2 * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <net/flow_dissector.h>
34 #include <net/sch_generic.h>
35 #include <net/pkt_cls.h>
36 #include <net/tc_act/tc_gact.h>
37 #include <net/tc_act/tc_skbedit.h>
38 #include <linux/mlx5/fs.h>
39 #include <linux/mlx5/device.h>
40 #include <linux/rhashtable.h>
41 #include <linux/refcount.h>
42 #include <linux/completion.h>
43 #include <net/tc_act/tc_mirred.h>
44 #include <net/tc_act/tc_vlan.h>
45 #include <net/tc_act/tc_tunnel_key.h>
46 #include <net/tc_act/tc_pedit.h>
47 #include <net/tc_act/tc_csum.h>
49 #include <net/ipv6_stubs.h>
54 #include "eswitch_offloads_chains.h"
57 #include "en/tc_tun.h"
58 #include "lib/devcom.h"
59 #include "lib/geneve.h"
60 #include "diag/en_tc_tracepoint.h"
62 struct mlx5_nic_flow_attr {
65 struct mlx5_modify_hdr *modify_hdr;
68 struct mlx5_flow_table *hairpin_ft;
69 struct mlx5_fc *counter;
72 #define MLX5E_TC_FLOW_BASE (MLX5E_TC_FLAG_LAST_EXPORTED_BIT + 1)
75 MLX5E_TC_FLOW_FLAG_INGRESS = MLX5E_TC_FLAG_INGRESS_BIT,
76 MLX5E_TC_FLOW_FLAG_EGRESS = MLX5E_TC_FLAG_EGRESS_BIT,
77 MLX5E_TC_FLOW_FLAG_ESWITCH = MLX5E_TC_FLAG_ESW_OFFLOAD_BIT,
78 MLX5E_TC_FLOW_FLAG_FT = MLX5E_TC_FLAG_FT_OFFLOAD_BIT,
79 MLX5E_TC_FLOW_FLAG_NIC = MLX5E_TC_FLAG_NIC_OFFLOAD_BIT,
80 MLX5E_TC_FLOW_FLAG_OFFLOADED = MLX5E_TC_FLOW_BASE,
81 MLX5E_TC_FLOW_FLAG_HAIRPIN = MLX5E_TC_FLOW_BASE + 1,
82 MLX5E_TC_FLOW_FLAG_HAIRPIN_RSS = MLX5E_TC_FLOW_BASE + 2,
83 MLX5E_TC_FLOW_FLAG_SLOW = MLX5E_TC_FLOW_BASE + 3,
84 MLX5E_TC_FLOW_FLAG_DUP = MLX5E_TC_FLOW_BASE + 4,
85 MLX5E_TC_FLOW_FLAG_NOT_READY = MLX5E_TC_FLOW_BASE + 5,
86 MLX5E_TC_FLOW_FLAG_DELETED = MLX5E_TC_FLOW_BASE + 6,
89 #define MLX5E_TC_MAX_SPLITS 1
91 /* Helper struct for accessing a struct containing list_head array.
100 * To access the containing struct from one of the list_head items:
101 * 1. Get the helper item from the list_head item using
103 * container_of(list_head item, helper struct type, list_head field)
104 * 2. Get the contining struct from the helper item and its index in the array:
105 * containing struct =
106 * container_of(helper item, containing struct type, helper field[index])
108 struct encap_flow_item {
109 struct mlx5e_encap_entry *e; /* attached encap instance */
110 struct list_head list;
114 struct mlx5e_tc_flow {
115 struct rhash_head node;
116 struct mlx5e_priv *priv;
119 struct mlx5_flow_handle *rule[MLX5E_TC_MAX_SPLITS + 1];
120 /* Flow can be associated with multiple encap IDs.
121 * The number of encaps is bounded by the number of supported
124 struct encap_flow_item encaps[MLX5_MAX_FLOW_FWD_VPORTS];
125 struct mlx5e_tc_flow *peer_flow;
126 struct mlx5e_mod_hdr_entry *mh; /* attached mod header instance */
127 struct list_head mod_hdr; /* flows sharing the same mod hdr ID */
128 struct mlx5e_hairpin_entry *hpe; /* attached hairpin instance */
129 struct list_head hairpin; /* flows sharing the same hairpin */
130 struct list_head peer; /* flows with peer flow */
131 struct list_head unready; /* flows not ready to be offloaded (e.g due to missing route) */
133 struct list_head tmp_list; /* temporary flow list used by neigh update */
135 struct rcu_head rcu_head;
136 struct completion init_done;
138 struct mlx5_esw_flow_attr esw_attr[0];
139 struct mlx5_nic_flow_attr nic_attr[0];
143 struct mlx5e_tc_flow_parse_attr {
144 const struct ip_tunnel_info *tun_info[MLX5_MAX_FLOW_FWD_VPORTS];
145 struct net_device *filter_dev;
146 struct mlx5_flow_spec spec;
147 int num_mod_hdr_actions;
148 int max_mod_hdr_actions;
149 void *mod_hdr_actions;
150 int mirred_ifindex[MLX5_MAX_FLOW_FWD_VPORTS];
153 #define MLX5E_TC_TABLE_NUM_GROUPS 4
154 #define MLX5E_TC_TABLE_MAX_GROUP_SIZE BIT(16)
156 struct mlx5e_hairpin {
157 struct mlx5_hairpin *pair;
159 struct mlx5_core_dev *func_mdev;
160 struct mlx5e_priv *func_priv;
165 struct mlx5e_rqt indir_rqt;
166 u32 indir_tirn[MLX5E_NUM_INDIR_TIRS];
167 struct mlx5e_ttc_table ttc;
170 struct mlx5e_hairpin_entry {
171 /* a node of a hash table which keeps all the hairpin entries */
172 struct hlist_node hairpin_hlist;
174 /* protects flows list */
175 spinlock_t flows_lock;
176 /* flows sharing the same hairpin */
177 struct list_head flows;
178 /* hpe's that were not fully initialized when dead peer update event
179 * function traversed them.
181 struct list_head dead_peer_wait_list;
185 struct mlx5e_hairpin *hp;
187 struct completion res_ready;
195 struct mlx5e_mod_hdr_entry {
196 /* a node of a hash table which keeps all the mod_hdr entries */
197 struct hlist_node mod_hdr_hlist;
199 /* protects flows list */
200 spinlock_t flows_lock;
201 /* flows sharing the same mod_hdr entry */
202 struct list_head flows;
204 struct mod_hdr_key key;
206 struct mlx5_modify_hdr *modify_hdr;
209 struct completion res_ready;
213 #define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)
215 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
216 struct mlx5e_tc_flow *flow);
218 static struct mlx5e_tc_flow *mlx5e_flow_get(struct mlx5e_tc_flow *flow)
220 if (!flow || !refcount_inc_not_zero(&flow->refcnt))
221 return ERR_PTR(-EINVAL);
225 static void mlx5e_flow_put(struct mlx5e_priv *priv,
226 struct mlx5e_tc_flow *flow)
228 if (refcount_dec_and_test(&flow->refcnt)) {
229 mlx5e_tc_del_flow(priv, flow);
230 kfree_rcu(flow, rcu_head);
234 static void __flow_flag_set(struct mlx5e_tc_flow *flow, unsigned long flag)
236 /* Complete all memory stores before setting bit. */
237 smp_mb__before_atomic();
238 set_bit(flag, &flow->flags);
241 #define flow_flag_set(flow, flag) __flow_flag_set(flow, MLX5E_TC_FLOW_FLAG_##flag)
243 static bool __flow_flag_test_and_set(struct mlx5e_tc_flow *flow,
246 /* test_and_set_bit() provides all necessary barriers */
247 return test_and_set_bit(flag, &flow->flags);
250 #define flow_flag_test_and_set(flow, flag) \
251 __flow_flag_test_and_set(flow, \
252 MLX5E_TC_FLOW_FLAG_##flag)
254 static void __flow_flag_clear(struct mlx5e_tc_flow *flow, unsigned long flag)
256 /* Complete all memory stores before clearing bit. */
257 smp_mb__before_atomic();
258 clear_bit(flag, &flow->flags);
261 #define flow_flag_clear(flow, flag) __flow_flag_clear(flow, \
262 MLX5E_TC_FLOW_FLAG_##flag)
264 static bool __flow_flag_test(struct mlx5e_tc_flow *flow, unsigned long flag)
266 bool ret = test_bit(flag, &flow->flags);
268 /* Read fields of flow structure only after checking flags. */
269 smp_mb__after_atomic();
273 #define flow_flag_test(flow, flag) __flow_flag_test(flow, \
274 MLX5E_TC_FLOW_FLAG_##flag)
276 static bool mlx5e_is_eswitch_flow(struct mlx5e_tc_flow *flow)
278 return flow_flag_test(flow, ESWITCH);
281 static bool mlx5e_is_ft_flow(struct mlx5e_tc_flow *flow)
283 return flow_flag_test(flow, FT);
286 static bool mlx5e_is_offloaded_flow(struct mlx5e_tc_flow *flow)
288 return flow_flag_test(flow, OFFLOADED);
291 static inline u32 hash_mod_hdr_info(struct mod_hdr_key *key)
293 return jhash(key->actions,
294 key->num_actions * MLX5_MH_ACT_SZ, 0);
297 static inline int cmp_mod_hdr_info(struct mod_hdr_key *a,
298 struct mod_hdr_key *b)
300 if (a->num_actions != b->num_actions)
303 return memcmp(a->actions, b->actions, a->num_actions * MLX5_MH_ACT_SZ);
306 static struct mod_hdr_tbl *
307 get_mod_hdr_table(struct mlx5e_priv *priv, int namespace)
309 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
311 return namespace == MLX5_FLOW_NAMESPACE_FDB ? &esw->offloads.mod_hdr :
312 &priv->fs.tc.mod_hdr;
315 static struct mlx5e_mod_hdr_entry *
316 mlx5e_mod_hdr_get(struct mod_hdr_tbl *tbl, struct mod_hdr_key *key, u32 hash_key)
318 struct mlx5e_mod_hdr_entry *mh, *found = NULL;
320 hash_for_each_possible(tbl->hlist, mh, mod_hdr_hlist, hash_key) {
321 if (!cmp_mod_hdr_info(&mh->key, key)) {
322 refcount_inc(&mh->refcnt);
331 static void mlx5e_mod_hdr_put(struct mlx5e_priv *priv,
332 struct mlx5e_mod_hdr_entry *mh,
335 struct mod_hdr_tbl *tbl = get_mod_hdr_table(priv, namespace);
337 if (!refcount_dec_and_mutex_lock(&mh->refcnt, &tbl->lock))
339 hash_del(&mh->mod_hdr_hlist);
340 mutex_unlock(&tbl->lock);
342 WARN_ON(!list_empty(&mh->flows));
343 if (mh->compl_result > 0)
344 mlx5_modify_header_dealloc(priv->mdev, mh->modify_hdr);
349 static int get_flow_name_space(struct mlx5e_tc_flow *flow)
351 return mlx5e_is_eswitch_flow(flow) ?
352 MLX5_FLOW_NAMESPACE_FDB : MLX5_FLOW_NAMESPACE_KERNEL;
354 static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
355 struct mlx5e_tc_flow *flow,
356 struct mlx5e_tc_flow_parse_attr *parse_attr)
358 int num_actions, actions_size, namespace, err;
359 struct mlx5e_mod_hdr_entry *mh;
360 struct mod_hdr_tbl *tbl;
361 struct mod_hdr_key key;
364 num_actions = parse_attr->num_mod_hdr_actions;
365 actions_size = MLX5_MH_ACT_SZ * num_actions;
367 key.actions = parse_attr->mod_hdr_actions;
368 key.num_actions = num_actions;
370 hash_key = hash_mod_hdr_info(&key);
372 namespace = get_flow_name_space(flow);
373 tbl = get_mod_hdr_table(priv, namespace);
375 mutex_lock(&tbl->lock);
376 mh = mlx5e_mod_hdr_get(tbl, &key, hash_key);
378 mutex_unlock(&tbl->lock);
379 wait_for_completion(&mh->res_ready);
381 if (mh->compl_result < 0) {
383 goto attach_header_err;
388 mh = kzalloc(sizeof(*mh) + actions_size, GFP_KERNEL);
390 mutex_unlock(&tbl->lock);
394 mh->key.actions = (void *)mh + sizeof(*mh);
395 memcpy(mh->key.actions, key.actions, actions_size);
396 mh->key.num_actions = num_actions;
397 spin_lock_init(&mh->flows_lock);
398 INIT_LIST_HEAD(&mh->flows);
399 refcount_set(&mh->refcnt, 1);
400 init_completion(&mh->res_ready);
402 hash_add(tbl->hlist, &mh->mod_hdr_hlist, hash_key);
403 mutex_unlock(&tbl->lock);
405 mh->modify_hdr = mlx5_modify_header_alloc(priv->mdev, namespace,
408 if (IS_ERR(mh->modify_hdr)) {
409 err = PTR_ERR(mh->modify_hdr);
410 mh->compl_result = err;
411 goto alloc_header_err;
413 mh->compl_result = 1;
414 complete_all(&mh->res_ready);
418 spin_lock(&mh->flows_lock);
419 list_add(&flow->mod_hdr, &mh->flows);
420 spin_unlock(&mh->flows_lock);
421 if (mlx5e_is_eswitch_flow(flow))
422 flow->esw_attr->modify_hdr = mh->modify_hdr;
424 flow->nic_attr->modify_hdr = mh->modify_hdr;
429 complete_all(&mh->res_ready);
431 mlx5e_mod_hdr_put(priv, mh, namespace);
435 static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
436 struct mlx5e_tc_flow *flow)
438 /* flow wasn't fully initialized */
442 spin_lock(&flow->mh->flows_lock);
443 list_del(&flow->mod_hdr);
444 spin_unlock(&flow->mh->flows_lock);
446 mlx5e_mod_hdr_put(priv, flow->mh, get_flow_name_space(flow));
451 struct mlx5_core_dev *mlx5e_hairpin_get_mdev(struct net *net, int ifindex)
453 struct net_device *netdev;
454 struct mlx5e_priv *priv;
456 netdev = __dev_get_by_index(net, ifindex);
457 priv = netdev_priv(netdev);
461 static int mlx5e_hairpin_create_transport(struct mlx5e_hairpin *hp)
463 u32 in[MLX5_ST_SZ_DW(create_tir_in)] = {0};
467 err = mlx5_core_alloc_transport_domain(hp->func_mdev, &hp->tdn);
471 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
473 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
474 MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn[0]);
475 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
477 err = mlx5_core_create_tir(hp->func_mdev, in, MLX5_ST_SZ_BYTES(create_tir_in), &hp->tirn);
484 mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
489 static void mlx5e_hairpin_destroy_transport(struct mlx5e_hairpin *hp)
491 mlx5_core_destroy_tir(hp->func_mdev, hp->tirn);
492 mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
495 static void mlx5e_hairpin_fill_rqt_rqns(struct mlx5e_hairpin *hp, void *rqtc)
497 u32 indirection_rqt[MLX5E_INDIR_RQT_SIZE], rqn;
498 struct mlx5e_priv *priv = hp->func_priv;
499 int i, ix, sz = MLX5E_INDIR_RQT_SIZE;
501 mlx5e_build_default_indir_rqt(indirection_rqt, sz,
504 for (i = 0; i < sz; i++) {
506 if (priv->rss_params.hfunc == ETH_RSS_HASH_XOR)
507 ix = mlx5e_bits_invert(i, ilog2(sz));
508 ix = indirection_rqt[ix];
509 rqn = hp->pair->rqn[ix];
510 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
514 static int mlx5e_hairpin_create_indirect_rqt(struct mlx5e_hairpin *hp)
516 int inlen, err, sz = MLX5E_INDIR_RQT_SIZE;
517 struct mlx5e_priv *priv = hp->func_priv;
518 struct mlx5_core_dev *mdev = priv->mdev;
522 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
523 in = kvzalloc(inlen, GFP_KERNEL);
527 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
529 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
530 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
532 mlx5e_hairpin_fill_rqt_rqns(hp, rqtc);
534 err = mlx5_core_create_rqt(mdev, in, inlen, &hp->indir_rqt.rqtn);
536 hp->indir_rqt.enabled = true;
542 static int mlx5e_hairpin_create_indirect_tirs(struct mlx5e_hairpin *hp)
544 struct mlx5e_priv *priv = hp->func_priv;
545 u32 in[MLX5_ST_SZ_DW(create_tir_in)];
549 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
550 struct mlx5e_tirc_config ttconfig = mlx5e_tirc_get_default_config(tt);
552 memset(in, 0, MLX5_ST_SZ_BYTES(create_tir_in));
553 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
555 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
556 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
557 MLX5_SET(tirc, tirc, indirect_table, hp->indir_rqt.rqtn);
558 mlx5e_build_indir_tir_ctx_hash(&priv->rss_params, &ttconfig, tirc, false);
560 err = mlx5_core_create_tir(hp->func_mdev, in,
561 MLX5_ST_SZ_BYTES(create_tir_in), &hp->indir_tirn[tt]);
563 mlx5_core_warn(hp->func_mdev, "create indirect tirs failed, %d\n", err);
564 goto err_destroy_tirs;
570 for (i = 0; i < tt; i++)
571 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[i]);
575 static void mlx5e_hairpin_destroy_indirect_tirs(struct mlx5e_hairpin *hp)
579 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
580 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[tt]);
583 static void mlx5e_hairpin_set_ttc_params(struct mlx5e_hairpin *hp,
584 struct ttc_params *ttc_params)
586 struct mlx5_flow_table_attr *ft_attr = &ttc_params->ft_attr;
589 memset(ttc_params, 0, sizeof(*ttc_params));
591 ttc_params->any_tt_tirn = hp->tirn;
593 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
594 ttc_params->indir_tirn[tt] = hp->indir_tirn[tt];
596 ft_attr->max_fte = MLX5E_TTC_TABLE_SIZE;
597 ft_attr->level = MLX5E_TC_TTC_FT_LEVEL;
598 ft_attr->prio = MLX5E_TC_PRIO;
601 static int mlx5e_hairpin_rss_init(struct mlx5e_hairpin *hp)
603 struct mlx5e_priv *priv = hp->func_priv;
604 struct ttc_params ttc_params;
607 err = mlx5e_hairpin_create_indirect_rqt(hp);
611 err = mlx5e_hairpin_create_indirect_tirs(hp);
613 goto err_create_indirect_tirs;
615 mlx5e_hairpin_set_ttc_params(hp, &ttc_params);
616 err = mlx5e_create_ttc_table(priv, &ttc_params, &hp->ttc);
618 goto err_create_ttc_table;
620 netdev_dbg(priv->netdev, "add hairpin: using %d channels rss ttc table id %x\n",
621 hp->num_channels, hp->ttc.ft.t->id);
