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,
1080 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,
1099 struct mlx5_esw_flow_attr *slow_attr)
1101 memcpy(slow_attr, flow->esw_attr, sizeof(*slow_attr));
1102 slow_attr->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1103 slow_attr->split_count = 0;
1104 slow_attr->flags |= MLX5_ESW_ATTR_FLAG_SLOW_PATH;
1105 mlx5e_tc_unoffload_fdb_rules(esw, flow, slow_attr);
1106 flow_flag_clear(flow, SLOW);
1109 /* Caller must obtain uplink_priv->unready_flows_lock mutex before calling this
1112 static void unready_flow_add(struct mlx5e_tc_flow *flow,
1113 struct list_head *unready_flows)
1115 flow_flag_set(flow, NOT_READY);
1116 list_add_tail(&flow->unready, unready_flows);
1119 /* Caller must obtain uplink_priv->unready_flows_lock mutex before calling this
1122 static void unready_flow_del(struct mlx5e_tc_flow *flow)
1124 list_del(&flow->unready);
1125 flow_flag_clear(flow, NOT_READY);
1128 static void add_unready_flow(struct mlx5e_tc_flow *flow)
1130 struct mlx5_rep_uplink_priv *uplink_priv;
1131 struct mlx5e_rep_priv *rpriv;
1132 struct mlx5_eswitch *esw;
1134 esw = flow->priv->mdev->priv.eswitch;
1135 rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1136 uplink_priv = &rpriv->uplink_priv;
1138 mutex_lock(&uplink_priv->unready_flows_lock);
1139 unready_flow_add(flow, &uplink_priv->unready_flows);
1140 mutex_unlock(&uplink_priv->unready_flows_lock);
1143 static void remove_unready_flow(struct mlx5e_tc_flow *flow)
1145 struct mlx5_rep_uplink_priv *uplink_priv;
1146 struct mlx5e_rep_priv *rpriv;
1147 struct mlx5_eswitch *esw;
1149 esw = flow->priv->mdev->priv.eswitch;
1150 rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1151 uplink_priv = &rpriv->uplink_priv;
1153 mutex_lock(&uplink_priv->unready_flows_lock);
1154 unready_flow_del(flow);
1155 mutex_unlock(&uplink_priv->unready_flows_lock);
1159 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
1160 struct mlx5e_tc_flow *flow,
1161 struct netlink_ext_ack *extack)
1163 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1164 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1165 struct mlx5e_tc_flow_parse_attr *parse_attr = attr->parse_attr;
1166 struct net_device *out_dev, *encap_dev = NULL;
1167 struct mlx5_fc *counter = NULL;
1168 struct mlx5e_rep_priv *rpriv;
1169 struct mlx5e_priv *out_priv;
1170 bool encap_valid = true;
1171 u32 max_prio, max_chain;
1175 if (!mlx5_esw_chains_prios_supported(esw) && attr->prio != 1) {
1176 NL_SET_ERR_MSG_MOD(extack,
1177 "E-switch priorities unsupported, upgrade FW");
1181 /* We check chain range only for tc flows.
1182 * For ft flows, we checked attr->chain was originally 0 and set it to
1183 * FDB_FT_CHAIN which is outside tc range.
1184 * See mlx5e_rep_setup_ft_cb().
1186 max_chain = mlx5_esw_chains_get_chain_range(esw);
1187 if (!mlx5e_is_ft_flow(flow) && attr->chain > max_chain) {
1188 NL_SET_ERR_MSG_MOD(extack,
1189 "Requested chain is out of supported range");
1193 max_prio = mlx5_esw_chains_get_prio_range(esw);
1194 if (attr->prio > max_prio) {
1195 NL_SET_ERR_MSG_MOD(extack,
1196 "Requested priority is out of supported range");
1200 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++) {
1203 if (!(attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP))
1206 mirred_ifindex = parse_attr->mirred_ifindex[out_index];
1207 out_dev = __dev_get_by_index(dev_net(priv->netdev),
1209 err = mlx5e_attach_encap(priv, flow, out_dev, out_index,
1210 extack, &encap_dev, &encap_valid);
1214 out_priv = netdev_priv(encap_dev);
1215 rpriv = out_priv->ppriv;
1216 attr->dests[out_index].rep = rpriv->rep;
1217 attr->dests[out_index].mdev = out_priv->mdev;
1220 err = mlx5_eswitch_add_vlan_action(esw, attr);
1224 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
1225 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
1226 kfree(parse_attr->mod_hdr_actions);
1231 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1232 counter = mlx5_fc_create(attr->counter_dev, true);
1233 if (IS_ERR(counter))
1234 return PTR_ERR(counter);
1236 attr->counter = counter;
1239 /* we get here if one of the following takes place:
1240 * (1) there's no error
1241 * (2) there's an encap action and we don't have valid neigh
1244 /* continue with goto slow path rule instead */
1245 struct mlx5_esw_flow_attr slow_attr;
1247 flow->rule[0] = mlx5e_tc_offload_to_slow_path(esw, flow, &parse_attr->spec, &slow_attr);
1249 flow->rule[0] = mlx5e_tc_offload_fdb_rules(esw, flow, &parse_attr->spec, attr);
1252 if (IS_ERR(flow->rule[0]))
1253 return PTR_ERR(flow->rule[0]);
1255 flow_flag_set(flow, OFFLOADED);
1260 static bool mlx5_flow_has_geneve_opt(struct mlx5e_tc_flow *flow)
1262 struct mlx5_flow_spec *spec = &flow->esw_attr->parse_attr->spec;
1263 void *headers_v = MLX5_ADDR_OF(fte_match_param,
1266 u32 geneve_tlv_opt_0_data = MLX5_GET(fte_match_set_misc3,
1268 geneve_tlv_option_0_data);
1270 return !!geneve_tlv_opt_0_data;
1273 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
1274 struct mlx5e_tc_flow *flow)
1276 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1277 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
1278 struct mlx5_esw_flow_attr slow_attr;
1281 if (flow_flag_test(flow, NOT_READY)) {
1282 remove_unready_flow(flow);
1283 kvfree(attr->parse_attr);
1287 if (mlx5e_is_offloaded_flow(flow)) {
1288 if (flow_flag_test(flow, SLOW))
1289 mlx5e_tc_unoffload_from_slow_path(esw, flow, &slow_attr);
1291 mlx5e_tc_unoffload_fdb_rules(esw, flow, attr);
1294 if (mlx5_flow_has_geneve_opt(flow))
1295 mlx5_geneve_tlv_option_del(priv->mdev->geneve);
1297 mlx5_eswitch_del_vlan_action(esw, attr);
1299 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++)
1300 if (attr->dests[out_index].flags & MLX5_ESW_DEST_ENCAP) {
1301 mlx5e_detach_encap(priv, flow, out_index);
1302 kfree(attr->parse_attr->tun_info[out_index]);
1304 kvfree(attr->parse_attr);
1306 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1307 mlx5e_detach_mod_hdr(priv, flow);
1309 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
1310 mlx5_fc_destroy(attr->counter_dev, attr->counter);
1313 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
1314 struct mlx5e_encap_entry *e,
1315 struct list_head *flow_list)
1317 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1318 struct mlx5_esw_flow_attr slow_attr, *esw_attr;
1319 struct mlx5_flow_handle *rule;
1320 struct mlx5_flow_spec *spec;
1321 struct mlx5e_tc_flow *flow;
1324 e->pkt_reformat = mlx5_packet_reformat_alloc(priv->mdev,
1326 e->encap_size, e->encap_header,
1327 MLX5_FLOW_NAMESPACE_FDB);
1328 if (IS_ERR(e->pkt_reformat)) {
1329 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %lu\n",
1330 PTR_ERR(e->pkt_reformat));
1333 e->flags |= MLX5_ENCAP_ENTRY_VALID;
1334 mlx5e_rep_queue_neigh_stats_work(priv);
1336 list_for_each_entry(flow, flow_list, tmp_list) {
1337 bool all_flow_encaps_valid = true;
1340 if (!mlx5e_is_offloaded_flow(flow))
1342 esw_attr = flow->esw_attr;
1343 spec = &esw_attr->parse_attr->spec;
1345 esw_attr->dests[flow->tmp_efi_index].pkt_reformat = e->pkt_reformat;
1346 esw_attr->dests[flow->tmp_efi_index].flags |= MLX5_ESW_DEST_ENCAP_VALID;
1347 /* Flow can be associated with multiple encap entries.
1348 * Before offloading the flow verify that all of them have
1349 * a valid neighbour.
1351 for (i = 0; i < MLX5_MAX_FLOW_FWD_VPORTS; i++) {
1352 if (!(esw_attr->dests[i].flags & MLX5_ESW_DEST_ENCAP))
1354 if (!(esw_attr->dests[i].flags & MLX5_ESW_DEST_ENCAP_VALID)) {
1355 all_flow_encaps_valid = false;
1359 /* Do not offload flows with unresolved neighbors */
1360 if (!all_flow_encaps_valid)
1362 /* update from slow path rule to encap rule */
1363 rule = mlx5e_tc_offload_fdb_rules(esw, flow, spec, esw_attr);
1365 err = PTR_ERR(rule);
1366 mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
1371 mlx5e_tc_unoffload_from_slow_path(esw, flow, &slow_attr);
1372 flow->rule[0] = rule;
1373 /* was unset when slow path rule removed */
1374 flow_flag_set(flow, OFFLOADED);
1378 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
1379 struct mlx5e_encap_entry *e,
1380 struct list_head *flow_list)
1382 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1383 struct mlx5_esw_flow_attr slow_attr;
1384 struct mlx5_flow_handle *rule;
1385 struct mlx5_flow_spec *spec;
1386 struct mlx5e_tc_flow *flow;
1389 list_for_each_entry(flow, flow_list, tmp_list) {
1390 if (!mlx5e_is_offloaded_flow(flow))
1392 spec = &flow->esw_attr->parse_attr->spec;
1394 /* update from encap rule to slow path rule */
1395 rule = mlx5e_tc_offload_to_slow_path(esw, flow, spec, &slow_attr);
1396 /* mark the flow's encap dest as non-valid */
1397 flow->esw_attr->dests[flow->tmp_efi_index].flags &= ~MLX5_ESW_DEST_ENCAP_VALID;
1400 err = PTR_ERR(rule);
1401 mlx5_core_warn(priv->mdev, "Failed to update slow path (encap) flow, %d\n",
1406 mlx5e_tc_unoffload_fdb_rules(esw, flow, flow->esw_attr);
1407 flow->rule[0] = rule;
1408 /* was unset when fast path rule removed */
1409 flow_flag_set(flow, OFFLOADED);
1412 /* we know that the encap is valid */
1413 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
1414 mlx5_packet_reformat_dealloc(priv->mdev, e->pkt_reformat);
1417 static struct mlx5_fc *mlx5e_tc_get_counter(struct mlx5e_tc_flow *flow)
1419 if (mlx5e_is_eswitch_flow(flow))
1420 return flow->esw_attr->counter;
1422 return flow->nic_attr->counter;
1425 /* Takes reference to all flows attached to encap and adds the flows to
1426 * flow_list using 'tmp_list' list_head in mlx5e_tc_flow.
