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 #include <net/flow_dissector.h>
34 #include <net/sch_generic.h>
35 #include <net/pkt_cls.h>
36 #include <net/tc_act/tc_gact.h>
37 #include <net/tc_act/tc_skbedit.h>
38 #include <linux/mlx5/fs.h>
39 #include <linux/mlx5/device.h>
40 #include <linux/rhashtable.h>
41 #include <net/switchdev.h>
42 #include <net/tc_act/tc_mirred.h>
43 #include <net/tc_act/tc_vlan.h>
44 #include <net/tc_act/tc_tunnel_key.h>
45 #include <net/tc_act/tc_pedit.h>
46 #include <net/tc_act/tc_csum.h>
47 #include <net/vxlan.h>
53 #include "lib/vxlan.h"
57 struct mlx5_nic_flow_attr {
63 struct mlx5_flow_table *hairpin_ft;
64 struct mlx5_fc *counter;
67 #define MLX5E_TC_FLOW_BASE (MLX5E_TC_LAST_EXPORTED_BIT + 1)
70 MLX5E_TC_FLOW_INGRESS = MLX5E_TC_INGRESS,
71 MLX5E_TC_FLOW_EGRESS = MLX5E_TC_EGRESS,
72 MLX5E_TC_FLOW_ESWITCH = BIT(MLX5E_TC_FLOW_BASE),
73 MLX5E_TC_FLOW_NIC = BIT(MLX5E_TC_FLOW_BASE + 1),
74 MLX5E_TC_FLOW_OFFLOADED = BIT(MLX5E_TC_FLOW_BASE + 2),
75 MLX5E_TC_FLOW_HAIRPIN = BIT(MLX5E_TC_FLOW_BASE + 3),
76 MLX5E_TC_FLOW_HAIRPIN_RSS = BIT(MLX5E_TC_FLOW_BASE + 4),
77 MLX5E_TC_FLOW_SLOW = BIT(MLX5E_TC_FLOW_BASE + 5),
80 #define MLX5E_TC_MAX_SPLITS 1
82 struct mlx5e_tc_flow {
83 struct rhash_head node;
84 struct mlx5e_priv *priv;
87 struct mlx5_flow_handle *rule[MLX5E_TC_MAX_SPLITS + 1];
88 struct list_head encap; /* flows sharing the same encap ID */
89 struct list_head mod_hdr; /* flows sharing the same mod hdr ID */
90 struct list_head hairpin; /* flows sharing the same hairpin */
92 struct mlx5_esw_flow_attr esw_attr[0];
93 struct mlx5_nic_flow_attr nic_attr[0];
97 struct mlx5e_tc_flow_parse_attr {
98 struct ip_tunnel_info tun_info;
99 struct mlx5_flow_spec spec;
100 int num_mod_hdr_actions;
101 void *mod_hdr_actions;
105 #define MLX5E_TC_TABLE_NUM_GROUPS 4
106 #define MLX5E_TC_TABLE_MAX_GROUP_SIZE BIT(16)
108 struct mlx5e_hairpin {
109 struct mlx5_hairpin *pair;
111 struct mlx5_core_dev *func_mdev;
112 struct mlx5e_priv *func_priv;
117 struct mlx5e_rqt indir_rqt;
118 u32 indir_tirn[MLX5E_NUM_INDIR_TIRS];
119 struct mlx5e_ttc_table ttc;
122 struct mlx5e_hairpin_entry {
123 /* a node of a hash table which keeps all the hairpin entries */
124 struct hlist_node hairpin_hlist;
126 /* flows sharing the same hairpin */
127 struct list_head flows;
131 struct mlx5e_hairpin *hp;
139 struct mlx5e_mod_hdr_entry {
140 /* a node of a hash table which keeps all the mod_hdr entries */
141 struct hlist_node mod_hdr_hlist;
143 /* flows sharing the same mod_hdr entry */
144 struct list_head flows;
146 struct mod_hdr_key key;
151 #define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)
153 static inline u32 hash_mod_hdr_info(struct mod_hdr_key *key)
155 return jhash(key->actions,
156 key->num_actions * MLX5_MH_ACT_SZ, 0);
159 static inline int cmp_mod_hdr_info(struct mod_hdr_key *a,
160 struct mod_hdr_key *b)
162 if (a->num_actions != b->num_actions)
165 return memcmp(a->actions, b->actions, a->num_actions * MLX5_MH_ACT_SZ);
168 static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
169 struct mlx5e_tc_flow *flow,
170 struct mlx5e_tc_flow_parse_attr *parse_attr)
172 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
173 int num_actions, actions_size, namespace, err;
174 struct mlx5e_mod_hdr_entry *mh;
175 struct mod_hdr_key key;
179 num_actions = parse_attr->num_mod_hdr_actions;
180 actions_size = MLX5_MH_ACT_SZ * num_actions;
182 key.actions = parse_attr->mod_hdr_actions;
183 key.num_actions = num_actions;
185 hash_key = hash_mod_hdr_info(&key);
187 if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
188 namespace = MLX5_FLOW_NAMESPACE_FDB;
189 hash_for_each_possible(esw->offloads.mod_hdr_tbl, mh,
190 mod_hdr_hlist, hash_key) {
191 if (!cmp_mod_hdr_info(&mh->key, &key)) {
197 namespace = MLX5_FLOW_NAMESPACE_KERNEL;
198 hash_for_each_possible(priv->fs.tc.mod_hdr_tbl, mh,
199 mod_hdr_hlist, hash_key) {
200 if (!cmp_mod_hdr_info(&mh->key, &key)) {
210 mh = kzalloc(sizeof(*mh) + actions_size, GFP_KERNEL);
214 mh->key.actions = (void *)mh + sizeof(*mh);
215 memcpy(mh->key.actions, key.actions, actions_size);
216 mh->key.num_actions = num_actions;
217 INIT_LIST_HEAD(&mh->flows);
219 err = mlx5_modify_header_alloc(priv->mdev, namespace,
226 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
227 hash_add(esw->offloads.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
229 hash_add(priv->fs.tc.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
232 list_add(&flow->mod_hdr, &mh->flows);
233 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
234 flow->esw_attr->mod_hdr_id = mh->mod_hdr_id;
236 flow->nic_attr->mod_hdr_id = mh->mod_hdr_id;
245 static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
246 struct mlx5e_tc_flow *flow)
248 struct list_head *next = flow->mod_hdr.next;
250 list_del(&flow->mod_hdr);
252 if (list_empty(next)) {
253 struct mlx5e_mod_hdr_entry *mh;
255 mh = list_entry(next, struct mlx5e_mod_hdr_entry, flows);
257 mlx5_modify_header_dealloc(priv->mdev, mh->mod_hdr_id);
258 hash_del(&mh->mod_hdr_hlist);
264 struct mlx5_core_dev *mlx5e_hairpin_get_mdev(struct net *net, int ifindex)
266 struct net_device *netdev;
267 struct mlx5e_priv *priv;
269 netdev = __dev_get_by_index(net, ifindex);
270 priv = netdev_priv(netdev);
274 static int mlx5e_hairpin_create_transport(struct mlx5e_hairpin *hp)
276 u32 in[MLX5_ST_SZ_DW(create_tir_in)] = {0};
280 err = mlx5_core_alloc_transport_domain(hp->func_mdev, &hp->tdn);
284 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
286 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
287 MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn[0]);
288 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
290 err = mlx5_core_create_tir(hp->func_mdev, in, MLX5_ST_SZ_BYTES(create_tir_in), &hp->tirn);
297 mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
302 static void mlx5e_hairpin_destroy_transport(struct mlx5e_hairpin *hp)
304 mlx5_core_destroy_tir(hp->func_mdev, hp->tirn);
305 mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
308 static void mlx5e_hairpin_fill_rqt_rqns(struct mlx5e_hairpin *hp, void *rqtc)
310 u32 indirection_rqt[MLX5E_INDIR_RQT_SIZE], rqn;
311 struct mlx5e_priv *priv = hp->func_priv;
312 int i, ix, sz = MLX5E_INDIR_RQT_SIZE;
314 mlx5e_build_default_indir_rqt(indirection_rqt, sz,
317 for (i = 0; i < sz; i++) {
319 if (priv->rss_params.hfunc == ETH_RSS_HASH_XOR)
320 ix = mlx5e_bits_invert(i, ilog2(sz));
321 ix = indirection_rqt[ix];
322 rqn = hp->pair->rqn[ix];
323 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
327 static int mlx5e_hairpin_create_indirect_rqt(struct mlx5e_hairpin *hp)
329 int inlen, err, sz = MLX5E_INDIR_RQT_SIZE;
330 struct mlx5e_priv *priv = hp->func_priv;
331 struct mlx5_core_dev *mdev = priv->mdev;
335 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
336 in = kvzalloc(inlen, GFP_KERNEL);
340 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
342 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
343 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
345 mlx5e_hairpin_fill_rqt_rqns(hp, rqtc);
347 err = mlx5_core_create_rqt(mdev, in, inlen, &hp->indir_rqt.rqtn);
349 hp->indir_rqt.enabled = true;
355 static int mlx5e_hairpin_create_indirect_tirs(struct mlx5e_hairpin *hp)
357 struct mlx5e_priv *priv = hp->func_priv;
358 u32 in[MLX5_ST_SZ_DW(create_tir_in)];
362 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
363 struct mlx5e_tirc_config ttconfig = mlx5e_tirc_get_default_config(tt);
365 memset(in, 0, MLX5_ST_SZ_BYTES(create_tir_in));
366 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
368 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
369 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
370 MLX5_SET(tirc, tirc, indirect_table, hp->indir_rqt.rqtn);
371 mlx5e_build_indir_tir_ctx_hash(&priv->rss_params, &ttconfig, tirc, false);
373 err = mlx5_core_create_tir(hp->func_mdev, in,
374 MLX5_ST_SZ_BYTES(create_tir_in), &hp->indir_tirn[tt]);
376 mlx5_core_warn(hp->func_mdev, "create indirect tirs failed, %d\n", err);
377 goto err_destroy_tirs;
383 for (i = 0; i < tt; i++)
384 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[i]);
388 static void mlx5e_hairpin_destroy_indirect_tirs(struct mlx5e_hairpin *hp)
392 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
393 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[tt]);
396 static void mlx5e_hairpin_set_ttc_params(struct mlx5e_hairpin *hp,
397 struct ttc_params *ttc_params)
399 struct mlx5_flow_table_attr *ft_attr = &ttc_params->ft_attr;
402 memset(ttc_params, 0, sizeof(*ttc_params));
404 ttc_params->any_tt_tirn = hp->tirn;
406 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
407 ttc_params->indir_tirn[tt] = hp->indir_tirn[tt];
409 ft_attr->max_fte = MLX5E_NUM_TT;
410 ft_attr->level = MLX5E_TC_TTC_FT_LEVEL;
411 ft_attr->prio = MLX5E_TC_PRIO;
414 static int mlx5e_hairpin_rss_init(struct mlx5e_hairpin *hp)
416 struct mlx5e_priv *priv = hp->func_priv;
417 struct ttc_params ttc_params;
420 err = mlx5e_hairpin_create_indirect_rqt(hp);
424 err = mlx5e_hairpin_create_indirect_tirs(hp);
426 goto err_create_indirect_tirs;
428 mlx5e_hairpin_set_ttc_params(hp, &ttc_params);
429 err = mlx5e_create_ttc_table(priv, &ttc_params, &hp->ttc);
431 goto err_create_ttc_table;
433 netdev_dbg(priv->netdev, "add hairpin: using %d channels rss ttc table id %x\n",
434 hp->num_channels, hp->ttc.ft.t->id);
438 err_create_ttc_table:
439 mlx5e_hairpin_destroy_indirect_tirs(hp);
440 err_create_indirect_tirs:
441 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
446 static void mlx5e_hairpin_rss_cleanup(struct mlx5e_hairpin *hp)
448 struct mlx5e_priv *priv = hp->func_priv;
450 mlx5e_destroy_ttc_table(priv, &hp->ttc);
451 mlx5e_hairpin_destroy_indirect_tirs(hp);
452 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
455 static struct mlx5e_hairpin *
456 mlx5e_hairpin_create(struct mlx5e_priv *priv, struct mlx5_hairpin_params *params,
459 struct mlx5_core_dev *func_mdev, *peer_mdev;
460 struct mlx5e_hairpin *hp;
461 struct mlx5_hairpin *pair;
464 hp = kzalloc(sizeof(*hp), GFP_KERNEL);
466 return ERR_PTR(-ENOMEM);
468 func_mdev = priv->mdev;
469 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
471 pair = mlx5_core_hairpin_create(func_mdev, peer_mdev, params);
474 goto create_pair_err;
477 hp->func_mdev = func_mdev;
478 hp->func_priv = priv;
479 hp->num_channels = params->num_channels;
481 err = mlx5e_hairpin_create_transport(hp);
483 goto create_transport_err;
485 if (hp->num_channels > 1) {
486 err = mlx5e_hairpin_rss_init(hp);
494 mlx5e_hairpin_destroy_transport(hp);
495 create_transport_err:
496 mlx5_core_hairpin_destroy(hp->pair);
502 static void mlx5e_hairpin_destroy(struct mlx5e_hairpin *hp)
504 if (hp->num_channels > 1)
505 mlx5e_hairpin_rss_cleanup(hp);
506 mlx5e_hairpin_destroy_transport(hp);
507 mlx5_core_hairpin_destroy(hp->pair);
