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/tc_act/tc_mirred.h>
42 #include <net/tc_act/tc_vlan.h>
43 #include <net/tc_act/tc_tunnel_key.h>
44 #include <net/tc_act/tc_pedit.h>
45 #include <net/tc_act/tc_csum.h>
47 #include <net/ipv6_stubs.h>
54 #include "en/tc_tun.h"
55 #include "lib/devcom.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 = MLX5E_TC_ESW_OFFLOAD,
73 MLX5E_TC_FLOW_NIC = MLX5E_TC_NIC_OFFLOAD,
74 MLX5E_TC_FLOW_OFFLOADED = BIT(MLX5E_TC_FLOW_BASE),
75 MLX5E_TC_FLOW_HAIRPIN = BIT(MLX5E_TC_FLOW_BASE + 1),
76 MLX5E_TC_FLOW_HAIRPIN_RSS = BIT(MLX5E_TC_FLOW_BASE + 2),
77 MLX5E_TC_FLOW_SLOW = BIT(MLX5E_TC_FLOW_BASE + 3),
78 MLX5E_TC_FLOW_DUP = BIT(MLX5E_TC_FLOW_BASE + 4),
79 MLX5E_TC_FLOW_NOT_READY = BIT(MLX5E_TC_FLOW_BASE + 5),
82 #define MLX5E_TC_MAX_SPLITS 1
84 /* Helper struct for accessing a struct containing list_head array.
93 * To access the containing struct from one of the list_head items:
94 * 1. Get the helper item from the list_head item using
96 * container_of(list_head item, helper struct type, list_head field)
97 * 2. Get the contining struct from the helper item and its index in the array:
99 * container_of(helper item, containing struct type, helper field[index])
101 struct encap_flow_item {
102 struct list_head list;
106 struct mlx5e_tc_flow {
107 struct rhash_head node;
108 struct mlx5e_priv *priv;
111 struct mlx5_flow_handle *rule[MLX5E_TC_MAX_SPLITS + 1];
112 /* Flow can be associated with multiple encap IDs.
113 * The number of encaps is bounded by the number of supported
116 struct encap_flow_item encaps[MLX5_MAX_FLOW_FWD_VPORTS];
117 struct mlx5e_tc_flow *peer_flow;
118 struct list_head mod_hdr; /* flows sharing the same mod hdr ID */
119 struct list_head hairpin; /* flows sharing the same hairpin */
120 struct list_head peer; /* flows with peer flow */
121 struct list_head unready; /* flows not ready to be offloaded (e.g due to missing route) */
123 struct mlx5_esw_flow_attr esw_attr[0];
124 struct mlx5_nic_flow_attr nic_attr[0];
128 struct mlx5e_tc_flow_parse_attr {
129 struct ip_tunnel_info tun_info[MLX5_MAX_FLOW_FWD_VPORTS];
130 struct net_device *filter_dev;
131 struct mlx5_flow_spec spec;
132 int num_mod_hdr_actions;
133 int max_mod_hdr_actions;
134 void *mod_hdr_actions;
135 int mirred_ifindex[MLX5_MAX_FLOW_FWD_VPORTS];
138 #define MLX5E_TC_TABLE_NUM_GROUPS 4
139 #define MLX5E_TC_TABLE_MAX_GROUP_SIZE BIT(16)
141 struct mlx5e_hairpin {
142 struct mlx5_hairpin *pair;
144 struct mlx5_core_dev *func_mdev;
145 struct mlx5e_priv *func_priv;
150 struct mlx5e_rqt indir_rqt;
151 u32 indir_tirn[MLX5E_NUM_INDIR_TIRS];
152 struct mlx5e_ttc_table ttc;
155 struct mlx5e_hairpin_entry {
156 /* a node of a hash table which keeps all the hairpin entries */
157 struct hlist_node hairpin_hlist;
159 /* flows sharing the same hairpin */
160 struct list_head flows;
164 struct mlx5e_hairpin *hp;
172 struct mlx5e_mod_hdr_entry {
173 /* a node of a hash table which keeps all the mod_hdr entries */
174 struct hlist_node mod_hdr_hlist;
176 /* flows sharing the same mod_hdr entry */
177 struct list_head flows;
179 struct mod_hdr_key key;
184 #define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)
186 static inline u32 hash_mod_hdr_info(struct mod_hdr_key *key)
188 return jhash(key->actions,
189 key->num_actions * MLX5_MH_ACT_SZ, 0);
192 static inline int cmp_mod_hdr_info(struct mod_hdr_key *a,
193 struct mod_hdr_key *b)
195 if (a->num_actions != b->num_actions)
198 return memcmp(a->actions, b->actions, a->num_actions * MLX5_MH_ACT_SZ);
201 static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
202 struct mlx5e_tc_flow *flow,
203 struct mlx5e_tc_flow_parse_attr *parse_attr)
205 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
206 int num_actions, actions_size, namespace, err;
207 struct mlx5e_mod_hdr_entry *mh;
208 struct mod_hdr_key key;
212 num_actions = parse_attr->num_mod_hdr_actions;
213 actions_size = MLX5_MH_ACT_SZ * num_actions;
215 key.actions = parse_attr->mod_hdr_actions;
216 key.num_actions = num_actions;
218 hash_key = hash_mod_hdr_info(&key);
220 if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
221 namespace = MLX5_FLOW_NAMESPACE_FDB;
222 hash_for_each_possible(esw->offloads.mod_hdr_tbl, mh,
223 mod_hdr_hlist, hash_key) {
224 if (!cmp_mod_hdr_info(&mh->key, &key)) {
230 namespace = MLX5_FLOW_NAMESPACE_KERNEL;
231 hash_for_each_possible(priv->fs.tc.mod_hdr_tbl, mh,
232 mod_hdr_hlist, hash_key) {
233 if (!cmp_mod_hdr_info(&mh->key, &key)) {
243 mh = kzalloc(sizeof(*mh) + actions_size, GFP_KERNEL);
247 mh->key.actions = (void *)mh + sizeof(*mh);
248 memcpy(mh->key.actions, key.actions, actions_size);
249 mh->key.num_actions = num_actions;
250 INIT_LIST_HEAD(&mh->flows);
252 err = mlx5_modify_header_alloc(priv->mdev, namespace,
259 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
260 hash_add(esw->offloads.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
262 hash_add(priv->fs.tc.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
265 list_add(&flow->mod_hdr, &mh->flows);
266 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
267 flow->esw_attr->mod_hdr_id = mh->mod_hdr_id;
269 flow->nic_attr->mod_hdr_id = mh->mod_hdr_id;
278 static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
279 struct mlx5e_tc_flow *flow)
281 struct list_head *next = flow->mod_hdr.next;
283 list_del(&flow->mod_hdr);
285 if (list_empty(next)) {
286 struct mlx5e_mod_hdr_entry *mh;
288 mh = list_entry(next, struct mlx5e_mod_hdr_entry, flows);
290 mlx5_modify_header_dealloc(priv->mdev, mh->mod_hdr_id);
291 hash_del(&mh->mod_hdr_hlist);
297 struct mlx5_core_dev *mlx5e_hairpin_get_mdev(struct net *net, int ifindex)
299 struct net_device *netdev;
300 struct mlx5e_priv *priv;
302 netdev = __dev_get_by_index(net, ifindex);
303 priv = netdev_priv(netdev);
307 static int mlx5e_hairpin_create_transport(struct mlx5e_hairpin *hp)
309 u32 in[MLX5_ST_SZ_DW(create_tir_in)] = {0};
313 err = mlx5_core_alloc_transport_domain(hp->func_mdev, &hp->tdn);
317 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
319 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
320 MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn[0]);
321 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
323 err = mlx5_core_create_tir(hp->func_mdev, in, MLX5_ST_SZ_BYTES(create_tir_in), &hp->tirn);
330 mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
335 static void mlx5e_hairpin_destroy_transport(struct mlx5e_hairpin *hp)
337 mlx5_core_destroy_tir(hp->func_mdev, hp->tirn);
338 mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
341 static void mlx5e_hairpin_fill_rqt_rqns(struct mlx5e_hairpin *hp, void *rqtc)
343 u32 indirection_rqt[MLX5E_INDIR_RQT_SIZE], rqn;
344 struct mlx5e_priv *priv = hp->func_priv;
345 int i, ix, sz = MLX5E_INDIR_RQT_SIZE;
347 mlx5e_build_default_indir_rqt(indirection_rqt, sz,
350 for (i = 0; i < sz; i++) {
352 if (priv->rss_params.hfunc == ETH_RSS_HASH_XOR)
353 ix = mlx5e_bits_invert(i, ilog2(sz));
354 ix = indirection_rqt[ix];
355 rqn = hp->pair->rqn[ix];
356 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
360 static int mlx5e_hairpin_create_indirect_rqt(struct mlx5e_hairpin *hp)
362 int inlen, err, sz = MLX5E_INDIR_RQT_SIZE;
363 struct mlx5e_priv *priv = hp->func_priv;
364 struct mlx5_core_dev *mdev = priv->mdev;
368 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
369 in = kvzalloc(inlen, GFP_KERNEL);
373 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
375 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
376 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
378 mlx5e_hairpin_fill_rqt_rqns(hp, rqtc);
380 err = mlx5_core_create_rqt(mdev, in, inlen, &hp->indir_rqt.rqtn);
382 hp->indir_rqt.enabled = true;
388 static int mlx5e_hairpin_create_indirect_tirs(struct mlx5e_hairpin *hp)
390 struct mlx5e_priv *priv = hp->func_priv;
391 u32 in[MLX5_ST_SZ_DW(create_tir_in)];
395 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
396 struct mlx5e_tirc_config ttconfig = mlx5e_tirc_get_default_config(tt);
398 memset(in, 0, MLX5_ST_SZ_BYTES(create_tir_in));
399 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
401 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
402 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
403 MLX5_SET(tirc, tirc, indirect_table, hp->indir_rqt.rqtn);
404 mlx5e_build_indir_tir_ctx_hash(&priv->rss_params, &ttconfig, tirc, false);
406 err = mlx5_core_create_tir(hp->func_mdev, in,
407 MLX5_ST_SZ_BYTES(create_tir_in), &hp->indir_tirn[tt]);
409 mlx5_core_warn(hp->func_mdev, "create indirect tirs failed, %d\n", err);
410 goto err_destroy_tirs;
416 for (i = 0; i < tt; i++)
417 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[i]);
421 static void mlx5e_hairpin_destroy_indirect_tirs(struct mlx5e_hairpin *hp)
425 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
426 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[tt]);
429 static void mlx5e_hairpin_set_ttc_params(struct mlx5e_hairpin *hp,
430 struct ttc_params *ttc_params)
432 struct mlx5_flow_table_attr *ft_attr = &ttc_params->ft_attr;
435 memset(ttc_params, 0, sizeof(*ttc_params));
437 ttc_params->any_tt_tirn = hp->tirn;
439 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
440 ttc_params->indir_tirn[tt] = hp->indir_tirn[tt];
442 ft_attr->max_fte = MLX5E_NUM_TT;
443 ft_attr->level = MLX5E_TC_TTC_FT_LEVEL;
444 ft_attr->prio = MLX5E_TC_PRIO;
447 static int mlx5e_hairpin_rss_init(struct mlx5e_hairpin *hp)
449 struct mlx5e_priv *priv = hp->func_priv;
450 struct ttc_params ttc_params;
453 err = mlx5e_hairpin_create_indirect_rqt(hp);
457 err = mlx5e_hairpin_create_indirect_tirs(hp);
459 goto err_create_indirect_tirs;
461 mlx5e_hairpin_set_ttc_params(hp, &ttc_params);
462 err = mlx5e_create_ttc_table(priv, &ttc_params, &hp->ttc);
464 goto err_create_ttc_table;
466 netdev_dbg(priv->netdev, "add hairpin: using %d channels rss ttc table id %x\n",
467 hp->num_channels, hp->ttc.ft.t->id);
471 err_create_ttc_table:
472 mlx5e_hairpin_destroy_indirect_tirs(hp);
473 err_create_indirect_tirs:
474 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
479 static void mlx5e_hairpin_rss_cleanup(struct mlx5e_hairpin *hp)
481 struct mlx5e_priv *priv = hp->func_priv;
483 mlx5e_destroy_ttc_table(priv, &hp->ttc);
484 mlx5e_hairpin_destroy_indirect_tirs(hp);
485 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
488 static struct mlx5e_hairpin *
489 mlx5e_hairpin_create(struct mlx5e_priv *priv, struct mlx5_hairpin_params *params,
492 struct mlx5_core_dev *func_mdev, *peer_mdev;
493 struct mlx5e_hairpin *hp;
494 struct mlx5_hairpin *pair;
497 hp = kzalloc(sizeof(*hp), GFP_KERNEL);
499 return ERR_PTR(-ENOMEM);
501 func_mdev = priv->mdev;
502 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
504 pair = mlx5_core_hairpin_create(func_mdev, peer_mdev, params);
507 goto create_pair_err;
510 hp->func_mdev = func_mdev;
511 hp->func_priv = priv;
512 hp->num_channels = params->num_channels;
514 err = mlx5e_hairpin_create_transport(hp);
516 goto create_transport_err;
518 if (hp->num_channels > 1) {
519 err = mlx5e_hairpin_rss_init(hp);
527 mlx5e_hairpin_destroy_transport(hp);
528 create_transport_err:
529 mlx5_core_hairpin_destroy(hp->pair);
535 static void mlx5e_hairpin_destroy(struct mlx5e_hairpin *hp)
537 if (hp->num_channels > 1)
538 mlx5e_hairpin_rss_cleanup(hp);
539 mlx5e_hairpin_destroy_transport(hp);
540 mlx5_core_hairpin_destroy(hp->pair);
544 static inline u32 hash_hairpin_info(u16 peer_vhca_id, u8 prio)
546 return (peer_vhca_id << 16 | prio);
549 static struct mlx5e_hairpin_entry *mlx5e_hairpin_get(struct mlx5e_priv *priv,
550 u16 peer_vhca_id, u8 prio)
552 struct mlx5e_hairpin_entry *hpe;
553 u32 hash_key = hash_hairpin_info(peer_vhca_id, prio);
555 hash_for_each_possible(priv->fs.tc.hairpin_tbl, hpe,
556 hairpin_hlist, hash_key) {
557 if (hpe->peer_vhca_id == peer_vhca_id && hpe->prio == prio)
564 #define UNKNOWN_MATCH_PRIO 8
