net/mlx5e: Return success when TC offloaded fdb actions parsed ok
[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 BIT(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 = &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                 if (neigh_used)
986                         break;
987         }
988
989         if (neigh_used) {
990                 nhe->reported_lastuse = jiffies;
991
992                 /* find the relevant neigh according to the cached device and
993                  * dst ip pair
994                  */
995                 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
996                 if (!n) {
997                         WARN(1, "The neighbour already freed\n");
998                         return;
999                 }
1000
1001                 neigh_event_send(n, NULL);
1002                 neigh_release(n);
1003         }
1004 }
1005
1006 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1007                                struct mlx5e_tc_flow *flow)
1008 {
1009         struct list_head *next = flow->encap.next;
1010
1011         list_del(&flow->encap);
1012         if (list_empty(next)) {
1013                 struct mlx5e_encap_entry *e;
1014
1015                 e = list_entry(next, struct mlx5e_encap_entry, flows);
1016                 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1017
1018                 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1019                         mlx5_encap_dealloc(priv->mdev, e->encap_id);
1020
1021                 hash_del_rcu(&e->encap_hlist);
1022                 kfree(e->encap_header);
1023                 kfree(e);
1024         }
1025 }
1026
1027 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1028                               struct mlx5e_tc_flow *flow)
1029 {
1030         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1031                 mlx5e_tc_del_fdb_flow(priv, flow);
1032         else
1033                 mlx5e_tc_del_nic_flow(priv, flow);
1034 }
1035
1036 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
1037                              struct tc_cls_flower_offload *f)
1038 {
1039         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1040                                        outer_headers);
1041         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1042                                        outer_headers);
1043         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1044                                     misc_parameters);
1045         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1046                                     misc_parameters);
1047
1048         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
1049         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
1050
1051         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
1052                 struct flow_dissector_key_keyid *key =
1053                         skb_flow_dissector_target(f->dissector,
1054                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
1055                                                   f->key);
1056                 struct flow_dissector_key_keyid *mask =
1057                         skb_flow_dissector_target(f->dissector,
1058                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
1059                                                   f->mask);
1060                 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
1061                          be32_to_cpu(mask->keyid));
1062                 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
1063                          be32_to_cpu(key->keyid));
1064         }
1065 }
1066
1067 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1068                              struct mlx5_flow_spec *spec,
1069                              struct tc_cls_flower_offload *f)
1070 {
1071         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1072                                        outer_headers);
1073         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1074                                        outer_headers);
1075
1076         struct flow_dissector_key_control *enc_control =
1077                 skb_flow_dissector_target(f->dissector,
1078                                           FLOW_DISSECTOR_KEY_ENC_CONTROL,
1079                                           f->key);
1080
1081         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
1082                 struct flow_dissector_key_ports *key =
1083                         skb_flow_dissector_target(f->dissector,
1084                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
1085                                                   f->key);
1086                 struct flow_dissector_key_ports *mask =
1087                         skb_flow_dissector_target(f->dissector,
1088                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
1089                                                   f->mask);
1090                 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1091                 struct mlx5e_rep_priv *uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1092                 struct net_device *up_dev = uplink_rpriv->netdev;
1093                 struct mlx5e_priv *up_priv = netdev_priv(up_dev);
1094
1095                 /* Full udp dst port must be given */
1096                 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
1097                         goto vxlan_match_offload_err;
1098
1099                 if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->dst)) &&
1100                     MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
1101                         parse_vxlan_attr(spec, f);
1102                 else {
1103                         netdev_warn(priv->netdev,
1104                                     "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
1105                         return -EOPNOTSUPP;
1106                 }
1107
1108                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1109                          udp_dport, ntohs(mask->dst));
1110                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1111                          udp_dport, ntohs(key->dst));
1112
1113                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1114                          udp_sport, ntohs(mask->src));
1115                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1116                          udp_sport, ntohs(key->src));
1117         } else { /* udp dst port must be given */
1118 vxlan_match_offload_err:
1119                 netdev_warn(priv->netdev,
1120                             "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
1121                 return -EOPNOTSUPP;
1122         }
1123
1124         if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1125                 struct flow_dissector_key_ipv4_addrs *key =
1126                         skb_flow_dissector_target(f->dissector,
1127                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1128                                                   f->key);
1129                 struct flow_dissector_key_ipv4_addrs *mask =
1130                         skb_flow_dissector_target(f->dissector,
1131                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1132                                                   f->mask);
1133                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1134                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
1135                          ntohl(mask->src));
1136                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1137                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
1138                          ntohl(key->src));
1139
1140                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1141                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1142                          ntohl(mask->dst));
1143                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1144                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1145                          ntohl(key->dst));
1146
1147                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1148                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
1149         } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1150                 struct flow_dissector_key_ipv6_addrs *key =
1151                         skb_flow_dissector_target(f->dissector,
1152                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1153                                                   f->key);
1154                 struct flow_dissector_key_ipv6_addrs *mask =
1155                         skb_flow_dissector_target(f->dissector,
1156                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1157                                                   f->mask);
1158
1159                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1160                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1161                        &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1162                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1163                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1164                        &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1165
1166                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1167                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1168                        &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1169                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1170                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1171                        &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1172
1173                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1174                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
1175         }
1176
1177         /* Enforce DMAC when offloading incoming tunneled flows.
