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/flow_offload.h>
35 #include <net/sch_generic.h>
36 #include <net/pkt_cls.h>
37 #include <net/tc_act/tc_gact.h>
38 #include <net/tc_act/tc_skbedit.h>
39 #include <linux/mlx5/fs.h>
40 #include <linux/mlx5/device.h>
41 #include <linux/rhashtable.h>
42 #include <linux/refcount.h>
43 #include <linux/completion.h>
44 #include <net/tc_act/tc_mirred.h>
45 #include <net/tc_act/tc_vlan.h>
46 #include <net/tc_act/tc_tunnel_key.h>
47 #include <net/tc_act/tc_pedit.h>
48 #include <net/tc_act/tc_csum.h>
49 #include <net/tc_act/tc_mpls.h>
50 #include <net/psample.h>
52 #include <net/ipv6_stubs.h>
53 #include <net/bareudp.h>
54 #include <net/bonding.h>
57 #include "en/rep/tc.h"
58 #include "en/rep/neigh.h"
63 #include "en/tc_tun.h"
64 #include "en/mapping.h"
66 #include "en/mod_hdr.h"
67 #include "en/tc_priv.h"
68 #include "en/tc_tun_encap.h"
69 #include "esw/sample.h"
70 #include "lib/devcom.h"
71 #include "lib/geneve.h"
72 #include "lib/fs_chains.h"
73 #include "diag/en_tc_tracepoint.h"
74 #include <asm/div64.h>
76 #define nic_chains(priv) ((priv)->fs.tc.chains)
77 #define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_add_copy_action_in_auto)
79 #define MLX5E_TC_TABLE_NUM_GROUPS 4
80 #define MLX5E_TC_TABLE_MAX_GROUP_SIZE BIT(18)
82 struct mlx5e_tc_attr_to_reg_mapping mlx5e_tc_attr_to_reg_mappings[] = {
84 .mfield = MLX5_ACTION_IN_FIELD_METADATA_REG_C_0,
89 .mfield = MLX5_ACTION_IN_FIELD_METADATA_REG_C_0,
94 .mfield = MLX5_ACTION_IN_FIELD_METADATA_REG_C_1,
96 .mlen = ((ESW_TUN_OPTS_BITS + ESW_TUN_ID_BITS) / 8),
97 .soffset = MLX5_BYTE_OFF(fte_match_param,
98 misc_parameters_2.metadata_reg_c_1),
100 [ZONE_TO_REG] = zone_to_reg_ct,
101 [ZONE_RESTORE_TO_REG] = zone_restore_to_reg_ct,
102 [CTSTATE_TO_REG] = ctstate_to_reg_ct,
103 [MARK_TO_REG] = mark_to_reg_ct,
104 [LABELS_TO_REG] = labels_to_reg_ct,
105 [FTEID_TO_REG] = fteid_to_reg_ct,
106 /* For NIC rules we store the retore metadata directly
107 * into reg_b that is passed to SW since we don't
108 * jump between steering domains.
110 [NIC_CHAIN_TO_REG] = {
111 .mfield = MLX5_ACTION_IN_FIELD_METADATA_REG_B,
115 [NIC_ZONE_RESTORE_TO_REG] = nic_zone_restore_to_reg_ct,
118 /* To avoid false lock dependency warning set the tc_ht lock
119 * class different than the lock class of the ht being used when deleting
120 * last flow from a group and then deleting a group, we get into del_sw_flow_group()
121 * which call rhashtable_destroy on fg->ftes_hash which will take ht->mutex but
122 * it's different than the ht->mutex here.
124 static struct lock_class_key tc_ht_lock_key;
126 static void mlx5e_put_flow_tunnel_id(struct mlx5e_tc_flow *flow);
129 mlx5e_tc_match_to_reg_match(struct mlx5_flow_spec *spec,
130 enum mlx5e_tc_attr_to_reg type,
134 int soffset = mlx5e_tc_attr_to_reg_mappings[type].soffset;
135 int match_len = mlx5e_tc_attr_to_reg_mappings[type].mlen;
136 void *headers_c = spec->match_criteria;
137 void *headers_v = spec->match_value;
140 fmask = headers_c + soffset;
141 fval = headers_v + soffset;
143 mask = (__force u32)(cpu_to_be32(mask)) >> (32 - (match_len * 8));
144 data = (__force u32)(cpu_to_be32(data)) >> (32 - (match_len * 8));
146 memcpy(fmask, &mask, match_len);
147 memcpy(fval, &data, match_len);
149 spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS_2;
153 mlx5e_tc_match_to_reg_get_match(struct mlx5_flow_spec *spec,
154 enum mlx5e_tc_attr_to_reg type,
158 int soffset = mlx5e_tc_attr_to_reg_mappings[type].soffset;
159 int match_len = mlx5e_tc_attr_to_reg_mappings[type].mlen;
160 void *headers_c = spec->match_criteria;
161 void *headers_v = spec->match_value;
164 fmask = headers_c + soffset;
165 fval = headers_v + soffset;
167 memcpy(mask, fmask, match_len);
168 memcpy(data, fval, match_len);
170 *mask = be32_to_cpu((__force __be32)(*mask << (32 - (match_len * 8))));
171 *data = be32_to_cpu((__force __be32)(*data << (32 - (match_len * 8))));
175 mlx5e_tc_match_to_reg_set_and_get_id(struct mlx5_core_dev *mdev,
176 struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts,
177 enum mlx5_flow_namespace_type ns,
178 enum mlx5e_tc_attr_to_reg type,
181 int moffset = mlx5e_tc_attr_to_reg_mappings[type].moffset;
182 int mfield = mlx5e_tc_attr_to_reg_mappings[type].mfield;
183 int mlen = mlx5e_tc_attr_to_reg_mappings[type].mlen;
187 err = alloc_mod_hdr_actions(mdev, ns, mod_hdr_acts);
191 modact = mod_hdr_acts->actions +
192 (mod_hdr_acts->num_actions * MLX5_MH_ACT_SZ);
194 /* Firmware has 5bit length field and 0 means 32bits */
198 MLX5_SET(set_action_in, modact, action_type, MLX5_ACTION_TYPE_SET);
199 MLX5_SET(set_action_in, modact, field, mfield);
200 MLX5_SET(set_action_in, modact, offset, moffset * 8);
201 MLX5_SET(set_action_in, modact, length, mlen * 8);
202 MLX5_SET(set_action_in, modact, data, data);
203 err = mod_hdr_acts->num_actions;
204 mod_hdr_acts->num_actions++;
209 static struct mlx5_tc_ct_priv *
210 get_ct_priv(struct mlx5e_priv *priv)
212 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
213 struct mlx5_rep_uplink_priv *uplink_priv;
214 struct mlx5e_rep_priv *uplink_rpriv;
216 if (is_mdev_switchdev_mode(priv->mdev)) {
217 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
218 uplink_priv = &uplink_rpriv->uplink_priv;
220 return uplink_priv->ct_priv;
223 return priv->fs.tc.ct;
226 #if IS_ENABLED(CONFIG_MLX5_TC_SAMPLE)
227 static struct mlx5_esw_psample *
228 get_sample_priv(struct mlx5e_priv *priv)
230 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
231 struct mlx5_rep_uplink_priv *uplink_priv;
232 struct mlx5e_rep_priv *uplink_rpriv;
234 if (is_mdev_switchdev_mode(priv->mdev)) {
235 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
236 uplink_priv = &uplink_rpriv->uplink_priv;
238 return uplink_priv->esw_psample;
245 struct mlx5_flow_handle *
246 mlx5_tc_rule_insert(struct mlx5e_priv *priv,
247 struct mlx5_flow_spec *spec,
248 struct mlx5_flow_attr *attr)
250 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
252 if (is_mdev_switchdev_mode(priv->mdev))
253 return mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
255 return mlx5e_add_offloaded_nic_rule(priv, spec, attr);
259 mlx5_tc_rule_delete(struct mlx5e_priv *priv,
260 struct mlx5_flow_handle *rule,
261 struct mlx5_flow_attr *attr)
263 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
265 if (is_mdev_switchdev_mode(priv->mdev)) {
266 mlx5_eswitch_del_offloaded_rule(esw, rule, attr);
271 mlx5e_del_offloaded_nic_rule(priv, rule, attr);
275 mlx5e_tc_match_to_reg_set(struct mlx5_core_dev *mdev,
276 struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts,
277 enum mlx5_flow_namespace_type ns,
278 enum mlx5e_tc_attr_to_reg type,
281 int ret = mlx5e_tc_match_to_reg_set_and_get_id(mdev, mod_hdr_acts, ns, type, data);
283 return ret < 0 ? ret : 0;
286 void mlx5e_tc_match_to_reg_mod_hdr_change(struct mlx5_core_dev *mdev,
287 struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts,
288 enum mlx5e_tc_attr_to_reg type,
289 int act_id, u32 data)
291 int moffset = mlx5e_tc_attr_to_reg_mappings[type].moffset;
292 int mfield = mlx5e_tc_attr_to_reg_mappings[type].mfield;
293 int mlen = mlx5e_tc_attr_to_reg_mappings[type].mlen;
296 modact = mod_hdr_acts->actions + (act_id * MLX5_MH_ACT_SZ);
298 /* Firmware has 5bit length field and 0 means 32bits */
302 MLX5_SET(set_action_in, modact, action_type, MLX5_ACTION_TYPE_SET);
303 MLX5_SET(set_action_in, modact, field, mfield);
304 MLX5_SET(set_action_in, modact, offset, moffset * 8);
305 MLX5_SET(set_action_in, modact, length, mlen * 8);
306 MLX5_SET(set_action_in, modact, data, data);
309 struct mlx5e_hairpin {
310 struct mlx5_hairpin *pair;
312 struct mlx5_core_dev *func_mdev;
313 struct mlx5e_priv *func_priv;
318 struct mlx5e_rqt indir_rqt;
319 u32 indir_tirn[MLX5E_NUM_INDIR_TIRS];
320 struct mlx5e_ttc_table ttc;
323 struct mlx5e_hairpin_entry {
324 /* a node of a hash table which keeps all the hairpin entries */
325 struct hlist_node hairpin_hlist;
327 /* protects flows list */
328 spinlock_t flows_lock;
329 /* flows sharing the same hairpin */
330 struct list_head flows;
331 /* hpe's that were not fully initialized when dead peer update event
332 * function traversed them.
334 struct list_head dead_peer_wait_list;
338 struct mlx5e_hairpin *hp;
340 struct completion res_ready;
343 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
344 struct mlx5e_tc_flow *flow);
346 struct mlx5e_tc_flow *mlx5e_flow_get(struct mlx5e_tc_flow *flow)
348 if (!flow || !refcount_inc_not_zero(&flow->refcnt))
349 return ERR_PTR(-EINVAL);
353 void mlx5e_flow_put(struct mlx5e_priv *priv, struct mlx5e_tc_flow *flow)
355 if (refcount_dec_and_test(&flow->refcnt)) {
356 mlx5e_tc_del_flow(priv, flow);
357 kfree_rcu(flow, rcu_head);
361 bool mlx5e_is_eswitch_flow(struct mlx5e_tc_flow *flow)
363 return flow_flag_test(flow, ESWITCH);
366 static bool mlx5e_is_ft_flow(struct mlx5e_tc_flow *flow)
368 return flow_flag_test(flow, FT);
371 bool mlx5e_is_offloaded_flow(struct mlx5e_tc_flow *flow)
373 return flow_flag_test(flow, OFFLOADED);
376 static int get_flow_name_space(struct mlx5e_tc_flow *flow)
378 return mlx5e_is_eswitch_flow(flow) ?
379 MLX5_FLOW_NAMESPACE_FDB : MLX5_FLOW_NAMESPACE_KERNEL;
382 static struct mod_hdr_tbl *
383 get_mod_hdr_table(struct mlx5e_priv *priv, struct mlx5e_tc_flow *flow)
385 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
387 return get_flow_name_space(flow) == MLX5_FLOW_NAMESPACE_FDB ?
388 &esw->offloads.mod_hdr :
389 &priv->fs.tc.mod_hdr;
392 static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
393 struct mlx5e_tc_flow *flow,
394 struct mlx5e_tc_flow_parse_attr *parse_attr)
396 struct mlx5_modify_hdr *modify_hdr;
397 struct mlx5e_mod_hdr_handle *mh;
399 mh = mlx5e_mod_hdr_attach(priv->mdev, get_mod_hdr_table(priv, flow),
400 get_flow_name_space(flow),
401 &parse_attr->mod_hdr_acts);
405 modify_hdr = mlx5e_mod_hdr_get(mh);
406 flow->attr->modify_hdr = modify_hdr;
412 static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
413 struct mlx5e_tc_flow *flow)
415 /* flow wasn't fully initialized */
419 mlx5e_mod_hdr_detach(priv->mdev, get_mod_hdr_table(priv, flow),
425 struct mlx5_core_dev *mlx5e_hairpin_get_mdev(struct net *net, int ifindex)
427 struct net_device *netdev;
428 struct mlx5e_priv *priv;
430 netdev = __dev_get_by_index(net, ifindex);
431 priv = netdev_priv(netdev);
435 static int mlx5e_hairpin_create_transport(struct mlx5e_hairpin *hp)
437 u32 in[MLX5_ST_SZ_DW(create_tir_in)] = {};
441 err = mlx5_core_alloc_transport_domain(hp->func_mdev, &hp->tdn);
445 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
447 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
448 MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn[0]);
449 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
451 err = mlx5_core_create_tir(hp->func_mdev, in, &hp->tirn);
458 mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
463 static void mlx5e_hairpin_destroy_transport(struct mlx5e_hairpin *hp)
465 mlx5_core_destroy_tir(hp->func_mdev, hp->tirn);
466 mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
469 static int mlx5e_hairpin_fill_rqt_rqns(struct mlx5e_hairpin *hp, void *rqtc)
471 struct mlx5e_priv *priv = hp->func_priv;
472 int i, ix, sz = MLX5E_INDIR_RQT_SIZE;
473 u32 *indirection_rqt, rqn;
475 indirection_rqt = kcalloc(sz, sizeof(*indirection_rqt), GFP_KERNEL);
476 if (!indirection_rqt)
479 mlx5e_build_default_indir_rqt(indirection_rqt, sz,
482 for (i = 0; i < sz; i++) {
484 if (priv->rss_params.hfunc == ETH_RSS_HASH_XOR)
485 ix = mlx5e_bits_invert(i, ilog2(sz));
486 ix = indirection_rqt[ix];
487 rqn = hp->pair->rqn[ix];
488 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
491 kfree(indirection_rqt);
495 static int mlx5e_hairpin_create_indirect_rqt(struct mlx5e_hairpin *hp)
497 int inlen, err, sz = MLX5E_INDIR_RQT_SIZE;
498 struct mlx5e_priv *priv = hp->func_priv;
499 struct mlx5_core_dev *mdev = priv->mdev;
503 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
504 in = kvzalloc(inlen, GFP_KERNEL);
508 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
510 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
511 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
513 err = mlx5e_hairpin_fill_rqt_rqns(hp, rqtc);
517 err = mlx5_core_create_rqt(mdev, in, inlen, &hp->indir_rqt.rqtn);
519 hp->indir_rqt.enabled = true;
526 static int mlx5e_hairpin_create_indirect_tirs(struct mlx5e_hairpin *hp)
528 struct mlx5e_priv *priv = hp->func_priv;
529 u32 in[MLX5_ST_SZ_DW(create_tir_in)];
533 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
534 struct mlx5e_tirc_config ttconfig = mlx5e_tirc_get_default_config(tt);
536 memset(in, 0, MLX5_ST_SZ_BYTES(create_tir_in));
537 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
539 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
540 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
541 MLX5_SET(tirc, tirc, indirect_table, hp->indir_rqt.rqtn);
542 mlx5e_build_indir_tir_ctx_hash(&priv->rss_params, &ttconfig, tirc, false);
544 err = mlx5_core_create_tir(hp->func_mdev, in,
545 &hp->indir_tirn[tt]);
547 mlx5_core_warn(hp->func_mdev, "create indirect tirs failed, %d\n", err);
548 goto err_destroy_tirs;
554 for (i = 0; i < tt; i++)
555 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[i]);
559 static void mlx5e_hairpin_destroy_indirect_tirs(struct mlx5e_hairpin *hp)
563 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
564 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[tt]);
567 static void mlx5e_hairpin_set_ttc_params(struct mlx5e_hairpin *hp,
568 struct ttc_params *ttc_params)
570 struct mlx5_flow_table_attr *ft_attr = &ttc_params->ft_attr;
573 memset(ttc_params, 0, sizeof(*ttc_params));
575 ttc_params->any_tt_tirn = hp->tirn;
577 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
578 ttc_params->indir_tirn[tt] = hp->indir_tirn[tt];
580 ft_attr->max_fte = MLX5E_TTC_TABLE_SIZE;
581 ft_attr->level = MLX5E_TC_TTC_FT_LEVEL;
582 ft_attr->prio = MLX5E_TC_PRIO;
585 static int mlx5e_hairpin_rss_init(struct mlx5e_hairpin *hp)
587 struct mlx5e_priv *priv = hp->func_priv;
588 struct ttc_params ttc_params;
591 err = mlx5e_hairpin_create_indirect_rqt(hp);
595 err = mlx5e_hairpin_create_indirect_tirs(hp);
597 goto err_create_indirect_tirs;
599 mlx5e_hairpin_set_ttc_params(hp, &ttc_params);
600 err = mlx5e_create_ttc_table(priv, &ttc_params, &hp->ttc);
602 goto err_create_ttc_table;
604 netdev_dbg(priv->netdev, "add hairpin: using %d channels rss ttc table id %x\n",
605 hp->num_channels, hp->ttc.ft.t->id);
609 err_create_ttc_table:
610 mlx5e_hairpin_destroy_indirect_tirs(hp);
611 err_create_indirect_tirs:
612 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
617 static void mlx5e_hairpin_rss_cleanup(struct mlx5e_hairpin *hp)
619 struct mlx5e_priv *priv = hp->func_priv;
621 mlx5e_destroy_ttc_table(priv, &hp->ttc);
622 mlx5e_hairpin_destroy_indirect_tirs(hp);
623 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
626 static struct mlx5e_hairpin *
627 mlx5e_hairpin_create(struct mlx5e_priv *priv, struct mlx5_hairpin_params *params,
630 struct mlx5_core_dev *func_mdev, *peer_mdev;
631 struct mlx5e_hairpin *hp;
632 struct mlx5_hairpin *pair;
635 hp = kzalloc(sizeof(*hp), GFP_KERNEL);
637 return ERR_PTR(-ENOMEM);
639 func_mdev = priv->mdev;
640 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
642 pair = mlx5_core_hairpin_create(func_mdev, peer_mdev, params);
645 goto create_pair_err;
648 hp->func_mdev = func_mdev;
649 hp->func_priv = priv;
650 hp->num_channels = params->num_channels;
652 err = mlx5e_hairpin_create_transport(hp);
654 goto create_transport_err;
656 if (hp->num_channels > 1) {
657 err = mlx5e_hairpin_rss_init(hp);
665 mlx5e_hairpin_destroy_transport(hp);
666 create_transport_err:
667 mlx5_core_hairpin_destroy(hp->pair);
673 static void mlx5e_hairpin_destroy(struct mlx5e_hairpin *hp)
675 if (hp->num_channels > 1)
676 mlx5e_hairpin_rss_cleanup(hp);
677 mlx5e_hairpin_destroy_transport(hp);
678 mlx5_core_hairpin_destroy(hp->pair);
682 static inline u32 hash_hairpin_info(u16 peer_vhca_id, u8 prio)
684 return (peer_vhca_id << 16 | prio);
687 static struct mlx5e_hairpin_entry *mlx5e_hairpin_get(struct mlx5e_priv *priv,
688 u16 peer_vhca_id, u8 prio)
690 struct mlx5e_hairpin_entry *hpe;
691 u32 hash_key = hash_hairpin_info(peer_vhca_id, prio);
693 hash_for_each_possible(priv->fs.tc.hairpin_tbl, hpe,
694 hairpin_hlist, hash_key) {
695 if (hpe->peer_vhca_id == peer_vhca_id && hpe->prio == prio) {
696 refcount_inc(&hpe->refcnt);
704 static void mlx5e_hairpin_put(struct mlx5e_priv *priv,
705 struct mlx5e_hairpin_entry *hpe)
