net/mlx5e: Offload tc vlan push/pop using HW action
[sfrench/cifs-2.6.git] / drivers / net / ethernet / mellanox / mlx5 / core / en_tc.c
1 /*
2  * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
3  *
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:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
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.
22  *
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
30  * SOFTWARE.
31  */
32
33 #include <net/flow_dissector.h>
34 #include <net/sch_generic.h>
35 #include <net/pkt_cls.h>
36 #include <net/tc_act/tc_gact.h>
37 #include <net/tc_act/tc_skbedit.h>
38 #include <linux/mlx5/fs.h>
39 #include <linux/mlx5/device.h>
40 #include <linux/rhashtable.h>
41 #include <net/switchdev.h>
42 #include <net/tc_act/tc_mirred.h>
43 #include <net/tc_act/tc_vlan.h>
44 #include <net/tc_act/tc_tunnel_key.h>
45 #include <net/tc_act/tc_pedit.h>
46 #include <net/tc_act/tc_csum.h>
47 #include <net/vxlan.h>
48 #include <net/arp.h>
49 #include "en.h"
50 #include "en_rep.h"
51 #include "en_tc.h"
52 #include "eswitch.h"
53 #include "vxlan.h"
54 #include "fs_core.h"
55
56 struct mlx5_nic_flow_attr {
57         u32 action;
58         u32 flow_tag;
59         u32 mod_hdr_id;
60         u32 hairpin_tirn;
61         struct mlx5_flow_table  *hairpin_ft;
62 };
63
64 enum {
65         MLX5E_TC_FLOW_ESWITCH   = BIT(0),
66         MLX5E_TC_FLOW_NIC       = BIT(1),
67         MLX5E_TC_FLOW_OFFLOADED = BIT(2),
68         MLX5E_TC_FLOW_HAIRPIN   = BIT(3),
69         MLX5E_TC_FLOW_HAIRPIN_RSS = BIT(4),
70 };
71
72 struct mlx5e_tc_flow {
73         struct rhash_head       node;
74         u64                     cookie;
75         u8                      flags;
76         struct mlx5_flow_handle *rule;
77         struct list_head        encap;   /* flows sharing the same encap ID */
78         struct list_head        mod_hdr; /* flows sharing the same mod hdr ID */
79         struct list_head        hairpin; /* flows sharing the same hairpin */
80         union {
81                 struct mlx5_esw_flow_attr esw_attr[0];
82                 struct mlx5_nic_flow_attr nic_attr[0];
83         };
84 };
85
86 struct mlx5e_tc_flow_parse_attr {
87         struct ip_tunnel_info tun_info;
88         struct mlx5_flow_spec spec;
89         int num_mod_hdr_actions;
90         void *mod_hdr_actions;
91         int mirred_ifindex;
92 };
93
94 enum {
95         MLX5_HEADER_TYPE_VXLAN = 0x0,
96         MLX5_HEADER_TYPE_NVGRE = 0x1,
97 };
98
99 #define MLX5E_TC_TABLE_NUM_GROUPS 4
100 #define MLX5E_TC_TABLE_MAX_GROUP_SIZE (1 << 16)
101
102 struct mlx5e_hairpin {
103         struct mlx5_hairpin *pair;
104
105         struct mlx5_core_dev *func_mdev;
106         struct mlx5e_priv *func_priv;
107         u32 tdn;
108         u32 tirn;
109
110         int num_channels;
111         struct mlx5e_rqt indir_rqt;
112         u32 indir_tirn[MLX5E_NUM_INDIR_TIRS];
113         struct mlx5e_ttc_table ttc;
114 };
115
116 struct mlx5e_hairpin_entry {
117         /* a node of a hash table which keeps all the  hairpin entries */
118         struct hlist_node hairpin_hlist;
119
120         /* flows sharing the same hairpin */
121         struct list_head flows;
122
123         u16 peer_vhca_id;
124         u8 prio;
125         struct mlx5e_hairpin *hp;
126 };
127
128 struct mod_hdr_key {
129         int num_actions;
130         void *actions;
131 };
132
133 struct mlx5e_mod_hdr_entry {
134         /* a node of a hash table which keeps all the mod_hdr entries */
135         struct hlist_node mod_hdr_hlist;
136
137         /* flows sharing the same mod_hdr entry */
138         struct list_head flows;
139
140         struct mod_hdr_key key;
141
142         u32 mod_hdr_id;
143 };
144
145 #define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)
146
147 static inline u32 hash_mod_hdr_info(struct mod_hdr_key *key)
148 {
149         return jhash(key->actions,
150                      key->num_actions * MLX5_MH_ACT_SZ, 0);
151 }
152
153 static inline int cmp_mod_hdr_info(struct mod_hdr_key *a,
154                                    struct mod_hdr_key *b)
155 {
156         if (a->num_actions != b->num_actions)
157                 return 1;
158
159         return memcmp(a->actions, b->actions, a->num_actions * MLX5_MH_ACT_SZ);
160 }
161
162 static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
163                                 struct mlx5e_tc_flow *flow,
164                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
165 {
166         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
167         int num_actions, actions_size, namespace, err;
168         struct mlx5e_mod_hdr_entry *mh;
169         struct mod_hdr_key key;
170         bool found = false;
171         u32 hash_key;
172
173         num_actions  = parse_attr->num_mod_hdr_actions;
174         actions_size = MLX5_MH_ACT_SZ * num_actions;
175
176         key.actions = parse_attr->mod_hdr_actions;
177         key.num_actions = num_actions;
178
179         hash_key = hash_mod_hdr_info(&key);
180
181         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
182                 namespace = MLX5_FLOW_NAMESPACE_FDB;
183                 hash_for_each_possible(esw->offloads.mod_hdr_tbl, mh,
184                                        mod_hdr_hlist, hash_key) {
185                         if (!cmp_mod_hdr_info(&mh->key, &key)) {
186                                 found = true;
187                                 break;
188                         }
189                 }
190         } else {
191                 namespace = MLX5_FLOW_NAMESPACE_KERNEL;
192                 hash_for_each_possible(priv->fs.tc.mod_hdr_tbl, mh,
193                                        mod_hdr_hlist, hash_key) {
194                         if (!cmp_mod_hdr_info(&mh->key, &key)) {
195                                 found = true;
196                                 break;
197                         }
198                 }
199         }
200
201         if (found)
202                 goto attach_flow;
203
204         mh = kzalloc(sizeof(*mh) + actions_size, GFP_KERNEL);
205         if (!mh)
206                 return -ENOMEM;
207
208         mh->key.actions = (void *)mh + sizeof(*mh);
209         memcpy(mh->key.actions, key.actions, actions_size);
210         mh->key.num_actions = num_actions;
211         INIT_LIST_HEAD(&mh->flows);
212
213         err = mlx5_modify_header_alloc(priv->mdev, namespace,
214                                        mh->key.num_actions,
215                                        mh->key.actions,
216                                        &mh->mod_hdr_id);
217         if (err)
218                 goto out_err;
219
220         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
221                 hash_add(esw->offloads.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
222         else
223                 hash_add(priv->fs.tc.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
224
225 attach_flow:
226         list_add(&flow->mod_hdr, &mh->flows);
227         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
228                 flow->esw_attr->mod_hdr_id = mh->mod_hdr_id;
229         else
230                 flow->nic_attr->mod_hdr_id = mh->mod_hdr_id;
231
232         return 0;
233
234 out_err:
235         kfree(mh);
236         return err;
237 }
238
239 static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
240                                  struct mlx5e_tc_flow *flow)
241 {
242         struct list_head *next = flow->mod_hdr.next;
243
244         list_del(&flow->mod_hdr);
245
246         if (list_empty(next)) {
247                 struct mlx5e_mod_hdr_entry *mh;
248
249                 mh = list_entry(next, struct mlx5e_mod_hdr_entry, flows);
250
251                 mlx5_modify_header_dealloc(priv->mdev, mh->mod_hdr_id);
252                 hash_del(&mh->mod_hdr_hlist);
253                 kfree(mh);
254         }
255 }
256
257 static
258 struct mlx5_core_dev *mlx5e_hairpin_get_mdev(struct net *net, int ifindex)
259 {
260         struct net_device *netdev;
261         struct mlx5e_priv *priv;
262
263         netdev = __dev_get_by_index(net, ifindex);
264         priv = netdev_priv(netdev);
265         return priv->mdev;
266 }
267
268 static int mlx5e_hairpin_create_transport(struct mlx5e_hairpin *hp)
269 {
270         u32 in[MLX5_ST_SZ_DW(create_tir_in)] = {0};
271         void *tirc;
272         int err;
273
274         err = mlx5_core_alloc_transport_domain(hp->func_mdev, &hp->tdn);
275         if (err)
276                 goto alloc_tdn_err;
277
278         tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
279
280         MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
281         MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn[0]);
282         MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
283
284         err = mlx5_core_create_tir(hp->func_mdev, in, MLX5_ST_SZ_BYTES(create_tir_in), &hp->tirn);
285         if (err)
286                 goto create_tir_err;
287
288         return 0;
289
290 create_tir_err:
291         mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
292 alloc_tdn_err:
293         return err;
294 }
295
296 static void mlx5e_hairpin_destroy_transport(struct mlx5e_hairpin *hp)
297 {
298         mlx5_core_destroy_tir(hp->func_mdev, hp->tirn);
299         mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
300 }
301
302 static void mlx5e_hairpin_fill_rqt_rqns(struct mlx5e_hairpin *hp, void *rqtc)
303 {
304         u32 indirection_rqt[MLX5E_INDIR_RQT_SIZE], rqn;
305         struct mlx5e_priv *priv = hp->func_priv;
306         int i, ix, sz = MLX5E_INDIR_RQT_SIZE;
307
308         mlx5e_build_default_indir_rqt(indirection_rqt, sz,
309                                       hp->num_channels);
310
311         for (i = 0; i < sz; i++) {
312                 ix = i;
313                 if (priv->channels.params.rss_hfunc == ETH_RSS_HASH_XOR)
314                         ix = mlx5e_bits_invert(i, ilog2(sz));
315                 ix = indirection_rqt[ix];
316                 rqn = hp->pair->rqn[ix];
317                 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
318         }
319 }
320
321 static int mlx5e_hairpin_create_indirect_rqt(struct mlx5e_hairpin *hp)
322 {
323         int inlen, err, sz = MLX5E_INDIR_RQT_SIZE;
324         struct mlx5e_priv *priv = hp->func_priv;
325         struct mlx5_core_dev *mdev = priv->mdev;
326         void *rqtc;
327         u32 *in;
328
329         inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
330         in = kvzalloc(inlen, GFP_KERNEL);
331         if (!in)
332                 return -ENOMEM;
333
334         rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
335
336         MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
337         MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
338
339         mlx5e_hairpin_fill_rqt_rqns(hp, rqtc);
340
341         err = mlx5_core_create_rqt(mdev, in, inlen, &hp->indir_rqt.rqtn);
342         if (!err)
343                 hp->indir_rqt.enabled = true;
344
345         kvfree(in);
346         return err;
347 }
348
349 static int mlx5e_hairpin_create_indirect_tirs(struct mlx5e_hairpin *hp)
350 {
351         struct mlx5e_priv *priv = hp->func_priv;
352         u32 in[MLX5_ST_SZ_DW(create_tir_in)];
353         int tt, i, err;
354         void *tirc;
355
356         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
357                 memset(in, 0, MLX5_ST_SZ_BYTES(create_tir_in));
358                 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
359
360                 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
361                 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
362                 MLX5_SET(tirc, tirc, indirect_table, hp->indir_rqt.rqtn);
363                 mlx5e_build_indir_tir_ctx_hash(&priv->channels.params, tt, tirc, false);
364
365                 err = mlx5_core_create_tir(hp->func_mdev, in,
366                                            MLX5_ST_SZ_BYTES(create_tir_in), &hp->indir_tirn[tt]);
367                 if (err) {
368                         mlx5_core_warn(hp->func_mdev, "create indirect tirs failed, %d\n", err);
369                         goto err_destroy_tirs;
370                 }
371         }
372         return 0;
373
374 err_destroy_tirs:
375         for (i = 0; i < tt; i++)
376                 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[i]);
377         return err;
378 }
379
380 static void mlx5e_hairpin_destroy_indirect_tirs(struct mlx5e_hairpin *hp)
381 {
382         int tt;
383
384         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
385                 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[tt]);
386 }
387
388 static void mlx5e_hairpin_set_ttc_params(struct mlx5e_hairpin *hp,
389                                          struct ttc_params *ttc_params)
390 {
391         struct mlx5_flow_table_attr *ft_attr = &ttc_params->ft_attr;
392         int tt;
393
394         memset(ttc_params, 0, sizeof(*ttc_params));
395
396         ttc_params->any_tt_tirn = hp->tirn;
397
398         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
399                 ttc_params->indir_tirn[tt] = hp->indir_tirn[tt];
400
401         ft_attr->max_fte = MLX5E_NUM_TT;
402         ft_attr->level = MLX5E_TC_TTC_FT_LEVEL;
403         ft_attr->prio = MLX5E_TC_PRIO;
404 }
405
406 static int mlx5e_hairpin_rss_init(struct mlx5e_hairpin *hp)
407 {
408         struct mlx5e_priv *priv = hp->func_priv;
409         struct ttc_params ttc_params;
410         int err;
411
412         err = mlx5e_hairpin_create_indirect_rqt(hp);
413         if (err)
414                 return err;
415
416         err = mlx5e_hairpin_create_indirect_tirs(hp);
417         if (err)
418                 goto err_create_indirect_tirs;
419
420         mlx5e_hairpin_set_ttc_params(hp, &ttc_params);