1428 void mlx5e_take_all_encap_flows(struct mlx5e_encap_entry *e, struct list_head *flow_list)
1430 struct encap_flow_item *efi;
1431 struct mlx5e_tc_flow *flow;
1433 list_for_each_entry(efi, &e->flows, list) {
1434 flow = container_of(efi, struct mlx5e_tc_flow, encaps[efi->index]);
1435 if (IS_ERR(mlx5e_flow_get(flow)))
1437 wait_for_completion(&flow->init_done);
1439 flow->tmp_efi_index = efi->index;
1440 list_add(&flow->tmp_list, flow_list);
1444 /* Iterate over tmp_list of flows attached to flow_list head. */
1445 void mlx5e_put_encap_flow_list(struct mlx5e_priv *priv, struct list_head *flow_list)
1447 struct mlx5e_tc_flow *flow, *tmp;
1449 list_for_each_entry_safe(flow, tmp, flow_list, tmp_list)
1450 mlx5e_flow_put(priv, flow);
1453 static struct mlx5e_encap_entry *
1454 mlx5e_get_next_valid_encap(struct mlx5e_neigh_hash_entry *nhe,
1455 struct mlx5e_encap_entry *e)
1457 struct mlx5e_encap_entry *next = NULL;
1462 /* find encap with non-zero reference counter value */
1464 list_next_or_null_rcu(&nhe->encap_list,
1466 struct mlx5e_encap_entry,
1468 list_first_or_null_rcu(&nhe->encap_list,
1469 struct mlx5e_encap_entry,
1472 next = list_next_or_null_rcu(&nhe->encap_list,
1474 struct mlx5e_encap_entry,
1476 if (mlx5e_encap_take(next))
1481 /* release starting encap */
1483 mlx5e_encap_put(netdev_priv(e->out_dev), e);
1487 /* wait for encap to be fully initialized */
1488 wait_for_completion(&next->res_ready);
1489 /* continue searching if encap entry is not in valid state after completion */
1490 if (!(next->flags & MLX5_ENCAP_ENTRY_VALID)) {
1498 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
1500 struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
1501 struct mlx5e_encap_entry *e = NULL;
1502 struct mlx5e_tc_flow *flow;
1503 struct mlx5_fc *counter;
1504 struct neigh_table *tbl;
1505 bool neigh_used = false;
1506 struct neighbour *n;
1509 if (m_neigh->family == AF_INET)
1511 #if IS_ENABLED(CONFIG_IPV6)
1512 else if (m_neigh->family == AF_INET6)
1513 tbl = ipv6_stub->nd_tbl;
1518 /* mlx5e_get_next_valid_encap() releases previous encap before returning
1521 while ((e = mlx5e_get_next_valid_encap(nhe, e)) != NULL) {
1522 struct mlx5e_priv *priv = netdev_priv(e->out_dev);
1523 struct encap_flow_item *efi, *tmp;
1524 struct mlx5_eswitch *esw;
1525 LIST_HEAD(flow_list);
1527 esw = priv->mdev->priv.eswitch;
1528 mutex_lock(&esw->offloads.encap_tbl_lock);
1529 list_for_each_entry_safe(efi, tmp, &e->flows, list) {
1530 flow = container_of(efi, struct mlx5e_tc_flow,
1531 encaps[efi->index]);
1532 if (IS_ERR(mlx5e_flow_get(flow)))
1534 list_add(&flow->tmp_list, &flow_list);
1536 if (mlx5e_is_offloaded_flow(flow)) {
1537 counter = mlx5e_tc_get_counter(flow);
1538 lastuse = mlx5_fc_query_lastuse(counter);
1539 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
1545 mutex_unlock(&esw->offloads.encap_tbl_lock);
1547 mlx5e_put_encap_flow_list(priv, &flow_list);
1549 /* release current encap before breaking the loop */
1550 mlx5e_encap_put(priv, e);
1555 trace_mlx5e_tc_update_neigh_used_value(nhe, neigh_used);
1558 nhe->reported_lastuse = jiffies;
1560 /* find the relevant neigh according to the cached device and
1563 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
1567 neigh_event_send(n, NULL);
1572 static void mlx5e_encap_dealloc(struct mlx5e_priv *priv, struct mlx5e_encap_entry *e)
1574 WARN_ON(!list_empty(&e->flows));
1576 if (e->compl_result > 0) {
1577 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1579 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1580 mlx5_packet_reformat_dealloc(priv->mdev, e->pkt_reformat);
1584 kfree(e->encap_header);
1588 void mlx5e_encap_put(struct mlx5e_priv *priv, struct mlx5e_encap_entry *e)
1590 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1592 if (!refcount_dec_and_mutex_lock(&e->refcnt, &esw->offloads.encap_tbl_lock))
1594 hash_del_rcu(&e->encap_hlist);
1595 mutex_unlock(&esw->offloads.encap_tbl_lock);
1597 mlx5e_encap_dealloc(priv, e);
1600 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1601 struct mlx5e_tc_flow *flow, int out_index)
1603 struct mlx5e_encap_entry *e = flow->encaps[out_index].e;
1604 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1606 /* flow wasn't fully initialized */
1610 mutex_lock(&esw->offloads.encap_tbl_lock);
1611 list_del(&flow->encaps[out_index].list);
1612 flow->encaps[out_index].e = NULL;
1613 if (!refcount_dec_and_test(&e->refcnt)) {
1614 mutex_unlock(&esw->offloads.encap_tbl_lock);
1617 hash_del_rcu(&e->encap_hlist);
1618 mutex_unlock(&esw->offloads.encap_tbl_lock);
1620 mlx5e_encap_dealloc(priv, e);
1623 static void __mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1625 struct mlx5_eswitch *esw = flow->priv->mdev->priv.eswitch;
1627 if (!flow_flag_test(flow, ESWITCH) ||
1628 !flow_flag_test(flow, DUP))
1631 mutex_lock(&esw->offloads.peer_mutex);
1632 list_del(&flow->peer);
1633 mutex_unlock(&esw->offloads.peer_mutex);
1635 flow_flag_clear(flow, DUP);
1637 if (refcount_dec_and_test(&flow->peer_flow->refcnt)) {
1638 mlx5e_tc_del_fdb_flow(flow->peer_flow->priv, flow->peer_flow);
1639 kfree(flow->peer_flow);
1642 flow->peer_flow = NULL;
1645 static void mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
1647 struct mlx5_core_dev *dev = flow->priv->mdev;
1648 struct mlx5_devcom *devcom = dev->priv.devcom;
1649 struct mlx5_eswitch *peer_esw;
1651 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1655 __mlx5e_tc_del_fdb_peer_flow(flow);
1656 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1659 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1660 struct mlx5e_tc_flow *flow)
1662 if (mlx5e_is_eswitch_flow(flow)) {
1663 mlx5e_tc_del_fdb_peer_flow(flow);
1664 mlx5e_tc_del_fdb_flow(priv, flow);
1666 mlx5e_tc_del_nic_flow(priv, flow);
1671 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1672 struct mlx5_flow_spec *spec,
1673 struct flow_cls_offload *f,
1674 struct net_device *filter_dev, u8 *match_level)
1676 struct netlink_ext_ack *extack = f->common.extack;
1677 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1679 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1681 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1684 err = mlx5e_tc_tun_parse(filter_dev, priv, spec, f,
1685 headers_c, headers_v, match_level);
1687 NL_SET_ERR_MSG_MOD(extack,
1688 "failed to parse tunnel attributes");
1692 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
1693 struct flow_match_control match;
1696 flow_rule_match_enc_control(rule, &match);
1697 addr_type = match.key->addr_type;
1699 /* For tunnel addr_type used same key id`s as for non-tunnel */
1700 if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1701 struct flow_match_ipv4_addrs match;
1703 flow_rule_match_enc_ipv4_addrs(rule, &match);
1704 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1705 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1706 ntohl(match.mask->src));
1707 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1708 src_ipv4_src_ipv6.ipv4_layout.ipv4,
1709 ntohl(match.key->src));
1711 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1712 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1713 ntohl(match.mask->dst));
1714 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1715 dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1716 ntohl(match.key->dst));
1718 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c,
1720 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1722 } else if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1723 struct flow_match_ipv6_addrs match;
1725 flow_rule_match_enc_ipv6_addrs(rule, &match);
1726 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1727 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1728 &match.mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout,
1730 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1731 src_ipv4_src_ipv6.ipv6_layout.ipv6),
1732 &match.key->src, MLX5_FLD_SZ_BYTES(ipv6_layout,
1735 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1736 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1737 &match.mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout,
1739 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1740 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1741 &match.key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout,
1744 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c,
1746 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1751 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_IP)) {
1752 struct flow_match_ip match;
1754 flow_rule_match_enc_ip(rule, &match);
1755 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn,
1756 match.mask->tos & 0x3);
1757 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn,
1758 match.key->tos & 0x3);
1760 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp,
1761 match.mask->tos >> 2);
1762 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp,
1763 match.key->tos >> 2);
1765 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit,
1767 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit,
1770 if (match.mask->ttl &&
1771 !MLX5_CAP_ESW_FLOWTABLE_FDB
1773 ft_field_support.outer_ipv4_ttl)) {
1774 NL_SET_ERR_MSG_MOD(extack,
1775 "Matching on TTL is not supported");
1781 /* Enforce DMAC when offloading incoming tunneled flows.
1782 * Flow counters require a match on the DMAC.
1784 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
1785 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
1786 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1787 dmac_47_16), priv->netdev->dev_addr);
1789 /* let software handle IP fragments */
1790 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1791 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
1796 static void *get_match_headers_criteria(u32 flags,
1797 struct mlx5_flow_spec *spec)
1799 return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
1800 MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1802 MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1806 static void *get_match_headers_value(u32 flags,
1807 struct mlx5_flow_spec *spec)
1809 return (flags & MLX5_FLOW_CONTEXT_ACTION_DECAP) ?