511 static inline u32 hash_hairpin_info(u16 peer_vhca_id, u8 prio)
513 return (peer_vhca_id << 16 | prio);
516 static struct mlx5e_hairpin_entry *mlx5e_hairpin_get(struct mlx5e_priv *priv,
517 u16 peer_vhca_id, u8 prio)
519 struct mlx5e_hairpin_entry *hpe;
520 u32 hash_key = hash_hairpin_info(peer_vhca_id, prio);
522 hash_for_each_possible(priv->fs.tc.hairpin_tbl, hpe,
523 hairpin_hlist, hash_key) {
524 if (hpe->peer_vhca_id == peer_vhca_id && hpe->prio == prio)
531 #define UNKNOWN_MATCH_PRIO 8
533 static int mlx5e_hairpin_get_prio(struct mlx5e_priv *priv,
534 struct mlx5_flow_spec *spec, u8 *match_prio,
535 struct netlink_ext_ack *extack)
537 void *headers_c, *headers_v;
538 u8 prio_val, prio_mask = 0;
541 #ifdef CONFIG_MLX5_CORE_EN_DCB
542 if (priv->dcbx_dp.trust_state != MLX5_QPTS_TRUST_PCP) {
543 NL_SET_ERR_MSG_MOD(extack,
544 "only PCP trust state supported for hairpin");
548 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, outer_headers);
549 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
551 vlan_present = MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag);
553 prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
554 prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
557 if (!vlan_present || !prio_mask) {
558 prio_val = UNKNOWN_MATCH_PRIO;
559 } else if (prio_mask != 0x7) {
560 NL_SET_ERR_MSG_MOD(extack,
561 "masked priority match not supported for hairpin");
565 *match_prio = prio_val;
569 static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
570 struct mlx5e_tc_flow *flow,
571 struct mlx5e_tc_flow_parse_attr *parse_attr,
572 struct netlink_ext_ack *extack)
574 int peer_ifindex = parse_attr->mirred_ifindex;
575 struct mlx5_hairpin_params params;
576 struct mlx5_core_dev *peer_mdev;
577 struct mlx5e_hairpin_entry *hpe;
578 struct mlx5e_hairpin *hp;
585 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
586 if (!MLX5_CAP_GEN(priv->mdev, hairpin) || !MLX5_CAP_GEN(peer_mdev, hairpin)) {
587 NL_SET_ERR_MSG_MOD(extack, "hairpin is not supported");
591 peer_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
592 err = mlx5e_hairpin_get_prio(priv, &parse_attr->spec, &match_prio,
596 hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
600 hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
604 INIT_LIST_HEAD(&hpe->flows);
605 hpe->peer_vhca_id = peer_id;
606 hpe->prio = match_prio;
608 params.log_data_size = 15;
609 params.log_data_size = min_t(u8, params.log_data_size,
610 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_wq_data_sz));
611 params.log_data_size = max_t(u8, params.log_data_size,
612 MLX5_CAP_GEN(priv->mdev, log_min_hairpin_wq_data_sz));
614 params.log_num_packets = params.log_data_size -
615 MLX5_MPWRQ_MIN_LOG_STRIDE_SZ(priv->mdev);
616 params.log_num_packets = min_t(u8, params.log_num_packets,
617 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_num_packets));
619 params.q_counter = priv->q_counter;
620 /* set hairpin pair per each 50Gbs share of the link */
621 mlx5e_port_max_linkspeed(priv->mdev, &link_speed);
622 link_speed = max_t(u32, link_speed, 50000);
623 link_speed64 = link_speed;
624 do_div(link_speed64, 50000);
625 params.num_channels = link_speed64;
627 hp = mlx5e_hairpin_create(priv, ¶ms, peer_ifindex);
630 goto create_hairpin_err;
633 netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
634 hp->tirn, hp->pair->rqn[0], hp->pair->peer_mdev->priv.name,
635 hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
638 hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
639 hash_hairpin_info(peer_id, match_prio));
642 if (hpe->hp->num_channels > 1) {
643 flow->flags |= MLX5E_TC_FLOW_HAIRPIN_RSS;
644 flow->nic_attr->hairpin_ft = hpe->hp->ttc.ft.t;
646 flow->nic_attr->hairpin_tirn = hpe->hp->tirn;
648 list_add(&flow->hairpin, &hpe->flows);
657 static void mlx5e_hairpin_flow_del(struct mlx5e_priv *priv,
658 struct mlx5e_tc_flow *flow)
660 struct list_head *next = flow->hairpin.next;
662 list_del(&flow->hairpin);
664 /* no more hairpin flows for us, release the hairpin pair */
665 if (list_empty(next)) {
666 struct mlx5e_hairpin_entry *hpe;
668 hpe = list_entry(next, struct mlx5e_hairpin_entry, flows);
670 netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
671 hpe->hp->pair->peer_mdev->priv.name);
673 mlx5e_hairpin_destroy(hpe->hp);
674 hash_del(&hpe->hairpin_hlist);
680 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
681 struct mlx5e_tc_flow_parse_attr *parse_attr,
682 struct mlx5e_tc_flow *flow,
683 struct netlink_ext_ack *extack)
685 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
686 struct mlx5_core_dev *dev = priv->mdev;
687 struct mlx5_flow_destination dest[2] = {};
688 struct mlx5_flow_act flow_act = {
689 .action = attr->action,
690 .flow_tag = attr->flow_tag,
692 .flags = FLOW_ACT_HAS_TAG | FLOW_ACT_NO_APPEND,
694 struct mlx5_fc *counter = NULL;
695 bool table_created = false;
696 int err, dest_ix = 0;
698 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN) {
699 err = mlx5e_hairpin_flow_add(priv, flow, parse_attr, extack);
701 goto err_add_hairpin_flow;
703 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN_RSS) {
704 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
705 dest[dest_ix].ft = attr->hairpin_ft;
707 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_TIR;
708 dest[dest_ix].tir_num = attr->hairpin_tirn;
711 } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
712 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
713 dest[dest_ix].ft = priv->fs.vlan.ft.t;
717 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
718 counter = mlx5_fc_create(dev, true);
719 if (IS_ERR(counter)) {
720 err = PTR_ERR(counter);
723 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
724 dest[dest_ix].counter_id = mlx5_fc_id(counter);
726 attr->counter = counter;
729 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
730 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
731 flow_act.modify_id = attr->mod_hdr_id;
732 kfree(parse_attr->mod_hdr_actions);
734 goto err_create_mod_hdr_id;
737 if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
738 int tc_grp_size, tc_tbl_size;
739 u32 max_flow_counter;
741 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
742 MLX5_CAP_GEN(dev, max_flow_counter_15_0);
744 tc_grp_size = min_t(int, max_flow_counter, MLX5E_TC_TABLE_MAX_GROUP_SIZE);
746 tc_tbl_size = min_t(int, tc_grp_size * MLX5E_TC_TABLE_NUM_GROUPS,
747 BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev, log_max_ft_size)));
750 mlx5_create_auto_grouped_flow_table(priv->fs.ns,
753 MLX5E_TC_TABLE_NUM_GROUPS,
754 MLX5E_TC_FT_LEVEL, 0);
755 if (IS_ERR(priv->fs.tc.t)) {
756 NL_SET_ERR_MSG_MOD(extack,
757 "Failed to create tc offload table\n");
758 netdev_err(priv->netdev,
759 "Failed to create tc offload table\n");
760 err = PTR_ERR(priv->fs.tc.t);
764 table_created = true;
767 if (attr->match_level != MLX5_MATCH_NONE)
768 parse_attr->spec.match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
770 flow->rule[0] = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
771 &flow_act, dest, dest_ix);
773 if (IS_ERR(flow->rule[0])) {
774 err = PTR_ERR(flow->rule[0]);
782 mlx5_destroy_flow_table(priv->fs.tc.t);
783 priv->fs.tc.t = NULL;
786 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
787 mlx5e_detach_mod_hdr(priv, flow);
788 err_create_mod_hdr_id:
789 mlx5_fc_destroy(dev, counter);
791 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
792 mlx5e_hairpin_flow_del(priv, flow);
793 err_add_hairpin_flow:
797 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
798 struct mlx5e_tc_flow *flow)
800 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
801 struct mlx5_fc *counter = NULL;
803 counter = attr->counter;
804 mlx5_del_flow_rules(flow->rule[0]);
805 mlx5_fc_destroy(priv->mdev, counter);
807 if (!mlx5e_tc_num_filters(priv) && priv->fs.tc.t) {
808 mlx5_destroy_flow_table(priv->fs.tc.t);
809 priv->fs.tc.t = NULL;
812 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
813 mlx5e_detach_mod_hdr(priv, flow);
815 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
816 mlx5e_hairpin_flow_del(priv, flow);
819 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
820 struct mlx5e_tc_flow *flow);
822 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
823 struct ip_tunnel_info *tun_info,
824 struct net_device *mirred_dev,
825 struct net_device **encap_dev,
826 struct mlx5e_tc_flow *flow,
827 struct netlink_ext_ack *extack);
829 static struct mlx5_flow_handle *
830 mlx5e_tc_offload_fdb_rules(struct mlx5_eswitch *esw,
831 struct mlx5e_tc_flow *flow,
832 struct mlx5_flow_spec *spec,
833 struct mlx5_esw_flow_attr *attr)
835 struct mlx5_flow_handle *rule;
837 rule = mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
841 if (attr->mirror_count) {
842 flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, spec, attr);
843 if (IS_ERR(flow->rule[1])) {
844 mlx5_eswitch_del_offloaded_rule(esw, rule, attr);
845 return flow->rule[1];
849 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
854 mlx5e_tc_unoffload_fdb_rules(struct mlx5_eswitch *esw,
855 struct mlx5e_tc_flow *flow,
856 struct mlx5_esw_flow_attr *attr)
858 flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
860 if (attr->mirror_count)
861 mlx5_eswitch_del_fwd_rule(esw, flow->rule[1], attr);
863 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
866 static struct mlx5_flow_handle *
867 mlx5e_tc_offload_to_slow_path(struct mlx5_eswitch *esw,
868 struct mlx5e_tc_flow *flow,
869 struct mlx5_flow_spec *spec,
870 struct mlx5_esw_flow_attr *slow_attr)
872 struct mlx5_flow_handle *rule;
874 memcpy(slow_attr, flow->esw_attr, sizeof(*slow_attr));
875 slow_attr->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST,
876 slow_attr->mirror_count = 0,
877 slow_attr->dest_chain = FDB_SLOW_PATH_CHAIN,
879 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, slow_attr);
881 flow->flags |= MLX5E_TC_FLOW_SLOW;
887 mlx5e_tc_unoffload_from_slow_path(struct mlx5_eswitch *esw,
888 struct mlx5e_tc_flow *flow,
889 struct mlx5_esw_flow_attr *slow_attr)
891 memcpy(slow_attr, flow->esw_attr, sizeof(*slow_attr));
892 mlx5e_tc_unoffload_fdb_rules(esw, flow, slow_attr);
893 flow->flags &= ~MLX5E_TC_FLOW_SLOW;
897 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
898 struct mlx5e_tc_flow_parse_attr *parse_attr,
899 struct mlx5e_tc_flow *flow,
900 struct netlink_ext_ack *extack)
902 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
903 u32 max_chain = mlx5_eswitch_get_chain_range(esw);
904 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
905 u16 max_prio = mlx5_eswitch_get_prio_range(esw);
906 struct net_device *out_dev, *encap_dev = NULL;
907 struct mlx5_fc *counter = NULL;
908 struct mlx5e_rep_priv *rpriv;
909 struct mlx5e_priv *out_priv;
910 int err = 0, encap_err = 0;
912 /* if prios are not supported, keep the old behaviour of using same prio
913 * for all offloaded rules.