566 static int mlx5e_hairpin_get_prio(struct mlx5e_priv *priv,
567 struct mlx5_flow_spec *spec, u8 *match_prio,
568 struct netlink_ext_ack *extack)
570 void *headers_c, *headers_v;
571 u8 prio_val, prio_mask = 0;
574 #ifdef CONFIG_MLX5_CORE_EN_DCB
575 if (priv->dcbx_dp.trust_state != MLX5_QPTS_TRUST_PCP) {
576 NL_SET_ERR_MSG_MOD(extack,
577 "only PCP trust state supported for hairpin");
581 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, outer_headers);
582 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
584 vlan_present = MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag);
586 prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
587 prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
590 if (!vlan_present || !prio_mask) {
591 prio_val = UNKNOWN_MATCH_PRIO;
592 } else if (prio_mask != 0x7) {
593 NL_SET_ERR_MSG_MOD(extack,
594 "masked priority match not supported for hairpin");
598 *match_prio = prio_val;
602 static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
603 struct mlx5e_tc_flow *flow,
604 struct mlx5e_tc_flow_parse_attr *parse_attr,
605 struct netlink_ext_ack *extack)
607 int peer_ifindex = parse_attr->mirred_ifindex[0];
608 struct mlx5_hairpin_params params;
609 struct mlx5_core_dev *peer_mdev;
610 struct mlx5e_hairpin_entry *hpe;
611 struct mlx5e_hairpin *hp;
618 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
619 if (!MLX5_CAP_GEN(priv->mdev, hairpin) || !MLX5_CAP_GEN(peer_mdev, hairpin)) {
620 NL_SET_ERR_MSG_MOD(extack, "hairpin is not supported");
624 peer_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
625 err = mlx5e_hairpin_get_prio(priv, &parse_attr->spec, &match_prio,
629 hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
633 hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
637 INIT_LIST_HEAD(&hpe->flows);
638 hpe->peer_vhca_id = peer_id;
639 hpe->prio = match_prio;
641 params.log_data_size = 15;
642 params.log_data_size = min_t(u8, params.log_data_size,
643 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_wq_data_sz));
644 params.log_data_size = max_t(u8, params.log_data_size,
645 MLX5_CAP_GEN(priv->mdev, log_min_hairpin_wq_data_sz));
647 params.log_num_packets = params.log_data_size -
648 MLX5_MPWRQ_MIN_LOG_STRIDE_SZ(priv->mdev);
649 params.log_num_packets = min_t(u8, params.log_num_packets,
650 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_num_packets));
652 params.q_counter = priv->q_counter;
653 /* set hairpin pair per each 50Gbs share of the link */
654 mlx5e_port_max_linkspeed(priv->mdev, &link_speed);
655 link_speed = max_t(u32, link_speed, 50000);
656 link_speed64 = link_speed;
657 do_div(link_speed64, 50000);
658 params.num_channels = link_speed64;
660 hp = mlx5e_hairpin_create(priv, ¶ms, peer_ifindex);
663 goto create_hairpin_err;
666 netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
667 hp->tirn, hp->pair->rqn[0],
668 dev_name(hp->pair->peer_mdev->device),
669 hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
672 hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
673 hash_hairpin_info(peer_id, match_prio));
676 if (hpe->hp->num_channels > 1) {
677 flow->flags |= MLX5E_TC_FLOW_HAIRPIN_RSS;
678 flow->nic_attr->hairpin_ft = hpe->hp->ttc.ft.t;
680 flow->nic_attr->hairpin_tirn = hpe->hp->tirn;
682 list_add(&flow->hairpin, &hpe->flows);
691 static void mlx5e_hairpin_flow_del(struct mlx5e_priv *priv,
692 struct mlx5e_tc_flow *flow)
694 struct list_head *next = flow->hairpin.next;
696 list_del(&flow->hairpin);
698 /* no more hairpin flows for us, release the hairpin pair */
699 if (list_empty(next)) {
700 struct mlx5e_hairpin_entry *hpe;
702 hpe = list_entry(next, struct mlx5e_hairpin_entry, flows);
704 netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
705 dev_name(hpe->hp->pair->peer_mdev->device));
707 mlx5e_hairpin_destroy(hpe->hp);
708 hash_del(&hpe->hairpin_hlist);
714 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
715 struct mlx5e_tc_flow_parse_attr *parse_attr,
716 struct mlx5e_tc_flow *flow,
717 struct netlink_ext_ack *extack)
719 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
720 struct mlx5_core_dev *dev = priv->mdev;
721 struct mlx5_flow_destination dest[2] = {};
722 struct mlx5_flow_act flow_act = {
723 .action = attr->action,
724 .flow_tag = attr->flow_tag,
726 .flags = FLOW_ACT_HAS_TAG | FLOW_ACT_NO_APPEND,
728 struct mlx5_fc *counter = NULL;
729 bool table_created = false;
730 int err, dest_ix = 0;
732 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN) {
733 err = mlx5e_hairpin_flow_add(priv, flow, parse_attr, extack);
735 goto err_add_hairpin_flow;
737 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN_RSS) {
738 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
739 dest[dest_ix].ft = attr->hairpin_ft;
741 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_TIR;
742 dest[dest_ix].tir_num = attr->hairpin_tirn;
745 } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
746 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
747 dest[dest_ix].ft = priv->fs.vlan.ft.t;
751 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
752 counter = mlx5_fc_create(dev, true);
753 if (IS_ERR(counter)) {
754 err = PTR_ERR(counter);
757 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
758 dest[dest_ix].counter_id = mlx5_fc_id(counter);
760 attr->counter = counter;
763 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
764 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
765 flow_act.modify_id = attr->mod_hdr_id;
766 kfree(parse_attr->mod_hdr_actions);
768 goto err_create_mod_hdr_id;
771 if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
772 int tc_grp_size, tc_tbl_size;
773 u32 max_flow_counter;
775 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
776 MLX5_CAP_GEN(dev, max_flow_counter_15_0);
778 tc_grp_size = min_t(int, max_flow_counter, MLX5E_TC_TABLE_MAX_GROUP_SIZE);
780 tc_tbl_size = min_t(int, tc_grp_size * MLX5E_TC_TABLE_NUM_GROUPS,
781 BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev, log_max_ft_size)));
784 mlx5_create_auto_grouped_flow_table(priv->fs.ns,
787 MLX5E_TC_TABLE_NUM_GROUPS,
788 MLX5E_TC_FT_LEVEL, 0);
789 if (IS_ERR(priv->fs.tc.t)) {
790 NL_SET_ERR_MSG_MOD(extack,
791 "Failed to create tc offload table\n");
792 netdev_err(priv->netdev,
793 "Failed to create tc offload table\n");
794 err = PTR_ERR(priv->fs.tc.t);
798 table_created = true;
801 if (attr->match_level != MLX5_MATCH_NONE)
802 parse_attr->spec.match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
804 flow->rule[0] = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
805 &flow_act, dest, dest_ix);
807 if (IS_ERR(flow->rule[0])) {
808 err = PTR_ERR(flow->rule[0]);
816 mlx5_destroy_flow_table(priv->fs.tc.t);
817 priv->fs.tc.t = NULL;
820 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
821 mlx5e_detach_mod_hdr(priv, flow);
822 err_create_mod_hdr_id:
823 mlx5_fc_destroy(dev, counter);
825 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
826 mlx5e_hairpin_flow_del(priv, flow);
827 err_add_hairpin_flow:
831 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
832 struct mlx5e_tc_flow *flow)
834 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
835 struct mlx5_fc *counter = NULL;
837 counter = attr->counter;
838 mlx5_del_flow_rules(flow->rule[0]);
839 mlx5_fc_destroy(priv->mdev, counter);
841 if (!mlx5e_tc_num_filters(priv, MLX5E_TC_NIC_OFFLOAD) && priv->fs.tc.t) {
842 mlx5_destroy_flow_table(priv->fs.tc.t);
843 priv->fs.tc.t = NULL;
846 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
847 mlx5e_detach_mod_hdr(priv, flow);
849 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
850 mlx5e_hairpin_flow_del(priv, flow);
853 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
854 struct mlx5e_tc_flow *flow, int out_index);
856 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
857 struct mlx5e_tc_flow *flow,
858 struct net_device *mirred_dev,
860 struct netlink_ext_ack *extack,
861 struct net_device **encap_dev,
864 static struct mlx5_flow_handle *
865 mlx5e_tc_offload_fdb_rules(struct mlx5_eswitch *esw,
866 struct mlx5e_tc_flow *flow,
867 struct mlx5_flow_spec *spec,
868 struct mlx5_esw_flow_attr *attr)
870 struct mlx5_flow_handle *rule;
872 rule = mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
876 if (attr->split_count) {
877 flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, spec, attr);
878 if (IS_ERR(flow->rule[1])) {
879 mlx5_eswitch_del_offloaded_rule(esw, rule, attr);
880 return flow->rule[1];
884 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
889 mlx5e_tc_unoffload_fdb_rules(struct mlx5_eswitch *esw,
890 struct mlx5e_tc_flow *flow,
891 struct mlx5_esw_flow_attr *attr)
893 flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
895 if (attr->split_count)
896 mlx5_eswitch_del_fwd_rule(esw, flow->rule[1], attr);
898 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
901 static struct mlx5_flow_handle *
902 mlx5e_tc_offload_to_slow_path(struct mlx5_eswitch *esw,
903 struct mlx5e_tc_flow *flow,
904 struct mlx5_flow_spec *spec,
905 struct mlx5_esw_flow_attr *slow_attr)
907 struct mlx5_flow_handle *rule;
909 memcpy(slow_attr, flow->esw_attr, sizeof(*slow_attr));
910 slow_attr->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
911 slow_attr->split_count = 0;
912 slow_attr->dest_chain = FDB_SLOW_PATH_CHAIN;
914 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, slow_attr);
916 flow->flags |= MLX5E_TC_FLOW_SLOW;
922 mlx5e_tc_unoffload_from_slow_path(struct mlx5_eswitch *esw,
923 struct mlx5e_tc_flow *flow,
924 struct mlx5_esw_flow_attr *slow_attr)
926 memcpy(slow_attr, flow->esw_attr, sizeof(*slow_attr));
927 slow_attr->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
928 slow_attr->split_count = 0;
929 slow_attr->dest_chain = FDB_SLOW_PATH_CHAIN;
930 mlx5e_tc_unoffload_fdb_rules(esw, flow, slow_attr);
931 flow->flags &= ~MLX5E_TC_FLOW_SLOW;
934 static void add_unready_flow(struct mlx5e_tc_flow *flow)
936 struct mlx5_rep_uplink_priv *uplink_priv;
937 struct mlx5e_rep_priv *rpriv;
938 struct mlx5_eswitch *esw;
940 esw = flow->priv->mdev->priv.eswitch;
941 rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
942 uplink_priv = &rpriv->uplink_priv;
944 flow->flags |= MLX5E_TC_FLOW_NOT_READY;
945 list_add_tail(&flow->unready, &uplink_priv->unready_flows);
948 static void remove_unready_flow(struct mlx5e_tc_flow *flow)
950 list_del(&flow->unready);
951 flow->flags &= ~MLX5E_TC_FLOW_NOT_READY;
955 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
956 struct mlx5e_tc_flow *flow,
957 struct netlink_ext_ack *extack)
959 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
960 u32 max_chain = mlx5_eswitch_get_chain_range(esw);
961 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
962 struct mlx5e_tc_flow_parse_attr *parse_attr = attr->parse_attr;
963 u16 max_prio = mlx5_eswitch_get_prio_range(esw);
964 struct net_device *out_dev, *encap_dev = NULL;
965 struct mlx5_fc *counter = NULL;
966 struct mlx5e_rep_priv *rpriv;
967 struct mlx5e_priv *out_priv;
968 bool encap_valid = true;
972 if (!mlx5_eswitch_prios_supported(esw) && attr->prio != 1) {
973 NL_SET_ERR_MSG(extack, "E-switch priorities unsupported, upgrade FW");
977 if (attr->chain > max_chain) {
978 NL_SET_ERR_MSG(extack, "Requested chain is out of supported range");
980 goto err_max_prio_chain;
983 if (attr->prio > max_prio) {
984 NL_SET_ERR_MSG(extack, "Requested priority is out of supported range");
986 goto err_max_prio_chain;
989 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++) {
992 if (!(attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP))
995 mirred_ifindex = parse_attr->mirred_ifindex[out_index];
996 out_dev = __dev_get_by_index(dev_net(priv->netdev),
998 err = mlx5e_attach_encap(priv, flow, out_dev, out_index,
999 extack, &encap_dev, &encap_valid);
1001 goto err_attach_encap;
1003 out_priv = netdev_priv(encap_dev);
1004 rpriv = out_priv->ppriv;
1005 attr->dests[out_index].rep = rpriv->rep;
1006 attr->dests[out_index].mdev = out_priv->mdev;