625 err_create_ttc_table:
626 mlx5e_hairpin_destroy_indirect_tirs(hp);
627 err_create_indirect_tirs:
628 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
633 static void mlx5e_hairpin_rss_cleanup(struct mlx5e_hairpin *hp)
635 struct mlx5e_priv *priv = hp->func_priv;
637 mlx5e_destroy_ttc_table(priv, &hp->ttc);
638 mlx5e_hairpin_destroy_indirect_tirs(hp);
639 mlx5e_destroy_rqt(priv, &hp->indir_rqt);
642 static struct mlx5e_hairpin *
643 mlx5e_hairpin_create(struct mlx5e_priv *priv, struct mlx5_hairpin_params *params,
646 struct mlx5_core_dev *func_mdev, *peer_mdev;
647 struct mlx5e_hairpin *hp;
648 struct mlx5_hairpin *pair;
651 hp = kzalloc(sizeof(*hp), GFP_KERNEL);
653 return ERR_PTR(-ENOMEM);
655 func_mdev = priv->mdev;
656 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
658 pair = mlx5_core_hairpin_create(func_mdev, peer_mdev, params);
661 goto create_pair_err;
664 hp->func_mdev = func_mdev;
665 hp->func_priv = priv;
666 hp->num_channels = params->num_channels;
668 err = mlx5e_hairpin_create_transport(hp);
670 goto create_transport_err;
672 if (hp->num_channels > 1) {
673 err = mlx5e_hairpin_rss_init(hp);
681 mlx5e_hairpin_destroy_transport(hp);
682 create_transport_err:
683 mlx5_core_hairpin_destroy(hp->pair);
689 static void mlx5e_hairpin_destroy(struct mlx5e_hairpin *hp)
691 if (hp->num_channels > 1)
692 mlx5e_hairpin_rss_cleanup(hp);
693 mlx5e_hairpin_destroy_transport(hp);
694 mlx5_core_hairpin_destroy(hp->pair);
698 static inline u32 hash_hairpin_info(u16 peer_vhca_id, u8 prio)
700 return (peer_vhca_id << 16 | prio);
703 static struct mlx5e_hairpin_entry *mlx5e_hairpin_get(struct mlx5e_priv *priv,
704 u16 peer_vhca_id, u8 prio)
706 struct mlx5e_hairpin_entry *hpe;
707 u32 hash_key = hash_hairpin_info(peer_vhca_id, prio);
709 hash_for_each_possible(priv->fs.tc.hairpin_tbl, hpe,
710 hairpin_hlist, hash_key) {
711 if (hpe->peer_vhca_id == peer_vhca_id && hpe->prio == prio) {
712 refcount_inc(&hpe->refcnt);
720 static void mlx5e_hairpin_put(struct mlx5e_priv *priv,
721 struct mlx5e_hairpin_entry *hpe)
723 /* no more hairpin flows for us, release the hairpin pair */
724 if (!refcount_dec_and_mutex_lock(&hpe->refcnt, &priv->fs.tc.hairpin_tbl_lock))
726 hash_del(&hpe->hairpin_hlist);
727 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
729 if (!IS_ERR_OR_NULL(hpe->hp)) {
730 netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
731 dev_name(hpe->hp->pair->peer_mdev->device));
733 mlx5e_hairpin_destroy(hpe->hp);
736 WARN_ON(!list_empty(&hpe->flows));
740 #define UNKNOWN_MATCH_PRIO 8
742 static int mlx5e_hairpin_get_prio(struct mlx5e_priv *priv,
743 struct mlx5_flow_spec *spec, u8 *match_prio,
744 struct netlink_ext_ack *extack)
746 void *headers_c, *headers_v;
747 u8 prio_val, prio_mask = 0;
750 #ifdef CONFIG_MLX5_CORE_EN_DCB
751 if (priv->dcbx_dp.trust_state != MLX5_QPTS_TRUST_PCP) {
752 NL_SET_ERR_MSG_MOD(extack,
753 "only PCP trust state supported for hairpin");
757 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, outer_headers);
758 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
760 vlan_present = MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag);
762 prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
763 prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
766 if (!vlan_present || !prio_mask) {
767 prio_val = UNKNOWN_MATCH_PRIO;
768 } else if (prio_mask != 0x7) {
769 NL_SET_ERR_MSG_MOD(extack,
770 "masked priority match not supported for hairpin");
774 *match_prio = prio_val;
778 static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
779 struct mlx5e_tc_flow *flow,
780 struct mlx5e_tc_flow_parse_attr *parse_attr,
781 struct netlink_ext_ack *extack)
783 int peer_ifindex = parse_attr->mirred_ifindex[0];
784 struct mlx5_hairpin_params params;
785 struct mlx5_core_dev *peer_mdev;
786 struct mlx5e_hairpin_entry *hpe;
787 struct mlx5e_hairpin *hp;
794 peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
795 if (!MLX5_CAP_GEN(priv->mdev, hairpin) || !MLX5_CAP_GEN(peer_mdev, hairpin)) {
796 NL_SET_ERR_MSG_MOD(extack, "hairpin is not supported");
800 peer_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
801 err = mlx5e_hairpin_get_prio(priv, &parse_attr->spec, &match_prio,
806 mutex_lock(&priv->fs.tc.hairpin_tbl_lock);
807 hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
809 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
810 wait_for_completion(&hpe->res_ready);
812 if (IS_ERR(hpe->hp)) {
819 hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
821 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
825 spin_lock_init(&hpe->flows_lock);
826 INIT_LIST_HEAD(&hpe->flows);
827 INIT_LIST_HEAD(&hpe->dead_peer_wait_list);
828 hpe->peer_vhca_id = peer_id;
829 hpe->prio = match_prio;
830 refcount_set(&hpe->refcnt, 1);
831 init_completion(&hpe->res_ready);
833 hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
834 hash_hairpin_info(peer_id, match_prio));
835 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
837 params.log_data_size = 15;
838 params.log_data_size = min_t(u8, params.log_data_size,
839 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_wq_data_sz));
840 params.log_data_size = max_t(u8, params.log_data_size,
841 MLX5_CAP_GEN(priv->mdev, log_min_hairpin_wq_data_sz));
843 params.log_num_packets = params.log_data_size -
844 MLX5_MPWRQ_MIN_LOG_STRIDE_SZ(priv->mdev);
845 params.log_num_packets = min_t(u8, params.log_num_packets,
846 MLX5_CAP_GEN(priv->mdev, log_max_hairpin_num_packets));
848 params.q_counter = priv->q_counter;
849 /* set hairpin pair per each 50Gbs share of the link */
850 mlx5e_port_max_linkspeed(priv->mdev, &link_speed);
851 link_speed = max_t(u32, link_speed, 50000);
852 link_speed64 = link_speed;
853 do_div(link_speed64, 50000);
854 params.num_channels = link_speed64;
856 hp = mlx5e_hairpin_create(priv, ¶ms, peer_ifindex);
858 complete_all(&hpe->res_ready);
864 netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
865 hp->tirn, hp->pair->rqn[0],
866 dev_name(hp->pair->peer_mdev->device),
867 hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
870 if (hpe->hp->num_channels > 1) {
871 flow_flag_set(flow, HAIRPIN_RSS);
872 flow->nic_attr->hairpin_ft = hpe->hp->ttc.ft.t;
874 flow->nic_attr->hairpin_tirn = hpe->hp->tirn;
878 spin_lock(&hpe->flows_lock);
879 list_add(&flow->hairpin, &hpe->flows);
880 spin_unlock(&hpe->flows_lock);
885 mlx5e_hairpin_put(priv, hpe);
889 static void mlx5e_hairpin_flow_del(struct mlx5e_priv *priv,
890 struct mlx5e_tc_flow *flow)
892 /* flow wasn't fully initialized */
896 spin_lock(&flow->hpe->flows_lock);
897 list_del(&flow->hairpin);
898 spin_unlock(&flow->hpe->flows_lock);
900 mlx5e_hairpin_put(priv, flow->hpe);
905 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
906 struct mlx5e_tc_flow_parse_attr *parse_attr,
907 struct mlx5e_tc_flow *flow,
908 struct netlink_ext_ack *extack)
910 struct mlx5_flow_context *flow_context = &parse_attr->spec.flow_context;
911 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
912 struct mlx5_core_dev *dev = priv->mdev;
913 struct mlx5_flow_destination dest[2] = {};
914 struct mlx5_flow_act flow_act = {
915 .action = attr->action,
916 .flags = FLOW_ACT_NO_APPEND,
918 struct mlx5_fc *counter = NULL;
919 int err, dest_ix = 0;
921 flow_context->flags |= FLOW_CONTEXT_HAS_TAG;
922 flow_context->flow_tag = attr->flow_tag;
924 if (flow_flag_test(flow, HAIRPIN)) {
925 err = mlx5e_hairpin_flow_add(priv, flow, parse_attr, extack);
929 if (flow_flag_test(flow, HAIRPIN_RSS)) {
930 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
931 dest[dest_ix].ft = attr->hairpin_ft;
933 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_TIR;
934 dest[dest_ix].tir_num = attr->hairpin_tirn;
937 } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
938 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
939 dest[dest_ix].ft = priv->fs.vlan.ft.t;
943 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
944 counter = mlx5_fc_create(dev, true);
946 return PTR_ERR(counter);
948 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
949 dest[dest_ix].counter_id = mlx5_fc_id(counter);
951 attr->counter = counter;
954 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
955 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
956 flow_act.modify_hdr = attr->modify_hdr;
957 kfree(parse_attr->mod_hdr_actions);
962 mutex_lock(&priv->fs.tc.t_lock);
963 if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
964 struct mlx5_flow_table_attr ft_attr = {};
965 int tc_grp_size, tc_tbl_size, tc_num_grps;
966 u32 max_flow_counter;
968 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
969 MLX5_CAP_GEN(dev, max_flow_counter_15_0);
971 tc_grp_size = min_t(int, max_flow_counter, MLX5E_TC_TABLE_MAX_GROUP_SIZE);
973 tc_tbl_size = min_t(int, tc_grp_size * MLX5E_TC_TABLE_NUM_GROUPS,
974 BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev, log_max_ft_size)));
975 tc_num_grps = MLX5E_TC_TABLE_NUM_GROUPS;
977 ft_attr.prio = MLX5E_TC_PRIO;
978 ft_attr.max_fte = tc_tbl_size;
979 ft_attr.level = MLX5E_TC_FT_LEVEL;
980 ft_attr.autogroup.max_num_groups = tc_num_grps;
982 mlx5_create_auto_grouped_flow_table(priv->fs.ns,
984 if (IS_ERR(priv->fs.tc.t)) {
985 mutex_unlock(&priv->fs.tc.t_lock);
986 NL_SET_ERR_MSG_MOD(extack,
987 "Failed to create tc offload table\n");
988 netdev_err(priv->netdev,
989 "Failed to create tc offload table\n");
990 return PTR_ERR(priv->fs.tc.t);
994 if (attr->match_level != MLX5_MATCH_NONE)
995 parse_attr->spec.match_criteria_enable |= MLX5_MATCH_OUTER_HEADERS;
997 flow->rule[0] = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
998 &flow_act, dest, dest_ix);
999 mutex_unlock(&priv->fs.tc.t_lock);
1001 return PTR_ERR_OR_ZERO(flow->rule[0]);
1004 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
1005 struct mlx5e_tc_flow *flow)
1007 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
1008 struct mlx5_fc *counter = NULL;
1010 counter = attr->counter;
1011 if (!IS_ERR_OR_NULL(flow->rule[0]))
1012 mlx5_del_flow_rules(flow->rule[0]);
1013 mlx5_fc_destroy(priv->mdev, counter);
1015 mutex_lock(&priv->fs.tc.t_lock);
1016 if (!mlx5e_tc_num_filters(priv, MLX5_TC_FLAG(NIC_OFFLOAD)) && priv->fs.tc.t) {
1017 mlx5_destroy_flow_table(priv->fs.tc.t);
1018 priv->fs.tc.t = NULL;
1020 mutex_unlock(&priv->fs.tc.t_lock);
1022 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1023 mlx5e_detach_mod_hdr(priv, flow);
1025 if (flow_flag_test(flow, HAIRPIN))
1026 mlx5e_hairpin_flow_del(priv, flow);
1029 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1030 struct mlx5e_tc_flow *flow, int out_index);
1032 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
1033 struct mlx5e_tc_flow *flow,
1034 struct net_device *mirred_dev,
1036 struct netlink_ext_ack *extack,
1037 struct net_device **encap_dev,
1040 static struct mlx5_flow_handle *
1041 mlx5e_tc_offload_fdb_rules(struct mlx5_eswitch *esw,
1042 struct mlx5e_tc_flow *flow,
1043 struct mlx5_flow_spec *spec,
1044 struct mlx5_esw_flow_attr *attr)
1046 struct mlx5_flow_handle *rule;
1048 rule = mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
1052 if (attr->split_count) {
1053 flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, spec, attr);
1054 if (IS_ERR(flow->rule[1])) {
1055 mlx5_eswitch_del_offloaded_rule(esw, rule, attr);
1056 return flow->rule[1];
1064 mlx5e_tc_unoffload_fdb_rules(struct mlx5_eswitch *esw,
1065 struct mlx5e_tc_flow *flow,
1066 struct mlx5_esw_flow_attr *attr)
1068 flow_flag_clear(flow, OFFLOADED);
1070 if (attr->split_count)
1071 mlx5_eswitch_del_fwd_rule(esw, flow->rule[1], attr);
1073 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
1076 static struct mlx5_flow_handle *
1077 mlx5e_tc_offload_to_slow_path(struct mlx5_eswitch *esw,
1078 struct mlx5e_tc_flow *flow,
1079 struct mlx5_flow_spec *spec)
1081 struct mlx5_esw_flow_attr slow_attr;
1082 struct mlx5_flow_handle *rule;
1084 memcpy(&slow_attr, flow->esw_attr, sizeof(slow_attr));
1085 slow_attr.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1086 slow_attr.split_count = 0;
1087 slow_attr.flags |= MLX5_ESW_ATTR_FLAG_SLOW_PATH;
1089 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, &slow_attr);
1091 flow_flag_set(flow, SLOW);
1097 mlx5e_tc_unoffload_from_slow_path(struct mlx5_eswitch *esw,
1098 struct mlx5e_tc_flow *flow)
1100 struct mlx5_esw_flow_attr slow_attr;
1102 memcpy(&slow_attr, flow->esw_attr, sizeof(slow_attr));
1103 slow_attr.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1104 slow_attr.split_count = 0;
1105 slow_attr.flags |= MLX5_ESW_ATTR_FLAG_SLOW_PATH;
1106 mlx5e_tc_unoffload_fdb_rules(esw, flow, &slow_attr);
1107 flow_flag_clear(flow, SLOW);
1110 /* Caller must obtain uplink_priv->unready_flows_lock mutex before calling this
1113 static void unready_flow_add(struct mlx5e_tc_flow *flow,
1114 struct list_head *unready_flows)
1116 flow_flag_set(flow, NOT_READY);
1117 list_add_tail(&flow->unready, unready_flows);
1120 /* Caller must obtain uplink_priv->unready_flows_lock mutex before calling this
1123 static void unready_flow_del(struct mlx5e_tc_flow *flow)
1125 list_del(&flow->unready);
1126 flow_flag_clear(flow, NOT_READY);
1129 static void add_unready_flow(struct mlx5e_tc_flow *flow)
1131 struct mlx5_rep_uplink_priv *uplink_priv;
1132 struct mlx5e_rep_priv *rpriv;
1133 struct mlx5_eswitch *esw;
1135 esw = flow->priv->mdev->priv.eswitch;
1136 rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1137 uplink_priv = &rpriv->uplink_priv;
1139 mutex_lock(&uplink_priv->unready_flows_lock);
1140 unready_flow_add(flow, &uplink_priv->unready_flows);
1141 mutex_unlock(&uplink_priv->unready_flows_lock);
1144 static void remove_unready_flow(struct mlx5e_tc_flow *flow)
1146 struct mlx5_rep_uplink_priv *uplink_priv;
1147 struct mlx5e_rep_priv *rpriv;
1148 struct mlx5_eswitch *esw;
1150 esw = flow->priv->mdev->priv.eswitch;
1151 rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1152 uplink_priv = &rpriv->uplink_priv;
1154 mutex_lock(&uplink_priv->unready_flows_lock);
1155 unready_flow_del(flow);
1156 mutex_unlock(&uplink_priv->unready_flows_lock);
1160 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
1161 struct mlx5e_tc_flow *flow,
1162 struct netlink_ext_ack *extack)
1164 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1165 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1166 struct mlx5e_tc_flow_parse_attr *parse_attr = attr->parse_attr;
1167 struct net_device *out_dev, *encap_dev = NULL;
1168 struct mlx5_fc *counter = NULL;
1169 struct mlx5e_rep_priv *rpriv;
1170 struct mlx5e_priv *out_priv;
1171 bool encap_valid = true;
1172 u32 max_prio, max_chain;
1176 if (!mlx5_esw_chains_prios_supported(esw) && attr->prio != 1) {
1177 NL_SET_ERR_MSG_MOD(extack,
1178 "E-switch priorities unsupported, upgrade FW");
1182 /* We check chain range only for tc flows.