1810 MLX5_ADDR_OF(fte_match_param, spec->match_value,
1812 MLX5_ADDR_OF(fte_match_param, spec->match_value,
1816 static int mlx5e_flower_parse_meta(struct net_device *filter_dev,
1817 struct flow_cls_offload *f)
1819 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1820 struct netlink_ext_ack *extack = f->common.extack;
1821 struct net_device *ingress_dev;
1822 struct flow_match_meta match;
1824 if (!flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_META))
1827 flow_rule_match_meta(rule, &match);
1828 if (match.mask->ingress_ifindex != 0xFFFFFFFF) {
1829 NL_SET_ERR_MSG_MOD(extack, "Unsupported ingress ifindex mask");
1833 ingress_dev = __dev_get_by_index(dev_net(filter_dev),
1834 match.key->ingress_ifindex);
1836 NL_SET_ERR_MSG_MOD(extack,
1837 "Can't find the ingress port to match on");
1841 if (ingress_dev != filter_dev) {
1842 NL_SET_ERR_MSG_MOD(extack,
1843 "Can't match on the ingress filter port");
1850 static int __parse_cls_flower(struct mlx5e_priv *priv,
1851 struct mlx5_flow_spec *spec,
1852 struct flow_cls_offload *f,
1853 struct net_device *filter_dev,
1854 u8 *inner_match_level, u8 *outer_match_level)
1856 struct netlink_ext_ack *extack = f->common.extack;
1857 void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1859 void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1861 void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1863 void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1865 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
1866 struct flow_dissector *dissector = rule->match.dissector;
1872 match_level = outer_match_level;
1874 if (dissector->used_keys &
1875 ~(BIT(FLOW_DISSECTOR_KEY_META) |
1876 BIT(FLOW_DISSECTOR_KEY_CONTROL) |
1877 BIT(FLOW_DISSECTOR_KEY_BASIC) |
1878 BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
1879 BIT(FLOW_DISSECTOR_KEY_VLAN) |
1880 BIT(FLOW_DISSECTOR_KEY_CVLAN) |
1881 BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
1882 BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
1883 BIT(FLOW_DISSECTOR_KEY_PORTS) |
1884 BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
1885 BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
1886 BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
1887 BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
1888 BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
1889 BIT(FLOW_DISSECTOR_KEY_TCP) |
1890 BIT(FLOW_DISSECTOR_KEY_IP) |
1891 BIT(FLOW_DISSECTOR_KEY_ENC_IP) |
1892 BIT(FLOW_DISSECTOR_KEY_ENC_OPTS))) {
1893 NL_SET_ERR_MSG_MOD(extack, "Unsupported key");
1894 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
1895 dissector->used_keys);
1899 if (mlx5e_get_tc_tun(filter_dev)) {
1900 if (parse_tunnel_attr(priv, spec, f, filter_dev,
1904 /* At this point, header pointers should point to the inner
1905 * headers, outer header were already set by parse_tunnel_attr
1907 match_level = inner_match_level;
1908 headers_c = get_match_headers_criteria(MLX5_FLOW_CONTEXT_ACTION_DECAP,
1910 headers_v = get_match_headers_value(MLX5_FLOW_CONTEXT_ACTION_DECAP,
1914 err = mlx5e_flower_parse_meta(filter_dev, f);
1918 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
1919 struct flow_match_basic match;
1921 flow_rule_match_basic(rule, &match);
1922 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1923 ntohs(match.mask->n_proto));
1924 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1925 ntohs(match.key->n_proto));
1927 if (match.mask->n_proto)
1928 *match_level = MLX5_MATCH_L2;
1930 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN) ||
1931 is_vlan_dev(filter_dev)) {
1932 struct flow_dissector_key_vlan filter_dev_mask;
1933 struct flow_dissector_key_vlan filter_dev_key;
1934 struct flow_match_vlan match;
1936 if (is_vlan_dev(filter_dev)) {
1937 match.key = &filter_dev_key;
1938 match.key->vlan_id = vlan_dev_vlan_id(filter_dev);
1939 match.key->vlan_tpid = vlan_dev_vlan_proto(filter_dev);
1940 match.key->vlan_priority = 0;
1941 match.mask = &filter_dev_mask;
1942 memset(match.mask, 0xff, sizeof(*match.mask));
1943 match.mask->vlan_priority = 0;
1945 flow_rule_match_vlan(rule, &match);
1947 if (match.mask->vlan_id ||
1948 match.mask->vlan_priority ||
1949 match.mask->vlan_tpid) {
1950 if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
1951 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1953 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1956 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1958 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1962 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid,
1963 match.mask->vlan_id);
1964 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid,
1965 match.key->vlan_id);
1967 MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio,
1968 match.mask->vlan_priority);
1969 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio,
1970 match.key->vlan_priority);
1972 *match_level = MLX5_MATCH_L2;
1974 } else if (*match_level != MLX5_MATCH_NONE) {
1975 /* cvlan_tag enabled in match criteria and
1976 * disabled in match value means both S & C tags
1977 * don't exist (untagged of both)
1979 MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
1980 *match_level = MLX5_MATCH_L2;
1983 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
1984 struct flow_match_vlan match;
1986 flow_rule_match_cvlan(rule, &match);
1987 if (match.mask->vlan_id ||
1988 match.mask->vlan_priority ||
1989 match.mask->vlan_tpid) {
1990 if (match.key->vlan_tpid == htons(ETH_P_8021AD)) {
1991 MLX5_SET(fte_match_set_misc, misc_c,
1992 outer_second_svlan_tag, 1);
1993 MLX5_SET(fte_match_set_misc, misc_v,
1994 outer_second_svlan_tag, 1);
1996 MLX5_SET(fte_match_set_misc, misc_c,
1997 outer_second_cvlan_tag, 1);
1998 MLX5_SET(fte_match_set_misc, misc_v,
1999 outer_second_cvlan_tag, 1);
2002 MLX5_SET(fte_match_set_misc, misc_c, outer_second_vid,
2003 match.mask->vlan_id);
2004 MLX5_SET(fte_match_set_misc, misc_v, outer_second_vid,
2005 match.key->vlan_id);
2006 MLX5_SET(fte_match_set_misc, misc_c, outer_second_prio,
2007 match.mask->vlan_priority);
2008 MLX5_SET(fte_match_set_misc, misc_v, outer_second_prio,
2009 match.key->vlan_priority);
2011 *match_level = MLX5_MATCH_L2;
2015 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2016 struct flow_match_eth_addrs match;
2018 flow_rule_match_eth_addrs(rule, &match);
2019 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2022 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2026 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2029 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2033 if (!is_zero_ether_addr(match.mask->src) ||
2034 !is_zero_ether_addr(match.mask->dst))
2035 *match_level = MLX5_MATCH_L2;
2038 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
2039 struct flow_match_control match;
2041 flow_rule_match_control(rule, &match);
2042 addr_type = match.key->addr_type;
2044 /* the HW doesn't support frag first/later */
2045 if (match.mask->flags & FLOW_DIS_FIRST_FRAG)
2048 if (match.mask->flags & FLOW_DIS_IS_FRAGMENT) {
2049 MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
2050 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
2051 match.key->flags & FLOW_DIS_IS_FRAGMENT);
2053 /* the HW doesn't need L3 inline to match on frag=no */
2054 if (!(match.key->flags & FLOW_DIS_IS_FRAGMENT))
2055 *match_level = MLX5_MATCH_L2;
2056 /* *** L2 attributes parsing up to here *** */
2058 *match_level = MLX5_MATCH_L3;
2062 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2063 struct flow_match_basic match;
2065 flow_rule_match_basic(rule, &match);
2066 ip_proto = match.key->ip_proto;
2068 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
2069 match.mask->ip_proto);
2070 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
2071 match.key->ip_proto);
2073 if (match.mask->ip_proto)
2074 *match_level = MLX5_MATCH_L3;
2077 if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2078 struct flow_match_ipv4_addrs match;
2080 flow_rule_match_ipv4_addrs(rule, &match);
2081 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2082 src_ipv4_src_ipv6.ipv4_layout.ipv4),
2083 &match.mask->src, sizeof(match.mask->src));
2084 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2085 src_ipv4_src_ipv6.ipv4_layout.ipv4),
2086 &match.key->src, sizeof(match.key->src));
2087 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2088 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2089 &match.mask->dst, sizeof(match.mask->dst));
2090 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2091 dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2092 &match.key->dst, sizeof(match.key->dst));
2094 if (match.mask->src || match.mask->dst)
2095 *match_level = MLX5_MATCH_L3;
2098 if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2099 struct flow_match_ipv6_addrs match;
2101 flow_rule_match_ipv6_addrs(rule, &match);
2102 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2103 src_ipv4_src_ipv6.ipv6_layout.ipv6),
2104 &match.mask->src, sizeof(match.mask->src));
2105 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2106 src_ipv4_src_ipv6.ipv6_layout.ipv6),
2107 &match.key->src, sizeof(match.key->src));
2109 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2110 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2111 &match.mask->dst, sizeof(match.mask->dst));
2112 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2113 dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2114 &match.key->dst, sizeof(match.key->dst));
2116 if (ipv6_addr_type(&match.mask->src) != IPV6_ADDR_ANY ||
2117 ipv6_addr_type(&match.mask->dst) != IPV6_ADDR_ANY)
2118 *match_level = MLX5_MATCH_L3;
2121 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
2122 struct flow_match_ip match;
2124 flow_rule_match_ip(rule, &match);
2125 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn,
2126 match.mask->tos & 0x3);
2127 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn,
2128 match.key->tos & 0x3);
2130 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp,
2131 match.mask->tos >> 2);
2132 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp,
2133 match.key->tos >> 2);
2135 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit,
2137 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit,
2140 if (match.mask->ttl &&
2141 !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
2142 ft_field_support.outer_ipv4_ttl)) {
2143 NL_SET_ERR_MSG_MOD(extack,
2144 "Matching on TTL is not supported");
2148 if (match.mask->tos || match.mask->ttl)
2149 *match_level = MLX5_MATCH_L3;
2152 /* *** L3 attributes parsing up to here *** */
2154 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) {
2155 struct flow_match_ports match;
2157 flow_rule_match_ports(rule, &match);
2160 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2161 tcp_sport, ntohs(match.mask->src));
2162 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2163 tcp_sport, ntohs(match.key->src));
2165 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2166 tcp_dport, ntohs(match.mask->dst));
2167 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2168 tcp_dport, ntohs(match.key->dst));
2172 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2173 udp_sport, ntohs(match.mask->src));
2174 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2175 udp_sport, ntohs(match.key->src));
2177 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2178 udp_dport, ntohs(match.mask->dst));
2179 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2180 udp_dport, ntohs(match.key->dst));
2183 NL_SET_ERR_MSG_MOD(extack,
2184 "Only UDP and TCP transports are supported for L4 matching");
2185 netdev_err(priv->netdev,
2186 "Only UDP and TCP transport are supported\n");
2190 if (match.mask->src || match.mask->dst)
2191 *match_level = MLX5_MATCH_L4;
2194 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_TCP)) {
2195 struct flow_match_tcp match;
2197 flow_rule_match_tcp(rule, &match);
2198 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
2199 ntohs(match.mask->flags));
2200 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
2201 ntohs(match.key->flags));
2203 if (match.mask->flags)
2204 *match_level = MLX5_MATCH_L4;
2210 static int parse_cls_flower(struct mlx5e_priv *priv,
2211 struct mlx5e_tc_flow *flow,
2212 struct mlx5_flow_spec *spec,
2213 struct flow_cls_offload *f,
2214 struct net_device *filter_dev)
2216 u8 inner_match_level, outer_match_level, non_tunnel_match_level;
2217 struct netlink_ext_ack *extack = f->common.extack;
2218 struct mlx5_core_dev *dev = priv->mdev;
2219 struct mlx5_eswitch *esw = dev->priv.eswitch;
2220 struct mlx5e_rep_priv *rpriv = priv->ppriv;
2221 struct mlx5_eswitch_rep *rep;
2222 bool is_eswitch_flow;
2225 inner_match_level = MLX5_MATCH_NONE;
2226 outer_match_level = MLX5_MATCH_NONE;
2228 err = __parse_cls_flower(priv, spec, f, filter_dev, &inner_match_level,
2229 &outer_match_level);
2230 non_tunnel_match_level = (inner_match_level == MLX5_MATCH_NONE) ?