915 if (!mlx5_eswitch_prios_supported(esw))
918 if (attr->chain > max_chain) {
919 NL_SET_ERR_MSG(extack, "Requested chain is out of supported range");
921 goto err_max_prio_chain;
924 if (attr->prio > max_prio) {
925 NL_SET_ERR_MSG(extack, "Requested priority is out of supported range");
927 goto err_max_prio_chain;
930 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT) {
931 out_dev = __dev_get_by_index(dev_net(priv->netdev),
932 attr->parse_attr->mirred_ifindex);
933 encap_err = mlx5e_attach_encap(priv, &parse_attr->tun_info,
934 out_dev, &encap_dev, flow,
936 if (encap_err && encap_err != -EAGAIN) {
938 goto err_attach_encap;
940 out_priv = netdev_priv(encap_dev);
941 rpriv = out_priv->ppriv;
942 attr->out_rep[attr->out_count] = rpriv->rep;
943 attr->out_mdev[attr->out_count++] = out_priv->mdev;
946 err = mlx5_eswitch_add_vlan_action(esw, attr);
950 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
951 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
952 kfree(parse_attr->mod_hdr_actions);
957 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
958 counter = mlx5_fc_create(esw->dev, true);
959 if (IS_ERR(counter)) {
960 err = PTR_ERR(counter);
961 goto err_create_counter;
964 attr->counter = counter;
967 /* we get here if (1) there's no error or when
968 * (2) there's an encap action and we're on -EAGAIN (no valid neigh)
970 if (encap_err == -EAGAIN) {
971 /* continue with goto slow path rule instead */
972 struct mlx5_esw_flow_attr slow_attr;
974 flow->rule[0] = mlx5e_tc_offload_to_slow_path(esw, flow, &parse_attr->spec, &slow_attr);
976 flow->rule[0] = mlx5e_tc_offload_fdb_rules(esw, flow, &parse_attr->spec, attr);
979 if (IS_ERR(flow->rule[0])) {
980 err = PTR_ERR(flow->rule[0]);
987 mlx5_fc_destroy(esw->dev, counter);
989 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
990 mlx5e_detach_mod_hdr(priv, flow);
992 mlx5_eswitch_del_vlan_action(esw, attr);
994 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT)
995 mlx5e_detach_encap(priv, flow);
1001 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
1002 struct mlx5e_tc_flow *flow)
1004 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1005 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1006 struct mlx5_esw_flow_attr slow_attr;
1008 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
1009 if (flow->flags & MLX5E_TC_FLOW_SLOW)
1010 mlx5e_tc_unoffload_from_slow_path(esw, flow, &slow_attr);
1012 mlx5e_tc_unoffload_fdb_rules(esw, flow, attr);
1015 mlx5_eswitch_del_vlan_action(esw, attr);
1017 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT) {
1018 mlx5e_detach_encap(priv, flow);
1019 kvfree(attr->parse_attr);
1022 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1023 mlx5e_detach_mod_hdr(priv, flow);
1025 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
1026 mlx5_fc_destroy(esw->dev, attr->counter);
1029 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
1030 struct mlx5e_encap_entry *e)
1032 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1033 struct mlx5_esw_flow_attr slow_attr, *esw_attr;
1034 struct mlx5_flow_handle *rule;
1035 struct mlx5_flow_spec *spec;
1036 struct mlx5e_tc_flow *flow;
1039 err = mlx5_packet_reformat_alloc(priv->mdev, e->tunnel_type,
1040 e->encap_size, e->encap_header,
1041 MLX5_FLOW_NAMESPACE_FDB,
1044 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %d\n",
1048 e->flags |= MLX5_ENCAP_ENTRY_VALID;
1049 mlx5e_rep_queue_neigh_stats_work(priv);
1051 list_for_each_entry(flow, &e->flows, encap) {
1052 esw_attr = flow->esw_attr;
1053 esw_attr->encap_id = e->encap_id;
1054 spec = &esw_attr->parse_attr->spec;
1056 /* update from slow path rule to encap rule */
1057 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, esw_attr);
1059 err = PTR_ERR(rule);
1060 mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
1065 mlx5e_tc_unoffload_from_slow_path(esw, flow, &slow_attr);
1066 flow->flags |= MLX5E_TC_FLOW_OFFLOADED; /* was unset when slow path rule removed */
1067 flow->rule[0] = rule;
1071 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
1072 struct mlx5e_encap_entry *e)
1074 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1075 struct mlx5_esw_flow_attr slow_attr;
1076 struct mlx5_flow_handle *rule;
1077 struct mlx5_flow_spec *spec;
1078 struct mlx5e_tc_flow *flow;
1081 list_for_each_entry(flow, &e->flows, encap) {
1082 spec = &flow->esw_attr->parse_attr->spec;
1084 /* update from encap rule to slow path rule */
1085 rule = mlx5e_tc_offload_to_slow_path(esw, flow, spec, &slow_attr);
1088 err = PTR_ERR(rule);
1089 mlx5_core_warn(priv->mdev, "Failed to update slow path (encap) flow, %d\n",
1094 mlx5e_tc_unoffload_fdb_rules(esw, flow, flow->esw_attr);
1095 flow->flags |= MLX5E_TC_FLOW_OFFLOADED; /* was unset when fast path rule removed */
1096 flow->rule[0] = rule;
1099 if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
1100 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
1101 mlx5_packet_reformat_dealloc(priv->mdev, e->encap_id);
1105 static struct mlx5_fc *mlx5e_tc_get_counter(struct mlx5e_tc_flow *flow)
1107 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1108 return flow->esw_attr->counter;
1110 return flow->nic_attr->counter;
1113 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
1115 struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
1116 u64 bytes, packets, lastuse = 0;
1117 struct mlx5e_tc_flow *flow;
1118 struct mlx5e_encap_entry *e;
1119 struct mlx5_fc *counter;
1120 struct neigh_table *tbl;
1121 bool neigh_used = false;
1122 struct neighbour *n;
1124 if (m_neigh->family == AF_INET)
1126 #if IS_ENABLED(CONFIG_IPV6)
1127 else if (m_neigh->family == AF_INET6)
1133 list_for_each_entry(e, &nhe->encap_list, encap_list) {
1134 if (!(e->flags & MLX5_ENCAP_ENTRY_VALID))
1136 list_for_each_entry(flow, &e->flows, encap) {
1137 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
1138 counter = mlx5e_tc_get_counter(flow);
1139 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
1140 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
1151 nhe->reported_lastuse = jiffies;
1153 /* find the relevant neigh according to the cached device and
1156 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
1160 neigh_event_send(n, NULL);
1165 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1166 struct mlx5e_tc_flow *flow)
1168 struct list_head *next = flow->encap.next;
1170 list_del(&flow->encap);
1171 if (list_empty(next)) {
1172 struct mlx5e_encap_entry *e;
1174 e = list_entry(next, struct mlx5e_encap_entry, flows);
1175 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1177 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1178 mlx5_packet_reformat_dealloc(priv->mdev, e->encap_id);
1180 hash_del_rcu(&e->encap_hlist);
1181 kfree(e->encap_header);
1186 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1187 struct mlx5e_tc_flow *flow)
1189 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1190 mlx5e_tc_del_fdb_flow(priv, flow);
1192 mlx5e_tc_del_nic_flow(priv, flow);
1195 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
1196 struct tc_cls_flower_offload *f)
1198 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1200 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1202 void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1204 void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1207 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
1208 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
1210 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
1211 struct flow_dissector_key_keyid *key =
1212 skb_flow_dissector_target(f->dissector,
1213 FLOW_DISSECTOR_KEY_ENC_KEYID,
1215 struct flow_dissector_key_keyid *mask =
1216 skb_flow_dissector_target(f->dissector,
1217 FLOW_DISSECTOR_KEY_ENC_KEYID,
1219 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
1220 be32_to_cpu(mask->keyid));
1221 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
1222 be32_to_cpu(key->keyid));
1226 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1227 struct mlx5_flow_spec *spec,
1228 struct tc_cls_flower_offload *f)
1230 struct netlink_ext_ack *extack = f->common.extack;
1231 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1233 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1236 struct flow_dissector_key_control *enc_control =
1237 skb_flow_dissector_target(f->dissector,
1238 FLOW_DISSECTOR_KEY_ENC_CONTROL,
1241 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
1242 struct flow_dissector_key_ports *key =
1243 skb_flow_dissector_target(f->dissector,
1244 FLOW_DISSECTOR_KEY_ENC_PORTS,
1246 struct flow_dissector_key_ports *mask =
1247 skb_flow_dissector_target(f->dissector,
1248 FLOW_DISSECTOR_KEY_ENC_PORTS,
1251 /* Full udp dst port must be given */
1252 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
1253 goto vxlan_match_offload_err;
1255 if (mlx5_vxlan_lookup_port(priv->mdev->vxlan, be16_to_cpu(key->dst)) &&
1256 MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
1257 parse_vxlan_attr(spec, f);
1259 NL_SET_ERR_MSG_MOD(extack,
1260 "port isn't an offloaded vxlan udp dport");
1261 netdev_warn(priv->netdev,
1262 "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
1266 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1267 udp_dport, ntohs(mask->dst));
1268 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1269 udp_dport, ntohs(key->dst));
1271 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1272 udp_sport, ntohs(mask->src));
1273 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1274 udp_sport, ntohs(key->src));
1275 } else { /* udp dst port must be given */
1276 vxlan_match_offload_err:
1277 NL_SET_ERR_MSG_MOD(extack,
1278 "IP tunnel decap offload supported only for vxlan, must set UDP dport");
1279 netdev_warn(priv->netdev,
1280 "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
1284 if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1285 struct flow_dissector_key_ipv4_addrs *key =
1286 skb_flow_dissector_target(f->dissector,
1287 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1289 struct flow_dissector_key_ipv4_addrs *mask =
1290 skb_flow_dissector_target(f->dissector,
1291 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1293 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1294 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1296 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1297 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1300 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1301 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1303 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1304 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1307 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1308 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
1309 } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1310 struct flow_dissector_key_ipv6_addrs *key =
1311 skb_flow_dissector_target(f->dissector,
1312 FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1314 struct flow_dissector_key_ipv6_addrs *mask =
1315 skb_flow_dissector_target(f->dissector,
1316 FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1319 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1320 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1321 &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1322 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1323 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1324 &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1326 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1327 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1328 &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1329 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1330 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1331 &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1333 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1334 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
1337 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_IP)) {
1338 struct flow_dissector_key_ip *key =
1339 skb_flow_dissector_target(f->dissector,
1340 FLOW_DISSECTOR_KEY_ENC_IP,
1342 struct flow_dissector_key_ip *mask =
1343 skb_flow_dissector_target(f->dissector,
1344 FLOW_DISSECTOR_KEY_ENC_IP,
1347 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn, mask->tos & 0x3);
1348 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, key->tos & 0x3);
1350 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp, mask->tos >> 2);
1351 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, key->tos >> 2);
1353 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit, mask->ttl);
1354 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit, key->ttl);
1357 !MLX5_CAP_ESW_FLOWTABLE_FDB
1359 ft_field_support.outer_ipv4_ttl)) {
1360 NL_SET_ERR_MSG_MOD(extack,
1361 "Matching on TTL is not supported");
1367 /* Enforce DMAC when offloading incoming tunneled flows.
