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->channels.params.rss_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->channels.params, &ttconfig, tirc, false);
372 err = mlx5_core_create_tir(hp->func_mdev, in,
373 MLX5_ST_SZ_BYTES(create_tir_in), &hp->indir_tirn[tt]);
375 mlx5_core_warn(hp->func_mdev, "create indirect tirs failed, %d\n", err);
376 goto err_destroy_tirs;
382 for (i = 0; i < tt; i++)
383 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[i]);
387 static void mlx5e_hairpin_destroy_indirect_tirs(struct mlx5e_hairpin *hp)
391 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
392 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[tt]);
395 static void mlx5e_hairpin_set_ttc_params(struct mlx5e_hairpin *hp,
396 struct ttc_params *ttc_params)
398 struct mlx5_flow_table_attr *ft_attr = &ttc_params->ft_attr;
401 memset(ttc_params, 0, sizeof(*ttc_params));
403 ttc_params->any_tt_tirn = hp->tirn;
405 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
406 ttc_params->indir_tirn[tt] = hp->indir_tirn[tt];
408 ft_attr->max_fte = MLX5E_NUM_TT;
409 ft_attr->level = MLX5E_TC_TTC_FT_LEVEL;
410 ft_attr->prio = MLX5E_TC_PRIO;
413 static int mlx5e_hairpin_rss_init(struct mlx5e_hairpin *hp)
415 struct mlx5e_priv *priv = hp->func_priv;
416 struct ttc_params ttc_params;
419 err = mlx5e_hairpin_create_indirect_rqt(hp);
423 err = mlx5e_hairpin_create_indirect_tirs(hp);
425 goto err_create_indirect_tirs;
427 mlx5e_hairpin_set_ttc_params(hp, &ttc_params);
428 err = mlx5e_create_ttc_table(priv, &ttc_params, &hp->ttc);
430 goto err_create_ttc_table;
432 netdev_dbg(priv->netdev, "add hairpin: using %d channels rss ttc table id %x\n",
433 hp->num_channels, hp->ttc.ft.t->id);
437 err_create_ttc_table:
438 mlx5e_hairpin_destroy_indirect_tirs(hp);
439 err_create_indirect_tirs:
440 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
445 static void mlx5e_hairpin_rss_cleanup(struct mlx5e_hairpin *hp)
447 struct mlx5e_priv *priv = hp->func_priv;
449 mlx5e_destroy_ttc_table(priv, &hp->ttc);
450 mlx5e_hairpin_destroy_indirect_tirs(hp);
451 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
454 static struct mlx5e_hairpin *
455 mlx5e_hairpin_create(struct mlx5e_priv *priv, struct mlx5_hairpin_params *params,
458 struct mlx5_core_dev *func_mdev, *peer_mdev;
459 struct mlx5e_hairpin *hp;
460 struct mlx5_hairpin *pair;
463 hp = kzalloc(sizeof(*hp), GFP_KERNEL);
465 return ERR_PTR(-ENOMEM);
467 func_mdev = priv->mdev;
468 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
470 pair = mlx5_core_hairpin_create(func_mdev, peer_mdev, params);
473 goto create_pair_err;
476 hp->func_mdev = func_mdev;
477 hp->func_priv = priv;
478 hp->num_channels = params->num_channels;
480 err = mlx5e_hairpin_create_transport(hp);
482 goto create_transport_err;
484 if (hp->num_channels > 1) {
485 err = mlx5e_hairpin_rss_init(hp);
493 mlx5e_hairpin_destroy_transport(hp);
494 create_transport_err:
495 mlx5_core_hairpin_destroy(hp->pair);
501 static void mlx5e_hairpin_destroy(struct mlx5e_hairpin *hp)
503 if (hp->num_channels > 1)
504 mlx5e_hairpin_rss_cleanup(hp);
505 mlx5e_hairpin_destroy_transport(hp);
506 mlx5_core_hairpin_destroy(hp->pair);
510 static inline u32 hash_hairpin_info(u16 peer_vhca_id, u8 prio)
512 return (peer_vhca_id << 16 | prio);
515 static struct mlx5e_hairpin_entry *mlx5e_hairpin_get(struct mlx5e_priv *priv,
516 u16 peer_vhca_id, u8 prio)
518 struct mlx5e_hairpin_entry *hpe;
519 u32 hash_key = hash_hairpin_info(peer_vhca_id, prio);
521 hash_for_each_possible(priv->fs.tc.hairpin_tbl, hpe,
522 hairpin_hlist, hash_key) {
523 if (hpe->peer_vhca_id == peer_vhca_id && hpe->prio == prio)
530 #define UNKNOWN_MATCH_PRIO 8
532 static int mlx5e_hairpin_get_prio(struct mlx5e_priv *priv,
533 struct mlx5_flow_spec *spec, u8 *match_prio,
534 struct netlink_ext_ack *extack)
536 void *headers_c, *headers_v;
537 u8 prio_val, prio_mask = 0;
540 #ifdef CONFIG_MLX5_CORE_EN_DCB
541 if (priv->dcbx_dp.trust_state != MLX5_QPTS_TRUST_PCP) {
542 NL_SET_ERR_MSG_MOD(extack,
543 "only PCP trust state supported for hairpin");
547 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, outer_headers);
548 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
550 vlan_present = MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag);
552 prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
553 prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
556 if (!vlan_present || !prio_mask) {
557 prio_val = UNKNOWN_MATCH_PRIO;
558 } else if (prio_mask != 0x7) {
559 NL_SET_ERR_MSG_MOD(extack,
560 "masked priority match not supported for hairpin");
564 *match_prio = prio_val;
568 static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
569 struct mlx5e_tc_flow *flow,
570 struct mlx5e_tc_flow_parse_attr *parse_attr,
571 struct netlink_ext_ack *extack)
573 int peer_ifindex = parse_attr->mirred_ifindex;
574 struct mlx5_hairpin_params params;
575 struct mlx5_core_dev *peer_mdev;
576 struct mlx5e_hairpin_entry *hpe;
577 struct mlx5e_hairpin *hp;
584 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
585 if (!MLX5_CAP_GEN(priv->mdev, hairpin) || !MLX5_CAP_GEN(peer_mdev, hairpin)) {
586 NL_SET_ERR_MSG_MOD(extack, "hairpin is not supported");
590 peer_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
591 err = mlx5e_hairpin_get_prio(priv, &parse_attr->spec, &match_prio,
595 hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
599 hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
603 INIT_LIST_HEAD(&hpe->flows);
604 hpe->peer_vhca_id = peer_id;
605 hpe->prio = match_prio;
607 params.log_data_size = 15;
608 params.log_data_size = min_t(u8, params.log_data_size,
609 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_wq_data_sz));
610 params.log_data_size = max_t(u8, params.log_data_size,
611 MLX5_CAP_GEN(priv->mdev, log_min_hairpin_wq_data_sz));
613 params.log_num_packets = params.log_data_size -
614 MLX5_MPWRQ_MIN_LOG_STRIDE_SZ(priv->mdev);
615 params.log_num_packets = min_t(u8, params.log_num_packets,
616 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_num_packets));
618 params.q_counter = priv->q_counter;
619 /* set hairpin pair per each 50Gbs share of the link */
620 mlx5e_port_max_linkspeed(priv->mdev, &link_speed);
621 link_speed = max_t(u32, link_speed, 50000);
622 link_speed64 = link_speed;
623 do_div(link_speed64, 50000);
624 params.num_channels = link_speed64;
626 hp = mlx5e_hairpin_create(priv, ¶ms, peer_ifindex);
629 goto create_hairpin_err;
632 netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
633 hp->tirn, hp->pair->rqn[0], hp->pair->peer_mdev->priv.name,
634 hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
637 hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
638 hash_hairpin_info(peer_id, match_prio));
641 if (hpe->hp->num_channels > 1) {
642 flow->flags |= MLX5E_TC_FLOW_HAIRPIN_RSS;
643 flow->nic_attr->hairpin_ft = hpe->hp->ttc.ft.t;
645 flow->nic_attr->hairpin_tirn = hpe->hp->tirn;
647 list_add(&flow->hairpin, &hpe->flows);
656 static void mlx5e_hairpin_flow_del(struct mlx5e_priv *priv,
657 struct mlx5e_tc_flow *flow)
659 struct list_head *next = flow->hairpin.next;
661 list_del(&flow->hairpin);
663 /* no more hairpin flows for us, release the hairpin pair */
664 if (list_empty(next)) {
665 struct mlx5e_hairpin_entry *hpe;
667 hpe = list_entry(next, struct mlx5e_hairpin_entry, flows);
669 netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
670 hpe->hp->pair->peer_mdev->priv.name);
672 mlx5e_hairpin_destroy(hpe->hp);
673 hash_del(&hpe->hairpin_hlist);
679 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
680 struct mlx5e_tc_flow_parse_attr *parse_attr,
681 struct mlx5e_tc_flow *flow,
682 struct netlink_ext_ack *extack)
684 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
685 struct mlx5_core_dev *dev = priv->mdev;
686 struct mlx5_flow_destination dest[2] = {};
687 struct mlx5_flow_act flow_act = {
688 .action = attr->action,
689 .flow_tag = attr->flow_tag,
691 .flags = FLOW_ACT_HAS_TAG | FLOW_ACT_NO_APPEND,
693 struct mlx5_fc *counter = NULL;
694 bool table_created = false;
695 int err, dest_ix = 0;
697 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN) {
698 err = mlx5e_hairpin_flow_add(priv, flow, parse_attr, extack);
700 goto err_add_hairpin_flow;
702 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN_RSS) {
703 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
704 dest[dest_ix].ft = attr->hairpin_ft;
706 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_TIR;
707 dest[dest_ix].tir_num = attr->hairpin_tirn;
710 } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
711 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
712 dest[dest_ix].ft = priv->fs.vlan.ft.t;
716 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
717 counter = mlx5_fc_create(dev, true);
718 if (IS_ERR(counter)) {
719 err = PTR_ERR(counter);
722 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
723 dest[dest_ix].counter_id = mlx5_fc_id(counter);
725 attr->counter = counter;
728 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