421         err = mlx5e_create_ttc_table(priv, &ttc_params, &hp->ttc);
422         if (err)
423                 goto err_create_ttc_table;
424
425         netdev_dbg(priv->netdev, "add hairpin: using %d channels rss ttc table id %x\n",
426                    hp->num_channels, hp->ttc.ft.t->id);
427
428         return 0;
429
430 err_create_ttc_table:
431         mlx5e_hairpin_destroy_indirect_tirs(hp);
432 err_create_indirect_tirs:
433         mlx5e_destroy_rqt(priv, &hp->indir_rqt);
434
435         return err;
436 }
437
438 static void mlx5e_hairpin_rss_cleanup(struct mlx5e_hairpin *hp)
439 {
440         struct mlx5e_priv *priv = hp->func_priv;
441
442         mlx5e_destroy_ttc_table(priv, &hp->ttc);
443         mlx5e_hairpin_destroy_indirect_tirs(hp);
444         mlx5e_destroy_rqt(priv, &hp->indir_rqt);
445 }
446
447 static struct mlx5e_hairpin *
448 mlx5e_hairpin_create(struct mlx5e_priv *priv, struct mlx5_hairpin_params *params,
449                      int peer_ifindex)
450 {
451         struct mlx5_core_dev *func_mdev, *peer_mdev;
452         struct mlx5e_hairpin *hp;
453         struct mlx5_hairpin *pair;
454         int err;
455
456         hp = kzalloc(sizeof(*hp), GFP_KERNEL);
457         if (!hp)
458                 return ERR_PTR(-ENOMEM);
459
460         func_mdev = priv->mdev;
461         peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
462
463         pair = mlx5_core_hairpin_create(func_mdev, peer_mdev, params);
464         if (IS_ERR(pair)) {
465                 err = PTR_ERR(pair);
466                 goto create_pair_err;
467         }
468         hp->pair = pair;
469         hp->func_mdev = func_mdev;
470         hp->func_priv = priv;
471         hp->num_channels = params->num_channels;
472
473         err = mlx5e_hairpin_create_transport(hp);
474         if (err)
475                 goto create_transport_err;
476
477         if (hp->num_channels > 1) {
478                 err = mlx5e_hairpin_rss_init(hp);
479                 if (err)
480                         goto rss_init_err;
481         }
482
483         return hp;
484
485 rss_init_err:
486         mlx5e_hairpin_destroy_transport(hp);
487 create_transport_err:
488         mlx5_core_hairpin_destroy(hp->pair);
489 create_pair_err:
490         kfree(hp);
491         return ERR_PTR(err);
492 }
493
494 static void mlx5e_hairpin_destroy(struct mlx5e_hairpin *hp)
495 {
496         if (hp->num_channels > 1)
497                 mlx5e_hairpin_rss_cleanup(hp);
498         mlx5e_hairpin_destroy_transport(hp);
499         mlx5_core_hairpin_destroy(hp->pair);
500         kvfree(hp);
501 }
502
503 static inline u32 hash_hairpin_info(u16 peer_vhca_id, u8 prio)
504 {
505         return (peer_vhca_id << 16 | prio);
506 }
507
508 static struct mlx5e_hairpin_entry *mlx5e_hairpin_get(struct mlx5e_priv *priv,
509                                                      u16 peer_vhca_id, u8 prio)
510 {
511         struct mlx5e_hairpin_entry *hpe;
512         u32 hash_key = hash_hairpin_info(peer_vhca_id, prio);
513
514         hash_for_each_possible(priv->fs.tc.hairpin_tbl, hpe,
515                                hairpin_hlist, hash_key) {
516                 if (hpe->peer_vhca_id == peer_vhca_id && hpe->prio == prio)
517                         return hpe;
518         }
519
520         return NULL;
521 }
522
523 #define UNKNOWN_MATCH_PRIO 8
524
525 static int mlx5e_hairpin_get_prio(struct mlx5e_priv *priv,
526                                   struct mlx5_flow_spec *spec, u8 *match_prio)
527 {
528         void *headers_c, *headers_v;
529         u8 prio_val, prio_mask = 0;
530         bool vlan_present;
531
532 #ifdef CONFIG_MLX5_CORE_EN_DCB
533         if (priv->dcbx_dp.trust_state != MLX5_QPTS_TRUST_PCP) {
534                 netdev_warn(priv->netdev,
535                             "only PCP trust state supported for hairpin\n");
536                 return -EOPNOTSUPP;
537         }
538 #endif
539         headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, outer_headers);
540         headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
541
542         vlan_present = MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag);
543         if (vlan_present) {
544                 prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
545                 prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
546         }
547
548         if (!vlan_present || !prio_mask) {
549                 prio_val = UNKNOWN_MATCH_PRIO;
550         } else if (prio_mask != 0x7) {
551                 netdev_warn(priv->netdev,
552                             "masked priority match not supported for hairpin\n");
553                 return -EOPNOTSUPP;
554         }
555
556         *match_prio = prio_val;
557         return 0;
558 }
559
560 static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
561                                   struct mlx5e_tc_flow *flow,
562                                   struct mlx5e_tc_flow_parse_attr *parse_attr)
563 {
564         int peer_ifindex = parse_attr->mirred_ifindex;
565         struct mlx5_hairpin_params params;
566         struct mlx5_core_dev *peer_mdev;
567         struct mlx5e_hairpin_entry *hpe;
568         struct mlx5e_hairpin *hp;
569         u64 link_speed64;
570         u32 link_speed;
571         u8 match_prio;
572         u16 peer_id;
573         int err;
574
575         peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
576         if (!MLX5_CAP_GEN(priv->mdev, hairpin) || !MLX5_CAP_GEN(peer_mdev, hairpin)) {
577                 netdev_warn(priv->netdev, "hairpin is not supported\n");
578                 return -EOPNOTSUPP;
579         }
580
581         peer_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
582         err = mlx5e_hairpin_get_prio(priv, &parse_attr->spec, &match_prio);
583         if (err)
584                 return err;
585         hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
586         if (hpe)
587                 goto attach_flow;
588
589         hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
590         if (!hpe)
591                 return -ENOMEM;
592
593         INIT_LIST_HEAD(&hpe->flows);
594         hpe->peer_vhca_id = peer_id;
595         hpe->prio = match_prio;
596
597         params.log_data_size = 15;
598         params.log_data_size = min_t(u8, params.log_data_size,
599                                      MLX5_CAP_GEN(priv->mdev, log_max_hairpin_wq_data_sz));
600         params.log_data_size = max_t(u8, params.log_data_size,
601                                      MLX5_CAP_GEN(priv->mdev, log_min_hairpin_wq_data_sz));
602
603         params.log_num_packets = params.log_data_size -
604                                  MLX5_MPWRQ_MIN_LOG_STRIDE_SZ(priv->mdev);
605         params.log_num_packets = min_t(u8, params.log_num_packets,
606                                        MLX5_CAP_GEN(priv->mdev, log_max_hairpin_num_packets));
607
608         params.q_counter = priv->q_counter;
609         /* set hairpin pair per each 50Gbs share of the link */
610         mlx5e_get_max_linkspeed(priv->mdev, &link_speed);
611         link_speed = max_t(u32, link_speed, 50000);
612         link_speed64 = link_speed;
613         do_div(link_speed64, 50000);
614         params.num_channels = link_speed64;
615
616         hp = mlx5e_hairpin_create(priv, &params, peer_ifindex);
617         if (IS_ERR(hp)) {
618                 err = PTR_ERR(hp);
619                 goto create_hairpin_err;
620         }
621
622         netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
623                    hp->tirn, hp->pair->rqn[0], hp->pair->peer_mdev->priv.name,
624                    hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
625
626         hpe->hp = hp;
627         hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
628                  hash_hairpin_info(peer_id, match_prio));
629
630 attach_flow:
631         if (hpe->hp->num_channels > 1) {
632                 flow->flags |= MLX5E_TC_FLOW_HAIRPIN_RSS;
633                 flow->nic_attr->hairpin_ft = hpe->hp->ttc.ft.t;
634         } else {
635                 flow->nic_attr->hairpin_tirn = hpe->hp->tirn;
636         }
637         list_add(&flow->hairpin, &hpe->flows);
638
639         return 0;
640
641 create_hairpin_err:
642         kfree(hpe);
643         return err;
644 }
645
646 static void mlx5e_hairpin_flow_del(struct mlx5e_priv *priv,
647                                    struct mlx5e_tc_flow *flow)
648 {
649         struct list_head *next = flow->hairpin.next;
650
651         list_del(&flow->hairpin);
652
653         /* no more hairpin flows for us, release the hairpin pair */
654         if (list_empty(next)) {
655                 struct mlx5e_hairpin_entry *hpe;
656
657                 hpe = list_entry(next, struct mlx5e_hairpin_entry, flows);
658
659                 netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
660                            hpe->hp->pair->peer_mdev->priv.name);
661
662                 mlx5e_hairpin_destroy(hpe->hp);
663                 hash_del(&hpe->hairpin_hlist);
664                 kfree(hpe);
665         }
666 }
667
668 static struct mlx5_flow_handle *
669 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
670                       struct mlx5e_tc_flow_parse_attr *parse_attr,
671                       struct mlx5e_tc_flow *flow)
672 {
673         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
674         struct mlx5_core_dev *dev = priv->mdev;
675         struct mlx5_flow_destination dest[2] = {};
676         struct mlx5_flow_act flow_act = {
677                 .action = attr->action,
678                 .has_flow_tag = true,
679                 .flow_tag = attr->flow_tag,
680                 .encap_id = 0,
681         };
682         struct mlx5_fc *counter = NULL;
683         struct mlx5_flow_handle *rule;
684         bool table_created = false;
685         int err, dest_ix = 0;
686
687         if (flow->flags & MLX5E_TC_FLOW_HAIRPIN) {
688                 err = mlx5e_hairpin_flow_add(priv, flow, parse_attr);
689                 if (err) {
690                         rule = ERR_PTR(err);
691                         goto err_add_hairpin_flow;
692                 }
693                 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN_RSS) {
694                         dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
695                         dest[dest_ix].ft = attr->hairpin_ft;
696                 } else {
697                         dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_TIR;
698                         dest[dest_ix].tir_num = attr->hairpin_tirn;
699                 }
700                 dest_ix++;
701         } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
702                 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
703                 dest[dest_ix].ft = priv->fs.vlan.ft.t;
704                 dest_ix++;
705         }
706
707         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
708                 counter = mlx5_fc_create(dev, true);
709                 if (IS_ERR(counter)) {
710                         rule = ERR_CAST(counter);
711                         goto err_fc_create;
712                 }
713                 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
714                 dest[dest_ix].counter = counter;
715                 dest_ix++;
716         }
717
718         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
719                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
720                 flow_act.modify_id = attr->mod_hdr_id;
721                 kfree(parse_attr->mod_hdr_actions);
722                 if (err) {
723                         rule = ERR_PTR(err);
724                         goto err_create_mod_hdr_id;
725                 }
726         }
727
728         if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
729                 int tc_grp_size, tc_tbl_size;
730                 u32 max_flow_counter;
731
732                 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
733                                     MLX5_CAP_GEN(dev, max_flow_counter_15_0);
734
735                 tc_grp_size = min_t(int, max_flow_counter, MLX5E_TC_TABLE_MAX_GROUP_SIZE);
736
737                 tc_tbl_size = min_t(int, tc_grp_size * MLX5E_TC_TABLE_NUM_GROUPS,
738                                     BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev, log_max_ft_size)));
739
740                 priv->fs.tc.t =
741                         mlx5_create_auto_grouped_flow_table(priv->fs.ns,
742                                                             MLX5E_TC_PRIO,
743                                                             tc_tbl_size,
744                                                             MLX5E_TC_TABLE_NUM_GROUPS,
745                                                             MLX5E_TC_FT_LEVEL, 0);
746                 if (IS_ERR(priv->fs.tc.t)) {
747                         netdev_err(priv->netdev,
748                                    "Failed to create tc offload table\n");
749                         rule = ERR_CAST(priv->fs.tc.t);
750                         goto err_create_ft;
751                 }
752
753                 table_created = true;
754         }
755
756         parse_attr->spec.match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
757         rule = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
758                                    &flow_act, dest, dest_ix);
759
760         if (IS_ERR(rule))
761                 goto err_add_rule;
762
763         return rule;
764
765 err_add_rule:
766         if (table_created) {
767                 mlx5_destroy_flow_table(priv->fs.tc.t);
768                 priv->fs.tc.t = NULL;
769         }
770 err_create_ft:
771         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
772                 mlx5e_detach_mod_hdr(priv, flow);
773 err_create_mod_hdr_id:
774         mlx5_fc_destroy(dev, counter);
775 err_fc_create:
776         if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
777                 mlx5e_hairpin_flow_del(priv, flow);
778 err_add_hairpin_flow:
779         return rule;
780 }
781
782 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
783                                   struct mlx5e_tc_flow *flow)
784 {
785         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
786         struct mlx5_fc *counter = NULL;
787
788         counter = mlx5_flow_rule_counter(flow->rule);
789         mlx5_del_flow_rules(flow->rule);
790         mlx5_fc_destroy(priv->mdev, counter);
791
792         if (!mlx5e_tc_num_filters(priv) && (priv->fs.tc.t)) {
793                 mlx5_destroy_flow_table(priv->fs.tc.t);
794                 priv->fs.tc.t = NULL;
795         }
796
797         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
798                 mlx5e_detach_mod_hdr(priv, flow);
799
800         if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
801                 mlx5e_hairpin_flow_del(priv, flow);
802 }
803
804 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
805                                struct mlx5e_tc_flow *flow);
806
807 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
808                               struct ip_tunnel_info *tun_info,
809                               struct net_device *mirred_dev,
810                               struct net_device **encap_dev,
811                               struct mlx5e_tc_flow *flow);
812
813 static struct mlx5_flow_handle *
814 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
815                       struct mlx5e_tc_flow_parse_attr *parse_attr,
816                       struct mlx5e_tc_flow *flow)
817 {
818         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
819         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
820         struct net_device *out_dev, *encap_dev = NULL;
821         struct mlx5_flow_handle *rule = NULL;
822         struct mlx5e_rep_priv *rpriv;
823         struct mlx5e_priv *out_priv;
824         int err;
825
826         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP) {
827                 out_dev = __dev_get_by_index(dev_net(priv->netdev),
828                                              attr->parse_attr->mirred_ifindex);
829                 err = mlx5e_attach_encap(priv, &parse_attr->tun_info,
830                                          out_dev, &encap_dev, flow);
831                 if (err) {
832                         rule = ERR_PTR(err);
833                         if (err != -EAGAIN)
834                                 goto err_attach_encap;
835                 }
836                 out_priv = netdev_priv(encap_dev);
837                 rpriv = out_priv->ppriv;
838                 attr->out_rep = rpriv->rep;
839         }
840
841         err = mlx5_eswitch_add_vlan_action(esw, attr);
842         if (err) {
843                 rule = ERR_PTR(err);
844                 goto err_add_vlan;
845         }
846
847         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
848                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
849                 kfree(parse_attr->mod_hdr_actions);
850                 if (err) {
851                         rule = ERR_PTR(err);
852                         goto err_mod_hdr;
853                 }
854         }
855
856         /* we get here if (1) there's no error (rule being null) or when
857          * (2) there's an encap action and we're on -EAGAIN (no valid neigh)
858          */
859         if (rule != ERR_PTR(-EAGAIN)) {
860                 rule = mlx5_eswitch_add_offloaded_rule(esw, &parse_attr->spec, attr);
861                 if (IS_ERR(rule))
862                         goto err_add_rule;
863         }
864         return rule;
865
866 err_add_rule:
867         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
868                 mlx5e_detach_mod_hdr(priv, flow);
869 err_mod_hdr:
870         mlx5_eswitch_del_vlan_action(esw, attr);
871 err_add_vlan:
872         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP)
873                 mlx5e_detach_encap(priv, flow);
874 err_attach_encap:
875         return rule;
876 }
877
878 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
879                                   struct mlx5e_tc_flow *flow)
880 {
881         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
882         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
883
884         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
885                 flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
886                 mlx5_eswitch_del_offloaded_rule(esw, flow->rule, attr);
887         }
888
889         mlx5_eswitch_del_vlan_action(esw, attr);
890
891         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP) {
892                 mlx5e_detach_encap(priv, flow);
893                 kvfree(attr->parse_attr);
894         }
895
896         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
897                 mlx5e_detach_mod_hdr(priv, flow);
898 }
899
900 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
901                               struct mlx5e_encap_entry *e)
902 {
903         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
904         struct mlx5_esw_flow_attr *esw_attr;
905         struct mlx5e_tc_flow *flow;
906         int err;
907
908         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
909                                e->encap_size, e->encap_header,
910                                &e->encap_id);
911         if (err) {
912                 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %d\n",
913                                err);
914                 return;
915         }
916         e->flags |= MLX5_ENCAP_ENTRY_VALID;
917         mlx5e_rep_queue_neigh_stats_work(priv);
918
919         list_for_each_entry(flow, &e->flows, encap) {
920                 esw_attr = flow->esw_attr;
921                 esw_attr->encap_id = e->encap_id;
922                 flow->rule = mlx5_eswitch_add_offloaded_rule(esw, &esw_attr->parse_attr->spec, esw_attr);
923                 if (IS_ERR(flow->rule)) {
924                         err = PTR_ERR(flow->rule);
925                         mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
926                                        err);
927                         continue;
928                 }
929                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
930         }
931 }
932
933 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
934                               struct mlx5e_encap_entry *e)
935 {
936         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
937         struct mlx5e_tc_flow *flow;
938
939         list_for_each_entry(flow, &e->flows, encap) {
940                 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
941                         flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
942                         mlx5_eswitch_del_offloaded_rule(esw, flow->rule, flow->esw_attr);
943                 }
944         }
945
946         if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
947                 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
948                 mlx5_encap_dealloc(priv->mdev, e->encap_id);
949         }
950 }
951
952 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
953 {
954         struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
955         u64 bytes, packets, lastuse = 0;
956         struct mlx5e_tc_flow *flow;
957         struct mlx5e_encap_entry *e;
958         struct mlx5_fc *counter;
959         struct neigh_table *tbl;
960         bool neigh_used = false;
961         struct neighbour *n;
962
963         if (m_neigh->family == AF_INET)
964                 tbl = &arp_tbl;
965 #if IS_ENABLED(CONFIG_IPV6)
966         else if (m_neigh->family == AF_INET6)
967                 tbl = ipv6_stub->nd_tbl;
968 #endif
969         else
970                 return;
971
972         list_for_each_entry(e, &nhe->encap_list, encap_list) {
973                 if (!(e->flags & MLX5_ENCAP_ENTRY_VALID))
974                         continue;
975                 list_for_each_entry(flow, &e->flows, encap) {
976                         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
977                                 counter = mlx5_flow_rule_counter(flow->rule);
978                                 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
979                                 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
980                                         neigh_used = true;
981                                         break;
982                                 }
983                         }
984                 }
985         }
986
987         if (neigh_used) {
988                 nhe->reported_lastuse = jiffies;
989
990                 /* find the relevant neigh according to the cached device and
991                  * dst ip pair
992                  */
993                 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
994                 if (!n) {
995                         WARN(1, "The neighbour already freed\n");
996                         return;
997                 }
998
999                 neigh_event_send(n, NULL);
1000                 neigh_release(n);
1001         }
1002 }
1003
1004 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1005                                struct mlx5e_tc_flow *flow)
1006 {
1007         struct list_head *next = flow->encap.next;
1008
1009         list_del(&flow->encap);
1010         if (list_empty(next)) {
1011                 struct mlx5e_encap_entry *e;
1012
1013                 e = list_entry(next, struct mlx5e_encap_entry, flows);
1014                 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1015
1016                 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1017                         mlx5_encap_dealloc(priv->mdev, e->encap_id);
1018
1019                 hash_del_rcu(&e->encap_hlist);
1020                 kfree(e->encap_header);
1021                 kfree(e);
1022         }
1023 }
1024
1025 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1026                               struct mlx5e_tc_flow *flow)
1027 {
1028         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1029                 mlx5e_tc_del_fdb_flow(priv, flow);
1030         else
1031                 mlx5e_tc_del_nic_flow(priv, flow);
1032 }
1033
1034 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
1035                              struct tc_cls_flower_offload *f)
1036 {
1037         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1038                                        outer_headers);
1039         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1040                                        outer_headers);
1041         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1042                                     misc_parameters);
1043         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1044                                     misc_parameters);
1045
1046         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
1047         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
1048
1049         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
1050                 struct flow_dissector_key_keyid *key =
1051                         skb_flow_dissector_target(f->dissector,
1052                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
1053                                                   f->key);
1054                 struct flow_dissector_key_keyid *mask =
1055                         skb_flow_dissector_target(f->dissector,
1056                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
1057                                                   f->mask);
1058                 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
1059                          be32_to_cpu(mask->keyid));
1060                 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
1061                          be32_to_cpu(key->keyid));
1062         }
1063 }
1064
1065 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1066                              struct mlx5_flow_spec *spec,
1067                              struct tc_cls_flower_offload *f)
1068 {
1069         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1070                                        outer_headers);
1071         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1072                                        outer_headers);
1073
1074         struct flow_dissector_key_control *enc_control =
1075                 skb_flow_dissector_target(f->dissector,
1076                                           FLOW_DISSECTOR_KEY_ENC_CONTROL,
1077                                           f->key);
1078
1079         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
1080                 struct flow_dissector_key_ports *key =
1081                         skb_flow_dissector_target(f->dissector,
1082                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
1083                                                   f->key);
1084                 struct flow_dissector_key_ports *mask =
1085                         skb_flow_dissector_target(f->dissector,
1086                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
1087                                                   f->mask);
1088                 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1089                 struct mlx5e_rep_priv *uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1090                 struct net_device *up_dev = uplink_rpriv->netdev;
1091                 struct mlx5e_priv *up_priv = netdev_priv(up_dev);
1092
1093                 /* Full udp dst port must be given */
1094                 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