1368 * Flow counters require a match on the DMAC.
1370 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
1371 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
1372 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1373 dmac_47_16), priv->netdev->dev_addr);
1375 /* let software handle IP fragments */
1376 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1377 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
1382 static int __parse_cls_flower(struct mlx5e_priv *priv,
1383 struct mlx5_flow_spec *spec,
1384 struct tc_cls_flower_offload *f,
1387 struct netlink_ext_ack *extack = f->common.extack;
1388 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1390 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1392 void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1394 void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1399 *match_level = MLX5_MATCH_NONE;
1401 if (f->dissector->used_keys &
1402 ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
1403 BIT(FLOW_DISSECTOR_KEY_BASIC) |
1404 BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
1405 BIT(FLOW_DISSECTOR_KEY_VLAN) |
1406 BIT(FLOW_DISSECTOR_KEY_CVLAN) |
1407 BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
1408 BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
1409 BIT(FLOW_DISSECTOR_KEY_PORTS) |
1410 BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
1411 BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
1412 BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
1413 BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
1414 BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
1415 BIT(FLOW_DISSECTOR_KEY_TCP) |
1416 BIT(FLOW_DISSECTOR_KEY_IP) |
1417 BIT(FLOW_DISSECTOR_KEY_ENC_IP))) {
1418 NL_SET_ERR_MSG_MOD(extack, "Unsupported key");
1419 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
1420 f->dissector->used_keys);
1424 if ((dissector_uses_key(f->dissector,
1425 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
1426 dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
1427 dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
1428 dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
1429 struct flow_dissector_key_control *key =
1430 skb_flow_dissector_target(f->dissector,
1431 FLOW_DISSECTOR_KEY_ENC_CONTROL,
1433 switch (key->addr_type) {
1434 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
1435 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
1436 if (parse_tunnel_attr(priv, spec, f))
1443 /* In decap flow, header pointers should point to the inner
1444 * headers, outer header were already set by parse_tunnel_attr
1446 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1448 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1452 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1453 struct flow_dissector_key_basic *key =
1454 skb_flow_dissector_target(f->dissector,
1455 FLOW_DISSECTOR_KEY_BASIC,
1457 struct flow_dissector_key_basic *mask =
1458 skb_flow_dissector_target(f->dissector,
1459 FLOW_DISSECTOR_KEY_BASIC,
1461 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1462 ntohs(mask->n_proto));
1463 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1464 ntohs(key->n_proto));
1467 *match_level = MLX5_MATCH_L2;
1470 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
1471 struct flow_dissector_key_vlan *key =
1472 skb_flow_dissector_target(f->dissector,
1473 FLOW_DISSECTOR_KEY_VLAN,
1475 struct flow_dissector_key_vlan *mask =
1476 skb_flow_dissector_target(f->dissector,
1477 FLOW_DISSECTOR_KEY_VLAN,
1479 if (mask->vlan_id || mask->vlan_priority || mask->vlan_tpid) {
1480 if (key->vlan_tpid == htons(ETH_P_8021AD)) {
1481 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1483 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1486 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1488 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1492 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
1493 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
1495 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
1496 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
1498 *match_level = MLX5_MATCH_L2;
1500 } else if (*match_level != MLX5_MATCH_NONE) {
1501 MLX5_SET(fte_match_set_lyr_2_4, headers_c, svlan_tag, 1);
1502 MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
1503 *match_level = MLX5_MATCH_L2;
1506 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CVLAN)) {
1507 struct flow_dissector_key_vlan *key =
1508 skb_flow_dissector_target(f->dissector,
1509 FLOW_DISSECTOR_KEY_CVLAN,
1511 struct flow_dissector_key_vlan *mask =
1512 skb_flow_dissector_target(f->dissector,
1513 FLOW_DISSECTOR_KEY_CVLAN,
1515 if (mask->vlan_id || mask->vlan_priority || mask->vlan_tpid) {
1516 if (key->vlan_tpid == htons(ETH_P_8021AD)) {
1517 MLX5_SET(fte_match_set_misc, misc_c,
1518 outer_second_svlan_tag, 1);
1519 MLX5_SET(fte_match_set_misc, misc_v,
1520 outer_second_svlan_tag, 1);
1522 MLX5_SET(fte_match_set_misc, misc_c,
1523 outer_second_cvlan_tag, 1);
1524 MLX5_SET(fte_match_set_misc, misc_v,
1525 outer_second_cvlan_tag, 1);
1528 MLX5_SET(fte_match_set_misc, misc_c, outer_second_vid,
1530 MLX5_SET(fte_match_set_misc, misc_v, outer_second_vid,
1532 MLX5_SET(fte_match_set_misc, misc_c, outer_second_prio,
1533 mask->vlan_priority);
1534 MLX5_SET(fte_match_set_misc, misc_v, outer_second_prio,
1535 key->vlan_priority);
1537 *match_level = MLX5_MATCH_L2;
1541 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
1542 struct flow_dissector_key_eth_addrs *key =
1543 skb_flow_dissector_target(f->dissector,
1544 FLOW_DISSECTOR_KEY_ETH_ADDRS,
1546 struct flow_dissector_key_eth_addrs *mask =
1547 skb_flow_dissector_target(f->dissector,
1548 FLOW_DISSECTOR_KEY_ETH_ADDRS,
1551 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1554 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1558 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1561 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1565 if (!is_zero_ether_addr(mask->src) || !is_zero_ether_addr(mask->dst))
1566 *match_level = MLX5_MATCH_L2;
1569 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
1570 struct flow_dissector_key_control *key =
1571 skb_flow_dissector_target(f->dissector,
1572 FLOW_DISSECTOR_KEY_CONTROL,
1575 struct flow_dissector_key_control *mask =
1576 skb_flow_dissector_target(f->dissector,
1577 FLOW_DISSECTOR_KEY_CONTROL,
1579 addr_type = key->addr_type;
1581 /* the HW doesn't support frag first/later */
1582 if (mask->flags & FLOW_DIS_FIRST_FRAG)
1585 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
1586 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1587 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
1588 key->flags & FLOW_DIS_IS_FRAGMENT);
1590 /* the HW doesn't need L3 inline to match on frag=no */
1591 if (!(key->flags & FLOW_DIS_IS_FRAGMENT))
1592 *match_level = MLX5_MATCH_L2;
1593 /* *** L2 attributes parsing up to here *** */
1595 *match_level = MLX5_MATCH_L3;
1599 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1600 struct flow_dissector_key_basic *key =
1601 skb_flow_dissector_target(f->dissector,
1602 FLOW_DISSECTOR_KEY_BASIC,
1604 struct flow_dissector_key_basic *mask =
1605 skb_flow_dissector_target(f->dissector,
1606 FLOW_DISSECTOR_KEY_BASIC,
1608 ip_proto = key->ip_proto;
1610 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
1612 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
1616 *match_level = MLX5_MATCH_L3;
1619 if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1620 struct flow_dissector_key_ipv4_addrs *key =
1621 skb_flow_dissector_target(f->dissector,
1622 FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1624 struct flow_dissector_key_ipv4_addrs *mask =
1625 skb_flow_dissector_target(f->dissector,
1626 FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1629 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1630 src_ipv4_src_ipv6.ipv4_layout.ipv4),
1631 &mask->src, sizeof(mask->src));
1632 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1633 src_ipv4_src_ipv6.ipv4_layout.ipv4),
1634 &key->src, sizeof(key->src));
1635 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1636 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1637 &mask->dst, sizeof(mask->dst));
1638 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1639 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1640 &key->dst, sizeof(key->dst));
1642 if (mask->src || mask->dst)
1643 *match_level = MLX5_MATCH_L3;
1646 if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1647 struct flow_dissector_key_ipv6_addrs *key =
1648 skb_flow_dissector_target(f->dissector,
1649 FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1651 struct flow_dissector_key_ipv6_addrs *mask =
1652 skb_flow_dissector_target(f->dissector,
1653 FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1656 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1657 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1658 &mask->src, sizeof(mask->src));
1659 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1660 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1661 &key->src, sizeof(key->src));
1663 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1664 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1665 &mask->dst, sizeof(mask->dst));
1666 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1667 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1668 &key->dst, sizeof(key->dst));
1670 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
1671 ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
1672 *match_level = MLX5_MATCH_L3;
1675 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_IP)) {
1676 struct flow_dissector_key_ip *key =
1677 skb_flow_dissector_target(f->dissector,
1678 FLOW_DISSECTOR_KEY_IP,
1680 struct flow_dissector_key_ip *mask =
1681 skb_flow_dissector_target(f->dissector,
1682 FLOW_DISSECTOR_KEY_IP,
1685 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn, mask->tos & 0x3);
1686 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, key->tos & 0x3);
1688 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp, mask->tos >> 2);
