
442 static u8 arfs_get_ip_proto(const struct sk_buff *skb)
444 return (skb->protocol == htons(ETH_P_IP)) ?
445 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
448 static struct arfs_table *arfs_get_table(struct mlx5e_arfs_tables *arfs,
449 u8 ip_proto, __be16 etype)
451 if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_TCP)
452 return &arfs->arfs_tables[ARFS_IPV4_TCP];
453 if (etype == htons(ETH_P_IP) && ip_proto == IPPROTO_UDP)
454 return &arfs->arfs_tables[ARFS_IPV4_UDP];
455 if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_TCP)
456 return &arfs->arfs_tables[ARFS_IPV6_TCP];
457 if (etype == htons(ETH_P_IPV6) && ip_proto == IPPROTO_UDP)
458 return &arfs->arfs_tables[ARFS_IPV6_UDP];
463 static struct mlx5_flow_handle *arfs_add_rule(struct mlx5e_priv *priv,
464 struct arfs_rule *arfs_rule)
466 struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
467 struct arfs_tuple *tuple = &arfs_rule->tuple;
468 struct mlx5_flow_handle *rule = NULL;
469 struct mlx5_flow_destination dest = {};
470 MLX5_DECLARE_FLOW_ACT(flow_act);
471 struct arfs_table *arfs_table;
472 struct mlx5_flow_spec *spec;
473 struct mlx5_flow_table *ft;
476 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
481 spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
482 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
483 outer_headers.ethertype);
484 MLX5_SET(fte_match_param, spec->match_value, outer_headers.ethertype,
485 ntohs(tuple->etype));
486 arfs_table = arfs_get_table(arfs, tuple->ip_proto, tuple->etype);
492 ft = arfs_table->ft.t;
493 if (tuple->ip_proto == IPPROTO_TCP) {
494 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
495 outer_headers.tcp_dport);
496 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
497 outer_headers.tcp_sport);
498 MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_dport,
499 ntohs(tuple->dst_port));
500 MLX5_SET(fte_match_param, spec->match_value, outer_headers.tcp_sport,
501 ntohs(tuple->src_port));
503 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
504 outer_headers.udp_dport);
505 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
506 outer_headers.udp_sport);
507 MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_dport,
508 ntohs(tuple->dst_port));
509 MLX5_SET(fte_match_param, spec->match_value, outer_headers.udp_sport,
510 ntohs(tuple->src_port));
512 if (tuple->etype == htons(ETH_P_IP)) {
513 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
514 outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4),
517 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
518 outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
521 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
522 outer_headers.src_ipv4_src_ipv6.ipv4_layout.ipv4);
523 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
524 outer_headers.dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
526 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
527 outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
530 memcpy(MLX5_ADDR_OF(fte_match_param, spec->match_value,
531 outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
534 memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
535 outer_headers.src_ipv4_src_ipv6.ipv6_layout.ipv6),
538 memset(MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
539 outer_headers.dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
543 dest.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
544 dest.tir_num = priv->direct_tir[arfs_rule->rxq].tirn;
545 rule = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
548 netdev_err(priv->netdev, "%s: add rule(filter id=%d, rq idx=%d) failed, err=%d\n",
549 __func__, arfs_rule->filter_id, arfs_rule->rxq, err);
554 return err ? ERR_PTR(err) : rule;
557 static void arfs_modify_rule_rq(struct mlx5e_priv *priv,
558 struct mlx5_flow_handle *rule, u16 rxq)
560 struct mlx5_flow_destination dst = {};
563 dst.type = MLX5_FLOW_DESTINATION_TYPE_TIR;
564 dst.tir_num = priv->direct_tir[rxq].tirn;
565 err = mlx5_modify_rule_destination(rule, &dst, NULL);
567 netdev_warn(priv->netdev,
568 "Failed to modfiy aRFS rule destination to rq=%d\n", rxq);
571 static void arfs_handle_work(struct work_struct *work)