1183 * For ft flows, we checked attr->chain was originally 0 and set it to
1184 * FDB_FT_CHAIN which is outside tc range.
1185 * See mlx5e_rep_setup_ft_cb().
1187 max_chain = mlx5_esw_chains_get_chain_range(esw);
1188 if (!mlx5e_is_ft_flow(flow) && attr->chain > max_chain) {
1189 NL_SET_ERR_MSG_MOD(extack,
1190 "Requested chain is out of supported range");
1194 max_prio = mlx5_esw_chains_get_prio_range(esw);
1195 if (attr->prio > max_prio) {
1196 NL_SET_ERR_MSG_MOD(extack,
1197 "Requested priority is out of supported range");
1201 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++) {
1204 if (!(attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP))
1207 mirred_ifindex = parse_attr->mirred_ifindex[out_index];
1208 out_dev = __dev_get_by_index(dev_net(priv->netdev),
1210 err = mlx5e_attach_encap(priv, flow, out_dev, out_index,
1211 extack, &encap_dev, &encap_valid);
1215 out_priv = netdev_priv(encap_dev);
1216 rpriv = out_priv->ppriv;
1217 attr->dests[out_index].rep = rpriv->rep;
1218 attr->dests[out_index].mdev = out_priv->mdev;
1221 err = mlx5_eswitch_add_vlan_action(esw, attr);
1225 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
1226 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
1227 kfree(parse_attr->mod_hdr_actions);
1232 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1233 counter = mlx5_fc_create(attr->counter_dev, true);
1234 if (IS_ERR(counter))
1235 return PTR_ERR(counter);
1237 attr->counter = counter;
1240 /* we get here if one of the following takes place:
1241 * (1) there's no error
1242 * (2) there's an encap action and we don't have valid neigh
1245 flow->rule[0] = mlx5e_tc_offload_to_slow_path(esw, flow, &parse_attr->spec);
1247 flow->rule[0] = mlx5e_tc_offload_fdb_rules(esw, flow, &parse_attr->spec, attr);
1249 if (IS_ERR(flow->rule[0]))
1250 return PTR_ERR(flow->rule[0]);
1252 flow_flag_set(flow, OFFLOADED);
1257 static bool mlx5_flow_has_geneve_opt(struct mlx5e_tc_flow *flow)
1259 struct mlx5_flow_spec *spec = &flow->esw_attr->parse_attr->spec;
1260 void *headers_v = MLX5_ADDR_OF(fte_match_param,
1263 u32 geneve_tlv_opt_0_data = MLX5_GET(fte_match_set_misc3,
1265 geneve_tlv_option_0_data);
1267 return !!geneve_tlv_opt_0_data;
1270 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
1271 struct mlx5e_tc_flow *flow)
1273 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1274 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1277 if (flow_flag_test(flow, NOT_READY)) {
1278 remove_unready_flow(flow);
1279 kvfree(attr->parse_attr);
1283 if (mlx5e_is_offloaded_flow(flow)) {
1284 if (flow_flag_test(flow, SLOW))
1285 mlx5e_tc_unoffload_from_slow_path(esw, flow);
1287 mlx5e_tc_unoffload_fdb_rules(esw, flow, attr);
1290 if (mlx5_flow_has_geneve_opt(flow))
1291 mlx5_geneve_tlv_option_del(priv->mdev->geneve);
1293 mlx5_eswitch_del_vlan_action(esw, attr);
1295 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++)
1296 if (attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP) {
1297 mlx5e_detach_encap(priv, flow, out_index);
1298 kfree(attr->parse_attr->tun_info[out_index]);
1300 kvfree(attr->parse_attr);
1302 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1303 mlx5e_detach_mod_hdr(priv, flow);
1305 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
1306 mlx5_fc_destroy(attr->counter_dev, attr->counter);
1309 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
1310 struct mlx5e_encap_entry *e,
1311 struct list_head *flow_list)
1313 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1314 struct mlx5_esw_flow_attr *esw_attr;
1315 struct mlx5_flow_handle *rule;
1316 struct mlx5_flow_spec *spec;
1317 struct mlx5e_tc_flow *flow;
1320 e->pkt_reformat = mlx5_packet_reformat_alloc(priv->mdev,
1322 e->encap_size, e->encap_header,
1323 MLX5_FLOW_NAMESPACE_FDB);
1324 if (IS_ERR(e->pkt_reformat)) {
1325 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %lu\n",
1326 PTR_ERR(e->pkt_reformat));
1329 e->flags |= MLX5_ENCAP_ENTRY_VALID;
1330 mlx5e_rep_queue_neigh_stats_work(priv);
1332 list_for_each_entry(flow, flow_list, tmp_list) {
1333 bool all_flow_encaps_valid = true;
1336 if (!mlx5e_is_offloaded_flow(flow))
1338 esw_attr = flow->esw_attr;
1339 spec = &esw_attr->parse_attr->spec;
1341 esw_attr->dests[flow->tmp_efi_index].pkt_reformat = e->pkt_reformat;
1342 esw_attr->dests[flow->tmp_efi_index].flags |= MLX5_ESW_DEST_ENCAP_VALID;
1343 /* Flow can be associated with multiple encap entries.
1344 * Before offloading the flow verify that all of them have
1345 * a valid neighbour.
1347 for (i = 0; i < MLX5_MAX_FLOW_FWD_VPORTS; i++) {
1348 if (!(esw_attr->dests[i].flags & MLX5_ESW_DEST_ENCAP))
1350 if (!(esw_attr->dests[i].flags & MLX5_ESW_DEST_ENCAP_VALID)) {
1351 all_flow_encaps_valid = false;
1355 /* Do not offload flows with unresolved neighbors */
1356 if (!all_flow_encaps_valid)
1358 /* update from slow path rule to encap rule */
1359 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, esw_attr);
1361 err = PTR_ERR(rule);
1362 mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
1367 mlx5e_tc_unoffload_from_slow_path(esw, flow);
1368 flow->rule[0] = rule;
1369 /* was unset when slow path rule removed */
1370 flow_flag_set(flow, OFFLOADED);
1374 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
1375 struct mlx5e_encap_entry *e,
1376 struct list_head *flow_list)
1378 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1379 struct mlx5_flow_handle *rule;
1380 struct mlx5_flow_spec *spec;
1381 struct mlx5e_tc_flow *flow;
1384 list_for_each_entry(flow, flow_list, tmp_list) {
1385 if (!mlx5e_is_offloaded_flow(flow))
1387 spec = &flow->esw_attr->parse_attr->spec;
1389 /* update from encap rule to slow path rule */
1390 rule = mlx5e_tc_offload_to_slow_path(esw, flow, spec);
1391 /* mark the flow's encap dest as non-valid */
1392 flow->esw_attr->dests[flow->tmp_efi_index].flags &= ~MLX5_ESW_DEST_ENCAP_VALID;
1395 err = PTR_ERR(rule);
1396 mlx5_core_warn(priv->mdev, "Failed to update slow path (encap) flow, %d\n",
1401 mlx5e_tc_unoffload_fdb_rules(esw, flow, flow->esw_attr);
1402 flow->rule[0] = rule;
1403 /* was unset when fast path rule removed */
1404 flow_flag_set(flow, OFFLOADED);
1407 /* we know that the encap is valid */
1408 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
1409 mlx5_packet_reformat_dealloc(priv->mdev, e->pkt_reformat);
1412 static struct mlx5_fc *mlx5e_tc_get_counter(struct mlx5e_tc_flow *flow)
1414 if (mlx5e_is_eswitch_flow(flow))
1415 return flow->esw_attr->counter;
1417 return flow->nic_attr->counter;
1420 /* Takes reference to all flows attached to encap and adds the flows to
1421 * flow_list using 'tmp_list' list_head in mlx5e_tc_flow.