729 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
730 flow_act.modify_id = attr->mod_hdr_id;
731 kfree(parse_attr->mod_hdr_actions);
733 goto err_create_mod_hdr_id;
736 if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
737 int tc_grp_size, tc_tbl_size;
738 u32 max_flow_counter;
740 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
741 MLX5_CAP_GEN(dev, max_flow_counter_15_0);
743 tc_grp_size = min_t(int, max_flow_counter, MLX5E_TC_TABLE_MAX_GROUP_SIZE);
745 tc_tbl_size = min_t(int, tc_grp_size * MLX5E_TC_TABLE_NUM_GROUPS,
746 BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev, log_max_ft_size)));
749 mlx5_create_auto_grouped_flow_table(priv->fs.ns,
752 MLX5E_TC_TABLE_NUM_GROUPS,
753 MLX5E_TC_FT_LEVEL, 0);
754 if (IS_ERR(priv->fs.tc.t)) {
755 NL_SET_ERR_MSG_MOD(extack,
756 "Failed to create tc offload table\n");
757 netdev_err(priv->netdev,
758 "Failed to create tc offload table\n");
759 err = PTR_ERR(priv->fs.tc.t);
763 table_created = true;
766 if (attr->match_level != MLX5_MATCH_NONE)
767 parse_attr->spec.match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
769 flow->rule[0] = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
770 &flow_act, dest, dest_ix);
772 if (IS_ERR(flow->rule[0])) {
773 err = PTR_ERR(flow->rule[0]);
781 mlx5_destroy_flow_table(priv->fs.tc.t);
782 priv->fs.tc.t = NULL;
785 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
786 mlx5e_detach_mod_hdr(priv, flow);
787 err_create_mod_hdr_id:
788 mlx5_fc_destroy(dev, counter);
790 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
791 mlx5e_hairpin_flow_del(priv, flow);
792 err_add_hairpin_flow:
796 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
797 struct mlx5e_tc_flow *flow)
799 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
800 struct mlx5_fc *counter = NULL;
802 counter = attr->counter;
803 mlx5_del_flow_rules(flow->rule[0]);
804 mlx5_fc_destroy(priv->mdev, counter);
806 if (!mlx5e_tc_num_filters(priv) && priv->fs.tc.t) {
807 mlx5_destroy_flow_table(priv->fs.tc.t);
808 priv->fs.tc.t = NULL;
811 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
812 mlx5e_detach_mod_hdr(priv, flow);
814 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
815 mlx5e_hairpin_flow_del(priv, flow);
818 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
819 struct mlx5e_tc_flow *flow);
821 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
822 struct ip_tunnel_info *tun_info,
823 struct net_device *mirred_dev,
824 struct net_device **encap_dev,
825 struct mlx5e_tc_flow *flow,
826 struct netlink_ext_ack *extack);
828 static struct mlx5_flow_handle *
829 mlx5e_tc_offload_fdb_rules(struct mlx5_eswitch *esw,
830 struct mlx5e_tc_flow *flow,
831 struct mlx5_flow_spec *spec,
832 struct mlx5_esw_flow_attr *attr)
834 struct mlx5_flow_handle *rule;
836 rule = mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
840 if (attr->mirror_count) {
841 flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, spec, attr);
842 if (IS_ERR(flow->rule[1])) {
843 mlx5_eswitch_del_offloaded_rule(esw, rule, attr);
844 return flow->rule[1];
848 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
853 mlx5e_tc_unoffload_fdb_rules(struct mlx5_eswitch *esw,
854 struct mlx5e_tc_flow *flow,
855 struct mlx5_esw_flow_attr *attr)
857 flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
859 if (attr->mirror_count)
860 mlx5_eswitch_del_fwd_rule(esw, flow->rule[1], attr);
862 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
865 static struct mlx5_flow_handle *
866 mlx5e_tc_offload_to_slow_path(struct mlx5_eswitch *esw,
867 struct mlx5e_tc_flow *flow,
868 struct mlx5_flow_spec *spec,
869 struct mlx5_esw_flow_attr *slow_attr)
871 struct mlx5_flow_handle *rule;
873 memcpy(slow_attr, flow->esw_attr, sizeof(*slow_attr));
874 slow_attr->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST,
875 slow_attr->mirror_count = 0,
876 slow_attr->dest_chain = FDB_SLOW_PATH_CHAIN,
878 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, slow_attr);
880 flow->flags |= MLX5E_TC_FLOW_SLOW;
886 mlx5e_tc_unoffload_from_slow_path(struct mlx5_eswitch *esw,
887 struct mlx5e_tc_flow *flow,
888 struct mlx5_esw_flow_attr *slow_attr)
890 memcpy(slow_attr, flow->esw_attr, sizeof(*slow_attr));
891 mlx5e_tc_unoffload_fdb_rules(esw, flow, slow_attr);
892 flow->flags &= ~MLX5E_TC_FLOW_SLOW;
896 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
897 struct mlx5e_tc_flow_parse_attr *parse_attr,
898 struct mlx5e_tc_flow *flow,
899 struct netlink_ext_ack *extack)
901 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
902 u32 max_chain = mlx5_eswitch_get_chain_range(esw);
903 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
904 u16 max_prio = mlx5_eswitch_get_prio_range(esw);
905 struct net_device *out_dev, *encap_dev = NULL;
906 struct mlx5_fc *counter = NULL;
907 struct mlx5e_rep_priv *rpriv;
908 struct mlx5e_priv *out_priv;
909 int err = 0, encap_err = 0;
911 /* if prios are not supported, keep the old behaviour of using same prio
912 * for all offloaded rules.
914 if (!mlx5_eswitch_prios_supported(esw))
917 if (attr->chain > max_chain) {
918 NL_SET_ERR_MSG(extack, "Requested chain is out of supported range");
920 goto err_max_prio_chain;
923 if (attr->prio > max_prio) {
924 NL_SET_ERR_MSG(extack, "Requested priority is out of supported range");
926 goto err_max_prio_chain;
929 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT) {
930 out_dev = __dev_get_by_index(dev_net(priv->netdev),
931 attr->parse_attr->mirred_ifindex);
932 encap_err = mlx5e_attach_encap(priv, &parse_attr->tun_info,
933 out_dev, &encap_dev, flow,
935 if (encap_err && encap_err != -EAGAIN) {
937 goto err_attach_encap;
939 out_priv = netdev_priv(encap_dev);
940 rpriv = out_priv->ppriv;
941 attr->out_rep[attr->out_count] = rpriv->rep;
942 attr->out_mdev[attr->out_count++] = out_priv->mdev;
945 err = mlx5_eswitch_add_vlan_action(esw, attr);
949 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
950 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
951 kfree(parse_attr->mod_hdr_actions);
956 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
957 counter = mlx5_fc_create(esw->dev, true);
958 if (IS_ERR(counter)) {
959 err = PTR_ERR(counter);
960 goto err_create_counter;
963 attr->counter = counter;
966 /* we get here if (1) there's no error or when
967 * (2) there's an encap action and we're on -EAGAIN (no valid neigh)
969 if (encap_err == -EAGAIN) {
970 /* continue with goto slow path rule instead */
971 struct mlx5_esw_flow_attr slow_attr;
973 flow->rule[0] = mlx5e_tc_offload_to_slow_path(esw, flow, &parse_attr->spec, &slow_attr);
975 flow->rule[0] = mlx5e_tc_offload_fdb_rules(esw, flow, &parse_attr->spec, attr);
978 if (IS_ERR(flow->rule[0])) {
979 err = PTR_ERR(flow->rule[0]);
986 mlx5_fc_destroy(esw->dev, counter);
988 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
989 mlx5e_detach_mod_hdr(priv, flow);
991 mlx5_eswitch_del_vlan_action(esw, attr);
993 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT)
994 mlx5e_detach_encap(priv, flow);
1000 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
1001 struct mlx5e_tc_flow *flow)
1003 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1004 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1005 struct mlx5_esw_flow_attr slow_attr;
1007 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
1008 if (flow->flags & MLX5E_TC_FLOW_SLOW)
1009 mlx5e_tc_unoffload_from_slow_path(esw, flow, &slow_attr);
1011 mlx5e_tc_unoffload_fdb_rules(esw, flow, attr);
1014 mlx5_eswitch_del_vlan_action(esw, attr);
1016 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT) {
1017 mlx5e_detach_encap(priv, flow);
1018 kvfree(attr->parse_attr);
1021 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1022 mlx5e_detach_mod_hdr(priv, flow);
1024 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
1025 mlx5_fc_destroy(esw->dev, attr->counter);
1028 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
1029 struct mlx5e_encap_entry *e)
1031 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1032 struct mlx5_esw_flow_attr slow_attr, *esw_attr;
1033 struct mlx5_flow_handle *rule;
1034 struct mlx5_flow_spec *spec;
1035 struct mlx5e_tc_flow *flow;
1038 err = mlx5_packet_reformat_alloc(priv->mdev, e->tunnel_type,
1039 e->encap_size, e->encap_header,
1040 MLX5_FLOW_NAMESPACE_FDB,
1043 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %d\n",
1047 e->flags |= MLX5_ENCAP_ENTRY_VALID;
1048 mlx5e_rep_queue_neigh_stats_work(priv);
1050 list_for_each_entry(flow, &e->flows, encap) {
1051 esw_attr = flow->esw_attr;
1052 esw_attr->encap_id = e->encap_id;
1053 spec = &esw_attr->parse_attr->spec;
1055 /* update from slow path rule to encap rule */
1056 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, esw_attr);
1058 err = PTR_ERR(rule);
1059 mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
1064 mlx5e_tc_unoffload_from_slow_path(esw, flow, &slow_attr);
1065 flow->flags |= MLX5E_TC_FLOW_OFFLOADED; /* was unset when slow path rule removed */