2231 outer_match_level : inner_match_level;
2233 is_eswitch_flow = mlx5e_is_eswitch_flow(flow);
2234 if (!err && is_eswitch_flow) {
2236 if (rep->vport != MLX5_VPORT_UPLINK &&
2237 (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
2238 esw->offloads.inline_mode < non_tunnel_match_level)) {
2239 NL_SET_ERR_MSG_MOD(extack,
2240 "Flow is not offloaded due to min inline setting");
2241 netdev_warn(priv->netdev,
2242 "Flow is not offloaded due to min inline setting, required %d actual %d\n",
2243 non_tunnel_match_level, esw->offloads.inline_mode);
2248 if (is_eswitch_flow) {
2249 flow->esw_attr->inner_match_level = inner_match_level;
2250 flow->esw_attr->outer_match_level = outer_match_level;
2252 flow->nic_attr->match_level = non_tunnel_match_level;
2258 struct pedit_headers {
2260 struct vlan_hdr vlan;
2267 struct pedit_headers_action {
2268 struct pedit_headers vals;
2269 struct pedit_headers masks;
2273 static int pedit_header_offsets[] = {
2274 [FLOW_ACT_MANGLE_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
2275 [FLOW_ACT_MANGLE_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
2276 [FLOW_ACT_MANGLE_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
2277 [FLOW_ACT_MANGLE_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
2278 [FLOW_ACT_MANGLE_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
2281 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
2283 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
2284 struct pedit_headers_action *hdrs)
2286 u32 *curr_pmask, *curr_pval;
2288 curr_pmask = (u32 *)(pedit_header(&hdrs->masks, hdr_type) + offset);
2289 curr_pval = (u32 *)(pedit_header(&hdrs->vals, hdr_type) + offset);
2291 if (*curr_pmask & mask) /* disallow acting twice on the same location */
2294 *curr_pmask |= mask;
2295 *curr_pval |= (val & mask);
2303 struct mlx5_fields {
2311 #define OFFLOAD(fw_field, field_bsize, field_mask, field, off, match_field) \
2312 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, field_bsize, field_mask, \
2313 offsetof(struct pedit_headers, field) + (off), \
2314 MLX5_BYTE_OFF(fte_match_set_lyr_2_4, match_field)}
2316 /* masked values are the same and there are no rewrites that do not have a
2319 #define SAME_VAL_MASK(type, valp, maskp, matchvalp, matchmaskp) ({ \
2320 type matchmaskx = *(type *)(matchmaskp); \
2321 type matchvalx = *(type *)(matchvalp); \
2322 type maskx = *(type *)(maskp); \
2323 type valx = *(type *)(valp); \
2325 (valx & maskx) == (matchvalx & matchmaskx) && !(maskx & (maskx ^ \
2329 static bool cmp_val_mask(void *valp, void *maskp, void *matchvalp,
2330 void *matchmaskp, u8 bsize)
2336 same = SAME_VAL_MASK(u8, valp, maskp, matchvalp, matchmaskp);
2339 same = SAME_VAL_MASK(u16, valp, maskp, matchvalp, matchmaskp);
2342 same = SAME_VAL_MASK(u32, valp, maskp, matchvalp, matchmaskp);
2349 static struct mlx5_fields fields[] = {
2350 OFFLOAD(DMAC_47_16, 32, U32_MAX, eth.h_dest[0], 0, dmac_47_16),
2351 OFFLOAD(DMAC_15_0, 16, U16_MAX, eth.h_dest[4], 0, dmac_15_0),
2352 OFFLOAD(SMAC_47_16, 32, U32_MAX, eth.h_source[0], 0, smac_47_16),
2353 OFFLOAD(SMAC_15_0, 16, U16_MAX, eth.h_source[4], 0, smac_15_0),
2354 OFFLOAD(ETHERTYPE, 16, U16_MAX, eth.h_proto, 0, ethertype),
2355 OFFLOAD(FIRST_VID, 16, U16_MAX, vlan.h_vlan_TCI, 0, first_vid),
2357 OFFLOAD(IP_DSCP, 8, 0xfc, ip4.tos, 0, ip_dscp),
2358 OFFLOAD(IP_TTL, 8, U8_MAX, ip4.ttl, 0, ttl_hoplimit),
2359 OFFLOAD(SIPV4, 32, U32_MAX, ip4.saddr, 0, src_ipv4_src_ipv6.ipv4_layout.ipv4),
2360 OFFLOAD(DIPV4, 32, U32_MAX, ip4.daddr, 0, dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2362 OFFLOAD(SIPV6_127_96, 32, U32_MAX, ip6.saddr.s6_addr32[0], 0,
2363 src_ipv4_src_ipv6.ipv6_layout.ipv6[0]),
2364 OFFLOAD(SIPV6_95_64, 32, U32_MAX, ip6.saddr.s6_addr32[1], 0,
2365 src_ipv4_src_ipv6.ipv6_layout.ipv6[4]),
2366 OFFLOAD(SIPV6_63_32, 32, U32_MAX, ip6.saddr.s6_addr32[2], 0,
2367 src_ipv4_src_ipv6.ipv6_layout.ipv6[8]),
2368 OFFLOAD(SIPV6_31_0, 32, U32_MAX, ip6.saddr.s6_addr32[3], 0,
2369 src_ipv4_src_ipv6.ipv6_layout.ipv6[12]),
2370 OFFLOAD(DIPV6_127_96, 32, U32_MAX, ip6.daddr.s6_addr32[0], 0,
2371 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[0]),
2372 OFFLOAD(DIPV6_95_64, 32, U32_MAX, ip6.daddr.s6_addr32[1], 0,
2373 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[4]),
2374 OFFLOAD(DIPV6_63_32, 32, U32_MAX, ip6.daddr.s6_addr32[2], 0,
2375 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[8]),
2376 OFFLOAD(DIPV6_31_0, 32, U32_MAX, ip6.daddr.s6_addr32[3], 0,
2377 dst_ipv4_dst_ipv6.ipv6_layout.ipv6[12]),
2378 OFFLOAD(IPV6_HOPLIMIT, 8, U8_MAX, ip6.hop_limit, 0, ttl_hoplimit),
2380 OFFLOAD(TCP_SPORT, 16, U16_MAX, tcp.source, 0, tcp_sport),
2381 OFFLOAD(TCP_DPORT, 16, U16_MAX, tcp.dest, 0, tcp_dport),
2382 /* in linux iphdr tcp_flags is 8 bits long */
2383 OFFLOAD(TCP_FLAGS, 8, U8_MAX, tcp.ack_seq, 5, tcp_flags),
2385 OFFLOAD(UDP_SPORT, 16, U16_MAX, udp.source, 0, udp_sport),
2386 OFFLOAD(UDP_DPORT, 16, U16_MAX, udp.dest, 0, udp_dport),
2389 /* On input attr->max_mod_hdr_actions tells how many HW actions can be parsed at
2390 * max from the SW pedit action. On success, attr->num_mod_hdr_actions
2391 * says how many HW actions were actually parsed.
2393 static int offload_pedit_fields(struct pedit_headers_action *hdrs,
2394 struct mlx5e_tc_flow_parse_attr *parse_attr,
2396 struct netlink_ext_ack *extack)
2398 struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
2399 int i, action_size, nactions, max_actions, first, last, next_z;
2400 void *headers_c, *headers_v, *action, *vals_p;
2401 u32 *s_masks_p, *a_masks_p, s_mask, a_mask;
2402 struct mlx5_fields *f;
2408 headers_c = get_match_headers_criteria(*action_flags, &parse_attr->spec);
2409 headers_v = get_match_headers_value(*action_flags, &parse_attr->spec);
2411 set_masks = &hdrs[0].masks;
2412 add_masks = &hdrs[1].masks;
2413 set_vals = &hdrs[0].vals;
2414 add_vals = &hdrs[1].vals;
2416 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2417 action = parse_attr->mod_hdr_actions +
2418 parse_attr->num_mod_hdr_actions * action_size;
2420 max_actions = parse_attr->max_mod_hdr_actions;
2421 nactions = parse_attr->num_mod_hdr_actions;
2423 for (i = 0; i < ARRAY_SIZE(fields); i++) {
2427 /* avoid seeing bits set from previous iterations */
2431 s_masks_p = (void *)set_masks + f->offset;
2432 a_masks_p = (void *)add_masks + f->offset;
2434 s_mask = *s_masks_p & f->field_mask;
2435 a_mask = *a_masks_p & f->field_mask;
2437 if (!s_mask && !a_mask) /* nothing to offload here */
2440 if (s_mask && a_mask) {
2441 NL_SET_ERR_MSG_MOD(extack,
2442 "can't set and add to the same HW field");
2443 printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
2447 if (nactions == max_actions) {
2448 NL_SET_ERR_MSG_MOD(extack,
2449 "too many pedit actions, can't offload");
2450 printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
2456 void *match_mask = headers_c + f->match_offset;
2457 void *match_val = headers_v + f->match_offset;
2459 cmd = MLX5_ACTION_TYPE_SET;
2461 vals_p = (void *)set_vals + f->offset;
2462 /* don't rewrite if we have a match on the same value */
2463 if (cmp_val_mask(vals_p, s_masks_p, match_val,
2464 match_mask, f->field_bsize))
2466 /* clear to denote we consumed this field */
2467 *s_masks_p &= ~f->field_mask;
2469 cmd = MLX5_ACTION_TYPE_ADD;
2471 vals_p = (void *)add_vals + f->offset;
2472 /* add 0 is no change */
2473 if ((*(u32 *)vals_p & f->field_mask) == 0)
2475 /* clear to denote we consumed this field */
2476 *a_masks_p &= ~f->field_mask;
2481 if (f->field_bsize == 32) {
2482 mask_be32 = *(__be32 *)&mask;
2483 mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
2484 } else if (f->field_bsize == 16) {
2485 mask_be16 = *(__be16 *)&mask;
2486 mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
2489 first = find_first_bit(&mask, f->field_bsize);
2490 next_z = find_next_zero_bit(&mask, f->field_bsize, first);
2491 last = find_last_bit(&mask, f->field_bsize);
2492 if (first < next_z && next_z < last) {
2493 NL_SET_ERR_MSG_MOD(extack,
2494 "rewrite of few sub-fields isn't supported");
2495 printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
2500 MLX5_SET(set_action_in, action, action_type, cmd);
2501 MLX5_SET(set_action_in, action, field, f->field);
2503 if (cmd == MLX5_ACTION_TYPE_SET) {
2506 /* if field is bit sized it can start not from first bit */
2507 start = find_first_bit((unsigned long *)&f->field_mask,
2510 MLX5_SET(set_action_in, action, offset, first - start);
2511 /* length is num of bits to be written, zero means length of 32 */
2512 MLX5_SET(set_action_in, action, length, (last - first + 1));
2515 if (f->field_bsize == 32)
2516 MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
2517 else if (f->field_bsize == 16)
2518 MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
2519 else if (f->field_bsize == 8)
2520 MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
2522 action += action_size;
2526 parse_attr->num_mod_hdr_actions = nactions;
2530 static int mlx5e_flow_namespace_max_modify_action(struct mlx5_core_dev *mdev,
2533 if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
2534 return MLX5_CAP_ESW_FLOWTABLE_FDB(mdev, max_modify_header_actions);
2535 else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
2536 return MLX5_CAP_FLOWTABLE_NIC_RX(mdev, max_modify_header_actions);
2539 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
2540 struct pedit_headers_action *hdrs,
2542 struct mlx5e_tc_flow_parse_attr *parse_attr)
2544 int nkeys, action_size, max_actions;
2546 nkeys = hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits +
2547 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits;
2548 action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
2550 max_actions = mlx5e_flow_namespace_max_modify_action(priv->mdev, namespace);
2551 /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
2552 max_actions = min(max_actions, nkeys * 16);
2554 parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
2555 if (!parse_attr->mod_hdr_actions)
2558 parse_attr->max_mod_hdr_actions = max_actions;
2562 static const struct pedit_headers zero_masks = {};
2564 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
2565 const struct flow_action_entry *act, int namespace,
2566 struct mlx5e_tc_flow_parse_attr *parse_attr,
2567 struct pedit_headers_action *hdrs,
2568 struct netlink_ext_ack *extack)
2570 u8 cmd = (act->id == FLOW_ACTION_MANGLE) ? 0 : 1;
2571 int err = -EOPNOTSUPP;
2572 u32 mask, val, offset;
2575 htype = act->mangle.htype;
2576 err = -EOPNOTSUPP; /* can't be all optimistic */
2578 if (htype == FLOW_ACT_MANGLE_UNSPEC) {
2579 NL_SET_ERR_MSG_MOD(extack, "legacy pedit isn't offloaded");
2583 if (!mlx5e_flow_namespace_max_modify_action(priv->mdev, namespace)) {
2584 NL_SET_ERR_MSG_MOD(extack,
2585 "The pedit offload action is not supported");
2589 mask = act->mangle.mask;
2590 val = act->mangle.val;
2591 offset = act->mangle.offset;
2593 err = set_pedit_val(htype, ~mask, val, offset, &hdrs[cmd]);
2604 static int alloc_tc_pedit_action(struct mlx5e_priv *priv, int namespace,
2605 struct mlx5e_tc_flow_parse_attr *parse_attr,
2606 struct pedit_headers_action *hdrs,
2608 struct netlink_ext_ack *extack)
2610 struct pedit_headers *cmd_masks;
2614 if (!parse_attr->mod_hdr_actions) {
2615 err = alloc_mod_hdr_actions(priv, hdrs, namespace, parse_attr);
2620 err = offload_pedit_fields(hdrs, parse_attr, action_flags, extack);
2622 goto out_dealloc_parsed_actions;
2624 for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
2625 cmd_masks = &hdrs[cmd].masks;
2626 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
2627 NL_SET_ERR_MSG_MOD(extack,
2628 "attempt to offload an unsupported field");
2629 netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
2630 print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
2631 16, 1, cmd_masks, sizeof(zero_masks), true);
2633 goto out_dealloc_parsed_actions;
2639 out_dealloc_parsed_actions:
2640 kfree(parse_attr->mod_hdr_actions);
2645 static bool csum_offload_supported(struct mlx5e_priv *priv,
2648 struct netlink_ext_ack *extack)
2650 u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
2651 TCA_CSUM_UPDATE_FLAG_UDP;
2653 /* The HW recalcs checksums only if re-writing headers */
2654 if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
2655 NL_SET_ERR_MSG_MOD(extack,
2656 "TC csum action is only offloaded with pedit");
2657 netdev_warn(priv->netdev,
2658 "TC csum action is only offloaded with pedit\n");
2662 if (update_flags & ~prot_flags) {
2663 NL_SET_ERR_MSG_MOD(extack,
2664 "can't offload TC csum action for some header/s");
2665 netdev_warn(priv->netdev,
2666 "can't offload TC csum action for some header/s - flags %#x\n",
2674 struct ip_ttl_word {
2680 struct ipv6_hoplimit_word {
2686 static bool is_action_keys_supported(const struct flow_action_entry *act)
2691 htype = act->mangle.htype;
2692 offset = act->mangle.offset;
2693 mask = ~act->mangle.mask;
2694 /* For IPv4 & IPv6 header check 4 byte word,
2695 * to determine that modified fields
2696 * are NOT ttl & hop_limit only.