1423 void mlx5e_take_all_encap_flows(struct mlx5e_encap_entry *e, struct list_head *flow_list)
1425 struct encap_flow_item *efi;
1426 struct mlx5e_tc_flow *flow;
1428 list_for_each_entry(efi, &e->flows, list) {
1429 flow = container_of(efi, struct mlx5e_tc_flow, encaps[efi->index]);
1430 if (IS_ERR(mlx5e_flow_get(flow)))
1432 wait_for_completion(&flow->init_done);
1434 flow->tmp_efi_index = efi->index;
1435 list_add(&flow->tmp_list, flow_list);
1439 /* Iterate over tmp_list of flows attached to flow_list head. */
1440 void mlx5e_put_encap_flow_list(struct mlx5e_priv *priv, struct list_head *flow_list)
1442 struct mlx5e_tc_flow *flow, *tmp;
1444 list_for_each_entry_safe(flow, tmp, flow_list, tmp_list)
1445 mlx5e_flow_put(priv, flow);
1448 static struct mlx5e_encap_entry *
1449 mlx5e_get_next_valid_encap(struct mlx5e_neigh_hash_entry *nhe,
1450 struct mlx5e_encap_entry *e)
1452 struct mlx5e_encap_entry *next = NULL;
1457 /* find encap with non-zero reference counter value */
1459 list_next_or_null_rcu(&nhe->encap_list,
1461 struct mlx5e_encap_entry,
1463 list_first_or_null_rcu(&nhe->encap_list,
1464 struct mlx5e_encap_entry,
1467 next = list_next_or_null_rcu(&nhe->encap_list,
1469 struct mlx5e_encap_entry,
1471 if (mlx5e_encap_take(next))
1476 /* release starting encap */
1478 mlx5e_encap_put(netdev_priv(e->out_dev), e);
1482 /* wait for encap to be fully initialized */
1483 wait_for_completion(&next->res_ready);
1484 /* continue searching if encap entry is not in valid state after completion */
1485 if (!(next->flags & MLX5_ENCAP_ENTRY_VALID)) {
1493 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
1495 struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
1496 struct mlx5e_encap_entry *e = NULL;
1497 struct mlx5e_tc_flow *flow;
1498 struct mlx5_fc *counter;
1499 struct neigh_table *tbl;
1500 bool neigh_used = false;
1501 struct neighbour *n;
1504 if (m_neigh->family == AF_INET)
1506 #if IS_ENABLED(CONFIG_IPV6)
1507 else if (m_neigh->family == AF_INET6)
1508 tbl = ipv6_stub->nd_tbl;
1513 /* mlx5e_get_next_valid_encap() releases previous encap before returning
1516 while ((e = mlx5e_get_next_valid_encap(nhe, e)) != NULL) {
1517 struct mlx5e_priv *priv = netdev_priv(e->out_dev);
1518 struct encap_flow_item *efi, *tmp;
1519 struct mlx5_eswitch *esw;
1520 LIST_HEAD(flow_list);
1522 esw = priv->mdev->priv.eswitch;
1523 mutex_lock(&esw->offloads.encap_tbl_lock);
1524 list_for_each_entry_safe(efi, tmp, &e->flows, list) {
1525 flow = container_of(efi, struct mlx5e_tc_flow,
1526 encaps[efi->index]);
1527 if (IS_ERR(mlx5e_flow_get(flow)))
1529 list_add(&flow->tmp_list, &flow_list);
1531 if (mlx5e_is_offloaded_flow(flow)) {
1532 counter = mlx5e_tc_get_counter(flow);
1533 lastuse = mlx5_fc_query_lastuse(counter);
1534 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
1540 mutex_unlock(&esw->offloads.encap_tbl_lock);
1542 mlx5e_put_encap_flow_list(priv, &flow_list);
1544 /* release current encap before breaking the loop */
1545 mlx5e_encap_put(priv, e);
1550 trace_mlx5e_tc_update_neigh_used_value(nhe, neigh_used);
1553 nhe->reported_lastuse = jiffies;
1555 /* find the relevant neigh according to the cached device and
1558 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
1562 neigh_event_send(n, NULL);
1567 static void mlx5e_encap_dealloc(struct mlx5e_priv *priv, struct mlx5e_encap_entry *e)
1569 WARN_ON(!list_empty(&e->flows));
1571 if (e->compl_result > 0) {
1572 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1574 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1575 mlx5_packet_reformat_dealloc(priv->mdev, e->pkt_reformat);
1579 kfree(e->encap_header);
1583 void mlx5e_encap_put(struct mlx5e_priv *priv, struct mlx5e_encap_entry *e)
1585 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1587 if (!refcount_dec_and_mutex_lock(&e->refcnt, &esw->offloads.encap_tbl_lock))
1589 hash_del_rcu(&e->encap_hlist);
1590 mutex_unlock(&esw->offloads.encap_tbl_lock);
1592 mlx5e_encap_dealloc(priv, e);
1595 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1596 struct mlx5e_tc_flow *flow, int out_index)
1598 struct mlx5e_encap_entry *e = flow->encaps[out_index].e;
1599 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1601 /* flow wasn't fully initialized */
1605 mutex_lock(&esw->offloads.encap_tbl_lock);
1606 list_del(&flow->encaps[out_index].list);
1607 flow->encaps[out_index].e = NULL;
1608 if (!refcount_dec_and_test(&e->refcnt)) {
1609 mutex_unlock(&esw->offloads.encap_tbl_lock);
1612 hash_del_rcu(&e->encap_hlist);
1613 mutex_unlock(&esw->offloads.encap_tbl_lock);
1615 mlx5e_encap_dealloc(priv, e);
1618 static void __mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1620 struct mlx5_eswitch *esw = flow->priv->mdev->priv.eswitch;
1622 if (!flow_flag_test(flow, ESWITCH) ||
1623 !flow_flag_test(flow, DUP))
1626 mutex_lock(&esw->offloads.peer_mutex);
1627 list_del(&flow->peer);
1628 mutex_unlock(&esw->offloads.peer_mutex);
1630 flow_flag_clear(flow, DUP);
1632 if (refcount_dec_and_test(&flow->peer_flow->refcnt)) {
1633 mlx5e_tc_del_fdb_flow(flow->peer_flow->priv, flow->peer_flow);
1634 kfree(flow->peer_flow);
1637 flow->peer_flow = NULL;
1640 static void mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1642 struct mlx5_core_dev *dev = flow->priv->mdev;
1643 struct mlx5_devcom *devcom = dev->priv.devcom;
1644 struct mlx5_eswitch *peer_esw;
1646 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1650 __mlx5e_tc_del_fdb_peer_flow(flow);
1651 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1654 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1655 struct mlx5e_tc_flow *flow)
1657 if (mlx5e_is_eswitch_flow(flow)) {
1658 mlx5e_tc_del_fdb_peer_flow(flow);
1659 mlx5e_tc_del_fdb_flow(priv, flow);
1661 mlx5e_tc_del_nic_flow(priv, flow);
1666 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1667 struct mlx5_flow_spec *spec,
1668 struct flow_cls_offload *f,
1669 struct net_device *filter_dev, u8 *match_level)
1671 struct netlink_ext_ack *extack = f->common.extack;
1672 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1674 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1676 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1679 err = mlx5e_tc_tun_parse(filter_dev, priv, spec, f,
1680 headers_c, headers_v, match_level);
1682 NL_SET_ERR_MSG_MOD(extack,
1683 "failed to parse tunnel attributes");
1687 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
1688 struct flow_match_control match;
1691 flow_rule_match_enc_control(rule, &match);
1692 addr_type = match.key->addr_type;
1694 /* For tunnel addr_type used same key id`s as for non-tunnel */
1695 if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1696 struct flow_match_ipv4_addrs match;
1698 flow_rule_match_enc_ipv4_addrs(rule, &match);
1699 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1700 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1701 ntohl(match.mask->src));
1702 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1703 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1704 ntohl(match.key->src));
1706 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1707 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1708 ntohl(match.mask->dst));
1709 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1710 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1711 ntohl(match.key->dst));
1713 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c,
1715 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1717 } else if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1718 struct flow_match_ipv6_addrs match;
1720 flow_rule_match_enc_ipv6_addrs(rule, &match);
1721 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1722 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1723 &match.mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout,
1725 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1726 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1727 &match.key->src, MLX5_FLD_SZ_BYTES(ipv6_layout,
1730 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1731 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1732 &match.mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout,
1734 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1735 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1736 &match.key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout,
1739 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c,
1741 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1746 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_IP)) {
1747 struct flow_match_ip match;
1749 flow_rule_match_enc_ip(rule, &match);
1750 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn,
1751 match.mask->tos & 0x3);
1752 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn,
1753 match.key->tos & 0x3);
1755 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp,
1756 match.mask->tos >> 2);
1757 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp,
1758 match.key->tos >> 2);
1760 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit,
1762 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit,
1765 if (match.mask->ttl &&
1766 !MLX5_CAP_ESW_FLOWTABLE_FDB
1768 ft_field_support.outer_ipv4_ttl)) {
1769 NL_SET_ERR_MSG_MOD(extack,
1770 "Matching on TTL is not supported");
1776 /* Enforce DMAC when offloading incoming tunneled flows.
1777 * Flow counters require a match on the DMAC.
1779 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
1780 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
1781 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1782 dmac_47_16), priv->netdev->dev_addr);
1784 /* let software handle IP fragments */
1785 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1786 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
1791 static void *get_match_headers_criteria(u32 flags,
1792 struct mlx5_flow_spec *spec)
1794 return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
1795 MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1797 MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1801 static void *get_match_headers_value(u32 flags,
1802 struct mlx5_flow_spec *spec)
1804 return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
1805 MLX5_ADDR_OF(fte_match_param, spec->match_value,
1807 MLX5_ADDR_OF(fte_match_param, spec->match_value,
1811 static int mlx5e_flower_parse_meta(struct net_device *filter_dev,
1812 struct flow_cls_offload *f)
1814 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1815 struct netlink_ext_ack *extack = f->common.extack;
1816 struct net_device *ingress_dev;
1817 struct flow_match_meta match;
1819 if (!flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_META))
1822 flow_rule_match_meta(rule, &match);
1823 if (match.mask->ingress_ifindex != 0xFFFFFFFF) {
1824 NL_SET_ERR_MSG_MOD(extack, "Unsupported ingress ifindex mask");
1828 ingress_dev = __dev_get_by_index(dev_net(filter_dev),
1829 match.key->ingress_ifindex);
1831 NL_SET_ERR_MSG_MOD(extack,
1832 "Can't find the ingress port to match on");
1836 if (ingress_dev != filter_dev) {
1837 NL_SET_ERR_MSG_MOD(extack,
1838 "Can't match on the ingress filter port");
1845 static int __parse_cls_flower(struct mlx5e_priv *priv,
1846 struct mlx5_flow_spec *spec,
1847 struct flow_cls_offload *f,
1848 struct net_device *filter_dev,
1849 u8 *inner_match_level, u8 *outer_match_level)
1851 struct netlink_ext_ack *extack = f->common.extack;
1852 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1854 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1856 void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1858 void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1860 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1861 struct flow_dissector *dissector = rule->match.dissector;
1867 match_level = outer_match_level;
1869 if (dissector->used_keys &
1870 ~(BIT(FLOW_DISSECTOR_KEY_META) |
1871 BIT(FLOW_DISSECTOR_KEY_CONTROL) |
1872 BIT(FLOW_DISSECTOR_KEY_BASIC) |
1873 BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
1874 BIT(FLOW_DISSECTOR_KEY_VLAN) |
1875 BIT(FLOW_DISSECTOR_KEY_CVLAN) |
1876 BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
1877 BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
1878 BIT(FLOW_DISSECTOR_KEY_PORTS) |
1879 BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
1880 BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
1881 BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
1882 BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
1883 BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
1884 BIT(FLOW_DISSECTOR_KEY_TCP) |
1885 BIT(FLOW_DISSECTOR_KEY_IP) |
1886 BIT(FLOW_DISSECTOR_KEY_ENC_IP) |
1887 BIT(FLOW_DISSECTOR_KEY_ENC_OPTS))) {
1888 NL_SET_ERR_MSG_MOD(extack, "Unsupported key");
1889 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
1890 dissector->used_keys);
1894 if (mlx5e_get_tc_tun(filter_dev)) {
1895 if (parse_tunnel_attr(priv, spec, f, filter_dev,
1899 /* At this point, header pointers should point to the inner
1900 * headers, outer header were already set by parse_tunnel_attr
1902 match_level = inner_match_level;
1903 headers_c = get_match_headers_criteria(MLX5_FLOW_CONTEXT_ACTION_DECAP,
1905 headers_v = get_match_headers_value(MLX5_FLOW_CONTEXT_ACTION_DECAP,
1909 err = mlx5e_flower_parse_meta(filter_dev, f);
1913 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
1914 struct flow_match_basic match;
1916 flow_rule_match_basic(rule, &match);
1917 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1918 ntohs(match.mask->n_proto));
1919 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1920 ntohs(match.key->n_proto));
1922 if (match.mask->n_proto)
1923 *match_level = MLX5_MATCH_L2;
1925 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN) ||
1926 is_vlan_dev(filter_dev)) {
1927 struct flow_dissector_key_vlan filter_dev_mask;
1928 struct flow_dissector_key_vlan filter_dev_key;
1929 struct flow_match_vlan match;
1931 if (is_vlan_dev(filter_dev)) {
1932 match.key = &filter_dev_key;
1933 match.key->vlan_id = vlan_dev_vlan_id(filter_dev);
1934 match.key->vlan_tpid = vlan_dev_vlan_proto(filter_dev);
1935 match.key->vlan_priority = 0;
1936 match.mask = &filter_dev_mask;
1937 memset(match.mask, 0xff, sizeof(*match.mask));
1938 match.mask->vlan_priority = 0;
1940 flow_rule_match_vlan(rule, &match);
1942 if (match.mask->vlan_id ||
1943 match.mask->vlan_priority ||
1944 match.mask->vlan_tpid) {
1945 if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
1946 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1948 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1951 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1953 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1957 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid,
1958 match.mask->vlan_id);
1959 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid,
1960 match.key->vlan_id);
1962 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio,
1963 match.mask->vlan_priority);
1964 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio,
1965 match.key->vlan_priority);
1967 *match_level = MLX5_MATCH_L2;
1969 } else if (*match_level != MLX5_MATCH_NONE) {
1970 /* cvlan_tag enabled in match criteria and
1971 * disabled in match value means both S & C tags
1972 * don't exist (untagged of both)
1974 MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
1975 *match_level = MLX5_MATCH_L2;
1978 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
1979 struct flow_match_vlan match;
1981 flow_rule_match_cvlan(rule, &match);
1982 if (match.mask->vlan_id ||
1983 match.mask->vlan_priority ||
1984 match.mask->vlan_tpid) {
1985 if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
1986 MLX5_SET(fte_match_set_misc, misc_c,
1987 outer_second_svlan_tag, 1);
1988 MLX5_SET(fte_match_set_misc, misc_v,
1989 outer_second_svlan_tag, 1);
1991 MLX5_SET(fte_match_set_misc, misc_c,
1992 outer_second_cvlan_tag, 1);
1993 MLX5_SET(fte_match_set_misc, misc_v,
1994 outer_second_cvlan_tag, 1);
1997 MLX5_SET(fte_match_set_misc, misc_c, outer_second_vid,
1998 match.mask->vlan_id);
1999 MLX5_SET(fte_match_set_misc, misc_v, outer_second_vid,
2000 match.key->vlan_id);
2001 MLX5_SET(fte_match_set_misc, misc_c, outer_second_prio,
2002 match.mask->vlan_priority);
2003 MLX5_SET(fte_match_set_misc, misc_v, outer_second_prio,
2004 match.key->vlan_priority);
2006 *match_level = MLX5_MATCH_L2;
2010 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2011 struct flow_match_eth_addrs match;
2013 flow_rule_match_eth_addrs(rule, &match);
2014 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2017 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2021 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2024 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2028 if (!is_zero_ether_addr(match.mask->src) ||
2029 !is_zero_ether_addr(match.mask->dst))
2030 *match_level = MLX5_MATCH_L2;
2033 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
2034 struct flow_match_control match;
2036 flow_rule_match_control(rule, &match);
2037 addr_type = match.key->addr_type;
2039 /* the HW doesn't support frag first/later */
2040 if (match.mask->flags & FLOW_DIS_FIRST_FRAG)
2043 if (match.mask->flags & FLOW_DIS_IS_FRAGMENT) {
2044 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
2045 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
2046 match.key->flags & FLOW_DIS_IS_FRAGMENT);
2048 /* the HW doesn't need L3 inline to match on frag=no */
2049 if (!(match.key->flags & FLOW_DIS_IS_FRAGMENT))
2050 *match_level = MLX5_MATCH_L2;
2051 /* *** L2 attributes parsing up to here *** */
2053 *match_level = MLX5_MATCH_L3;
2057 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2058 struct flow_match_basic match;
2060 flow_rule_match_basic(rule, &match);
2061 ip_proto = match.key->ip_proto;
2063 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
2064 match.mask->ip_proto);
2065 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
2066 match.key->ip_proto);
2068 if (match.mask->ip_proto)
2069 *match_level = MLX5_MATCH_L3;
2072 if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2073 struct flow_match_ipv4_addrs match;
2075 flow_rule_match_ipv4_addrs(rule, &match);
2076 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2077 src_ipv4_src_ipv6.ipv4_layout.ipv4),
2078 &match.mask->src, sizeof(match.mask->src));
2079 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2080 src_ipv4_src_ipv6.ipv4_layout.ipv4),
2081 &match.key->src, sizeof(match.key->src));
2082 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2083 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2084 &match.mask->dst, sizeof(match.mask->dst));
2085 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2086 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2087 &match.key->dst, sizeof(match.key->dst));
2089 if (match.mask->src || match.mask->dst)
2090 *match_level = MLX5_MATCH_L3;
2093 if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2094 struct flow_match_ipv6_addrs match;
2096 flow_rule_match_ipv6_addrs(rule, &match);
2097 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2098 src_ipv4_src_ipv6.ipv6_layout.ipv6),
2099 &match.mask->src, sizeof(match.mask->src));
2100 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2101 src_ipv4_src_ipv6.ipv6_layout.ipv6),
2102 &match.key->src, sizeof(match.key->src));
2104 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2105 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2106 &match.mask->dst, sizeof(match.mask->dst));
2107 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2108 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2109 &match.key->dst, sizeof(match.key->dst));
2111 if (ipv6_addr_type(&match.mask->src) != IPV6_ADDR_ANY ||
2112 ipv6_addr_type(&match.mask->dst) != IPV6_ADDR_ANY)
2113 *match_level = MLX5_MATCH_L3;
2116 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
2117 struct flow_match_ip match;
2119 flow_rule_match_ip(rule, &match);
2120 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn,
2121 match.mask->tos & 0x3);
2122 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn,
2123 match.key->tos & 0x3);
2125 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp,
2126 match.mask->tos >> 2);
2127 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp,
2128 match.key->tos >> 2);
2130 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit,
2132 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit,
2135 if (match.mask->ttl &&
2136 !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
2137 ft_field_support.outer_ipv4_ttl)) {
2138 NL_SET_ERR_MSG_MOD(extack,
2139 "Matching on TTL is not supported");
2143 if (match.mask->tos || match.mask->ttl)
2144 *match_level = MLX5_MATCH_L3;
2147 /* *** L3 attributes parsing up to here *** */
2149 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) {
2150 struct flow_match_ports match;
2152 flow_rule_match_ports(rule, &match);
2155 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2156 tcp_sport, ntohs(match.mask->src));
2157 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2158 tcp_sport, ntohs(match.key->src));
2160 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2161 tcp_dport, ntohs(match.mask->dst));
2162 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2163 tcp_dport, ntohs(match.key->dst));
2167 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2168 udp_sport, ntohs(match.mask->src));
2169 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2170 udp_sport, ntohs(match.key->src));
2172 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2173 udp_dport, ntohs(match.mask->dst));
2174 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2175 udp_dport, ntohs(match.key->dst));
2178 NL_SET_ERR_MSG_MOD(extack,
2179 "Only UDP and TCP transports are supported for L4 matching");
2180 netdev_err(priv->netdev,
2181 "Only UDP and TCP transport are supported\n");
2185 if (match.mask->src || match.mask->dst)
2186 *match_level = MLX5_MATCH_L4;
2189 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_TCP)) {
2190 struct flow_match_tcp match;
2192 flow_rule_match_tcp(rule, &match);
2193 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
2194 ntohs(match.mask->flags));
2195 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
2196 ntohs(match.key->flags));
2198 if (match.mask->flags)
2199 *match_level = MLX5_MATCH_L4;
2205 static int parse_cls_flower(struct mlx5e_priv *priv,
2206 struct mlx5e_tc_flow *flow,
2207 struct mlx5_flow_spec *spec,
2208 struct flow_cls_offload *f,
2209 struct net_device *filter_dev)
2211 u8 inner_match_level, outer_match_level, non_tunnel_match_level;
2212 struct netlink_ext_ack *extack = f->common.extack;
2213 struct mlx5_core_dev *dev = priv->mdev;
2214 struct mlx5_eswitch *esw = dev->priv.eswitch;
2215 struct mlx5e_rep_priv *rpriv = priv->ppriv;
2216 struct mlx5_eswitch_rep *rep;
2217 bool is_eswitch_flow;
2220 inner_match_level = MLX5_MATCH_NONE;
2221 outer_match_level = MLX5_MATCH_NONE;
2223 err = __parse_cls_flower(priv, spec, f, filter_dev, &inner_match_level,
2224 &outer_match_level);
2225 non_tunnel_match_level = (inner_match_level == MLX5_MATCH_NONE) ?