573 struct arfs_rule *arfs_rule = container_of(work,
576 struct mlx5e_priv *priv = arfs_rule->priv;
577 struct mlx5_flow_handle *rule;
579 mutex_lock(&priv->state_lock);
580 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
581 spin_lock_bh(&priv->fs.arfs.arfs_lock);
582 hlist_del(&arfs_rule->hlist);
583 spin_unlock_bh(&priv->fs.arfs.arfs_lock);
585 mutex_unlock(&priv->state_lock);
589 mutex_unlock(&priv->state_lock);
591 if (!arfs_rule->rule) {
592 rule = arfs_add_rule(priv, arfs_rule);
595 arfs_rule->rule = rule;
597 arfs_modify_rule_rq(priv, arfs_rule->rule,
601 arfs_may_expire_flow(priv);
604 /* return L4 destination port from ip4/6 packets */
605 static __be16 arfs_get_dst_port(const struct sk_buff *skb)
607 char *transport_header;
609 transport_header = skb_transport_header(skb);
610 if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
611 return ((struct tcphdr *)transport_header)->dest;
612 return ((struct udphdr *)transport_header)->dest;
615 /* return L4 source port from ip4/6 packets */
616 static __be16 arfs_get_src_port(const struct sk_buff *skb)
618 char *transport_header;
620 transport_header = skb_transport_header(skb);
621 if (arfs_get_ip_proto(skb) == IPPROTO_TCP)
622 return ((struct tcphdr *)transport_header)->source;
623 return ((struct udphdr *)transport_header)->source;
626 static struct arfs_rule *arfs_alloc_rule(struct mlx5e_priv *priv,
627 struct arfs_table *arfs_t,
628 const struct sk_buff *skb,
629 u16 rxq, u32 flow_id)
631 struct arfs_rule *rule;
632 struct arfs_tuple *tuple;
634 rule = kzalloc(sizeof(*rule), GFP_ATOMIC);
640 INIT_WORK(&rule->arfs_work, arfs_handle_work);
642 tuple = &rule->tuple;
643 tuple->etype = skb->protocol;
644 if (tuple->etype == htons(ETH_P_IP)) {
645 tuple->src_ipv4 = ip_hdr(skb)->saddr;
646 tuple->dst_ipv4 = ip_hdr(skb)->daddr;
648 memcpy(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
649 sizeof(struct in6_addr));
650 memcpy(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
651 sizeof(struct in6_addr));
653 tuple->ip_proto = arfs_get_ip_proto(skb);
654 tuple->src_port = arfs_get_src_port(skb);
655 tuple->dst_port = arfs_get_dst_port(skb);
657 rule->flow_id = flow_id;
658 rule->filter_id = priv->fs.arfs.last_filter_id++ % RPS_NO_FILTER;
660 hlist_add_head(&rule->hlist,
661 arfs_hash_bucket(arfs_t, tuple->src_port,
666 static bool arfs_cmp_ips(struct arfs_tuple *tuple,
667 const struct sk_buff *skb)
669 if (tuple->etype == htons(ETH_P_IP) &&
670 tuple->src_ipv4 == ip_hdr(skb)->saddr &&
671 tuple->dst_ipv4 == ip_hdr(skb)->daddr)
673 if (tuple->etype == htons(ETH_P_IPV6) &&
674 (!memcmp(&tuple->src_ipv6, &ipv6_hdr(skb)->saddr,
675 sizeof(struct in6_addr))) &&
676 (!memcmp(&tuple->dst_ipv6, &ipv6_hdr(skb)->daddr,
677 sizeof(struct in6_addr))))
682 static struct arfs_rule *arfs_find_rule(struct arfs_table *arfs_t,
683 const struct sk_buff *skb)
685 struct arfs_rule *arfs_rule;
686 struct hlist_head *head;
687 __be16 src_port = arfs_get_src_port(skb);
688 __be16 dst_port = arfs_get_dst_port(skb);
690 head = arfs_hash_bucket(arfs_t, src_port, dst_port);
691 hlist_for_each_entry(arfs_rule, head, hlist) {
692 if (arfs_rule->tuple.src_port == src_port &&
693 arfs_rule->tuple.dst_port == dst_port &&
694 arfs_cmp_ips(&arfs_rule->tuple, skb)) {
702 int mlx5e_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
703 u16 rxq_index, u32 flow_id)
705 struct mlx5e_priv *priv = netdev_priv(dev);
706 struct mlx5e_arfs_tables *arfs = &priv->fs.arfs;
707 struct arfs_table *arfs_t;
708 struct arfs_rule *arfs_rule;
710 if (skb->protocol != htons(ETH_P_IP) &&
711 skb->protocol != htons(ETH_P_IPV6))
712 return -EPROTONOSUPPORT;
714 arfs_t = arfs_get_table(arfs, arfs_get_ip_proto(skb), skb->protocol);
716 return -EPROTONOSUPPORT;
718 spin_lock_bh(&arfs->arfs_lock);
719 arfs_rule = arfs_find_rule(arfs_t, skb);
721 if (arfs_rule->rxq == rxq_index) {
722 spin_unlock_bh(&arfs->arfs_lock);
723 return arfs_rule->filter_id;
725 arfs_rule->rxq = rxq_index;
727 arfs_rule = arfs_alloc_rule(priv, arfs_t, skb,
730 spin_unlock_bh(&arfs->arfs_lock);
734 queue_work(priv->fs.arfs.wq, &arfs_rule->arfs_work);
735 spin_unlock_bh(&arfs->arfs_lock);
736 return arfs_rule->filter_id;