2698 if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP4) {
2699 struct ip_ttl_word *ttl_word =
2700 (struct ip_ttl_word *)&mask;
2702 if (offset != offsetof(struct iphdr, ttl) ||
2703 ttl_word->protocol ||
2707 } else if (htype == FLOW_ACT_MANGLE_HDR_TYPE_IP6) {
2708 struct ipv6_hoplimit_word *hoplimit_word =
2709 (struct ipv6_hoplimit_word *)&mask;
2711 if (offset != offsetof(struct ipv6hdr, payload_len) ||
2712 hoplimit_word->payload_len ||
2713 hoplimit_word->nexthdr) {
2720 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
2721 struct flow_action *flow_action,
2723 struct netlink_ext_ack *extack)
2725 const struct flow_action_entry *act;
2726 bool modify_ip_header;
2732 headers_v = get_match_headers_value(actions, spec);
2733 ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
2735 /* for non-IP we only re-write MACs, so we're okay */
2736 if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
2739 modify_ip_header = false;
2740 flow_action_for_each(i, act, flow_action) {
2741 if (act->id != FLOW_ACTION_MANGLE &&
2742 act->id != FLOW_ACTION_ADD)
2745 if (is_action_keys_supported(act)) {
2746 modify_ip_header = true;
2751 ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
2752 if (modify_ip_header && ip_proto != IPPROTO_TCP &&
2753 ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
2754 NL_SET_ERR_MSG_MOD(extack,
2755 "can't offload re-write of non TCP/UDP");
2756 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
2764 static bool actions_match_supported(struct mlx5e_priv *priv,
2765 struct flow_action *flow_action,
2766 struct mlx5e_tc_flow_parse_attr *parse_attr,
2767 struct mlx5e_tc_flow *flow,
2768 struct netlink_ext_ack *extack)
2772 if (mlx5e_is_eswitch_flow(flow))
2773 actions = flow->esw_attr->action;
2775 actions = flow->nic_attr->action;
2777 if (flow_flag_test(flow, EGRESS) &&
2778 !((actions & MLX5_FLOW_CONTEXT_ACTION_DECAP) ||
2779 (actions & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) ||
2780 (actions & MLX5_FLOW_CONTEXT_ACTION_DROP)))
2783 if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
2784 return modify_header_match_supported(&parse_attr->spec,
2785 flow_action, actions,
2791 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
2793 struct mlx5_core_dev *fmdev, *pmdev;
2794 u64 fsystem_guid, psystem_guid;
2797 pmdev = peer_priv->mdev;
2799 fsystem_guid = mlx5_query_nic_system_image_guid(fmdev);
2800 psystem_guid = mlx5_query_nic_system_image_guid(pmdev);
2802 return (fsystem_guid == psystem_guid);
2805 static int add_vlan_rewrite_action(struct mlx5e_priv *priv, int namespace,
2806 const struct flow_action_entry *act,
2807 struct mlx5e_tc_flow_parse_attr *parse_attr,
2808 struct pedit_headers_action *hdrs,
2809 u32 *action, struct netlink_ext_ack *extack)
2811 u16 mask16 = VLAN_VID_MASK;
2812 u16 val16 = act->vlan.vid & VLAN_VID_MASK;
2813 const struct flow_action_entry pedit_act = {
2814 .id = FLOW_ACTION_MANGLE,
2815 .mangle.htype = FLOW_ACT_MANGLE_HDR_TYPE_ETH,
2816 .mangle.offset = offsetof(struct vlan_ethhdr, h_vlan_TCI),
2817 .mangle.mask = ~(u32)be16_to_cpu(*(__be16 *)&mask16),
2818 .mangle.val = (u32)be16_to_cpu(*(__be16 *)&val16),
2820 u8 match_prio_mask, match_prio_val;
2821 void *headers_c, *headers_v;
2824 headers_c = get_match_headers_criteria(*action, &parse_attr->spec);
2825 headers_v = get_match_headers_value(*action, &parse_attr->spec);
2827 if (!(MLX5_GET(fte_match_set_lyr_2_4, headers_c, cvlan_tag) &&
2828 MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag))) {
2829 NL_SET_ERR_MSG_MOD(extack,
2830 "VLAN rewrite action must have VLAN protocol match");
2834 match_prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
2835 match_prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
2836 if (act->vlan.prio != (match_prio_val & match_prio_mask)) {
2837 NL_SET_ERR_MSG_MOD(extack,
2838 "Changing VLAN prio is not supported");
2842 err = parse_tc_pedit_action(priv, &pedit_act, namespace, parse_attr,
2844 *action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2850 add_vlan_prio_tag_rewrite_action(struct mlx5e_priv *priv,
2851 struct mlx5e_tc_flow_parse_attr *parse_attr,
2852 struct pedit_headers_action *hdrs,
2853 u32 *action, struct netlink_ext_ack *extack)
2855 const struct flow_action_entry prio_tag_act = {
2858 MLX5_GET(fte_match_set_lyr_2_4,
2859 get_match_headers_value(*action,
2862 MLX5_GET(fte_match_set_lyr_2_4,
2863 get_match_headers_criteria(*action,
2868 return add_vlan_rewrite_action(priv, MLX5_FLOW_NAMESPACE_FDB,
2869 &prio_tag_act, parse_attr, hdrs, action,
2873 static int parse_tc_nic_actions(struct mlx5e_priv *priv,
2874 struct flow_action *flow_action,
2875 struct mlx5e_tc_flow_parse_attr *parse_attr,
2876 struct mlx5e_tc_flow *flow,
2877 struct netlink_ext_ack *extack)
2879 struct mlx5_nic_flow_attr *attr = flow->nic_attr;
2880 struct pedit_headers_action hdrs[2] = {};
2881 const struct flow_action_entry *act;
2885 if (!flow_action_has_entries(flow_action))
2888 attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
2890 flow_action_for_each(i, act, flow_action) {
2892 case FLOW_ACTION_ACCEPT:
2893 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2894 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2896 case FLOW_ACTION_DROP:
2897 action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
2898 if (MLX5_CAP_FLOWTABLE(priv->mdev,
2899 flow_table_properties_nic_receive.flow_counter))
2900 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
2902 case FLOW_ACTION_MANGLE:
2903 case FLOW_ACTION_ADD:
2904 err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_KERNEL,
2905 parse_attr, hdrs, extack);
2909 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
2910 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2912 case FLOW_ACTION_VLAN_MANGLE:
2913 err = add_vlan_rewrite_action(priv,
2914 MLX5_FLOW_NAMESPACE_KERNEL,
2915 act, parse_attr, hdrs,
2921 case FLOW_ACTION_CSUM:
2922 if (csum_offload_supported(priv, action,
2928 case FLOW_ACTION_REDIRECT: {
2929 struct net_device *peer_dev = act->dev;
2931 if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
2932 same_hw_devs(priv, netdev_priv(peer_dev))) {
2933 parse_attr->mirred_ifindex[0] = peer_dev->ifindex;
2934 flow_flag_set(flow, HAIRPIN);
2935 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2936 MLX5_FLOW_CONTEXT_ACTION_COUNT;
2938 NL_SET_ERR_MSG_MOD(extack,
2939 "device is not on same HW, can't offload");
2940 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
2946 case FLOW_ACTION_MARK: {
2947 u32 mark = act->mark;
2949 if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
2950 NL_SET_ERR_MSG_MOD(extack,
2951 "Bad flow mark - only 16 bit is supported");
2955 attr->flow_tag = mark;
2956 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2960 NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
2965 if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
2966 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
2967 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_KERNEL,
2968 parse_attr, hdrs, &action, extack);
2971 /* in case all pedit actions are skipped, remove the MOD_HDR
2974 if (parse_attr->num_mod_hdr_actions == 0) {
2975 action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2976 kfree(parse_attr->mod_hdr_actions);
2980 attr->action = action;
2981 if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
2988 const struct ip_tunnel_key *ip_tun_key;
2989 struct mlx5e_tc_tunnel *tc_tunnel;
2992 static inline int cmp_encap_info(struct encap_key *a,
2993 struct encap_key *b)
2995 return memcmp(a->ip_tun_key, b->ip_tun_key, sizeof(*a->ip_tun_key)) ||
2996 a->tc_tunnel->tunnel_type != b->tc_tunnel->tunnel_type;
2999 static inline int hash_encap_info(struct encap_key *key)
3001 return jhash(key->ip_tun_key, sizeof(*key->ip_tun_key),
3002 key->tc_tunnel->tunnel_type);
3006 static bool is_merged_eswitch_dev(struct mlx5e_priv *priv,
3007 struct net_device *peer_netdev)
3009 struct mlx5e_priv *peer_priv;
3011 peer_priv = netdev_priv(peer_netdev);
3013 return (MLX5_CAP_ESW(priv->mdev, merged_eswitch) &&
3014 mlx5e_eswitch_rep(priv->netdev) &&
3015 mlx5e_eswitch_rep(peer_netdev) &&
3016 same_hw_devs(priv, peer_priv));
3021 bool mlx5e_encap_take(struct mlx5e_encap_entry *e)
3023 return refcount_inc_not_zero(&e->refcnt);
3026 static struct mlx5e_encap_entry *
3027 mlx5e_encap_get(struct mlx5e_priv *priv, struct encap_key *key,
3030 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3031 struct mlx5e_encap_entry *e;
3032 struct encap_key e_key;
3034 hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
3035 encap_hlist, hash_key) {
3036 e_key.ip_tun_key = &e->tun_info->key;
3037 e_key.tc_tunnel = e->tunnel;
3038 if (!cmp_encap_info(&e_key, key) &&
3039 mlx5e_encap_take(e))
3046 static struct ip_tunnel_info *dup_tun_info(const struct ip_tunnel_info *tun_info)
3048 size_t tun_size = sizeof(*tun_info) + tun_info->options_len;
3050 return kmemdup(tun_info, tun_size, GFP_KERNEL);
3053 static bool is_duplicated_encap_entry(struct mlx5e_priv *priv,
3054 struct mlx5e_tc_flow *flow,
3056 struct mlx5e_encap_entry *e,
3057 struct netlink_ext_ack *extack)
3061 for (i = 0; i < out_index; i++) {
3062 if (flow->encaps[i].e != e)
3064 NL_SET_ERR_MSG_MOD(extack, "can't duplicate encap action");
3065 netdev_err(priv->netdev, "can't duplicate encap action\n");
3072 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
3073 struct mlx5e_tc_flow *flow,
3074 struct net_device *mirred_dev,
3076 struct netlink_ext_ack *extack,
3077 struct net_device **encap_dev,
3080 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3081 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3082 struct mlx5e_tc_flow_parse_attr *parse_attr;
3083 const struct ip_tunnel_info *tun_info;
3084 struct encap_key key;
3085 struct mlx5e_encap_entry *e;
3086 unsigned short family;
3090 parse_attr = attr->parse_attr;
3091 tun_info = parse_attr->tun_info[out_index];
3092 family = ip_tunnel_info_af(tun_info);
3093 key.ip_tun_key = &tun_info->key;
3094 key.tc_tunnel = mlx5e_get_tc_tun(mirred_dev);
3095 if (!key.tc_tunnel) {
3096 NL_SET_ERR_MSG_MOD(extack, "Unsupported tunnel");
3100 hash_key = hash_encap_info(&key);
3102 mutex_lock(&esw->offloads.encap_tbl_lock);
3103 e = mlx5e_encap_get(priv, &key, hash_key);
3105 /* must verify if encap is valid or not */
3107 /* Check that entry was not already attached to this flow */
3108 if (is_duplicated_encap_entry(priv, flow, out_index, e, extack)) {
3113 mutex_unlock(&esw->offloads.encap_tbl_lock);
3114 wait_for_completion(&e->res_ready);
3116 /* Protect against concurrent neigh update. */
3117 mutex_lock(&esw->offloads.encap_tbl_lock);
3118 if (e->compl_result < 0) {