1178          * Flow counters require a match on the DMAC.
1179          */
1180         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
1181         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
1182         ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1183                                      dmac_47_16), priv->netdev->dev_addr);
1184
1185         /* let software handle IP fragments */
1186         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1187         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
1188
1189         return 0;
1190 }
1191
1192 static int __parse_cls_flower(struct mlx5e_priv *priv,
1193                               struct mlx5_flow_spec *spec,
1194                               struct tc_cls_flower_offload *f,
1195                               u8 *min_inline)
1196 {
1197         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1198                                        outer_headers);
1199         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1200                                        outer_headers);
1201         u16 addr_type = 0;
1202         u8 ip_proto = 0;
1203
1204         *min_inline = MLX5_INLINE_MODE_L2;
1205
1206         if (f->dissector->used_keys &
1207             ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
1208               BIT(FLOW_DISSECTOR_KEY_BASIC) |
1209               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
1210               BIT(FLOW_DISSECTOR_KEY_VLAN) |
1211               BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
1212               BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
1213               BIT(FLOW_DISSECTOR_KEY_PORTS) |
1214               BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
1215               BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
1216               BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
1217               BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
1218               BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
1219               BIT(FLOW_DISSECTOR_KEY_TCP) |
1220               BIT(FLOW_DISSECTOR_KEY_IP))) {
1221                 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
1222                             f->dissector->used_keys);
1223                 return -EOPNOTSUPP;
1224         }
1225
1226         if ((dissector_uses_key(f->dissector,
1227                                 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
1228              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
1229              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
1230             dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
1231                 struct flow_dissector_key_control *key =
1232                         skb_flow_dissector_target(f->dissector,
1233                                                   FLOW_DISSECTOR_KEY_ENC_CONTROL,
1234                                                   f->key);
1235                 switch (key->addr_type) {
1236                 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
1237                 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
1238                         if (parse_tunnel_attr(priv, spec, f))
1239                                 return -EOPNOTSUPP;
1240                         break;
1241                 default:
1242                         return -EOPNOTSUPP;
1243                 }
1244
1245                 /* In decap flow, header pointers should point to the inner
1246                  * headers, outer header were already set by parse_tunnel_attr
1247                  */
1248                 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1249                                          inner_headers);
1250                 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1251                                          inner_headers);
1252         }
1253
1254         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
1255                 struct flow_dissector_key_control *key =
1256                         skb_flow_dissector_target(f->dissector,
1257                                                   FLOW_DISSECTOR_KEY_CONTROL,
1258                                                   f->key);
1259
1260                 struct flow_dissector_key_control *mask =
1261                         skb_flow_dissector_target(f->dissector,
1262                                                   FLOW_DISSECTOR_KEY_CONTROL,
1263                                                   f->mask);
1264                 addr_type = key->addr_type;
1265
1266                 /* the HW doesn't support frag first/later */
1267                 if (mask->flags & FLOW_DIS_FIRST_FRAG)
1268                         return -EOPNOTSUPP;
1269
1270                 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
1271                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1272                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
1273                                  key->flags & FLOW_DIS_IS_FRAGMENT);
1274
1275                         /* the HW doesn't need L3 inline to match on frag=no */
1276                         if (key->flags & FLOW_DIS_IS_FRAGMENT)
1277                                 *min_inline = MLX5_INLINE_MODE_IP;
1278                 }
1279         }
1280
1281         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1282                 struct flow_dissector_key_basic *key =