1009 err = mlx5_eswitch_add_vlan_action(esw, attr);
1013 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
1014 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
1015 kfree(parse_attr->mod_hdr_actions);
1020 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1021 counter = mlx5_fc_create(attr->counter_dev, true);
1022 if (IS_ERR(counter)) {
1023 err = PTR_ERR(counter);
1024 goto err_create_counter;
1027 attr->counter = counter;
1030 /* we get here if one of the following takes place:
1031 * (1) there's no error
1032 * (2) there's an encap action and we don't have valid neigh
1035 /* continue with goto slow path rule instead */
1036 struct mlx5_esw_flow_attr slow_attr;
1038 flow->rule[0] = mlx5e_tc_offload_to_slow_path(esw, flow, &parse_attr->spec, &slow_attr);
1040 flow->rule[0] = mlx5e_tc_offload_fdb_rules(esw, flow, &parse_attr->spec, attr);
1043 if (IS_ERR(flow->rule[0])) {
1044 err = PTR_ERR(flow->rule[0]);
1051 mlx5_fc_destroy(attr->counter_dev, counter);
1053 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1054 mlx5e_detach_mod_hdr(priv, flow);
1056 mlx5_eswitch_del_vlan_action(esw, attr);
1058 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++)
1059 if (attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP)
1060 mlx5e_detach_encap(priv, flow, out_index);
1066 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
1067 struct mlx5e_tc_flow *flow)
1069 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1070 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1071 struct mlx5_esw_flow_attr slow_attr;
1074 if (flow->flags & MLX5E_TC_FLOW_NOT_READY) {
1075 remove_unready_flow(flow);
1076 kvfree(attr->parse_attr);
1080 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
1081 if (flow->flags & MLX5E_TC_FLOW_SLOW)
1082 mlx5e_tc_unoffload_from_slow_path(esw, flow, &slow_attr);
1084 mlx5e_tc_unoffload_fdb_rules(esw, flow, attr);
1087 mlx5_eswitch_del_vlan_action(esw, attr);
1089 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++)
1090 if (attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP)
1091 mlx5e_detach_encap(priv, flow, out_index);
1092 kvfree(attr->parse_attr);
1094 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1095 mlx5e_detach_mod_hdr(priv, flow);
1097 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
1098 mlx5_fc_destroy(attr->counter_dev, attr->counter);
1101 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
1102 struct mlx5e_encap_entry *e)
1104 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1105 struct mlx5_esw_flow_attr slow_attr, *esw_attr;
1106 struct mlx5_flow_handle *rule;
1107 struct mlx5_flow_spec *spec;
1108 struct encap_flow_item *efi;
1109 struct mlx5e_tc_flow *flow;
1112 err = mlx5_packet_reformat_alloc(priv->mdev,
1114 e->encap_size, e->encap_header,
1115 MLX5_FLOW_NAMESPACE_FDB,
1118 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %d\n",
1122 e->flags |= MLX5_ENCAP_ENTRY_VALID;
1123 mlx5e_rep_queue_neigh_stats_work(priv);
1125 list_for_each_entry(efi, &e->flows, list) {
1126 bool all_flow_encaps_valid = true;
1129 flow = container_of(efi, struct mlx5e_tc_flow, encaps[efi->index]);
1130 esw_attr = flow->esw_attr;
1131 spec = &esw_attr->parse_attr->spec;
1133 esw_attr->dests[efi->index].encap_id = e->encap_id;
1134 esw_attr->dests[efi->index].flags |= MLX5_ESW_DEST_ENCAP_VALID;
1135 /* Flow can be associated with multiple encap entries.
1136 * Before offloading the flow verify that all of them have
1137 * a valid neighbour.
1139 for (i = 0; i < MLX5_MAX_FLOW_FWD_VPORTS; i++) {
1140 if (!(esw_attr->dests[i].flags & MLX5_ESW_DEST_ENCAP))
1142 if (!(esw_attr->dests[i].flags & MLX5_ESW_DEST_ENCAP_VALID)) {
1143 all_flow_encaps_valid = false;
1147 /* Do not offload flows with unresolved neighbors */
1148 if (!all_flow_encaps_valid)
1150 /* update from slow path rule to encap rule */
1151 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, esw_attr);
1153 err = PTR_ERR(rule);
1154 mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
1159 mlx5e_tc_unoffload_from_slow_path(esw, flow, &slow_attr);
1160 flow->flags |= MLX5E_TC_FLOW_OFFLOADED; /* was unset when slow path rule removed */
1161 flow->rule[0] = rule;
1165 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
1166 struct mlx5e_encap_entry *e)
1168 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1169 struct mlx5_esw_flow_attr slow_attr;
1170 struct mlx5_flow_handle *rule;
1171 struct mlx5_flow_spec *spec;
1172 struct encap_flow_item *efi;
1173 struct mlx5e_tc_flow *flow;
1176 list_for_each_entry(efi, &e->flows, list) {
1177 flow = container_of(efi, struct mlx5e_tc_flow, encaps[efi->index]);
1178 spec = &flow->esw_attr->parse_attr->spec;
1180 /* update from encap rule to slow path rule */
1181 rule = mlx5e_tc_offload_to_slow_path(esw, flow, spec, &slow_attr);
1182 /* mark the flow's encap dest as non-valid */
1183 flow->esw_attr->dests[efi->index].flags &= ~MLX5_ESW_DEST_ENCAP_VALID;
1186 err = PTR_ERR(rule);
1187 mlx5_core_warn(priv->mdev, "Failed to update slow path (encap) flow, %d\n",
1192 mlx5e_tc_unoffload_fdb_rules(esw, flow, flow->esw_attr);
1193 flow->flags |= MLX5E_TC_FLOW_OFFLOADED; /* was unset when fast path rule removed */
1194 flow->rule[0] = rule;
1197 /* we know that the encap is valid */
1198 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
1199 mlx5_packet_reformat_dealloc(priv->mdev, e->encap_id);
1202 static struct mlx5_fc *mlx5e_tc_get_counter(struct mlx5e_tc_flow *flow)
1204 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1205 return flow->esw_attr->counter;
1207 return flow->nic_attr->counter;
1210 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
1212 struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
1213 u64 bytes, packets, lastuse = 0;
1214 struct mlx5e_tc_flow *flow;
1215 struct mlx5e_encap_entry *e;
1216 struct mlx5_fc *counter;
1217 struct neigh_table *tbl;
1218 bool neigh_used = false;
1219 struct neighbour *n;
1221 if (m_neigh->family == AF_INET)
1223 #if IS_ENABLED(CONFIG_IPV6)
1224 else if (m_neigh->family == AF_INET6)
1230 list_for_each_entry(e, &nhe->encap_list, encap_list) {
1231 struct encap_flow_item *efi;
1232 if (!(e->flags & MLX5_ENCAP_ENTRY_VALID))
1234 list_for_each_entry(efi, &e->flows, list) {
1235 flow = container_of(efi, struct mlx5e_tc_flow,
1236 encaps[efi->index]);
1237 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
1238 counter = mlx5e_tc_get_counter(flow);
1239 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
1240 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
1251 nhe->reported_lastuse = jiffies;
1253 /* find the relevant neigh according to the cached device and
1256 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
1260 neigh_event_send(n, NULL);
1265 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1266 struct mlx5e_tc_flow *flow, int out_index)
1268 struct list_head *next = flow->encaps[out_index].list.next;
1270 list_del(&flow->encaps[out_index].list);
1271 if (list_empty(next)) {
1272 struct mlx5e_encap_entry *e;
1274 e = list_entry(next, struct mlx5e_encap_entry, flows);
1275 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1277 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1278 mlx5_packet_reformat_dealloc(priv->mdev, e->encap_id);
1280 hash_del_rcu(&e->encap_hlist);
1281 kfree(e->encap_header);
1286 static void __mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1288 struct mlx5_eswitch *esw = flow->priv->mdev->priv.eswitch;
1290 if (!(flow->flags & MLX5E_TC_FLOW_ESWITCH) ||
1291 !(flow->flags & MLX5E_TC_FLOW_DUP))
1294 mutex_lock(&esw->offloads.peer_mutex);
1295 list_del(&flow->peer);
1296 mutex_unlock(&esw->offloads.peer_mutex);
1298 flow->flags &= ~MLX5E_TC_FLOW_DUP;
1300 mlx5e_tc_del_fdb_flow(flow->peer_flow->priv, flow->peer_flow);
1301 kvfree(flow->peer_flow);
1302 flow->peer_flow = NULL;
1305 static void mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1307 struct mlx5_core_dev *dev = flow->priv->mdev;
1308 struct mlx5_devcom *devcom = dev->priv.devcom;
1309 struct mlx5_eswitch *peer_esw;
1311 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1315 __mlx5e_tc_del_fdb_peer_flow(flow);
1316 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1319 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1320 struct mlx5e_tc_flow *flow)
1322 if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
1323 mlx5e_tc_del_fdb_peer_flow(flow);
1324 mlx5e_tc_del_fdb_flow(priv, flow);
1326 mlx5e_tc_del_nic_flow(priv, flow);
1331 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1332 struct mlx5_flow_spec *spec,
1333 struct tc_cls_flower_offload *f,
1334 struct net_device *filter_dev, u8 *match_level)
1336 struct netlink_ext_ack *extack = f->common.extack;
1337 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1339 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1341 struct flow_rule *rule = tc_cls_flower_offload_flow_rule(f);
1342 struct flow_match_control enc_control;
1345 err = mlx5e_tc_tun_parse(filter_dev, priv, spec, f,
1346 headers_c, headers_v, match_level);
1348 NL_SET_ERR_MSG_MOD(extack,
1349 "failed to parse tunnel attributes");
1353 flow_rule_match_enc_control(rule, &enc_control);
1355 if (enc_control.key->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1356 struct flow_match_ipv4_addrs match;
1358 flow_rule_match_enc_ipv4_addrs(rule, &match);
1359 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1360 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1361 ntohl(match.mask->src));
1362 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1363 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1364 ntohl(match.key->src));
1366 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1367 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1368 ntohl(match.mask->dst));
1369 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1370 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1371 ntohl(match.key->dst));
1373 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1374 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
1375 } else if (enc_control.key->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1376 struct flow_match_ipv6_addrs match;
1378 flow_rule_match_enc_ipv6_addrs(rule, &match);
1379 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1380 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1381 &match.mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1382 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1383 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1384 &match.key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1386 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1387 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1388 &match.mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1389 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1390 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1391 &match.key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1393 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1394 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
1397 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_IP)) {
1398 struct flow_match_ip match;
1400 flow_rule_match_enc_ip(rule, &match);
1401 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn,
1402 match.mask->tos & 0x3);
1403 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn,
1404 match.key->tos & 0x3);
1406 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp,
1407 match.mask->tos >> 2);
1408 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp,
1409 match.key->tos >> 2);
1411 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit,
1413 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit,
1416 if (match.mask->ttl &&
1417 !MLX5_CAP_ESW_FLOWTABLE_FDB
1419 ft_field_support.outer_ipv4_ttl)) {
1420 NL_SET_ERR_MSG_MOD(extack,
1421 "Matching on TTL is not supported");
1427 /* Enforce DMAC when offloading incoming tunneled flows.