2226 outer_match_level : inner_match_level;
2228 is_eswitch_flow = mlx5e_is_eswitch_flow(flow);
2229 if (!err && is_eswitch_flow) {
2231 if (rep->vport != MLX5_VPORT_UPLINK &&
2232 (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
2233 esw->offloads.inline_mode < non_tunnel_match_level)) {
2234 NL_SET_ERR_MSG_MOD(extack,
2235 "Flow is not offloaded due to min inline setting");
2236 netdev_warn(priv->netdev,
2237 "Flow is not offloaded due to min inline setting, required %d actual %d\n",
2238 non_tunnel_match_level, esw->offloads.inline_mode);
2243 if (is_eswitch_flow) {
2244 flow->esw_attr->inner_match_level = inner_match_level;
2245 flow->esw_attr->outer_match_level = outer_match_level;
2247 flow->nic_attr->match_level = non_tunnel_match_level;
2253 struct pedit_headers {
2255 struct vlan_hdr vlan;
2262 struct pedit_headers_action {
2263 struct pedit_headers vals;
2264 struct pedit_headers masks;
2268 static int pedit_header_offsets[] = {
2269 [FLOW_ACT_MANGLE_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
2270 [FLOW_ACT_MANGLE_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
2271 [FLOW_ACT_MANGLE_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
2272 [FLOW_ACT_MANGLE_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
2273 [FLOW_ACT_MANGLE_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
2276 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
2278 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
2279 struct pedit_headers_action *hdrs)
2281 u32 *curr_pmask, *curr_pval;
2283 curr_pmask = (u32 *)(pedit_header(&hdrs->masks, hdr_type) + offset);
2284 curr_pval = (u32 *)(pedit_header(&hdrs->vals, hdr_type) + offset);
2286 if (*curr_pmask & mask) /* disallow acting twice on the same location */
2289 *curr_pmask |= mask;
2290 *curr_pval |= (val & mask);
2298 struct mlx5_fields {
2306 #define OFFLOAD(fw_field, field_bsize, field_mask, field, off, match_field) \
2307 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, field_bsize, field_mask, \
2308 offsetof(struct pedit_headers, field) + (off), \
2309 MLX5_BYTE_OFF(fte_match_set_lyr_2_4, match_field)}
2311 /* masked values are the same and there are no rewrites that do not have a
2314 #define SAME_VAL_MASK(type, valp, maskp, matchvalp, matchmaskp) ({ \
2315 type matchmaskx = *(type *)(matchmaskp); \
2316 type matchvalx = *(type *)(matchvalp); \
2317 type maskx = *(type *)(maskp); \
2318 type valx = *(type *)(valp); \
2320 (valx & maskx) == (matchvalx & matchmaskx) && !(maskx & (maskx ^ \
2324 static bool cmp_val_mask(void *valp, void *maskp, void *matchvalp,
2325 void *matchmaskp, u8 bsize)
2331 same = SAME_VAL_MASK(u8, valp, maskp, matchvalp, matchmaskp);
2334 same = SAME_VAL_MASK(u16, valp, maskp, matchvalp, matchmaskp);
2337 same = SAME_VAL_MASK(u32, valp, maskp, matchvalp, matchmaskp);
2344 static struct mlx5_fields fields[] = {
2345 OFFLOAD(DMAC_47_16, 32, U32_MAX, eth.h_dest[0], 0, dmac_47_16),
2346 OFFLOAD(DMAC_15_0, 16, U16_MAX, eth.h_dest[4], 0, dmac_15_0),
2347 OFFLOAD(SMAC_47_16, 32, U32_MAX, eth.h_source[0], 0, smac_47_16),
2348 OFFLOAD(SMAC_15_0, 16, U16_MAX, eth.h_source[4], 0, smac_15_0),
2349 OFFLOAD(ETHERTYPE, 16, U16_MAX, eth.h_proto, 0, ethertype),
2350 OFFLOAD(FIRST_VID, 16, U16_MAX, vlan.h_vlan_TCI, 0, first_vid),
2352 OFFLOAD(IP_DSCP, 8, 0xfc, ip4.tos, 0, ip_dscp),
2353 OFFLOAD(IP_TTL, 8, U8_MAX, ip4.ttl, 0, ttl_hoplimit),
2354 OFFLOAD(SIPV4, 32, U32_MAX, ip4.saddr, 0, src_ipv4_src_ipv6.ipv4_layout.ipv4),
2355 OFFLOAD(DIPV4, 32, U32_MAX, ip4.daddr, 0, dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2357 OFFLOAD(SIPV6_127_96, 32, U32_MAX, ip6.saddr.s6_addr32[0], 0,
2358 src_ipv4_src_ipv6.ipv6_layout.ipv6[0]),
2359 OFFLOAD(SIPV6_95_64, 32, U32_MAX, ip6.saddr.s6_addr32[1], 0,
2360 src_ipv4_src_ipv6.ipv6_layout.ipv6[4]),
2361 OFFLOAD(SIPV6_63_32, 32, U32_MAX, ip6.saddr.s6_addr32[2], 0,
2362 src_ipv4_src_ipv6.ipv6_layout.ipv6[8]),
2363 OFFLOAD(SIPV6_31_0, 32, U32_MAX, ip6.saddr.s6_addr32[3], 0,
2364 src_ipv4_src_ipv6.ipv6_layout.ipv6[12]),
2365 OFFLOAD(DIPV6_127_96, 32, U32_MAX, ip6.daddr.s6_addr32[0], 0,
2366 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[0]),
2367 OFFLOAD(DIPV6_95_64, 32, U32_MAX, ip6.daddr.s6_addr32[1], 0,
2368 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[4]),
2369 OFFLOAD(DIPV6_63_32, 32, U32_MAX, ip6.daddr.s6_addr32[2], 0,
2370 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[8]),
2371 OFFLOAD(DIPV6_31_0, 32, U32_MAX, ip6.daddr.s6_addr32[3], 0,
2372 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[12]),
2373 OFFLOAD(IPV6_HOPLIMIT, 8, U8_MAX, ip6.hop_limit, 0, ttl_hoplimit),
2375 OFFLOAD(TCP_SPORT, 16, U16_MAX, tcp.source, 0, tcp_sport),
2376 OFFLOAD(TCP_DPORT, 16, U16_MAX, tcp.dest, 0, tcp_dport),
2377 /* in linux iphdr tcp_flags is 8 bits long */
2378 OFFLOAD(TCP_FLAGS, 8, U8_MAX, tcp.ack_seq, 5, tcp_flags),
2380 OFFLOAD(UDP_SPORT, 16, U16_MAX, udp.source, 0, udp_sport),
2381 OFFLOAD(UDP_DPORT, 16, U16_MAX, udp.dest, 0, udp_dport),
2384 /* On input attr->max_mod_hdr_actions tells how many HW actions can be parsed at
2385 * max from the SW pedit action. On success, attr->num_mod_hdr_actions
2386 * says how many HW actions were actually parsed.
2388 static int offload_pedit_fields(struct pedit_headers_action *hdrs,
2389 struct mlx5e_tc_flow_parse_attr *parse_attr,
2391 struct netlink_ext_ack *extack)
2393 struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
2394 int i, action_size, nactions, max_actions, first, last, next_z;
2395 void *headers_c, *headers_v, *action, *vals_p;
2396 u32 *s_masks_p, *a_masks_p, s_mask, a_mask;
2397 struct mlx5_fields *f;
2403 headers_c = get_match_headers_criteria(*action_flags, &parse_attr->spec);
2404 headers_v = get_match_headers_value(*action_flags, &parse_attr->spec);
2406 set_masks = &hdrs[0].masks;
2407 add_masks = &hdrs[1].masks;
2408 set_vals = &hdrs[0].vals;
2409 add_vals = &hdrs[1].vals;
2411 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2412 action = parse_attr->mod_hdr_actions +
2413 parse_attr->num_mod_hdr_actions * action_size;
2415 max_actions = parse_attr->max_mod_hdr_actions;
2416 nactions = parse_attr->num_mod_hdr_actions;
2418 for (i = 0; i < ARRAY_SIZE(fields); i++) {
2422 /* avoid seeing bits set from previous iterations */
2426 s_masks_p = (void *)set_masks + f->offset;
2427 a_masks_p = (void *)add_masks + f->offset;
2429 s_mask = *s_masks_p & f->field_mask;
2430 a_mask = *a_masks_p & f->field_mask;
2432 if (!s_mask && !a_mask) /* nothing to offload here */
2435 if (s_mask && a_mask) {
2436 NL_SET_ERR_MSG_MOD(extack,
2437 "can't set and add to the same HW field");
2438 printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
2442 if (nactions == max_actions) {
2443 NL_SET_ERR_MSG_MOD(extack,
2444 "too many pedit actions, can't offload");
2445 printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
2451 void *match_mask = headers_c + f->match_offset;
2452 void *match_val = headers_v + f->match_offset;
2454 cmd = MLX5_ACTION_TYPE_SET;
2456 vals_p = (void *)set_vals + f->offset;
2457 /* don't rewrite if we have a match on the same value */
2458 if (cmp_val_mask(vals_p, s_masks_p, match_val,
2459 match_mask, f->field_bsize))
2461 /* clear to denote we consumed this field */
2462 *s_masks_p &= ~f->field_mask;
2464 cmd = MLX5_ACTION_TYPE_ADD;
2466 vals_p = (void *)add_vals + f->offset;
2467 /* add 0 is no change */
2468 if ((*(u32 *)vals_p & f->field_mask) == 0)
2470 /* clear to denote we consumed this field */
2471 *a_masks_p &= ~f->field_mask;
2476 if (f->field_bsize == 32) {
2477 mask_be32 = *(__be32 *)&mask;
2478 mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
2479 } else if (f->field_bsize == 16) {
2480 mask_be16 = *(__be16 *)&mask;
2481 mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
2484 first = find_first_bit(&mask, f->field_bsize);
2485 next_z = find_next_zero_bit(&mask, f->field_bsize, first);
2486 last = find_last_bit(&mask, f->field_bsize);
2487 if (first < next_z && next_z < last) {
2488 NL_SET_ERR_MSG_MOD(extack,
2489 "rewrite of few sub-fields isn't supported");
2490 printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
2495 MLX5_SET(set_action_in, action, action_type, cmd);
2496 MLX5_SET(set_action_in, action, field, f->field);
2498 if (cmd == MLX5_ACTION_TYPE_SET) {
2501 /* if field is bit sized it can start not from first bit */
2502 start = find_first_bit((unsigned long *)&f->field_mask,
2505 MLX5_SET(set_action_in, action, offset, first - start);
2506 /* length is num of bits to be written, zero means length of 32 */
2507 MLX5_SET(set_action_in, action, length, (last - first + 1));
2510 if (f->field_bsize == 32)
2511 MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
2512 else if (f->field_bsize == 16)
2513 MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
2514 else if (f->field_bsize == 8)
2515 MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
2517 action += action_size;
2521 parse_attr->num_mod_hdr_actions = nactions;
2525 static int mlx5e_flow_namespace_max_modify_action(struct mlx5_core_dev *mdev,
2528 if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
2529 return MLX5_CAP_ESW_FLOWTABLE_FDB(mdev, max_modify_header_actions);
2530 else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
2531 return MLX5_CAP_FLOWTABLE_NIC_RX(mdev, max_modify_header_actions);
2534 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
2535 struct pedit_headers_action *hdrs,
2537 struct mlx5e_tc_flow_parse_attr *parse_attr)
2539 int nkeys, action_size, max_actions;
2541 nkeys = hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits +
2542 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits;
2543 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2545 max_actions = mlx5e_flow_namespace_max_modify_action(priv->mdev, namespace);
2546 /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
2547 max_actions = min(max_actions, nkeys * 16);
2549 parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
2550 if (!parse_attr->mod_hdr_actions)
2553 parse_attr->max_mod_hdr_actions = max_actions;
2557 static const struct pedit_headers zero_masks = {};
2559 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
2560 const struct flow_action_entry *act, int namespace,
2561 struct pedit_headers_action *hdrs,
2562 struct netlink_ext_ack *extack)
2564 u8 cmd = (act->id == FLOW_ACTION_MANGLE) ? 0 : 1;
2565 int err = -EOPNOTSUPP;
2566 u32 mask, val, offset;
2569 htype = act->mangle.htype;
2570 err = -EOPNOTSUPP; /* can't be all optimistic */
2572 if (htype == FLOW_ACT_MANGLE_UNSPEC) {
2573 NL_SET_ERR_MSG_MOD(extack, "legacy pedit isn't offloaded");
2577 if (!mlx5e_flow_namespace_max_modify_action(priv->mdev, namespace)) {
2578 NL_SET_ERR_MSG_MOD(extack,
2579 "The pedit offload action is not supported");
2583 mask = act->mangle.mask;
2584 val = act->mangle.val;
2585 offset = act->mangle.offset;
2587 err = set_pedit_val(htype, ~mask, val, offset, &hdrs[cmd]);
2598 static int alloc_tc_pedit_action(struct mlx5e_priv *priv, int namespace,
2599 struct mlx5e_tc_flow_parse_attr *parse_attr,
2600 struct pedit_headers_action *hdrs,
2602 struct netlink_ext_ack *extack)
2604 struct pedit_headers *cmd_masks;
2608 if (!parse_attr->mod_hdr_actions) {
2609 err = alloc_mod_hdr_actions(priv, hdrs, namespace, parse_attr);
2614 err = offload_pedit_fields(hdrs, parse_attr, action_flags, extack);
2616 goto out_dealloc_parsed_actions;
2618 for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
2619 cmd_masks = &hdrs[cmd].masks;
2620 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
2621 NL_SET_ERR_MSG_MOD(extack,
2622 "attempt to offload an unsupported field");
2623 netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
2624 print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
2625 16, 1, cmd_masks, sizeof(zero_masks), true);
2627 goto out_dealloc_parsed_actions;
2633 out_dealloc_parsed_actions:
2634 kfree(parse_attr->mod_hdr_actions);
2639 static bool csum_offload_supported(struct mlx5e_priv *priv,
2642 struct netlink_ext_ack *extack)
2644 u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
2645 TCA_CSUM_UPDATE_FLAG_UDP;
2647 /* The HW recalcs checksums only if re-writing headers */
2648 if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
2649 NL_SET_ERR_MSG_MOD(extack,
2650 "TC csum action is only offloaded with pedit");
2651 netdev_warn(priv->netdev,
2652 "TC csum action is only offloaded with pedit\n");
2656 if (update_flags & ~prot_flags) {
2657 NL_SET_ERR_MSG_MOD(extack,
2658 "can't offload TC csum action for some header/s");
2659 netdev_warn(priv->netdev,
2660 "can't offload TC csum action for some header/s - flags %#x\n",
2668 struct ip_ttl_word {
2674 struct ipv6_hoplimit_word {
2680 static bool is_action_keys_supported(const struct flow_action_entry *act)
2685 htype = act->mangle.htype;
2686 offset = act->mangle.offset;
2687 mask = ~act->mangle.mask;
2688 /* For IPv4 & IPv6 header check 4 byte word,
2689 * to determine that modified fields
2690 * are NOT ttl & hop_limit only.