707 /* no more hairpin flows for us, release the hairpin pair */
708 if (!refcount_dec_and_mutex_lock(&hpe->refcnt, &priv->fs.tc.hairpin_tbl_lock))
710 hash_del(&hpe->hairpin_hlist);
711 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
713 if (!IS_ERR_OR_NULL(hpe->hp)) {
714 netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
715 dev_name(hpe->hp->pair->peer_mdev->device));
717 mlx5e_hairpin_destroy(hpe->hp);
720 WARN_ON(!list_empty(&hpe->flows));
724 #define UNKNOWN_MATCH_PRIO 8
726 static int mlx5e_hairpin_get_prio(struct mlx5e_priv *priv,
727 struct mlx5_flow_spec *spec, u8 *match_prio,
728 struct netlink_ext_ack *extack)
730 void *headers_c, *headers_v;
731 u8 prio_val, prio_mask = 0;
734 #ifdef CONFIG_MLX5_CORE_EN_DCB
735 if (priv->dcbx_dp.trust_state != MLX5_QPTS_TRUST_PCP) {
736 NL_SET_ERR_MSG_MOD(extack,
737 "only PCP trust state supported for hairpin");
741 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, outer_headers);
742 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
744 vlan_present = MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag);
746 prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
747 prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
750 if (!vlan_present || !prio_mask) {
751 prio_val = UNKNOWN_MATCH_PRIO;
752 } else if (prio_mask != 0x7) {
753 NL_SET_ERR_MSG_MOD(extack,
754 "masked priority match not supported for hairpin");
758 *match_prio = prio_val;
762 static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
763 struct mlx5e_tc_flow *flow,
764 struct mlx5e_tc_flow_parse_attr *parse_attr,
765 struct netlink_ext_ack *extack)
767 int peer_ifindex = parse_attr->mirred_ifindex[0];
768 struct mlx5_hairpin_params params;
769 struct mlx5_core_dev *peer_mdev;
770 struct mlx5e_hairpin_entry *hpe;
771 struct mlx5e_hairpin *hp;
778 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
779 if (!MLX5_CAP_GEN(priv->mdev, hairpin) || !MLX5_CAP_GEN(peer_mdev, hairpin)) {
780 NL_SET_ERR_MSG_MOD(extack, "hairpin is not supported");
784 peer_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
785 err = mlx5e_hairpin_get_prio(priv, &parse_attr->spec, &match_prio,
790 mutex_lock(&priv->fs.tc.hairpin_tbl_lock);
791 hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
793 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
794 wait_for_completion(&hpe->res_ready);
796 if (IS_ERR(hpe->hp)) {
803 hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
805 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
809 spin_lock_init(&hpe->flows_lock);
810 INIT_LIST_HEAD(&hpe->flows);
811 INIT_LIST_HEAD(&hpe->dead_peer_wait_list);
812 hpe->peer_vhca_id = peer_id;
813 hpe->prio = match_prio;
814 refcount_set(&hpe->refcnt, 1);
815 init_completion(&hpe->res_ready);
817 hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
818 hash_hairpin_info(peer_id, match_prio));
819 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
821 params.log_data_size = 15;
822 params.log_data_size = min_t(u8, params.log_data_size,
823 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_wq_data_sz));
824 params.log_data_size = max_t(u8, params.log_data_size,
825 MLX5_CAP_GEN(priv->mdev, log_min_hairpin_wq_data_sz));
827 params.log_num_packets = params.log_data_size -
828 MLX5_MPWRQ_MIN_LOG_STRIDE_SZ(priv->mdev);
829 params.log_num_packets = min_t(u8, params.log_num_packets,
830 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_num_packets));
832 params.q_counter = priv->q_counter;
833 /* set hairpin pair per each 50Gbs share of the link */
834 mlx5e_port_max_linkspeed(priv->mdev, &link_speed);
835 link_speed = max_t(u32, link_speed, 50000);
836 link_speed64 = link_speed;
837 do_div(link_speed64, 50000);
838 params.num_channels = link_speed64;
840 hp = mlx5e_hairpin_create(priv, ¶ms, peer_ifindex);
842 complete_all(&hpe->res_ready);
848 netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
849 hp->tirn, hp->pair->rqn[0],
850 dev_name(hp->pair->peer_mdev->device),
851 hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
854 if (hpe->hp->num_channels > 1) {
855 flow_flag_set(flow, HAIRPIN_RSS);
856 flow->attr->nic_attr->hairpin_ft = hpe->hp->ttc.ft.t;
858 flow->attr->nic_attr->hairpin_tirn = hpe->hp->tirn;
862 spin_lock(&hpe->flows_lock);
863 list_add(&flow->hairpin, &hpe->flows);
864 spin_unlock(&hpe->flows_lock);
869 mlx5e_hairpin_put(priv, hpe);
873 static void mlx5e_hairpin_flow_del(struct mlx5e_priv *priv,
874 struct mlx5e_tc_flow *flow)
876 /* flow wasn't fully initialized */
880 spin_lock(&flow->hpe->flows_lock);
881 list_del(&flow->hairpin);
882 spin_unlock(&flow->hpe->flows_lock);
884 mlx5e_hairpin_put(priv, flow->hpe);
888 struct mlx5_flow_handle *
889 mlx5e_add_offloaded_nic_rule(struct mlx5e_priv *priv,
890 struct mlx5_flow_spec *spec,
891 struct mlx5_flow_attr *attr)
893 struct mlx5_flow_context *flow_context = &spec->flow_context;
894 struct mlx5_fs_chains *nic_chains = nic_chains(priv);
895 struct mlx5_nic_flow_attr *nic_attr = attr->nic_attr;
896 struct mlx5e_tc_table *tc = &priv->fs.tc;
897 struct mlx5_flow_destination dest[2] = {};
898 struct mlx5_flow_act flow_act = {
899 .action = attr->action,
900 .flags = FLOW_ACT_NO_APPEND,
902 struct mlx5_flow_handle *rule;
903 struct mlx5_flow_table *ft;
906 flow_context->flags |= FLOW_CONTEXT_HAS_TAG;
907 flow_context->flow_tag = nic_attr->flow_tag;
910 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
911 dest[dest_ix].ft = attr->dest_ft;
913 } else if (nic_attr->hairpin_ft) {
914 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
915 dest[dest_ix].ft = nic_attr->hairpin_ft;
917 } else if (nic_attr->hairpin_tirn) {
918 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_TIR;
919 dest[dest_ix].tir_num = nic_attr->hairpin_tirn;
921 } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
922 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
923 if (attr->dest_chain) {
924 dest[dest_ix].ft = mlx5_chains_get_table(nic_chains,
927 if (IS_ERR(dest[dest_ix].ft))
928 return ERR_CAST(dest[dest_ix].ft);
930 dest[dest_ix].ft = mlx5e_vlan_get_flowtable(priv->fs.vlan);
935 if (dest[0].type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
936 MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, ignore_flow_level))
937 flow_act.flags |= FLOW_ACT_IGNORE_FLOW_LEVEL;
939 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
940 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
941 dest[dest_ix].counter_id = mlx5_fc_id(attr->counter);
945 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
946 flow_act.modify_hdr = attr->modify_hdr;
948 mutex_lock(&tc->t_lock);
949 if (IS_ERR_OR_NULL(tc->t)) {
950 /* Create the root table here if doesn't exist yet */
952 mlx5_chains_get_table(nic_chains, 0, 1, MLX5E_TC_FT_LEVEL);
955 mutex_unlock(&tc->t_lock);
956 netdev_err(priv->netdev,
957 "Failed to create tc offload table\n");
958 rule = ERR_CAST(priv->fs.tc.t);
962 mutex_unlock(&tc->t_lock);
964 if (attr->chain || attr->prio)
965 ft = mlx5_chains_get_table(nic_chains,
966 attr->chain, attr->prio,
976 if (attr->outer_match_level != MLX5_MATCH_NONE)
977 spec->match_criteria_enable |= MLX5_MATCH_OUTER_HEADERS;
979 rule = mlx5_add_flow_rules(ft, spec,
980 &flow_act, dest, dest_ix);
987 if (attr->chain || attr->prio)
988 mlx5_chains_put_table(nic_chains,
989 attr->chain, attr->prio,
992 if (attr->dest_chain)
993 mlx5_chains_put_table(nic_chains,
997 return ERR_CAST(rule);
1001 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
1002 struct mlx5e_tc_flow_parse_attr *parse_attr,
1003 struct mlx5e_tc_flow *flow,
1004 struct netlink_ext_ack *extack)
1006 struct mlx5_flow_attr *attr = flow->attr;
1007 struct mlx5_core_dev *dev = priv->mdev;
1008 struct mlx5_fc *counter = NULL;
1011 if (flow_flag_test(flow, HAIRPIN)) {
1012 err = mlx5e_hairpin_flow_add(priv, flow, parse_attr, extack);
1017 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1018 counter = mlx5_fc_create(dev, true);
1019 if (IS_ERR(counter))
1020 return PTR_ERR(counter);
1022 attr->counter = counter;
1025 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
1026 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
1027 dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
1032 if (flow_flag_test(flow, CT))
1033 flow->rule[0] = mlx5_tc_ct_flow_offload(get_ct_priv(priv), flow, &parse_attr->spec,
1034 attr, &parse_attr->mod_hdr_acts);
1036 flow->rule[0] = mlx5e_add_offloaded_nic_rule(priv, &parse_attr->spec,
1039 return PTR_ERR_OR_ZERO(flow->rule[0]);
1042 void mlx5e_del_offloaded_nic_rule(struct mlx5e_priv *priv,
1043 struct mlx5_flow_handle *rule,
1044 struct mlx5_flow_attr *attr)
1046 struct mlx5_fs_chains *nic_chains = nic_chains(priv);
1048 mlx5_del_flow_rules(rule);
1050 if (attr->chain || attr->prio)
1051 mlx5_chains_put_table(nic_chains, attr->chain, attr->prio,
1054 if (attr->dest_chain)
1055 mlx5_chains_put_table(nic_chains, attr->dest_chain, 1,
1059 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
1060 struct mlx5e_tc_flow *flow)
1062 struct mlx5_flow_attr *attr = flow->attr;
1063 struct mlx5e_tc_table *tc = &priv->fs.tc;
1065 flow_flag_clear(flow, OFFLOADED);
1067 if (flow_flag_test(flow, CT))
1068 mlx5_tc_ct_delete_flow(get_ct_priv(flow->priv), flow, attr);
1069 else if (!IS_ERR_OR_NULL(flow->rule[0]))
1070 mlx5e_del_offloaded_nic_rule(priv, flow->rule[0], attr);
1072 /* Remove root table if no rules are left to avoid
1073 * extra steering hops.
1075 mutex_lock(&priv->fs.tc.t_lock);
1076 if (!mlx5e_tc_num_filters(priv, MLX5_TC_FLAG(NIC_OFFLOAD)) &&
1077 !IS_ERR_OR_NULL(tc->t)) {
1078 mlx5_chains_put_table(nic_chains(priv), 0, 1, MLX5E_TC_FT_LEVEL);
1079 priv->fs.tc.t = NULL;
1081 mutex_unlock(&priv->fs.tc.t_lock);
1083 kvfree(attr->parse_attr);
1085 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1086 mlx5e_detach_mod_hdr(priv, flow);
1088 mlx5_fc_destroy(priv->mdev, attr->counter);
1090 if (flow_flag_test(flow, HAIRPIN))
1091 mlx5e_hairpin_flow_del(priv, flow);
1096 struct mlx5_flow_handle *
1097 mlx5e_tc_offload_fdb_rules(struct mlx5_eswitch *esw,
1098 struct mlx5e_tc_flow *flow,
1099 struct mlx5_flow_spec *spec,
1100 struct mlx5_flow_attr *attr)
1102 struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts;
1103 struct mlx5_flow_handle *rule;
1105 if (attr->flags & MLX5_ESW_ATTR_FLAG_SLOW_PATH)
1106 return mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
1108 if (flow_flag_test(flow, CT)) {
1109 mod_hdr_acts = &attr->parse_attr->mod_hdr_acts;
1111 rule = mlx5_tc_ct_flow_offload(get_ct_priv(flow->priv),
1114 #if IS_ENABLED(CONFIG_MLX5_TC_SAMPLE)
1115 } else if (flow_flag_test(flow, SAMPLE)) {
1116 rule = mlx5_esw_sample_offload(get_sample_priv(flow->priv), spec, attr);
1119 rule = mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
1125 if (attr->esw_attr->split_count) {
1126 flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, spec, attr);
1127 if (IS_ERR(flow->rule[1])) {
1128 if (flow_flag_test(flow, CT))
1129 mlx5_tc_ct_delete_flow(get_ct_priv(flow->priv), flow, attr);
1131 mlx5_eswitch_del_offloaded_rule(esw, rule, attr);
1132 return flow->rule[1];
1139 void mlx5e_tc_unoffload_fdb_rules(struct mlx5_eswitch *esw,
1140 struct mlx5e_tc_flow *flow,
1141 struct mlx5_flow_attr *attr)
1143 flow_flag_clear(flow, OFFLOADED);
1145 if (attr->flags & MLX5_ESW_ATTR_FLAG_SLOW_PATH)
1146 goto offload_rule_0;
1148 if (flow_flag_test(flow, CT)) {
1149 mlx5_tc_ct_delete_flow(get_ct_priv(flow->priv), flow, attr);
1153 #if IS_ENABLED(CONFIG_MLX5_TC_SAMPLE)
1154 if (flow_flag_test(flow, SAMPLE)) {
1155 mlx5_esw_sample_unoffload(get_sample_priv(flow->priv), flow->rule[0], attr);
1160 if (attr->esw_attr->split_count)
1161 mlx5_eswitch_del_fwd_rule(esw, flow->rule[1], attr);
1164 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
1167 struct mlx5_flow_handle *
1168 mlx5e_tc_offload_to_slow_path(struct mlx5_eswitch *esw,
1169 struct mlx5e_tc_flow *flow,
1170 struct mlx5_flow_spec *spec)
1172 struct mlx5_flow_attr *slow_attr;
1173 struct mlx5_flow_handle *rule;
1175 slow_attr = mlx5_alloc_flow_attr(MLX5_FLOW_NAMESPACE_FDB);
1177 return ERR_PTR(-ENOMEM);
1179 memcpy(slow_attr, flow->attr, ESW_FLOW_ATTR_SZ);
1180 slow_attr->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1181 slow_attr->esw_attr->split_count = 0;
1182 slow_attr->flags |= MLX5_ESW_ATTR_FLAG_SLOW_PATH;
1184 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, slow_attr);
1186 flow_flag_set(flow, SLOW);
1193 void mlx5e_tc_unoffload_from_slow_path(struct mlx5_eswitch *esw,
1194 struct mlx5e_tc_flow *flow)
1196 struct mlx5_flow_attr *slow_attr;
1198 slow_attr = mlx5_alloc_flow_attr(MLX5_FLOW_NAMESPACE_FDB);
1200 mlx5_core_warn(flow->priv->mdev, "Unable to alloc attr to unoffload slow path rule\n");
1204 memcpy(slow_attr, flow->attr, ESW_FLOW_ATTR_SZ);
1205 slow_attr->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1206 slow_attr->esw_attr->split_count = 0;
1207 slow_attr->flags |= MLX5_ESW_ATTR_FLAG_SLOW_PATH;
1208 mlx5e_tc_unoffload_fdb_rules(esw, flow, slow_attr);
1209 flow_flag_clear(flow, SLOW);
1213 /* Caller must obtain uplink_priv->unready_flows_lock mutex before calling this
1216 static void unready_flow_add(struct mlx5e_tc_flow *flow,
1217 struct list_head *unready_flows)
1219 flow_flag_set(flow, NOT_READY);
1220 list_add_tail(&flow->unready, unready_flows);
1223 /* Caller must obtain uplink_priv->unready_flows_lock mutex before calling this
1226 static void unready_flow_del(struct mlx5e_tc_flow *flow)
1228 list_del(&flow->unready);
1229 flow_flag_clear(flow, NOT_READY);
1232 static void add_unready_flow(struct mlx5e_tc_flow *flow)
1234 struct mlx5_rep_uplink_priv *uplink_priv;
1235 struct mlx5e_rep_priv *rpriv;
1236 struct mlx5_eswitch *esw;
1238 esw = flow->priv->mdev->priv.eswitch;
1239 rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1240 uplink_priv = &rpriv->uplink_priv;
1242 mutex_lock(&uplink_priv->unready_flows_lock);
1243 unready_flow_add(flow, &uplink_priv->unready_flows);
1244 mutex_unlock(&uplink_priv->unready_flows_lock);
1247 static void remove_unready_flow(struct mlx5e_tc_flow *flow)
1249 struct mlx5_rep_uplink_priv *uplink_priv;
1250 struct mlx5e_rep_priv *rpriv;
1251 struct mlx5_eswitch *esw;
1253 esw = flow->priv->mdev->priv.eswitch;
1254 rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1255 uplink_priv = &rpriv->uplink_priv;
1257 mutex_lock(&uplink_priv->unready_flows_lock);
1258 unready_flow_del(flow);
1259 mutex_unlock(&uplink_priv->unready_flows_lock);
1262 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv);
1264 bool mlx5e_tc_is_vf_tunnel(struct net_device *out_dev, struct net_device *route_dev)
1266 struct mlx5_core_dev *out_mdev, *route_mdev;
1267 struct mlx5e_priv *out_priv, *route_priv;
1269 out_priv = netdev_priv(out_dev);
1270 out_mdev = out_priv->mdev;
1271 route_priv = netdev_priv(route_dev);
1272 route_mdev = route_priv->mdev;
1274 if (out_mdev->coredev_type != MLX5_COREDEV_PF ||
1275 route_mdev->coredev_type != MLX5_COREDEV_VF)
1278 return same_hw_devs(out_priv, route_priv);
1281 int mlx5e_tc_query_route_vport(struct net_device *out_dev, struct net_device *route_dev, u16 *vport)
1283 struct mlx5e_priv *out_priv, *route_priv;
1284 struct mlx5_core_dev *route_mdev;
1285 struct mlx5_eswitch *esw;
1289 out_priv = netdev_priv(out_dev);
1290 esw = out_priv->mdev->priv.eswitch;
1291 route_priv = netdev_priv(route_dev);
1292 route_mdev = route_priv->mdev;
1294 vhca_id = MLX5_CAP_GEN(route_mdev, vhca_id);
1295 err = mlx5_eswitch_vhca_id_to_vport(esw, vhca_id, vport);
1299 int mlx5e_tc_add_flow_mod_hdr(struct mlx5e_priv *priv,
1300 struct mlx5e_tc_flow_parse_attr *parse_attr,
1301 struct mlx5e_tc_flow *flow)
1303 struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts = &parse_attr->mod_hdr_acts;
1304 struct mlx5_modify_hdr *mod_hdr;
1306 mod_hdr = mlx5_modify_header_alloc(priv->mdev,
1307 get_flow_name_space(flow),
1308 mod_hdr_acts->num_actions,
1309 mod_hdr_acts->actions);
1310 if (IS_ERR(mod_hdr))
1311 return PTR_ERR(mod_hdr);
1313 WARN_ON(flow->attr->modify_hdr);
1314 flow->attr->modify_hdr = mod_hdr;
1320 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
1321 struct mlx5e_tc_flow *flow,
1322 struct netlink_ext_ack *extack)
1324 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1325 struct net_device *out_dev, *encap_dev = NULL;
1326 struct mlx5e_tc_flow_parse_attr *parse_attr;
1327 struct mlx5_flow_attr *attr = flow->attr;
1328 bool vf_tun = false, encap_valid = true;
1329 struct mlx5_esw_flow_attr *esw_attr;
1330 struct mlx5_fc *counter = NULL;
1331 struct mlx5e_rep_priv *rpriv;
1332 struct mlx5e_priv *out_priv;
1333 u32 max_prio, max_chain;
1337 /* We check chain range only for tc flows.