1283                         skb_flow_dissector_target(f->dissector,
1284                                                   FLOW_DISSECTOR_KEY_BASIC,
1285                                                   f->key);
1286                 struct flow_dissector_key_basic *mask =
1287                         skb_flow_dissector_target(f->dissector,
1288                                                   FLOW_DISSECTOR_KEY_BASIC,
1289                                                   f->mask);
1290                 ip_proto = key->ip_proto;
1291
1292                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1293                          ntohs(mask->n_proto));
1294                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1295                          ntohs(key->n_proto));
1296
1297                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
1298                          mask->ip_proto);
1299                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
1300                          key->ip_proto);
1301
1302                 if (mask->ip_proto)
1303                         *min_inline = MLX5_INLINE_MODE_IP;
1304         }
1305
1306         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
1307                 struct flow_dissector_key_eth_addrs *key =
1308                         skb_flow_dissector_target(f->dissector,
1309                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
1310                                                   f->key);
1311                 struct flow_dissector_key_eth_addrs *mask =
1312                         skb_flow_dissector_target(f->dissector,
1313                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
1314                                                   f->mask);
1315
1316                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1317                                              dmac_47_16),
1318                                 mask->dst);
1319                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1320                                              dmac_47_16),
1321                                 key->dst);
1322
1323                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1324                                              smac_47_16),
1325                                 mask->src);
1326                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1327                                              smac_47_16),
1328                                 key->src);
1329         }
1330
1331         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
1332                 struct flow_dissector_key_vlan *key =
1333                         skb_flow_dissector_target(f->dissector,
1334                                                   FLOW_DISSECTOR_KEY_VLAN,
1335                                                   f->key);
1336                 struct flow_dissector_key_vlan *mask =
1337                         skb_flow_dissector_target(f->dissector,
1338                                                   FLOW_DISSECTOR_KEY_VLAN,
1339                                                   f->mask);
1340                 if (mask->vlan_id || mask->vlan_priority) {
1341                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
1342                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
1343
1344                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
1345                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
1346
1347                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
1348                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
1349                 }
1350         }
1351
1352         if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1353                 struct flow_dissector_key_ipv4_addrs *key =
1354                         skb_flow_dissector_target(f->dissector,
1355                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1356                                                   f->key);
1357                 struct flow_dissector_key_ipv4_addrs *mask =
1358                         skb_flow_dissector_target(f->dissector,
1359                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1360                                                   f->mask);
1361
1362                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1363                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
1364                        &mask->src, sizeof(mask->src));
1365                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1366                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
1367                        &key->src, sizeof(key->src));
1368                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1369                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1370                        &mask->dst, sizeof(mask->dst));
1371                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1372                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1373                        &key->dst, sizeof(key->dst));
1374
1375                 if (mask->src || mask->dst)
1376                         *min_inline = MLX5_INLINE_MODE_IP;
1377         }
1378
1379         if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1380                 struct flow_dissector_key_ipv6_addrs *key =
1381                         skb_flow_dissector_target(f->dissector,