1095                         goto vxlan_match_offload_err;
1096
1097                 if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->dst)) &&
1098                     MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
1099                         parse_vxlan_attr(spec, f);
1100                 else {
1101                         netdev_warn(priv->netdev,
1102                                     "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
1103                         return -EOPNOTSUPP;
1104                 }
1105
1106                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1107                          udp_dport, ntohs(mask->dst));
1108                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1109                          udp_dport, ntohs(key->dst));
1110
1111                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1112                          udp_sport, ntohs(mask->src));
1113                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1114                          udp_sport, ntohs(key->src));
1115         } else { /* udp dst port must be given */
1116 vxlan_match_offload_err:
1117                 netdev_warn(priv->netdev,
1118                             "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
1119                 return -EOPNOTSUPP;
1120         }
1121
1122         if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1123                 struct flow_dissector_key_ipv4_addrs *key =
1124                         skb_flow_dissector_target(f->dissector,
1125                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1126                                                   f->key);
1127                 struct flow_dissector_key_ipv4_addrs *mask =
1128                         skb_flow_dissector_target(f->dissector,
1129                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1130                                                   f->mask);
1131                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1132                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
1133                          ntohl(mask->src));
1134                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1135                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
1136                          ntohl(key->src));
1137
1138                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1139                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1140                          ntohl(mask->dst));
1141                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1142                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1143                          ntohl(key->dst));
1144
1145                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1146                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
1147         } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1148                 struct flow_dissector_key_ipv6_addrs *key =
1149                         skb_flow_dissector_target(f->dissector,
1150                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1151                                                   f->key);
1152                 struct flow_dissector_key_ipv6_addrs *mask =
1153                         skb_flow_dissector_target(f->dissector,
1154                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1155                                                   f->mask);
1156
1157                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1158                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1159                        &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1160                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1161                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1162                        &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1163
1164                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1165                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1166                        &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1167                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1168                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1169                        &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1170
1171                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1172                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
1173         }
1174
1175         /* Enforce DMAC when offloading incoming tunneled flows.
1176          * Flow counters require a match on the DMAC.
1177          */
1178         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
1179         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
1180         ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1181                                      dmac_47_16), priv->netdev->dev_addr);
1182
1183         /* let software handle IP fragments */
1184         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1185         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
1186
1187         return 0;
1188 }
1189
1190 static int __parse_cls_flower(struct mlx5e_priv *priv,
1191                               struct mlx5_flow_spec *spec,
1192                               struct tc_cls_flower_offload *f,
1193                               u8 *min_inline)
1194 {
1195         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1196                                        outer_headers);
1197         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1198                                        outer_headers);
1199         u16 addr_type = 0;
1200         u8 ip_proto = 0;
1201
1202         *min_inline = MLX5_INLINE_MODE_L2;
1203
1204         if (f->dissector->used_keys &
1205             ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
1206               BIT(FLOW_DISSECTOR_KEY_BASIC) |
1207               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
1208               BIT(FLOW_DISSECTOR_KEY_VLAN) |
1209               BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
1210               BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
1211               BIT(FLOW_DISSECTOR_KEY_PORTS) |
1212               BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
1213               BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
1214               BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
1215               BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
1216               BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
1217               BIT(FLOW_DISSECTOR_KEY_TCP) |
1218               BIT(FLOW_DISSECTOR_KEY_IP))) {
1219                 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
1220                             f->dissector->used_keys);
1221                 return -EOPNOTSUPP;
1222         }
1223
1224         if ((dissector_uses_key(f->dissector,
1225                                 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
1226              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
1227              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
1228             dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
1229                 struct flow_dissector_key_control *key =
1230                         skb_flow_dissector_target(f->dissector,
1231                                                   FLOW_DISSECTOR_KEY_ENC_CONTROL,
1232                                                   f->key);
1233                 switch (key->addr_type) {
1234                 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
1235                 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
1236                         if (parse_tunnel_attr(priv, spec, f))
1237                                 return -EOPNOTSUPP;
1238                         break;
1239                 default:
1240                         return -EOPNOTSUPP;
1241                 }
1242
1243                 /* In decap flow, header pointers should point to the inner
1244                  * headers, outer header were already set by parse_tunnel_attr
1245                  */
1246                 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1247                                          inner_headers);
1248                 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1249                                          inner_headers);
1250         }
1251
1252         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
1253                 struct flow_dissector_key_control *key =
1254                         skb_flow_dissector_target(f->dissector,
1255                                                   FLOW_DISSECTOR_KEY_CONTROL,
1256                                                   f->key);
1257
1258                 struct flow_dissector_key_control *mask =
1259                         skb_flow_dissector_target(f->dissector,
1260                                                   FLOW_DISSECTOR_KEY_CONTROL,
1261                                                   f->mask);
1262                 addr_type = key->addr_type;
1263
1264                 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
1265                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1266                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
1267                                  key->flags & FLOW_DIS_IS_FRAGMENT);
1268
1269                         /* the HW doesn't need L3 inline to match on frag=no */
1270                         if (key->flags & FLOW_DIS_IS_FRAGMENT)
1271                                 *min_inline = MLX5_INLINE_MODE_IP;
1272                 }
1273         }
1274
1275         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1276                 struct flow_dissector_key_basic *key =
1277                         skb_flow_dissector_target(f->dissector,
1278                                                   FLOW_DISSECTOR_KEY_BASIC,
1279                                                   f->key);
1280                 struct flow_dissector_key_basic *mask =
1281                         skb_flow_dissector_target(f->dissector,
1282                                                   FLOW_DISSECTOR_KEY_BASIC,
1283                                                   f->mask);
1284                 ip_proto = key->ip_proto;
1285
1286                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1287                          ntohs(mask->n_proto));
1288                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1289                          ntohs(key->n_proto));
1290
1291                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
1292                          mask->ip_proto);
1293                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
1294                          key->ip_proto);
1295
1296                 if (mask->ip_proto)
1297                         *min_inline = MLX5_INLINE_MODE_IP;
1298         }
1299
1300         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
1301                 struct flow_dissector_key_eth_addrs *key =
1302                         skb_flow_dissector_target(f->dissector,
1303                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
1304                                                   f->key);
1305                 struct flow_dissector_key_eth_addrs *mask =
1306                         skb_flow_dissector_target(f->dissector,
1307                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
1308                                                   f->mask);
1309
1310                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1311                                              dmac_47_16),
1312                                 mask->dst);
1313                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1314                                              dmac_47_16),
1315                                 key->dst);
1316
1317                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1318                                              smac_47_16),
1319                                 mask->src);
1320                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1321                                              smac_47_16),
1322                                 key->src);
1323         }
1324
1325         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
1326                 struct flow_dissector_key_vlan *key =
1327                         skb_flow_dissector_target(f->dissector,
1328                                                   FLOW_DISSECTOR_KEY_VLAN,
1329                                                   f->key);
1330                 struct flow_dissector_key_vlan *mask =
1331                         skb_flow_dissector_target(f->dissector,
1332                                                   FLOW_DISSECTOR_KEY_VLAN,
1333                                                   f->mask);
1334                 if (mask->vlan_id || mask->vlan_priority) {
1335                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
1336                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
1337
1338                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
1339                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
1340
1341                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
1342                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
1343                 }
1344         }
1345
1346         if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1347                 struct flow_dissector_key_ipv4_addrs *key =
1348                         skb_flow_dissector_target(f->dissector,