1689 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, key->tos >> 2);
1691 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit, mask->ttl);
1692 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit, key->ttl);
1695 !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
1696 ft_field_support.outer_ipv4_ttl)) {
1697 NL_SET_ERR_MSG_MOD(extack,
1698 "Matching on TTL is not supported");
1702 if (mask->tos || mask->ttl)
1703 *match_level = MLX5_MATCH_L3;
1706 /* *** L3 attributes parsing up to here *** */
1708 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
1709 struct flow_dissector_key_ports *key =
1710 skb_flow_dissector_target(f->dissector,
1711 FLOW_DISSECTOR_KEY_PORTS,
1713 struct flow_dissector_key_ports *mask =
1714 skb_flow_dissector_target(f->dissector,
1715 FLOW_DISSECTOR_KEY_PORTS,
1719 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1720 tcp_sport, ntohs(mask->src));
1721 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1722 tcp_sport, ntohs(key->src));
1724 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1725 tcp_dport, ntohs(mask->dst));
1726 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1727 tcp_dport, ntohs(key->dst));
1731 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1732 udp_sport, ntohs(mask->src));
1733 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1734 udp_sport, ntohs(key->src));
1736 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1737 udp_dport, ntohs(mask->dst));
1738 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1739 udp_dport, ntohs(key->dst));
1742 NL_SET_ERR_MSG_MOD(extack,
1743 "Only UDP and TCP transports are supported for L4 matching");
1744 netdev_err(priv->netdev,
1745 "Only UDP and TCP transport are supported\n");
1749 if (mask->src || mask->dst)
1750 *match_level = MLX5_MATCH_L4;
1753 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_TCP)) {
1754 struct flow_dissector_key_tcp *key =
1755 skb_flow_dissector_target(f->dissector,
1756 FLOW_DISSECTOR_KEY_TCP,
1758 struct flow_dissector_key_tcp *mask =
1759 skb_flow_dissector_target(f->dissector,
1760 FLOW_DISSECTOR_KEY_TCP,
1763 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
1764 ntohs(mask->flags));
1765 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
1769 *match_level = MLX5_MATCH_L4;
1775 static int parse_cls_flower(struct mlx5e_priv *priv,
1776 struct mlx5e_tc_flow *flow,
1777 struct mlx5_flow_spec *spec,
1778 struct tc_cls_flower_offload *f)
1780 struct netlink_ext_ack *extack = f->common.extack;
1781 struct mlx5_core_dev *dev = priv->mdev;
1782 struct mlx5_eswitch *esw = dev->priv.eswitch;
1783 struct mlx5e_rep_priv *rpriv = priv->ppriv;
1784 struct mlx5_eswitch_rep *rep;
1788 err = __parse_cls_flower(priv, spec, f, &match_level);
1790 if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH)) {
1792 if (rep->vport != FDB_UPLINK_VPORT &&
1793 (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
1794 esw->offloads.inline_mode < match_level)) {
1795 NL_SET_ERR_MSG_MOD(extack,
1796 "Flow is not offloaded due to min inline setting");
1797 netdev_warn(priv->netdev,
1798 "Flow is not offloaded due to min inline setting, required %d actual %d\n",
1799 match_level, esw->offloads.inline_mode);
1804 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1805 flow->esw_attr->match_level = match_level;
1807 flow->nic_attr->match_level = match_level;
1812 struct pedit_headers {
1820 static int pedit_header_offsets[] = {
1821 [TCA_PEDIT_KEY_EX_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
1822 [TCA_PEDIT_KEY_EX_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
1823 [TCA_PEDIT_KEY_EX_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
1824 [TCA_PEDIT_KEY_EX_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
1825 [TCA_PEDIT_KEY_EX_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
1828 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
1830 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
1831 struct pedit_headers *masks,
1832 struct pedit_headers *vals)
1834 u32 *curr_pmask, *curr_pval;
1836 if (hdr_type >= __PEDIT_HDR_TYPE_MAX)
1839 curr_pmask = (u32 *)(pedit_header(masks, hdr_type) + offset);
1840 curr_pval = (u32 *)(pedit_header(vals, hdr_type) + offset);
1842 if (*curr_pmask & mask) /* disallow acting twice on the same location */
1845 *curr_pmask |= mask;
1846 *curr_pval |= (val & mask);
1854 struct mlx5_fields {
1860 #define OFFLOAD(fw_field, size, field, off) \
1861 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, size, offsetof(struct pedit_headers, field) + (off)}
1863 static struct mlx5_fields fields[] = {
1864 OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0),
1865 OFFLOAD(DMAC_15_0, 2, eth.h_dest[4], 0),
1866 OFFLOAD(SMAC_47_16, 4, eth.h_source[0], 0),
1867 OFFLOAD(SMAC_15_0, 2, eth.h_source[4], 0),
1868 OFFLOAD(ETHERTYPE, 2, eth.h_proto, 0),
1870 OFFLOAD(IP_TTL, 1, ip4.ttl, 0),
1871 OFFLOAD(SIPV4, 4, ip4.saddr, 0),
1872 OFFLOAD(DIPV4, 4, ip4.daddr, 0),
1874 OFFLOAD(SIPV6_127_96, 4, ip6.saddr.s6_addr32[0], 0),
1875 OFFLOAD(SIPV6_95_64, 4, ip6.saddr.s6_addr32[1], 0),
1876 OFFLOAD(SIPV6_63_32, 4, ip6.saddr.s6_addr32[2], 0),
1877 OFFLOAD(SIPV6_31_0, 4, ip6.saddr.s6_addr32[3], 0),
1878 OFFLOAD(DIPV6_127_96, 4, ip6.daddr.s6_addr32[0], 0),
1879 OFFLOAD(DIPV6_95_64, 4, ip6.daddr.s6_addr32[1], 0),
1880 OFFLOAD(DIPV6_63_32, 4, ip6.daddr.s6_addr32[2], 0),
1881 OFFLOAD(DIPV6_31_0, 4, ip6.daddr.s6_addr32[3], 0),
1882 OFFLOAD(IPV6_HOPLIMIT, 1, ip6.hop_limit, 0),
1884 OFFLOAD(TCP_SPORT, 2, tcp.source, 0),
1885 OFFLOAD(TCP_DPORT, 2, tcp.dest, 0),
1886 OFFLOAD(TCP_FLAGS, 1, tcp.ack_seq, 5),
1888 OFFLOAD(UDP_SPORT, 2, udp.source, 0),
1889 OFFLOAD(UDP_DPORT, 2, udp.dest, 0),
1892 /* On input attr->num_mod_hdr_actions tells how many HW actions can be parsed at
1893 * max from the SW pedit action. On success, it says how many HW actions were
1896 static int offload_pedit_fields(struct pedit_headers *masks,
1897 struct pedit_headers *vals,
1898 struct mlx5e_tc_flow_parse_attr *parse_attr,
1899 struct netlink_ext_ack *extack)
1901 struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
1902 int i, action_size, nactions, max_actions, first, last, next_z;
1903 void *s_masks_p, *a_masks_p, *vals_p;
1904 struct mlx5_fields *f;
1905 u8 cmd, field_bsize;
1912 set_masks = &masks[TCA_PEDIT_KEY_EX_CMD_SET];
1913 add_masks = &masks[TCA_PEDIT_KEY_EX_CMD_ADD];
1914 set_vals = &vals[TCA_PEDIT_KEY_EX_CMD_SET];
1915 add_vals = &vals[TCA_PEDIT_KEY_EX_CMD_ADD];
1917 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1918 action = parse_attr->mod_hdr_actions;
1919 max_actions = parse_attr->num_mod_hdr_actions;
1922 for (i = 0; i < ARRAY_SIZE(fields); i++) {
1924 /* avoid seeing bits set from previous iterations */
1928 s_masks_p = (void *)set_masks + f->offset;
1929 a_masks_p = (void *)add_masks + f->offset;
1931 memcpy(&s_mask, s_masks_p, f->size);
1932 memcpy(&a_mask, a_masks_p, f->size);
1934 if (!s_mask && !a_mask) /* nothing to offload here */
1937 if (s_mask && a_mask) {
1938 NL_SET_ERR_MSG_MOD(extack,
1939 "can't set and add to the same HW field");
1940 printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
1944 if (nactions == max_actions) {
1945 NL_SET_ERR_MSG_MOD(extack,
1946 "too many pedit actions, can't offload");
1947 printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
1952 cmd = MLX5_ACTION_TYPE_SET;
1954 vals_p = (void *)set_vals + f->offset;
1955 /* clear to denote we consumed this field */
1956 memset(s_masks_p, 0, f->size);
1958 cmd = MLX5_ACTION_TYPE_ADD;
1960 vals_p = (void *)add_vals + f->offset;
1961 /* clear to denote we consumed this field */
1962 memset(a_masks_p, 0, f->size);
1965 field_bsize = f->size * BITS_PER_BYTE;
1967 if (field_bsize == 32) {
1968 mask_be32 = *(__be32 *)&mask;
1969 mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
1970 } else if (field_bsize == 16) {
1971 mask_be16 = *(__be16 *)&mask;
1972 mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
1975 first = find_first_bit(&mask, field_bsize);
1976 next_z = find_next_zero_bit(&mask, field_bsize, first);
1977 last = find_last_bit(&mask, field_bsize);
1978 if (first < next_z && next_z < last) {
1979 NL_SET_ERR_MSG_MOD(extack,
1980 "rewrite of few sub-fields isn't supported");
1981 printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
1986 MLX5_SET(set_action_in, action, action_type, cmd);
1987 MLX5_SET(set_action_in, action, field, f->field);
1989 if (cmd == MLX5_ACTION_TYPE_SET) {
1990 MLX5_SET(set_action_in, action, offset, first);
1991 /* length is num of bits to be written, zero means length of 32 */
1992 MLX5_SET(set_action_in, action, length, (last - first + 1));
1995 if (field_bsize == 32)
1996 MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
1997 else if (field_bsize == 16)
1998 MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
1999 else if (field_bsize == 8)
2000 MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
2002 action += action_size;
2006 parse_attr->num_mod_hdr_actions = nactions;
2010 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
2011 const struct tc_action *a, int namespace,
2012 struct mlx5e_tc_flow_parse_attr *parse_attr)
2014 int nkeys, action_size, max_actions;
2016 nkeys = tcf_pedit_nkeys(a);
2017 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2019 if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
2020 max_actions = MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev, max_modify_header_actions);
2021 else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
2022 max_actions = MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, max_modify_header_actions);
2024 /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
2025 max_actions = min(max_actions, nkeys * 16);
2027 parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
2028 if (!parse_attr->mod_hdr_actions)
2031 parse_attr->num_mod_hdr_actions = max_actions;
2035 static const struct pedit_headers zero_masks = {};
2037 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
2038 const struct tc_action *a, int namespace,
2039 struct mlx5e_tc_flow_parse_attr *parse_attr,
2040 struct netlink_ext_ack *extack)