1428 * Flow counters require a match on the DMAC.
1430 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
1431 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
1432 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1433 dmac_47_16), priv->netdev->dev_addr);
1435 /* let software handle IP fragments */
1436 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1437 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
1442 static void *get_match_headers_criteria(u32 flags,
1443 struct mlx5_flow_spec *spec)
1445 return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
1446 MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1448 MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1452 static void *get_match_headers_value(u32 flags,
1453 struct mlx5_flow_spec *spec)
1455 return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
1456 MLX5_ADDR_OF(fte_match_param, spec->match_value,
1458 MLX5_ADDR_OF(fte_match_param, spec->match_value,
1462 static int __parse_cls_flower(struct mlx5e_priv *priv,
1463 struct mlx5_flow_spec *spec,
1464 struct tc_cls_flower_offload *f,
1465 struct net_device *filter_dev,
1466 u8 *match_level, u8 *tunnel_match_level)
1468 struct netlink_ext_ack *extack = f->common.extack;
1469 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1471 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1473 void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1475 void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1477 struct flow_rule *rule = tc_cls_flower_offload_flow_rule(f);
1478 struct flow_dissector *dissector = rule->match.dissector;
1482 *match_level = MLX5_MATCH_NONE;
1484 if (dissector->used_keys &
1485 ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
1486 BIT(FLOW_DISSECTOR_KEY_BASIC) |
1487 BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
1488 BIT(FLOW_DISSECTOR_KEY_VLAN) |
1489 BIT(FLOW_DISSECTOR_KEY_CVLAN) |
1490 BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
1491 BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
1492 BIT(FLOW_DISSECTOR_KEY_PORTS) |
1493 BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
1494 BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
1495 BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
1496 BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
1497 BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
1498 BIT(FLOW_DISSECTOR_KEY_TCP) |
1499 BIT(FLOW_DISSECTOR_KEY_IP) |
1500 BIT(FLOW_DISSECTOR_KEY_ENC_IP))) {
1501 NL_SET_ERR_MSG_MOD(extack, "Unsupported key");
1502 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
1503 dissector->used_keys);
1507 if ((flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
1508 flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
1509 flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
1510 flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
1511 struct flow_match_control match;
1513 flow_rule_match_enc_control(rule, &match);
1514 switch (match.key->addr_type) {
1515 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
1516 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
1517 if (parse_tunnel_attr(priv, spec, f, filter_dev, tunnel_match_level))
1524 /* In decap flow, header pointers should point to the inner
1525 * headers, outer header were already set by parse_tunnel_attr
1527 headers_c = get_match_headers_criteria(MLX5_FLOW_CONTEXT_ACTION_DECAP,
1529 headers_v = get_match_headers_value(MLX5_FLOW_CONTEXT_ACTION_DECAP,
1533 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
1534 struct flow_match_basic match;
1536 flow_rule_match_basic(rule, &match);
1537 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1538 ntohs(match.mask->n_proto));
1539 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1540 ntohs(match.key->n_proto));
1542 if (match.mask->n_proto)
1543 *match_level = MLX5_MATCH_L2;
1545 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN) ||
1546 is_vlan_dev(filter_dev)) {
1547 struct flow_dissector_key_vlan filter_dev_mask;
1548 struct flow_dissector_key_vlan filter_dev_key;
1549 struct flow_match_vlan match;
1551 if (is_vlan_dev(filter_dev)) {
1552 match.key = &filter_dev_key;
1553 match.key->vlan_id = vlan_dev_vlan_id(filter_dev);
1554 match.key->vlan_tpid = vlan_dev_vlan_proto(filter_dev);
1555 match.key->vlan_priority = 0;
1556 match.mask = &filter_dev_mask;
1557 memset(match.mask, 0xff, sizeof(*match.mask));
1558 match.mask->vlan_priority = 0;
1560 flow_rule_match_vlan(rule, &match);
1562 if (match.mask->vlan_id ||
1563 match.mask->vlan_priority ||
1564 match.mask->vlan_tpid) {
1565 if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
1566 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1568 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1571 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1573 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1577 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid,
1578 match.mask->vlan_id);
1579 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid,
1580 match.key->vlan_id);
1582 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio,
1583 match.mask->vlan_priority);
1584 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio,
1585 match.key->vlan_priority);
1587 *match_level = MLX5_MATCH_L2;
1589 } else if (*match_level != MLX5_MATCH_NONE) {
1590 MLX5_SET(fte_match_set_lyr_2_4, headers_c, svlan_tag, 1);
1591 MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
1592 *match_level = MLX5_MATCH_L2;
1595 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
1596 struct flow_match_vlan match;
1598 flow_rule_match_cvlan(rule, &match);
1599 if (match.mask->vlan_id ||
1600 match.mask->vlan_priority ||
1601 match.mask->vlan_tpid) {
1602 if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
1603 MLX5_SET(fte_match_set_misc, misc_c,
1604 outer_second_svlan_tag, 1);
1605 MLX5_SET(fte_match_set_misc, misc_v,
1606 outer_second_svlan_tag, 1);
1608 MLX5_SET(fte_match_set_misc, misc_c,
1609 outer_second_cvlan_tag, 1);
1610 MLX5_SET(fte_match_set_misc, misc_v,
1611 outer_second_cvlan_tag, 1);
1614 MLX5_SET(fte_match_set_misc, misc_c, outer_second_vid,
1615 match.mask->vlan_id);
1616 MLX5_SET(fte_match_set_misc, misc_v, outer_second_vid,
1617 match.key->vlan_id);
1618 MLX5_SET(fte_match_set_misc, misc_c, outer_second_prio,
1619 match.mask->vlan_priority);
1620 MLX5_SET(fte_match_set_misc, misc_v, outer_second_prio,
1621 match.key->vlan_priority);
1623 *match_level = MLX5_MATCH_L2;
1627 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
1628 struct flow_match_eth_addrs match;
1630 flow_rule_match_eth_addrs(rule, &match);
1631 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1634 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1638 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1641 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1645 if (!is_zero_ether_addr(match.mask->src) ||
1646 !is_zero_ether_addr(match.mask->dst))
1647 *match_level = MLX5_MATCH_L2;
1650 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
1651 struct flow_match_control match;
1653 flow_rule_match_control(rule, &match);
1654 addr_type = match.key->addr_type;
1656 /* the HW doesn't support frag first/later */
1657 if (match.mask->flags & FLOW_DIS_FIRST_FRAG)
1660 if (match.mask->flags & FLOW_DIS_IS_FRAGMENT) {
1661 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1662 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
1663 match.key->flags & FLOW_DIS_IS_FRAGMENT);
1665 /* the HW doesn't need L3 inline to match on frag=no */
1666 if (!(match.key->flags & FLOW_DIS_IS_FRAGMENT))
1667 *match_level = MLX5_MATCH_L2;
1668 /* *** L2 attributes parsing up to here *** */
1670 *match_level = MLX5_MATCH_L3;
1674 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
1675 struct flow_match_basic match;
1677 flow_rule_match_basic(rule, &match);
1678 ip_proto = match.key->ip_proto;
1680 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
1681 match.mask->ip_proto);
1682 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
1683 match.key->ip_proto);
1685 if (match.mask->ip_proto)
1686 *match_level = MLX5_MATCH_L3;
1689 if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1690 struct flow_match_ipv4_addrs match;
1692 flow_rule_match_ipv4_addrs(rule, &match);
1693 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1694 src_ipv4_src_ipv6.ipv4_layout.ipv4),
1695 &match.mask->src, sizeof(match.mask->src));
1696 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1697 src_ipv4_src_ipv6.ipv4_layout.ipv4),
1698 &match.key->src, sizeof(match.key->src));
1699 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1700 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1701 &match.mask->dst, sizeof(match.mask->dst));
1702 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1703 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1704 &match.key->dst, sizeof(match.key->dst));
1706 if (match.mask->src || match.mask->dst)
1707 *match_level = MLX5_MATCH_L3;
1710 if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1711 struct flow_match_ipv6_addrs match;
1713 flow_rule_match_ipv6_addrs(rule, &match);
1714 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1715 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1716 &match.mask->src, sizeof(match.mask->src));
1717 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1718 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1719 &match.key->src, sizeof(match.key->src));
1721 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1722 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1723 &match.mask->dst, sizeof(match.mask->dst));
1724 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1725 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1726 &match.key->dst, sizeof(match.key->dst));
1728 if (ipv6_addr_type(&match.mask->src) != IPV6_ADDR_ANY ||
1729 ipv6_addr_type(&match.mask->dst) != IPV6_ADDR_ANY)
1730 *match_level = MLX5_MATCH_L3;
1733 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
1734 struct flow_match_ip match;
1736 flow_rule_match_ip(rule, &match);
1737 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn,
1738 match.mask->tos & 0x3);
1739 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn,
1740 match.key->tos & 0x3);
1742 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp,
1743 match.mask->tos >> 2);
1744 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp,
1745 match.key->tos >> 2);
1747 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit,
1749 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit,
1752 if (match.mask->ttl &&
1753 !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
1754 ft_field_support.outer_ipv4_ttl)) {
1755 NL_SET_ERR_MSG_MOD(extack,
1756 "Matching on TTL is not supported");
1760 if (match.mask->tos || match.mask->ttl)
1761 *match_level = MLX5_MATCH_L3;
1764 /* *** L3 attributes parsing up to here *** */
1766 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) {
1767 struct flow_match_ports match;
1769 flow_rule_match_ports(rule, &match);
1772 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1773 tcp_sport, ntohs(match.mask->src));
1774 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1775 tcp_sport, ntohs(match.key->src));
1777 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1778 tcp_dport, ntohs(match.mask->dst));
1779 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1780 tcp_dport, ntohs(match.key->dst));
1784 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1785 udp_sport, ntohs(match.mask->src));
1786 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1787 udp_sport, ntohs(match.key->src));
1789 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1790 udp_dport, ntohs(match.mask->dst));
1791 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1792 udp_dport, ntohs(match.key->dst));
1795 NL_SET_ERR_MSG_MOD(extack,
1796 "Only UDP and TCP transports are supported for L4 matching");
1797 netdev_err(priv->netdev,
1798 "Only UDP and TCP transport are supported\n");
1802 if (match.mask->src || match.mask->dst)
1803 *match_level = MLX5_MATCH_L4;
1806 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_TCP)) {
1807 struct flow_match_tcp match;
1809 flow_rule_match_tcp(rule, &match);
1810 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
1811 ntohs(match.mask->flags));
1812 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
1813 ntohs(match.key->flags));
1815 if (match.mask->flags)
1816 *match_level = MLX5_MATCH_L4;
1822 static int parse_cls_flower(struct mlx5e_priv *priv,
1823 struct mlx5e_tc_flow *flow,
1824 struct mlx5_flow_spec *spec,
1825 struct tc_cls_flower_offload *f,
1826 struct net_device *filter_dev)
1828 struct netlink_ext_ack *extack = f->common.extack;
1829 struct mlx5_core_dev *dev = priv->mdev;