3125 e = kzalloc(sizeof(*e), GFP_KERNEL);
3131 refcount_set(&e->refcnt, 1);
3132 init_completion(&e->res_ready);
3134 tun_info = dup_tun_info(tun_info);
3139 e->tun_info = tun_info;
3140 err = mlx5e_tc_tun_init_encap_attr(mirred_dev, priv, e, extack);
3144 INIT_LIST_HEAD(&e->flows);
3145 hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
3146 mutex_unlock(&esw->offloads.encap_tbl_lock);
3148 if (family == AF_INET)
3149 err = mlx5e_tc_tun_create_header_ipv4(priv, mirred_dev, e);
3150 else if (family == AF_INET6)
3151 err = mlx5e_tc_tun_create_header_ipv6(priv, mirred_dev, e);
3153 /* Protect against concurrent neigh update. */
3154 mutex_lock(&esw->offloads.encap_tbl_lock);
3155 complete_all(&e->res_ready);
3157 e->compl_result = err;
3160 e->compl_result = 1;
3163 flow->encaps[out_index].e = e;
3164 list_add(&flow->encaps[out_index].list, &e->flows);
3165 flow->encaps[out_index].index = out_index;
3166 *encap_dev = e->out_dev;
3167 if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
3168 attr->dests[out_index].pkt_reformat = e->pkt_reformat;
3169 attr->dests[out_index].flags |= MLX5_ESW_DEST_ENCAP_VALID;
3170 *encap_valid = true;
3172 *encap_valid = false;
3174 mutex_unlock(&esw->offloads.encap_tbl_lock);
3179 mutex_unlock(&esw->offloads.encap_tbl_lock);
3181 mlx5e_encap_put(priv, e);
3185 mutex_unlock(&esw->offloads.encap_tbl_lock);
3191 static int parse_tc_vlan_action(struct mlx5e_priv *priv,
3192 const struct flow_action_entry *act,
3193 struct mlx5_esw_flow_attr *attr,
3196 u8 vlan_idx = attr->total_vlan;
3198 if (vlan_idx >= MLX5_FS_VLAN_DEPTH)
3202 case FLOW_ACTION_VLAN_POP:
3204 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
3205 MLX5_FS_VLAN_DEPTH))
3208 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2;
3210 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3213 case FLOW_ACTION_VLAN_PUSH:
3214 attr->vlan_vid[vlan_idx] = act->vlan.vid;
3215 attr->vlan_prio[vlan_idx] = act->vlan.prio;
3216 attr->vlan_proto[vlan_idx] = act->vlan.proto;
3217 if (!attr->vlan_proto[vlan_idx])
3218 attr->vlan_proto[vlan_idx] = htons(ETH_P_8021Q);
3221 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev,
3222 MLX5_FS_VLAN_DEPTH))
3225 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2;
3227 if (!mlx5_eswitch_vlan_actions_supported(priv->mdev, 1) &&
3228 (act->vlan.proto != htons(ETH_P_8021Q) ||
3232 *action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
3239 attr->total_vlan = vlan_idx + 1;
3244 static int add_vlan_push_action(struct mlx5e_priv *priv,
3245 struct mlx5_esw_flow_attr *attr,
3246 struct net_device **out_dev,
3249 struct net_device *vlan_dev = *out_dev;
3250 struct flow_action_entry vlan_act = {
3251 .id = FLOW_ACTION_VLAN_PUSH,
3252 .vlan.vid = vlan_dev_vlan_id(vlan_dev),
3253 .vlan.proto = vlan_dev_vlan_proto(vlan_dev),
3258 err = parse_tc_vlan_action(priv, &vlan_act, attr, action);
3262 *out_dev = dev_get_by_index_rcu(dev_net(vlan_dev),
3263 dev_get_iflink(vlan_dev));
3264 if (is_vlan_dev(*out_dev))
3265 err = add_vlan_push_action(priv, attr, out_dev, action);
3270 static int add_vlan_pop_action(struct mlx5e_priv *priv,
3271 struct mlx5_esw_flow_attr *attr,
3274 int nest_level = attr->parse_attr->filter_dev->lower_level;
3275 struct flow_action_entry vlan_act = {
3276 .id = FLOW_ACTION_VLAN_POP,
3280 while (nest_level--) {
3281 err = parse_tc_vlan_action(priv, &vlan_act, attr, action);
3289 bool mlx5e_is_valid_eswitch_fwd_dev(struct mlx5e_priv *priv,
3290 struct net_device *out_dev)
3292 if (is_merged_eswitch_dev(priv, out_dev))
3295 return mlx5e_eswitch_rep(out_dev) &&
3296 same_hw_devs(priv, netdev_priv(out_dev));
3299 static bool is_duplicated_output_device(struct net_device *dev,
3300 struct net_device *out_dev,
3301 int *ifindexes, int if_count,
3302 struct netlink_ext_ack *extack)
3306 for (i = 0; i < if_count; i++) {
3307 if (ifindexes[i] == out_dev->ifindex) {
3308 NL_SET_ERR_MSG_MOD(extack,
3309 "can't duplicate output to same device");
3310 netdev_err(dev, "can't duplicate output to same device: %s\n",
3319 static int parse_tc_fdb_actions(struct mlx5e_priv *priv,
3320 struct flow_action *flow_action,
3321 struct mlx5e_tc_flow *flow,
3322 struct netlink_ext_ack *extack)
3324 struct pedit_headers_action hdrs[2] = {};
3325 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3326 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3327 struct mlx5e_tc_flow_parse_attr *parse_attr = attr->parse_attr;
3328 struct mlx5e_rep_priv *rpriv = priv->ppriv;
3329 const struct ip_tunnel_info *info = NULL;
3330 int ifindexes[MLX5_MAX_FLOW_FWD_VPORTS];
3331 bool ft_flow = mlx5e_is_ft_flow(flow);
3332 const struct flow_action_entry *act;
3333 int err, i, if_count = 0;
3337 if (!flow_action_has_entries(flow_action))
3340 flow_action_for_each(i, act, flow_action) {
3342 case FLOW_ACTION_DROP:
3343 action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
3344 MLX5_FLOW_CONTEXT_ACTION_COUNT;
3346 case FLOW_ACTION_MANGLE:
3347 case FLOW_ACTION_ADD:
3348 err = parse_tc_pedit_action(priv, act, MLX5_FLOW_NAMESPACE_FDB,
3349 parse_attr, hdrs, extack);
3353 action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3354 attr->split_count = attr->out_count;
3356 case FLOW_ACTION_CSUM:
3357 if (csum_offload_supported(priv, action,
3358 act->csum_flags, extack))
3362 case FLOW_ACTION_REDIRECT:
3363 case FLOW_ACTION_MIRRED: {
3364 struct mlx5e_priv *out_priv;
3365 struct net_device *out_dev;
3369 /* out_dev is NULL when filters with
3370 * non-existing mirred device are replayed to
3376 if (ft_flow && out_dev == priv->netdev) {
3377 /* Ignore forward to self rules generated
3378 * by adding both mlx5 devs to the flow table
3379 * block on a normal nft offload setup.
3384 if (attr->out_count >= MLX5_MAX_FLOW_FWD_VPORTS) {
3385 NL_SET_ERR_MSG_MOD(extack,
3386 "can't support more output ports, can't offload forwarding");
3387 netdev_warn(priv->netdev,
3388 "can't support more than %d output ports, can't offload forwarding\n",
3393 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
3394 MLX5_FLOW_CONTEXT_ACTION_COUNT;
3396 parse_attr->mirred_ifindex[attr->out_count] =
3398 parse_attr->tun_info[attr->out_count] = dup_tun_info(info);
3399 if (!parse_attr->tun_info[attr->out_count])
3402 attr->dests[attr->out_count].flags |=
3403 MLX5_ESW_DEST_ENCAP;
3405 /* attr->dests[].rep is resolved when we
3408 } else if (netdev_port_same_parent_id(priv->netdev, out_dev)) {
3409 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3410 struct net_device *uplink_dev = mlx5_eswitch_uplink_get_proto_dev(esw, REP_ETH);
3411 struct net_device *uplink_upper;
3412 struct mlx5e_rep_priv *rep_priv;
3414 if (is_duplicated_output_device(priv->netdev,
3421 ifindexes[if_count] = out_dev->ifindex;
3426 netdev_master_upper_dev_get_rcu(uplink_dev);
3428 netif_is_lag_master(uplink_upper) &&
3429 uplink_upper == out_dev)
3430 out_dev = uplink_dev;
3433 if (is_vlan_dev(out_dev)) {
3434 err = add_vlan_push_action(priv, attr,
3441 if (is_vlan_dev(parse_attr->filter_dev)) {
3442 err = add_vlan_pop_action(priv, attr,
3448 /* Don't allow forwarding between uplink.
3450 * Input vport was stored esw_attr->in_rep.
3451 * In LAG case, *priv* is the private data of
3452 * uplink which may be not the input vport.
3454 rep_priv = mlx5e_rep_to_rep_priv(attr->in_rep);
3455 if (mlx5e_eswitch_uplink_rep(rep_priv->netdev) &&
3456 mlx5e_eswitch_uplink_rep(out_dev)) {
3457 NL_SET_ERR_MSG_MOD(extack,
3458 "devices are both uplink, can't offload forwarding");
3459 pr_err("devices %s %s are both uplink, can't offload forwarding\n",
3460 priv->netdev->name, out_dev->name);
3464 if (!mlx5e_is_valid_eswitch_fwd_dev(priv, out_dev)) {
3465 NL_SET_ERR_MSG_MOD(extack,
3466 "devices are not on same switch HW, can't offload forwarding");
3467 netdev_warn(priv->netdev,
3468 "devices %s %s not on same switch HW, can't offload forwarding\n",
3474 out_priv = netdev_priv(out_dev);
3475 rpriv = out_priv->ppriv;
3476 attr->dests[attr->out_count].rep = rpriv->rep;
3477 attr->dests[attr->out_count].mdev = out_priv->mdev;
3479 } else if (parse_attr->filter_dev != priv->netdev) {
3480 /* All mlx5 devices are called to configure
3481 * high level device filters. Therefore, the
3482 * *attempt* to install a filter on invalid
3483 * eswitch should not trigger an explicit error
3487 NL_SET_ERR_MSG_MOD(extack,
3488 "devices are not on same switch HW, can't offload forwarding");
3489 netdev_warn(priv->netdev,
3490 "devices %s %s not on same switch HW, can't offload forwarding\n",
3497 case FLOW_ACTION_TUNNEL_ENCAP:
3505 case FLOW_ACTION_VLAN_PUSH:
3506 case FLOW_ACTION_VLAN_POP:
3507 if (act->id == FLOW_ACTION_VLAN_PUSH &&
3508 (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP)) {
3509 /* Replace vlan pop+push with vlan modify */
3510 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3511 err = add_vlan_rewrite_action(priv,
3512 MLX5_FLOW_NAMESPACE_FDB,
3513 act, parse_attr, hdrs,
3516 err = parse_tc_vlan_action(priv, act, attr, &action);
3521 attr->split_count = attr->out_count;
3523 case FLOW_ACTION_VLAN_MANGLE:
3524 err = add_vlan_rewrite_action(priv,
3525 MLX5_FLOW_NAMESPACE_FDB,
3526 act, parse_attr, hdrs,
3531 attr->split_count = attr->out_count;
3533 case FLOW_ACTION_TUNNEL_DECAP:
3534 action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
3536 case FLOW_ACTION_GOTO: {
3537 u32 dest_chain = act->chain_index;
3538 u32 max_chain = mlx5_esw_chains_get_chain_range(esw);
3541 NL_SET_ERR_MSG_MOD(extack, "Goto action is not supported");
3544 if (!mlx5_esw_chains_backwards_supported(esw) &&
3545 dest_chain <= attr->chain) {
3546 NL_SET_ERR_MSG_MOD(extack,
3547 "Goto earlier chain isn't supported");
3550 if (dest_chain > max_chain) {
3551 NL_SET_ERR_MSG_MOD(extack,
3552 "Requested destination chain is out of supported range");
3555 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
3556 attr->dest_chain = dest_chain;
3560 NL_SET_ERR_MSG_MOD(extack, "The offload action is not supported");
3565 if (MLX5_CAP_GEN(esw->dev, prio_tag_required) &&
3566 action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) {
3567 /* For prio tag mode, replace vlan pop with rewrite vlan prio
3570 action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
3571 err = add_vlan_prio_tag_rewrite_action(priv, parse_attr, hdrs,
3577 if (hdrs[TCA_PEDIT_KEY_EX_CMD_SET].pedits ||
3578 hdrs[TCA_PEDIT_KEY_EX_CMD_ADD].pedits) {
3579 err = alloc_tc_pedit_action(priv, MLX5_FLOW_NAMESPACE_FDB,
3580 parse_attr, hdrs, &action, extack);
3583 /* in case all pedit actions are skipped, remove the MOD_HDR
3584 * flag. we might have set split_count either by pedit or
3585 * pop/push. if there is no pop/push either, reset it too.