2692 if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP4) {
2693 struct ip_ttl_word *ttl_word =
2694 (struct ip_ttl_word *)&mask;
2696 if (offset != offsetof(struct iphdr, ttl) ||
2697 ttl_word->protocol ||
2701 } else if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP6) {
2702 struct ipv6_hoplimit_word *hoplimit_word =
2703 (struct ipv6_hoplimit_word *)&mask;
2705 if (offset != offsetof(struct ipv6hdr, payload_len) ||
2706 hoplimit_word->payload_len ||
2707 hoplimit_word->nexthdr) {
2714 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
2715 struct flow_action *flow_action,
2717 struct netlink_ext_ack *extack)
2719 const struct flow_action_entry *act;
2720 bool modify_ip_header;
2726 headers_v = get_match_headers_value(actions, spec);
2727 ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
2729 /* for non-IP we only re-write MACs, so we're okay */
2730 if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
2733 modify_ip_header = false;
2734 flow_action_for_each(i, act, flow_action) {
2735 if (act->id != FLOW_ACTION_MANGLE &&
2736 act->id != FLOW_ACTION_ADD)
2739 if (is_action_keys_supported(act)) {
2740 modify_ip_header = true;
2745 ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
2746 if (modify_ip_header && ip_proto != IPPROTO_TCP &&
2747 ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
2748 NL_SET_ERR_MSG_MOD(extack,
2749 "can't offload re-write of non TCP/UDP");
2750 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
2758 static bool actions_match_supported(struct mlx5e_priv *priv,
2759 struct flow_action *flow_action,
2760 struct mlx5e_tc_flow_parse_attr *parse_attr,
2761 struct mlx5e_tc_flow *flow,
2762 struct netlink_ext_ack *extack)
2766 if (mlx5e_is_eswitch_flow(flow))
2767 actions = flow->esw_attr->action;
2769 actions = flow->nic_attr->action;
2771 if (flow_flag_test(flow, EGRESS) &&
2772 !((actions & MLX5_FLOW_CONTEXT_ACTION_DECAP) ||
2773 (actions & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) ||
2774 (actions & MLX5_FLOW_CONTEXT_ACTION_DROP)))
2777 if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
2778 return modify_header_match_supported(&parse_attr->spec,
2779 flow_action, actions,
2785 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
2787 struct mlx5_core_dev *fmdev, *pmdev;
2788 u64 fsystem_guid, psystem_guid;
2791 pmdev = peer_priv->mdev;
2793 fsystem_guid = mlx5_query_nic_system_image_guid(fmdev);
2794 psystem_guid = mlx5_query_nic_system_image_guid(pmdev);
2796 return (fsystem_guid == psystem_guid);
2799 static int add_vlan_rewrite_action(struct mlx5e_priv *priv, int namespace,
2800 const struct flow_action_entry *act,
2801 struct mlx5e_tc_flow_parse_attr *parse_attr,
2802 struct pedit_headers_action *hdrs,
2803 u32 *action, struct netlink_ext_ack *extack)
2805 u16 mask16 = VLAN_VID_MASK;
2806 u16 val16 = act->vlan.vid & VLAN_VID_MASK;
2807 const struct flow_action_entry pedit_act = {
2808 .id = FLOW_ACTION_MANGLE,
2809 .mangle.htype = FLOW_ACT_MANGLE_HDR_TYPE_ETH,
2810 .mangle.offset = offsetof(struct vlan_ethhdr, h_vlan_TCI),
2811 .mangle.mask = ~(u32)be16_to_cpu(*(__be16 *)&mask16),
2812 .mangle.val = (u32)be16_to_cpu(*(__be16 *)&val16),
2814 u8 match_prio_mask, match_prio_val;
2815 void *headers_c, *headers_v;
2818 headers_c = get_match_headers_criteria(*action, &parse_attr->spec);
2819 headers_v = get_match_headers_value(*action, &parse_attr->spec);
2821 if (!(MLX5_GET(fte_match_set_lyr_2_4, headers_c, cvlan_tag) &&
2822 MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag))) {
2823 NL_SET_ERR_MSG_MOD(extack,
2824 "VLAN rewrite action must have VLAN protocol match");
2828 match_prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
2829 match_prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
2830 if (act->vlan.prio != (match_prio_val & match_prio_mask)) {
2831 NL_SET_ERR_MSG_MOD(extack,
2832 "Changing VLAN prio is not supported");
2836 err = parse_tc_pedit_action(priv, &pedit_act, namespace, hdrs, NULL);
2837 *action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2843 add_vlan_prio_tag_rewrite_action(struct mlx5e_priv *priv,
2844 struct mlx5e_tc_flow_parse_attr *parse_attr,
2845 struct pedit_headers_action *hdrs,
2846 u32 *action, struct netlink_ext_ack *extack)
2848 const struct flow_action_entry prio_tag_act = {
2851 MLX5_GET(fte_match_set_lyr_2_4,
2852 get_match_headers_value(*action,
2855 MLX5_GET(fte_match_set_lyr_2_4,
2856 get_match_headers_criteria(*action,
2861 return add_vlan_rewrite_action(priv, MLX5_FLOW_NAMESPACE_FDB,
2862 &prio_tag_act, parse_attr, hdrs, action,
2866 static int parse_tc_nic_actions(struct mlx5e_priv *priv,
2867 struct flow_action *flow_action,
2868 struct mlx5e_tc_flow_parse_attr *parse_attr,
2869 struct mlx5e_tc_flow *flow,
2870 struct netlink_ext_ack *extack)
2872 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
2873 struct pedit_headers_action hdrs[2] = {};
2874 const struct flow_action_entry *act;
2878 if (!flow_action_has_entries(flow_action))
2881 if (!flow_action_hw_stats_types_check(flow_action, extack,
2882 FLOW_ACTION_HW_STATS_TYPE_DELAYED))
2885 attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
2887 flow_action_for_each(i, act, flow_action) {
2889 case FLOW_ACTION_ACCEPT:
2890 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2891 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2893 case FLOW_ACTION_DROP:
2894 action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
2895 if (MLX5_CAP_FLOWTABLE(priv->mdev,
2896 flow_table_properties_nic_receive.flow_counter))
2897 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
2899 case FLOW_ACTION_MANGLE:
2900 case FLOW_ACTION_ADD:
2901 err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_KERNEL,
2906 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
2907 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2909 case FLOW_ACTION_VLAN_MANGLE:
2910 err = add_vlan_rewrite_action(priv,
2911 MLX5_FLOW_NAMESPACE_KERNEL,
2912 act, parse_attr, hdrs,
2918 case FLOW_ACTION_CSUM:
2919 if (csum_offload_supported(priv, action,
2925 case FLOW_ACTION_REDIRECT: {
2926 struct net_device *peer_dev = act->dev;
2928 if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
2929 same_hw_devs(priv, netdev_priv(peer_dev))) {
2930 parse_attr->mirred_ifindex[0] = peer_dev->ifindex;
2931 flow_flag_set(flow, HAIRPIN);
2932 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2933 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2935 NL_SET_ERR_MSG_MOD(extack,
2936 "device is not on same HW, can't offload");
2937 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
2943 case FLOW_ACTION_MARK: {
2944 u32 mark = act->mark;
2946 if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
2947 NL_SET_ERR_MSG_MOD(extack,
2948 "Bad flow mark - only 16 bit is supported");
2952 attr->flow_tag = mark;
2953 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2957 NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
2962 if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
2963 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
2964 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_KERNEL,
2965 parse_attr, hdrs, &action, extack);
2968 /* in case all pedit actions are skipped, remove the MOD_HDR
2971 if (parse_attr->num_mod_hdr_actions == 0) {
2972 action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2973 kfree(parse_attr->mod_hdr_actions);
2977 attr->action = action;
2978 if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
2985 const struct ip_tunnel_key *ip_tun_key;
2986 struct mlx5e_tc_tunnel *tc_tunnel;
2989 static inline int cmp_encap_info(struct encap_key *a,
2990 struct encap_key *b)
2992 return memcmp(a->ip_tun_key, b->ip_tun_key, sizeof(*a->ip_tun_key)) ||
2993 a->tc_tunnel->tunnel_type != b->tc_tunnel->tunnel_type;
2996 static inline int hash_encap_info(struct encap_key *key)
2998 return jhash(key->ip_tun_key, sizeof(*key->ip_tun_key),
2999 key->tc_tunnel->tunnel_type);
3003 static bool is_merged_eswitch_dev(struct mlx5e_priv *priv,
3004 struct net_device *peer_netdev)
3006 struct mlx5e_priv *peer_priv;
3008 peer_priv = netdev_priv(peer_netdev);
3010 return (MLX5_CAP_ESW(priv->mdev, merged_eswitch) &&
3011 mlx5e_eswitch_rep(priv->netdev) &&
3012 mlx5e_eswitch_rep(peer_netdev) &&
3013 same_hw_devs(priv, peer_priv));
3018 bool mlx5e_encap_take(struct mlx5e_encap_entry *e)
3020 return refcount_inc_not_zero(&e->refcnt);
3023 static struct mlx5e_encap_entry *
3024 mlx5e_encap_get(struct mlx5e_priv *priv, struct encap_key *key,
3027 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3028 struct mlx5e_encap_entry *e;
3029 struct encap_key e_key;
3031 hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
3032 encap_hlist, hash_key) {
3033 e_key.ip_tun_key = &e->tun_info->key;
3034 e_key.tc_tunnel = e->tunnel;
3035 if (!cmp_encap_info(&e_key, key) &&
3036 mlx5e_encap_take(e))
3043 static struct ip_tunnel_info *dup_tun_info(const struct ip_tunnel_info *tun_info)
3045 size_t tun_size = sizeof(*tun_info) + tun_info->options_len;
3047 return kmemdup(tun_info, tun_size, GFP_KERNEL);
3050 static bool is_duplicated_encap_entry(struct mlx5e_priv *priv,
3051 struct mlx5e_tc_flow *flow,
3053 struct mlx5e_encap_entry *e,
3054 struct netlink_ext_ack *extack)
3058 for (i = 0; i < out_index; i++) {
3059 if (flow->encaps[i].e != e)
3061 NL_SET_ERR_MSG_MOD(extack, "can't duplicate encap action");
3062 netdev_err(priv->netdev, "can't duplicate encap action\n");
3069 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
3070 struct mlx5e_tc_flow *flow,
3071 struct net_device *mirred_dev,
3073 struct netlink_ext_ack *extack,
3074 struct net_device **encap_dev,
3077 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3078 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3079 struct mlx5e_tc_flow_parse_attr *parse_attr;
3080 const struct ip_tunnel_info *tun_info;
3081 struct encap_key key;
3082 struct mlx5e_encap_entry *e;
3083 unsigned short family;
3087 parse_attr = attr->parse_attr;
3088 tun_info = parse_attr->tun_info[out_index];
3089 family = ip_tunnel_info_af(tun_info);
3090 key.ip_tun_key = &tun_info->key;
3091 key.tc_tunnel = mlx5e_get_tc_tun(mirred_dev);
3092 if (!key.tc_tunnel) {
3093 NL_SET_ERR_MSG_MOD(extack, "Unsupported tunnel");
3097 hash_key = hash_encap_info(&key);
3099 mutex_lock(&esw->offloads.encap_tbl_lock);
3100 e = mlx5e_encap_get(priv, &key, hash_key);
3102 /* must verify if encap is valid or not */
3104 /* Check that entry was not already attached to this flow */
3105 if (is_duplicated_encap_entry(priv, flow, out_index, e, extack)) {
3110 mutex_unlock(&esw->offloads.encap_tbl_lock);
3111 wait_for_completion(&e->res_ready);
3113 /* Protect against concurrent neigh update. */
3114 mutex_lock(&esw->offloads.encap_tbl_lock);
3115 if (e->compl_result < 0) {
3122 e = kzalloc(sizeof(*e), GFP_KERNEL);
3128 refcount_set(&e->refcnt, 1);
3129 init_completion(&e->res_ready);
3131 tun_info = dup_tun_info(tun_info);
3136 e->tun_info = tun_info;
3137 err = mlx5e_tc_tun_init_encap_attr(mirred_dev, priv, e, extack);
3141 INIT_LIST_HEAD(&e->flows);
3142 hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
3143 mutex_unlock(&esw->offloads.encap_tbl_lock);
3145 if (family == AF_INET)
3146 err = mlx5e_tc_tun_create_header_ipv4(priv, mirred_dev, e);
3147 else if (family == AF_INET6)
3148 err = mlx5e_tc_tun_create_header_ipv6(priv, mirred_dev, e);
3150 /* Protect against concurrent neigh update. */
3151 mutex_lock(&esw->offloads.encap_tbl_lock);
3152 complete_all(&e->res_ready);
3154 e->compl_result = err;
3157 e->compl_result = 1;
3160 flow->encaps[out_index].e = e;
3161 list_add(&flow->encaps[out_index].list, &e->flows);
3162 flow->encaps[out_index].index = out_index;
3163 *encap_dev = e->out_dev;
3164 if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
3165 attr->dests[out_index].pkt_reformat = e->pkt_reformat;
3166 attr->dests[out_index].flags |= MLX5_ESW_DEST_ENCAP_VALID;
3167 *encap_valid = true;
3169 *encap_valid = false;
3171 mutex_unlock(&esw->offloads.encap_tbl_lock);
3176 mutex_unlock(&esw->offloads.encap_tbl_lock);
3178 mlx5e_encap_put(priv, e);
3182 mutex_unlock(&esw->offloads.encap_tbl_lock);
3188 static int parse_tc_vlan_action(struct mlx5e_priv *priv,
3189 const struct flow_action_entry *act,
3190 struct mlx5_esw_flow_attr *attr,
3193 u8 vlan_idx = attr->total_vlan;
3195 if (vlan_idx >= MLX5_FS_VLAN_DEPTH)
3199 case FLOW_ACTION_VLAN_POP:
3201 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
3202 MLX5_FS_VLAN_DEPTH))
3205 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2;
3207 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3210 case FLOW_ACTION_VLAN_PUSH:
3211 attr->vlan_vid[vlan_idx] = act->vlan.vid;
3212 attr->vlan_prio[vlan_idx] = act->vlan.prio;
3213 attr->vlan_proto[vlan_idx] = act->vlan.proto;
3214 if (!attr->vlan_proto[vlan_idx])
3215 attr->vlan_proto[vlan_idx] = htons(ETH_P_8021Q);
3218 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
3219 MLX5_FS_VLAN_DEPTH))
3222 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2;
3224 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev, 1) &&
3225 (act->vlan.proto != htons(ETH_P_8021Q) ||
3229 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
3236 attr->total_vlan = vlan_idx + 1;
3241 static int add_vlan_push_action(struct mlx5e_priv *priv,
3242 struct mlx5_esw_flow_attr *attr,
3243 struct net_device **out_dev,
3246 struct net_device *vlan_dev = *out_dev;
3247 struct flow_action_entry vlan_act = {
3248 .id = FLOW_ACTION_VLAN_PUSH,
3249 .vlan.vid = vlan_dev_vlan_id(vlan_dev),
3250 .vlan.proto = vlan_dev_vlan_proto(vlan_dev),
3255 err = parse_tc_vlan_action(priv, &vlan_act, attr, action);
3259 *out_dev = dev_get_by_index_rcu(dev_net(vlan_dev),
3260 dev_get_iflink(vlan_dev));
3261 if (is_vlan_dev(*out_dev))
3262 err = add_vlan_push_action(priv, attr, out_dev, action);
3267 static int add_vlan_pop_action(struct mlx5e_priv *priv,
3268 struct mlx5_esw_flow_attr *attr,
3271 int nest_level = attr->parse_attr->filter_dev->lower_level;
3272 struct flow_action_entry vlan_act = {
3273 .id = FLOW_ACTION_VLAN_POP,
3277 while (nest_level--) {
3278 err = parse_tc_vlan_action(priv, &vlan_act, attr, action);
3286 bool mlx5e_is_valid_eswitch_fwd_dev(struct mlx5e_priv *priv,
3287 struct net_device *out_dev)
3289 if (is_merged_eswitch_dev(priv, out_dev))
3292 return mlx5e_eswitch_rep(out_dev) &&
3293 same_hw_devs(priv, netdev_priv(out_dev));
3296 static bool is_duplicated_output_device(struct net_device *dev,
3297 struct net_device *out_dev,
3298 int *ifindexes, int if_count,
3299 struct netlink_ext_ack *extack)
3303 for (i = 0; i < if_count; i++) {
3304 if (ifindexes[i] == out_dev->ifindex) {
3305 NL_SET_ERR_MSG_MOD(extack,
3306 "can't duplicate output to same device");
3307 netdev_err(dev, "can't duplicate output to same device: %s\n",
3316 static int mlx5_validate_goto_chain(struct mlx5_eswitch *esw,
3317 struct mlx5e_tc_flow *flow,
3318 const struct flow_action_entry *act,
3320 struct netlink_ext_ack *extack)
3322 u32 max_chain = mlx5_esw_chains_get_chain_range(esw);
3323 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3324 bool ft_flow = mlx5e_is_ft_flow(flow);
3325 u32 dest_chain = act->chain_index;
3328 NL_SET_ERR_MSG_MOD(extack, "Goto action is not supported");
3332 if (!mlx5_esw_chains_backwards_supported(esw) &&
3333 dest_chain <= attr->chain) {
3334 NL_SET_ERR_MSG_MOD(extack,
3335 "Goto lower numbered chain isn't supported");
3338 if (dest_chain > max_chain) {
3339 NL_SET_ERR_MSG_MOD(extack,
3340 "Requested destination chain is out of supported range");
3344 if (actions & (MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
3345 MLX5_FLOW_CONTEXT_ACTION_DECAP) &&
3346 !MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, reformat_and_fwd_to_table)) {
3347 NL_SET_ERR_MSG_MOD(extack,
3348 "Goto chain is not allowed if action has reformat or decap");
3355 static int parse_tc_fdb_actions(struct mlx5e_priv *priv,
3356 struct flow_action *flow_action,
3357 struct mlx5e_tc_flow *flow,
3358 struct netlink_ext_ack *extack)
3360 struct pedit_headers_action hdrs[2] = {};
3361 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3362 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3363 struct mlx5e_tc_flow_parse_attr *parse_attr = attr->parse_attr;
3364 struct mlx5e_rep_priv *rpriv = priv->ppriv;
3365 const struct ip_tunnel_info *info = NULL;
3366 int ifindexes[MLX5_MAX_FLOW_FWD_VPORTS];
3367 bool ft_flow = mlx5e_is_ft_flow(flow);
3368 const struct flow_action_entry *act;
3369 int err, i, if_count = 0;
3373 if (!flow_action_has_entries(flow_action))
3376 if (!flow_action_hw_stats_types_check(flow_action, extack,
3377 FLOW_ACTION_HW_STATS_TYPE_DELAYED))
3380 flow_action_for_each(i, act, flow_action) {
3382 case FLOW_ACTION_DROP:
3383 action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
3384 MLX5_FLOW_CONTEXT_ACTION_COUNT;
3386 case FLOW_ACTION_MANGLE:
3387 case FLOW_ACTION_ADD:
3388 err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_FDB,
3393 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3394 attr->split_count = attr->out_count;
3396 case FLOW_ACTION_CSUM:
3397 if (csum_offload_supported(priv, action,
3398 act->csum_flags, extack))
3402 case FLOW_ACTION_REDIRECT:
3403 case FLOW_ACTION_MIRRED: {
3404 struct mlx5e_priv *out_priv;
3405 struct net_device *out_dev;
3409 /* out_dev is NULL when filters with
3410 * non-existing mirred device are replayed to
3416 if (ft_flow && out_dev == priv->netdev) {
3417 /* Ignore forward to self rules generated
3418 * by adding both mlx5 devs to the flow table
3419 * block on a normal nft offload setup.