1066 flow->rule[0] = rule;
1070 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
1071 struct mlx5e_encap_entry *e)
1073 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1074 struct mlx5_esw_flow_attr slow_attr;
1075 struct mlx5_flow_handle *rule;
1076 struct mlx5_flow_spec *spec;
1077 struct mlx5e_tc_flow *flow;
1080 list_for_each_entry(flow, &e->flows, encap) {
1081 spec = &flow->esw_attr->parse_attr->spec;
1083 /* update from encap rule to slow path rule */
1084 rule = mlx5e_tc_offload_to_slow_path(esw, flow, spec, &slow_attr);
1087 err = PTR_ERR(rule);
1088 mlx5_core_warn(priv->mdev, "Failed to update slow path (encap) flow, %d\n",
1093 mlx5e_tc_unoffload_fdb_rules(esw, flow, flow->esw_attr);
1094 flow->flags |= MLX5E_TC_FLOW_OFFLOADED; /* was unset when fast path rule removed */
1095 flow->rule[0] = rule;
1098 if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
1099 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
1100 mlx5_packet_reformat_dealloc(priv->mdev, e->encap_id);
1104 static struct mlx5_fc *mlx5e_tc_get_counter(struct mlx5e_tc_flow *flow)
1106 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1107 return flow->esw_attr->counter;
1109 return flow->nic_attr->counter;
1112 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
1114 struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
1115 u64 bytes, packets, lastuse = 0;
1116 struct mlx5e_tc_flow *flow;
1117 struct mlx5e_encap_entry *e;
1118 struct mlx5_fc *counter;
1119 struct neigh_table *tbl;
1120 bool neigh_used = false;
1121 struct neighbour *n;
1123 if (m_neigh->family == AF_INET)
1125 #if IS_ENABLED(CONFIG_IPV6)
1126 else if (m_neigh->family == AF_INET6)
1132 list_for_each_entry(e, &nhe->encap_list, encap_list) {
1133 if (!(e->flags & MLX5_ENCAP_ENTRY_VALID))
1135 list_for_each_entry(flow, &e->flows, encap) {
1136 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
1137 counter = mlx5e_tc_get_counter(flow);
1138 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
1139 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
1150 nhe->reported_lastuse = jiffies;
1152 /* find the relevant neigh according to the cached device and
1155 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
1159 neigh_event_send(n, NULL);
1164 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1165 struct mlx5e_tc_flow *flow)
1167 struct list_head *next = flow->encap.next;
1169 list_del(&flow->encap);
1170 if (list_empty(next)) {
1171 struct mlx5e_encap_entry *e;
1173 e = list_entry(next, struct mlx5e_encap_entry, flows);
1174 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1176 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1177 mlx5_packet_reformat_dealloc(priv->mdev, e->encap_id);
1179 hash_del_rcu(&e->encap_hlist);
1180 kfree(e->encap_header);
1185 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1186 struct mlx5e_tc_flow *flow)
1188 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1189 mlx5e_tc_del_fdb_flow(priv, flow);
1191 mlx5e_tc_del_nic_flow(priv, flow);
1194 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
1195 struct tc_cls_flower_offload *f)
1197 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1199 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1201 void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1203 void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1206 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
1207 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
1209 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
1210 struct flow_dissector_key_keyid *key =
1211 skb_flow_dissector_target(f->dissector,
1212 FLOW_DISSECTOR_KEY_ENC_KEYID,
1214 struct flow_dissector_key_keyid *mask =
1215 skb_flow_dissector_target(f->dissector,
1216 FLOW_DISSECTOR_KEY_ENC_KEYID,
1218 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
1219 be32_to_cpu(mask->keyid));
1220 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
1221 be32_to_cpu(key->keyid));
1225 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1226 struct mlx5_flow_spec *spec,
1227 struct tc_cls_flower_offload *f)
1229 struct netlink_ext_ack *extack = f->common.extack;
1230 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1232 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1235 struct flow_dissector_key_control *enc_control =
1236 skb_flow_dissector_target(f->dissector,
1237 FLOW_DISSECTOR_KEY_ENC_CONTROL,
1240 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
1241 struct flow_dissector_key_ports *key =
1242 skb_flow_dissector_target(f->dissector,
1243 FLOW_DISSECTOR_KEY_ENC_PORTS,
1245 struct flow_dissector_key_ports *mask =
1246 skb_flow_dissector_target(f->dissector,
1247 FLOW_DISSECTOR_KEY_ENC_PORTS,
1250 /* Full udp dst port must be given */
1251 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
1252 goto vxlan_match_offload_err;
1254 if (mlx5_vxlan_lookup_port(priv->mdev->vxlan, be16_to_cpu(key->dst)) &&
1255 MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
1256 parse_vxlan_attr(spec, f);
1258 NL_SET_ERR_MSG_MOD(extack,
1259 "port isn't an offloaded vxlan udp dport");
1260 netdev_warn(priv->netdev,
1261 "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
1265 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1266 udp_dport, ntohs(mask->dst));
1267 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1268 udp_dport, ntohs(key->dst));
1270 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1271 udp_sport, ntohs(mask->src));
1272 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1273 udp_sport, ntohs(key->src));
1274 } else { /* udp dst port must be given */
1275 vxlan_match_offload_err:
1276 NL_SET_ERR_MSG_MOD(extack,
1277 "IP tunnel decap offload supported only for vxlan, must set UDP dport");
1278 netdev_warn(priv->netdev,
1279 "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
1283 if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1284 struct flow_dissector_key_ipv4_addrs *key =
1285 skb_flow_dissector_target(f->dissector,
1286 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1288 struct flow_dissector_key_ipv4_addrs *mask =
1289 skb_flow_dissector_target(f->dissector,
1290 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1292 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1293 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1295 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1296 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1299 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1300 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1302 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1303 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1306 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1307 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
1308 } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1309 struct flow_dissector_key_ipv6_addrs *key =
1310 skb_flow_dissector_target(f->dissector,
1311 FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1313 struct flow_dissector_key_ipv6_addrs *mask =
1314 skb_flow_dissector_target(f->dissector,
1315 FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1318 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1319 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1320 &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1321 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1322 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1323 &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1325 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1326 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1327 &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1328 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1329 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1330 &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1332 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1333 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
1336 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_IP)) {
1337 struct flow_dissector_key_ip *key =
1338 skb_flow_dissector_target(f->dissector,
1339 FLOW_DISSECTOR_KEY_ENC_IP,
1341 struct flow_dissector_key_ip *mask =
1342 skb_flow_dissector_target(f->dissector,
1343 FLOW_DISSECTOR_KEY_ENC_IP,
1346 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn, mask->tos & 0x3);
1347 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, key->tos & 0x3);
1349 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp, mask->tos >> 2);
1350 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, key->tos >> 2);
1352 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit, mask->ttl);
1353 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit, key->ttl);
1356 !MLX5_CAP_ESW_FLOWTABLE_FDB
1358 ft_field_support.outer_ipv4_ttl)) {
1359 NL_SET_ERR_MSG_MOD(extack,
1360 "Matching on TTL is not supported");
1366 /* Enforce DMAC when offloading incoming tunneled flows.
1367 * Flow counters require a match on the DMAC.