3587 if (parse_attr->num_mod_hdr_actions == 0) {
3588 action &= ~MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
3589 kfree(parse_attr->mod_hdr_actions);
3590 if (!((action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP) ||
3591 (action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH)))
3592 attr->split_count = 0;
3596 attr->action = action;
3597 if (!actions_match_supported(priv, flow_action, parse_attr, flow, extack))
3600 if (attr->dest_chain) {
3601 if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
3602 NL_SET_ERR_MSG_MOD(extack,
3603 "Mirroring goto chain rules isn't supported");
3606 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3609 if (!(attr->action &
3610 (MLX5_FLOW_CONTEXT_ACTION_FWD_DEST | MLX5_FLOW_CONTEXT_ACTION_DROP))) {
3611 NL_SET_ERR_MSG_MOD(extack,
3612 "Rule must have at least one forward/drop action");
3616 if (attr->split_count > 0 && !mlx5_esw_has_fwd_fdb(priv->mdev)) {
3617 NL_SET_ERR_MSG_MOD(extack,
3618 "current firmware doesn't support split rule for port mirroring");
3619 netdev_warn_once(priv->netdev, "current firmware doesn't support split rule for port mirroring\n");
3626 static void get_flags(int flags, unsigned long *flow_flags)
3628 unsigned long __flow_flags = 0;
3630 if (flags & MLX5_TC_FLAG(INGRESS))
3631 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_INGRESS);
3632 if (flags & MLX5_TC_FLAG(EGRESS))
3633 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_EGRESS);
3635 if (flags & MLX5_TC_FLAG(ESW_OFFLOAD))
3636 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_ESWITCH);
3637 if (flags & MLX5_TC_FLAG(NIC_OFFLOAD))
3638 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_NIC);
3639 if (flags & MLX5_TC_FLAG(FT_OFFLOAD))
3640 __flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_FT);
3642 *flow_flags = __flow_flags;
3645 static const struct rhashtable_params tc_ht_params = {
3646 .head_offset = offsetof(struct mlx5e_tc_flow, node),
3647 .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
3648 .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
3649 .automatic_shrinking = true,
3652 static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv,
3653 unsigned long flags)
3655 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3656 struct mlx5e_rep_priv *uplink_rpriv;
3658 if (flags & MLX5_TC_FLAG(ESW_OFFLOAD)) {
3659 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
3660 return &uplink_rpriv->uplink_priv.tc_ht;
3661 } else /* NIC offload */
3662 return &priv->fs.tc.ht;
3665 static bool is_peer_flow_needed(struct mlx5e_tc_flow *flow)
3667 struct mlx5_esw_flow_attr *attr = flow->esw_attr;
3668 bool is_rep_ingress = attr->in_rep->vport != MLX5_VPORT_UPLINK &&
3669 flow_flag_test(flow, INGRESS);
3670 bool act_is_encap = !!(attr->action &
3671 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT);
3672 bool esw_paired = mlx5_devcom_is_paired(attr->in_mdev->priv.devcom,
3673 MLX5_DEVCOM_ESW_OFFLOADS);
3678 if ((mlx5_lag_is_sriov(attr->in_mdev) ||
3679 mlx5_lag_is_multipath(attr->in_mdev)) &&
3680 (is_rep_ingress || act_is_encap))
3687 mlx5e_alloc_flow(struct mlx5e_priv *priv, int attr_size,
3688 struct flow_cls_offload *f, unsigned long flow_flags,
3689 struct mlx5e_tc_flow_parse_attr **__parse_attr,
3690 struct mlx5e_tc_flow **__flow)
3692 struct mlx5e_tc_flow_parse_attr *parse_attr;
3693 struct mlx5e_tc_flow *flow;
3696 flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
3697 parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
3698 if (!parse_attr || !flow) {
3703 flow->cookie = f->cookie;
3704 flow->flags = flow_flags;
3706 for (out_index = 0; out_index < MLX5_MAX_FLOW_FWD_VPORTS; out_index++)
3707 INIT_LIST_HEAD(&flow->encaps[out_index].list);
3708 INIT_LIST_HEAD(&flow->mod_hdr);
3709 INIT_LIST_HEAD(&flow->hairpin);
3710 refcount_set(&flow->refcnt, 1);
3711 init_completion(&flow->init_done);
3714 *__parse_attr = parse_attr;
3725 mlx5e_flow_esw_attr_init(struct mlx5_esw_flow_attr *esw_attr,
3726 struct mlx5e_priv *priv,
3727 struct mlx5e_tc_flow_parse_attr *parse_attr,
3728 struct flow_cls_offload *f,
3729 struct mlx5_eswitch_rep *in_rep,
3730 struct mlx5_core_dev *in_mdev)
3732 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3734 esw_attr->parse_attr = parse_attr;
3735 esw_attr->chain = f->common.chain_index;
3736 esw_attr->prio = f->common.prio;
3738 esw_attr->in_rep = in_rep;
3739 esw_attr->in_mdev = in_mdev;
3741 if (MLX5_CAP_ESW(esw->dev, counter_eswitch_affinity) ==
3742 MLX5_COUNTER_SOURCE_ESWITCH)
3743 esw_attr->counter_dev = in_mdev;
3745 esw_attr->counter_dev = priv->mdev;
3748 static struct mlx5e_tc_flow *
3749 __mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
3750 struct flow_cls_offload *f,
3751 unsigned long flow_flags,
3752 struct net_device *filter_dev,
3753 struct mlx5_eswitch_rep *in_rep,
3754 struct mlx5_core_dev *in_mdev)
3756 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
3757 struct netlink_ext_ack *extack = f->common.extack;
3758 struct mlx5e_tc_flow_parse_attr *parse_attr;
3759 struct mlx5e_tc_flow *flow;
3762 flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_ESWITCH);
3763 attr_size = sizeof(struct mlx5_esw_flow_attr);
3764 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3765 &parse_attr, &flow);
3769 parse_attr->filter_dev = filter_dev;
3770 mlx5e_flow_esw_attr_init(flow->esw_attr,
3772 f, in_rep, in_mdev);
3774 err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
3779 err = parse_tc_fdb_actions(priv, &rule->action, flow, extack);
3783 err = mlx5e_tc_add_fdb_flow(priv, flow, extack);
3784 complete_all(&flow->init_done);
3786 if (!(err == -ENETUNREACH && mlx5_lag_is_multipath(in_mdev)))
3789 add_unready_flow(flow);
3795 mlx5e_flow_put(priv, flow);
3797 return ERR_PTR(err);
3800 static int mlx5e_tc_add_fdb_peer_flow(struct flow_cls_offload *f,
3801 struct mlx5e_tc_flow *flow,
3802 unsigned long flow_flags)
3804 struct mlx5e_priv *priv = flow->priv, *peer_priv;
3805 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch, *peer_esw;
3806 struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
3807 struct mlx5e_tc_flow_parse_attr *parse_attr;
3808 struct mlx5e_rep_priv *peer_urpriv;
3809 struct mlx5e_tc_flow *peer_flow;
3810 struct mlx5_core_dev *in_mdev;
3813 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
3817 peer_urpriv = mlx5_eswitch_get_uplink_priv(peer_esw, REP_ETH);
3818 peer_priv = netdev_priv(peer_urpriv->netdev);
3820 /* in_mdev is assigned of which the packet originated from.
3821 * So packets redirected to uplink use the same mdev of the
3822 * original flow and packets redirected from uplink use the
3825 if (flow->esw_attr->in_rep->vport == MLX5_VPORT_UPLINK)
3826 in_mdev = peer_priv->mdev;
3828 in_mdev = priv->mdev;
3830 parse_attr = flow->esw_attr->parse_attr;
3831 peer_flow = __mlx5e_add_fdb_flow(peer_priv, f, flow_flags,
3832 parse_attr->filter_dev,
3833 flow->esw_attr->in_rep, in_mdev);
3834 if (IS_ERR(peer_flow)) {
3835 err = PTR_ERR(peer_flow);
3839 flow->peer_flow = peer_flow;
3840 flow_flag_set(flow, DUP);
3841 mutex_lock(&esw->offloads.peer_mutex);
3842 list_add_tail(&flow->peer, &esw->offloads.peer_flows);
3843 mutex_unlock(&esw->offloads.peer_mutex);
3846 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
3851 mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
3852 struct flow_cls_offload *f,
3853 unsigned long flow_flags,
3854 struct net_device *filter_dev,
3855 struct mlx5e_tc_flow **__flow)
3857 struct mlx5e_rep_priv *rpriv = priv->ppriv;
3858 struct mlx5_eswitch_rep *in_rep = rpriv->rep;
3859 struct mlx5_core_dev *in_mdev = priv->mdev;
3860 struct mlx5e_tc_flow *flow;
3863 flow = __mlx5e_add_fdb_flow(priv, f, flow_flags, filter_dev, in_rep,
3866 return PTR_ERR(flow);
3868 if (is_peer_flow_needed(flow)) {
3869 err = mlx5e_tc_add_fdb_peer_flow(f, flow, flow_flags);
3871 mlx5e_tc_del_fdb_flow(priv, flow);
3885 mlx5e_add_nic_flow(struct mlx5e_priv *priv,
3886 struct flow_cls_offload *f,
3887 unsigned long flow_flags,
3888 struct net_device *filter_dev,
3889 struct mlx5e_tc_flow **__flow)
3891 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
3892 struct netlink_ext_ack *extack = f->common.extack;
3893 struct mlx5e_tc_flow_parse_attr *parse_attr;
3894 struct mlx5e_tc_flow *flow;
3897 /* multi-chain not supported for NIC rules */
3898 if (!tc_cls_can_offload_and_chain0(priv->netdev, &f->common))
3901 flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_NIC);
3902 attr_size = sizeof(struct mlx5_nic_flow_attr);
3903 err = mlx5e_alloc_flow(priv, attr_size, f, flow_flags,
3904 &parse_attr, &flow);
3908 parse_attr->filter_dev = filter_dev;
3909 err = parse_cls_flower(flow->priv, flow, &parse_attr->spec,
3914 err = parse_tc_nic_actions(priv, &rule->action, parse_attr, flow, extack);
3918 err = mlx5e_tc_add_nic_flow(priv, parse_attr, flow, extack);
3922 flow_flag_set(flow, OFFLOADED);
3929 mlx5e_flow_put(priv, flow);
3936 mlx5e_tc_add_flow(struct mlx5e_priv *priv,
3937 struct flow_cls_offload *f,
3938 unsigned long flags,
3939 struct net_device *filter_dev,
3940 struct mlx5e_tc_flow **flow)
3942 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
3943 unsigned long flow_flags;
3946 get_flags(flags, &flow_flags);
3948 if (!tc_can_offload_extack(priv->netdev, f->common.extack))
3951 if (esw && esw->mode == MLX5_ESWITCH_OFFLOADS)
3952 err = mlx5e_add_fdb_flow(priv, f, flow_flags,
3955 err = mlx5e_add_nic_flow(priv, f, flow_flags,
3961 int mlx5e_configure_flower(struct net_device *dev, struct mlx5e_priv *priv,
3962 struct flow_cls_offload *f, unsigned long flags)
3964 struct netlink_ext_ack *extack = f->common.extack;
3965 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
3966 struct mlx5e_tc_flow *flow;
3970 flow = rhashtable_lookup(tc_ht, &f->cookie, tc_ht_params);
3973 NL_SET_ERR_MSG_MOD(extack,
3974 "flow cookie already exists, ignoring");
3975 netdev_warn_once(priv->netdev,
3976 "flow cookie %lx already exists, ignoring\n",
3982 trace_mlx5e_configure_flower(f);
3983 err = mlx5e_tc_add_flow(priv, f, flags, dev, &flow);
3987 err = rhashtable_lookup_insert_fast(tc_ht, &flow->node, tc_ht_params);
3994 mlx5e_flow_put(priv, flow);
3999 static bool same_flow_direction(struct mlx5e_tc_flow *flow, int flags)
4001 bool dir_ingress = !!(flags & MLX5_TC_FLAG(INGRESS));
4002 bool dir_egress = !!(flags & MLX5_TC_FLAG(EGRESS));
4004 return flow_flag_test(flow, INGRESS) == dir_ingress &&
4005 flow_flag_test(flow, EGRESS) == dir_egress;
4008 int mlx5e_delete_flower(struct net_device *dev, struct mlx5e_priv *priv,
4009 struct flow_cls_offload *f, unsigned long flags)
4011 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4012 struct mlx5e_tc_flow *flow;
4016 flow = rhashtable_lookup(tc_ht, &f->cookie, tc_ht_params);
4017 if (!flow || !same_flow_direction(flow, flags)) {
4022 /* Only delete the flow if it doesn't have MLX5E_TC_FLOW_DELETED flag
4025 if (flow_flag_test_and_set(flow, DELETED)) {
4029 rhashtable_remove_fast(tc_ht, &flow->node, tc_ht_params);
4032 trace_mlx5e_delete_flower(f);
4033 mlx5e_flow_put(priv, flow);
4042 int mlx5e_stats_flower(struct net_device *dev, struct mlx5e_priv *priv,
4043 struct flow_cls_offload *f, unsigned long flags)
4045 struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
4046 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4047 struct mlx5_eswitch *peer_esw;
4048 struct mlx5e_tc_flow *flow;
4049 struct mlx5_fc *counter;
4056 flow = mlx5e_flow_get(rhashtable_lookup(tc_ht, &f->cookie,
4060 return PTR_ERR(flow);
4062 if (!same_flow_direction(flow, flags)) {
4067 if (mlx5e_is_offloaded_flow(flow)) {
4068 counter = mlx5e_tc_get_counter(flow);
4072 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
4075 /* Under multipath it's possible for one rule to be currently
4076 * un-offloaded while the other rule is offloaded.