3424 if (attr->out_count >= MLX5_MAX_FLOW_FWD_VPORTS) {
3425 NL_SET_ERR_MSG_MOD(extack,
3426 "can't support more output ports, can't offload forwarding");
3427 netdev_warn(priv->netdev,
3428 "can't support more than %d output ports, can't offload forwarding\n",
3433 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
3434 MLX5_FLOW_CONTEXT_ACTION_COUNT;
3436 parse_attr->mirred_ifindex[attr->out_count] =
3438 parse_attr->tun_info[attr->out_count] = dup_tun_info(info);
3439 if (!parse_attr->tun_info[attr->out_count])
3442 attr->dests[attr->out_count].flags |=
3443 MLX5_ESW_DEST_ENCAP;
3445 /* attr->dests[].rep is resolved when we
3448 } else if (netdev_port_same_parent_id(priv->netdev, out_dev)) {
3449 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3450 struct net_device *uplink_dev = mlx5_eswitch_uplink_get_proto_dev(esw, REP_ETH);
3451 struct net_device *uplink_upper;
3452 struct mlx5e_rep_priv *rep_priv;
3454 if (is_duplicated_output_device(priv->netdev,
3461 ifindexes[if_count] = out_dev->ifindex;
3466 netdev_master_upper_dev_get_rcu(uplink_dev);
3468 netif_is_lag_master(uplink_upper) &&
3469 uplink_upper == out_dev)
3470 out_dev = uplink_dev;
3473 if (is_vlan_dev(out_dev)) {
3474 err = add_vlan_push_action(priv, attr,
3481 if (is_vlan_dev(parse_attr->filter_dev)) {
3482 err = add_vlan_pop_action(priv, attr,
3488 /* Don't allow forwarding between uplink.
3490 * Input vport was stored esw_attr->in_rep.
3491 * In LAG case, *priv* is the private data of
3492 * uplink which may be not the input vport.
3494 rep_priv = mlx5e_rep_to_rep_priv(attr->in_rep);
3495 if (mlx5e_eswitch_uplink_rep(rep_priv->netdev) &&
3496 mlx5e_eswitch_uplink_rep(out_dev)) {
3497 NL_SET_ERR_MSG_MOD(extack,
3498 "devices are both uplink, can't offload forwarding");
3499 pr_err("devices %s %s are both uplink, can't offload forwarding\n",
3500 priv->netdev->name, out_dev->name);
3504 if (!mlx5e_is_valid_eswitch_fwd_dev(priv, out_dev)) {
3505 NL_SET_ERR_MSG_MOD(extack,
3506 "devices are not on same switch HW, can't offload forwarding");
3507 netdev_warn(priv->netdev,
3508 "devices %s %s not on same switch HW, can't offload forwarding\n",
3514 out_priv = netdev_priv(out_dev);
3515 rpriv = out_priv->ppriv;
3516 attr->dests[attr->out_count].rep = rpriv->rep;
3517 attr->dests[attr->out_count].mdev = out_priv->mdev;
3519 } else if (parse_attr->filter_dev != priv->netdev) {
3520 /* All mlx5 devices are called to configure
3521 * high level device filters. Therefore, the
3522 * *attempt* to install a filter on invalid
3523 * eswitch should not trigger an explicit error
3527 NL_SET_ERR_MSG_MOD(extack,
3528 "devices are not on same switch HW, can't offload forwarding");
3529 netdev_warn(priv->netdev,
3530 "devices %s %s not on same switch HW, can't offload forwarding\n",
3537 case FLOW_ACTION_TUNNEL_ENCAP:
3545 case FLOW_ACTION_VLAN_PUSH:
3546 case FLOW_ACTION_VLAN_POP:
3547 if (act->id == FLOW_ACTION_VLAN_PUSH &&
3548 (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP)) {
3549 /* Replace vlan pop+push with vlan modify */
3550 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3551 err = add_vlan_rewrite_action(priv,
3552 MLX5_FLOW_NAMESPACE_FDB,
3553 act, parse_attr, hdrs,
3556 err = parse_tc_vlan_action(priv, act, attr, &action);
3561 attr->split_count = attr->out_count;
3563 case FLOW_ACTION_VLAN_MANGLE:
3564 err = add_vlan_rewrite_action(priv,
3565 MLX5_FLOW_NAMESPACE_FDB,
3566 act, parse_attr, hdrs,
3571 attr->split_count = attr->out_count;
3573 case FLOW_ACTION_TUNNEL_DECAP:
3574 action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
3576 case FLOW_ACTION_GOTO:
3577 err = mlx5_validate_goto_chain(esw, flow, act, action,
3582 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
3583 attr->dest_chain = act->chain_index;
3586 NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
3591 if (MLX5_CAP_GEN(esw->dev, prio_tag_required) &&
3592 action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) {
3593 /* For prio tag mode, replace vlan pop with rewrite vlan prio
3596 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3597 err = add_vlan_prio_tag_rewrite_action(priv, parse_attr, hdrs,
3603 if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
3604 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
3605 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_FDB,
3606 parse_attr, hdrs, &action, extack);
3609 /* in case all pedit actions are skipped, remove the MOD_HDR
3610 * flag. we might have set split_count either by pedit or
3611 * pop/push. if there is no pop/push either, reset it too.
3613 if (parse_attr->num_mod_hdr_actions == 0) {
3614 action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3615 kfree(parse_attr->mod_hdr_actions);
3616 if (!((action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) ||
3617 (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH)))
3618 attr->split_count = 0;
3622 attr->action = action;
3623 if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
3626 if (attr->dest_chain) {
3627 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
3628 NL_SET_ERR_MSG_MOD(extack,
3629 "Mirroring goto chain rules isn't supported");
3632 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3635 if (!(attr->action &
3636 (MLX5_FLOW_CONTEXT_ACTION_FWD_DEST | MLX5_FLOW_CONTEXT_ACTION_DROP))) {
3637 NL_SET_ERR_MSG_MOD(extack,
3638 "Rule must have at least one forward/drop action");
3642 if (attr->split_count > 0 && !mlx5_esw_has_fwd_fdb(priv->mdev)) {
3643 NL_SET_ERR_MSG_MOD(extack,
3644 "current firmware doesn't support split rule for port mirroring");
3645 netdev_warn_once(priv->netdev, "current firmware doesn't support split rule for port mirroring\n");
3652 static void get_flags(int flags, unsigned long *flow_flags)
3654 unsigned long __flow_flags = 0;
3656 if (flags & MLX5_TC_FLAG(INGRESS))
3657 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_INGRESS);
3658 if (flags & MLX5_TC_FLAG(EGRESS))
3659 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_EGRESS);
3661 if (flags & MLX5_TC_FLAG(ESW_OFFLOAD))
3662 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_ESWITCH);
3663 if (flags & MLX5_TC_FLAG(NIC_OFFLOAD))
3664 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_NIC);
3665 if (flags & MLX5_TC_FLAG(FT_OFFLOAD))
3666 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_FT);
3668 *flow_flags = __flow_flags;
3671 static const struct rhashtable_params tc_ht_params = {
3672 .head_offset = offsetof(struct mlx5e_tc_flow, node),
3673 .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
3674 .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
3675 .automatic_shrinking = true,
3678 static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv,
3679 unsigned long flags)
3681 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3682 struct mlx5e_rep_priv *uplink_rpriv;
3684 if (flags & MLX5_TC_FLAG(ESW_OFFLOAD)) {
3685 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
3686 return &uplink_rpriv->uplink_priv.tc_ht;
3687 } else /* NIC offload */
3688 return &priv->fs.tc.ht;
3691 static bool is_peer_flow_needed(struct mlx5e_tc_flow *flow)
3693 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3694 bool is_rep_ingress = attr->in_rep->vport != MLX5_VPORT_UPLINK &&
3695 flow_flag_test(flow, INGRESS);
3696 bool act_is_encap = !!(attr->action &
3697 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT);
3698 bool esw_paired = mlx5_devcom_is_paired(attr->in_mdev->priv.devcom,
3699 MLX5_DEVCOM_ESW_OFFLOADS);
3704 if ((mlx5_lag_is_sriov(attr->in_mdev) ||
3705 mlx5_lag_is_multipath(attr->in_mdev)) &&
3706 (is_rep_ingress || act_is_encap))
3713 mlx5e_alloc_flow(struct mlx5e_priv *priv, int attr_size,
3714 struct flow_cls_offload *f, unsigned long flow_flags,
3715 struct mlx5e_tc_flow_parse_attr **__parse_attr,
3716 struct mlx5e_tc_flow **__flow)
3718 struct mlx5e_tc_flow_parse_attr *parse_attr;
3719 struct mlx5e_tc_flow *flow;
3722 flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
3723 parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
3724 if (!parse_attr || !flow) {
3729 flow->cookie = f->cookie;
3730 flow->flags = flow_flags;
3732 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++)
3733 INIT_LIST_HEAD(&flow->encaps[out_index].list);
3734 INIT_LIST_HEAD(&flow->mod_hdr);
3735 INIT_LIST_HEAD(&flow->hairpin);
3736 refcount_set(&flow->refcnt, 1);
3737 init_completion(&flow->init_done);
3740 *__parse_attr = parse_attr;
3751 mlx5e_flow_esw_attr_init(struct mlx5_esw_flow_attr *esw_attr,
3752 struct mlx5e_priv *priv,
3753 struct mlx5e_tc_flow_parse_attr *parse_attr,
3754 struct flow_cls_offload *f,
3755 struct mlx5_eswitch_rep *in_rep,
3756 struct mlx5_core_dev *in_mdev)
3758 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3760 esw_attr->parse_attr = parse_attr;
3761 esw_attr->chain = f->common.chain_index;
3762 esw_attr->prio = f->common.prio;
3764 esw_attr->in_rep = in_rep;
3765 esw_attr->in_mdev = in_mdev;
3767 if (MLX5_CAP_ESW(esw->dev, counter_eswitch_affinity) ==
3768 MLX5_COUNTER_SOURCE_ESWITCH)
3769 esw_attr->counter_dev = in_mdev;
3771 esw_attr->counter_dev = priv->mdev;
3774 static struct mlx5e_tc_flow *
3775 __mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
3776 struct flow_cls_offload *f,
3777 unsigned long flow_flags,
3778 struct net_device *filter_dev,
3779 struct mlx5_eswitch_rep *in_rep,
3780 struct mlx5_core_dev *in_mdev)
3782 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
3783 struct netlink_ext_ack *extack = f->common.extack;
3784 struct mlx5e_tc_flow_parse_attr *parse_attr;
3785 struct mlx5e_tc_flow *flow;
3788 flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_ESWITCH);
3789 attr_size = sizeof(struct mlx5_esw_flow_attr);
3790 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3791 &parse_attr, &flow);
3795 parse_attr->filter_dev = filter_dev;
3796 mlx5e_flow_esw_attr_init(flow->esw_attr,
3798 f, in_rep, in_mdev);
3800 err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
3805 err = parse_tc_fdb_actions(priv, &rule->action, flow, extack);
3809 err = mlx5e_tc_add_fdb_flow(priv, flow, extack);
3810 complete_all(&flow->init_done);
3812 if (!(err == -ENETUNREACH && mlx5_lag_is_multipath(in_mdev)))
3815 add_unready_flow(flow);
3821 mlx5e_flow_put(priv, flow);
3823 return ERR_PTR(err);
3826 static int mlx5e_tc_add_fdb_peer_flow(struct flow_cls_offload *f,
3827 struct mlx5e_tc_flow *flow,
3828 unsigned long flow_flags)
3830 struct mlx5e_priv *priv = flow->priv, *peer_priv;
3831 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch, *peer_esw;
3832 struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
3833 struct mlx5e_tc_flow_parse_attr *parse_attr;
3834 struct mlx5e_rep_priv *peer_urpriv;
3835 struct mlx5e_tc_flow *peer_flow;
3836 struct mlx5_core_dev *in_mdev;
3839 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
3843 peer_urpriv = mlx5_eswitch_get_uplink_priv(peer_esw, REP_ETH);
3844 peer_priv = netdev_priv(peer_urpriv->netdev);
3846 /* in_mdev is assigned of which the packet originated from.