1382                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1383                                                   f->key);
1384                 struct flow_dissector_key_ipv6_addrs *mask =
1385                         skb_flow_dissector_target(f->dissector,
1386                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1387                                                   f->mask);
1388
1389                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1390                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1391                        &mask->src, sizeof(mask->src));
1392                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1393                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1394                        &key->src, sizeof(key->src));
1395
1396                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1397                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1398                        &mask->dst, sizeof(mask->dst));
1399                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1400                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1401                        &key->dst, sizeof(key->dst));
1402
1403                 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
1404                     ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
1405                         *min_inline = MLX5_INLINE_MODE_IP;
1406         }
1407
1408         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_IP)) {
1409                 struct flow_dissector_key_ip *key =
1410                         skb_flow_dissector_target(f->dissector,
1411                                                   FLOW_DISSECTOR_KEY_IP,
1412                                                   f->key);
1413                 struct flow_dissector_key_ip *mask =
1414                         skb_flow_dissector_target(f->dissector,
1415                                                   FLOW_DISSECTOR_KEY_IP,
1416                                                   f->mask);
1417
1418                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn, mask->tos & 0x3);
1419                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, key->tos & 0x3);
1420
1421                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp, mask->tos >> 2);
1422                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, key->tos  >> 2);
1423
1424                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit, mask->ttl);
1425                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit, key->ttl);
1426
1427                 if (mask->ttl &&
1428                     !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
1429                                                 ft_field_support.outer_ipv4_ttl))
1430                         return -EOPNOTSUPP;
1431
1432                 if (mask->tos || mask->ttl)
1433                         *min_inline = MLX5_INLINE_MODE_IP;
1434         }
1435
1436         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
1437                 struct flow_dissector_key_ports *key =
1438                         skb_flow_dissector_target(f->dissector,
1439                                                   FLOW_DISSECTOR_KEY_PORTS,
1440                                                   f->key);
1441                 struct flow_dissector_key_ports *mask =
1442                         skb_flow_dissector_target(f->dissector,
1443                                                   FLOW_DISSECTOR_KEY_PORTS,
1444                                                   f->mask);
1445                 switch (ip_proto) {
1446                 case IPPROTO_TCP:
1447                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1448                                  tcp_sport, ntohs(mask->src));
1449                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1450                                  tcp_sport, ntohs(key->src));
1451
1452                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1453                                  tcp_dport, ntohs(mask->dst));
1454                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1455                                  tcp_dport, ntohs(key->dst));
1456                         break;
1457
1458                 case IPPROTO_UDP:
1459                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1460                                  udp_sport, ntohs(mask->src));
1461                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1462                                  udp_sport, ntohs(key->src));
1463
1464                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1465                                  udp_dport, ntohs(mask->dst));
1466                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1467                                  udp_dport, ntohs(key->dst));
1468                         break;
1469                 default:
1470                         netdev_err(priv->netdev,
1471                                    "Only UDP and TCP transport are supported\n");
1472                         return -EINVAL;
1473                 }
1474
1475                 if (mask->src || mask->dst)
1476                         *min_inline = MLX5_INLINE_MODE_TCP_UDP;
1477         }
1478
1479         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_TCP)) {
1480                 struct flow_dissector_key_tcp *key =