1338 * For ft flows, we checked attr->chain was originally 0 and set it to
1339 * FDB_FT_CHAIN which is outside tc range.
1340 * See mlx5e_rep_setup_ft_cb().
1342 max_chain = mlx5_chains_get_chain_range(esw_chains(esw));
1343 if (!mlx5e_is_ft_flow(flow) && attr->chain > max_chain) {
1344 NL_SET_ERR_MSG_MOD(extack,
1345 "Requested chain is out of supported range");
1350 max_prio = mlx5_chains_get_prio_range(esw_chains(esw));
1351 if (attr->prio > max_prio) {
1352 NL_SET_ERR_MSG_MOD(extack,
1353 "Requested priority is out of supported range");
1358 if (flow_flag_test(flow, TUN_RX)) {
1359 err = mlx5e_attach_decap_route(priv, flow);
1364 if (flow_flag_test(flow, L3_TO_L2_DECAP)) {
1365 err = mlx5e_attach_decap(priv, flow, extack);
1370 parse_attr = attr->parse_attr;
1371 esw_attr = attr->esw_attr;
1373 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++) {
1376 if (!(esw_attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP))
1379 mirred_ifindex = parse_attr->mirred_ifindex[out_index];
1380 out_dev = __dev_get_by_index(dev_net(priv->netdev),
1382 err = mlx5e_attach_encap(priv, flow, out_dev, out_index,
1383 extack, &encap_dev, &encap_valid);
1387 if (esw_attr->dests[out_index].flags &
1388 MLX5_ESW_DEST_CHAIN_WITH_SRC_PORT_CHANGE)
1390 out_priv = netdev_priv(encap_dev);
1391 rpriv = out_priv->ppriv;
1392 esw_attr->dests[out_index].rep = rpriv->rep;
1393 esw_attr->dests[out_index].mdev = out_priv->mdev;
1396 err = mlx5_eswitch_add_vlan_action(esw, attr);
1400 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR &&
1401 !(attr->ct_attr.ct_action & TCA_CT_ACT_CLEAR)) {
1403 err = mlx5e_tc_add_flow_mod_hdr(priv, parse_attr, flow);
1407 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
1413 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1414 counter = mlx5_fc_create(esw_attr->counter_dev, true);
1415 if (IS_ERR(counter)) {
1416 err = PTR_ERR(counter);
1420 attr->counter = counter;
1423 /* we get here if one of the following takes place:
1424 * (1) there's no error
1425 * (2) there's an encap action and we don't have valid neigh
1428 flow->rule[0] = mlx5e_tc_offload_to_slow_path(esw, flow, &parse_attr->spec);
1430 flow->rule[0] = mlx5e_tc_offload_fdb_rules(esw, flow, &parse_attr->spec, attr);
1432 if (IS_ERR(flow->rule[0])) {
1433 err = PTR_ERR(flow->rule[0]);
1436 flow_flag_set(flow, OFFLOADED);
1441 flow_flag_set(flow, FAILED);
1445 static bool mlx5_flow_has_geneve_opt(struct mlx5e_tc_flow *flow)
1447 struct mlx5_flow_spec *spec = &flow->attr->parse_attr->spec;
1448 void *headers_v = MLX5_ADDR_OF(fte_match_param,
1451 u32 geneve_tlv_opt_0_data = MLX5_GET(fte_match_set_misc3,
1453 geneve_tlv_option_0_data);
1455 return !!geneve_tlv_opt_0_data;
1458 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
1459 struct mlx5e_tc_flow *flow)
1461 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1462 struct mlx5_flow_attr *attr = flow->attr;
1463 struct mlx5_esw_flow_attr *esw_attr;
1464 bool vf_tun = false;
1467 esw_attr = attr->esw_attr;
1468 mlx5e_put_flow_tunnel_id(flow);
1470 if (flow_flag_test(flow, NOT_READY))
1471 remove_unready_flow(flow);
1473 if (mlx5e_is_offloaded_flow(flow)) {
1474 if (flow_flag_test(flow, SLOW))
1475 mlx5e_tc_unoffload_from_slow_path(esw, flow);
1477 mlx5e_tc_unoffload_fdb_rules(esw, flow, attr);
1480 if (mlx5_flow_has_geneve_opt(flow))
1481 mlx5_geneve_tlv_option_del(priv->mdev->geneve);
1483 mlx5_eswitch_del_vlan_action(esw, attr);
1485 if (flow->decap_route)
1486 mlx5e_detach_decap_route(priv, flow);
1488 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++) {
1489 if (esw_attr->dests[out_index].flags &
1490 MLX5_ESW_DEST_CHAIN_WITH_SRC_PORT_CHANGE)
1492 if (esw_attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP) {
1493 mlx5e_detach_encap(priv, flow, out_index);
1494 kfree(attr->parse_attr->tun_info[out_index]);
1498 mlx5_tc_ct_match_del(get_ct_priv(priv), &flow->attr->ct_attr);
1500 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
1501 dealloc_mod_hdr_actions(&attr->parse_attr->mod_hdr_acts);
1502 if (vf_tun && attr->modify_hdr)
1503 mlx5_modify_header_dealloc(priv->mdev, attr->modify_hdr);
1505 mlx5e_detach_mod_hdr(priv, flow);
1507 kvfree(attr->parse_attr);
1508 kvfree(attr->esw_attr->rx_tun_attr);
1510 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
1511 mlx5_fc_destroy(esw_attr->counter_dev, attr->counter);
1513 if (flow_flag_test(flow, L3_TO_L2_DECAP))
1514 mlx5e_detach_decap(priv, flow);
1516 kfree(flow->attr->esw_attr->sample);
1520 struct mlx5_fc *mlx5e_tc_get_counter(struct mlx5e_tc_flow *flow)
1522 return flow->attr->counter;
1525 /* Iterate over tmp_list of flows attached to flow_list head. */
1526 void mlx5e_put_flow_list(struct mlx5e_priv *priv, struct list_head *flow_list)
1528 struct mlx5e_tc_flow *flow, *tmp;
1530 list_for_each_entry_safe(flow, tmp, flow_list, tmp_list)
1531 mlx5e_flow_put(priv, flow);
1534 static void __mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1536 struct mlx5_eswitch *esw = flow->priv->mdev->priv.eswitch;
1538 if (!flow_flag_test(flow, ESWITCH) ||
1539 !flow_flag_test(flow, DUP))
1542 mutex_lock(&esw->offloads.peer_mutex);
1543 list_del(&flow->peer);
1544 mutex_unlock(&esw->offloads.peer_mutex);
1546 flow_flag_clear(flow, DUP);
1548 if (refcount_dec_and_test(&flow->peer_flow->refcnt)) {
1549 mlx5e_tc_del_fdb_flow(flow->peer_flow->priv, flow->peer_flow);
1550 kfree(flow->peer_flow);
1553 flow->peer_flow = NULL;
1556 static void mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1558 struct mlx5_core_dev *dev = flow->priv->mdev;
1559 struct mlx5_devcom *devcom = dev->priv.devcom;
1560 struct mlx5_eswitch *peer_esw;
1562 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1566 __mlx5e_tc_del_fdb_peer_flow(flow);
1567 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1570 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1571 struct mlx5e_tc_flow *flow)
1573 if (mlx5e_is_eswitch_flow(flow)) {
1574 mlx5e_tc_del_fdb_peer_flow(flow);
1575 mlx5e_tc_del_fdb_flow(priv, flow);
1577 mlx5e_tc_del_nic_flow(priv, flow);
1581 static int flow_has_tc_fwd_action(struct flow_cls_offload *f)
1583 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1584 struct flow_action *flow_action = &rule->action;
1585 const struct flow_action_entry *act;
1588 flow_action_for_each(i, act, flow_action) {
1590 case FLOW_ACTION_GOTO:
1601 enc_opts_is_dont_care_or_full_match(struct mlx5e_priv *priv,
1602 struct flow_dissector_key_enc_opts *opts,
1603 struct netlink_ext_ack *extack,
1606 struct geneve_opt *opt;
1611 while (opts->len > off) {
1612 opt = (struct geneve_opt *)&opts->data[off];
1614 if (!(*dont_care) || opt->opt_class || opt->type ||
1615 memchr_inv(opt->opt_data, 0, opt->length * 4)) {
1618 if (opt->opt_class != htons(U16_MAX) ||
1619 opt->type != U8_MAX) {
1620 NL_SET_ERR_MSG(extack,
1621 "Partial match of tunnel options in chain > 0 isn't supported");
1622 netdev_warn(priv->netdev,
1623 "Partial match of tunnel options in chain > 0 isn't supported");
1628 off += sizeof(struct geneve_opt) + opt->length * 4;
1634 #define COPY_DISSECTOR(rule, diss_key, dst)\
1636 struct flow_rule *__rule = (rule);\
1637 typeof(dst) __dst = dst;\
1640 skb_flow_dissector_target(__rule->match.dissector,\
1642 __rule->match.key),\
1646 static int mlx5e_get_flow_tunnel_id(struct mlx5e_priv *priv,
1647 struct mlx5e_tc_flow *flow,
1648 struct flow_cls_offload *f,
1649 struct net_device *filter_dev)
1651 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1652 struct netlink_ext_ack *extack = f->common.extack;
1653 struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts;
1654 struct flow_match_enc_opts enc_opts_match;
1655 struct tunnel_match_enc_opts tun_enc_opts;
1656 struct mlx5_rep_uplink_priv *uplink_priv;
1657 struct mlx5_flow_attr *attr = flow->attr;
1658 struct mlx5e_rep_priv *uplink_rpriv;
1659 struct tunnel_match_key tunnel_key;
1660 bool enc_opts_is_dont_care = true;
1661 u32 tun_id, enc_opts_id = 0;
1662 struct mlx5_eswitch *esw;
1666 esw = priv->mdev->priv.eswitch;
1667 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1668 uplink_priv = &uplink_rpriv->uplink_priv;
1670 memset(&tunnel_key, 0, sizeof(tunnel_key));
1671 COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL,
1672 &tunnel_key.enc_control);
1673 if (tunnel_key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS)
1674 COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1675 &tunnel_key.enc_ipv4);
1677 COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1678 &tunnel_key.enc_ipv6);
1679 COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_IP, &tunnel_key.enc_ip);
1680 COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_PORTS,
1681 &tunnel_key.enc_tp);
1682 COPY_DISSECTOR(rule, FLOW_DISSECTOR_KEY_ENC_KEYID,
1683 &tunnel_key.enc_key_id);
1684 tunnel_key.filter_ifindex = filter_dev->ifindex;
1686 err = mapping_add(uplink_priv->tunnel_mapping, &tunnel_key, &tun_id);
1690 flow_rule_match_enc_opts(rule, &enc_opts_match);
1691 err = enc_opts_is_dont_care_or_full_match(priv,
1692 enc_opts_match.mask,
1694 &enc_opts_is_dont_care);
1698 if (!enc_opts_is_dont_care) {
1699 memset(&tun_enc_opts, 0, sizeof(tun_enc_opts));
1700 memcpy(&tun_enc_opts.key, enc_opts_match.key,
1701 sizeof(*enc_opts_match.key));
1702 memcpy(&tun_enc_opts.mask, enc_opts_match.mask,
1703 sizeof(*enc_opts_match.mask));
1705 err = mapping_add(uplink_priv->tunnel_enc_opts_mapping,
1706 &tun_enc_opts, &enc_opts_id);
1711 value = tun_id << ENC_OPTS_BITS | enc_opts_id;
1712 mask = enc_opts_id ? TUNNEL_ID_MASK :
1713 (TUNNEL_ID_MASK & ~ENC_OPTS_BITS_MASK);
1716 mlx5e_tc_match_to_reg_match(&attr->parse_attr->spec,
1717 TUNNEL_TO_REG, value, mask);
1719 mod_hdr_acts = &attr->parse_attr->mod_hdr_acts;
1720 err = mlx5e_tc_match_to_reg_set(priv->mdev,
1721 mod_hdr_acts, MLX5_FLOW_NAMESPACE_FDB,
1722 TUNNEL_TO_REG, value);
1726 attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
1729 flow->tunnel_id = value;
1734 mapping_remove(uplink_priv->tunnel_enc_opts_mapping,
1737 mapping_remove(uplink_priv->tunnel_mapping, tun_id);
1741 static void mlx5e_put_flow_tunnel_id(struct mlx5e_tc_flow *flow)
1743 u32 enc_opts_id = flow->tunnel_id & ENC_OPTS_BITS_MASK;
1744 u32 tun_id = flow->tunnel_id >> ENC_OPTS_BITS;
1745 struct mlx5_rep_uplink_priv *uplink_priv;
1746 struct mlx5e_rep_priv *uplink_rpriv;
1747 struct mlx5_eswitch *esw;
1749 esw = flow->priv->mdev->priv.eswitch;
1750 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1751 uplink_priv = &uplink_rpriv->uplink_priv;
1754 mapping_remove(uplink_priv->tunnel_mapping, tun_id);
1756 mapping_remove(uplink_priv->tunnel_enc_opts_mapping,
1760 u32 mlx5e_tc_get_flow_tun_id(struct mlx5e_tc_flow *flow)
1762 return flow->tunnel_id;
1765 void mlx5e_tc_set_ethertype(struct mlx5_core_dev *mdev,
1766 struct flow_match_basic *match, bool outer,
1767 void *headers_c, void *headers_v)
1769 bool ip_version_cap;
1771 ip_version_cap = outer ?
1772 MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
1773 ft_field_support.outer_ip_version) :
1774 MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
1775 ft_field_support.inner_ip_version);
1777 if (ip_version_cap && match->mask->n_proto == htons(0xFFFF) &&
1778 (match->key->n_proto == htons(ETH_P_IP) ||
1779 match->key->n_proto == htons(ETH_P_IPV6))) {
1780 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_version);
1781 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version,
1782 match->key->n_proto == htons(ETH_P_IP) ? 4 : 6);
1784 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1785 ntohs(match->mask->n_proto));
1786 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1787 ntohs(match->key->n_proto));
1791 u8 mlx5e_tc_get_ip_version(struct mlx5_flow_spec *spec, bool outer)
1798 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
1800 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, inner_headers);
1802 ip_version = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_version);
1803 /* Return ip_version converted from ethertype anyway */
1805 ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
1806 if (ethertype == ETH_P_IP || ethertype == ETH_P_ARP)
1808 else if (ethertype == ETH_P_IPV6)
1814 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1815 struct mlx5e_tc_flow *flow,
1816 struct mlx5_flow_spec *spec,
1817 struct flow_cls_offload *f,
1818 struct net_device *filter_dev,
1822 struct mlx5e_tc_tunnel *tunnel = mlx5e_get_tc_tun(filter_dev);
1823 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1824 struct netlink_ext_ack *extack = f->common.extack;
1825 bool needs_mapping, sets_mapping;
1828 if (!mlx5e_is_eswitch_flow(flow))
1831 needs_mapping = !!flow->attr->chain;
1832 sets_mapping = !flow->attr->chain && flow_has_tc_fwd_action(f);
1833 *match_inner = !needs_mapping;
1835 if ((needs_mapping || sets_mapping) &&
1836 !mlx5_eswitch_reg_c1_loopback_enabled(esw)) {
1837 NL_SET_ERR_MSG(extack,
1838 "Chains on tunnel devices isn't supported without register loopback support");
1839 netdev_warn(priv->netdev,
1840 "Chains on tunnel devices isn't supported without register loopback support");
1844 if (!flow->attr->chain) {
1845 err = mlx5e_tc_tun_parse(filter_dev, priv, spec, f,
1848 NL_SET_ERR_MSG_MOD(extack,
1849 "Failed to parse tunnel attributes");
1850 netdev_warn(priv->netdev,
1851 "Failed to parse tunnel attributes");
1855 /* With mpls over udp we decapsulate using packet reformat
1858 if (!netif_is_bareudp(filter_dev))
1859 flow->attr->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
1860 err = mlx5e_tc_set_attr_rx_tun(flow, spec);
1863 } else if (tunnel && tunnel->tunnel_type == MLX5E_TC_TUNNEL_TYPE_VXLAN) {
1864 struct mlx5_flow_spec *tmp_spec;
1866 tmp_spec = kvzalloc(sizeof(*tmp_spec), GFP_KERNEL);
1868 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate memory for vxlan tmp spec");
1869 netdev_warn(priv->netdev, "Failed to allocate memory for vxlan tmp spec");
1872 memcpy(tmp_spec, spec, sizeof(*tmp_spec));
1874 err = mlx5e_tc_tun_parse(filter_dev, priv, tmp_spec, f, match_level);
1877 NL_SET_ERR_MSG_MOD(extack, "Failed to parse tunnel attributes");
1878 netdev_warn(priv->netdev, "Failed to parse tunnel attributes");
1881 err = mlx5e_tc_set_attr_rx_tun(flow, tmp_spec);
1887 if (!needs_mapping && !sets_mapping)
1890 return mlx5e_get_flow_tunnel_id(priv, flow, f, filter_dev);
1893 static void *get_match_inner_headers_criteria(struct mlx5_flow_spec *spec)
1895 return MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1899 static void *get_match_inner_headers_value(struct mlx5_flow_spec *spec)
1901 return MLX5_ADDR_OF(fte_match_param, spec->match_value,
1905 static void *get_match_outer_headers_criteria(struct mlx5_flow_spec *spec)
1907 return MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1911 static void *get_match_outer_headers_value(struct mlx5_flow_spec *spec)
1913 return MLX5_ADDR_OF(fte_match_param, spec->match_value,
1917 static void *get_match_headers_value(u32 flags,
1918 struct mlx5_flow_spec *spec)
1920 return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
1921 get_match_inner_headers_value(spec) :
1922 get_match_outer_headers_value(spec);
1925 static void *get_match_headers_criteria(u32 flags,
1926 struct mlx5_flow_spec *spec)
1928 return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
1929 get_match_inner_headers_criteria(spec) :
1930 get_match_outer_headers_criteria(spec);
1933 static int mlx5e_flower_parse_meta(struct net_device *filter_dev,
1934 struct flow_cls_offload *f)
1936 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1937 struct netlink_ext_ack *extack = f->common.extack;
1938 struct net_device *ingress_dev;
1939 struct flow_match_meta match;
1941 if (!flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_META))
1944 flow_rule_match_meta(rule, &match);
1945 if (!match.mask->ingress_ifindex)
1948 if (match.mask->ingress_ifindex != 0xFFFFFFFF) {
1949 NL_SET_ERR_MSG_MOD(extack, "Unsupported ingress ifindex mask");
1953 ingress_dev = __dev_get_by_index(dev_net(filter_dev),
1954 match.key->ingress_ifindex);
1956 NL_SET_ERR_MSG_MOD(extack,
1957 "Can't find the ingress port to match on");
1961 if (ingress_dev != filter_dev) {
1962 NL_SET_ERR_MSG_MOD(extack,
1963 "Can't match on the ingress filter port");
1970 static bool skip_key_basic(struct net_device *filter_dev,
1971 struct flow_cls_offload *f)
1973 /* When doing mpls over udp decap, the user needs to provide
1974 * MPLS_UC as the protocol in order to be able to match on mpls
1975 * label fields. However, the actual ethertype is IP so we want to
1976 * avoid matching on this, otherwise we'll fail the match.