3847 * So packets redirected to uplink use the same mdev of the
3848 * original flow and packets redirected from uplink use the
3851 if (flow->esw_attr->in_rep->vport == MLX5_VPORT_UPLINK)
3852 in_mdev = peer_priv->mdev;
3854 in_mdev = priv->mdev;
3856 parse_attr = flow->esw_attr->parse_attr;
3857 peer_flow = __mlx5e_add_fdb_flow(peer_priv, f, flow_flags,
3858 parse_attr->filter_dev,
3859 flow->esw_attr->in_rep, in_mdev);
3860 if (IS_ERR(peer_flow)) {
3861 err = PTR_ERR(peer_flow);
3865 flow->peer_flow = peer_flow;
3866 flow_flag_set(flow, DUP);
3867 mutex_lock(&esw->offloads.peer_mutex);
3868 list_add_tail(&flow->peer, &esw->offloads.peer_flows);
3869 mutex_unlock(&esw->offloads.peer_mutex);
3872 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
3877 mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
3878 struct flow_cls_offload *f,
3879 unsigned long flow_flags,
3880 struct net_device *filter_dev,
3881 struct mlx5e_tc_flow **__flow)
3883 struct mlx5e_rep_priv *rpriv = priv->ppriv;
3884 struct mlx5_eswitch_rep *in_rep = rpriv->rep;
3885 struct mlx5_core_dev *in_mdev = priv->mdev;
3886 struct mlx5e_tc_flow *flow;
3889 flow = __mlx5e_add_fdb_flow(priv, f, flow_flags, filter_dev, in_rep,
3892 return PTR_ERR(flow);
3894 if (is_peer_flow_needed(flow)) {
3895 err = mlx5e_tc_add_fdb_peer_flow(f, flow, flow_flags);
3897 mlx5e_tc_del_fdb_flow(priv, flow);
3911 mlx5e_add_nic_flow(struct mlx5e_priv *priv,
3912 struct flow_cls_offload *f,
3913 unsigned long flow_flags,
3914 struct net_device *filter_dev,
3915 struct mlx5e_tc_flow **__flow)
3917 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
3918 struct netlink_ext_ack *extack = f->common.extack;
3919 struct mlx5e_tc_flow_parse_attr *parse_attr;
3920 struct mlx5e_tc_flow *flow;
3923 /* multi-chain not supported for NIC rules */
3924 if (!tc_cls_can_offload_and_chain0(priv->netdev, &f->common))
3927 flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_NIC);
3928 attr_size = sizeof(struct mlx5_nic_flow_attr);
3929 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3930 &parse_attr, &flow);
3934 parse_attr->filter_dev = filter_dev;
3935 err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
3940 err = parse_tc_nic_actions(priv, &rule->action, parse_attr, flow, extack);
3944 err = mlx5e_tc_add_nic_flow(priv, parse_attr, flow, extack);
3948 flow_flag_set(flow, OFFLOADED);
3955 mlx5e_flow_put(priv, flow);
3962 mlx5e_tc_add_flow(struct mlx5e_priv *priv,
3963 struct flow_cls_offload *f,
3964 unsigned long flags,
3965 struct net_device *filter_dev,
3966 struct mlx5e_tc_flow **flow)
3968 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3969 unsigned long flow_flags;
3972 get_flags(flags, &flow_flags);
3974 if (!tc_can_offload_extack(priv->netdev, f->common.extack))
3977 if (esw && esw->mode == MLX5_ESWITCH_OFFLOADS)
3978 err = mlx5e_add_fdb_flow(priv, f, flow_flags,
3981 err = mlx5e_add_nic_flow(priv, f, flow_flags,
3987 int mlx5e_configure_flower(struct net_device *dev, struct mlx5e_priv *priv,
3988 struct flow_cls_offload *f, unsigned long flags)
3990 struct netlink_ext_ack *extack = f->common.extack;
3991 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
3992 struct mlx5e_tc_flow *flow;
3996 flow = rhashtable_lookup(tc_ht, &f->cookie, tc_ht_params);
3999 NL_SET_ERR_MSG_MOD(extack,
4000 "flow cookie already exists, ignoring");
4001 netdev_warn_once(priv->netdev,
4002 "flow cookie %lx already exists, ignoring\n",
4008 trace_mlx5e_configure_flower(f);
4009 err = mlx5e_tc_add_flow(priv, f, flags, dev, &flow);
4013 err = rhashtable_lookup_insert_fast(tc_ht, &flow->node, tc_ht_params);
4020 mlx5e_flow_put(priv, flow);
4025 static bool same_flow_direction(struct mlx5e_tc_flow *flow, int flags)
4027 bool dir_ingress = !!(flags & MLX5_TC_FLAG(INGRESS));
4028 bool dir_egress = !!(flags & MLX5_TC_FLAG(EGRESS));
4030 return flow_flag_test(flow, INGRESS) == dir_ingress &&
4031 flow_flag_test(flow, EGRESS) == dir_egress;
4034 int mlx5e_delete_flower(struct net_device *dev, struct mlx5e_priv *priv,
4035 struct flow_cls_offload *f, unsigned long flags)
4037 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4038 struct mlx5e_tc_flow *flow;
4042 flow = rhashtable_lookup(tc_ht, &f->cookie, tc_ht_params);
4043 if (!flow || !same_flow_direction(flow, flags)) {
4048 /* Only delete the flow if it doesn't have MLX5E_TC_FLOW_DELETED flag
4051 if (flow_flag_test_and_set(flow, DELETED)) {
4055 rhashtable_remove_fast(tc_ht, &flow->node, tc_ht_params);
4058 trace_mlx5e_delete_flower(f);
4059 mlx5e_flow_put(priv, flow);
4068 int mlx5e_stats_flower(struct net_device *dev, struct mlx5e_priv *priv,
4069 struct flow_cls_offload *f, unsigned long flags)
4071 struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
4072 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4073 struct mlx5_eswitch *peer_esw;
4074 struct mlx5e_tc_flow *flow;
4075 struct mlx5_fc *counter;
4082 flow = mlx5e_flow_get(rhashtable_lookup(tc_ht, &f->cookie,
4086 return PTR_ERR(flow);
4088 if (!same_flow_direction(flow, flags)) {
4093 if (mlx5e_is_offloaded_flow(flow)) {
4094 counter = mlx5e_tc_get_counter(flow);
4098 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
4101 /* Under multipath it's possible for one rule to be currently
4102 * un-offloaded while the other rule is offloaded.
4104 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4108 if (flow_flag_test(flow, DUP) &&
4109 flow_flag_test(flow->peer_flow, OFFLOADED)) {
4114 counter = mlx5e_tc_get_counter(flow->peer_flow);
4116 goto no_peer_counter;
4117 mlx5_fc_query_cached(counter, &bytes2, &packets2, &lastuse2);
4120 packets += packets2;
4121 lastuse = max_t(u64, lastuse, lastuse2);
4125 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4127 flow_stats_update(&f->stats, bytes, packets, lastuse);
4128 trace_mlx5e_stats_flower(f);
4130 mlx5e_flow_put(priv, flow);
4134 static int apply_police_params(struct mlx5e_priv *priv, u32 rate,
4135 struct netlink_ext_ack *extack)
4137 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4138 struct mlx5_eswitch *esw;
4143 vport_num = rpriv->rep->vport;
4144 if (vport_num >= MLX5_VPORT_ECPF) {
4145 NL_SET_ERR_MSG_MOD(extack,
4146 "Ingress rate limit is supported only for Eswitch ports connected to VFs");
4150 esw = priv->mdev->priv.eswitch;
4151 /* rate is given in bytes/sec.
4152 * First convert to bits/sec and then round to the nearest mbit/secs.
4153 * mbit means million bits.
4154 * Moreover, if rate is non zero we choose to configure to a minimum of
4157 rate_mbps = rate ? max_t(u32, (rate * 8 + 500000) / 1000000, 1) : 0;
4158 err = mlx5_esw_modify_vport_rate(esw, vport_num, rate_mbps);
4160 NL_SET_ERR_MSG_MOD(extack, "failed applying action to hardware");
4165 static int scan_tc_matchall_fdb_actions(struct mlx5e_priv *priv,
4166 struct flow_action *flow_action,
4167 struct netlink_ext_ack *extack)
4169 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4170 const struct flow_action_entry *act;
4174 if (!flow_action_has_entries(flow_action)) {
4175 NL_SET_ERR_MSG_MOD(extack, "matchall called with no action");
4179 if (!flow_offload_has_one_action(flow_action)) {
4180 NL_SET_ERR_MSG_MOD(extack, "matchall policing support only a single action");
4184 if (!flow_action_basic_hw_stats_types_check(flow_action, extack))
4187 flow_action_for_each(i, act, flow_action) {
4189 case FLOW_ACTION_POLICE:
4190 err = apply_police_params(priv, act->police.rate_bytes_ps, extack);
4194 rpriv->prev_vf_vport_stats = priv->stats.vf_vport;
4197 NL_SET_ERR_MSG_MOD(extack, "mlx5 supports only police action for matchall");
4205 int mlx5e_tc_configure_matchall(struct mlx5e_priv *priv,
4206 struct tc_cls_matchall_offload *ma)
4208 struct netlink_ext_ack *extack = ma->common.extack;
4210 if (ma->common.prio != 1) {
4211 NL_SET_ERR_MSG_MOD(extack, "only priority 1 is supported");
4215 return scan_tc_matchall_fdb_actions(priv, &ma->rule->action, extack);
4218 int mlx5e_tc_delete_matchall(struct mlx5e_priv *priv,
4219 struct tc_cls_matchall_offload *ma)
4221 struct netlink_ext_ack *extack = ma->common.extack;
4223 return apply_police_params(priv, 0, extack);
4226 void mlx5e_tc_stats_matchall(struct mlx5e_priv *priv,
4227 struct tc_cls_matchall_offload *ma)
4229 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4230 struct rtnl_link_stats64 cur_stats;
4234 cur_stats = priv->stats.vf_vport;
4235 dpkts = cur_stats.rx_packets - rpriv->prev_vf_vport_stats.rx_packets;
4236 dbytes = cur_stats.rx_bytes - rpriv->prev_vf_vport_stats.rx_bytes;
4237 rpriv->prev_vf_vport_stats = cur_stats;
4238 flow_stats_update(&ma->stats, dpkts, dbytes, jiffies);
4241 static void mlx5e_tc_hairpin_update_dead_peer(struct mlx5e_priv *priv,
4242 struct mlx5e_priv *peer_priv)
4244 struct mlx5_core_dev *peer_mdev = peer_priv->mdev;
4245 struct mlx5e_hairpin_entry *hpe, *tmp;
4246 LIST_HEAD(init_wait_list);
4250 if (!same_hw_devs(priv, peer_priv))
4253 peer_vhca_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
4255 mutex_lock(&priv->fs.tc.hairpin_tbl_lock);
4256 hash_for_each(priv->fs.tc.hairpin_tbl, bkt, hpe, hairpin_hlist)
4257 if (refcount_inc_not_zero(&hpe->refcnt))
4258 list_add(&hpe->dead_peer_wait_list, &init_wait_list);
4259 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
4261 list_for_each_entry_safe(hpe, tmp, &init_wait_list, dead_peer_wait_list) {
4262 wait_for_completion(&hpe->res_ready);
4263 if (!IS_ERR_OR_NULL(hpe->hp) && hpe->peer_vhca_id == peer_vhca_id)
4264 hpe->hp->pair->peer_gone = true;
4266 mlx5e_hairpin_put(priv, hpe);
4270 static int mlx5e_tc_netdev_event(struct notifier_block *this,
4271 unsigned long event, void *ptr)
4273 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4274 struct mlx5e_flow_steering *fs;
4275 struct mlx5e_priv *peer_priv;
4276 struct mlx5e_tc_table *tc;
4277 struct mlx5e_priv *priv;
4279 if (ndev->netdev_ops != &mlx5e_netdev_ops ||
4280 event != NETDEV_UNREGISTER ||
4281 ndev->reg_state == NETREG_REGISTERED)
4284 tc = container_of(this, struct mlx5e_tc_table, netdevice_nb);
4285 fs = container_of(tc, struct mlx5e_flow_steering, tc);
4286 priv = container_of(fs, struct mlx5e_priv, fs);
4287 peer_priv = netdev_priv(ndev);
4288 if (priv == peer_priv ||
4289 !(priv->netdev->features & NETIF_F_HW_TC))
4292 mlx5e_tc_hairpin_update_dead_peer(priv, peer_priv);
4297 int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
4299 struct mlx5e_tc_table *tc = &priv->fs.tc;
4302 mutex_init(&tc->t_lock);
4303 mutex_init(&tc->mod_hdr.lock);
4304 hash_init(tc->mod_hdr.hlist);
4305 mutex_init(&tc->hairpin_tbl_lock);
4306 hash_init(tc->hairpin_tbl);
4308 err = rhashtable_init(&tc->ht, &tc_ht_params);
4312 tc->netdevice_nb.notifier_call = mlx5e_tc_netdev_event;
4313 err = register_netdevice_notifier_dev_net(priv->netdev,
4317 tc->netdevice_nb.notifier_call = NULL;
4318 mlx5_core_warn(priv->mdev, "Failed to register netdev notifier\n");
4324 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
4326 struct mlx5e_tc_flow *flow = ptr;
4327 struct mlx5e_priv *priv = flow->priv;
4329 mlx5e_tc_del_flow(priv, flow);
4333 void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
4335 struct mlx5e_tc_table *tc = &priv->fs.tc;
4337 if (tc->netdevice_nb.notifier_call)
4338 unregister_netdevice_notifier_dev_net(priv->netdev,
4342 mutex_destroy(&tc->mod_hdr.lock);
4343 mutex_destroy(&tc->hairpin_tbl_lock);
4345 rhashtable_destroy(&tc->ht);
4347 if (!IS_ERR_OR_NULL(tc->t)) {
4348 mlx5_destroy_flow_table(tc->t);
4351 mutex_destroy(&tc->t_lock);
4354 int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
4356 return rhashtable_init(tc_ht, &tc_ht_params);
4359 void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
4361 rhashtable_free_and_destroy(tc_ht, _mlx5e_tc_del_flow, NULL);
4364 int mlx5e_tc_num_filters(struct mlx5e_priv *priv, unsigned long flags)
4366 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4368 return atomic_read(&tc_ht->nelems);
4371 void mlx5e_tc_clean_fdb_peer_flows(struct mlx5_eswitch *esw)
4373 struct mlx5e_tc_flow *flow, *tmp;
4375 list_for_each_entry_safe(flow, tmp, &esw->offloads.peer_flows, peer)
4376 __mlx5e_tc_del_fdb_peer_flow(flow);
4379 void mlx5e_tc_reoffload_flows_work(struct work_struct *work)
4381 struct mlx5_rep_uplink_priv *rpriv =
4382 container_of(work, struct mlx5_rep_uplink_priv,
4383 reoffload_flows_work);
4384 struct mlx5e_tc_flow *flow, *tmp;
4386 mutex_lock(&rpriv->unready_flows_lock);
4387 list_for_each_entry_safe(flow, tmp, &rpriv->unready_flows, unready) {
4388 if (!mlx5e_tc_add_fdb_flow(flow->priv, flow, NULL))
4389 unready_flow_del(flow);
4391 mutex_unlock(&rpriv->unready_flows_lock);