1369 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
1370 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
1371 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1372 dmac_47_16), priv->netdev->dev_addr);
1374 /* let software handle IP fragments */
1375 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1376 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
1381 static int __parse_cls_flower(struct mlx5e_priv *priv,
1382 struct mlx5_flow_spec *spec,
1383 struct tc_cls_flower_offload *f,
1386 struct netlink_ext_ack *extack = f->common.extack;
1387 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1389 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1391 void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1393 void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1398 *match_level = MLX5_MATCH_NONE;
1400 if (f->dissector->used_keys &
1401 ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
1402 BIT(FLOW_DISSECTOR_KEY_BASIC) |
1403 BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
1404 BIT(FLOW_DISSECTOR_KEY_VLAN) |
1405 BIT(FLOW_DISSECTOR_KEY_CVLAN) |
1406 BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
1407 BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
1408 BIT(FLOW_DISSECTOR_KEY_PORTS) |
1409 BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
1410 BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
1411 BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
1412 BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
1413 BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
1414 BIT(FLOW_DISSECTOR_KEY_TCP) |
1415 BIT(FLOW_DISSECTOR_KEY_IP) |
1416 BIT(FLOW_DISSECTOR_KEY_ENC_IP))) {
1417 NL_SET_ERR_MSG_MOD(extack, "Unsupported key");
1418 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
1419 f->dissector->used_keys);
1423 if ((dissector_uses_key(f->dissector,
1424 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
1425 dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
1426 dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
1427 dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
1428 struct flow_dissector_key_control *key =
1429 skb_flow_dissector_target(f->dissector,
1430 FLOW_DISSECTOR_KEY_ENC_CONTROL,
1432 switch (key->addr_type) {
1433 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
1434 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
1435 if (parse_tunnel_attr(priv, spec, f))
1442 /* In decap flow, header pointers should point to the inner
1443 * headers, outer header were already set by parse_tunnel_attr
1445 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1447 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1451 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1452 struct flow_dissector_key_basic *key =
1453 skb_flow_dissector_target(f->dissector,
1454 FLOW_DISSECTOR_KEY_BASIC,
1456 struct flow_dissector_key_basic *mask =
1457 skb_flow_dissector_target(f->dissector,
1458 FLOW_DISSECTOR_KEY_BASIC,
1460 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1461 ntohs(mask->n_proto));
1462 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1463 ntohs(key->n_proto));
1466 *match_level = MLX5_MATCH_L2;
1469 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
1470 struct flow_dissector_key_vlan *key =
1471 skb_flow_dissector_target(f->dissector,
1472 FLOW_DISSECTOR_KEY_VLAN,
1474 struct flow_dissector_key_vlan *mask =
1475 skb_flow_dissector_target(f->dissector,
1476 FLOW_DISSECTOR_KEY_VLAN,
1478 if (mask->vlan_id || mask->vlan_priority || mask->vlan_tpid) {
1479 if (key->vlan_tpid == htons(ETH_P_8021AD)) {
1480 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1482 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1485 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1487 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1491 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
1492 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
1494 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
1495 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
1497 *match_level = MLX5_MATCH_L2;
1499 } else if (*match_level != MLX5_MATCH_NONE) {
1500 MLX5_SET(fte_match_set_lyr_2_4, headers_c, svlan_tag, 1);
1501 MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
1502 *match_level = MLX5_MATCH_L2;
1505 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CVLAN)) {
1506 struct flow_dissector_key_vlan *key =
1507 skb_flow_dissector_target(f->dissector,
1508 FLOW_DISSECTOR_KEY_CVLAN,
1510 struct flow_dissector_key_vlan *mask =
1511 skb_flow_dissector_target(f->dissector,
1512 FLOW_DISSECTOR_KEY_CVLAN,
1514 if (mask->vlan_id || mask->vlan_priority || mask->vlan_tpid) {
1515 if (key->vlan_tpid == htons(ETH_P_8021AD)) {
1516 MLX5_SET(fte_match_set_misc, misc_c,
1517 outer_second_svlan_tag, 1);
1518 MLX5_SET(fte_match_set_misc, misc_v,
1519 outer_second_svlan_tag, 1);
1521 MLX5_SET(fte_match_set_misc, misc_c,
1522 outer_second_cvlan_tag, 1);
1523 MLX5_SET(fte_match_set_misc, misc_v,
1524 outer_second_cvlan_tag, 1);
1527 MLX5_SET(fte_match_set_misc, misc_c, outer_second_vid,
1529 MLX5_SET(fte_match_set_misc, misc_v, outer_second_vid,
1531 MLX5_SET(fte_match_set_misc, misc_c, outer_second_prio,
1532 mask->vlan_priority);
1533 MLX5_SET(fte_match_set_misc, misc_v, outer_second_prio,
1534 key->vlan_priority);
1536 *match_level = MLX5_MATCH_L2;
1540 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
1541 struct flow_dissector_key_eth_addrs *key =
1542 skb_flow_dissector_target(f->dissector,
1543 FLOW_DISSECTOR_KEY_ETH_ADDRS,
1545 struct flow_dissector_key_eth_addrs *mask =
1546 skb_flow_dissector_target(f->dissector,
1547 FLOW_DISSECTOR_KEY_ETH_ADDRS,
1550 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1553 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1557 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1560 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1564 if (!is_zero_ether_addr(mask->src) || !is_zero_ether_addr(mask->dst))
1565 *match_level = MLX5_MATCH_L2;
1568 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
1569 struct flow_dissector_key_control *key =
1570 skb_flow_dissector_target(f->dissector,
1571 FLOW_DISSECTOR_KEY_CONTROL,
1574 struct flow_dissector_key_control *mask =
1575 skb_flow_dissector_target(f->dissector,
1576 FLOW_DISSECTOR_KEY_CONTROL,
1578 addr_type = key->addr_type;
1580 /* the HW doesn't support frag first/later */
1581 if (mask->flags & FLOW_DIS_FIRST_FRAG)
1584 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
1585 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1586 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
1587 key->flags & FLOW_DIS_IS_FRAGMENT);
1589 /* the HW doesn't need L3 inline to match on frag=no */
1590 if (!(key->flags & FLOW_DIS_IS_FRAGMENT))
1591 *match_level = MLX5_MATCH_L2;
1592 /* *** L2 attributes parsing up to here *** */
1594 *match_level = MLX5_MATCH_L3;
1598 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1599 struct flow_dissector_key_basic *key =
1600 skb_flow_dissector_target(f->dissector,
1601 FLOW_DISSECTOR_KEY_BASIC,
1603 struct flow_dissector_key_basic *mask =
1604 skb_flow_dissector_target(f->dissector,
1605 FLOW_DISSECTOR_KEY_BASIC,
1607 ip_proto = key->ip_proto;
1609 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
1611 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
1615 *match_level = MLX5_MATCH_L3;
1618 if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1619 struct flow_dissector_key_ipv4_addrs *key =
1620 skb_flow_dissector_target(f->dissector,
1621 FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1623 struct flow_dissector_key_ipv4_addrs *mask =
1624 skb_flow_dissector_target(f->dissector,
1625 FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1628 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1629 src_ipv4_src_ipv6.ipv4_layout.ipv4),
1630 &mask->src, sizeof(mask->src));
1631 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1632 src_ipv4_src_ipv6.ipv4_layout.ipv4),
1633 &key->src, sizeof(key->src));
1634 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1635 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1636 &mask->dst, sizeof(mask->dst));
1637 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1638 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1639 &key->dst, sizeof(key->dst));
1641 if (mask->src || mask->dst)
1642 *match_level = MLX5_MATCH_L3;
1645 if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1646 struct flow_dissector_key_ipv6_addrs *key =
1647 skb_flow_dissector_target(f->dissector,
1648 FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1650 struct flow_dissector_key_ipv6_addrs *mask =
1651 skb_flow_dissector_target(f->dissector,
1652 FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1655 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1656 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1657 &mask->src, sizeof(mask->src));
1658 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1659 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1660 &key->src, sizeof(key->src));
1662 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1663 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1664 &mask->dst, sizeof(mask->dst));
1665 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1666 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1667 &key->dst, sizeof(key->dst));
1669 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
1670 ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
1671 *match_level = MLX5_MATCH_L3;
1674 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_IP)) {
1675 struct flow_dissector_key_ip *key =
1676 skb_flow_dissector_target(f->dissector,
1677 FLOW_DISSECTOR_KEY_IP,
1679 struct flow_dissector_key_ip *mask =
1680 skb_flow_dissector_target(f->dissector,
1681 FLOW_DISSECTOR_KEY_IP,
1684 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn, mask->tos & 0x3);
1685 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, key->tos & 0x3);
1687 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp, mask->tos >> 2);