1481                         skb_flow_dissector_target(f->dissector,
1482                                                   FLOW_DISSECTOR_KEY_TCP,
1483                                                   f->key);
1484                 struct flow_dissector_key_tcp *mask =
1485                         skb_flow_dissector_target(f->dissector,
1486                                                   FLOW_DISSECTOR_KEY_TCP,
1487                                                   f->mask);
1488
1489                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
1490                          ntohs(mask->flags));
1491                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
1492                          ntohs(key->flags));
1493
1494                 if (mask->flags)
1495                         *min_inline = MLX5_INLINE_MODE_TCP_UDP;
1496         }
1497
1498         return 0;
1499 }
1500
1501 static int parse_cls_flower(struct mlx5e_priv *priv,
1502                             struct mlx5e_tc_flow *flow,
1503                             struct mlx5_flow_spec *spec,
1504                             struct tc_cls_flower_offload *f)
1505 {
1506         struct mlx5_core_dev *dev = priv->mdev;
1507         struct mlx5_eswitch *esw = dev->priv.eswitch;
1508         struct mlx5e_rep_priv *rpriv = priv->ppriv;
1509         struct mlx5_eswitch_rep *rep;
1510         u8 min_inline;
1511         int err;
1512
1513         err = __parse_cls_flower(priv, spec, f, &min_inline);
1514
1515         if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH)) {
1516                 rep = rpriv->rep;
1517                 if (rep->vport != FDB_UPLINK_VPORT &&
1518                     (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
1519                     esw->offloads.inline_mode < min_inline)) {
1520                         netdev_warn(priv->netdev,
1521                                     "Flow is not offloaded due to min inline setting, required %d actual %d\n",
1522                                     min_inline, esw->offloads.inline_mode);
1523                         return -EOPNOTSUPP;
1524                 }
1525         }
1526
1527         return err;
1528 }
1529
1530 struct pedit_headers {
1531         struct ethhdr  eth;
1532         struct iphdr   ip4;
1533         struct ipv6hdr ip6;
1534         struct tcphdr  tcp;
1535         struct udphdr  udp;
1536 };
1537
1538 static int pedit_header_offsets[] = {
1539         [TCA_PEDIT_KEY_EX_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
1540         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
1541         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
1542         [TCA_PEDIT_KEY_EX_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
1543         [TCA_PEDIT_KEY_EX_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
1544 };
1545
1546 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
1547
1548 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
1549                          struct pedit_headers *masks,
1550                          struct pedit_headers *vals)
1551 {
1552         u32 *curr_pmask, *curr_pval;
1553
1554         if (hdr_type >= __PEDIT_HDR_TYPE_MAX)
1555                 goto out_err;
1556
1557         curr_pmask = (u32 *)(pedit_header(masks, hdr_type) + offset);
1558         curr_pval  = (u32 *)(pedit_header(vals, hdr_type) + offset);
1559
1560         if (*curr_pmask & mask)  /* disallow acting twice on the same location */
1561                 goto out_err;
1562
1563         *curr_pmask |= mask;
1564         *curr_pval  |= (val & mask);
1565
1566         return 0;
1567
1568 out_err:
1569         return -EOPNOTSUPP;
1570 }
1571
1572 struct mlx5_fields {
1573         u8  field;
1574         u8  size;
1575         u32 offset;
1576 };
1577
1578 #define OFFLOAD(fw_field, size, field, off) \
1579                 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, size, offsetof(struct pedit_headers, field) + (off)}
1580
1581 static struct mlx5_fields fields[] = {
1582         OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0),
1583         OFFLOAD(DMAC_15_0,  2, eth.h_dest[4], 0),
1584         OFFLOAD(SMAC_47_16, 4, eth.h_source[0], 0),
1585         OFFLOAD(SMAC_15_0,  2, eth.h_source[4], 0),
1586         OFFLOAD(ETHERTYPE,  2, eth.h_proto, 0),
1587
1588         OFFLOAD(IP_TTL, 1, ip4.ttl,   0),
1589         OFFLOAD(SIPV4,  4, ip4.saddr, 0),
1590         OFFLOAD(DIPV4,  4, ip4.daddr, 0),
1591
1592         OFFLOAD(SIPV6_127_96, 4, ip6.saddr.s6_addr32[0], 0),
1593         OFFLOAD(SIPV6_95_64,  4, ip6.saddr.s6_addr32[1], 0),
1594         OFFLOAD(SIPV6_63_32,  4, ip6.saddr.s6_addr32[2], 0),
1595         OFFLOAD(SIPV6_31_0,   4, ip6.saddr.s6_addr32[3], 0),
1596         OFFLOAD(DIPV6_127_96, 4, ip6.daddr.s6_addr32[0], 0),
1597         OFFLOAD(DIPV6_95_64,  4, ip6.daddr.s6_addr32[1], 0),
1598         OFFLOAD(DIPV6_63_32,  4, ip6.daddr.s6_addr32[2], 0),
1599         OFFLOAD(DIPV6_31_0,   4, ip6.daddr.s6_addr32[3], 0),
1600         OFFLOAD(IPV6_HOPLIMIT, 1, ip6.hop_limit, 0),
1601
1602         OFFLOAD(TCP_SPORT, 2, tcp.source,  0),
1603         OFFLOAD(TCP_DPORT, 2, tcp.dest,    0),
1604         OFFLOAD(TCP_FLAGS, 1, tcp.ack_seq, 5),
1605
1606         OFFLOAD(UDP_SPORT, 2, udp.source, 0),
1607         OFFLOAD(UDP_DPORT, 2, udp.dest,   0),
1608 };
1609
1610 /* On input attr->num_mod_hdr_actions tells how many HW actions can be parsed at
1611  * max from the SW pedit action. On success, it says how many HW actions were
1612  * actually parsed.