1978 if (netif_is_bareudp(filter_dev) && f->common.chain_index == 0)
1984 static int __parse_cls_flower(struct mlx5e_priv *priv,
1985 struct mlx5e_tc_flow *flow,
1986 struct mlx5_flow_spec *spec,
1987 struct flow_cls_offload *f,
1988 struct net_device *filter_dev,
1989 u8 *inner_match_level, u8 *outer_match_level)
1991 struct netlink_ext_ack *extack = f->common.extack;
1992 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1994 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1996 void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1998 void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
2000 void *misc_c_3 = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
2002 void *misc_v_3 = MLX5_ADDR_OF(fte_match_param, spec->match_value,
2004 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
2005 struct flow_dissector *dissector = rule->match.dissector;
2011 match_level = outer_match_level;
2013 if (dissector->used_keys &
2014 ~(BIT(FLOW_DISSECTOR_KEY_META) |
2015 BIT(FLOW_DISSECTOR_KEY_CONTROL) |
2016 BIT(FLOW_DISSECTOR_KEY_BASIC) |
2017 BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
2018 BIT(FLOW_DISSECTOR_KEY_VLAN) |
2019 BIT(FLOW_DISSECTOR_KEY_CVLAN) |
2020 BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
2021 BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
2022 BIT(FLOW_DISSECTOR_KEY_PORTS) |
2023 BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
2024 BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
2025 BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
2026 BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
2027 BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
2028 BIT(FLOW_DISSECTOR_KEY_TCP) |
2029 BIT(FLOW_DISSECTOR_KEY_IP) |
2030 BIT(FLOW_DISSECTOR_KEY_CT) |
2031 BIT(FLOW_DISSECTOR_KEY_ENC_IP) |
2032 BIT(FLOW_DISSECTOR_KEY_ENC_OPTS) |
2033 BIT(FLOW_DISSECTOR_KEY_ICMP) |
2034 BIT(FLOW_DISSECTOR_KEY_MPLS))) {
2035 NL_SET_ERR_MSG_MOD(extack, "Unsupported key");
2036 netdev_dbg(priv->netdev, "Unsupported key used: 0x%x\n",
2037 dissector->used_keys);
2041 if (mlx5e_get_tc_tun(filter_dev)) {
2042 bool match_inner = false;
2044 err = parse_tunnel_attr(priv, flow, spec, f, filter_dev,
2045 outer_match_level, &match_inner);
2050 /* header pointers should point to the inner headers
2051 * if the packet was decapsulated already.
2052 * outer headers are set by parse_tunnel_attr.
2054 match_level = inner_match_level;
2055 headers_c = get_match_inner_headers_criteria(spec);
2056 headers_v = get_match_inner_headers_value(spec);
2060 err = mlx5e_flower_parse_meta(filter_dev, f);
2064 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC) &&
2065 !skip_key_basic(filter_dev, f)) {
2066 struct flow_match_basic match;
2068 flow_rule_match_basic(rule, &match);
2069 mlx5e_tc_set_ethertype(priv->mdev, &match,
2070 match_level == outer_match_level,
2071 headers_c, headers_v);
2073 if (match.mask->n_proto)
2074 *match_level = MLX5_MATCH_L2;
2076 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN) ||
2077 is_vlan_dev(filter_dev)) {
2078 struct flow_dissector_key_vlan filter_dev_mask;
2079 struct flow_dissector_key_vlan filter_dev_key;
2080 struct flow_match_vlan match;
2082 if (is_vlan_dev(filter_dev)) {
2083 match.key = &filter_dev_key;
2084 match.key->vlan_id = vlan_dev_vlan_id(filter_dev);
2085 match.key->vlan_tpid = vlan_dev_vlan_proto(filter_dev);
2086 match.key->vlan_priority = 0;
2087 match.mask = &filter_dev_mask;
2088 memset(match.mask, 0xff, sizeof(*match.mask));
2089 match.mask->vlan_priority = 0;
2091 flow_rule_match_vlan(rule, &match);
2093 if (match.mask->vlan_id ||
2094 match.mask->vlan_priority ||
2095 match.mask->vlan_tpid) {
2096 if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
2097 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2099 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2102 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2104 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2108 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid,
2109 match.mask->vlan_id);
2110 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid,
2111 match.key->vlan_id);
2113 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio,
2114 match.mask->vlan_priority);
2115 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio,
2116 match.key->vlan_priority);
2118 *match_level = MLX5_MATCH_L2;
2120 } else if (*match_level != MLX5_MATCH_NONE) {
2121 /* cvlan_tag enabled in match criteria and
2122 * disabled in match value means both S & C tags
2123 * don't exist (untagged of both)
2125 MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
2126 *match_level = MLX5_MATCH_L2;
2129 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
2130 struct flow_match_vlan match;
2132 flow_rule_match_cvlan(rule, &match);
2133 if (match.mask->vlan_id ||
2134 match.mask->vlan_priority ||
2135 match.mask->vlan_tpid) {
2136 if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
2137 MLX5_SET(fte_match_set_misc, misc_c,
2138 outer_second_svlan_tag, 1);
2139 MLX5_SET(fte_match_set_misc, misc_v,
2140 outer_second_svlan_tag, 1);
2142 MLX5_SET(fte_match_set_misc, misc_c,
2143 outer_second_cvlan_tag, 1);
2144 MLX5_SET(fte_match_set_misc, misc_v,
2145 outer_second_cvlan_tag, 1);
2148 MLX5_SET(fte_match_set_misc, misc_c, outer_second_vid,
2149 match.mask->vlan_id);
2150 MLX5_SET(fte_match_set_misc, misc_v, outer_second_vid,
2151 match.key->vlan_id);
2152 MLX5_SET(fte_match_set_misc, misc_c, outer_second_prio,
2153 match.mask->vlan_priority);
2154 MLX5_SET(fte_match_set_misc, misc_v, outer_second_prio,
2155 match.key->vlan_priority);
2157 *match_level = MLX5_MATCH_L2;
2158 spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS;
2162 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2163 struct flow_match_eth_addrs match;
2165 flow_rule_match_eth_addrs(rule, &match);
2166 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2169 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2173 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2176 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2180 if (!is_zero_ether_addr(match.mask->src) ||
2181 !is_zero_ether_addr(match.mask->dst))
2182 *match_level = MLX5_MATCH_L2;
2185 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
2186 struct flow_match_control match;
2188 flow_rule_match_control(rule, &match);
2189 addr_type = match.key->addr_type;
2191 /* the HW doesn't support frag first/later */
2192 if (match.mask->flags & FLOW_DIS_FIRST_FRAG)
2195 if (match.mask->flags & FLOW_DIS_IS_FRAGMENT) {
2196 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
2197 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
2198 match.key->flags & FLOW_DIS_IS_FRAGMENT);
2200 /* the HW doesn't need L3 inline to match on frag=no */
2201 if (!(match.key->flags & FLOW_DIS_IS_FRAGMENT))
2202 *match_level = MLX5_MATCH_L2;
2203 /* *** L2 attributes parsing up to here *** */
2205 *match_level = MLX5_MATCH_L3;
2209 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2210 struct flow_match_basic match;
2212 flow_rule_match_basic(rule, &match);
2213 ip_proto = match.key->ip_proto;
2215 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
2216 match.mask->ip_proto);
2217 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
2218 match.key->ip_proto);
2220 if (match.mask->ip_proto)
2221 *match_level = MLX5_MATCH_L3;
2224 if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2225 struct flow_match_ipv4_addrs match;
2227 flow_rule_match_ipv4_addrs(rule, &match);
2228 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2229 src_ipv4_src_ipv6.ipv4_layout.ipv4),
2230 &match.mask->src, sizeof(match.mask->src));
2231 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2232 src_ipv4_src_ipv6.ipv4_layout.ipv4),
2233 &match.key->src, sizeof(match.key->src));
2234 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2235 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2236 &match.mask->dst, sizeof(match.mask->dst));
2237 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2238 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2239 &match.key->dst, sizeof(match.key->dst));
2241 if (match.mask->src || match.mask->dst)
2242 *match_level = MLX5_MATCH_L3;
2245 if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2246 struct flow_match_ipv6_addrs match;
2248 flow_rule_match_ipv6_addrs(rule, &match);
2249 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2250 src_ipv4_src_ipv6.ipv6_layout.ipv6),
2251 &match.mask->src, sizeof(match.mask->src));
2252 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2253 src_ipv4_src_ipv6.ipv6_layout.ipv6),
2254 &match.key->src, sizeof(match.key->src));
2256 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2257 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2258 &match.mask->dst, sizeof(match.mask->dst));
2259 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2260 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2261 &match.key->dst, sizeof(match.key->dst));
2263 if (ipv6_addr_type(&match.mask->src) != IPV6_ADDR_ANY ||
2264 ipv6_addr_type(&match.mask->dst) != IPV6_ADDR_ANY)
2265 *match_level = MLX5_MATCH_L3;
2268 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
2269 struct flow_match_ip match;
2271 flow_rule_match_ip(rule, &match);
2272 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn,
2273 match.mask->tos & 0x3);
2274 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn,
2275 match.key->tos & 0x3);
2277 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp,
2278 match.mask->tos >> 2);
2279 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp,
2280 match.key->tos >> 2);
2282 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit,
2284 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit,
2287 if (match.mask->ttl &&
2288 !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
2289 ft_field_support.outer_ipv4_ttl)) {
2290 NL_SET_ERR_MSG_MOD(extack,
2291 "Matching on TTL is not supported");
2295 if (match.mask->tos || match.mask->ttl)
2296 *match_level = MLX5_MATCH_L3;
2299 /* *** L3 attributes parsing up to here *** */
2301 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) {
2302 struct flow_match_ports match;
2304 flow_rule_match_ports(rule, &match);
2307 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2308 tcp_sport, ntohs(match.mask->src));
2309 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2310 tcp_sport, ntohs(match.key->src));
2312 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2313 tcp_dport, ntohs(match.mask->dst));
2314 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2315 tcp_dport, ntohs(match.key->dst));
2319 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2320 udp_sport, ntohs(match.mask->src));
2321 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2322 udp_sport, ntohs(match.key->src));
2324 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2325 udp_dport, ntohs(match.mask->dst));
2326 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2327 udp_dport, ntohs(match.key->dst));
2330 NL_SET_ERR_MSG_MOD(extack,
2331 "Only UDP and TCP transports are supported for L4 matching");
2332 netdev_err(priv->netdev,
2333 "Only UDP and TCP transport are supported\n");
2337 if (match.mask->src || match.mask->dst)
2338 *match_level = MLX5_MATCH_L4;
2341 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_TCP)) {
2342 struct flow_match_tcp match;
2344 flow_rule_match_tcp(rule, &match);
2345 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
2346 ntohs(match.mask->flags));
2347 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
2348 ntohs(match.key->flags));
2350 if (match.mask->flags)
2351 *match_level = MLX5_MATCH_L4;
2353 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ICMP)) {
2354 struct flow_match_icmp match;
2356 flow_rule_match_icmp(rule, &match);
2359 if (!(MLX5_CAP_GEN(priv->mdev, flex_parser_protocols) &
2360 MLX5_FLEX_PROTO_ICMP))
2362 MLX5_SET(fte_match_set_misc3, misc_c_3, icmp_type,
2364 MLX5_SET(fte_match_set_misc3, misc_v_3, icmp_type,
2366 MLX5_SET(fte_match_set_misc3, misc_c_3, icmp_code,
2368 MLX5_SET(fte_match_set_misc3, misc_v_3, icmp_code,
2371 case IPPROTO_ICMPV6:
2372 if (!(MLX5_CAP_GEN(priv->mdev, flex_parser_protocols) &
2373 MLX5_FLEX_PROTO_ICMPV6))
2375 MLX5_SET(fte_match_set_misc3, misc_c_3, icmpv6_type,
2377 MLX5_SET(fte_match_set_misc3, misc_v_3, icmpv6_type,
2379 MLX5_SET(fte_match_set_misc3, misc_c_3, icmpv6_code,
2381 MLX5_SET(fte_match_set_misc3, misc_v_3, icmpv6_code,
2385 NL_SET_ERR_MSG_MOD(extack,
2386 "Code and type matching only with ICMP and ICMPv6");
2387 netdev_err(priv->netdev,
2388 "Code and type matching only with ICMP and ICMPv6\n");
2391 if (match.mask->code || match.mask->type) {
2392 *match_level = MLX5_MATCH_L4;
2393 spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS_3;
2396 /* Currenlty supported only for MPLS over UDP */
2397 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_MPLS) &&
2398 !netif_is_bareudp(filter_dev)) {
2399 NL_SET_ERR_MSG_MOD(extack,
2400 "Matching on MPLS is supported only for MPLS over UDP");
2401 netdev_err(priv->netdev,
2402 "Matching on MPLS is supported only for MPLS over UDP\n");
2409 static int parse_cls_flower(struct mlx5e_priv *priv,
2410 struct mlx5e_tc_flow *flow,
2411 struct mlx5_flow_spec *spec,
2412 struct flow_cls_offload *f,
2413 struct net_device *filter_dev)
2415 u8 inner_match_level, outer_match_level, non_tunnel_match_level;
2416 struct netlink_ext_ack *extack = f->common.extack;
2417 struct mlx5_core_dev *dev = priv->mdev;
2418 struct mlx5_eswitch *esw = dev->priv.eswitch;
2419 struct mlx5e_rep_priv *rpriv = priv->ppriv;
2420 struct mlx5_eswitch_rep *rep;
2421 bool is_eswitch_flow;
2424 inner_match_level = MLX5_MATCH_NONE;
2425 outer_match_level = MLX5_MATCH_NONE;
2427 err = __parse_cls_flower(priv, flow, spec, f, filter_dev,
2428 &inner_match_level, &outer_match_level);
2429 non_tunnel_match_level = (inner_match_level == MLX5_MATCH_NONE) ?