1688 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, key->tos >> 2);
1690 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit, mask->ttl);
1691 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit, key->ttl);
1694 !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
1695 ft_field_support.outer_ipv4_ttl)) {
1696 NL_SET_ERR_MSG_MOD(extack,
1697 "Matching on TTL is not supported");
1701 if (mask->tos || mask->ttl)
1702 *match_level = MLX5_MATCH_L3;
1705 /* *** L3 attributes parsing up to here *** */
1707 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
1708 struct flow_dissector_key_ports *key =
1709 skb_flow_dissector_target(f->dissector,
1710 FLOW_DISSECTOR_KEY_PORTS,
1712 struct flow_dissector_key_ports *mask =
1713 skb_flow_dissector_target(f->dissector,
1714 FLOW_DISSECTOR_KEY_PORTS,
1718 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1719 tcp_sport, ntohs(mask->src));
1720 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1721 tcp_sport, ntohs(key->src));
1723 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1724 tcp_dport, ntohs(mask->dst));
1725 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1726 tcp_dport, ntohs(key->dst));
1730 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1731 udp_sport, ntohs(mask->src));
1732 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1733 udp_sport, ntohs(key->src));
1735 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1736 udp_dport, ntohs(mask->dst));
1737 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1738 udp_dport, ntohs(key->dst));
1741 NL_SET_ERR_MSG_MOD(extack,
1742 "Only UDP and TCP transports are supported for L4 matching");
1743 netdev_err(priv->netdev,
1744 "Only UDP and TCP transport are supported\n");
1748 if (mask->src || mask->dst)
1749 *match_level = MLX5_MATCH_L4;
1752 if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_TCP)) {
1753 struct flow_dissector_key_tcp *key =
1754 skb_flow_dissector_target(f->dissector,
1755 FLOW_DISSECTOR_KEY_TCP,
1757 struct flow_dissector_key_tcp *mask =
1758 skb_flow_dissector_target(f->dissector,
1759 FLOW_DISSECTOR_KEY_TCP,
1762 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
1763 ntohs(mask->flags));
1764 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
1768 *match_level = MLX5_MATCH_L4;
1774 static int parse_cls_flower(struct mlx5e_priv *priv,
1775 struct mlx5e_tc_flow *flow,
1776 struct mlx5_flow_spec *spec,
1777 struct tc_cls_flower_offload *f)
1779 struct netlink_ext_ack *extack = f->common.extack;
1780 struct mlx5_core_dev *dev = priv->mdev;
1781 struct mlx5_eswitch *esw = dev->priv.eswitch;
1782 struct mlx5e_rep_priv *rpriv = priv->ppriv;
1783 struct mlx5_eswitch_rep *rep;
1787 err = __parse_cls_flower(priv, spec, f, &match_level);
1789 if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH)) {
1791 if (rep->vport != FDB_UPLINK_VPORT &&
1792 (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
1793 esw->offloads.inline_mode < match_level)) {
1794 NL_SET_ERR_MSG_MOD(extack,
1795 "Flow is not offloaded due to min inline setting");
1796 netdev_warn(priv->netdev,
1797 "Flow is not offloaded due to min inline setting, required %d actual %d\n",
1798 match_level, esw->offloads.inline_mode);
1803 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1804 flow->esw_attr->match_level = match_level;
1806 flow->nic_attr->match_level = match_level;
1811 struct pedit_headers {
1819 static int pedit_header_offsets[] = {
1820 [TCA_PEDIT_KEY_EX_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
1821 [TCA_PEDIT_KEY_EX_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
1822 [TCA_PEDIT_KEY_EX_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
1823 [TCA_PEDIT_KEY_EX_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
1824 [TCA_PEDIT_KEY_EX_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
1827 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
1829 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
1830 struct pedit_headers *masks,
1831 struct pedit_headers *vals)
1833 u32 *curr_pmask, *curr_pval;
1835 if (hdr_type >= __PEDIT_HDR_TYPE_MAX)
1838 curr_pmask = (u32 *)(pedit_header(masks, hdr_type) + offset);
1839 curr_pval = (u32 *)(pedit_header(vals, hdr_type) + offset);
1841 if (*curr_pmask & mask) /* disallow acting twice on the same location */
1844 *curr_pmask |= mask;
1845 *curr_pval |= (val & mask);
1853 struct mlx5_fields {
1859 #define OFFLOAD(fw_field, size, field, off) \
1860 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, size, offsetof(struct pedit_headers, field) + (off)}
1862 static struct mlx5_fields fields[] = {
1863 OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0),
1864 OFFLOAD(DMAC_15_0, 2, eth.h_dest[4], 0),
1865 OFFLOAD(SMAC_47_16, 4, eth.h_source[0], 0),
1866 OFFLOAD(SMAC_15_0, 2, eth.h_source[4], 0),
1867 OFFLOAD(ETHERTYPE, 2, eth.h_proto, 0),
1869 OFFLOAD(IP_TTL, 1, ip4.ttl, 0),
1870 OFFLOAD(SIPV4, 4, ip4.saddr, 0),
1871 OFFLOAD(DIPV4, 4, ip4.daddr, 0),
1873 OFFLOAD(SIPV6_127_96, 4, ip6.saddr.s6_addr32[0], 0),
1874 OFFLOAD(SIPV6_95_64, 4, ip6.saddr.s6_addr32[1], 0),
1875 OFFLOAD(SIPV6_63_32, 4, ip6.saddr.s6_addr32[2], 0),
1876 OFFLOAD(SIPV6_31_0, 4, ip6.saddr.s6_addr32[3], 0),
1877 OFFLOAD(DIPV6_127_96, 4, ip6.daddr.s6_addr32[0], 0),
1878 OFFLOAD(DIPV6_95_64, 4, ip6.daddr.s6_addr32[1], 0),
1879 OFFLOAD(DIPV6_63_32, 4, ip6.daddr.s6_addr32[2], 0),
1880 OFFLOAD(DIPV6_31_0, 4, ip6.daddr.s6_addr32[3], 0),
1881 OFFLOAD(IPV6_HOPLIMIT, 1, ip6.hop_limit, 0),
1883 OFFLOAD(TCP_SPORT, 2, tcp.source, 0),
1884 OFFLOAD(TCP_DPORT, 2, tcp.dest, 0),
1885 OFFLOAD(TCP_FLAGS, 1, tcp.ack_seq, 5),
1887 OFFLOAD(UDP_SPORT, 2, udp.source, 0),
1888 OFFLOAD(UDP_DPORT, 2, udp.dest, 0),
1891 /* On input attr->num_mod_hdr_actions tells how many HW actions can be parsed at
1892 * max from the SW pedit action. On success, it says how many HW actions were
1895 static int offload_pedit_fields(struct pedit_headers *masks,
1896 struct pedit_headers *vals,
1897 struct mlx5e_tc_flow_parse_attr *parse_attr,
1898 struct netlink_ext_ack *extack)
1900 struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
1901 int i, action_size, nactions, max_actions, first, last, next_z;
1902 void *s_masks_p, *a_masks_p, *vals_p;
1903 struct mlx5_fields *f;
1904 u8 cmd, field_bsize;
1911 set_masks = &masks[TCA_PEDIT_KEY_EX_CMD_SET];
1912 add_masks = &masks[TCA_PEDIT_KEY_EX_CMD_ADD];
1913 set_vals = &vals[TCA_PEDIT_KEY_EX_CMD_SET];
1914 add_vals = &vals[TCA_PEDIT_KEY_EX_CMD_ADD];
1916 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1917 action = parse_attr->mod_hdr_actions;
1918 max_actions = parse_attr->num_mod_hdr_actions;
1921 for (i = 0; i < ARRAY_SIZE(fields); i++) {
1923 /* avoid seeing bits set from previous iterations */
1927 s_masks_p = (void *)set_masks + f->offset;
1928 a_masks_p = (void *)add_masks + f->offset;
1930 memcpy(&s_mask, s_masks_p, f->size);
1931 memcpy(&a_mask, a_masks_p, f->size);
1933 if (!s_mask && !a_mask) /* nothing to offload here */
1936 if (s_mask && a_mask) {
1937 NL_SET_ERR_MSG_MOD(extack,
1938 "can't set and add to the same HW field");
1939 printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
1943 if (nactions == max_actions) {
1944 NL_SET_ERR_MSG_MOD(extack,
1945 "too many pedit actions, can't offload");
1946 printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
1951 cmd = MLX5_ACTION_TYPE_SET;
1953 vals_p = (void *)set_vals + f->offset;
1954 /* clear to denote we consumed this field */
1955 memset(s_masks_p, 0, f->size);
1957 cmd = MLX5_ACTION_TYPE_ADD;
1959 vals_p = (void *)add_vals + f->offset;
1960 /* clear to denote we consumed this field */
1961 memset(a_masks_p, 0, f->size);
1964 field_bsize = f->size * BITS_PER_BYTE;
1966 if (field_bsize == 32) {
1967 mask_be32 = *(__be32 *)&mask;
1968 mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
1969 } else if (field_bsize == 16) {
1970 mask_be16 = *(__be16 *)&mask;
1971 mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
1974 first = find_first_bit(&mask, field_bsize);
1975 next_z = find_next_zero_bit(&mask, field_bsize, first);
1976 last = find_last_bit(&mask, field_bsize);
1977 if (first < next_z && next_z < last) {
1978 NL_SET_ERR_MSG_MOD(extack,
1979 "rewrite of few sub-fields isn't supported");
1980 printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
1985 MLX5_SET(set_action_in, action, action_type, cmd);
1986 MLX5_SET(set_action_in, action, field, f->field);
1988 if (cmd == MLX5_ACTION_TYPE_SET) {
1989 MLX5_SET(set_action_in, action, offset, first);
1990 /* length is num of bits to be written, zero means length of 32 */
1991 MLX5_SET(set_action_in, action, length, (last - first + 1));
1994 if (field_bsize == 32)
1995 MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
1996 else if (field_bsize == 16)
1997 MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
1998 else if (field_bsize == 8)
1999 MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
2001 action += action_size;
2005 parse_attr->num_mod_hdr_actions = nactions;
2009 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
2010 const struct tc_action *a, int namespace,
2011 struct mlx5e_tc_flow_parse_attr *parse_attr)
2013 int nkeys, action_size, max_actions;
2015 nkeys = tcf_pedit_nkeys(a);
2016 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2018 if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
2019 max_actions = MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev, max_modify_header_actions);
2020 else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
2021 max_actions = MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, max_modify_header_actions);
2023 /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
2024 max_actions = min(max_actions, nkeys * 16);
2026 parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
2027 if (!parse_attr->mod_hdr_actions)
2030 parse_attr->num_mod_hdr_actions = max_actions;
2034 static const struct pedit_headers zero_masks = {};
2036 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
2037 const struct tc_action *a, int namespace,
2038 struct mlx5e_tc_flow_parse_attr *parse_attr,
2039 struct netlink_ext_ack *extack)