1349                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1350                                                   f->key);
1351                 struct flow_dissector_key_ipv4_addrs *mask =
1352                         skb_flow_dissector_target(f->dissector,
1353                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1354                                                   f->mask);
1355
1356                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1357                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
1358                        &mask->src, sizeof(mask->src));
1359                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1360                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
1361                        &key->src, sizeof(key->src));
1362                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1363                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1364                        &mask->dst, sizeof(mask->dst));
1365                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1366                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1367                        &key->dst, sizeof(key->dst));
1368
1369                 if (mask->src || mask->dst)
1370                         *min_inline = MLX5_INLINE_MODE_IP;
1371         }
1372
1373         if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1374                 struct flow_dissector_key_ipv6_addrs *key =
1375                         skb_flow_dissector_target(f->dissector,
1376                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1377                                                   f->key);
1378                 struct flow_dissector_key_ipv6_addrs *mask =
1379                         skb_flow_dissector_target(f->dissector,
1380                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1381                                                   f->mask);
1382
1383                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1384                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1385                        &mask->src, sizeof(mask->src));
1386                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1387                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1388                        &key->src, sizeof(key->src));
1389
1390                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1391                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1392                        &mask->dst, sizeof(mask->dst));
1393                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1394                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1395                        &key->dst, sizeof(key->dst));
1396
1397                 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
1398                     ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
1399                         *min_inline = MLX5_INLINE_MODE_IP;
1400         }
1401
1402         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_IP)) {
1403                 struct flow_dissector_key_ip *key =
1404                         skb_flow_dissector_target(f->dissector,
1405                                                   FLOW_DISSECTOR_KEY_IP,
1406                                                   f->key);
1407                 struct flow_dissector_key_ip *mask =
1408                         skb_flow_dissector_target(f->dissector,
1409                                                   FLOW_DISSECTOR_KEY_IP,
1410                                                   f->mask);
1411
1412                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn, mask->tos & 0x3);
1413                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, key->tos & 0x3);
1414
1415                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp, mask->tos >> 2);
1416                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, key->tos  >> 2);
1417
1418                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit, mask->ttl);
1419                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit, key->ttl);
1420
1421                 if (mask->ttl &&
1422                     !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
1423                                                 ft_field_support.outer_ipv4_ttl))
1424                         return -EOPNOTSUPP;
1425
1426                 if (mask->tos || mask->ttl)
1427                         *min_inline = MLX5_INLINE_MODE_IP;
1428         }
1429
1430         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
1431                 struct flow_dissector_key_ports *key =
1432                         skb_flow_dissector_target(f->dissector,
1433                                                   FLOW_DISSECTOR_KEY_PORTS,
1434                                                   f->key);
1435                 struct flow_dissector_key_ports *mask =
1436                         skb_flow_dissector_target(f->dissector,
1437                                                   FLOW_DISSECTOR_KEY_PORTS,
1438                                                   f->mask);
1439                 switch (ip_proto) {
1440                 case IPPROTO_TCP:
1441                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1442                                  tcp_sport, ntohs(mask->src));
1443                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1444                                  tcp_sport, ntohs(key->src));
1445
1446                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1447                                  tcp_dport, ntohs(mask->dst));
1448                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1449                                  tcp_dport, ntohs(key->dst));
1450                         break;
1451
1452                 case IPPROTO_UDP:
1453                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1454                                  udp_sport, ntohs(mask->src));
1455                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1456                                  udp_sport, ntohs(key->src));
1457
1458                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1459                                  udp_dport, ntohs(mask->dst));
1460                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1461                                  udp_dport, ntohs(key->dst));
1462                         break;
1463                 default:
1464                         netdev_err(priv->netdev,
1465                                    "Only UDP and TCP transport are supported\n");
1466                         return -EINVAL;
1467                 }
1468
1469                 if (mask->src || mask->dst)
1470                         *min_inline = MLX5_INLINE_MODE_TCP_UDP;
1471         }
1472
1473         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_TCP)) {
1474                 struct flow_dissector_key_tcp *key =
1475                         skb_flow_dissector_target(f->dissector,
1476                                                   FLOW_DISSECTOR_KEY_TCP,
1477                                                   f->key);
1478                 struct flow_dissector_key_tcp *mask =
1479                         skb_flow_dissector_target(f->dissector,
1480                                                   FLOW_DISSECTOR_KEY_TCP,
1481                                                   f->mask);
1482
1483                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
1484                          ntohs(mask->flags));
1485                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
1486                          ntohs(key->flags));
1487
1488                 if (mask->flags)
1489                         *min_inline = MLX5_INLINE_MODE_TCP_UDP;
1490         }
1491
1492         return 0;
1493 }
1494
1495 static int parse_cls_flower(struct mlx5e_priv *priv,
1496                             struct mlx5e_tc_flow *flow,
1497                             struct mlx5_flow_spec *spec,
1498                             struct tc_cls_flower_offload *f)
1499 {
1500         struct mlx5_core_dev *dev = priv->mdev;
1501         struct mlx5_eswitch *esw = dev->priv.eswitch;
1502         struct mlx5e_rep_priv *rpriv = priv->ppriv;
1503         struct mlx5_eswitch_rep *rep;
1504         u8 min_inline;
1505         int err;
1506
1507         err = __parse_cls_flower(priv, spec, f, &min_inline);
1508
1509         if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH)) {
1510                 rep = rpriv->rep;
1511                 if (rep->vport != FDB_UPLINK_VPORT &&
1512                     (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
1513                     esw->offloads.inline_mode < min_inline)) {
1514                         netdev_warn(priv->netdev,
1515                                     "Flow is not offloaded due to min inline setting, required %d actual %d\n",
1516                                     min_inline, esw->offloads.inline_mode);
1517                         return -EOPNOTSUPP;
1518                 }
1519         }
1520
1521         return err;
1522 }
1523
1524 struct pedit_headers {
1525         struct ethhdr  eth;
1526         struct iphdr   ip4;
1527         struct ipv6hdr ip6;
1528         struct tcphdr  tcp;
1529         struct udphdr  udp;
1530 };
1531
1532 static int pedit_header_offsets[] = {
1533         [TCA_PEDIT_KEY_EX_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
1534         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
1535         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
1536         [TCA_PEDIT_KEY_EX_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
1537         [TCA_PEDIT_KEY_EX_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
1538 };
1539
1540 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
1541
1542 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
1543                          struct pedit_headers *masks,
1544                          struct pedit_headers *vals)
1545 {
1546         u32 *curr_pmask, *curr_pval;
1547
1548         if (hdr_type >= __PEDIT_HDR_TYPE_MAX)
1549                 goto out_err;
1550
1551         curr_pmask = (u32 *)(pedit_header(masks, hdr_type) + offset);
1552         curr_pval  = (u32 *)(pedit_header(vals, hdr_type) + offset);
1553
1554         if (*curr_pmask & mask)  /* disallow acting twice on the same location */
1555                 goto out_err;
1556
1557         *curr_pmask |= mask;
1558         *curr_pval  |= (val & mask);
1559
1560         return 0;
1561
1562 out_err:
1563         return -EOPNOTSUPP;
1564 }
1565
1566 struct mlx5_fields {
1567         u8  field;
1568         u8  size;
1569         u32 offset;
1570 };
1571
1572 #define OFFLOAD(fw_field, size, field, off) \
1573                 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, size, offsetof(struct pedit_headers, field) + (off)}
1574
1575 static struct mlx5_fields fields[] = {
1576         OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0),
1577         OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0),
1578         OFFLOAD(DMAC_15_0,  2, eth.h_dest[4], 0),
1579         OFFLOAD(SMAC_47_16, 4, eth.h_source[0], 0),
1580         OFFLOAD(SMAC_15_0,  2, eth.h_source[4], 0),
1581         OFFLOAD(ETHERTYPE,  2, eth.h_proto, 0),
1582
1583         OFFLOAD(IP_TTL, 1, ip4.ttl,   0),
1584         OFFLOAD(SIPV4,  4, ip4.saddr, 0),
1585         OFFLOAD(DIPV4,  4, ip4.daddr, 0),
1586
1587         OFFLOAD(SIPV6_127_96, 4, ip6.saddr.s6_addr32[0], 0),
1588         OFFLOAD(SIPV6_95_64,  4, ip6.saddr.s6_addr32[1], 0),
1589         OFFLOAD(SIPV6_63_32,  4, ip6.saddr.s6_addr32[2], 0),
1590         OFFLOAD(SIPV6_31_0,   4, ip6.saddr.s6_addr32[3], 0),
1591         OFFLOAD(DIPV6_127_96, 4, ip6.daddr.s6_addr32[0], 0),
1592         OFFLOAD(DIPV6_95_64,  4, ip6.daddr.s6_addr32[1], 0),
1593         OFFLOAD(DIPV6_63_32,  4, ip6.daddr.s6_addr32[2], 0),
1594         OFFLOAD(DIPV6_31_0,   4, ip6.daddr.s6_addr32[3], 0),
1595         OFFLOAD(IPV6_HOPLIMIT, 1, ip6.hop_limit, 0),
1596
1597         OFFLOAD(TCP_SPORT, 2, tcp.source,  0),
1598         OFFLOAD(TCP_DPORT, 2, tcp.dest,    0),
1599         OFFLOAD(TCP_FLAGS, 1, tcp.ack_seq, 5),
1600
1601         OFFLOAD(UDP_SPORT, 2, udp.source, 0),
1602         OFFLOAD(UDP_DPORT, 2, udp.dest,   0),
1603 };
1604
1605 /* On input attr->num_mod_hdr_actions tells how many HW actions can be parsed at
1606  * max from the SW pedit action. On success, it says how many HW actions were
1607  * actually parsed.
1608  */
1609 static int offload_pedit_fields(struct pedit_headers *masks,
1610                                 struct pedit_headers *vals,
1611                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
1612 {
1613         struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
1614         int i, action_size, nactions, max_actions, first, last, next_z;
1615         void *s_masks_p, *a_masks_p, *vals_p;
1616         struct mlx5_fields *f;
1617         u8 cmd, field_bsize;
1618         u32 s_mask, a_mask;
1619         unsigned long mask;
1620         __be32 mask_be32;
1621         __be16 mask_be16;
1622         void *action;
1623
1624         set_masks = &masks[TCA_PEDIT_KEY_EX_CMD_SET];
1625         add_masks = &masks[TCA_PEDIT_KEY_EX_CMD_ADD];
1626         set_vals = &vals[TCA_PEDIT_KEY_EX_CMD_SET];
1627         add_vals = &vals[TCA_PEDIT_KEY_EX_CMD_ADD];
1628
1629         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1630         action = parse_attr->mod_hdr_actions;
1631         max_actions = parse_attr->num_mod_hdr_actions;
1632         nactions = 0;
1633
1634         for (i = 0; i < ARRAY_SIZE(fields); i++) {
1635                 f = &fields[i];