1613  */
1614 static int offload_pedit_fields(struct pedit_headers *masks,
1615                                 struct pedit_headers *vals,
1616                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
1617 {
1618         struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
1619         int i, action_size, nactions, max_actions, first, last, next_z;
1620         void *s_masks_p, *a_masks_p, *vals_p;
1621         struct mlx5_fields *f;
1622         u8 cmd, field_bsize;
1623         u32 s_mask, a_mask;
1624         unsigned long mask;
1625         __be32 mask_be32;
1626         __be16 mask_be16;
1627         void *action;
1628
1629         set_masks = &masks[TCA_PEDIT_KEY_EX_CMD_SET];
1630         add_masks = &masks[TCA_PEDIT_KEY_EX_CMD_ADD];
1631         set_vals = &vals[TCA_PEDIT_KEY_EX_CMD_SET];
1632         add_vals = &vals[TCA_PEDIT_KEY_EX_CMD_ADD];
1633
1634         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1635         action = parse_attr->mod_hdr_actions;
1636         max_actions = parse_attr->num_mod_hdr_actions;
1637         nactions = 0;
1638
1639         for (i = 0; i < ARRAY_SIZE(fields); i++) {
1640                 f = &fields[i];
1641                 /* avoid seeing bits set from previous iterations */
1642                 s_mask = 0;
1643                 a_mask = 0;
1644
1645                 s_masks_p = (void *)set_masks + f->offset;
1646                 a_masks_p = (void *)add_masks + f->offset;
1647
1648                 memcpy(&s_mask, s_masks_p, f->size);
1649                 memcpy(&a_mask, a_masks_p, f->size);
1650
1651                 if (!s_mask && !a_mask) /* nothing to offload here */
1652                         continue;
1653
1654                 if (s_mask && a_mask) {
1655                         printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
1656                         return -EOPNOTSUPP;
1657                 }
1658
1659                 if (nactions == max_actions) {
1660                         printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
1661                         return -EOPNOTSUPP;
1662                 }
1663
1664                 if (s_mask) {
1665                         cmd  = MLX5_ACTION_TYPE_SET;
1666                         mask = s_mask;
1667                         vals_p = (void *)set_vals + f->offset;
1668                         /* clear to denote we consumed this field */
1669                         memset(s_masks_p, 0, f->size);
1670                 } else {
1671                         cmd  = MLX5_ACTION_TYPE_ADD;
1672                         mask = a_mask;
1673                         vals_p = (void *)add_vals + f->offset;
1674                         /* clear to denote we consumed this field */
1675                         memset(a_masks_p, 0, f->size);
1676                 }
1677
1678                 field_bsize = f->size * BITS_PER_BYTE;
1679
1680                 if (field_bsize == 32) {
1681                         mask_be32 = *(__be32 *)&mask;
1682                         mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
1683                 } else if (field_bsize == 16) {
1684                         mask_be16 = *(__be16 *)&mask;
1685                         mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
1686                 }
1687
1688                 first = find_first_bit(&mask, field_bsize);
1689                 next_z = find_next_zero_bit(&mask, field_bsize, first);
1690                 last  = find_last_bit(&mask, field_bsize);
1691                 if (first < next_z && next_z < last) {
1692                         printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
1693                                mask);
1694                         return -EOPNOTSUPP;
1695                 }
1696
1697                 MLX5_SET(set_action_in, action, action_type, cmd);
1698                 MLX5_SET(set_action_in, action, field, f->field);
1699
1700                 if (cmd == MLX5_ACTION_TYPE_SET) {
1701                         MLX5_SET(set_action_in, action, offset, first);
1702                         /* length is num of bits to be written, zero means length of 32 */
1703                         MLX5_SET(set_action_in, action, length, (last - first + 1));
1704                 }
1705
1706                 if (field_bsize == 32)
1707                         MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
1708                 else if (field_bsize == 16)
1709                         MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
1710                 else if (field_bsize == 8)
1711                         MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
1712
1713                 action += action_size;
1714                 nactions++;
1715         }
1716
1717         parse_attr->num_mod_hdr_actions = nactions;
1718         return 0;
1719 }
1720
1721 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