2041 struct pedit_headers masks[__PEDIT_CMD_MAX], vals[__PEDIT_CMD_MAX], *cmd_masks;
2042 int nkeys, i, err = -EOPNOTSUPP;
2043 u32 mask, val, offset;
2046 nkeys = tcf_pedit_nkeys(a);
2048 memset(masks, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
2049 memset(vals, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
2051 for (i = 0; i < nkeys; i++) {
2052 htype = tcf_pedit_htype(a, i);
2053 cmd = tcf_pedit_cmd(a, i);
2054 err = -EOPNOTSUPP; /* can't be all optimistic */
2056 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK) {
2057 NL_SET_ERR_MSG_MOD(extack,
2058 "legacy pedit isn't offloaded");
2062 if (cmd != TCA_PEDIT_KEY_EX_CMD_SET && cmd != TCA_PEDIT_KEY_EX_CMD_ADD) {
2063 NL_SET_ERR_MSG_MOD(extack, "pedit cmd isn't offloaded");
2067 mask = tcf_pedit_mask(a, i);
2068 val = tcf_pedit_val(a, i);
2069 offset = tcf_pedit_offset(a, i);
2071 err = set_pedit_val(htype, ~mask, val, offset, &masks[cmd], &vals[cmd]);
2076 err = alloc_mod_hdr_actions(priv, a, namespace, parse_attr);
2080 err = offload_pedit_fields(masks, vals, parse_attr, extack);
2082 goto out_dealloc_parsed_actions;
2084 for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
2085 cmd_masks = &masks[cmd];
2086 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
2087 NL_SET_ERR_MSG_MOD(extack,
2088 "attempt to offload an unsupported field");
2089 netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
2090 print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
2091 16, 1, cmd_masks, sizeof(zero_masks), true);
2093 goto out_dealloc_parsed_actions;
2099 out_dealloc_parsed_actions:
2100 kfree(parse_attr->mod_hdr_actions);
2105 static bool csum_offload_supported(struct mlx5e_priv *priv,
2108 struct netlink_ext_ack *extack)
2110 u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
2111 TCA_CSUM_UPDATE_FLAG_UDP;
2113 /* The HW recalcs checksums only if re-writing headers */
2114 if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
2115 NL_SET_ERR_MSG_MOD(extack,
2116 "TC csum action is only offloaded with pedit");
2117 netdev_warn(priv->netdev,
2118 "TC csum action is only offloaded with pedit\n");
2122 if (update_flags & ~prot_flags) {
2123 NL_SET_ERR_MSG_MOD(extack,
2124 "can't offload TC csum action for some header/s");
2125 netdev_warn(priv->netdev,
2126 "can't offload TC csum action for some header/s - flags %#x\n",
2134 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
2135 struct tcf_exts *exts,
2136 struct netlink_ext_ack *extack)
2138 const struct tc_action *a;
2139 bool modify_ip_header;
2146 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
2147 ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
2149 /* for non-IP we only re-write MACs, so we're okay */
2150 if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
2153 modify_ip_header = false;
2154 tcf_exts_for_each_action(i, a, exts) {
2157 if (!is_tcf_pedit(a))
2160 nkeys = tcf_pedit_nkeys(a);
2161 for (k = 0; k < nkeys; k++) {
2162 htype = tcf_pedit_htype(a, k);
2163 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP4 ||
2164 htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP6) {
2165 modify_ip_header = true;
2171 ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
2172 if (modify_ip_header && ip_proto != IPPROTO_TCP &&
2173 ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
2174 NL_SET_ERR_MSG_MOD(extack,
2175 "can't offload re-write of non TCP/UDP");
2176 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
2184 static bool actions_match_supported(struct mlx5e_priv *priv,
2185 struct tcf_exts *exts,
2186 struct mlx5e_tc_flow_parse_attr *parse_attr,
2187 struct mlx5e_tc_flow *flow,
2188 struct netlink_ext_ack *extack)
2192 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
2193 actions = flow->esw_attr->action;
2195 actions = flow->nic_attr->action;
2197 if (flow->flags & MLX5E_TC_FLOW_EGRESS &&
2198 !(actions & MLX5_FLOW_CONTEXT_ACTION_DECAP))
2201 if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
2202 return modify_header_match_supported(&parse_attr->spec, exts,
2208 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
2210 struct mlx5_core_dev *fmdev, *pmdev;
2211 u64 fsystem_guid, psystem_guid;
2214 pmdev = peer_priv->mdev;
2216 fsystem_guid = mlx5_query_nic_system_image_guid(fmdev);
2217 psystem_guid = mlx5_query_nic_system_image_guid(pmdev);
2219 return (fsystem_guid == psystem_guid);
2222 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
2223 struct mlx5e_tc_flow_parse_attr *parse_attr,
2224 struct mlx5e_tc_flow *flow,
2225 struct netlink_ext_ack *extack)
2227 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
2228 const struct tc_action *a;
2233 if (!tcf_exts_has_actions(exts))
2236 attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
2238 tcf_exts_for_each_action(i, a, exts) {
2239 if (is_tcf_gact_shot(a)) {
2240 action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
2241 if (MLX5_CAP_FLOWTABLE(priv->mdev,
2242 flow_table_properties_nic_receive.flow_counter))
2243 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
2247 if (is_tcf_pedit(a)) {
2248 err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_KERNEL,
2249 parse_attr, extack);
2253 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
2254 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2258 if (is_tcf_csum(a)) {
2259 if (csum_offload_supported(priv, action,
2260 tcf_csum_update_flags(a),
2267 if (is_tcf_mirred_egress_redirect(a)) {
2268 struct net_device *peer_dev = tcf_mirred_dev(a);
2270 if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
2271 same_hw_devs(priv, netdev_priv(peer_dev))) {
2272 parse_attr->mirred_ifindex = peer_dev->ifindex;
2273 flow->flags |= MLX5E_TC_FLOW_HAIRPIN;
2274 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2275 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2277 NL_SET_ERR_MSG_MOD(extack,
2278 "device is not on same HW, can't offload");
2279 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
2286 if (is_tcf_skbedit_mark(a)) {
2287 u32 mark = tcf_skbedit_mark(a);
2289 if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
2290 NL_SET_ERR_MSG_MOD(extack,
2291 "Bad flow mark - only 16 bit is supported");
2295 attr->flow_tag = mark;
2296 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2303 attr->action = action;
2304 if (!actions_match_supported(priv, exts, parse_attr, flow, extack))
2310 static inline int cmp_encap_info(struct ip_tunnel_key *a,
2311 struct ip_tunnel_key *b)
2313 return memcmp(a, b, sizeof(*a));
2316 static inline int hash_encap_info(struct ip_tunnel_key *key)
2318 return jhash(key, sizeof(*key), 0);
2321 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
2322 struct net_device *mirred_dev,
2323 struct net_device **out_dev,
2325 struct neighbour **out_n,
2328 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2329 struct mlx5e_rep_priv *uplink_rpriv;
2331 struct neighbour *n = NULL;
2333 #if IS_ENABLED(CONFIG_INET)
2336 rt = ip_route_output_key(dev_net(mirred_dev), fl4);
2337 ret = PTR_ERR_OR_ZERO(rt);
2343 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2344 /* if the egress device isn't on the same HW e-switch, we use the uplink */
2345 if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
2346 *out_dev = uplink_rpriv->netdev;
2348 *out_dev = rt->dst.dev;
2351 *out_ttl = ip4_dst_hoplimit(&rt->dst);
2352 n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
2361 static bool is_merged_eswitch_dev(struct mlx5e_priv *priv,
2362 struct net_device *peer_netdev)
2364 struct mlx5e_priv *peer_priv;
2366 peer_priv = netdev_priv(peer_netdev);
2368 return (MLX5_CAP_ESW(priv->mdev, merged_eswitch) &&
2369 (priv->netdev->netdev_ops == peer_netdev->netdev_ops) &&
2370 same_hw_devs(priv, peer_priv) &&
2371 MLX5_VPORT_MANAGER(peer_priv->mdev) &&
2372 (peer_priv->mdev->priv.eswitch->mode == SRIOV_OFFLOADS));
2375 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
2376 struct net_device *mirred_dev,
2377 struct net_device **out_dev,
2379 struct neighbour **out_n,
2382 struct neighbour *n = NULL;
2383 struct dst_entry *dst;
2385 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
2386 struct mlx5e_rep_priv *uplink_rpriv;
2387 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2390 ret = ipv6_stub->ipv6_dst_lookup(dev_net(mirred_dev), NULL, &dst,
2396 *out_ttl = ip6_dst_hoplimit(dst);
2398 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2399 /* if the egress device isn't on the same HW e-switch, we use the uplink */
2400 if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
2401 *out_dev = uplink_rpriv->netdev;
2403 *out_dev = dst->dev;
2408 n = dst_neigh_lookup(dst, &fl6->daddr);
2417 static void gen_vxlan_header_ipv4(struct net_device *out_dev,
2418 char buf[], int encap_size,
2419 unsigned char h_dest[ETH_ALEN],
2423 __be16 udp_dst_port,
2426 struct ethhdr *eth = (struct ethhdr *)buf;
2427 struct iphdr *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
2428 struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
2429 struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2431 memset(buf, 0, encap_size);
2433 ether_addr_copy(eth->h_dest, h_dest);
2434 ether_addr_copy(eth->h_source, out_dev->dev_addr);
2435 eth->h_proto = htons(ETH_P_IP);
2442 ip->protocol = IPPROTO_UDP;
2446 udp->dest = udp_dst_port;
2447 vxh->vx_flags = VXLAN_HF_VNI;
2448 vxh->vx_vni = vxlan_vni_field(vx_vni);
2451 static void gen_vxlan_header_ipv6(struct net_device *out_dev,
2452 char buf[], int encap_size,
2453 unsigned char h_dest[ETH_ALEN],
2455 struct in6_addr *daddr,
2456 struct in6_addr *saddr,
2457 __be16 udp_dst_port,
2460 struct ethhdr *eth = (struct ethhdr *)buf;
2461 struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
2462 struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
2463 struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2465 memset(buf, 0, encap_size);
2467 ether_addr_copy(eth->h_dest, h_dest);
2468 ether_addr_copy(eth->h_source, out_dev->dev_addr);
2469 eth->h_proto = htons(ETH_P_IPV6);
2471 ip6_flow_hdr(ip6h, tos, 0);
2472 /* the HW fills up ipv6 payload len */
2473 ip6h->nexthdr = IPPROTO_UDP;
2474 ip6h->hop_limit = ttl;