1636                 /* avoid seeing bits set from previous iterations */
1637                 s_mask = 0;
1638                 a_mask = 0;
1639
1640                 s_masks_p = (void *)set_masks + f->offset;
1641                 a_masks_p = (void *)add_masks + f->offset;
1642
1643                 memcpy(&s_mask, s_masks_p, f->size);
1644                 memcpy(&a_mask, a_masks_p, f->size);
1645
1646                 if (!s_mask && !a_mask) /* nothing to offload here */
1647                         continue;
1648
1649                 if (s_mask && a_mask) {
1650                         printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
1651                         return -EOPNOTSUPP;
1652                 }
1653
1654                 if (nactions == max_actions) {
1655                         printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
1656                         return -EOPNOTSUPP;
1657                 }
1658
1659                 if (s_mask) {
1660                         cmd  = MLX5_ACTION_TYPE_SET;
1661                         mask = s_mask;
1662                         vals_p = (void *)set_vals + f->offset;
1663                         /* clear to denote we consumed this field */
1664                         memset(s_masks_p, 0, f->size);
1665                 } else {
1666                         cmd  = MLX5_ACTION_TYPE_ADD;
1667                         mask = a_mask;
1668                         vals_p = (void *)add_vals + f->offset;
1669                         /* clear to denote we consumed this field */
1670                         memset(a_masks_p, 0, f->size);
1671                 }
1672
1673                 field_bsize = f->size * BITS_PER_BYTE;
1674
1675                 if (field_bsize == 32) {
1676                         mask_be32 = *(__be32 *)&mask;
1677                         mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
1678                 } else if (field_bsize == 16) {
1679                         mask_be16 = *(__be16 *)&mask;
1680                         mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
1681                 }
1682
1683                 first = find_first_bit(&mask, field_bsize);
1684                 next_z = find_next_zero_bit(&mask, field_bsize, first);
1685                 last  = find_last_bit(&mask, field_bsize);
1686                 if (first < next_z && next_z < last) {
1687                         printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
1688                                mask);
1689                         return -EOPNOTSUPP;
1690                 }
1691
1692                 MLX5_SET(set_action_in, action, action_type, cmd);
1693                 MLX5_SET(set_action_in, action, field, f->field);
1694
1695                 if (cmd == MLX5_ACTION_TYPE_SET) {
1696                         MLX5_SET(set_action_in, action, offset, first);
1697                         /* length is num of bits to be written, zero means length of 32 */
1698                         MLX5_SET(set_action_in, action, length, (last - first + 1));
1699                 }
1700
1701                 if (field_bsize == 32)
1702                         MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
1703                 else if (field_bsize == 16)
1704                         MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
1705                 else if (field_bsize == 8)
1706                         MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
1707
1708                 action += action_size;
1709                 nactions++;
1710         }
1711
1712         parse_attr->num_mod_hdr_actions = nactions;
1713         return 0;
1714 }
1715
1716 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
1717                                  const struct tc_action *a, int namespace,
1718                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1719 {
1720         int nkeys, action_size, max_actions;
1721
1722         nkeys = tcf_pedit_nkeys(a);
1723         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1724
1725         if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
1726                 max_actions = MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev, max_modify_header_actions);
1727         else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
1728                 max_actions = MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, max_modify_header_actions);
1729
1730         /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
1731         max_actions = min(max_actions, nkeys * 16);
1732
1733         parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
1734         if (!parse_attr->mod_hdr_actions)
1735                 return -ENOMEM;
1736
1737         parse_attr->num_mod_hdr_actions = max_actions;
1738         return 0;
1739 }
1740
1741 static const struct pedit_headers zero_masks = {};
1742
1743 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
1744                                  const struct tc_action *a, int namespace,
1745                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1746 {
1747         struct pedit_headers masks[__PEDIT_CMD_MAX], vals[__PEDIT_CMD_MAX], *cmd_masks;
1748         int nkeys, i, err = -EOPNOTSUPP;
1749         u32 mask, val, offset;
1750         u8 cmd, htype;
1751
1752         nkeys = tcf_pedit_nkeys(a);
1753
1754         memset(masks, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1755         memset(vals,  0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1756
1757         for (i = 0; i < nkeys; i++) {
1758                 htype = tcf_pedit_htype(a, i);
1759                 cmd = tcf_pedit_cmd(a, i);
1760                 err = -EOPNOTSUPP; /* can't be all optimistic */
1761
1762                 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK) {
1763                         printk(KERN_WARNING "mlx5: legacy pedit isn't offloaded\n");
1764                         goto out_err;
1765                 }
1766
1767                 if (cmd != TCA_PEDIT_KEY_EX_CMD_SET && cmd != TCA_PEDIT_KEY_EX_CMD_ADD) {
1768                         printk(KERN_WARNING "mlx5: pedit cmd %d isn't offloaded\n", cmd);
1769                         goto out_err;
1770                 }
1771
1772                 mask = tcf_pedit_mask(a, i);
1773                 val = tcf_pedit_val(a, i);
1774                 offset = tcf_pedit_offset(a, i);
1775
1776                 err = set_pedit_val(htype, ~mask, val, offset, &masks[cmd], &vals[cmd]);
1777                 if (err)
1778                         goto out_err;
1779         }
1780
1781         err = alloc_mod_hdr_actions(priv, a, namespace, parse_attr);
1782         if (err)
1783                 goto out_err;
1784
1785         err = offload_pedit_fields(masks, vals, parse_attr);
1786         if (err < 0)
1787                 goto out_dealloc_parsed_actions;
1788
1789         for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
1790                 cmd_masks = &masks[cmd];
1791                 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
1792                         printk(KERN_WARNING "mlx5: attempt to offload an unsupported field (cmd %d)\n",
1793                                cmd);
1794                         print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
1795                                        16, 1, cmd_masks, sizeof(zero_masks), true);
1796                         err = -EOPNOTSUPP;
1797                         goto out_dealloc_parsed_actions;
1798                 }
1799         }
1800
1801         return 0;
1802
1803 out_dealloc_parsed_actions:
1804         kfree(parse_attr->mod_hdr_actions);
1805 out_err:
1806         return err;
1807 }
1808
1809 static bool csum_offload_supported(struct mlx5e_priv *priv, u32 action, u32 update_flags)
1810 {
1811         u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
1812                          TCA_CSUM_UPDATE_FLAG_UDP;
1813
1814         /*  The HW recalcs checksums only if re-writing headers */
1815         if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
1816                 netdev_warn(priv->netdev,
1817                             "TC csum action is only offloaded with pedit\n");
1818                 return false;
1819         }
1820
1821         if (update_flags & ~prot_flags) {
1822                 netdev_warn(priv->netdev,
1823                             "can't offload TC csum action for some header/s - flags %#x\n",
1824                             update_flags);
1825                 return false;
1826         }
1827
1828         return true;
1829 }
1830
1831 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
1832                                           struct tcf_exts *exts)
1833 {
1834         const struct tc_action *a;
1835         bool modify_ip_header;
1836         LIST_HEAD(actions);
1837         u8 htype, ip_proto;
1838         void *headers_v;
1839         u16 ethertype;
1840         int nkeys, i;
1841
1842         headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
1843         ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
1844
1845         /* for non-IP we only re-write MACs, so we're okay */
1846         if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
1847                 goto out_ok;
1848
1849         modify_ip_header = false;
1850         tcf_exts_to_list(exts, &actions);
1851         list_for_each_entry(a, &actions, list) {
1852                 if (!is_tcf_pedit(a))
1853                         continue;
1854
1855                 nkeys = tcf_pedit_nkeys(a);
1856                 for (i = 0; i < nkeys; i++) {
1857                         htype = tcf_pedit_htype(a, i);
1858                         if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP4 ||
1859                             htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP6) {
1860                                 modify_ip_header = true;
1861                                 break;
1862                         }
1863                 }
1864         }
1865
1866         ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
1867         if (modify_ip_header && ip_proto != IPPROTO_TCP && ip_proto != IPPROTO_UDP) {
1868                 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
1869                 return false;
1870         }
1871
1872 out_ok:
1873         return true;
1874 }
1875
1876 static bool actions_match_supported(struct mlx5e_priv *priv,
1877                                     struct tcf_exts *exts,
1878                                     struct mlx5e_tc_flow_parse_attr *parse_attr,
1879                                     struct mlx5e_tc_flow *flow)
1880 {
1881         u32 actions;
1882
1883         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1884                 actions = flow->esw_attr->action;
1885         else
1886                 actions = flow->nic_attr->action;
1887
1888         if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1889                 return modify_header_match_supported(&parse_attr->spec, exts);
1890
1891         return true;
1892 }
1893
1894 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
1895 {
1896         struct mlx5_core_dev *fmdev, *pmdev;
1897         u16 func_id, peer_id;
1898
1899         fmdev = priv->mdev;
1900         pmdev = peer_priv->mdev;
1901
1902         func_id = (u16)((fmdev->pdev->bus->number << 8) | PCI_SLOT(fmdev->pdev->devfn));
1903         peer_id = (u16)((pmdev->pdev->bus->number << 8) | PCI_SLOT(pmdev->pdev->devfn));
1904
1905         return (func_id == peer_id);
1906 }
1907
1908 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
1909                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
1910                                 struct mlx5e_tc_flow *flow)
1911 {
1912         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
1913         const struct tc_action *a;
1914         LIST_HEAD(actions);
1915         int err;
1916
1917         if (!tcf_exts_has_actions(exts))
1918                 return -EINVAL;
1919
1920         attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
1921         attr->action = 0;
1922
1923         tcf_exts_to_list(exts, &actions);
1924         list_for_each_entry(a, &actions, list) {
1925                 if (is_tcf_gact_shot(a)) {
1926                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
1927                         if (MLX5_CAP_FLOWTABLE(priv->mdev,
1928                                                flow_table_properties_nic_receive.flow_counter))
1929                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
1930                         continue;
1931                 }
1932
1933                 if (is_tcf_pedit(a)) {
1934                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_KERNEL,
1935                                                     parse_attr);
1936                         if (err)
1937                                 return err;
1938
1939                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
1940                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1941                         continue;
1942                 }
1943
1944                 if (is_tcf_csum(a)) {
1945                         if (csum_offload_supported(priv, attr->action,
1946                                                    tcf_csum_update_flags(a)))
1947                                 continue;
1948
1949                         return -EOPNOTSUPP;
1950                 }
1951
1952                 if (is_tcf_mirred_egress_redirect(a)) {
1953                         struct net_device *peer_dev = tcf_mirred_dev(a);
1954
1955                         if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
1956                             same_hw_devs(priv, netdev_priv(peer_dev))) {
1957                                 parse_attr->mirred_ifindex = peer_dev->ifindex;
1958                                 flow->flags |= MLX5E_TC_FLOW_HAIRPIN;