2430 outer_match_level : inner_match_level;
2432 is_eswitch_flow = mlx5e_is_eswitch_flow(flow);
2433 if (!err && is_eswitch_flow) {
2435 if (rep->vport != MLX5_VPORT_UPLINK &&
2436 (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
2437 esw->offloads.inline_mode < non_tunnel_match_level)) {
2438 NL_SET_ERR_MSG_MOD(extack,
2439 "Flow is not offloaded due to min inline setting");
2440 netdev_warn(priv->netdev,
2441 "Flow is not offloaded due to min inline setting, required %d actual %d\n",
2442 non_tunnel_match_level, esw->offloads.inline_mode);
2447 flow->attr->inner_match_level = inner_match_level;
2448 flow->attr->outer_match_level = outer_match_level;
2454 struct pedit_headers {
2456 struct vlan_hdr vlan;
2463 struct pedit_headers_action {
2464 struct pedit_headers vals;
2465 struct pedit_headers masks;
2469 static int pedit_header_offsets[] = {
2470 [FLOW_ACT_MANGLE_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
2471 [FLOW_ACT_MANGLE_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
2472 [FLOW_ACT_MANGLE_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
2473 [FLOW_ACT_MANGLE_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
2474 [FLOW_ACT_MANGLE_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
2477 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
2479 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
2480 struct pedit_headers_action *hdrs)
2482 u32 *curr_pmask, *curr_pval;
2484 curr_pmask = (u32 *)(pedit_header(&hdrs->masks, hdr_type) + offset);
2485 curr_pval = (u32 *)(pedit_header(&hdrs->vals, hdr_type) + offset);
2487 if (*curr_pmask & mask) /* disallow acting twice on the same location */
2490 *curr_pmask |= mask;
2491 *curr_pval |= (val & mask);
2499 struct mlx5_fields {
2507 #define OFFLOAD(fw_field, field_bsize, field_mask, field, off, match_field) \
2508 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, field_bsize, field_mask, \
2509 offsetof(struct pedit_headers, field) + (off), \
2510 MLX5_BYTE_OFF(fte_match_set_lyr_2_4, match_field)}
2512 /* masked values are the same and there are no rewrites that do not have a
2515 #define SAME_VAL_MASK(type, valp, maskp, matchvalp, matchmaskp) ({ \
2516 type matchmaskx = *(type *)(matchmaskp); \
2517 type matchvalx = *(type *)(matchvalp); \
2518 type maskx = *(type *)(maskp); \
2519 type valx = *(type *)(valp); \
2521 (valx & maskx) == (matchvalx & matchmaskx) && !(maskx & (maskx ^ \
2525 static bool cmp_val_mask(void *valp, void *maskp, void *matchvalp,
2526 void *matchmaskp, u8 bsize)
2532 same = SAME_VAL_MASK(u8, valp, maskp, matchvalp, matchmaskp);
2535 same = SAME_VAL_MASK(u16, valp, maskp, matchvalp, matchmaskp);
2538 same = SAME_VAL_MASK(u32, valp, maskp, matchvalp, matchmaskp);
2545 static struct mlx5_fields fields[] = {
2546 OFFLOAD(DMAC_47_16, 32, U32_MAX, eth.h_dest[0], 0, dmac_47_16),
2547 OFFLOAD(DMAC_15_0, 16, U16_MAX, eth.h_dest[4], 0, dmac_15_0),
2548 OFFLOAD(SMAC_47_16, 32, U32_MAX, eth.h_source[0], 0, smac_47_16),
2549 OFFLOAD(SMAC_15_0, 16, U16_MAX, eth.h_source[4], 0, smac_15_0),
2550 OFFLOAD(ETHERTYPE, 16, U16_MAX, eth.h_proto, 0, ethertype),
2551 OFFLOAD(FIRST_VID, 16, U16_MAX, vlan.h_vlan_TCI, 0, first_vid),
2553 OFFLOAD(IP_DSCP, 8, 0xfc, ip4.tos, 0, ip_dscp),
2554 OFFLOAD(IP_TTL, 8, U8_MAX, ip4.ttl, 0, ttl_hoplimit),
2555 OFFLOAD(SIPV4, 32, U32_MAX, ip4.saddr, 0, src_ipv4_src_ipv6.ipv4_layout.ipv4),
2556 OFFLOAD(DIPV4, 32, U32_MAX, ip4.daddr, 0, dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2558 OFFLOAD(SIPV6_127_96, 32, U32_MAX, ip6.saddr.s6_addr32[0], 0,
2559 src_ipv4_src_ipv6.ipv6_layout.ipv6[0]),
2560 OFFLOAD(SIPV6_95_64, 32, U32_MAX, ip6.saddr.s6_addr32[1], 0,
2561 src_ipv4_src_ipv6.ipv6_layout.ipv6[4]),
2562 OFFLOAD(SIPV6_63_32, 32, U32_MAX, ip6.saddr.s6_addr32[2], 0,
2563 src_ipv4_src_ipv6.ipv6_layout.ipv6[8]),
2564 OFFLOAD(SIPV6_31_0, 32, U32_MAX, ip6.saddr.s6_addr32[3], 0,
2565 src_ipv4_src_ipv6.ipv6_layout.ipv6[12]),
2566 OFFLOAD(DIPV6_127_96, 32, U32_MAX, ip6.daddr.s6_addr32[0], 0,
2567 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[0]),
2568 OFFLOAD(DIPV6_95_64, 32, U32_MAX, ip6.daddr.s6_addr32[1], 0,
2569 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[4]),
2570 OFFLOAD(DIPV6_63_32, 32, U32_MAX, ip6.daddr.s6_addr32[2], 0,
2571 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[8]),
2572 OFFLOAD(DIPV6_31_0, 32, U32_MAX, ip6.daddr.s6_addr32[3], 0,
2573 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[12]),
2574 OFFLOAD(IPV6_HOPLIMIT, 8, U8_MAX, ip6.hop_limit, 0, ttl_hoplimit),
2575 OFFLOAD(IP_DSCP, 16, 0xc00f, ip6, 0, ip_dscp),
2577 OFFLOAD(TCP_SPORT, 16, U16_MAX, tcp.source, 0, tcp_sport),
2578 OFFLOAD(TCP_DPORT, 16, U16_MAX, tcp.dest, 0, tcp_dport),
2579 /* in linux iphdr tcp_flags is 8 bits long */
2580 OFFLOAD(TCP_FLAGS, 8, U8_MAX, tcp.ack_seq, 5, tcp_flags),
2582 OFFLOAD(UDP_SPORT, 16, U16_MAX, udp.source, 0, udp_sport),
2583 OFFLOAD(UDP_DPORT, 16, U16_MAX, udp.dest, 0, udp_dport),
2586 static unsigned long mask_to_le(unsigned long mask, int size)
2592 mask_be32 = (__force __be32)(mask);
2593 mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
2594 } else if (size == 16) {
2595 mask_be32 = (__force __be32)(mask);
2596 mask_be16 = *(__be16 *)&mask_be32;
2597 mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
2602 static int offload_pedit_fields(struct mlx5e_priv *priv,
2604 struct pedit_headers_action *hdrs,
2605 struct mlx5e_tc_flow_parse_attr *parse_attr,
2607 struct netlink_ext_ack *extack)
2609 struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
2610 int i, action_size, first, last, next_z;
2611 void *headers_c, *headers_v, *action, *vals_p;
2612 u32 *s_masks_p, *a_masks_p, s_mask, a_mask;
2613 struct mlx5e_tc_mod_hdr_acts *mod_acts;
2614 struct mlx5_fields *f;
2615 unsigned long mask, field_mask;
2619 mod_acts = &parse_attr->mod_hdr_acts;
2620 headers_c = get_match_headers_criteria(*action_flags, &parse_attr->spec);
2621 headers_v = get_match_headers_value(*action_flags, &parse_attr->spec);
2623 set_masks = &hdrs[0].masks;
2624 add_masks = &hdrs[1].masks;
2625 set_vals = &hdrs[0].vals;
2626 add_vals = &hdrs[1].vals;
2628 action_size = MLX5_UN_SZ_BYTES(set_add_copy_action_in_auto);
2630 for (i = 0; i < ARRAY_SIZE(fields); i++) {
2634 /* avoid seeing bits set from previous iterations */
2638 s_masks_p = (void *)set_masks + f->offset;
2639 a_masks_p = (void *)add_masks + f->offset;
2641 s_mask = *s_masks_p & f->field_mask;
2642 a_mask = *a_masks_p & f->field_mask;
2644 if (!s_mask && !a_mask) /* nothing to offload here */
2647 if (s_mask && a_mask) {
2648 NL_SET_ERR_MSG_MOD(extack,
2649 "can't set and add to the same HW field");
2650 printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
2656 void *match_mask = headers_c + f->match_offset;
2657 void *match_val = headers_v + f->match_offset;
2659 cmd = MLX5_ACTION_TYPE_SET;
2661 vals_p = (void *)set_vals + f->offset;
2662 /* don't rewrite if we have a match on the same value */
2663 if (cmp_val_mask(vals_p, s_masks_p, match_val,
2664 match_mask, f->field_bsize))
2666 /* clear to denote we consumed this field */
2667 *s_masks_p &= ~f->field_mask;
2669 cmd = MLX5_ACTION_TYPE_ADD;
2671 vals_p = (void *)add_vals + f->offset;
2672 /* add 0 is no change */
2673 if ((*(u32 *)vals_p & f->field_mask) == 0)
2675 /* clear to denote we consumed this field */
2676 *a_masks_p &= ~f->field_mask;
2681 mask = mask_to_le(mask, f->field_bsize);
2683 first = find_first_bit(&mask, f->field_bsize);
2684 next_z = find_next_zero_bit(&mask, f->field_bsize, first);
2685 last = find_last_bit(&mask, f->field_bsize);
2686 if (first < next_z && next_z < last) {
2687 NL_SET_ERR_MSG_MOD(extack,
2688 "rewrite of few sub-fields isn't supported");
2689 printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
2694 err = alloc_mod_hdr_actions(priv->mdev, namespace, mod_acts);
2696 NL_SET_ERR_MSG_MOD(extack,
2697 "too many pedit actions, can't offload");
2698 mlx5_core_warn(priv->mdev,
2699 "mlx5: parsed %d pedit actions, can't do more\n",
2700 mod_acts->num_actions);
2704 action = mod_acts->actions +
2705 (mod_acts->num_actions * action_size);
2706 MLX5_SET(set_action_in, action, action_type, cmd);
2707 MLX5_SET(set_action_in, action, field, f->field);
2709 if (cmd == MLX5_ACTION_TYPE_SET) {
2712 field_mask = mask_to_le(f->field_mask, f->field_bsize);
2714 /* if field is bit sized it can start not from first bit */
2715 start = find_first_bit(&field_mask, f->field_bsize);
2717 MLX5_SET(set_action_in, action, offset, first - start);
2718 /* length is num of bits to be written, zero means length of 32 */
2719 MLX5_SET(set_action_in, action, length, (last - first + 1));
2722 if (f->field_bsize == 32)
2723 MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
2724 else if (f->field_bsize == 16)
2725 MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
2726 else if (f->field_bsize == 8)
2727 MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
2729 ++mod_acts->num_actions;
2735 static int mlx5e_flow_namespace_max_modify_action(struct mlx5_core_dev *mdev,
2738 if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
2739 return MLX5_CAP_ESW_FLOWTABLE_FDB(mdev, max_modify_header_actions);
2740 else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
2741 return MLX5_CAP_FLOWTABLE_NIC_RX(mdev, max_modify_header_actions);
2744 int alloc_mod_hdr_actions(struct mlx5_core_dev *mdev,
2746 struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts)
2748 int action_size, new_num_actions, max_hw_actions;
2749 size_t new_sz, old_sz;
2752 if (mod_hdr_acts->num_actions < mod_hdr_acts->max_actions)
2755 action_size = MLX5_UN_SZ_BYTES(set_add_copy_action_in_auto);
2757 max_hw_actions = mlx5e_flow_namespace_max_modify_action(mdev,
2759 new_num_actions = min(max_hw_actions,
2760 mod_hdr_acts->actions ?
2761 mod_hdr_acts->max_actions * 2 : 1);
2762 if (mod_hdr_acts->max_actions == new_num_actions)
2765 new_sz = action_size * new_num_actions;
2766 old_sz = mod_hdr_acts->max_actions * action_size;
2767 ret = krealloc(mod_hdr_acts->actions, new_sz, GFP_KERNEL);
2771 memset(ret + old_sz, 0, new_sz - old_sz);
2772 mod_hdr_acts->actions = ret;
2773 mod_hdr_acts->max_actions = new_num_actions;
2778 void dealloc_mod_hdr_actions(struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts)
2780 kfree(mod_hdr_acts->actions);
2781 mod_hdr_acts->actions = NULL;
2782 mod_hdr_acts->num_actions = 0;
2783 mod_hdr_acts->max_actions = 0;
2786 static const struct pedit_headers zero_masks = {};
2789 parse_pedit_to_modify_hdr(struct mlx5e_priv *priv,
2790 const struct flow_action_entry *act, int namespace,
2791 struct mlx5e_tc_flow_parse_attr *parse_attr,
2792 struct pedit_headers_action *hdrs,
2793 struct netlink_ext_ack *extack)
2795 u8 cmd = (act->id == FLOW_ACTION_MANGLE) ? 0 : 1;
2796 int err = -EOPNOTSUPP;
2797 u32 mask, val, offset;
2800 htype = act->mangle.htype;
2801 err = -EOPNOTSUPP; /* can't be all optimistic */
2803 if (htype == FLOW_ACT_MANGLE_UNSPEC) {
2804 NL_SET_ERR_MSG_MOD(extack, "legacy pedit isn't offloaded");
2808 if (!mlx5e_flow_namespace_max_modify_action(priv->mdev, namespace)) {
2809 NL_SET_ERR_MSG_MOD(extack,
2810 "The pedit offload action is not supported");
2814 mask = act->mangle.mask;
2815 val = act->mangle.val;
2816 offset = act->mangle.offset;
2818 err = set_pedit_val(htype, ~mask, val, offset, &hdrs[cmd]);
2830 parse_pedit_to_reformat(struct mlx5e_priv *priv,
2831 const struct flow_action_entry *act,
2832 struct mlx5e_tc_flow_parse_attr *parse_attr,
2833 struct netlink_ext_ack *extack)
2835 u32 mask, val, offset;
2838 if (act->id != FLOW_ACTION_MANGLE)
2841 if (act->mangle.htype != FLOW_ACT_MANGLE_HDR_TYPE_ETH) {
2842 NL_SET_ERR_MSG_MOD(extack, "Only Ethernet modification is supported");
2846 mask = ~act->mangle.mask;
2847 val = act->mangle.val;
2848 offset = act->mangle.offset;
2849 p = (u32 *)&parse_attr->eth;
2850 *(p + (offset >> 2)) |= (val & mask);
2855 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
2856 const struct flow_action_entry *act, int namespace,
2857 struct mlx5e_tc_flow_parse_attr *parse_attr,
2858 struct pedit_headers_action *hdrs,
2859 struct mlx5e_tc_flow *flow,
2860 struct netlink_ext_ack *extack)
2862 if (flow && flow_flag_test(flow, L3_TO_L2_DECAP))
2863 return parse_pedit_to_reformat(priv, act, parse_attr, extack);
2865 return parse_pedit_to_modify_hdr(priv, act, namespace,
2866 parse_attr, hdrs, extack);
2869 static int alloc_tc_pedit_action(struct mlx5e_priv *priv, int namespace,
2870 struct mlx5e_tc_flow_parse_attr *parse_attr,
2871 struct pedit_headers_action *hdrs,
2873 struct netlink_ext_ack *extack)
2875 struct pedit_headers *cmd_masks;
2879 err = offload_pedit_fields(priv, namespace, hdrs, parse_attr,
2880 action_flags, extack);
2882 goto out_dealloc_parsed_actions;
2884 for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
2885 cmd_masks = &hdrs[cmd].masks;
2886 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
2887 NL_SET_ERR_MSG_MOD(extack,
2888 "attempt to offload an unsupported field");
2889 netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
2890 print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
2891 16, 1, cmd_masks, sizeof(zero_masks), true);
2893 goto out_dealloc_parsed_actions;
2899 out_dealloc_parsed_actions:
2900 dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
2904 static bool csum_offload_supported(struct mlx5e_priv *priv,
2907 struct netlink_ext_ack *extack)
2909 u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
2910 TCA_CSUM_UPDATE_FLAG_UDP;
2912 /* The HW recalcs checksums only if re-writing headers */
2913 if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
2914 NL_SET_ERR_MSG_MOD(extack,
2915 "TC csum action is only offloaded with pedit");
2916 netdev_warn(priv->netdev,
2917 "TC csum action is only offloaded with pedit\n");
2921 if (update_flags & ~prot_flags) {
2922 NL_SET_ERR_MSG_MOD(extack,
2923 "can't offload TC csum action for some header/s");
2924 netdev_warn(priv->netdev,
2925 "can't offload TC csum action for some header/s - flags %#x\n",
2933 struct ip_ttl_word {
2939 struct ipv6_hoplimit_word {
2945 static int is_action_keys_supported(const struct flow_action_entry *act,
2946 bool ct_flow, bool *modify_ip_header,
2948 struct netlink_ext_ack *extack)
2953 htype = act->mangle.htype;
2954 offset = act->mangle.offset;
2955 mask = ~act->mangle.mask;
2956 /* For IPv4 & IPv6 header check 4 byte word,
2957 * to determine that modified fields
2958 * are NOT ttl & hop_limit only.
2960 if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP4) {
2961 struct ip_ttl_word *ttl_word =
2962 (struct ip_ttl_word *)&mask;
2964 if (offset != offsetof(struct iphdr, ttl) ||
2965 ttl_word->protocol ||
2967 *modify_ip_header = true;
2970 if (offset >= offsetof(struct iphdr, saddr))
2971 *modify_tuple = true;
2973 if (ct_flow && *modify_tuple) {
2974 NL_SET_ERR_MSG_MOD(extack,
2975 "can't offload re-write of ipv4 address with action ct");
2978 } else if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP6) {
2979 struct ipv6_hoplimit_word *hoplimit_word =
2980 (struct ipv6_hoplimit_word *)&mask;
2982 if (offset != offsetof(struct ipv6hdr, payload_len) ||
2983 hoplimit_word->payload_len ||
2984 hoplimit_word->nexthdr) {
2985 *modify_ip_header = true;
2988 if (ct_flow && offset >= offsetof(struct ipv6hdr, saddr))
2989 *modify_tuple = true;
2991 if (ct_flow && *modify_tuple) {
2992 NL_SET_ERR_MSG_MOD(extack,
2993 "can't offload re-write of ipv6 address with action ct");
2996 } else if (htype == FLOW_ACT_MANGLE_HDR_TYPE_TCP ||
2997 htype == FLOW_ACT_MANGLE_HDR_TYPE_UDP) {
2998 *modify_tuple = true;
3000 NL_SET_ERR_MSG_MOD(extack,
3001 "can't offload re-write of transport header ports with action ct");
3009 static bool modify_tuple_supported(bool modify_tuple, bool ct_clear,
3010 bool ct_flow, struct netlink_ext_ack *extack,
3011 struct mlx5e_priv *priv,
3012 struct mlx5_flow_spec *spec)
3014 if (!modify_tuple || ct_clear)
3018 NL_SET_ERR_MSG_MOD(extack,
3019 "can't offload tuple modification with non-clear ct()");
3020 netdev_info(priv->netdev,
3021 "can't offload tuple modification with non-clear ct()");
3025 /* Add ct_state=-trk match so it will be offloaded for non ct flows
3026 * (or after clear action), as otherwise, since the tuple is changed,
3027 * we can't restore ct state
3029 if (mlx5_tc_ct_add_no_trk_match(spec)) {
3030 NL_SET_ERR_MSG_MOD(extack,
3031 "can't offload tuple modification with ct matches and no ct(clear) action");
3032 netdev_info(priv->netdev,
3033 "can't offload tuple modification with ct matches and no ct(clear) action");
3040 static bool modify_header_match_supported(struct mlx5e_priv *priv,
3041 struct mlx5_flow_spec *spec,
3042 struct flow_action *flow_action,
3043 u32 actions, bool ct_flow,
3045 struct netlink_ext_ack *extack)
3047 const struct flow_action_entry *act;
3048 bool modify_ip_header, modify_tuple;
3055 headers_c = get_match_headers_criteria(actions, spec);
3056 headers_v = get_match_headers_value(actions, spec);
3057 ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
3059 /* for non-IP we only re-write MACs, so we're okay */
3060 if (MLX5_GET(fte_match_set_lyr_2_4, headers_c, ip_version) == 0 &&
3061 ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
3064 modify_ip_header = false;
3065 modify_tuple = false;
3066 flow_action_for_each(i, act, flow_action) {
3067 if (act->id != FLOW_ACTION_MANGLE &&
3068 act->id != FLOW_ACTION_ADD)
3071 err = is_action_keys_supported(act, ct_flow,
3073 &modify_tuple, extack);
3078 if (!modify_tuple_supported(modify_tuple, ct_clear, ct_flow, extack,
3082 ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
3083 if (modify_ip_header && ip_proto != IPPROTO_TCP &&
3084 ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
3085 NL_SET_ERR_MSG_MOD(extack,
3086 "can't offload re-write of non TCP/UDP");
3087 netdev_info(priv->netdev, "can't offload re-write of ip proto %d\n",
3096 static bool actions_match_supported(struct mlx5e_priv *priv,
3097 struct flow_action *flow_action,
3098 struct mlx5e_tc_flow_parse_attr *parse_attr,
3099 struct mlx5e_tc_flow *flow,
3100 struct netlink_ext_ack *extack)
3102 bool ct_flow = false, ct_clear = false;
3105 ct_clear = flow->attr->ct_attr.ct_action &
3107 ct_flow = flow_flag_test(flow, CT) && !ct_clear;
3108 actions = flow->attr->action;
3110 if (mlx5e_is_eswitch_flow(flow)) {
3111 if (flow->attr->esw_attr->split_count && ct_flow &&
3112 !MLX5_CAP_GEN(flow->attr->esw_attr->in_mdev, reg_c_preserve)) {
3113 /* All registers used by ct are cleared when using
3116 NL_SET_ERR_MSG_MOD(extack,
3117 "Can't offload mirroring with action ct");
3122 if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
3123 return modify_header_match_supported(priv, &parse_attr->spec,
3124 flow_action, actions,
3131 static bool same_port_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
3133 return priv->mdev == peer_priv->mdev;
3136 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
3138 struct mlx5_core_dev *fmdev, *pmdev;
3139 u64 fsystem_guid, psystem_guid;
3142 pmdev = peer_priv->mdev;
3144 fsystem_guid = mlx5_query_nic_system_image_guid(fmdev);
3145 psystem_guid = mlx5_query_nic_system_image_guid(pmdev);
3147 return (fsystem_guid == psystem_guid);
3150 static bool same_vf_reps(struct mlx5e_priv *priv,
3151 struct net_device *out_dev)
3153 return mlx5e_eswitch_vf_rep(priv->netdev) &&
3154 priv->netdev == out_dev;
3157 static int add_vlan_rewrite_action(struct mlx5e_priv *priv, int namespace,
3158 const struct flow_action_entry *act,
3159 struct mlx5e_tc_flow_parse_attr *parse_attr,
3160 struct pedit_headers_action *hdrs,
3161 u32 *action, struct netlink_ext_ack *extack)
3163 u16 mask16 = VLAN_VID_MASK;
3164 u16 val16 = act->vlan.vid & VLAN_VID_MASK;
3165 const struct flow_action_entry pedit_act = {
3166 .id = FLOW_ACTION_MANGLE,
3167 .mangle.htype = FLOW_ACT_MANGLE_HDR_TYPE_ETH,
3168 .mangle.offset = offsetof(struct vlan_ethhdr, h_vlan_TCI),
3169 .mangle.mask = ~(u32)be16_to_cpu(*(__be16 *)&mask16),
3170 .mangle.val = (u32)be16_to_cpu(*(__be16 *)&val16),
3172 u8 match_prio_mask, match_prio_val;
3173 void *headers_c, *headers_v;
3176 headers_c = get_match_headers_criteria(*action, &parse_attr->spec);
3177 headers_v = get_match_headers_value(*action, &parse_attr->spec);
3179 if (!(MLX5_GET(fte_match_set_lyr_2_4, headers_c, cvlan_tag) &&
3180 MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag))) {
3181 NL_SET_ERR_MSG_MOD(extack,
3182 "VLAN rewrite action must have VLAN protocol match");
3186 match_prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
3187 match_prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
3188 if (act->vlan.prio != (match_prio_val & match_prio_mask)) {
3189 NL_SET_ERR_MSG_MOD(extack,
3190 "Changing VLAN prio is not supported");
3194 err = parse_tc_pedit_action(priv, &pedit_act, namespace, parse_attr, hdrs, NULL, extack);
3195 *action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3201 add_vlan_prio_tag_rewrite_action(struct mlx5e_priv *priv,
3202 struct mlx5e_tc_flow_parse_attr *parse_attr,
3203 struct pedit_headers_action *hdrs,
3204 u32 *action, struct netlink_ext_ack *extack)
3206 const struct flow_action_entry prio_tag_act = {
3209 MLX5_GET(fte_match_set_lyr_2_4,
3210 get_match_headers_value(*action,
3213 MLX5_GET(fte_match_set_lyr_2_4,
3214 get_match_headers_criteria(*action,
3219 return add_vlan_rewrite_action(priv, MLX5_FLOW_NAMESPACE_FDB,
3220 &prio_tag_act, parse_attr, hdrs, action,
3224 static int validate_goto_chain(struct mlx5e_priv *priv,
3225 struct mlx5e_tc_flow *flow,
3226 const struct flow_action_entry *act,
3228 struct netlink_ext_ack *extack)
3230 bool is_esw = mlx5e_is_eswitch_flow(flow);
3231 struct mlx5_flow_attr *attr = flow->attr;
3232 bool ft_flow = mlx5e_is_ft_flow(flow);
3233 u32 dest_chain = act->chain_index;
3234 struct mlx5_fs_chains *chains;
3235 struct mlx5_eswitch *esw;
3236 u32 reformat_and_fwd;
3239 esw = priv->mdev->priv.eswitch;
3240 chains = is_esw ? esw_chains(esw) : nic_chains(priv);
3241 max_chain = mlx5_chains_get_chain_range(chains);
3242 reformat_and_fwd = is_esw ?