2475 ip6h->daddr = *daddr;
2476 ip6h->saddr = *saddr;
2478 udp->dest = udp_dst_port;
2479 vxh->vx_flags = VXLAN_HF_VNI;
2480 vxh->vx_vni = vxlan_vni_field(vx_vni);
2483 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
2484 struct net_device *mirred_dev,
2485 struct mlx5e_encap_entry *e)
2487 int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2488 int ipv4_encap_size = ETH_HLEN + sizeof(struct iphdr) + VXLAN_HLEN;
2489 struct ip_tunnel_key *tun_key = &e->tun_info.key;
2490 struct net_device *out_dev;
2491 struct neighbour *n = NULL;
2492 struct flowi4 fl4 = {};
2493 u8 nud_state, tos, ttl;
2497 if (max_encap_size < ipv4_encap_size) {
2498 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2499 ipv4_encap_size, max_encap_size);
2503 encap_header = kzalloc(ipv4_encap_size, GFP_KERNEL);
2507 switch (e->tunnel_type) {
2508 case MLX5_REFORMAT_TYPE_L2_TO_VXLAN:
2509 fl4.flowi4_proto = IPPROTO_UDP;
2510 fl4.fl4_dport = tun_key->tp_dst;
2520 fl4.flowi4_tos = tun_key->tos;
2521 fl4.daddr = tun_key->u.ipv4.dst;
2522 fl4.saddr = tun_key->u.ipv4.src;
2524 err = mlx5e_route_lookup_ipv4(priv, mirred_dev, &out_dev,
2529 /* used by mlx5e_detach_encap to lookup a neigh hash table
2530 * entry in the neigh hash table when a user deletes a rule
2532 e->m_neigh.dev = n->dev;
2533 e->m_neigh.family = n->ops->family;
2534 memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2535 e->out_dev = out_dev;
2537 /* It's importent to add the neigh to the hash table before checking
2538 * the neigh validity state. So if we'll get a notification, in case the
2539 * neigh changes it's validity state, we would find the relevant neigh
2542 err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2546 read_lock_bh(&n->lock);
2547 nud_state = n->nud_state;
2548 ether_addr_copy(e->h_dest, n->ha);
2549 read_unlock_bh(&n->lock);
2551 switch (e->tunnel_type) {
2552 case MLX5_REFORMAT_TYPE_L2_TO_VXLAN:
2553 gen_vxlan_header_ipv4(out_dev, encap_header,
2554 ipv4_encap_size, e->h_dest, tos, ttl,
2556 fl4.saddr, tun_key->tp_dst,
2557 tunnel_id_to_key32(tun_key->tun_id));
2561 goto destroy_neigh_entry;
2563 e->encap_size = ipv4_encap_size;
2564 e->encap_header = encap_header;
2566 if (!(nud_state & NUD_VALID)) {
2567 neigh_event_send(n, NULL);
2572 err = mlx5_packet_reformat_alloc(priv->mdev, e->tunnel_type,
2573 ipv4_encap_size, encap_header,
2574 MLX5_FLOW_NAMESPACE_FDB,
2577 goto destroy_neigh_entry;
2579 e->flags |= MLX5_ENCAP_ENTRY_VALID;
2580 mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2584 destroy_neigh_entry:
2585 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2587 kfree(encap_header);
2594 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
2595 struct net_device *mirred_dev,
2596 struct mlx5e_encap_entry *e)
2598 int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2599 int ipv6_encap_size = ETH_HLEN + sizeof(struct ipv6hdr) + VXLAN_HLEN;
2600 struct ip_tunnel_key *tun_key = &e->tun_info.key;
2601 struct net_device *out_dev;
2602 struct neighbour *n = NULL;
2603 struct flowi6 fl6 = {};
2604 u8 nud_state, tos, ttl;
2608 if (max_encap_size < ipv6_encap_size) {
2609 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2610 ipv6_encap_size, max_encap_size);
2614 encap_header = kzalloc(ipv6_encap_size, GFP_KERNEL);
2618 switch (e->tunnel_type) {
2619 case MLX5_REFORMAT_TYPE_L2_TO_VXLAN:
2620 fl6.flowi6_proto = IPPROTO_UDP;
2621 fl6.fl6_dport = tun_key->tp_dst;
2631 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
2632 fl6.daddr = tun_key->u.ipv6.dst;
2633 fl6.saddr = tun_key->u.ipv6.src;
2635 err = mlx5e_route_lookup_ipv6(priv, mirred_dev, &out_dev,
2640 /* used by mlx5e_detach_encap to lookup a neigh hash table
2641 * entry in the neigh hash table when a user deletes a rule
2643 e->m_neigh.dev = n->dev;
2644 e->m_neigh.family = n->ops->family;
2645 memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2646 e->out_dev = out_dev;
2648 /* It's importent to add the neigh to the hash table before checking
2649 * the neigh validity state. So if we'll get a notification, in case the
2650 * neigh changes it's validity state, we would find the relevant neigh
2653 err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2657 read_lock_bh(&n->lock);
2658 nud_state = n->nud_state;
2659 ether_addr_copy(e->h_dest, n->ha);
2660 read_unlock_bh(&n->lock);
2662 switch (e->tunnel_type) {
2663 case MLX5_REFORMAT_TYPE_L2_TO_VXLAN:
2664 gen_vxlan_header_ipv6(out_dev, encap_header,
2665 ipv6_encap_size, e->h_dest, tos, ttl,
2667 &fl6.saddr, tun_key->tp_dst,
2668 tunnel_id_to_key32(tun_key->tun_id));
2672 goto destroy_neigh_entry;
2675 e->encap_size = ipv6_encap_size;
2676 e->encap_header = encap_header;
2678 if (!(nud_state & NUD_VALID)) {
2679 neigh_event_send(n, NULL);
2684 err = mlx5_packet_reformat_alloc(priv->mdev, e->tunnel_type,
2685 ipv6_encap_size, encap_header,
2686 MLX5_FLOW_NAMESPACE_FDB,
2689 goto destroy_neigh_entry;
2691 e->flags |= MLX5_ENCAP_ENTRY_VALID;
2692 mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2696 destroy_neigh_entry:
2697 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2699 kfree(encap_header);
2706 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
2707 struct ip_tunnel_info *tun_info,
2708 struct net_device *mirred_dev,
2709 struct net_device **encap_dev,
2710 struct mlx5e_tc_flow *flow,
2711 struct netlink_ext_ack *extack)
2713 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2714 unsigned short family = ip_tunnel_info_af(tun_info);
2715 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2716 struct ip_tunnel_key *key = &tun_info->key;
2717 struct mlx5e_encap_entry *e;
2718 int tunnel_type, err = 0;
2722 /* udp dst port must be set */
2723 if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
2724 goto vxlan_encap_offload_err;
2726 /* setting udp src port isn't supported */
2727 if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
2728 vxlan_encap_offload_err:
2729 NL_SET_ERR_MSG_MOD(extack,
2730 "must set udp dst port and not set udp src port");
2731 netdev_warn(priv->netdev,
2732 "must set udp dst port and not set udp src port\n");
2736 if (mlx5_vxlan_lookup_port(priv->mdev->vxlan, be16_to_cpu(key->tp_dst)) &&
2737 MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
2738 tunnel_type = MLX5_REFORMAT_TYPE_L2_TO_VXLAN;
2740 NL_SET_ERR_MSG_MOD(extack,
2741 "port isn't an offloaded vxlan udp dport");
2742 netdev_warn(priv->netdev,
2743 "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
2747 hash_key = hash_encap_info(key);
2749 hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
2750 encap_hlist, hash_key) {
2751 if (!cmp_encap_info(&e->tun_info.key, key)) {
2757 /* must verify if encap is valid or not */
2761 e = kzalloc(sizeof(*e), GFP_KERNEL);
2765 e->tun_info = *tun_info;
2766 e->tunnel_type = tunnel_type;
2767 INIT_LIST_HEAD(&e->flows);
2769 if (family == AF_INET)
2770 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e);
2771 else if (family == AF_INET6)
2772 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e);
2774 if (err && err != -EAGAIN)
2777 hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
2780 list_add(&flow->encap, &e->flows);
2781 *encap_dev = e->out_dev;
2782 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
2783 attr->encap_id = e->encap_id;
2794 static int parse_tc_vlan_action(struct mlx5e_priv *priv,
2795 const struct tc_action *a,
2796 struct mlx5_esw_flow_attr *attr,
2799 u8 vlan_idx = attr->total_vlan;
2801 if (vlan_idx >= MLX5_FS_VLAN_DEPTH)
2804 if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
2806 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
2807 MLX5_FS_VLAN_DEPTH))
2810 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2;
2812 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2814 } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
2815 attr->vlan_vid[vlan_idx] = tcf_vlan_push_vid(a);
2816 attr->vlan_prio[vlan_idx] = tcf_vlan_push_prio(a);
2817 attr->vlan_proto[vlan_idx] = tcf_vlan_push_proto(a);
2818 if (!attr->vlan_proto[vlan_idx])
2819 attr->vlan_proto[vlan_idx] = htons(ETH_P_8021Q);
2822 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
2823 MLX5_FS_VLAN_DEPTH))
2826 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2;
2828 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev, 1) &&
2829 (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q) ||
2830 tcf_vlan_push_prio(a)))
2833 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
2835 } else { /* action is TCA_VLAN_ACT_MODIFY */
2839 attr->total_vlan = vlan_idx + 1;
2844 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
2845 struct mlx5e_tc_flow_parse_attr *parse_attr,
2846 struct mlx5e_tc_flow *flow,
2847 struct netlink_ext_ack *extack)
2849 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2850 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2851 struct mlx5e_rep_priv *rpriv = priv->ppriv;
2852 struct ip_tunnel_info *info = NULL;
2853 const struct tc_action *a;
2859 if (!tcf_exts_has_actions(exts))
2862 attr->in_rep = rpriv->rep;
2863 attr->in_mdev = priv->mdev;
2865 tcf_exts_for_each_action(i, a, exts) {
2866 if (is_tcf_gact_shot(a)) {
2867 action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
2868 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2872 if (is_tcf_pedit(a)) {
2873 err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_FDB,
2874 parse_attr, extack);
2878 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2879 attr->mirror_count = attr->out_count;
2883 if (is_tcf_csum(a)) {
2884 if (csum_offload_supported(priv, action,
2885 tcf_csum_update_flags(a),
2892 if (is_tcf_mirred_egress_redirect(a) || is_tcf_mirred_egress_mirror(a)) {
2893 struct mlx5e_priv *out_priv;
2894 struct net_device *out_dev;
2896 out_dev = tcf_mirred_dev(a);
2898 if (attr->out_count >= MLX5_MAX_FLOW_FWD_VPORTS) {
2899 NL_SET_ERR_MSG_MOD(extack,
2900 "can't support more output ports, can't offload forwarding");
2901 pr_err("can't support more than %d output ports, can't offload forwarding\n",
2906 if (switchdev_port_same_parent_id(priv->netdev,
2908 is_merged_eswitch_dev(priv, out_dev)) {