2042 struct pedit_headers masks[__PEDIT_CMD_MAX], vals[__PEDIT_CMD_MAX], *cmd_masks;
2043 int nkeys, i, err = -EOPNOTSUPP;
2044 u32 mask, val, offset;
2047 nkeys = tcf_pedit_nkeys(a);
2049 memset(masks, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
2050 memset(vals, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
2052 for (i = 0; i < nkeys; i++) {
2053 htype = tcf_pedit_htype(a, i);
2054 cmd = tcf_pedit_cmd(a, i);
2055 err = -EOPNOTSUPP; /* can't be all optimistic */
2057 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK) {
2058 NL_SET_ERR_MSG_MOD(extack,
2059 "legacy pedit isn't offloaded");
2063 if (cmd != TCA_PEDIT_KEY_EX_CMD_SET && cmd != TCA_PEDIT_KEY_EX_CMD_ADD) {
2064 NL_SET_ERR_MSG_MOD(extack, "pedit cmd isn't offloaded");
2068 mask = tcf_pedit_mask(a, i);
2069 val = tcf_pedit_val(a, i);
2070 offset = tcf_pedit_offset(a, i);
2072 err = set_pedit_val(htype, ~mask, val, offset, &masks[cmd], &vals[cmd]);
2077 err = alloc_mod_hdr_actions(priv, a, namespace, parse_attr);
2081 err = offload_pedit_fields(masks, vals, parse_attr, extack);
2083 goto out_dealloc_parsed_actions;
2085 for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
2086 cmd_masks = &masks[cmd];
2087 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
2088 NL_SET_ERR_MSG_MOD(extack,
2089 "attempt to offload an unsupported field");
2090 netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
2091 print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
2092 16, 1, cmd_masks, sizeof(zero_masks), true);
2094 goto out_dealloc_parsed_actions;
2100 out_dealloc_parsed_actions:
2101 kfree(parse_attr->mod_hdr_actions);
2106 static bool csum_offload_supported(struct mlx5e_priv *priv,
2109 struct netlink_ext_ack *extack)
2111 u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
2112 TCA_CSUM_UPDATE_FLAG_UDP;
2114 /* The HW recalcs checksums only if re-writing headers */
2115 if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
2116 NL_SET_ERR_MSG_MOD(extack,
2117 "TC csum action is only offloaded with pedit");
2118 netdev_warn(priv->netdev,
2119 "TC csum action is only offloaded with pedit\n");
2123 if (update_flags & ~prot_flags) {
2124 NL_SET_ERR_MSG_MOD(extack,
2125 "can't offload TC csum action for some header/s");
2126 netdev_warn(priv->netdev,
2127 "can't offload TC csum action for some header/s - flags %#x\n",
2135 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
2136 struct tcf_exts *exts,
2137 struct netlink_ext_ack *extack)
2139 const struct tc_action *a;
2140 bool modify_ip_header;
2147 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
2148 ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
2150 /* for non-IP we only re-write MACs, so we're okay */
2151 if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
2154 modify_ip_header = false;
2155 tcf_exts_for_each_action(i, a, exts) {
2158 if (!is_tcf_pedit(a))
2161 nkeys = tcf_pedit_nkeys(a);
2162 for (k = 0; k < nkeys; k++) {
2163 htype = tcf_pedit_htype(a, k);
2164 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP4 ||
2165 htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP6) {
2166 modify_ip_header = true;
2172 ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
2173 if (modify_ip_header && ip_proto != IPPROTO_TCP &&
2174 ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
2175 NL_SET_ERR_MSG_MOD(extack,
2176 "can't offload re-write of non TCP/UDP");
2177 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
2185 static bool actions_match_supported(struct mlx5e_priv *priv,
2186 struct tcf_exts *exts,
2187 struct mlx5e_tc_flow_parse_attr *parse_attr,
2188 struct mlx5e_tc_flow *flow,
2189 struct netlink_ext_ack *extack)
2193 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
2194 actions = flow->esw_attr->action;
2196 actions = flow->nic_attr->action;
2198 if (flow->flags & MLX5E_TC_FLOW_EGRESS &&
2199 !(actions & MLX5_FLOW_CONTEXT_ACTION_DECAP))
2202 if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
2203 return modify_header_match_supported(&parse_attr->spec, exts,
2209 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
2211 struct mlx5_core_dev *fmdev, *pmdev;
2212 u64 fsystem_guid, psystem_guid;
2215 pmdev = peer_priv->mdev;
2217 fsystem_guid = mlx5_query_nic_system_image_guid(fmdev);
2218 psystem_guid = mlx5_query_nic_system_image_guid(pmdev);
2220 return (fsystem_guid == psystem_guid);
2223 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
2224 struct mlx5e_tc_flow_parse_attr *parse_attr,
2225 struct mlx5e_tc_flow *flow,
2226 struct netlink_ext_ack *extack)
2228 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
2229 const struct tc_action *a;
2234 if (!tcf_exts_has_actions(exts))
2237 attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
2239 tcf_exts_for_each_action(i, a, exts) {
2240 if (is_tcf_gact_shot(a)) {
2241 action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
2242 if (MLX5_CAP_FLOWTABLE(priv->mdev,
2243 flow_table_properties_nic_receive.flow_counter))
2244 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
2248 if (is_tcf_pedit(a)) {
2249 err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_KERNEL,
2250 parse_attr, extack);
2254 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
2255 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2259 if (is_tcf_csum(a)) {
2260 if (csum_offload_supported(priv, action,
2261 tcf_csum_update_flags(a),
2268 if (is_tcf_mirred_egress_redirect(a)) {
2269 struct net_device *peer_dev = tcf_mirred_dev(a);
2271 if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
2272 same_hw_devs(priv, netdev_priv(peer_dev))) {
2273 parse_attr->mirred_ifindex = peer_dev->ifindex;
2274 flow->flags |= MLX5E_TC_FLOW_HAIRPIN;
2275 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2276 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2278 NL_SET_ERR_MSG_MOD(extack,
2279 "device is not on same HW, can't offload");
2280 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
2287 if (is_tcf_skbedit_mark(a)) {
2288 u32 mark = tcf_skbedit_mark(a);
2290 if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
2291 NL_SET_ERR_MSG_MOD(extack,
2292 "Bad flow mark - only 16 bit is supported");
2296 attr->flow_tag = mark;
2297 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2304 attr->action = action;
2305 if (!actions_match_supported(priv, exts, parse_attr, flow, extack))
2311 static inline int cmp_encap_info(struct ip_tunnel_key *a,
2312 struct ip_tunnel_key *b)
2314 return memcmp(a, b, sizeof(*a));
2317 static inline int hash_encap_info(struct ip_tunnel_key *key)
2319 return jhash(key, sizeof(*key), 0);
2322 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
2323 struct net_device *mirred_dev,
2324 struct net_device **out_dev,
2326 struct neighbour **out_n,
2329 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2330 struct mlx5e_rep_priv *uplink_rpriv;
2332 struct neighbour *n = NULL;
2334 #if IS_ENABLED(CONFIG_INET)
2337 rt = ip_route_output_key(dev_net(mirred_dev), fl4);
2338 ret = PTR_ERR_OR_ZERO(rt);
2344 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2345 /* if the egress device isn't on the same HW e-switch, we use the uplink */
2346 if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
2347 *out_dev = uplink_rpriv->netdev;
2349 *out_dev = rt->dst.dev;
2352 *out_ttl = ip4_dst_hoplimit(&rt->dst);
2353 n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
2362 static bool is_merged_eswitch_dev(struct mlx5e_priv *priv,
2363 struct net_device *peer_netdev)
2365 struct mlx5e_priv *peer_priv;
2367 peer_priv = netdev_priv(peer_netdev);
2369 return (MLX5_CAP_ESW(priv->mdev, merged_eswitch) &&
2370 (priv->netdev->netdev_ops == peer_netdev->netdev_ops) &&
2371 same_hw_devs(priv, peer_priv) &&
2372 MLX5_VPORT_MANAGER(peer_priv->mdev) &&
2373 (peer_priv->mdev->priv.eswitch->mode == SRIOV_OFFLOADS));
2376 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
2377 struct net_device *mirred_dev,
2378 struct net_device **out_dev,
2380 struct neighbour **out_n,
2383 struct neighbour *n = NULL;
2384 struct dst_entry *dst;
2386 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
2387 struct mlx5e_rep_priv *uplink_rpriv;
2388 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2391 ret = ipv6_stub->ipv6_dst_lookup(dev_net(mirred_dev), NULL, &dst,
2397 *out_ttl = ip6_dst_hoplimit(dst);
2399 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2400 /* if the egress device isn't on the same HW e-switch, we use the uplink */
2401 if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
2402 *out_dev = uplink_rpriv->netdev;
2404 *out_dev = dst->dev;
2409 n = dst_neigh_lookup(dst, &fl6->daddr);
2418 static void gen_vxlan_header_ipv4(struct net_device *out_dev,
2419 char buf[], int encap_size,
2420 unsigned char h_dest[ETH_ALEN],
2424 __be16 udp_dst_port,
2427 struct ethhdr *eth = (struct ethhdr *)buf;
2428 struct iphdr *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
2429 struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
2430 struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2432 memset(buf, 0, encap_size);
2434 ether_addr_copy(eth->h_dest, h_dest);
2435 ether_addr_copy(eth->h_source, out_dev->dev_addr);
2436 eth->h_proto = htons(ETH_P_IP);
2443 ip->protocol = IPPROTO_UDP;
2447 udp->dest = udp_dst_port;
2448 vxh->vx_flags = VXLAN_HF_VNI;
2449 vxh->vx_vni = vxlan_vni_field(vx_vni);
2452 static void gen_vxlan_header_ipv6(struct net_device *out_dev,
2453 char buf[], int encap_size,
2454 unsigned char h_dest[ETH_ALEN],
2456 struct in6_addr *daddr,
2457 struct in6_addr *saddr,
2458 __be16 udp_dst_port,
2461 struct ethhdr *eth = (struct ethhdr *)buf;
2462 struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
2463 struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
2464 struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2466 memset(buf, 0, encap_size);
2468 ether_addr_copy(eth->h_dest, h_dest);
2469 ether_addr_copy(eth->h_source, out_dev->dev_addr);
2470 eth->h_proto = htons(ETH_P_IPV6);
2472 ip6_flow_hdr(ip6h, tos, 0);
2473 /* the HW fills up ipv6 payload len */
2474 ip6h->nexthdr = IPPROTO_UDP;
2475 ip6h->hop_limit = ttl;