1959                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1960                                                 MLX5_FLOW_CONTEXT_ACTION_COUNT;
1961                         } else {
1962                                 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
1963                                             peer_dev->name);
1964                                 return -EINVAL;
1965                         }
1966                         continue;
1967                 }
1968
1969                 if (is_tcf_skbedit_mark(a)) {
1970                         u32 mark = tcf_skbedit_mark(a);
1971
1972                         if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
1973                                 netdev_warn(priv->netdev, "Bad flow mark - only 16 bit is supported: 0x%x\n",
1974                                             mark);
1975                                 return -EINVAL;
1976                         }
1977
1978                         attr->flow_tag = mark;
1979                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1980                         continue;
1981                 }
1982
1983                 return -EINVAL;
1984         }
1985
1986         if (!actions_match_supported(priv, exts, parse_attr, flow))
1987                 return -EOPNOTSUPP;
1988
1989         return 0;
1990 }
1991
1992 static inline int cmp_encap_info(struct ip_tunnel_key *a,
1993                                  struct ip_tunnel_key *b)
1994 {
1995         return memcmp(a, b, sizeof(*a));
1996 }
1997
1998 static inline int hash_encap_info(struct ip_tunnel_key *key)
1999 {
2000         return jhash(key, sizeof(*key), 0);
2001 }
2002
2003 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
2004                                    struct net_device *mirred_dev,
2005                                    struct net_device **out_dev,
2006                                    struct flowi4 *fl4,
2007                                    struct neighbour **out_n,
2008                                    int *out_ttl)
2009 {
2010         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2011         struct mlx5e_rep_priv *uplink_rpriv;
2012         struct rtable *rt;
2013         struct neighbour *n = NULL;
2014
2015 #if IS_ENABLED(CONFIG_INET)
2016         int ret;
2017
2018         rt = ip_route_output_key(dev_net(mirred_dev), fl4);
2019         ret = PTR_ERR_OR_ZERO(rt);
2020         if (ret)
2021                 return ret;
2022 #else
2023         return -EOPNOTSUPP;
2024 #endif
2025         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2026         /* if the egress device isn't on the same HW e-switch, we use the uplink */
2027         if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
2028                 *out_dev = uplink_rpriv->netdev;
2029         else
2030                 *out_dev = rt->dst.dev;
2031
2032         *out_ttl = ip4_dst_hoplimit(&rt->dst);
2033         n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
2034         ip_rt_put(rt);
2035         if (!n)
2036                 return -ENOMEM;
2037
2038         *out_n = n;
2039         return 0;
2040 }
2041
2042 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
2043                                    struct net_device *mirred_dev,
2044                                    struct net_device **out_dev,
2045                                    struct flowi6 *fl6,
2046                                    struct neighbour **out_n,
2047                                    int *out_ttl)
2048 {
2049         struct neighbour *n = NULL;
2050         struct dst_entry *dst;
2051
2052 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
2053         struct mlx5e_rep_priv *uplink_rpriv;
2054         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2055         int ret;
2056
2057         ret = ipv6_stub->ipv6_dst_lookup(dev_net(mirred_dev), NULL, &dst,
2058                                          fl6);
2059         if (ret < 0)
2060                 return ret;
2061
2062         *out_ttl = ip6_dst_hoplimit(dst);
2063
2064         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2065         /* if the egress device isn't on the same HW e-switch, we use the uplink */
2066         if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
2067                 *out_dev = uplink_rpriv->netdev;
2068         else
2069                 *out_dev = dst->dev;
2070 #else
2071         return -EOPNOTSUPP;
2072 #endif
2073
2074         n = dst_neigh_lookup(dst, &fl6->daddr);
2075         dst_release(dst);
2076         if (!n)
2077                 return -ENOMEM;
2078
2079         *out_n = n;
2080         return 0;
2081 }
2082
2083 static void gen_vxlan_header_ipv4(struct net_device *out_dev,
2084                                   char buf[], int encap_size,
2085                                   unsigned char h_dest[ETH_ALEN],
2086                                   int ttl,
2087                                   __be32 daddr,
2088                                   __be32 saddr,
2089                                   __be16 udp_dst_port,
2090                                   __be32 vx_vni)
2091 {
2092         struct ethhdr *eth = (struct ethhdr *)buf;
2093         struct iphdr  *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
2094         struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
2095         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2096
2097         memset(buf, 0, encap_size);
2098
2099         ether_addr_copy(eth->h_dest, h_dest);
2100         ether_addr_copy(eth->h_source, out_dev->dev_addr);
2101         eth->h_proto = htons(ETH_P_IP);
2102
2103         ip->daddr = daddr;
2104         ip->saddr = saddr;
2105
2106         ip->ttl = ttl;
2107         ip->protocol = IPPROTO_UDP;
2108         ip->version = 0x4;
2109         ip->ihl = 0x5;
2110
2111         udp->dest = udp_dst_port;
2112         vxh->vx_flags = VXLAN_HF_VNI;
2113         vxh->vx_vni = vxlan_vni_field(vx_vni);
2114 }
2115
2116 static void gen_vxlan_header_ipv6(struct net_device *out_dev,
2117                                   char buf[], int encap_size,
2118                                   unsigned char h_dest[ETH_ALEN],
2119                                   int ttl,
2120                                   struct in6_addr *daddr,
2121                                   struct in6_addr *saddr,
2122                                   __be16 udp_dst_port,
2123                                   __be32 vx_vni)
2124 {
2125         struct ethhdr *eth = (struct ethhdr *)buf;
2126         struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
2127         struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
2128         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2129
2130         memset(buf, 0, encap_size);
2131
2132         ether_addr_copy(eth->h_dest, h_dest);
2133         ether_addr_copy(eth->h_source, out_dev->dev_addr);
2134         eth->h_proto = htons(ETH_P_IPV6);
2135
2136         ip6_flow_hdr(ip6h, 0, 0);
2137         /* the HW fills up ipv6 payload len */
2138         ip6h->nexthdr     = IPPROTO_UDP;
2139         ip6h->hop_limit   = ttl;
2140         ip6h->daddr       = *daddr;
2141         ip6h->saddr       = *saddr;
2142
2143         udp->dest = udp_dst_port;
2144         vxh->vx_flags = VXLAN_HF_VNI;
2145         vxh->vx_vni = vxlan_vni_field(vx_vni);
2146 }
2147
2148 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
2149                                           struct net_device *mirred_dev,
2150                                           struct mlx5e_encap_entry *e)
2151 {
2152         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2153         int ipv4_encap_size = ETH_HLEN + sizeof(struct iphdr) + VXLAN_HLEN;
2154         struct ip_tunnel_key *tun_key = &e->tun_info.key;
2155         struct net_device *out_dev;
2156         struct neighbour *n = NULL;
2157         struct flowi4 fl4 = {};
2158         char *encap_header;
2159         int ttl, err;
2160         u8 nud_state;
2161
2162         if (max_encap_size < ipv4_encap_size) {
2163                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2164                                ipv4_encap_size, max_encap_size);
2165                 return -EOPNOTSUPP;
2166         }
2167
2168         encap_header = kzalloc(ipv4_encap_size, GFP_KERNEL);
2169         if (!encap_header)
2170                 return -ENOMEM;
2171
2172         switch (e->tunnel_type) {
2173         case MLX5_HEADER_TYPE_VXLAN:
2174                 fl4.flowi4_proto = IPPROTO_UDP;
2175                 fl4.fl4_dport = tun_key->tp_dst;
2176                 break;
2177         default:
2178                 err = -EOPNOTSUPP;
2179                 goto free_encap;
2180         }
2181         fl4.flowi4_tos = tun_key->tos;
2182         fl4.daddr = tun_key->u.ipv4.dst;
2183         fl4.saddr = tun_key->u.ipv4.src;
2184
2185         err = mlx5e_route_lookup_ipv4(priv, mirred_dev, &out_dev,
2186                                       &fl4, &n, &ttl);
2187         if (err)
2188                 goto free_encap;
2189
2190         /* used by mlx5e_detach_encap to lookup a neigh hash table
2191          * entry in the neigh hash table when a user deletes a rule
2192          */
2193         e->m_neigh.dev = n->dev;
2194         e->m_neigh.family = n->ops->family;
2195         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2196         e->out_dev = out_dev;
2197
2198         /* It's importent to add the neigh to the hash table before checking
2199          * the neigh validity state. So if we'll get a notification, in case the
2200          * neigh changes it's validity state, we would find the relevant neigh
2201          * in the hash.
2202          */
2203         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2204         if (err)
2205                 goto free_encap;
2206
2207         read_lock_bh(&n->lock);
2208         nud_state = n->nud_state;
2209         ether_addr_copy(e->h_dest, n->ha);
2210         read_unlock_bh(&n->lock);
2211
2212         switch (e->tunnel_type) {
2213         case MLX5_HEADER_TYPE_VXLAN:
2214                 gen_vxlan_header_ipv4(out_dev, encap_header,
2215                                       ipv4_encap_size, e->h_dest, ttl,
2216                                       fl4.daddr,
2217                                       fl4.saddr, tun_key->tp_dst,
2218                                       tunnel_id_to_key32(tun_key->tun_id));
2219                 break;
2220         default:
2221                 err = -EOPNOTSUPP;
2222                 goto destroy_neigh_entry;
2223         }
2224         e->encap_size = ipv4_encap_size;
2225         e->encap_header = encap_header;
2226
2227         if (!(nud_state & NUD_VALID)) {
2228                 neigh_event_send(n, NULL);
2229                 err = -EAGAIN;
2230                 goto out;
2231         }
2232
2233         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
2234                                ipv4_encap_size, encap_header, &e->encap_id);
2235         if (err)
2236                 goto destroy_neigh_entry;
2237
2238         e->flags |= MLX5_ENCAP_ENTRY_VALID;
2239         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2240         neigh_release(n);
2241         return err;
2242
2243 destroy_neigh_entry:
2244         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2245 free_encap:
2246         kfree(encap_header);
2247 out:
2248         if (n)
2249                 neigh_release(n);
2250         return err;
2251 }
2252
2253 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
2254                                           struct net_device *mirred_dev,
2255                                           struct mlx5e_encap_entry *e)
2256 {
2257         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2258         int ipv6_encap_size = ETH_HLEN + sizeof(struct ipv6hdr) + VXLAN_HLEN;
2259         struct ip_tunnel_key *tun_key = &e->tun_info.key;
2260         struct net_device *out_dev;
2261         struct neighbour *n = NULL;
2262         struct flowi6 fl6 = {};
2263         char *encap_header;
2264         int err, ttl = 0;
2265         u8 nud_state;
2266
2267         if (max_encap_size < ipv6_encap_size) {
2268                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2269                                ipv6_encap_size, max_encap_size);
2270                 return -EOPNOTSUPP;
2271         }
2272
2273         encap_header = kzalloc(ipv6_encap_size, GFP_KERNEL);
2274         if (!encap_header)
2275                 return -ENOMEM;
2276
2277         switch (e->tunnel_type) {
2278         case MLX5_HEADER_TYPE_VXLAN:
2279                 fl6.flowi6_proto = IPPROTO_UDP;
2280                 fl6.fl6_dport = tun_key->tp_dst;
2281                 break;
2282         default:
2283                 err = -EOPNOTSUPP;
2284                 goto free_encap;
2285         }
2286
2287         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
2288         fl6.daddr = tun_key->u.ipv6.dst;
2289         fl6.saddr = tun_key->u.ipv6.src;
2290
2291         err = mlx5e_route_lookup_ipv6(priv, mirred_dev, &out_dev,
2292                                       &fl6, &n, &ttl);
2293         if (err)
2294                 goto free_encap;
2295
2296         /* used by mlx5e_detach_encap to lookup a neigh hash table
2297          * entry in the neigh hash table when a user deletes a rule
2298          */
2299         e->m_neigh.dev = n->dev;
2300         e->m_neigh.family = n->ops->family;
2301         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2302         e->out_dev = out_dev;
2303
2304         /* It's importent to add the neigh to the hash table before checking
2305          * the neigh validity state. So if we'll get a notification, in case the
2306          * neigh changes it's validity state, we would find the relevant neigh
2307          * in the hash.