2909 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2910 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2911 out_priv = netdev_priv(out_dev);
2912 rpriv = out_priv->ppriv;
2913 attr->out_rep[attr->out_count] = rpriv->rep;
2914 attr->out_mdev[attr->out_count++] = out_priv->mdev;
2916 parse_attr->mirred_ifindex = out_dev->ifindex;
2917 parse_attr->tun_info = *info;
2918 attr->parse_attr = parse_attr;
2919 action |= MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
2920 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2921 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2922 /* attr->out_rep is resolved when we handle encap */
2924 NL_SET_ERR_MSG_MOD(extack,
2925 "devices are not on same switch HW, can't offload forwarding");
2926 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
2927 priv->netdev->name, out_dev->name);
2933 if (is_tcf_tunnel_set(a)) {
2934 info = tcf_tunnel_info(a);
2939 attr->mirror_count = attr->out_count;
2943 if (is_tcf_vlan(a)) {
2944 err = parse_tc_vlan_action(priv, a, attr, &action);
2949 attr->mirror_count = attr->out_count;
2953 if (is_tcf_tunnel_release(a)) {
2954 action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
2958 if (is_tcf_gact_goto_chain(a)) {
2959 u32 dest_chain = tcf_gact_goto_chain_index(a);
2960 u32 max_chain = mlx5_eswitch_get_chain_range(esw);
2962 if (dest_chain <= attr->chain) {
2963 NL_SET_ERR_MSG(extack, "Goto earlier chain isn't supported");
2966 if (dest_chain > max_chain) {
2967 NL_SET_ERR_MSG(extack, "Requested destination chain is out of supported range");
2970 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2971 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2972 attr->dest_chain = dest_chain;
2980 attr->action = action;
2981 if (!actions_match_supported(priv, exts, parse_attr, flow, extack))
2984 if (attr->mirror_count > 0 && !mlx5_esw_has_fwd_fdb(priv->mdev)) {
2985 NL_SET_ERR_MSG_MOD(extack,
2986 "current firmware doesn't support split rule for port mirroring");
2987 netdev_warn_once(priv->netdev, "current firmware doesn't support split rule for port mirroring\n");
2994 static void get_flags(int flags, u16 *flow_flags)
2996 u16 __flow_flags = 0;
2998 if (flags & MLX5E_TC_INGRESS)
2999 __flow_flags |= MLX5E_TC_FLOW_INGRESS;
3000 if (flags & MLX5E_TC_EGRESS)
3001 __flow_flags |= MLX5E_TC_FLOW_EGRESS;
3003 *flow_flags = __flow_flags;
3006 static const struct rhashtable_params tc_ht_params = {
3007 .head_offset = offsetof(struct mlx5e_tc_flow, node),
3008 .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
3009 .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
3010 .automatic_shrinking = true,
3013 static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv)
3015 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3016 struct mlx5e_rep_priv *uplink_rpriv;
3018 if (MLX5_VPORT_MANAGER(priv->mdev) && esw->mode == SRIOV_OFFLOADS) {
3019 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
3020 return &uplink_rpriv->tc_ht;
3022 return &priv->fs.tc.ht;
3026 mlx5e_alloc_flow(struct mlx5e_priv *priv, int attr_size,
3027 struct tc_cls_flower_offload *f, u16 flow_flags,
3028 struct mlx5e_tc_flow_parse_attr **__parse_attr,
3029 struct mlx5e_tc_flow **__flow)
3031 struct mlx5e_tc_flow_parse_attr *parse_attr;
3032 struct mlx5e_tc_flow *flow;
3035 flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
3036 parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
3037 if (!parse_attr || !flow) {
3042 flow->cookie = f->cookie;
3043 flow->flags = flow_flags;
3046 err = parse_cls_flower(priv, flow, &parse_attr->spec, f);
3051 *__parse_attr = parse_attr;
3062 mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
3063 struct tc_cls_flower_offload *f,
3065 struct mlx5e_tc_flow **__flow)
3067 struct netlink_ext_ack *extack = f->common.extack;
3068 struct mlx5e_tc_flow_parse_attr *parse_attr;
3069 struct mlx5e_tc_flow *flow;
3072 flow_flags |= MLX5E_TC_FLOW_ESWITCH;
3073 attr_size = sizeof(struct mlx5_esw_flow_attr);
3074 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3075 &parse_attr, &flow);
3079 flow->esw_attr->chain = f->common.chain_index;
3080 flow->esw_attr->prio = TC_H_MAJ(f->common.prio) >> 16;
3081 err = parse_tc_fdb_actions(priv, f->exts, parse_attr, flow, extack);
3085 err = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow, extack);
3089 if (!(flow->esw_attr->action &
3090 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT))
3105 mlx5e_add_nic_flow(struct mlx5e_priv *priv,
3106 struct tc_cls_flower_offload *f,
3108 struct mlx5e_tc_flow **__flow)
3110 struct netlink_ext_ack *extack = f->common.extack;
3111 struct mlx5e_tc_flow_parse_attr *parse_attr;
3112 struct mlx5e_tc_flow *flow;
3115 /* multi-chain not supported for NIC rules */
3116 if (!tc_cls_can_offload_and_chain0(priv->netdev, &f->common))
3119 flow_flags |= MLX5E_TC_FLOW_NIC;
3120 attr_size = sizeof(struct mlx5_nic_flow_attr);
3121 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3122 &parse_attr, &flow);
3126 err = parse_tc_nic_actions(priv, f->exts, parse_attr, flow, extack);
3130 err = mlx5e_tc_add_nic_flow(priv, parse_attr, flow, extack);
3134 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
3148 mlx5e_tc_add_flow(struct mlx5e_priv *priv,
3149 struct tc_cls_flower_offload *f,
3151 struct mlx5e_tc_flow **flow)
3153 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3157 get_flags(flags, &flow_flags);
3159 if (!tc_can_offload_extack(priv->netdev, f->common.extack))
3162 if (esw && esw->mode == SRIOV_OFFLOADS)
3163 err = mlx5e_add_fdb_flow(priv, f, flow_flags, flow);
3165 err = mlx5e_add_nic_flow(priv, f, flow_flags, flow);
3170 int mlx5e_configure_flower(struct mlx5e_priv *priv,
3171 struct tc_cls_flower_offload *f, int flags)
3173 struct netlink_ext_ack *extack = f->common.extack;
3174 struct rhashtable *tc_ht = get_tc_ht(priv);
3175 struct mlx5e_tc_flow *flow;
3178 flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
3180 NL_SET_ERR_MSG_MOD(extack,
3181 "flow cookie already exists, ignoring");
3182 netdev_warn_once(priv->netdev,
3183 "flow cookie %lx already exists, ignoring\n",
3188 err = mlx5e_tc_add_flow(priv, f, flags, &flow);
3192 err = rhashtable_insert_fast(tc_ht, &flow->node, tc_ht_params);
3199 mlx5e_tc_del_flow(priv, flow);
3205 #define DIRECTION_MASK (MLX5E_TC_INGRESS | MLX5E_TC_EGRESS)
3206 #define FLOW_DIRECTION_MASK (MLX5E_TC_FLOW_INGRESS | MLX5E_TC_FLOW_EGRESS)
3208 static bool same_flow_direction(struct mlx5e_tc_flow *flow, int flags)
3210 if ((flow->flags & FLOW_DIRECTION_MASK) == (flags & DIRECTION_MASK))
3216 int mlx5e_delete_flower(struct mlx5e_priv *priv,
3217 struct tc_cls_flower_offload *f, int flags)
3219 struct rhashtable *tc_ht = get_tc_ht(priv);
3220 struct mlx5e_tc_flow *flow;
3222 flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
3223 if (!flow || !same_flow_direction(flow, flags))
3226 rhashtable_remove_fast(tc_ht, &flow->node, tc_ht_params);
3228 mlx5e_tc_del_flow(priv, flow);
3235 int mlx5e_stats_flower(struct mlx5e_priv *priv,
3236 struct tc_cls_flower_offload *f, int flags)
3238 struct rhashtable *tc_ht = get_tc_ht(priv);
3239 struct mlx5e_tc_flow *flow;
3240 struct mlx5_fc *counter;
3245 flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
3246 if (!flow || !same_flow_direction(flow, flags))
3249 if (!(flow->flags & MLX5E_TC_FLOW_OFFLOADED))
3252 counter = mlx5e_tc_get_counter(flow);
3256 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
3258 tcf_exts_stats_update(f->exts, bytes, packets, lastuse);
3263 static void mlx5e_tc_hairpin_update_dead_peer(struct mlx5e_priv *priv,
3264 struct mlx5e_priv *peer_priv)
3266 struct mlx5_core_dev *peer_mdev = peer_priv->mdev;
3267 struct mlx5e_hairpin_entry *hpe;
3271 if (!same_hw_devs(priv, peer_priv))
3274 peer_vhca_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
3276 hash_for_each(priv->fs.tc.hairpin_tbl, bkt, hpe, hairpin_hlist) {
3277 if (hpe->peer_vhca_id == peer_vhca_id)
3278 hpe->hp->pair->peer_gone = true;
3282 static int mlx5e_tc_netdev_event(struct notifier_block *this,
3283 unsigned long event, void *ptr)
3285 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3286 struct mlx5e_flow_steering *fs;
3287 struct mlx5e_priv *peer_priv;
3288 struct mlx5e_tc_table *tc;
3289 struct mlx5e_priv *priv;
3291 if (ndev->netdev_ops != &mlx5e_netdev_ops ||
3292 event != NETDEV_UNREGISTER ||
3293 ndev->reg_state == NETREG_REGISTERED)
3296 tc = container_of(this, struct mlx5e_tc_table, netdevice_nb);
3297 fs = container_of(tc, struct mlx5e_flow_steering, tc);
3298 priv = container_of(fs, struct mlx5e_priv, fs);
3299 peer_priv = netdev_priv(ndev);
3300 if (priv == peer_priv ||
3301 !(priv->netdev->features & NETIF_F_HW_TC))
3304 mlx5e_tc_hairpin_update_dead_peer(priv, peer_priv);
3309 int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
3311 struct mlx5e_tc_table *tc = &priv->fs.tc;
3314 hash_init(tc->mod_hdr_tbl);
3315 hash_init(tc->hairpin_tbl);
3317 err = rhashtable_init(&tc->ht, &tc_ht_params);
3321 tc->netdevice_nb.notifier_call = mlx5e_tc_netdev_event;
3322 if (register_netdevice_notifier(&tc->netdevice_nb)) {
3323 tc->netdevice_nb.notifier_call = NULL;
3324 mlx5_core_warn(priv->mdev, "Failed to register netdev notifier\n");
3330 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
3332 struct mlx5e_tc_flow *flow = ptr;
3333 struct mlx5e_priv *priv = flow->priv;
3335 mlx5e_tc_del_flow(priv, flow);
3339 void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
3341 struct mlx5e_tc_table *tc = &priv->fs.tc;
3343 if (tc->netdevice_nb.notifier_call)
3344 unregister_netdevice_notifier(&tc->netdevice_nb);
3346 rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, NULL);
3348 if (!IS_ERR_OR_NULL(tc->t)) {
3349 mlx5_destroy_flow_table(tc->t);
3354 int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
3356 return rhashtable_init(tc_ht, &tc_ht_params);
3359 void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
3361 rhashtable_free_and_destroy(tc_ht, _mlx5e_tc_del_flow, NULL);
3364 int mlx5e_tc_num_filters(struct mlx5e_priv *priv)
3366 struct rhashtable *tc_ht = get_tc_ht(priv);
3368 return atomic_read(&tc_ht->nelems);