1830 struct mlx5_eswitch *esw = dev->priv.eswitch;
1831 struct mlx5e_rep_priv *rpriv = priv->ppriv;
1832 u8 match_level, tunnel_match_level = MLX5_MATCH_NONE;
1833 struct mlx5_eswitch_rep *rep;
1836 err = __parse_cls_flower(priv, spec, f, filter_dev, &match_level, &tunnel_match_level);
1838 if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH)) {
1840 if (rep->vport != MLX5_VPORT_UPLINK &&
1841 (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
1842 esw->offloads.inline_mode < match_level)) {
1843 NL_SET_ERR_MSG_MOD(extack,
1844 "Flow is not offloaded due to min inline setting");
1845 netdev_warn(priv->netdev,
1846 "Flow is not offloaded due to min inline setting, required %d actual %d\n",
1847 match_level, esw->offloads.inline_mode);
1852 if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
1853 flow->esw_attr->match_level = match_level;
1854 flow->esw_attr->tunnel_match_level = tunnel_match_level;
1856 flow->nic_attr->match_level = match_level;
1862 struct pedit_headers {
1864 struct vlan_hdr vlan;
1871 struct pedit_headers_action {
1872 struct pedit_headers vals;
1873 struct pedit_headers masks;
1877 static int pedit_header_offsets[] = {
1878 [FLOW_ACT_MANGLE_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
1879 [FLOW_ACT_MANGLE_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
1880 [FLOW_ACT_MANGLE_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
1881 [FLOW_ACT_MANGLE_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
1882 [FLOW_ACT_MANGLE_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
1885 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
1887 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
1888 struct pedit_headers_action *hdrs)
1890 u32 *curr_pmask, *curr_pval;
1892 curr_pmask = (u32 *)(pedit_header(&hdrs->masks, hdr_type) + offset);
1893 curr_pval = (u32 *)(pedit_header(&hdrs->vals, hdr_type) + offset);
1895 if (*curr_pmask & mask) /* disallow acting twice on the same location */
1898 *curr_pmask |= mask;
1899 *curr_pval |= (val & mask);
1907 struct mlx5_fields {
1914 #define OFFLOAD(fw_field, size, field, off, match_field) \
1915 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, size, \
1916 offsetof(struct pedit_headers, field) + (off), \
1917 MLX5_BYTE_OFF(fte_match_set_lyr_2_4, match_field)}
1919 /* masked values are the same and there are no rewrites that do not have a
1922 #define SAME_VAL_MASK(type, valp, maskp, matchvalp, matchmaskp) ({ \
1923 type matchmaskx = *(type *)(matchmaskp); \
1924 type matchvalx = *(type *)(matchvalp); \
1925 type maskx = *(type *)(maskp); \
1926 type valx = *(type *)(valp); \
1928 (valx & maskx) == (matchvalx & matchmaskx) && !(maskx & (maskx ^ \
1932 static bool cmp_val_mask(void *valp, void *maskp, void *matchvalp,
1933 void *matchmaskp, int size)
1939 same = SAME_VAL_MASK(u8, valp, maskp, matchvalp, matchmaskp);
1942 same = SAME_VAL_MASK(u16, valp, maskp, matchvalp, matchmaskp);
1945 same = SAME_VAL_MASK(u32, valp, maskp, matchvalp, matchmaskp);
1952 static struct mlx5_fields fields[] = {
1953 OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0, dmac_47_16),
1954 OFFLOAD(DMAC_15_0, 2, eth.h_dest[4], 0, dmac_15_0),
1955 OFFLOAD(SMAC_47_16, 4, eth.h_source[0], 0, smac_47_16),
1956 OFFLOAD(SMAC_15_0, 2, eth.h_source[4], 0, smac_15_0),
1957 OFFLOAD(ETHERTYPE, 2, eth.h_proto, 0, ethertype),
1958 OFFLOAD(FIRST_VID, 2, vlan.h_vlan_TCI, 0, first_vid),
1960 OFFLOAD(IP_TTL, 1, ip4.ttl, 0, ttl_hoplimit),
1961 OFFLOAD(SIPV4, 4, ip4.saddr, 0, src_ipv4_src_ipv6.ipv4_layout.ipv4),
1962 OFFLOAD(DIPV4, 4, ip4.daddr, 0, dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1964 OFFLOAD(SIPV6_127_96, 4, ip6.saddr.s6_addr32[0], 0,
1965 src_ipv4_src_ipv6.ipv6_layout.ipv6[0]),
1966 OFFLOAD(SIPV6_95_64, 4, ip6.saddr.s6_addr32[1], 0,
1967 src_ipv4_src_ipv6.ipv6_layout.ipv6[4]),
1968 OFFLOAD(SIPV6_63_32, 4, ip6.saddr.s6_addr32[2], 0,
1969 src_ipv4_src_ipv6.ipv6_layout.ipv6[8]),
1970 OFFLOAD(SIPV6_31_0, 4, ip6.saddr.s6_addr32[3], 0,
1971 src_ipv4_src_ipv6.ipv6_layout.ipv6[12]),
1972 OFFLOAD(DIPV6_127_96, 4, ip6.daddr.s6_addr32[0], 0,
1973 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[0]),
1974 OFFLOAD(DIPV6_95_64, 4, ip6.daddr.s6_addr32[1], 0,
1975 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[4]),
1976 OFFLOAD(DIPV6_63_32, 4, ip6.daddr.s6_addr32[2], 0,
1977 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[8]),
1978 OFFLOAD(DIPV6_31_0, 4, ip6.daddr.s6_addr32[3], 0,
1979 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[12]),
1980 OFFLOAD(IPV6_HOPLIMIT, 1, ip6.hop_limit, 0, ttl_hoplimit),
1982 OFFLOAD(TCP_SPORT, 2, tcp.source, 0, tcp_sport),
1983 OFFLOAD(TCP_DPORT, 2, tcp.dest, 0, tcp_dport),
1984 OFFLOAD(TCP_FLAGS, 1, tcp.ack_seq, 5, tcp_flags),
1986 OFFLOAD(UDP_SPORT, 2, udp.source, 0, udp_sport),
1987 OFFLOAD(UDP_DPORT, 2, udp.dest, 0, udp_dport),
1990 /* On input attr->max_mod_hdr_actions tells how many HW actions can be parsed at
1991 * max from the SW pedit action. On success, attr->num_mod_hdr_actions
1992 * says how many HW actions were actually parsed.
1994 static int offload_pedit_fields(struct pedit_headers_action *hdrs,
1995 struct mlx5e_tc_flow_parse_attr *parse_attr,
1997 struct netlink_ext_ack *extack)
1999 struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
2000 void *headers_c = get_match_headers_criteria(*action_flags,
2002 void *headers_v = get_match_headers_value(*action_flags,
2004 int i, action_size, nactions, max_actions, first, last, next_z;
2005 void *s_masks_p, *a_masks_p, *vals_p;
2006 struct mlx5_fields *f;
2007 u8 cmd, field_bsize;
2014 set_masks = &hdrs[0].masks;
2015 add_masks = &hdrs[1].masks;
2016 set_vals = &hdrs[0].vals;
2017 add_vals = &hdrs[1].vals;
2019 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2020 action = parse_attr->mod_hdr_actions +
2021 parse_attr->num_mod_hdr_actions * action_size;
2023 max_actions = parse_attr->max_mod_hdr_actions;
2024 nactions = parse_attr->num_mod_hdr_actions;
2026 for (i = 0; i < ARRAY_SIZE(fields); i++) {
2030 /* avoid seeing bits set from previous iterations */
2034 s_masks_p = (void *)set_masks + f->offset;
2035 a_masks_p = (void *)add_masks + f->offset;
2037 memcpy(&s_mask, s_masks_p, f->size);
2038 memcpy(&a_mask, a_masks_p, f->size);
2040 if (!s_mask && !a_mask) /* nothing to offload here */
2043 if (s_mask && a_mask) {
2044 NL_SET_ERR_MSG_MOD(extack,
2045 "can't set and add to the same HW field");
2046 printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
2050 if (nactions == max_actions) {
2051 NL_SET_ERR_MSG_MOD(extack,
2052 "too many pedit actions, can't offload");
2053 printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
2059 void *match_mask = headers_c + f->match_offset;
2060 void *match_val = headers_v + f->match_offset;
2062 cmd = MLX5_ACTION_TYPE_SET;
2064 vals_p = (void *)set_vals + f->offset;
2065 /* don't rewrite if we have a match on the same value */
2066 if (cmp_val_mask(vals_p, s_masks_p, match_val,
2067 match_mask, f->size))
2069 /* clear to denote we consumed this field */
2070 memset(s_masks_p, 0, f->size);
2074 cmd = MLX5_ACTION_TYPE_ADD;
2076 vals_p = (void *)add_vals + f->offset;
2077 /* add 0 is no change */
2078 if (!memcmp(vals_p, &zero, f->size))
2080 /* clear to denote we consumed this field */
2081 memset(a_masks_p, 0, f->size);
2086 field_bsize = f->size * BITS_PER_BYTE;
2088 if (field_bsize == 32) {
2089 mask_be32 = *(__be32 *)&mask;
2090 mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
2091 } else if (field_bsize == 16) {
2092 mask_be16 = *(__be16 *)&mask;
2093 mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
2096 first = find_first_bit(&mask, field_bsize);
2097 next_z = find_next_zero_bit(&mask, field_bsize, first);
2098 last = find_last_bit(&mask, field_bsize);
2099 if (first < next_z && next_z < last) {
2100 NL_SET_ERR_MSG_MOD(extack,
2101 "rewrite of few sub-fields isn't supported");
2102 printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
2107 MLX5_SET(set_action_in, action, action_type, cmd);
2108 MLX5_SET(set_action_in, action, field, f->field);
2110 if (cmd == MLX5_ACTION_TYPE_SET) {
2111 MLX5_SET(set_action_in, action, offset, first);
2112 /* length is num of bits to be written, zero means length of 32 */
2113 MLX5_SET(set_action_in, action, length, (last - first + 1));
2116 if (field_bsize == 32)
2117 MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
2118 else if (field_bsize == 16)
2119 MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
2120 else if (field_bsize == 8)
2121 MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
2123 action += action_size;
2127 parse_attr->num_mod_hdr_actions = nactions;
2131 static int mlx5e_flow_namespace_max_modify_action(struct mlx5_core_dev *mdev,
2134 if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
2135 return MLX5_CAP_ESW_FLOWTABLE_FDB(mdev, max_modify_header_actions);
2136 else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
2137 return MLX5_CAP_FLOWTABLE_NIC_RX(mdev, max_modify_header_actions);
2140 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
2141 struct pedit_headers_action *hdrs,
2143 struct mlx5e_tc_flow_parse_attr *parse_attr)
2145 int nkeys, action_size, max_actions;
2147 nkeys = hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits +
2148 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits;
2149 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2151 max_actions = mlx5e_flow_namespace_max_modify_action(priv->mdev, namespace);
2152 /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
2153 max_actions = min(max_actions, nkeys * 16);
2155 parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
2156 if (!parse_attr->mod_hdr_actions)
2159 parse_attr->max_mod_hdr_actions = max_actions;
2163 static const struct pedit_headers zero_masks = {};
2165 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
2166 const struct flow_action_entry *act, int namespace,
2167 struct mlx5e_tc_flow_parse_attr *parse_attr,
2168 struct pedit_headers_action *hdrs,
2169 struct netlink_ext_ack *extack)
2171 u8 cmd = (act->id == FLOW_ACTION_MANGLE) ? 0 : 1;
2172 int err = -EOPNOTSUPP;
2173 u32 mask, val, offset;
2176 htype = act->mangle.htype;
2177 err = -EOPNOTSUPP; /* can't be all optimistic */
2179 if (htype == FLOW_ACT_MANGLE_UNSPEC) {
2180 NL_SET_ERR_MSG_MOD(extack, "legacy pedit isn't offloaded");
2184 if (!mlx5e_flow_namespace_max_modify_action(priv->mdev, namespace)) {
2185 NL_SET_ERR_MSG_MOD(extack,
2186 "The pedit offload action is not supported");
2190 mask = act->mangle.mask;
2191 val = act->mangle.val;
2192 offset = act->mangle.offset;
2194 err = set_pedit_val(htype, ~mask, val, offset, &hdrs[cmd]);
2205 static int alloc_tc_pedit_action(struct mlx5e_priv *priv, int namespace,
2206 struct mlx5e_tc_flow_parse_attr *parse_attr,
2207 struct pedit_headers_action *hdrs,
2209 struct netlink_ext_ack *extack)
2211 struct pedit_headers *cmd_masks;
2215 if (!parse_attr->mod_hdr_actions) {
2216 err = alloc_mod_hdr_actions(priv, hdrs, namespace, parse_attr);
2221 err = offload_pedit_fields(hdrs, parse_attr, action_flags, extack);
2223 goto out_dealloc_parsed_actions;
2225 for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
2226 cmd_masks = &hdrs[cmd].masks;
2227 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
2228 NL_SET_ERR_MSG_MOD(extack,
2229 "attempt to offload an unsupported field");
2230 netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
2231 print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
2232 16, 1, cmd_masks, sizeof(zero_masks), true);
2234 goto out_dealloc_parsed_actions;
2240 out_dealloc_parsed_actions:
2241 kfree(parse_attr->mod_hdr_actions);
2246 static bool csum_offload_supported(struct mlx5e_priv *priv,
2249 struct netlink_ext_ack *extack)
2251 u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
2252 TCA_CSUM_UPDATE_FLAG_UDP;
2254 /* The HW recalcs checksums only if re-writing headers */
2255 if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
2256 NL_SET_ERR_MSG_MOD(extack,
2257 "TC csum action is only offloaded with pedit");
2258 netdev_warn(priv->netdev,
2259 "TC csum action is only offloaded with pedit\n");
2263 if (update_flags & ~prot_flags) {
2264 NL_SET_ERR_MSG_MOD(extack,
2265 "can't offload TC csum action for some header/s");
2266 netdev_warn(priv->netdev,
2267 "can't offload TC csum action for some header/s - flags %#x\n",
2275 struct ip_ttl_word {
2281 struct ipv6_hoplimit_word {
2287 static bool is_action_keys_supported(const struct flow_action_entry *act)
2292 htype = act->mangle.htype;
2293 offset = act->mangle.offset;
2294 mask = ~act->mangle.mask;
2295 /* For IPv4 & IPv6 header check 4 byte word,
2296 * to determine that modified fields
2297 * are NOT ttl & hop_limit only.