2308          */
2309         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2310         if (err)
2311                 goto free_encap;
2312
2313         read_lock_bh(&n->lock);
2314         nud_state = n->nud_state;
2315         ether_addr_copy(e->h_dest, n->ha);
2316         read_unlock_bh(&n->lock);
2317
2318         switch (e->tunnel_type) {
2319         case MLX5_HEADER_TYPE_VXLAN:
2320                 gen_vxlan_header_ipv6(out_dev, encap_header,
2321                                       ipv6_encap_size, e->h_dest, ttl,
2322                                       &fl6.daddr,
2323                                       &fl6.saddr, tun_key->tp_dst,
2324                                       tunnel_id_to_key32(tun_key->tun_id));
2325                 break;
2326         default:
2327                 err = -EOPNOTSUPP;
2328                 goto destroy_neigh_entry;
2329         }
2330
2331         e->encap_size = ipv6_encap_size;
2332         e->encap_header = encap_header;
2333
2334         if (!(nud_state & NUD_VALID)) {
2335                 neigh_event_send(n, NULL);
2336                 err = -EAGAIN;
2337                 goto out;
2338         }
2339
2340         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
2341                                ipv6_encap_size, encap_header, &e->encap_id);
2342         if (err)
2343                 goto destroy_neigh_entry;
2344
2345         e->flags |= MLX5_ENCAP_ENTRY_VALID;
2346         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2347         neigh_release(n);
2348         return err;
2349
2350 destroy_neigh_entry:
2351         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2352 free_encap:
2353         kfree(encap_header);
2354 out:
2355         if (n)
2356                 neigh_release(n);
2357         return err;
2358 }
2359
2360 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
2361                               struct ip_tunnel_info *tun_info,
2362                               struct net_device *mirred_dev,
2363                               struct net_device **encap_dev,
2364                               struct mlx5e_tc_flow *flow)
2365 {
2366         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2367         struct mlx5e_rep_priv *uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw,
2368                                                                            REP_ETH);
2369         struct net_device *up_dev = uplink_rpriv->netdev;
2370         unsigned short family = ip_tunnel_info_af(tun_info);
2371         struct mlx5e_priv *up_priv = netdev_priv(up_dev);
2372         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2373         struct ip_tunnel_key *key = &tun_info->key;
2374         struct mlx5e_encap_entry *e;
2375         int tunnel_type, err = 0;
2376         uintptr_t hash_key;
2377         bool found = false;
2378
2379         /* udp dst port must be set */
2380         if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
2381                 goto vxlan_encap_offload_err;
2382
2383         /* setting udp src port isn't supported */
2384         if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
2385 vxlan_encap_offload_err:
2386                 netdev_warn(priv->netdev,
2387                             "must set udp dst port and not set udp src port\n");
2388                 return -EOPNOTSUPP;
2389         }
2390
2391         if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->tp_dst)) &&
2392             MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
2393                 tunnel_type = MLX5_HEADER_TYPE_VXLAN;
2394         } else {
2395                 netdev_warn(priv->netdev,
2396                             "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
2397                 return -EOPNOTSUPP;
2398         }
2399
2400         hash_key = hash_encap_info(key);
2401
2402         hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
2403                                    encap_hlist, hash_key) {
2404                 if (!cmp_encap_info(&e->tun_info.key, key)) {
2405                         found = true;
2406                         break;
2407                 }
2408         }
2409
2410         /* must verify if encap is valid or not */
2411         if (found)
2412                 goto attach_flow;
2413
2414         e = kzalloc(sizeof(*e), GFP_KERNEL);
2415         if (!e)
2416                 return -ENOMEM;
2417
2418         e->tun_info = *tun_info;
2419         e->tunnel_type = tunnel_type;
2420         INIT_LIST_HEAD(&e->flows);
2421
2422         if (family == AF_INET)
2423                 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e);
2424         else if (family == AF_INET6)
2425                 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e);
2426
2427         if (err && err != -EAGAIN)
2428                 goto out_err;
2429
2430         hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
2431
2432 attach_flow:
2433         list_add(&flow->encap, &e->flows);
2434         *encap_dev = e->out_dev;
2435         if (e->flags & MLX5_ENCAP_ENTRY_VALID)
2436                 attr->encap_id = e->encap_id;
2437         else
2438                 err = -EAGAIN;
2439
2440         return err;
2441
2442 out_err:
2443         kfree(e);
2444         return err;
2445 }
2446
2447 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
2448                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
2449                                 struct mlx5e_tc_flow *flow)
2450 {
2451         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2452         struct mlx5e_rep_priv *rpriv = priv->ppriv;
2453         struct ip_tunnel_info *info = NULL;
2454         const struct tc_action *a;
2455         LIST_HEAD(actions);
2456         bool encap = false;
2457         int err = 0;
2458
2459         if (!tcf_exts_has_actions(exts))
2460                 return -EINVAL;
2461
2462         memset(attr, 0, sizeof(*attr));
2463         attr->in_rep = rpriv->rep;
2464
2465         tcf_exts_to_list(exts, &actions);
2466         list_for_each_entry(a, &actions, list) {
2467                 if (is_tcf_gact_shot(a)) {
2468                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
2469                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
2470                         continue;
2471                 }
2472
2473                 if (is_tcf_pedit(a)) {
2474                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_FDB,
2475                                                     parse_attr);
2476                         if (err)
2477                                 return err;
2478
2479                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2480                         continue;
2481                 }
2482
2483                 if (is_tcf_csum(a)) {
2484                         if (csum_offload_supported(priv, attr->action,
2485                                                    tcf_csum_update_flags(a)))
2486                                 continue;
2487
2488                         return -EOPNOTSUPP;
2489                 }
2490
2491                 if (is_tcf_mirred_egress_redirect(a)) {
2492                         struct net_device *out_dev;
2493                         struct mlx5e_priv *out_priv;
2494
2495                         out_dev = tcf_mirred_dev(a);
2496
2497                         if (switchdev_port_same_parent_id(priv->netdev,
2498                                                           out_dev)) {
2499                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2500                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
2501                                 out_priv = netdev_priv(out_dev);
2502                                 rpriv = out_priv->ppriv;
2503                                 attr->out_rep = rpriv->rep;
2504                         } else if (encap) {
2505                                 parse_attr->mirred_ifindex = out_dev->ifindex;
2506                                 parse_attr->tun_info = *info;
2507                                 attr->parse_attr = parse_attr;
2508                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_ENCAP |
2509                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2510                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
2511                                 /* attr->out_rep is resolved when we handle encap */
2512                         } else {
2513                                 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
2514                                        priv->netdev->name, out_dev->name);
2515                                 return -EINVAL;
2516                         }
2517                         continue;
2518                 }
2519
2520                 if (is_tcf_tunnel_set(a)) {
2521                         info = tcf_tunnel_info(a);
2522                         if (info)
2523                                 encap = true;
2524                         else
2525                                 return -EOPNOTSUPP;
2526                         continue;
2527                 }
2528
2529                 if (is_tcf_vlan(a)) {
2530                         if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
2531                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2532                         } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
2533                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
2534                                 attr->vlan_vid = tcf_vlan_push_vid(a);
2535                                 if (mlx5_eswitch_vlan_actions_supported(priv->mdev)) {
2536                                         attr->vlan_prio = tcf_vlan_push_prio(a);
2537                                         attr->vlan_proto = tcf_vlan_push_proto(a);
2538                                         if (!attr->vlan_proto)
2539                                                 attr->vlan_proto = htons(ETH_P_8021Q);
2540                                 } else if (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q) ||
2541                                            tcf_vlan_push_prio(a)) {
2542                                         return -EOPNOTSUPP;
2543                                 }
2544                         } else { /* action is TCA_VLAN_ACT_MODIFY */
2545                                 return -EOPNOTSUPP;
2546                         }
2547                         continue;
2548                 }
2549
2550                 if (is_tcf_tunnel_release(a)) {
2551                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
2552                         continue;
2553                 }
2554
2555                 return -EINVAL;
2556         }
2557
2558         if (!actions_match_supported(priv, exts, parse_attr, flow))
2559                 return -EOPNOTSUPP;
2560
2561         return err;
2562 }
2563
2564 int mlx5e_configure_flower(struct mlx5e_priv *priv,
2565                            struct tc_cls_flower_offload *f)
2566 {
2567         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2568         struct mlx5e_tc_flow_parse_attr *parse_attr;
2569         struct mlx5e_tc_table *tc = &priv->fs.tc;
2570         struct mlx5e_tc_flow *flow;
2571         int attr_size, err = 0;
2572         u8 flow_flags = 0;
2573
2574         if (esw && esw->mode == SRIOV_OFFLOADS) {
2575                 flow_flags = MLX5E_TC_FLOW_ESWITCH;
2576                 attr_size  = sizeof(struct mlx5_esw_flow_attr);
2577         } else {
2578                 flow_flags = MLX5E_TC_FLOW_NIC;
2579                 attr_size  = sizeof(struct mlx5_nic_flow_attr);
2580         }
2581
2582         flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
2583         parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
2584         if (!parse_attr || !flow) {
2585                 err = -ENOMEM;
2586                 goto err_free;
2587         }
2588
2589         flow->cookie = f->cookie;
2590         flow->flags = flow_flags;
2591
2592         err = parse_cls_flower(priv, flow, &parse_attr->spec, f);
2593         if (err < 0)
2594                 goto err_free;
2595
2596         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
2597                 err = parse_tc_fdb_actions(priv, f->exts, parse_attr, flow);
2598                 if (err < 0)
2599                         goto err_free;
2600                 flow->rule = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow);
2601         } else {
2602                 err = parse_tc_nic_actions(priv, f->exts, parse_attr, flow);
2603                 if (err < 0)
2604                         goto err_free;
2605                 flow->rule = mlx5e_tc_add_nic_flow(priv, parse_attr, flow);
2606         }
2607
2608         if (IS_ERR(flow->rule)) {
2609                 err = PTR_ERR(flow->rule);
2610                 if (err != -EAGAIN)
2611                         goto err_free;
2612         }
2613
2614         if (err != -EAGAIN)
2615                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
2616
2617         err = rhashtable_insert_fast(&tc->ht, &flow->node,
2618                                      tc->ht_params);
2619         if (err)
2620                 goto err_del_rule;
2621
2622         if (flow->flags & MLX5E_TC_FLOW_ESWITCH &&
2623             !(flow->esw_attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP))
2624                 kvfree(parse_attr);
2625         return err;
2626
2627 err_del_rule:
2628         mlx5e_tc_del_flow(priv, flow);
2629
2630 err_free:
2631         kvfree(parse_attr);
2632         kfree(flow);
2633         return err;
2634 }
2635
2636 int mlx5e_delete_flower(struct mlx5e_priv *priv,
2637                         struct tc_cls_flower_offload *f)
2638 {
2639         struct mlx5e_tc_flow *flow;
2640         struct mlx5e_tc_table *tc = &priv->fs.tc;
2641
2642         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
2643                                       tc->ht_params);
2644         if (!flow)
2645                 return -EINVAL;
2646
2647         rhashtable_remove_fast(&tc->ht, &flow->node, tc->ht_params);
2648
2649         mlx5e_tc_del_flow(priv, flow);
2650
2651         kfree(flow);
2652
2653         return 0;
2654 }
2655
2656 int mlx5e_stats_flower(struct mlx5e_priv *priv,
2657                        struct tc_cls_flower_offload *f)
2658 {
2659         struct mlx5e_tc_table *tc = &priv->fs.tc;
2660         struct mlx5e_tc_flow *flow;
2661         struct mlx5_fc *counter;
2662         u64 bytes;
2663         u64 packets;
2664         u64 lastuse;
2665
2666         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
2667                                       tc->ht_params);
2668         if (!flow)
2669                 return -EINVAL;
2670
2671         if (!(flow->flags & MLX5E_TC_FLOW_OFFLOADED))
2672                 return 0;
2673
2674         counter = mlx5_flow_rule_counter(flow->rule);
2675         if (!counter)
2676                 return 0;
2677
2678         mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
2679
2680         tcf_exts_stats_update(f->exts, bytes, packets, lastuse);
2681
2682         return 0;
2683 }
2684
2685 static const struct rhashtable_params mlx5e_tc_flow_ht_params = {
2686         .head_offset = offsetof(struct mlx5e_tc_flow, node),
2687         .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
2688         .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
2689         .automatic_shrinking = true,
2690 };
2691
2692 int mlx5e_tc_init(struct mlx5e_priv *priv)
2693 {
2694         struct mlx5e_tc_table *tc = &priv->fs.tc;
2695
2696         hash_init(tc->mod_hdr_tbl);
2697         hash_init(tc->hairpin_tbl);
2698
2699         tc->ht_params = mlx5e_tc_flow_ht_params;
2700         return rhashtable_init(&tc->ht, &tc->ht_params);
2701 }
2702
2703 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
2704 {
2705         struct mlx5e_tc_flow *flow = ptr;
2706         struct mlx5e_priv *priv = arg;
2707
2708         mlx5e_tc_del_flow(priv, flow);
2709         kfree(flow);
2710 }
2711
2712 void mlx5e_tc_cleanup(struct mlx5e_priv *priv)
2713 {
2714         struct mlx5e_tc_table *tc = &priv->fs.tc;
2715
2716         rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, priv);
2717
2718         if (!IS_ERR_OR_NULL(tc->t)) {
2719                 mlx5_destroy_flow_table(tc->t);
2720                 tc->t = NULL;
2721         }
2722 }