3243 MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev, reformat_and_fwd_to_table) :
3244 MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, reformat_and_fwd_to_table);
3247 NL_SET_ERR_MSG_MOD(extack, "Goto action is not supported");
3251 if (!mlx5_chains_backwards_supported(chains) &&
3252 dest_chain <= attr->chain) {
3253 NL_SET_ERR_MSG_MOD(extack,
3254 "Goto lower numbered chain isn't supported");
3258 if (dest_chain > max_chain) {
3259 NL_SET_ERR_MSG_MOD(extack,
3260 "Requested destination chain is out of supported range");
3264 if (actions & (MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
3265 MLX5_FLOW_CONTEXT_ACTION_DECAP) &&
3266 !reformat_and_fwd) {
3267 NL_SET_ERR_MSG_MOD(extack,
3268 "Goto chain is not allowed if action has reformat or decap");
3275 static int parse_tc_nic_actions(struct mlx5e_priv *priv,
3276 struct flow_action *flow_action,
3277 struct mlx5e_tc_flow_parse_attr *parse_attr,
3278 struct mlx5e_tc_flow *flow,
3279 struct netlink_ext_ack *extack)
3281 struct mlx5_flow_attr *attr = flow->attr;
3282 struct pedit_headers_action hdrs[2] = {};
3283 const struct flow_action_entry *act;
3284 struct mlx5_nic_flow_attr *nic_attr;
3288 if (!flow_action_has_entries(flow_action))
3291 if (!flow_action_hw_stats_check(flow_action, extack,
3292 FLOW_ACTION_HW_STATS_DELAYED_BIT))
3295 nic_attr = attr->nic_attr;
3297 nic_attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
3299 flow_action_for_each(i, act, flow_action) {
3301 case FLOW_ACTION_ACCEPT:
3302 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
3303 MLX5_FLOW_CONTEXT_ACTION_COUNT;
3305 case FLOW_ACTION_DROP:
3306 action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
3307 if (MLX5_CAP_FLOWTABLE(priv->mdev,
3308 flow_table_properties_nic_receive.flow_counter))
3309 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
3311 case FLOW_ACTION_MANGLE:
3312 case FLOW_ACTION_ADD:
3313 err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_KERNEL,
3314 parse_attr, hdrs, NULL, extack);
3318 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3320 case FLOW_ACTION_VLAN_MANGLE:
3321 err = add_vlan_rewrite_action(priv,
3322 MLX5_FLOW_NAMESPACE_KERNEL,
3323 act, parse_attr, hdrs,
3329 case FLOW_ACTION_CSUM:
3330 if (csum_offload_supported(priv, action,
3336 case FLOW_ACTION_REDIRECT: {
3337 struct net_device *peer_dev = act->dev;
3339 if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
3340 same_hw_devs(priv, netdev_priv(peer_dev))) {
3341 parse_attr->mirred_ifindex[0] = peer_dev->ifindex;
3342 flow_flag_set(flow, HAIRPIN);
3343 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
3344 MLX5_FLOW_CONTEXT_ACTION_COUNT;
3346 NL_SET_ERR_MSG_MOD(extack,
3347 "device is not on same HW, can't offload");
3348 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
3354 case FLOW_ACTION_MARK: {
3355 u32 mark = act->mark;
3357 if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
3358 NL_SET_ERR_MSG_MOD(extack,
3359 "Bad flow mark - only 16 bit is supported");
3363 nic_attr->flow_tag = mark;
3364 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3367 case FLOW_ACTION_GOTO:
3368 err = validate_goto_chain(priv, flow, act, action,
3373 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
3374 attr->dest_chain = act->chain_index;
3376 case FLOW_ACTION_CT:
3377 err = mlx5_tc_ct_parse_action(get_ct_priv(priv), attr, act, extack);
3381 flow_flag_set(flow, CT);
3384 NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
3389 if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
3390 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
3391 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_KERNEL,
3392 parse_attr, hdrs, &action, extack);
3395 /* in case all pedit actions are skipped, remove the MOD_HDR
3398 if (parse_attr->mod_hdr_acts.num_actions == 0) {
3399 action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3400 dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
3404 attr->action = action;
3406 if (attr->dest_chain) {
3407 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
3408 NL_SET_ERR_MSG(extack, "Mirroring goto chain rules isn't supported");
3411 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3414 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
3415 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3417 if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
3423 static bool is_merged_eswitch_vfs(struct mlx5e_priv *priv,
3424 struct net_device *peer_netdev)
3426 struct mlx5e_priv *peer_priv;
3428 peer_priv = netdev_priv(peer_netdev);
3430 return (MLX5_CAP_ESW(priv->mdev, merged_eswitch) &&
3431 mlx5e_eswitch_vf_rep(priv->netdev) &&
3432 mlx5e_eswitch_vf_rep(peer_netdev) &&
3433 same_hw_devs(priv, peer_priv));
3436 static int parse_tc_vlan_action(struct mlx5e_priv *priv,
3437 const struct flow_action_entry *act,
3438 struct mlx5_esw_flow_attr *attr,
3441 u8 vlan_idx = attr->total_vlan;
3443 if (vlan_idx >= MLX5_FS_VLAN_DEPTH)
3447 case FLOW_ACTION_VLAN_POP:
3449 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
3450 MLX5_FS_VLAN_DEPTH))
3453 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2;
3455 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3458 case FLOW_ACTION_VLAN_PUSH:
3459 attr->vlan_vid[vlan_idx] = act->vlan.vid;
3460 attr->vlan_prio[vlan_idx] = act->vlan.prio;
3461 attr->vlan_proto[vlan_idx] = act->vlan.proto;
3462 if (!attr->vlan_proto[vlan_idx])
3463 attr->vlan_proto[vlan_idx] = htons(ETH_P_8021Q);
3466 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
3467 MLX5_FS_VLAN_DEPTH))
3470 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2;
3472 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev, 1) &&
3473 (act->vlan.proto != htons(ETH_P_8021Q) ||
3477 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
3484 attr->total_vlan = vlan_idx + 1;
3489 static struct net_device *get_fdb_out_dev(struct net_device *uplink_dev,
3490 struct net_device *out_dev)
3492 struct net_device *fdb_out_dev = out_dev;
3493 struct net_device *uplink_upper;
3496 uplink_upper = netdev_master_upper_dev_get_rcu(uplink_dev);
3497 if (uplink_upper && netif_is_lag_master(uplink_upper) &&
3498 uplink_upper == out_dev) {
3499 fdb_out_dev = uplink_dev;
3500 } else if (netif_is_lag_master(out_dev)) {
3501 fdb_out_dev = bond_option_active_slave_get_rcu(netdev_priv(out_dev));
3503 (!mlx5e_eswitch_rep(fdb_out_dev) ||
3504 !netdev_port_same_parent_id(fdb_out_dev, uplink_dev)))
3511 static int add_vlan_push_action(struct mlx5e_priv *priv,
3512 struct mlx5_flow_attr *attr,
3513 struct net_device **out_dev,
3516 struct net_device *vlan_dev = *out_dev;
3517 struct flow_action_entry vlan_act = {
3518 .id = FLOW_ACTION_VLAN_PUSH,
3519 .vlan.vid = vlan_dev_vlan_id(vlan_dev),
3520 .vlan.proto = vlan_dev_vlan_proto(vlan_dev),
3525 err = parse_tc_vlan_action(priv, &vlan_act, attr->esw_attr, action);
3529 *out_dev = dev_get_by_index_rcu(dev_net(vlan_dev),
3530 dev_get_iflink(vlan_dev));
3531 if (is_vlan_dev(*out_dev))
3532 err = add_vlan_push_action(priv, attr, out_dev, action);
3537 static int add_vlan_pop_action(struct mlx5e_priv *priv,
3538 struct mlx5_flow_attr *attr,
3541 struct flow_action_entry vlan_act = {
3542 .id = FLOW_ACTION_VLAN_POP,
3544 int nest_level, err = 0;
3546 nest_level = attr->parse_attr->filter_dev->lower_level -
3547 priv->netdev->lower_level;
3548 while (nest_level--) {
3549 err = parse_tc_vlan_action(priv, &vlan_act, attr->esw_attr, action);
3557 static bool same_hw_reps(struct mlx5e_priv *priv,
3558 struct net_device *peer_netdev)
3560 struct mlx5e_priv *peer_priv;
3562 peer_priv = netdev_priv(peer_netdev);
3564 return mlx5e_eswitch_rep(priv->netdev) &&
3565 mlx5e_eswitch_rep(peer_netdev) &&
3566 same_hw_devs(priv, peer_priv);
3569 static bool is_lag_dev(struct mlx5e_priv *priv,
3570 struct net_device *peer_netdev)
3572 return ((mlx5_lag_is_sriov(priv->mdev) ||
3573 mlx5_lag_is_multipath(priv->mdev)) &&
3574 same_hw_reps(priv, peer_netdev));
3577 bool mlx5e_is_valid_eswitch_fwd_dev(struct mlx5e_priv *priv,
3578 struct net_device *out_dev)
3580 if (is_merged_eswitch_vfs(priv, out_dev))
3583 if (is_lag_dev(priv, out_dev))
3586 return mlx5e_eswitch_rep(out_dev) &&
3587 same_port_devs(priv, netdev_priv(out_dev));
3590 static bool is_duplicated_output_device(struct net_device *dev,
3591 struct net_device *out_dev,
3592 int *ifindexes, int if_count,
3593 struct netlink_ext_ack *extack)
3597 for (i = 0; i < if_count; i++) {
3598 if (ifindexes[i] == out_dev->ifindex) {
3599 NL_SET_ERR_MSG_MOD(extack,
3600 "can't duplicate output to same device");
3601 netdev_err(dev, "can't duplicate output to same device: %s\n",
3610 static int verify_uplink_forwarding(struct mlx5e_priv *priv,
3611 struct mlx5e_tc_flow *flow,
3612 struct net_device *out_dev,
3613 struct netlink_ext_ack *extack)
3615 struct mlx5_esw_flow_attr *attr = flow->attr->esw_attr;
3616 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3617 struct mlx5e_rep_priv *rep_priv;
3619 /* Forwarding non encapsulated traffic between
3620 * uplink ports is allowed only if
3621 * termination_table_raw_traffic cap is set.
3623 * Input vport was stored attr->in_rep.
3624 * In LAG case, *priv* is the private data of
3625 * uplink which may be not the input vport.
3627 rep_priv = mlx5e_rep_to_rep_priv(attr->in_rep);
3629 if (!(mlx5e_eswitch_uplink_rep(rep_priv->netdev) &&
3630 mlx5e_eswitch_uplink_rep(out_dev)))
3633 if (!MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev,
3634 termination_table_raw_traffic)) {
3635 NL_SET_ERR_MSG_MOD(extack,
3636 "devices are both uplink, can't offload forwarding");
3637 pr_err("devices %s %s are both uplink, can't offload forwarding\n",
3638 priv->netdev->name, out_dev->name);
3640 } else if (out_dev != rep_priv->netdev) {
3641 NL_SET_ERR_MSG_MOD(extack,
3642 "devices are not the same uplink, can't offload forwarding");
3643 pr_err("devices %s %s are both uplink but not the same, can't offload forwarding\n",
3644 priv->netdev->name, out_dev->name);
3650 static int parse_tc_fdb_actions(struct mlx5e_priv *priv,
3651 struct flow_action *flow_action,
3652 struct mlx5e_tc_flow *flow,
3653 struct netlink_ext_ack *extack,
3654 struct net_device *filter_dev)
3656 struct pedit_headers_action hdrs[2] = {};
3657 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3658 struct mlx5e_tc_flow_parse_attr *parse_attr;
3659 struct mlx5e_rep_priv *rpriv = priv->ppriv;
3660 const struct ip_tunnel_info *info = NULL;
3661 struct mlx5_flow_attr *attr = flow->attr;
3662 int ifindexes[MLX5_MAX_FLOW_FWD_VPORTS];
3663 bool ft_flow = mlx5e_is_ft_flow(flow);
3664 const struct flow_action_entry *act;
3665 struct mlx5_esw_flow_attr *esw_attr;
3666 struct mlx5_sample_attr sample = {};
3667 bool encap = false, decap = false;
3668 u32 action = attr->action;
3669 int err, i, if_count = 0;
3670 bool mpls_push = false;
3672 if (!flow_action_has_entries(flow_action))
3675 if (!flow_action_hw_stats_check(flow_action, extack,
3676 FLOW_ACTION_HW_STATS_DELAYED_BIT))
3679 esw_attr = attr->esw_attr;
3680 parse_attr = attr->parse_attr;
3682 flow_action_for_each(i, act, flow_action) {
3684 case FLOW_ACTION_DROP:
3685 action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
3686 MLX5_FLOW_CONTEXT_ACTION_COUNT;
3688 case FLOW_ACTION_TRAP:
3689 if (!flow_offload_has_one_action(flow_action)) {
3690 NL_SET_ERR_MSG_MOD(extack,
3691 "action trap is supported as a sole action only");
3694 action |= (MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
3695 MLX5_FLOW_CONTEXT_ACTION_COUNT);
3696 attr->flags |= MLX5_ESW_ATTR_FLAG_SLOW_PATH;
3698 case FLOW_ACTION_MPLS_PUSH:
3699 if (!MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
3700 reformat_l2_to_l3_tunnel) ||
3701 act->mpls_push.proto != htons(ETH_P_MPLS_UC)) {
3702 NL_SET_ERR_MSG_MOD(extack,
3703 "mpls push is supported only for mpls_uc protocol");
3708 case FLOW_ACTION_MPLS_POP:
3709 /* we only support mpls pop if it is the first action
3710 * and the filter net device is bareudp. Subsequent
3711 * actions can be pedit and the last can be mirred
3715 NL_SET_ERR_MSG_MOD(extack,
3716 "mpls pop supported only as first action");
3719 if (!netif_is_bareudp(filter_dev)) {
3720 NL_SET_ERR_MSG_MOD(extack,
3721 "mpls pop supported only on bareudp devices");
3725 parse_attr->eth.h_proto = act->mpls_pop.proto;
3726 action |= MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
3727 flow_flag_set(flow, L3_TO_L2_DECAP);
3729 case FLOW_ACTION_MANGLE:
3730 case FLOW_ACTION_ADD:
3731 err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_FDB,
3732 parse_attr, hdrs, flow, extack);
3736 if (!flow_flag_test(flow, L3_TO_L2_DECAP)) {
3737 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3738 esw_attr->split_count = esw_attr->out_count;
3741 case FLOW_ACTION_CSUM:
3742 if (csum_offload_supported(priv, action,
3743 act->csum_flags, extack))
3747 case FLOW_ACTION_REDIRECT:
3748 case FLOW_ACTION_MIRRED: {
3749 struct mlx5e_priv *out_priv;
3750 struct net_device *out_dev;
3754 /* out_dev is NULL when filters with
3755 * non-existing mirred device are replayed to
3761 if (mpls_push && !netif_is_bareudp(out_dev)) {
3762 NL_SET_ERR_MSG_MOD(extack,
3763 "mpls is supported only through a bareudp device");
3767 if (ft_flow && out_dev == priv->netdev) {
3768 /* Ignore forward to self rules generated
3769 * by adding both mlx5 devs to the flow table
3770 * block on a normal nft offload setup.