2299 if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP4) {
2300 struct ip_ttl_word *ttl_word =
2301 (struct ip_ttl_word *)&mask;
2303 if (offset != offsetof(struct iphdr, ttl) ||
2304 ttl_word->protocol ||
2308 } else if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP6) {
2309 struct ipv6_hoplimit_word *hoplimit_word =
2310 (struct ipv6_hoplimit_word *)&mask;
2312 if (offset != offsetof(struct ipv6hdr, payload_len) ||
2313 hoplimit_word->payload_len ||
2314 hoplimit_word->nexthdr) {
2321 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
2322 struct flow_action *flow_action,
2324 struct netlink_ext_ack *extack)
2326 const struct flow_action_entry *act;
2327 bool modify_ip_header;
2333 headers_v = get_match_headers_value(actions, spec);
2334 ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
2336 /* for non-IP we only re-write MACs, so we're okay */
2337 if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
2340 modify_ip_header = false;
2341 flow_action_for_each(i, act, flow_action) {
2342 if (act->id != FLOW_ACTION_MANGLE &&
2343 act->id != FLOW_ACTION_ADD)
2346 if (is_action_keys_supported(act)) {
2347 modify_ip_header = true;
2352 ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
2353 if (modify_ip_header && ip_proto != IPPROTO_TCP &&
2354 ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
2355 NL_SET_ERR_MSG_MOD(extack,
2356 "can't offload re-write of non TCP/UDP");
2357 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
2365 static bool actions_match_supported(struct mlx5e_priv *priv,
2366 struct flow_action *flow_action,
2367 struct mlx5e_tc_flow_parse_attr *parse_attr,
2368 struct mlx5e_tc_flow *flow,
2369 struct netlink_ext_ack *extack)
2373 if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
2374 actions = flow->esw_attr->action;
2376 actions = flow->nic_attr->action;
2378 if (flow->flags & MLX5E_TC_FLOW_EGRESS &&
2379 !((actions & MLX5_FLOW_CONTEXT_ACTION_DECAP) ||
2380 (actions & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP)))
2383 if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
2384 return modify_header_match_supported(&parse_attr->spec,
2385 flow_action, actions,
2391 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
2393 struct mlx5_core_dev *fmdev, *pmdev;
2394 u64 fsystem_guid, psystem_guid;
2397 pmdev = peer_priv->mdev;
2399 fsystem_guid = mlx5_query_nic_system_image_guid(fmdev);
2400 psystem_guid = mlx5_query_nic_system_image_guid(pmdev);
2402 return (fsystem_guid == psystem_guid);
2405 static int add_vlan_rewrite_action(struct mlx5e_priv *priv, int namespace,
2406 const struct flow_action_entry *act,
2407 struct mlx5e_tc_flow_parse_attr *parse_attr,
2408 struct pedit_headers_action *hdrs,
2409 u32 *action, struct netlink_ext_ack *extack)
2411 u16 mask16 = VLAN_VID_MASK;
2412 u16 val16 = act->vlan.vid & VLAN_VID_MASK;
2413 const struct flow_action_entry pedit_act = {
2414 .id = FLOW_ACTION_MANGLE,
2415 .mangle.htype = FLOW_ACT_MANGLE_HDR_TYPE_ETH,
2416 .mangle.offset = offsetof(struct vlan_ethhdr, h_vlan_TCI),
2417 .mangle.mask = ~(u32)be16_to_cpu(*(__be16 *)&mask16),
2418 .mangle.val = (u32)be16_to_cpu(*(__be16 *)&val16),
2420 u8 match_prio_mask, match_prio_val;
2421 void *headers_c, *headers_v;
2424 headers_c = get_match_headers_criteria(*action, &parse_attr->spec);
2425 headers_v = get_match_headers_value(*action, &parse_attr->spec);
2427 if (!(MLX5_GET(fte_match_set_lyr_2_4, headers_c, cvlan_tag) &&
2428 MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag))) {
2429 NL_SET_ERR_MSG_MOD(extack,
2430 "VLAN rewrite action must have VLAN protocol match");
2434 match_prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
2435 match_prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
2436 if (act->vlan.prio != (match_prio_val & match_prio_mask)) {
2437 NL_SET_ERR_MSG_MOD(extack,
2438 "Changing VLAN prio is not supported");
2442 err = parse_tc_pedit_action(priv, &pedit_act, namespace, parse_attr,
2444 *action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2450 add_vlan_prio_tag_rewrite_action(struct mlx5e_priv *priv,
2451 struct mlx5e_tc_flow_parse_attr *parse_attr,
2452 struct pedit_headers_action *hdrs,
2453 u32 *action, struct netlink_ext_ack *extack)
2455 const struct flow_action_entry prio_tag_act = {
2458 MLX5_GET(fte_match_set_lyr_2_4,
2459 get_match_headers_value(*action,
2462 MLX5_GET(fte_match_set_lyr_2_4,
2463 get_match_headers_criteria(*action,
2468 return add_vlan_rewrite_action(priv, MLX5_FLOW_NAMESPACE_FDB,
2469 &prio_tag_act, parse_attr, hdrs, action,
2473 static int parse_tc_nic_actions(struct mlx5e_priv *priv,
2474 struct flow_action *flow_action,
2475 struct mlx5e_tc_flow_parse_attr *parse_attr,
2476 struct mlx5e_tc_flow *flow,
2477 struct netlink_ext_ack *extack)
2479 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
2480 struct pedit_headers_action hdrs[2] = {};
2481 const struct flow_action_entry *act;
2485 if (!flow_action_has_entries(flow_action))
2488 attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
2490 flow_action_for_each(i, act, flow_action) {
2492 case FLOW_ACTION_DROP:
2493 action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
2494 if (MLX5_CAP_FLOWTABLE(priv->mdev,
2495 flow_table_properties_nic_receive.flow_counter))
2496 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
2498 case FLOW_ACTION_MANGLE:
2499 case FLOW_ACTION_ADD:
2500 err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_KERNEL,
2501 parse_attr, hdrs, extack);
2505 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
2506 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2508 case FLOW_ACTION_VLAN_MANGLE:
2509 err = add_vlan_rewrite_action(priv,
2510 MLX5_FLOW_NAMESPACE_KERNEL,
2511 act, parse_attr, hdrs,
2517 case FLOW_ACTION_CSUM:
2518 if (csum_offload_supported(priv, action,
2524 case FLOW_ACTION_REDIRECT: {
2525 struct net_device *peer_dev = act->dev;
2527 if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
2528 same_hw_devs(priv, netdev_priv(peer_dev))) {
2529 parse_attr->mirred_ifindex[0] = peer_dev->ifindex;
2530 flow->flags |= MLX5E_TC_FLOW_HAIRPIN;
2531 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2532 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2534 NL_SET_ERR_MSG_MOD(extack,
2535 "device is not on same HW, can't offload");
2536 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
2542 case FLOW_ACTION_MARK: {
2543 u32 mark = act->mark;
2545 if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
2546 NL_SET_ERR_MSG_MOD(extack,
2547 "Bad flow mark - only 16 bit is supported");
2551 attr->flow_tag = mark;
2552 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2556 NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
2561 if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
2562 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
2563 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_KERNEL,
2564 parse_attr, hdrs, &action, extack);
2567 /* in case all pedit actions are skipped, remove the MOD_HDR
2570 if (parse_attr->num_mod_hdr_actions == 0) {
2571 action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2572 kfree(parse_attr->mod_hdr_actions);
2576 attr->action = action;
2577 if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
2584 struct ip_tunnel_key *ip_tun_key;
2588 static inline int cmp_encap_info(struct encap_key *a,
2589 struct encap_key *b)
2591 return memcmp(a->ip_tun_key, b->ip_tun_key, sizeof(*a->ip_tun_key)) ||
2592 a->tunnel_type != b->tunnel_type;
2595 static inline int hash_encap_info(struct encap_key *key)
2597 return jhash(key->ip_tun_key, sizeof(*key->ip_tun_key),
2602 static bool is_merged_eswitch_dev(struct mlx5e_priv *priv,
2603 struct net_device *peer_netdev)
2605 struct mlx5e_priv *peer_priv;
2607 peer_priv = netdev_priv(peer_netdev);
2609 return (MLX5_CAP_ESW(priv->mdev, merged_eswitch) &&
2610 mlx5e_eswitch_rep(priv->netdev) &&
2611 mlx5e_eswitch_rep(peer_netdev) &&
2612 same_hw_devs(priv, peer_priv));
2617 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
2618 struct mlx5e_tc_flow *flow,
2619 struct net_device *mirred_dev,
2621 struct netlink_ext_ack *extack,
2622 struct net_device **encap_dev,
2625 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2626 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2627 struct mlx5e_tc_flow_parse_attr *parse_attr;
2628 struct ip_tunnel_info *tun_info;
2629 struct encap_key key, e_key;
2630 struct mlx5e_encap_entry *e;
2631 unsigned short family;
2636 parse_attr = attr->parse_attr;
2637 tun_info = &parse_attr->tun_info[out_index];
2638 family = ip_tunnel_info_af(tun_info);
2639 key.ip_tun_key = &tun_info->key;
2640 key.tunnel_type = mlx5e_tc_tun_get_type(mirred_dev);
2642 hash_key = hash_encap_info(&key);
2644 hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
2645 encap_hlist, hash_key) {
2646 e_key.ip_tun_key = &e->tun_info.key;
2647 e_key.tunnel_type = e->tunnel_type;
2648 if (!cmp_encap_info(&e_key, &key)) {
2654 /* must verify if encap is valid or not */
2658 e = kzalloc(sizeof(*e), GFP_KERNEL);
2662 e->tun_info = *tun_info;
2663 err = mlx5e_tc_tun_init_encap_attr(mirred_dev, priv, e, extack);
2667 INIT_LIST_HEAD(&e->flows);
2669 if (family == AF_INET)
2670 err = mlx5e_tc_tun_create_header_ipv4(priv, mirred_dev, e);
2671 else if (family == AF_INET6)
2672 err = mlx5e_tc_tun_create_header_ipv6(priv, mirred_dev, e);
2677 hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
2680 list_add(&flow->encaps[out_index].list, &e->flows);
2681 flow->encaps[out_index].index = out_index;
2682 *encap_dev = e->out_dev;
2683 if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
2684 attr->dests[out_index].encap_id = e->encap_id;
2685 attr->dests[out_index].flags |= MLX5_ESW_DEST_ENCAP_VALID;
2686 *encap_valid = true;
2688 *encap_valid = false;
2698 static int parse_tc_vlan_action(struct mlx5e_priv *priv,
2699 const struct flow_action_entry *act,
2700 struct mlx5_esw_flow_attr *attr,
2703 u8 vlan_idx = attr->total_vlan;
2705 if (vlan_idx >= MLX5_FS_VLAN_DEPTH)
2709 case FLOW_ACTION_VLAN_POP:
2711 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
2712 MLX5_FS_VLAN_DEPTH))
2715 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2;
2717 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2720 case FLOW_ACTION_VLAN_PUSH:
2721 attr->vlan_vid[vlan_idx] = act->vlan.vid;
2722 attr->vlan_prio[vlan_idx] = act->vlan.prio;
2723 attr->vlan_proto[vlan_idx] = act->vlan.proto;
2724 if (!attr->vlan_proto[vlan_idx])
2725 attr->vlan_proto[vlan_idx] = htons(ETH_P_8021Q);
2728 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
2729 MLX5_FS_VLAN_DEPTH))
2732 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2;
2734 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev, 1) &&
2735 (act->vlan.proto != htons(ETH_P_8021Q) ||
2739 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
2746 attr->total_vlan = vlan_idx + 1;
2751 static int add_vlan_push_action(struct mlx5e_priv *priv,
2752 struct mlx5_esw_flow_attr *attr,
2753 struct net_device **out_dev,
2756 struct net_device *vlan_dev = *out_dev;
2757 struct flow_action_entry vlan_act = {
2758 .id = FLOW_ACTION_VLAN_PUSH,
2759 .vlan.vid = vlan_dev_vlan_id(vlan_dev),
2760 .vlan.proto = vlan_dev_vlan_proto(vlan_dev),
2765 err = parse_tc_vlan_action(priv, &vlan_act, attr, action);
2769 *out_dev = dev_get_by_index_rcu(dev_net(vlan_dev),