2476 ip6h->daddr = *daddr;
2477 ip6h->saddr = *saddr;
2479 udp->dest = udp_dst_port;
2480 vxh->vx_flags = VXLAN_HF_VNI;
2481 vxh->vx_vni = vxlan_vni_field(vx_vni);
2484 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
2485 struct net_device *mirred_dev,
2486 struct mlx5e_encap_entry *e)
2488 int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2489 int ipv4_encap_size = ETH_HLEN + sizeof(struct iphdr) + VXLAN_HLEN;
2490 struct ip_tunnel_key *tun_key = &e->tun_info.key;
2491 struct net_device *out_dev;
2492 struct neighbour *n = NULL;
2493 struct flowi4 fl4 = {};
2494 u8 nud_state, tos, ttl;
2498 if (max_encap_size < ipv4_encap_size) {
2499 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2500 ipv4_encap_size, max_encap_size);
2504 encap_header = kzalloc(ipv4_encap_size, GFP_KERNEL);
2508 switch (e->tunnel_type) {
2509 case MLX5_REFORMAT_TYPE_L2_TO_VXLAN:
2510 fl4.flowi4_proto = IPPROTO_UDP;
2511 fl4.fl4_dport = tun_key->tp_dst;
2521 fl4.flowi4_tos = tun_key->tos;
2522 fl4.daddr = tun_key->u.ipv4.dst;
2523 fl4.saddr = tun_key->u.ipv4.src;
2525 err = mlx5e_route_lookup_ipv4(priv, mirred_dev, &out_dev,
2530 /* used by mlx5e_detach_encap to lookup a neigh hash table
2531 * entry in the neigh hash table when a user deletes a rule
2533 e->m_neigh.dev = n->dev;
2534 e->m_neigh.family = n->ops->family;
2535 memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2536 e->out_dev = out_dev;
2538 /* It's importent to add the neigh to the hash table before checking
2539 * the neigh validity state. So if we'll get a notification, in case the
2540 * neigh changes it's validity state, we would find the relevant neigh
2543 err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2547 read_lock_bh(&n->lock);
2548 nud_state = n->nud_state;
2549 ether_addr_copy(e->h_dest, n->ha);
2550 read_unlock_bh(&n->lock);
2552 switch (e->tunnel_type) {
2553 case MLX5_REFORMAT_TYPE_L2_TO_VXLAN:
2554 gen_vxlan_header_ipv4(out_dev, encap_header,
2555 ipv4_encap_size, e->h_dest, tos, ttl,
2557 fl4.saddr, tun_key->tp_dst,
2558 tunnel_id_to_key32(tun_key->tun_id));
2562 goto destroy_neigh_entry;
2564 e->encap_size = ipv4_encap_size;
2565 e->encap_header = encap_header;
2567 if (!(nud_state & NUD_VALID)) {
2568 neigh_event_send(n, NULL);
2573 err = mlx5_packet_reformat_alloc(priv->mdev, e->tunnel_type,
2574 ipv4_encap_size, encap_header,
2575 MLX5_FLOW_NAMESPACE_FDB,
2578 goto destroy_neigh_entry;
2580 e->flags |= MLX5_ENCAP_ENTRY_VALID;
2581 mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2585 destroy_neigh_entry:
2586 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2588 kfree(encap_header);
2595 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
2596 struct net_device *mirred_dev,
2597 struct mlx5e_encap_entry *e)
2599 int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2600 int ipv6_encap_size = ETH_HLEN + sizeof(struct ipv6hdr) + VXLAN_HLEN;
2601 struct ip_tunnel_key *tun_key = &e->tun_info.key;
2602 struct net_device *out_dev;
2603 struct neighbour *n = NULL;
2604 struct flowi6 fl6 = {};
2605 u8 nud_state, tos, ttl;
2609 if (max_encap_size < ipv6_encap_size) {
2610 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2611 ipv6_encap_size, max_encap_size);
2615 encap_header = kzalloc(ipv6_encap_size, GFP_KERNEL);
2619 switch (e->tunnel_type) {
2620 case MLX5_REFORMAT_TYPE_L2_TO_VXLAN:
2621 fl6.flowi6_proto = IPPROTO_UDP;
2622 fl6.fl6_dport = tun_key->tp_dst;
2632 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
2633 fl6.daddr = tun_key->u.ipv6.dst;
2634 fl6.saddr = tun_key->u.ipv6.src;
2636 err = mlx5e_route_lookup_ipv6(priv, mirred_dev, &out_dev,
2641 /* used by mlx5e_detach_encap to lookup a neigh hash table
2642 * entry in the neigh hash table when a user deletes a rule
2644 e->m_neigh.dev = n->dev;
2645 e->m_neigh.family = n->ops->family;
2646 memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2647 e->out_dev = out_dev;
2649 /* It's importent to add the neigh to the hash table before checking
2650 * the neigh validity state. So if we'll get a notification, in case the
2651 * neigh changes it's validity state, we would find the relevant neigh
2654 err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2658 read_lock_bh(&n->lock);
2659 nud_state = n->nud_state;
2660 ether_addr_copy(e->h_dest, n->ha);
2661 read_unlock_bh(&n->lock);
2663 switch (e->tunnel_type) {
2664 case MLX5_REFORMAT_TYPE_L2_TO_VXLAN:
2665 gen_vxlan_header_ipv6(out_dev, encap_header,
2666 ipv6_encap_size, e->h_dest, tos, ttl,
2668 &fl6.saddr, tun_key->tp_dst,
2669 tunnel_id_to_key32(tun_key->tun_id));
2673 goto destroy_neigh_entry;
2676 e->encap_size = ipv6_encap_size;
2677 e->encap_header = encap_header;
2679 if (!(nud_state & NUD_VALID)) {
2680 neigh_event_send(n, NULL);
2685 err = mlx5_packet_reformat_alloc(priv->mdev, e->tunnel_type,
2686 ipv6_encap_size, encap_header,
2687 MLX5_FLOW_NAMESPACE_FDB,
2690 goto destroy_neigh_entry;
2692 e->flags |= MLX5_ENCAP_ENTRY_VALID;
2693 mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2697 destroy_neigh_entry:
2698 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2700 kfree(encap_header);
2707 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
2708 struct ip_tunnel_info *tun_info,
2709 struct net_device *mirred_dev,
2710 struct net_device **encap_dev,
2711 struct mlx5e_tc_flow *flow,
2712 struct netlink_ext_ack *extack)
2714 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2715 unsigned short family = ip_tunnel_info_af(tun_info);
2716 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2717 struct ip_tunnel_key *key = &tun_info->key;
2718 struct mlx5e_encap_entry *e;
2719 int tunnel_type, err = 0;
2723 /* udp dst port must be set */
2724 if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
2725 goto vxlan_encap_offload_err;
2727 /* setting udp src port isn't supported */
2728 if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
2729 vxlan_encap_offload_err:
2730 NL_SET_ERR_MSG_MOD(extack,
2731 "must set udp dst port and not set udp src port");
2732 netdev_warn(priv->netdev,
2733 "must set udp dst port and not set udp src port\n");
2737 if (mlx5_vxlan_lookup_port(priv->mdev->vxlan, be16_to_cpu(key->tp_dst)) &&
2738 MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
2739 tunnel_type = MLX5_REFORMAT_TYPE_L2_TO_VXLAN;
2741 NL_SET_ERR_MSG_MOD(extack,
2742 "port isn't an offloaded vxlan udp dport");
2743 netdev_warn(priv->netdev,
2744 "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
2748 hash_key = hash_encap_info(key);
2750 hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
2751 encap_hlist, hash_key) {
2752 if (!cmp_encap_info(&e->tun_info.key, key)) {
2758 /* must verify if encap is valid or not */
2762 e = kzalloc(sizeof(*e), GFP_KERNEL);
2766 e->tun_info = *tun_info;
2767 e->tunnel_type = tunnel_type;
2768 INIT_LIST_HEAD(&e->flows);
2770 if (family == AF_INET)
2771 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e);
2772 else if (family == AF_INET6)
2773 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e);
2775 if (err && err != -EAGAIN)
2778 hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
2781 list_add(&flow->encap, &e->flows);
2782 *encap_dev = e->out_dev;
2783 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
2784 attr->encap_id = e->encap_id;
2795 static int parse_tc_vlan_action(struct mlx5e_priv *priv,
2796 const struct tc_action *a,
2797 struct mlx5_esw_flow_attr *attr,
2800 u8 vlan_idx = attr->total_vlan;
2802 if (vlan_idx >= MLX5_FS_VLAN_DEPTH)
2805 if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
2807 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
2808 MLX5_FS_VLAN_DEPTH))
2811 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2;
2813 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2815 } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
2816 attr->vlan_vid[vlan_idx] = tcf_vlan_push_vid(a);
2817 attr->vlan_prio[vlan_idx] = tcf_vlan_push_prio(a);
2818 attr->vlan_proto[vlan_idx] = tcf_vlan_push_proto(a);
2819 if (!attr->vlan_proto[vlan_idx])
2820 attr->vlan_proto[vlan_idx] = htons(ETH_P_8021Q);
2823 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
2824 MLX5_FS_VLAN_DEPTH))
2827 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2;
2829 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev, 1) &&
2830 (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q) ||
2831 tcf_vlan_push_prio(a)))
2834 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
2836 } else { /* action is TCA_VLAN_ACT_MODIFY */
2840 attr->total_vlan = vlan_idx + 1;
2845 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
2846 struct mlx5e_tc_flow_parse_attr *parse_attr,
2847 struct mlx5e_tc_flow *flow,
2848 struct netlink_ext_ack *extack)
2850 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2851 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2852 struct mlx5e_rep_priv *rpriv = priv->ppriv;
2853 struct ip_tunnel_info *info = NULL;
2854 const struct tc_action *a;
2860 if (!tcf_exts_has_actions(exts))
2863 attr->in_rep = rpriv->rep;
2864 attr->in_mdev = priv->mdev;
2866 tcf_exts_for_each_action(i, a, exts) {
2867 if (is_tcf_gact_shot(a)) {
2868 action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
2869 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2873 if (is_tcf_pedit(a)) {
2874 err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_FDB,
2875 parse_attr, extack);
2879 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2880 attr->mirror_count = attr->out_count;
2884 if (is_tcf_csum(a)) {
2885 if (csum_offload_supported(priv, action,
2886 tcf_csum_update_flags(a),
2893 if (is_tcf_mirred_egress_redirect(a) || is_tcf_mirred_egress_mirror(a)) {
2894 struct mlx5e_priv *out_priv;
2895 struct net_device *out_dev;
2897 out_dev = tcf_mirred_dev(a);
2899 if (attr->out_count >= MLX5_MAX_FLOW_FWD_VPORTS) {
2900 NL_SET_ERR_MSG_MOD(extack,
2901 "can't support more output ports, can't offload forwarding");
2902 pr_err("can't support more than %d output ports, can't offload forwarding\n",
2907 if (switchdev_port_same_parent_id(priv->netdev,
2909 is_merged_eswitch_dev(priv, out_dev)) {