3775 if (esw_attr->out_count >= MLX5_MAX_FLOW_FWD_VPORTS) {
3776 NL_SET_ERR_MSG_MOD(extack,
3777 "can't support more output ports, can't offload forwarding");
3778 netdev_warn(priv->netdev,
3779 "can't support more than %d output ports, can't offload forwarding\n",
3780 esw_attr->out_count);
3784 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
3785 MLX5_FLOW_CONTEXT_ACTION_COUNT;
3787 parse_attr->mirred_ifindex[esw_attr->out_count] =
3789 parse_attr->tun_info[esw_attr->out_count] =
3790 mlx5e_dup_tun_info(info);
3791 if (!parse_attr->tun_info[esw_attr->out_count])
3794 esw_attr->dests[esw_attr->out_count].flags |=
3795 MLX5_ESW_DEST_ENCAP;
3796 esw_attr->out_count++;
3797 /* attr->dests[].rep is resolved when we
3800 } else if (netdev_port_same_parent_id(priv->netdev, out_dev)) {
3801 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3802 struct net_device *uplink_dev = mlx5_eswitch_uplink_get_proto_dev(esw, REP_ETH);
3804 if (is_duplicated_output_device(priv->netdev,
3811 ifindexes[if_count] = out_dev->ifindex;
3814 out_dev = get_fdb_out_dev(uplink_dev, out_dev);
3818 if (is_vlan_dev(out_dev)) {
3819 err = add_vlan_push_action(priv, attr,
3826 if (is_vlan_dev(parse_attr->filter_dev)) {
3827 err = add_vlan_pop_action(priv, attr,
3833 err = verify_uplink_forwarding(priv, flow, out_dev, extack);
3837 if (!mlx5e_is_valid_eswitch_fwd_dev(priv, out_dev)) {
3838 NL_SET_ERR_MSG_MOD(extack,
3839 "devices are not on same switch HW, can't offload forwarding");
3843 if (same_vf_reps(priv, out_dev)) {
3844 NL_SET_ERR_MSG_MOD(extack,
3845 "can't forward from a VF to itself");
3849 out_priv = netdev_priv(out_dev);
3850 rpriv = out_priv->ppriv;
3851 esw_attr->dests[esw_attr->out_count].rep = rpriv->rep;
3852 esw_attr->dests[esw_attr->out_count].mdev = out_priv->mdev;
3853 esw_attr->out_count++;
3854 } else if (parse_attr->filter_dev != priv->netdev) {
3855 /* All mlx5 devices are called to configure
3856 * high level device filters. Therefore, the
3857 * *attempt* to install a filter on invalid
3858 * eswitch should not trigger an explicit error
3862 NL_SET_ERR_MSG_MOD(extack,
3863 "devices are not on same switch HW, can't offload forwarding");
3864 netdev_warn(priv->netdev,
3865 "devices %s %s not on same switch HW, can't offload forwarding\n",
3872 case FLOW_ACTION_TUNNEL_ENCAP:
3880 case FLOW_ACTION_VLAN_PUSH:
3881 case FLOW_ACTION_VLAN_POP:
3882 if (act->id == FLOW_ACTION_VLAN_PUSH &&
3883 (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP)) {
3884 /* Replace vlan pop+push with vlan modify */
3885 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3886 err = add_vlan_rewrite_action(priv,
3887 MLX5_FLOW_NAMESPACE_FDB,
3888 act, parse_attr, hdrs,
3891 err = parse_tc_vlan_action(priv, act, esw_attr, &action);
3896 esw_attr->split_count = esw_attr->out_count;
3898 case FLOW_ACTION_VLAN_MANGLE:
3899 err = add_vlan_rewrite_action(priv,
3900 MLX5_FLOW_NAMESPACE_FDB,
3901 act, parse_attr, hdrs,
3906 esw_attr->split_count = esw_attr->out_count;
3908 case FLOW_ACTION_TUNNEL_DECAP:
3911 case FLOW_ACTION_GOTO:
3912 err = validate_goto_chain(priv, flow, act, action,
3917 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
3918 attr->dest_chain = act->chain_index;
3920 case FLOW_ACTION_CT:
3921 if (flow_flag_test(flow, SAMPLE)) {
3922 NL_SET_ERR_MSG_MOD(extack, "Sample action with connection tracking is not supported");
3925 err = mlx5_tc_ct_parse_action(get_ct_priv(priv), attr, act, extack);
3929 flow_flag_set(flow, CT);
3930 esw_attr->split_count = esw_attr->out_count;
3932 case FLOW_ACTION_SAMPLE:
3933 if (flow_flag_test(flow, CT)) {
3934 NL_SET_ERR_MSG_MOD(extack, "Sample action with connection tracking is not supported");
3937 sample.rate = act->sample.rate;
3938 sample.group_num = act->sample.psample_group->group_num;
3939 if (act->sample.truncate)
3940 sample.trunc_size = act->sample.trunc_size;
3941 flow_flag_set(flow, SAMPLE);
3944 NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
3949 /* always set IP version for indirect table handling */
3950 attr->ip_version = mlx5e_tc_get_ip_version(&parse_attr->spec, true);
3952 if (MLX5_CAP_GEN(esw->dev, prio_tag_required) &&
3953 action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) {
3954 /* For prio tag mode, replace vlan pop with rewrite vlan prio
3957 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3958 err = add_vlan_prio_tag_rewrite_action(priv, parse_attr, hdrs,
3964 if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
3965 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
3966 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_FDB,
3967 parse_attr, hdrs, &action, extack);
3970 /* in case all pedit actions are skipped, remove the MOD_HDR
3971 * flag. we might have set split_count either by pedit or
3972 * pop/push. if there is no pop/push either, reset it too.
3974 if (parse_attr->mod_hdr_acts.num_actions == 0) {
3975 action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3976 dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
3977 if (!((action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) ||
3978 (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH)))
3979 esw_attr->split_count = 0;
3983 attr->action = action;
3984 if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
3987 if (attr->dest_chain) {
3989 /* It can be supported if we'll create a mapping for
3990 * the tunnel device only (without tunnel), and set
3991 * this tunnel id with this decap flow.
3993 * On restore (miss), we'll just set this saved tunnel
3997 NL_SET_ERR_MSG(extack,
3998 "Decap with goto isn't supported");
3999 netdev_warn(priv->netdev,
4000 "Decap with goto isn't supported");
4004 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
4007 if (!(attr->action &
4008 (MLX5_FLOW_CONTEXT_ACTION_FWD_DEST | MLX5_FLOW_CONTEXT_ACTION_DROP))) {
4009 NL_SET_ERR_MSG_MOD(extack,
4010 "Rule must have at least one forward/drop action");
4014 if (esw_attr->split_count > 0 && !mlx5_esw_has_fwd_fdb(priv->mdev)) {
4015 NL_SET_ERR_MSG_MOD(extack,
4016 "current firmware doesn't support split rule for port mirroring");
4017 netdev_warn_once(priv->netdev, "current firmware doesn't support split rule for port mirroring\n");
4021 /* Allocate sample attribute only when there is a sample action and
4022 * no errors after parsing.
4024 if (flow_flag_test(flow, SAMPLE)) {
4025 esw_attr->sample = kzalloc(sizeof(*esw_attr->sample), GFP_KERNEL);
4026 if (!esw_attr->sample)
4028 *esw_attr->sample = sample;
4034 static void get_flags(int flags, unsigned long *flow_flags)
4036 unsigned long __flow_flags = 0;
4038 if (flags & MLX5_TC_FLAG(INGRESS))
4039 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_INGRESS);
4040 if (flags & MLX5_TC_FLAG(EGRESS))
4041 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_EGRESS);
4043 if (flags & MLX5_TC_FLAG(ESW_OFFLOAD))
4044 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_ESWITCH);
4045 if (flags & MLX5_TC_FLAG(NIC_OFFLOAD))
4046 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_NIC);
4047 if (flags & MLX5_TC_FLAG(FT_OFFLOAD))
4048 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_FT);
4050 *flow_flags = __flow_flags;
4053 static const struct rhashtable_params tc_ht_params = {
4054 .head_offset = offsetof(struct mlx5e_tc_flow, node),
4055 .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
4056 .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
4057 .automatic_shrinking = true,
4060 static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv,
4061 unsigned long flags)
4063 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
4064 struct mlx5e_rep_priv *uplink_rpriv;
4066 if (flags & MLX5_TC_FLAG(ESW_OFFLOAD)) {
4067 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
4068 return &uplink_rpriv->uplink_priv.tc_ht;
4069 } else /* NIC offload */
4070 return &priv->fs.tc.ht;
4073 static bool is_peer_flow_needed(struct mlx5e_tc_flow *flow)
4075 struct mlx5_esw_flow_attr *esw_attr = flow->attr->esw_attr;
4076 struct mlx5_flow_attr *attr = flow->attr;
4077 bool is_rep_ingress = esw_attr->in_rep->vport != MLX5_VPORT_UPLINK &&
4078 flow_flag_test(flow, INGRESS);
4079 bool act_is_encap = !!(attr->action &
4080 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT);
4081 bool esw_paired = mlx5_devcom_is_paired(esw_attr->in_mdev->priv.devcom,
4082 MLX5_DEVCOM_ESW_OFFLOADS);
4087 if ((mlx5_lag_is_sriov(esw_attr->in_mdev) ||
4088 mlx5_lag_is_multipath(esw_attr->in_mdev)) &&
4089 (is_rep_ingress || act_is_encap))
4095 struct mlx5_flow_attr *
4096 mlx5_alloc_flow_attr(enum mlx5_flow_namespace_type type)
4098 u32 ex_attr_size = (type == MLX5_FLOW_NAMESPACE_FDB) ?
4099 sizeof(struct mlx5_esw_flow_attr) :
4100 sizeof(struct mlx5_nic_flow_attr);
4101 struct mlx5_flow_attr *attr;
4103 return kzalloc(sizeof(*attr) + ex_attr_size, GFP_KERNEL);
4107 mlx5e_alloc_flow(struct mlx5e_priv *priv, int attr_size,
4108 struct flow_cls_offload *f, unsigned long flow_flags,
4109 struct mlx5e_tc_flow_parse_attr **__parse_attr,
4110 struct mlx5e_tc_flow **__flow)
4112 struct mlx5e_tc_flow_parse_attr *parse_attr;
4113 struct mlx5_flow_attr *attr;
4114 struct mlx5e_tc_flow *flow;
4118 flow = kzalloc(sizeof(*flow), GFP_KERNEL);
4119 parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
4120 if (!parse_attr || !flow)
4123 flow->flags = flow_flags;
4124 flow->cookie = f->cookie;
4127 attr = mlx5_alloc_flow_attr(get_flow_name_space(flow));
4133 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++)
4134 INIT_LIST_HEAD(&flow->encaps[out_index].list);
4135 INIT_LIST_HEAD(&flow->hairpin);
4136 INIT_LIST_HEAD(&flow->l3_to_l2_reformat);
4137 refcount_set(&flow->refcnt, 1);
4138 init_completion(&flow->init_done);
4141 *__parse_attr = parse_attr;
4152 mlx5e_flow_attr_init(struct mlx5_flow_attr *attr,
4153 struct mlx5e_tc_flow_parse_attr *parse_attr,
4154 struct flow_cls_offload *f)
4156 attr->parse_attr = parse_attr;
4157 attr->chain = f->common.chain_index;
4158 attr->prio = f->common.prio;
4162 mlx5e_flow_esw_attr_init(struct mlx5_flow_attr *attr,
4163 struct mlx5e_priv *priv,
4164 struct mlx5e_tc_flow_parse_attr *parse_attr,
4165 struct flow_cls_offload *f,
4166 struct mlx5_eswitch_rep *in_rep,
4167 struct mlx5_core_dev *in_mdev)
4169 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
4170 struct mlx5_esw_flow_attr *esw_attr = attr->esw_attr;
4172 mlx5e_flow_attr_init(attr, parse_attr, f);
4174 esw_attr->in_rep = in_rep;
4175 esw_attr->in_mdev = in_mdev;
4177 if (MLX5_CAP_ESW(esw->dev, counter_eswitch_affinity) ==
4178 MLX5_COUNTER_SOURCE_ESWITCH)
4179 esw_attr->counter_dev = in_mdev;
4181 esw_attr->counter_dev = priv->mdev;
4184 static struct mlx5e_tc_flow *
4185 __mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
4186 struct flow_cls_offload *f,
4187 unsigned long flow_flags,
4188 struct net_device *filter_dev,
4189 struct mlx5_eswitch_rep *in_rep,
4190 struct mlx5_core_dev *in_mdev)
4192 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
4193 struct netlink_ext_ack *extack = f->common.extack;
4194 struct mlx5e_tc_flow_parse_attr *parse_attr;
4195 struct mlx5e_tc_flow *flow;
4198 flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_ESWITCH);
4199 attr_size = sizeof(struct mlx5_esw_flow_attr);
4200 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
4201 &parse_attr, &flow);
4205 parse_attr->filter_dev = filter_dev;
4206 mlx5e_flow_esw_attr_init(flow->attr,
4208 f, in_rep, in_mdev);
4210 err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
4215 /* actions validation depends on parsing the ct matches first */
4216 err = mlx5_tc_ct_match_add(get_ct_priv(priv), &parse_attr->spec, f,
4217 &flow->attr->ct_attr, extack);
4221 err = parse_tc_fdb_actions(priv, &rule->action, flow, extack, filter_dev);
4225 err = mlx5e_tc_add_fdb_flow(priv, flow, extack);
4226 complete_all(&flow->init_done);
4228 if (!(err == -ENETUNREACH && mlx5_lag_is_multipath(in_mdev)))
4231 add_unready_flow(flow);
4237 mlx5e_flow_put(priv, flow);
4239 return ERR_PTR(err);
4242 static int mlx5e_tc_add_fdb_peer_flow(struct flow_cls_offload *f,
4243 struct mlx5e_tc_flow *flow,
4244 unsigned long flow_flags)
4246 struct mlx5e_priv *priv = flow->priv, *peer_priv;
4247 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch, *peer_esw;
4248 struct mlx5_esw_flow_attr *attr = flow->attr->esw_attr;
4249 struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
4250 struct mlx5e_tc_flow_parse_attr *parse_attr;
4251 struct mlx5e_rep_priv *peer_urpriv;
4252 struct mlx5e_tc_flow *peer_flow;
4253 struct mlx5_core_dev *in_mdev;
4256 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4260 peer_urpriv = mlx5_eswitch_get_uplink_priv(peer_esw, REP_ETH);
4261 peer_priv = netdev_priv(peer_urpriv->netdev);
4263 /* in_mdev is assigned of which the packet originated from.
4264 * So packets redirected to uplink use the same mdev of the
4265 * original flow and packets redirected from uplink use the
4268 if (attr->in_rep->vport == MLX5_VPORT_UPLINK)
4269 in_mdev = peer_priv->mdev;
4271 in_mdev = priv->mdev;
4273 parse_attr = flow->attr->parse_attr;
4274 peer_flow = __mlx5e_add_fdb_flow(peer_priv, f, flow_flags,
4275 parse_attr->filter_dev,
4276 attr->in_rep, in_mdev);
4277 if (IS_ERR(peer_flow)) {
4278 err = PTR_ERR(peer_flow);
4282 flow->peer_flow = peer_flow;
4283 flow_flag_set(flow, DUP);
4284 mutex_lock(&esw->offloads.peer_mutex);
4285 list_add_tail(&flow->peer, &esw->offloads.peer_flows);
4286 mutex_unlock(&esw->offloads.peer_mutex);
4289 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4294 mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
4295 struct flow_cls_offload *f,
4296 unsigned long flow_flags,
4297 struct net_device *filter_dev,
4298 struct mlx5e_tc_flow **__flow)
4300 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4301 struct mlx5_eswitch_rep *in_rep = rpriv->rep;
4302 struct mlx5_core_dev *in_mdev = priv->mdev;
4303 struct mlx5e_tc_flow *flow;
4306 flow = __mlx5e_add_fdb_flow(priv, f, flow_flags, filter_dev, in_rep,
4309 return PTR_ERR(flow);
4311 if (is_peer_flow_needed(flow)) {
4312 err = mlx5e_tc_add_fdb_peer_flow(f, flow, flow_flags);
4314 mlx5e_tc_del_fdb_flow(priv, flow);
4328 mlx5e_add_nic_flow(struct mlx5e_priv *priv,
4329 struct flow_cls_offload *f,
4330 unsigned long flow_flags,
4331 struct net_device *filter_dev,
4332 struct mlx5e_tc_flow **__flow)
4334 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
4335 struct netlink_ext_ack *extack = f->common.extack;
4336 struct mlx5e_tc_flow_parse_attr *parse_attr;
4337 struct mlx5e_tc_flow *flow;
4340 if (!MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, ignore_flow_level)) {
4341 if (!tc_cls_can_offload_and_chain0(priv->netdev, &f->common))
4343 } else if (!tc_can_offload_extack(priv->netdev, f->common.extack)) {
4347 flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_NIC);
4348 attr_size = sizeof(struct mlx5_nic_flow_attr);
4349 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
4350 &parse_attr, &flow);
4354 parse_attr->filter_dev = filter_dev;
4355 mlx5e_flow_attr_init(flow->attr, parse_attr, f);
4357 err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
4362 err = mlx5_tc_ct_match_add(get_ct_priv(priv), &parse_attr->spec, f,
4363 &flow->attr->ct_attr, extack);
4367 err = parse_tc_nic_actions(priv, &rule->action, parse_attr, flow, extack);
4371 err = mlx5e_tc_add_nic_flow(priv, parse_attr, flow, extack);
4375 flow_flag_set(flow, OFFLOADED);
4381 flow_flag_set(flow, FAILED);
4382 dealloc_mod_hdr_actions(&parse_attr->mod_hdr_acts);
4383 mlx5e_flow_put(priv, flow);
4389 mlx5e_tc_add_flow(struct mlx5e_priv *priv,
4390 struct flow_cls_offload *f,
4391 unsigned long flags,
4392 struct net_device *filter_dev,
4393 struct mlx5e_tc_flow **flow)
4395 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
4396 unsigned long flow_flags;
4399 get_flags(flags, &flow_flags);
4401 if (!tc_can_offload_extack(priv->netdev, f->common.extack))
4404 if (esw && esw->mode == MLX5_ESWITCH_OFFLOADS)
4405 err = mlx5e_add_fdb_flow(priv, f, flow_flags,
4408 err = mlx5e_add_nic_flow(priv, f, flow_flags,
4414 static bool is_flow_rule_duplicate_allowed(struct net_device *dev,
4415 struct mlx5e_rep_priv *rpriv)
4417 /* Offloaded flow rule is allowed to duplicate on non-uplink representor
4418 * sharing tc block with other slaves of a lag device. Rpriv can be NULL if this
4419 * function is called from NIC mode.
4421 return netif_is_lag_port(dev) && rpriv && rpriv->rep->vport != MLX5_VPORT_UPLINK;
4424 int mlx5e_configure_flower(struct net_device *dev, struct mlx5e_priv *priv,
4425 struct flow_cls_offload *f, unsigned long flags)
4427 struct netlink_ext_ack *extack = f->common.extack;
4428 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4429 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4430 struct mlx5e_tc_flow *flow;
4433 if (!mlx5_esw_hold(priv->mdev))
4436 mlx5_esw_get(priv->mdev);
4439 flow = rhashtable_lookup(tc_ht, &f->cookie, tc_ht_params);
4441 /* Same flow rule offloaded to non-uplink representor sharing tc block,
4444 if (is_flow_rule_duplicate_allowed(dev, rpriv) && flow->orig_dev != dev)
4447 NL_SET_ERR_MSG_MOD(extack,
4448 "flow cookie already exists, ignoring");
4449 netdev_warn_once(priv->netdev,
4450 "flow cookie %lx already exists, ignoring\n",
4460 trace_mlx5e_configure_flower(f);
4461 err = mlx5e_tc_add_flow(priv, f, flags, dev, &flow);
4465 /* Flow rule offloaded to non-uplink representor sharing tc block,
4466 * set the flow's owner dev.