2770 dev_get_iflink(vlan_dev));
2771 if (is_vlan_dev(*out_dev))
2772 err = add_vlan_push_action(priv, attr, out_dev, action);
2777 static int add_vlan_pop_action(struct mlx5e_priv *priv,
2778 struct mlx5_esw_flow_attr *attr,
2781 int nest_level = vlan_get_encap_level(attr->parse_attr->filter_dev);
2782 struct flow_action_entry vlan_act = {
2783 .id = FLOW_ACTION_VLAN_POP,
2787 while (nest_level--) {
2788 err = parse_tc_vlan_action(priv, &vlan_act, attr, action);
2796 static int parse_tc_fdb_actions(struct mlx5e_priv *priv,
2797 struct flow_action *flow_action,
2798 struct mlx5e_tc_flow *flow,
2799 struct netlink_ext_ack *extack)
2801 struct pedit_headers_action hdrs[2] = {};
2802 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2803 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2804 struct mlx5e_tc_flow_parse_attr *parse_attr = attr->parse_attr;
2805 struct mlx5e_rep_priv *rpriv = priv->ppriv;
2806 const struct ip_tunnel_info *info = NULL;
2807 const struct flow_action_entry *act;
2812 if (!flow_action_has_entries(flow_action))
2815 attr->in_rep = rpriv->rep;
2816 attr->in_mdev = priv->mdev;
2818 flow_action_for_each(i, act, flow_action) {
2820 case FLOW_ACTION_DROP:
2821 action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
2822 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2824 case FLOW_ACTION_MANGLE:
2825 case FLOW_ACTION_ADD:
2826 err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_FDB,
2827 parse_attr, hdrs, extack);
2831 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2832 attr->split_count = attr->out_count;
2834 case FLOW_ACTION_CSUM:
2835 if (csum_offload_supported(priv, action,
2836 act->csum_flags, extack))
2840 case FLOW_ACTION_REDIRECT:
2841 case FLOW_ACTION_MIRRED: {
2842 struct mlx5e_priv *out_priv;
2843 struct net_device *out_dev;
2847 /* out_dev is NULL when filters with
2848 * non-existing mirred device are replayed to
2854 if (attr->out_count >= MLX5_MAX_FLOW_FWD_VPORTS) {
2855 NL_SET_ERR_MSG_MOD(extack,
2856 "can't support more output ports, can't offload forwarding");
2857 pr_err("can't support more than %d output ports, can't offload forwarding\n",
2862 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2863 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2864 if (netdev_port_same_parent_id(priv->netdev,
2866 is_merged_eswitch_dev(priv, out_dev)) {
2867 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2868 struct net_device *uplink_dev = mlx5_eswitch_uplink_get_proto_dev(esw, REP_ETH);
2869 struct net_device *uplink_upper = netdev_master_upper_dev_get(uplink_dev);
2872 netif_is_lag_master(uplink_upper) &&
2873 uplink_upper == out_dev)
2874 out_dev = uplink_dev;
2876 if (is_vlan_dev(out_dev)) {
2877 err = add_vlan_push_action(priv, attr,
2883 if (is_vlan_dev(parse_attr->filter_dev)) {
2884 err = add_vlan_pop_action(priv, attr,
2890 if (!mlx5e_eswitch_rep(out_dev))
2893 out_priv = netdev_priv(out_dev);
2894 rpriv = out_priv->ppriv;
2895 attr->dests[attr->out_count].rep = rpriv->rep;
2896 attr->dests[attr->out_count].mdev = out_priv->mdev;
2899 parse_attr->mirred_ifindex[attr->out_count] =
2901 parse_attr->tun_info[attr->out_count] = *info;
2903 attr->dests[attr->out_count].flags |=
2904 MLX5_ESW_DEST_ENCAP;
2906 /* attr->dests[].rep is resolved when we
2909 } else if (parse_attr->filter_dev != priv->netdev) {
2910 /* All mlx5 devices are called to configure
2911 * high level device filters. Therefore, the
2912 * *attempt* to install a filter on invalid
2913 * eswitch should not trigger an explicit error
2917 NL_SET_ERR_MSG_MOD(extack,
2918 "devices are not on same switch HW, can't offload forwarding");
2919 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
2920 priv->netdev->name, out_dev->name);
2925 case FLOW_ACTION_TUNNEL_ENCAP:
2933 case FLOW_ACTION_VLAN_PUSH:
2934 case FLOW_ACTION_VLAN_POP:
2935 if (act->id == FLOW_ACTION_VLAN_PUSH &&
2936 (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP)) {
2937 /* Replace vlan pop+push with vlan modify */
2938 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2939 err = add_vlan_rewrite_action(priv,
2940 MLX5_FLOW_NAMESPACE_FDB,
2941 act, parse_attr, hdrs,
2944 err = parse_tc_vlan_action(priv, act, attr, &action);
2949 attr->split_count = attr->out_count;
2951 case FLOW_ACTION_VLAN_MANGLE:
2952 err = add_vlan_rewrite_action(priv,
2953 MLX5_FLOW_NAMESPACE_FDB,
2954 act, parse_attr, hdrs,
2959 attr->split_count = attr->out_count;
2961 case FLOW_ACTION_TUNNEL_DECAP:
2962 action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
2964 case FLOW_ACTION_GOTO: {
2965 u32 dest_chain = act->chain_index;
2966 u32 max_chain = mlx5_eswitch_get_chain_range(esw);
2968 if (dest_chain <= attr->chain) {
2969 NL_SET_ERR_MSG(extack, "Goto earlier chain isn't supported");
2972 if (dest_chain > max_chain) {
2973 NL_SET_ERR_MSG(extack, "Requested destination chain is out of supported range");
2976 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
2977 attr->dest_chain = dest_chain;
2981 NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
2986 if (MLX5_CAP_GEN(esw->dev, prio_tag_required) &&
2987 action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) {
2988 /* For prio tag mode, replace vlan pop with rewrite vlan prio
2991 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2992 err = add_vlan_prio_tag_rewrite_action(priv, parse_attr, hdrs,
2998 if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
2999 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
3000 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_FDB,
3001 parse_attr, hdrs, &action, extack);
3004 /* in case all pedit actions are skipped, remove the MOD_HDR
3005 * flag. we might have set split_count either by pedit or
3006 * pop/push. if there is no pop/push either, reset it too.
3008 if (parse_attr->num_mod_hdr_actions == 0) {
3009 action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3010 kfree(parse_attr->mod_hdr_actions);
3011 if (!((action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) ||
3012 (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH)))
3013 attr->split_count = 0;
3017 attr->action = action;
3018 if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
3021 if (attr->dest_chain) {
3022 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
3023 NL_SET_ERR_MSG(extack, "Mirroring goto chain rules isn't supported");
3026 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3029 if (attr->split_count > 0 && !mlx5_esw_has_fwd_fdb(priv->mdev)) {
3030 NL_SET_ERR_MSG_MOD(extack,
3031 "current firmware doesn't support split rule for port mirroring");
3032 netdev_warn_once(priv->netdev, "current firmware doesn't support split rule for port mirroring\n");
3039 static void get_flags(int flags, u16 *flow_flags)
3041 u16 __flow_flags = 0;
3043 if (flags & MLX5E_TC_INGRESS)
3044 __flow_flags |= MLX5E_TC_FLOW_INGRESS;
3045 if (flags & MLX5E_TC_EGRESS)
3046 __flow_flags |= MLX5E_TC_FLOW_EGRESS;
3048 if (flags & MLX5E_TC_ESW_OFFLOAD)
3049 __flow_flags |= MLX5E_TC_FLOW_ESWITCH;
3050 if (flags & MLX5E_TC_NIC_OFFLOAD)
3051 __flow_flags |= MLX5E_TC_FLOW_NIC;
3053 *flow_flags = __flow_flags;
3056 static const struct rhashtable_params tc_ht_params = {
3057 .head_offset = offsetof(struct mlx5e_tc_flow, node),
3058 .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
3059 .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
3060 .automatic_shrinking = true,
3063 static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv, int flags)
3065 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3066 struct mlx5e_rep_priv *uplink_rpriv;
3068 if (flags & MLX5E_TC_ESW_OFFLOAD) {
3069 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
3070 return &uplink_rpriv->uplink_priv.tc_ht;
3071 } else /* NIC offload */
3072 return &priv->fs.tc.ht;
3075 static bool is_peer_flow_needed(struct mlx5e_tc_flow *flow)
3077 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3078 bool is_rep_ingress = attr->in_rep->vport != MLX5_VPORT_UPLINK &&
3079 flow->flags & MLX5E_TC_FLOW_INGRESS;
3080 bool act_is_encap = !!(attr->action &
3081 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT);
3082 bool esw_paired = mlx5_devcom_is_paired(attr->in_mdev->priv.devcom,
3083 MLX5_DEVCOM_ESW_OFFLOADS);
3088 if ((mlx5_lag_is_sriov(attr->in_mdev) ||
3089 mlx5_lag_is_multipath(attr->in_mdev)) &&
3090 (is_rep_ingress || act_is_encap))
3097 mlx5e_alloc_flow(struct mlx5e_priv *priv, int attr_size,
3098 struct tc_cls_flower_offload *f, u16 flow_flags,
3099 struct mlx5e_tc_flow_parse_attr **__parse_attr,
3100 struct mlx5e_tc_flow **__flow)
3102 struct mlx5e_tc_flow_parse_attr *parse_attr;
3103 struct mlx5e_tc_flow *flow;
3106 flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
3107 parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
3108 if (!parse_attr || !flow) {
3113 flow->cookie = f->cookie;
3114 flow->flags = flow_flags;
3118 *__parse_attr = parse_attr;
3129 mlx5e_flow_esw_attr_init(struct mlx5_esw_flow_attr *esw_attr,
3130 struct mlx5e_priv *priv,
3131 struct mlx5e_tc_flow_parse_attr *parse_attr,
3132 struct tc_cls_flower_offload *f,
3133 struct mlx5_eswitch_rep *in_rep,
3134 struct mlx5_core_dev *in_mdev)
3136 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3138 esw_attr->parse_attr = parse_attr;
3139 esw_attr->chain = f->common.chain_index;
3140 esw_attr->prio = TC_H_MAJ(f->common.prio) >> 16;
3142 esw_attr->in_rep = in_rep;
3143 esw_attr->in_mdev = in_mdev;
3145 if (MLX5_CAP_ESW(esw->dev, counter_eswitch_affinity) ==
3146 MLX5_COUNTER_SOURCE_ESWITCH)
3147 esw_attr->counter_dev = in_mdev;
3149 esw_attr->counter_dev = priv->mdev;
3152 static struct mlx5e_tc_flow *
3153 __mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
3154 struct tc_cls_flower_offload *f,
3156 struct net_device *filter_dev,
3157 struct mlx5_eswitch_rep *in_rep,
3158 struct mlx5_core_dev *in_mdev)
3160 struct flow_rule *rule = tc_cls_flower_offload_flow_rule(f);
3161 struct netlink_ext_ack *extack = f->common.extack;
3162 struct mlx5e_tc_flow_parse_attr *parse_attr;
3163 struct mlx5e_tc_flow *flow;
3166 flow_flags |= MLX5E_TC_FLOW_ESWITCH;
3167 attr_size = sizeof(struct mlx5_esw_flow_attr);
3168 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3169 &parse_attr, &flow);
3173 parse_attr->filter_dev = filter_dev;
3174 mlx5e_flow_esw_attr_init(flow->esw_attr,
3176 f, in_rep, in_mdev);
3178 err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
3183 err = parse_tc_fdb_actions(priv, &rule->action, flow, extack);
3187 err = mlx5e_tc_add_fdb_flow(priv, flow, extack);
3189 if (!(err == -ENETUNREACH && mlx5_lag_is_multipath(in_mdev)))
3192 add_unready_flow(flow);
3201 return ERR_PTR(err);
3204 static int mlx5e_tc_add_fdb_peer_flow(struct tc_cls_flower_offload *f,
3205 struct mlx5e_tc_flow *flow,
3208 struct mlx5e_priv *priv = flow->priv, *peer_priv;
3209 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch, *peer_esw;
3210 struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
3211 struct mlx5e_tc_flow_parse_attr *parse_attr;
3212 struct mlx5e_rep_priv *peer_urpriv;
3213 struct mlx5e_tc_flow *peer_flow;
3214 struct mlx5_core_dev *in_mdev;
3217 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
3221 peer_urpriv = mlx5_eswitch_get_uplink_priv(peer_esw, REP_ETH);
3222 peer_priv = netdev_priv(peer_urpriv->netdev);
3224 /* in_mdev is assigned of which the packet originated from.