2910 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2911 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2912 out_priv = netdev_priv(out_dev);
2913 rpriv = out_priv->ppriv;
2914 attr->out_rep[attr->out_count] = rpriv->rep;
2915 attr->out_mdev[attr->out_count++] = out_priv->mdev;
2917 parse_attr->mirred_ifindex = out_dev->ifindex;
2918 parse_attr->tun_info = *info;
2919 attr->parse_attr = parse_attr;
2920 action |= MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
2921 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2922 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2923 /* attr->out_rep is resolved when we handle encap */
2925 NL_SET_ERR_MSG_MOD(extack,
2926 "devices are not on same switch HW, can't offload forwarding");
2927 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
2928 priv->netdev->name, out_dev->name);
2934 if (is_tcf_tunnel_set(a)) {
2935 info = tcf_tunnel_info(a);
2940 attr->mirror_count = attr->out_count;
2944 if (is_tcf_vlan(a)) {
2945 err = parse_tc_vlan_action(priv, a, attr, &action);
2950 attr->mirror_count = attr->out_count;
2954 if (is_tcf_tunnel_release(a)) {
2955 action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
2959 if (is_tcf_gact_goto_chain(a)) {
2960 u32 dest_chain = tcf_gact_goto_chain_index(a);
2961 u32 max_chain = mlx5_eswitch_get_chain_range(esw);
2963 if (dest_chain <= attr->chain) {
2964 NL_SET_ERR_MSG(extack, "Goto earlier chain isn't supported");
2967 if (dest_chain > max_chain) {
2968 NL_SET_ERR_MSG(extack, "Requested destination chain is out of supported range");
2971 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2972 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2973 attr->dest_chain = dest_chain;
2981 attr->action = action;
2982 if (!actions_match_supported(priv, exts, parse_attr, flow, extack))
2985 if (attr->mirror_count > 0 && !mlx5_esw_has_fwd_fdb(priv->mdev)) {
2986 NL_SET_ERR_MSG_MOD(extack,
2987 "current firmware doesn't support split rule for port mirroring");
2988 netdev_warn_once(priv->netdev, "current firmware doesn't support split rule for port mirroring\n");
2995 static void get_flags(int flags, u16 *flow_flags)
2997 u16 __flow_flags = 0;
2999 if (flags & MLX5E_TC_INGRESS)
3000 __flow_flags |= MLX5E_TC_FLOW_INGRESS;
3001 if (flags & MLX5E_TC_EGRESS)
3002 __flow_flags |= MLX5E_TC_FLOW_EGRESS;
3004 *flow_flags = __flow_flags;
3007 static const struct rhashtable_params tc_ht_params = {
3008 .head_offset = offsetof(struct mlx5e_tc_flow, node),
3009 .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
3010 .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
3011 .automatic_shrinking = true,
3014 static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv)
3016 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3017 struct mlx5e_rep_priv *uplink_rpriv;
3019 if (MLX5_VPORT_MANAGER(priv->mdev) && esw->mode == SRIOV_OFFLOADS) {
3020 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
3021 return &uplink_rpriv->tc_ht;
3023 return &priv->fs.tc.ht;
3027 mlx5e_alloc_flow(struct mlx5e_priv *priv, int attr_size,
3028 struct tc_cls_flower_offload *f, u16 flow_flags,
3029 struct mlx5e_tc_flow_parse_attr **__parse_attr,
3030 struct mlx5e_tc_flow **__flow)
3032 struct mlx5e_tc_flow_parse_attr *parse_attr;
3033 struct mlx5e_tc_flow *flow;
3036 flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
3037 parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
3038 if (!parse_attr || !flow) {
3043 flow->cookie = f->cookie;
3044 flow->flags = flow_flags;
3047 err = parse_cls_flower(priv, flow, &parse_attr->spec, f);
3052 *__parse_attr = parse_attr;
3063 mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
3064 struct tc_cls_flower_offload *f,
3066 struct mlx5e_tc_flow **__flow)
3068 struct netlink_ext_ack *extack = f->common.extack;
3069 struct mlx5e_tc_flow_parse_attr *parse_attr;
3070 struct mlx5e_tc_flow *flow;
3073 flow_flags |= MLX5E_TC_FLOW_ESWITCH;
3074 attr_size = sizeof(struct mlx5_esw_flow_attr);
3075 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3076 &parse_attr, &flow);
3080 flow->esw_attr->chain = f->common.chain_index;
3081 flow->esw_attr->prio = TC_H_MAJ(f->common.prio) >> 16;
3082 err = parse_tc_fdb_actions(priv, f->exts, parse_attr, flow, extack);
3086 err = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow, extack);
3090 if (!(flow->esw_attr->action &
3091 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT))
3106 mlx5e_add_nic_flow(struct mlx5e_priv *priv,
3107 struct tc_cls_flower_offload *f,
3109 struct mlx5e_tc_flow **__flow)
3111 struct netlink_ext_ack *extack = f->common.extack;
3112 struct mlx5e_tc_flow_parse_attr *parse_attr;
3113 struct mlx5e_tc_flow *flow;
3116 /* multi-chain not supported for NIC rules */
3117 if (!tc_cls_can_offload_and_chain0(priv->netdev, &f->common))
3120 flow_flags |= MLX5E_TC_FLOW_NIC;
3121 attr_size = sizeof(struct mlx5_nic_flow_attr);
3122 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3123 &parse_attr, &flow);
3127 err = parse_tc_nic_actions(priv, f->exts, parse_attr, flow, extack);
3131 err = mlx5e_tc_add_nic_flow(priv, parse_attr, flow, extack);
3135 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
3149 mlx5e_tc_add_flow(struct mlx5e_priv *priv,
3150 struct tc_cls_flower_offload *f,
3152 struct mlx5e_tc_flow **flow)
3154 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3158 get_flags(flags, &flow_flags);
3160 if (!tc_can_offload_extack(priv->netdev, f->common.extack))
3163 if (esw && esw->mode == SRIOV_OFFLOADS)
3164 err = mlx5e_add_fdb_flow(priv, f, flow_flags, flow);
3166 err = mlx5e_add_nic_flow(priv, f, flow_flags, flow);
3171 int mlx5e_configure_flower(struct mlx5e_priv *priv,
3172 struct tc_cls_flower_offload *f, int flags)
3174 struct netlink_ext_ack *extack = f->common.extack;
3175 struct rhashtable *tc_ht = get_tc_ht(priv);
3176 struct mlx5e_tc_flow *flow;
3179 flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
3181 NL_SET_ERR_MSG_MOD(extack,
3182 "flow cookie already exists, ignoring");
3183 netdev_warn_once(priv->netdev,
3184 "flow cookie %lx already exists, ignoring\n",
3189 err = mlx5e_tc_add_flow(priv, f, flags, &flow);
3193 err = rhashtable_insert_fast(tc_ht, &flow->node, tc_ht_params);
3200 mlx5e_tc_del_flow(priv, flow);
3206 #define DIRECTION_MASK (MLX5E_TC_INGRESS | MLX5E_TC_EGRESS)
3207 #define FLOW_DIRECTION_MASK (MLX5E_TC_FLOW_INGRESS | MLX5E_TC_FLOW_EGRESS)
3209 static bool same_flow_direction(struct mlx5e_tc_flow *flow, int flags)
3211 if ((flow->flags & FLOW_DIRECTION_MASK) == (flags & DIRECTION_MASK))
3217 int mlx5e_delete_flower(struct mlx5e_priv *priv,
3218 struct tc_cls_flower_offload *f, int flags)
3220 struct rhashtable *tc_ht = get_tc_ht(priv);
3221 struct mlx5e_tc_flow *flow;
3223 flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
3224 if (!flow || !same_flow_direction(flow, flags))
3227 rhashtable_remove_fast(tc_ht, &flow->node, tc_ht_params);
3229 mlx5e_tc_del_flow(priv, flow);
3236 int mlx5e_stats_flower(struct mlx5e_priv *priv,
3237 struct tc_cls_flower_offload *f, int flags)
3239 struct rhashtable *tc_ht = get_tc_ht(priv);
3240 struct mlx5e_tc_flow *flow;
3241 struct mlx5_fc *counter;
3246 flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
3247 if (!flow || !same_flow_direction(flow, flags))
3250 if (!(flow->flags & MLX5E_TC_FLOW_OFFLOADED))
3253 counter = mlx5e_tc_get_counter(flow);
3257 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
3259 tcf_exts_stats_update(f->exts, bytes, packets, lastuse);
3264 static void mlx5e_tc_hairpin_update_dead_peer(struct mlx5e_priv *priv,
3265 struct mlx5e_priv *peer_priv)
3267 struct mlx5_core_dev *peer_mdev = peer_priv->mdev;
3268 struct mlx5e_hairpin_entry *hpe;
3272 if (!same_hw_devs(priv, peer_priv))
3275 peer_vhca_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
3277 hash_for_each(priv->fs.tc.hairpin_tbl, bkt, hpe, hairpin_hlist) {
3278 if (hpe->peer_vhca_id == peer_vhca_id)
3279 hpe->hp->pair->peer_gone = true;
3283 static int mlx5e_tc_netdev_event(struct notifier_block *this,
3284 unsigned long event, void *ptr)
3286 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3287 struct mlx5e_flow_steering *fs;
3288 struct mlx5e_priv *peer_priv;
3289 struct mlx5e_tc_table *tc;
3290 struct mlx5e_priv *priv;
3292 if (ndev->netdev_ops != &mlx5e_netdev_ops ||
3293 event != NETDEV_UNREGISTER ||
3294 ndev->reg_state == NETREG_REGISTERED)
3297 tc = container_of(this, struct mlx5e_tc_table, netdevice_nb);
3298 fs = container_of(tc, struct mlx5e_flow_steering, tc);
3299 priv = container_of(fs, struct mlx5e_priv, fs);
3300 peer_priv = netdev_priv(ndev);
3301 if (priv == peer_priv ||
3302 !(priv->netdev->features & NETIF_F_HW_TC))
3305 mlx5e_tc_hairpin_update_dead_peer(priv, peer_priv);
3310 int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
3312 struct mlx5e_tc_table *tc = &priv->fs.tc;
3315 hash_init(tc->mod_hdr_tbl);
3316 hash_init(tc->hairpin_tbl);
3318 err = rhashtable_init(&tc->ht, &tc_ht_params);
3322 tc->netdevice_nb.notifier_call = mlx5e_tc_netdev_event;
3323 if (register_netdevice_notifier(&tc->netdevice_nb)) {
3324 tc->netdevice_nb.notifier_call = NULL;
3325 mlx5_core_warn(priv->mdev, "Failed to register netdev notifier\n");
3331 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
3333 struct mlx5e_tc_flow *flow = ptr;
3334 struct mlx5e_priv *priv = flow->priv;
3336 mlx5e_tc_del_flow(priv, flow);
3340 void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
3342 struct mlx5e_tc_table *tc = &priv->fs.tc;
3344 if (tc->netdevice_nb.notifier_call)
3345 unregister_netdevice_notifier(&tc->netdevice_nb);
3347 rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, NULL);
3349 if (!IS_ERR_OR_NULL(tc->t)) {
3350 mlx5_destroy_flow_table(tc->t);
3355 int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
3357 return rhashtable_init(tc_ht, &tc_ht_params);
3360 void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
3362 rhashtable_free_and_destroy(tc_ht, _mlx5e_tc_del_flow, NULL);
3365 int mlx5e_tc_num_filters(struct mlx5e_priv *priv)
3367 struct rhashtable *tc_ht = get_tc_ht(priv);
3369 return atomic_read(&tc_ht->nelems);