4078 peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4082 if (flow_flag_test(flow, DUP) &&
4083 flow_flag_test(flow->peer_flow, OFFLOADED)) {
4088 counter = mlx5e_tc_get_counter(flow->peer_flow);
4090 goto no_peer_counter;
4091 mlx5_fc_query_cached(counter, &bytes2, &packets2, &lastuse2);
4094 packets += packets2;
4095 lastuse = max_t(u64, lastuse, lastuse2);
4099 mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
4101 flow_stats_update(&f->stats, bytes, packets, lastuse);
4102 trace_mlx5e_stats_flower(f);
4104 mlx5e_flow_put(priv, flow);
4108 static int apply_police_params(struct mlx5e_priv *priv, u32 rate,
4109 struct netlink_ext_ack *extack)
4111 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4112 struct mlx5_eswitch *esw;
4117 vport_num = rpriv->rep->vport;
4118 if (vport_num >= MLX5_VPORT_ECPF) {
4119 NL_SET_ERR_MSG_MOD(extack,
4120 "Ingress rate limit is supported only for Eswitch ports connected to VFs");
4124 esw = priv->mdev->priv.eswitch;
4125 /* rate is given in bytes/sec.
4126 * First convert to bits/sec and then round to the nearest mbit/secs.
4127 * mbit means million bits.
4128 * Moreover, if rate is non zero we choose to configure to a minimum of
4131 rate_mbps = rate ? max_t(u32, (rate * 8 + 500000) / 1000000, 1) : 0;
4132 err = mlx5_esw_modify_vport_rate(esw, vport_num, rate_mbps);
4134 NL_SET_ERR_MSG_MOD(extack, "failed applying action to hardware");
4139 static int scan_tc_matchall_fdb_actions(struct mlx5e_priv *priv,
4140 struct flow_action *flow_action,
4141 struct netlink_ext_ack *extack)
4143 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4144 const struct flow_action_entry *act;
4148 if (!flow_action_has_entries(flow_action)) {
4149 NL_SET_ERR_MSG_MOD(extack, "matchall called with no action");
4153 if (!flow_offload_has_one_action(flow_action)) {
4154 NL_SET_ERR_MSG_MOD(extack, "matchall policing support only a single action");
4158 flow_action_for_each(i, act, flow_action) {
4160 case FLOW_ACTION_POLICE:
4161 err = apply_police_params(priv, act->police.rate_bytes_ps, extack);
4165 rpriv->prev_vf_vport_stats = priv->stats.vf_vport;
4168 NL_SET_ERR_MSG_MOD(extack, "mlx5 supports only police action for matchall");
4176 int mlx5e_tc_configure_matchall(struct mlx5e_priv *priv,
4177 struct tc_cls_matchall_offload *ma)
4179 struct netlink_ext_ack *extack = ma->common.extack;
4181 if (ma->common.prio != 1) {
4182 NL_SET_ERR_MSG_MOD(extack, "only priority 1 is supported");
4186 return scan_tc_matchall_fdb_actions(priv, &ma->rule->action, extack);
4189 int mlx5e_tc_delete_matchall(struct mlx5e_priv *priv,
4190 struct tc_cls_matchall_offload *ma)
4192 struct netlink_ext_ack *extack = ma->common.extack;
4194 return apply_police_params(priv, 0, extack);
4197 void mlx5e_tc_stats_matchall(struct mlx5e_priv *priv,
4198 struct tc_cls_matchall_offload *ma)
4200 struct mlx5e_rep_priv *rpriv = priv->ppriv;
4201 struct rtnl_link_stats64 cur_stats;
4205 cur_stats = priv->stats.vf_vport;
4206 dpkts = cur_stats.rx_packets - rpriv->prev_vf_vport_stats.rx_packets;
4207 dbytes = cur_stats.rx_bytes - rpriv->prev_vf_vport_stats.rx_bytes;
4208 rpriv->prev_vf_vport_stats = cur_stats;
4209 flow_stats_update(&ma->stats, dpkts, dbytes, jiffies);
4212 static void mlx5e_tc_hairpin_update_dead_peer(struct mlx5e_priv *priv,
4213 struct mlx5e_priv *peer_priv)
4215 struct mlx5_core_dev *peer_mdev = peer_priv->mdev;
4216 struct mlx5e_hairpin_entry *hpe, *tmp;
4217 LIST_HEAD(init_wait_list);
4221 if (!same_hw_devs(priv, peer_priv))
4224 peer_vhca_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
4226 mutex_lock(&priv->fs.tc.hairpin_tbl_lock);
4227 hash_for_each(priv->fs.tc.hairpin_tbl, bkt, hpe, hairpin_hlist)
4228 if (refcount_inc_not_zero(&hpe->refcnt))
4229 list_add(&hpe->dead_peer_wait_list, &init_wait_list);
4230 mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
4232 list_for_each_entry_safe(hpe, tmp, &init_wait_list, dead_peer_wait_list) {
4233 wait_for_completion(&hpe->res_ready);
4234 if (!IS_ERR_OR_NULL(hpe->hp) && hpe->peer_vhca_id == peer_vhca_id)
4235 hpe->hp->pair->peer_gone = true;
4237 mlx5e_hairpin_put(priv, hpe);
4241 static int mlx5e_tc_netdev_event(struct notifier_block *this,
4242 unsigned long event, void *ptr)
4244 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4245 struct mlx5e_flow_steering *fs;
4246 struct mlx5e_priv *peer_priv;
4247 struct mlx5e_tc_table *tc;
4248 struct mlx5e_priv *priv;
4250 if (ndev->netdev_ops != &mlx5e_netdev_ops ||
4251 event != NETDEV_UNREGISTER ||
4252 ndev->reg_state == NETREG_REGISTERED)
4255 tc = container_of(this, struct mlx5e_tc_table, netdevice_nb);
4256 fs = container_of(tc, struct mlx5e_flow_steering, tc);
4257 priv = container_of(fs, struct mlx5e_priv, fs);
4258 peer_priv = netdev_priv(ndev);
4259 if (priv == peer_priv ||
4260 !(priv->netdev->features & NETIF_F_HW_TC))
4263 mlx5e_tc_hairpin_update_dead_peer(priv, peer_priv);
4268 int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
4270 struct mlx5e_tc_table *tc = &priv->fs.tc;
4273 mutex_init(&tc->t_lock);
4274 mutex_init(&tc->mod_hdr.lock);
4275 hash_init(tc->mod_hdr.hlist);
4276 mutex_init(&tc->hairpin_tbl_lock);
4277 hash_init(tc->hairpin_tbl);
4279 err = rhashtable_init(&tc->ht, &tc_ht_params);
4283 tc->netdevice_nb.notifier_call = mlx5e_tc_netdev_event;
4284 err = register_netdevice_notifier_dev_net(priv->netdev,
4288 tc->netdevice_nb.notifier_call = NULL;
4289 mlx5_core_warn(priv->mdev, "Failed to register netdev notifier\n");
4295 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
4297 struct mlx5e_tc_flow *flow = ptr;
4298 struct mlx5e_priv *priv = flow->priv;
4300 mlx5e_tc_del_flow(priv, flow);
4304 void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
4306 struct mlx5e_tc_table *tc = &priv->fs.tc;
4308 if (tc->netdevice_nb.notifier_call)
4309 unregister_netdevice_notifier_dev_net(priv->netdev,
4313 mutex_destroy(&tc->mod_hdr.lock);
4314 mutex_destroy(&tc->hairpin_tbl_lock);
4316 rhashtable_destroy(&tc->ht);
4318 if (!IS_ERR_OR_NULL(tc->t)) {
4319 mlx5_destroy_flow_table(tc->t);
4322 mutex_destroy(&tc->t_lock);
4325 int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
4327 return rhashtable_init(tc_ht, &tc_ht_params);
4330 void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
4332 rhashtable_free_and_destroy(tc_ht, _mlx5e_tc_del_flow, NULL);
4335 int mlx5e_tc_num_filters(struct mlx5e_priv *priv, unsigned long flags)
4337 struct rhashtable *tc_ht = get_tc_ht(priv, flags);
4339 return atomic_read(&tc_ht->nelems);
4342 void mlx5e_tc_clean_fdb_peer_flows(struct mlx5_eswitch *esw)
4344 struct mlx5e_tc_flow *flow, *tmp;
4346 list_for_each_entry_safe(flow, tmp, &esw->offloads.peer_flows, peer)
4347 __mlx5e_tc_del_fdb_peer_flow(flow);
4350 void mlx5e_tc_reoffload_flows_work(struct work_struct *work)
4352 struct mlx5_rep_uplink_priv *rpriv =
4353 container_of(work, struct mlx5_rep_uplink_priv,
4354 reoffload_flows_work);
4355 struct mlx5e_tc_flow *flow, *tmp;
4357 mutex_lock(&rpriv->unready_flows_lock);
4358 list_for_each_entry_safe(flow, tmp, &rpriv->unready_flows, unready) {
4359 if (!mlx5e_tc_add_fdb_flow(flow->priv, flow, NULL))
4360 unready_flow_del(flow);
4362 mutex_unlock(&rpriv->unready_flows_lock);