3225 * So packets redirected to uplink use the same mdev of the
3226 * original flow and packets redirected from uplink use the
3229 if (flow->esw_attr->in_rep->vport == MLX5_VPORT_UPLINK)
3230 in_mdev = peer_priv->mdev;
3232 in_mdev = priv->mdev;
3234 parse_attr = flow->esw_attr->parse_attr;
3235 peer_flow = __mlx5e_add_fdb_flow(peer_priv, f, flow_flags,
3236 parse_attr->filter_dev,
3237 flow->esw_attr->in_rep, in_mdev);
3238 if (IS_ERR(peer_flow)) {
3239 err = PTR_ERR(peer_flow);
3243 flow->peer_flow = peer_flow;
3244 flow->flags |= MLX5E_TC_FLOW_DUP;
3245 mutex_lock(&esw->offloads.peer_mutex);
3246 list_add_tail(&flow->peer, &esw->offloads.peer_flows);
3247 mutex_unlock(&esw->offloads.peer_mutex);
3250 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
3255 mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
3256 struct tc_cls_flower_offload *f,
3258 struct net_device *filter_dev,
3259 struct mlx5e_tc_flow **__flow)
3261 struct mlx5e_rep_priv *rpriv = priv->ppriv;
3262 struct mlx5_eswitch_rep *in_rep = rpriv->rep;
3263 struct mlx5_core_dev *in_mdev = priv->mdev;
3264 struct mlx5e_tc_flow *flow;
3267 flow = __mlx5e_add_fdb_flow(priv, f, flow_flags, filter_dev, in_rep,
3270 return PTR_ERR(flow);
3272 if (is_peer_flow_needed(flow)) {
3273 err = mlx5e_tc_add_fdb_peer_flow(f, flow, flow_flags);
3275 mlx5e_tc_del_fdb_flow(priv, flow);
3289 mlx5e_add_nic_flow(struct mlx5e_priv *priv,
3290 struct tc_cls_flower_offload *f,
3292 struct net_device *filter_dev,
3293 struct mlx5e_tc_flow **__flow)
3295 struct flow_rule *rule = tc_cls_flower_offload_flow_rule(f);
3296 struct netlink_ext_ack *extack = f->common.extack;
3297 struct mlx5e_tc_flow_parse_attr *parse_attr;
3298 struct mlx5e_tc_flow *flow;
3301 /* multi-chain not supported for NIC rules */
3302 if (!tc_cls_can_offload_and_chain0(priv->netdev, &f->common))
3305 flow_flags |= MLX5E_TC_FLOW_NIC;
3306 attr_size = sizeof(struct mlx5_nic_flow_attr);
3307 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3308 &parse_attr, &flow);
3312 parse_attr->filter_dev = filter_dev;
3313 err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
3318 err = parse_tc_nic_actions(priv, &rule->action, parse_attr, flow, extack);
3322 err = mlx5e_tc_add_nic_flow(priv, parse_attr, flow, extack);
3326 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
3340 mlx5e_tc_add_flow(struct mlx5e_priv *priv,
3341 struct tc_cls_flower_offload *f,
3343 struct net_device *filter_dev,
3344 struct mlx5e_tc_flow **flow)
3346 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3350 get_flags(flags, &flow_flags);
3352 if (!tc_can_offload_extack(priv->netdev, f->common.extack))
3355 if (esw && esw->mode == SRIOV_OFFLOADS)
3356 err = mlx5e_add_fdb_flow(priv, f, flow_flags,
3359 err = mlx5e_add_nic_flow(priv, f, flow_flags,
3365 int mlx5e_configure_flower(struct net_device *dev, struct mlx5e_priv *priv,
3366 struct tc_cls_flower_offload *f, int flags)
3368 struct netlink_ext_ack *extack = f->common.extack;
3369 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
3370 struct mlx5e_tc_flow *flow;
3373 flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
3375 NL_SET_ERR_MSG_MOD(extack,
3376 "flow cookie already exists, ignoring");
3377 netdev_warn_once(priv->netdev,
3378 "flow cookie %lx already exists, ignoring\n",
3384 err = mlx5e_tc_add_flow(priv, f, flags, dev, &flow);
3388 err = rhashtable_insert_fast(tc_ht, &flow->node, tc_ht_params);
3395 mlx5e_tc_del_flow(priv, flow);
3401 #define DIRECTION_MASK (MLX5E_TC_INGRESS | MLX5E_TC_EGRESS)
3402 #define FLOW_DIRECTION_MASK (MLX5E_TC_FLOW_INGRESS | MLX5E_TC_FLOW_EGRESS)
3404 static bool same_flow_direction(struct mlx5e_tc_flow *flow, int flags)
3406 if ((flow->flags & FLOW_DIRECTION_MASK) == (flags & DIRECTION_MASK))
3412 int mlx5e_delete_flower(struct net_device *dev, struct mlx5e_priv *priv,
3413 struct tc_cls_flower_offload *f, int flags)
3415 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
3416 struct mlx5e_tc_flow *flow;
3418 flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
3419 if (!flow || !same_flow_direction(flow, flags))
3422 rhashtable_remove_fast(tc_ht, &flow->node, tc_ht_params);
3424 mlx5e_tc_del_flow(priv, flow);
3431 int mlx5e_stats_flower(struct net_device *dev, struct mlx5e_priv *priv,
3432 struct tc_cls_flower_offload *f, int flags)
3434 struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
3435 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
3436 struct mlx5_eswitch *peer_esw;
3437 struct mlx5e_tc_flow *flow;
3438 struct mlx5_fc *counter;
3443 flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
3444 if (!flow || !same_flow_direction(flow, flags))
3447 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
3448 counter = mlx5e_tc_get_counter(flow);
3452 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
3455 /* Under multipath it's possible for one rule to be currently
3456 * un-offloaded while the other rule is offloaded.
3458 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
3462 if ((flow->flags & MLX5E_TC_FLOW_DUP) &&
3463 (flow->peer_flow->flags & MLX5E_TC_FLOW_OFFLOADED)) {
3468 counter = mlx5e_tc_get_counter(flow->peer_flow);
3470 goto no_peer_counter;
3471 mlx5_fc_query_cached(counter, &bytes2, &packets2, &lastuse2);
3474 packets += packets2;
3475 lastuse = max_t(u64, lastuse, lastuse2);
3479 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
3481 flow_stats_update(&f->stats, bytes, packets, lastuse);
3486 static void mlx5e_tc_hairpin_update_dead_peer(struct mlx5e_priv *priv,
3487 struct mlx5e_priv *peer_priv)
3489 struct mlx5_core_dev *peer_mdev = peer_priv->mdev;
3490 struct mlx5e_hairpin_entry *hpe;
3494 if (!same_hw_devs(priv, peer_priv))
3497 peer_vhca_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
3499 hash_for_each(priv->fs.tc.hairpin_tbl, bkt, hpe, hairpin_hlist) {
3500 if (hpe->peer_vhca_id == peer_vhca_id)
3501 hpe->hp->pair->peer_gone = true;
3505 static int mlx5e_tc_netdev_event(struct notifier_block *this,
3506 unsigned long event, void *ptr)
3508 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3509 struct mlx5e_flow_steering *fs;
3510 struct mlx5e_priv *peer_priv;
3511 struct mlx5e_tc_table *tc;
3512 struct mlx5e_priv *priv;
3514 if (ndev->netdev_ops != &mlx5e_netdev_ops ||
3515 event != NETDEV_UNREGISTER ||
3516 ndev->reg_state == NETREG_REGISTERED)
3519 tc = container_of(this, struct mlx5e_tc_table, netdevice_nb);
3520 fs = container_of(tc, struct mlx5e_flow_steering, tc);
3521 priv = container_of(fs, struct mlx5e_priv, fs);
3522 peer_priv = netdev_priv(ndev);
3523 if (priv == peer_priv ||
3524 !(priv->netdev->features & NETIF_F_HW_TC))
3527 mlx5e_tc_hairpin_update_dead_peer(priv, peer_priv);
3532 int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
3534 struct mlx5e_tc_table *tc = &priv->fs.tc;
3537 hash_init(tc->mod_hdr_tbl);
3538 hash_init(tc->hairpin_tbl);
3540 err = rhashtable_init(&tc->ht, &tc_ht_params);
3544 tc->netdevice_nb.notifier_call = mlx5e_tc_netdev_event;
3545 if (register_netdevice_notifier(&tc->netdevice_nb)) {
3546 tc->netdevice_nb.notifier_call = NULL;
3547 mlx5_core_warn(priv->mdev, "Failed to register netdev notifier\n");
3553 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
3555 struct mlx5e_tc_flow *flow = ptr;
3556 struct mlx5e_priv *priv = flow->priv;
3558 mlx5e_tc_del_flow(priv, flow);
3562 void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
3564 struct mlx5e_tc_table *tc = &priv->fs.tc;
3566 if (tc->netdevice_nb.notifier_call)
3567 unregister_netdevice_notifier(&tc->netdevice_nb);
3569 rhashtable_destroy(&tc->ht);
3571 if (!IS_ERR_OR_NULL(tc->t)) {
3572 mlx5_destroy_flow_table(tc->t);
3577 int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
3579 return rhashtable_init(tc_ht, &tc_ht_params);
3582 void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
3584 rhashtable_free_and_destroy(tc_ht, _mlx5e_tc_del_flow, NULL);
3587 int mlx5e_tc_num_filters(struct mlx5e_priv *priv, int flags)
3589 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
3591 return atomic_read(&tc_ht->nelems);
3594 void mlx5e_tc_clean_fdb_peer_flows(struct mlx5_eswitch *esw)
3596 struct mlx5e_tc_flow *flow, *tmp;
3598 list_for_each_entry_safe(flow, tmp, &esw->offloads.peer_flows, peer)
3599 __mlx5e_tc_del_fdb_peer_flow(flow);
3602 void mlx5e_tc_reoffload_flows_work(struct work_struct *work)
3604 struct mlx5_rep_uplink_priv *rpriv =
3605 container_of(work, struct mlx5_rep_uplink_priv,
3606 reoffload_flows_work);
3607 struct mlx5e_tc_flow *flow, *tmp;
3610 list_for_each_entry_safe(flow, tmp, &rpriv->unready_flows, unready) {
3611 if (!mlx5e_tc_add_fdb_flow(flow->priv, flow, NULL))
3612 remove_unready_flow(flow);