4468 if (is_flow_rule_duplicate_allowed(dev, rpriv))
4469 flow->orig_dev = dev;
4471 err = rhashtable_lookup_insert_fast(tc_ht, &flow->node, tc_ht_params);
4475 mlx5_esw_release(priv->mdev);
4479 mlx5e_flow_put(priv, flow);
4481 mlx5_esw_put(priv->mdev);
4482 mlx5_esw_release(priv->mdev);
4486 static bool same_flow_direction(struct mlx5e_tc_flow *flow, int flags)
4488 bool dir_ingress = !!(flags & MLX5_TC_FLAG(INGRESS));
4489 bool dir_egress = !!(flags & MLX5_TC_FLAG(EGRESS));
4491 return flow_flag_test(flow, INGRESS) == dir_ingress &&
4492 flow_flag_test(flow, EGRESS) == dir_egress;
4495 int mlx5e_delete_flower(struct net_device *dev, struct mlx5e_priv *priv,
4496 struct flow_cls_offload *f, unsigned long flags)
4498 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4499 struct mlx5e_tc_flow *flow;
4503 flow = rhashtable_lookup(tc_ht, &f->cookie, tc_ht_params);
4504 if (!flow || !same_flow_direction(flow, flags)) {
4509 /* Only delete the flow if it doesn't have MLX5E_TC_FLOW_DELETED flag
4512 if (flow_flag_test_and_set(flow, DELETED)) {
4516 rhashtable_remove_fast(tc_ht, &flow->node, tc_ht_params);
4519 trace_mlx5e_delete_flower(f);
4520 mlx5e_flow_put(priv, flow);
4522 mlx5_esw_put(priv->mdev);
4530 int mlx5e_stats_flower(struct net_device *dev, struct mlx5e_priv *priv,
4531 struct flow_cls_offload *f, unsigned long flags)
4533 struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
4534 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4535 struct mlx5_eswitch *peer_esw;
4536 struct mlx5e_tc_flow *flow;
4537 struct mlx5_fc *counter;
4544 flow = mlx5e_flow_get(rhashtable_lookup(tc_ht, &f->cookie,
4548 return PTR_ERR(flow);
4550 if (!same_flow_direction(flow, flags)) {
4555 if (mlx5e_is_offloaded_flow(flow) || flow_flag_test(flow, CT)) {
4556 counter = mlx5e_tc_get_counter(flow);
4560 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
4563 /* Under multipath it's possible for one rule to be currently
4564 * un-offloaded while the other rule is offloaded.
4566 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4570 if (flow_flag_test(flow, DUP) &&
4571 flow_flag_test(flow->peer_flow, OFFLOADED)) {
4576 counter = mlx5e_tc_get_counter(flow->peer_flow);
4578 goto no_peer_counter;
4579 mlx5_fc_query_cached(counter, &bytes2, &packets2, &lastuse2);
4582 packets += packets2;
4583 lastuse = max_t(u64, lastuse, lastuse2);
4587 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4589 flow_stats_update(&f->stats, bytes, packets, 0, lastuse,
4590 FLOW_ACTION_HW_STATS_DELAYED);
4591 trace_mlx5e_stats_flower(f);
4593 mlx5e_flow_put(priv, flow);
4597 static int apply_police_params(struct mlx5e_priv *priv, u64 rate,
4598 struct netlink_ext_ack *extack)
4600 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4601 struct mlx5_eswitch *esw;
4606 vport_num = rpriv->rep->vport;
4607 if (vport_num >= MLX5_VPORT_ECPF) {
4608 NL_SET_ERR_MSG_MOD(extack,
4609 "Ingress rate limit is supported only for Eswitch ports connected to VFs");
4613 esw = priv->mdev->priv.eswitch;
4614 /* rate is given in bytes/sec.
4615 * First convert to bits/sec and then round to the nearest mbit/secs.
4616 * mbit means million bits.
4617 * Moreover, if rate is non zero we choose to configure to a minimum of
4621 rate = (rate * BITS_PER_BYTE) + 500000;
4622 do_div(rate, 1000000);
4623 rate_mbps = max_t(u32, rate, 1);
4626 err = mlx5_esw_modify_vport_rate(esw, vport_num, rate_mbps);
4628 NL_SET_ERR_MSG_MOD(extack, "failed applying action to hardware");
4633 static int scan_tc_matchall_fdb_actions(struct mlx5e_priv *priv,
4634 struct flow_action *flow_action,
4635 struct netlink_ext_ack *extack)
4637 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4638 const struct flow_action_entry *act;
4642 if (!flow_action_has_entries(flow_action)) {
4643 NL_SET_ERR_MSG_MOD(extack, "matchall called with no action");
4647 if (!flow_offload_has_one_action(flow_action)) {
4648 NL_SET_ERR_MSG_MOD(extack, "matchall policing support only a single action");
4652 if (!flow_action_basic_hw_stats_check(flow_action, extack))
4655 flow_action_for_each(i, act, flow_action) {
4657 case FLOW_ACTION_POLICE:
4658 if (act->police.rate_pkt_ps) {
4659 NL_SET_ERR_MSG_MOD(extack, "QoS offload not support packets per second");
4662 err = apply_police_params(priv, act->police.rate_bytes_ps, extack);
4666 rpriv->prev_vf_vport_stats = priv->stats.vf_vport;
4669 NL_SET_ERR_MSG_MOD(extack, "mlx5 supports only police action for matchall");
4677 int mlx5e_tc_configure_matchall(struct mlx5e_priv *priv,
4678 struct tc_cls_matchall_offload *ma)
4680 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
4681 struct netlink_ext_ack *extack = ma->common.extack;
4683 if (!mlx5_esw_qos_enabled(esw)) {
4684 NL_SET_ERR_MSG_MOD(extack, "QoS is not supported on this device");
4688 if (ma->common.prio != 1) {
4689 NL_SET_ERR_MSG_MOD(extack, "only priority 1 is supported");
4693 return scan_tc_matchall_fdb_actions(priv, &ma->rule->action, extack);
4696 int mlx5e_tc_delete_matchall(struct mlx5e_priv *priv,
4697 struct tc_cls_matchall_offload *ma)
4699 struct netlink_ext_ack *extack = ma->common.extack;
4701 return apply_police_params(priv, 0, extack);
4704 void mlx5e_tc_stats_matchall(struct mlx5e_priv *priv,
4705 struct tc_cls_matchall_offload *ma)
4707 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4708 struct rtnl_link_stats64 cur_stats;
4712 cur_stats = priv->stats.vf_vport;
4713 dpkts = cur_stats.rx_packets - rpriv->prev_vf_vport_stats.rx_packets;
4714 dbytes = cur_stats.rx_bytes - rpriv->prev_vf_vport_stats.rx_bytes;
4715 rpriv->prev_vf_vport_stats = cur_stats;
4716 flow_stats_update(&ma->stats, dbytes, dpkts, 0, jiffies,
4717 FLOW_ACTION_HW_STATS_DELAYED);
4720 static void mlx5e_tc_hairpin_update_dead_peer(struct mlx5e_priv *priv,
4721 struct mlx5e_priv *peer_priv)
4723 struct mlx5_core_dev *peer_mdev = peer_priv->mdev;
4724 struct mlx5e_hairpin_entry *hpe, *tmp;
4725 LIST_HEAD(init_wait_list);
4729 if (!same_hw_devs(priv, peer_priv))
4732 peer_vhca_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
4734 mutex_lock(&priv->fs.tc.hairpin_tbl_lock);
4735 hash_for_each(priv->fs.tc.hairpin_tbl, bkt, hpe, hairpin_hlist)
4736 if (refcount_inc_not_zero(&hpe->refcnt))
4737 list_add(&hpe->dead_peer_wait_list, &init_wait_list);
4738 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
4740 list_for_each_entry_safe(hpe, tmp, &init_wait_list, dead_peer_wait_list) {
4741 wait_for_completion(&hpe->res_ready);
4742 if (!IS_ERR_OR_NULL(hpe->hp) && hpe->peer_vhca_id == peer_vhca_id)
4743 hpe->hp->pair->peer_gone = true;
4745 mlx5e_hairpin_put(priv, hpe);
4749 static int mlx5e_tc_netdev_event(struct notifier_block *this,
4750 unsigned long event, void *ptr)
4752 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4753 struct mlx5e_flow_steering *fs;
4754 struct mlx5e_priv *peer_priv;
4755 struct mlx5e_tc_table *tc;
4756 struct mlx5e_priv *priv;
4758 if (ndev->netdev_ops != &mlx5e_netdev_ops ||
4759 event != NETDEV_UNREGISTER ||
4760 ndev->reg_state == NETREG_REGISTERED)
4763 tc = container_of(this, struct mlx5e_tc_table, netdevice_nb);
4764 fs = container_of(tc, struct mlx5e_flow_steering, tc);
4765 priv = container_of(fs, struct mlx5e_priv, fs);
4766 peer_priv = netdev_priv(ndev);
4767 if (priv == peer_priv ||
4768 !(priv->netdev->features & NETIF_F_HW_TC))
4771 mlx5e_tc_hairpin_update_dead_peer(priv, peer_priv);
4776 static int mlx5e_tc_nic_get_ft_size(struct mlx5_core_dev *dev)
4778 int tc_grp_size, tc_tbl_size;
4779 u32 max_flow_counter;
4781 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
4782 MLX5_CAP_GEN(dev, max_flow_counter_15_0);
4784 tc_grp_size = min_t(int, max_flow_counter, MLX5E_TC_TABLE_MAX_GROUP_SIZE);
4786 tc_tbl_size = min_t(int, tc_grp_size * MLX5E_TC_TABLE_NUM_GROUPS,
4787 BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev, log_max_ft_size)));
4792 int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
4794 struct mlx5e_tc_table *tc = &priv->fs.tc;
4795 struct mlx5_core_dev *dev = priv->mdev;
4796 struct mapping_ctx *chains_mapping;
4797 struct mlx5_chains_attr attr = {};
4800 mlx5e_mod_hdr_tbl_init(&tc->mod_hdr);
4801 mutex_init(&tc->t_lock);
4802 mutex_init(&tc->hairpin_tbl_lock);
4803 hash_init(tc->hairpin_tbl);
4805 err = rhashtable_init(&tc->ht, &tc_ht_params);
4809 lockdep_set_class(&tc->ht.mutex, &tc_ht_lock_key);
4811 chains_mapping = mapping_create(sizeof(struct mlx5_mapped_obj),
4812 MLX5E_TC_TABLE_CHAIN_TAG_MASK, true);
4813 if (IS_ERR(chains_mapping)) {
4814 err = PTR_ERR(chains_mapping);
4817 tc->mapping = chains_mapping;
4819 if (MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, ignore_flow_level))
4820 attr.flags = MLX5_CHAINS_AND_PRIOS_SUPPORTED |
4821 MLX5_CHAINS_IGNORE_FLOW_LEVEL_SUPPORTED;
4822 attr.ns = MLX5_FLOW_NAMESPACE_KERNEL;
4823 attr.max_ft_sz = mlx5e_tc_nic_get_ft_size(dev);
4824 attr.max_grp_num = MLX5E_TC_TABLE_NUM_GROUPS;
4825 attr.default_ft = mlx5e_vlan_get_flowtable(priv->fs.vlan);
4826 attr.mapping = chains_mapping;
4828 tc->chains = mlx5_chains_create(dev, &attr);
4829 if (IS_ERR(tc->chains)) {
4830 err = PTR_ERR(tc->chains);
4834 tc->ct = mlx5_tc_ct_init(priv, tc->chains, &priv->fs.tc.mod_hdr,
4835 MLX5_FLOW_NAMESPACE_KERNEL);
4837 tc->netdevice_nb.notifier_call = mlx5e_tc_netdev_event;
4838 err = register_netdevice_notifier_dev_net(priv->netdev,
4842 tc->netdevice_nb.notifier_call = NULL;
4843 mlx5_core_warn(priv->mdev, "Failed to register netdev notifier\n");
4850 mlx5_tc_ct_clean(tc->ct);
4851 mlx5_chains_destroy(tc->chains);
4853 mapping_destroy(chains_mapping);
4855 rhashtable_destroy(&tc->ht);
4859 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
4861 struct mlx5e_tc_flow *flow = ptr;
4862 struct mlx5e_priv *priv = flow->priv;
4864 mlx5e_tc_del_flow(priv, flow);
4868 void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
4870 struct mlx5e_tc_table *tc = &priv->fs.tc;
4872 if (tc->netdevice_nb.notifier_call)
4873 unregister_netdevice_notifier_dev_net(priv->netdev,
4877 mlx5e_mod_hdr_tbl_destroy(&tc->mod_hdr);
4878 mutex_destroy(&tc->hairpin_tbl_lock);
4880 rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, NULL);
4882 if (!IS_ERR_OR_NULL(tc->t)) {
4883 mlx5_chains_put_table(tc->chains, 0, 1, MLX5E_TC_FT_LEVEL);
4886 mutex_destroy(&tc->t_lock);
4888 mlx5_tc_ct_clean(tc->ct);
4889 mapping_destroy(tc->mapping);
4890 mlx5_chains_destroy(tc->chains);
4893 int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
4895 const size_t sz_enc_opts = sizeof(struct tunnel_match_enc_opts);
4896 struct mlx5_rep_uplink_priv *uplink_priv;
4897 struct mlx5e_rep_priv *rpriv;
4898 struct mapping_ctx *mapping;
4899 struct mlx5_eswitch *esw;
4900 struct mlx5e_priv *priv;
4903 uplink_priv = container_of(tc_ht, struct mlx5_rep_uplink_priv, tc_ht);
4904 rpriv = container_of(uplink_priv, struct mlx5e_rep_priv, uplink_priv);
4905 priv = netdev_priv(rpriv->netdev);
4906 esw = priv->mdev->priv.eswitch;
4908 uplink_priv->ct_priv = mlx5_tc_ct_init(netdev_priv(priv->netdev),
4910 &esw->offloads.mod_hdr,
4911 MLX5_FLOW_NAMESPACE_FDB);
4913 #if IS_ENABLED(CONFIG_MLX5_TC_SAMPLE)
4914 uplink_priv->esw_psample = mlx5_esw_sample_init(netdev_priv(priv->netdev));
4917 mapping = mapping_create(sizeof(struct tunnel_match_key),
4918 TUNNEL_INFO_BITS_MASK, true);
4919 if (IS_ERR(mapping)) {
4920 err = PTR_ERR(mapping);
4921 goto err_tun_mapping;
4923 uplink_priv->tunnel_mapping = mapping;
4925 /* 0xFFF is reserved for stack devices slow path table mark */
4926 mapping = mapping_create(sz_enc_opts, ENC_OPTS_BITS_MASK - 1, true);
4927 if (IS_ERR(mapping)) {
4928 err = PTR_ERR(mapping);
4929 goto err_enc_opts_mapping;
4931 uplink_priv->tunnel_enc_opts_mapping = mapping;
4933 err = rhashtable_init(tc_ht, &tc_ht_params);
4937 lockdep_set_class(&tc_ht->mutex, &tc_ht_lock_key);
4939 uplink_priv->encap = mlx5e_tc_tun_init(priv);
4940 if (IS_ERR(uplink_priv->encap)) {
4941 err = PTR_ERR(uplink_priv->encap);
4942 goto err_register_fib_notifier;
4947 err_register_fib_notifier:
4948 rhashtable_destroy(tc_ht);
4950 mapping_destroy(uplink_priv->tunnel_enc_opts_mapping);
4951 err_enc_opts_mapping:
4952 mapping_destroy(uplink_priv->tunnel_mapping);
4954 #if IS_ENABLED(CONFIG_MLX5_TC_SAMPLE)
4955 mlx5_esw_sample_cleanup(uplink_priv->esw_psample);
4957 mlx5_tc_ct_clean(uplink_priv->ct_priv);
4958 netdev_warn(priv->netdev,
4959 "Failed to initialize tc (eswitch), err: %d", err);
4963 void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
4965 struct mlx5_rep_uplink_priv *uplink_priv;
4967 uplink_priv = container_of(tc_ht, struct mlx5_rep_uplink_priv, tc_ht);
4969 rhashtable_free_and_destroy(tc_ht, _mlx5e_tc_del_flow, NULL);
4970 mlx5e_tc_tun_cleanup(uplink_priv->encap);
4972 mapping_destroy(uplink_priv->tunnel_enc_opts_mapping);
4973 mapping_destroy(uplink_priv->tunnel_mapping);
4975 #if IS_ENABLED(CONFIG_MLX5_TC_SAMPLE)
4976 mlx5_esw_sample_cleanup(uplink_priv->esw_psample);
4978 mlx5_tc_ct_clean(uplink_priv->ct_priv);
4981 int mlx5e_tc_num_filters(struct mlx5e_priv *priv, unsigned long flags)
4983 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4985 return atomic_read(&tc_ht->nelems);
4988 void mlx5e_tc_clean_fdb_peer_flows(struct mlx5_eswitch *esw)
4990 struct mlx5e_tc_flow *flow, *tmp;
4992 list_for_each_entry_safe(flow, tmp, &esw->offloads.peer_flows, peer)
4993 __mlx5e_tc_del_fdb_peer_flow(flow);
4996 void mlx5e_tc_reoffload_flows_work(struct work_struct *work)
4998 struct mlx5_rep_uplink_priv *rpriv =
4999 container_of(work, struct mlx5_rep_uplink_priv,
5000 reoffload_flows_work);
5001 struct mlx5e_tc_flow *flow, *tmp;
5003 mutex_lock(&rpriv->unready_flows_lock);
5004 list_for_each_entry_safe(flow, tmp, &rpriv->unready_flows, unready) {
5005 if (!mlx5e_tc_add_fdb_flow(flow->priv, flow, NULL))
5006 unready_flow_del(flow);
5008 mutex_unlock(&rpriv->unready_flows_lock);
5011 static int mlx5e_setup_tc_cls_flower(struct mlx5e_priv *priv,
5012 struct flow_cls_offload *cls_flower,
5013 unsigned long flags)
5015 switch (cls_flower->command) {
5016 case FLOW_CLS_REPLACE:
5017 return mlx5e_configure_flower(priv->netdev, priv, cls_flower,
5019 case FLOW_CLS_DESTROY:
5020 return mlx5e_delete_flower(priv->netdev, priv, cls_flower,
5022 case FLOW_CLS_STATS:
5023 return mlx5e_stats_flower(priv->netdev, priv, cls_flower,
5030 int mlx5e_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
5033 unsigned long flags = MLX5_TC_FLAG(INGRESS);
5034 struct mlx5e_priv *priv = cb_priv;
5036 if (!priv->netdev || !netif_device_present(priv->netdev))
5039 if (mlx5e_is_uplink_rep(priv))
5040 flags |= MLX5_TC_FLAG(ESW_OFFLOAD);
5042 flags |= MLX5_TC_FLAG(NIC_OFFLOAD);
5045 case TC_SETUP_CLSFLOWER:
5046 return mlx5e_setup_tc_cls_flower(priv, type_data, flags);
5052 bool mlx5e_tc_update_skb(struct mlx5_cqe64 *cqe,
5053 struct sk_buff *skb)
5055 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5056 u32 chain = 0, chain_tag, reg_b, zone_restore_id;
5057 struct mlx5e_priv *priv = netdev_priv(skb->dev);
5058 struct mlx5e_tc_table *tc = &priv->fs.tc;
5059 struct mlx5_mapped_obj mapped_obj;
5060 struct tc_skb_ext *tc_skb_ext;
5063 reg_b = be32_to_cpu(cqe->ft_metadata);
5065 chain_tag = reg_b & MLX5E_TC_TABLE_CHAIN_TAG_MASK;
5067 err = mapping_find(tc->mapping, chain_tag, &mapped_obj);
5069 netdev_dbg(priv->netdev,
5070 "Couldn't find chain for chain tag: %d, err: %d\n",
5075 if (mapped_obj.type == MLX5_MAPPED_OBJ_CHAIN) {
5076 chain = mapped_obj.chain;
5077 tc_skb_ext = skb_ext_add(skb, TC_SKB_EXT);
5078 if (WARN_ON(!tc_skb_ext))
5081 tc_skb_ext->chain = chain;
5083 zone_restore_id = (reg_b >> REG_MAPPING_SHIFT(NIC_ZONE_RESTORE_TO_REG)) &
5086 if (!mlx5e_tc_ct_restore_flow(tc->ct, skb,
5090 netdev_dbg(priv->netdev, "Invalid mapped object type: %d\n", mapped_obj.type);
5093 #endif /* CONFIG_NET_TC_SKB_EXT */