1722                                  const struct tc_action *a, int namespace,
1723                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1724 {
1725         int nkeys, action_size, max_actions;
1726
1727         nkeys = tcf_pedit_nkeys(a);
1728         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1729
1730         if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
1731                 max_actions = MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev, max_modify_header_actions);
1732         else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
1733                 max_actions = MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, max_modify_header_actions);
1734
1735         /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
1736         max_actions = min(max_actions, nkeys * 16);
1737
1738         parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
1739         if (!parse_attr->mod_hdr_actions)
1740                 return -ENOMEM;
1741
1742         parse_attr->num_mod_hdr_actions = max_actions;
1743         return 0;
1744 }
1745
1746 static const struct pedit_headers zero_masks = {};
1747
1748 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
1749                                  const struct tc_action *a, int namespace,
1750                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1751 {
1752         struct pedit_headers masks[__PEDIT_CMD_MAX], vals[__PEDIT_CMD_MAX], *cmd_masks;
1753         int nkeys, i, err = -EOPNOTSUPP;
1754         u32 mask, val, offset;
1755         u8 cmd, htype;
1756
1757         nkeys = tcf_pedit_nkeys(a);
1758
1759         memset(masks, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1760         memset(vals,  0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1761
1762         for (i = 0; i < nkeys; i++) {
1763                 htype = tcf_pedit_htype(a, i);
1764                 cmd = tcf_pedit_cmd(a, i);
1765                 err = -EOPNOTSUPP; /* can't be all optimistic */
1766
1767                 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK) {
1768                         netdev_warn(priv->netdev, "legacy pedit isn't offloaded\n");
1769                         goto out_err;
1770                 }
1771
1772                 if (cmd != TCA_PEDIT_KEY_EX_CMD_SET && cmd != TCA_PEDIT_KEY_EX_CMD_ADD) {
1773                         netdev_warn(priv->netdev, "pedit cmd %d isn't offloaded\n", cmd);
1774                         goto out_err;
1775                 }
1776
1777                 mask = tcf_pedit_mask(a, i);
1778                 val = tcf_pedit_val(a, i);
1779                 offset = tcf_pedit_offset(a, i);
1780
1781                 err = set_pedit_val(htype, ~mask, val, offset, &masks[cmd], &vals[cmd]);
1782                 if (err)
1783                         goto out_err;
1784         }
1785
1786         err = alloc_mod_hdr_actions(priv, a, namespace, parse_attr);
1787         if (err)
1788                 goto out_err;
1789
1790         err = offload_pedit_fields(masks, vals, parse_attr);
1791         if (err < 0)
1792                 goto out_dealloc_parsed_actions;
1793
1794         for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
1795                 cmd_masks = &masks[cmd];
1796                 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
1797                         netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
1798                         print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
1799                                        16, 1, cmd_masks, sizeof(zero_masks), true);
1800                         err = -EOPNOTSUPP;
1801                         goto out_dealloc_parsed_actions;
1802                 }
1803         }
1804
1805         return 0;
1806
1807 out_dealloc_parsed_actions:
1808         kfree(parse_attr->mod_hdr_actions);
1809 out_err:
1810         return err;
1811 }
1812
1813 static bool csum_offload_supported(struct mlx5e_priv *priv, u32 action, u32 update_flags)
1814 {
1815         u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
1816                          TCA_CSUM_UPDATE_FLAG_UDP;
1817
1818         /*  The HW recalcs checksums only if re-writing headers */
1819         if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
1820                 netdev_warn(priv->netdev,
1821                             "TC csum action is only offloaded with pedit\n");
1822                 return false;
1823         }
1824
1825         if (update_flags & ~prot_flags) {
1826                 netdev_warn(priv->netdev,
1827                             "can't offload TC csum action for some header/s - flags %#x\n",
1828                             update_flags);
1829                 return false;
1830         }
1831
1832         return true;
1833 }
1834
1835 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
1836                                           struct tcf_exts *exts)
1837 {
1838         const struct tc_action *a;
1839         bool modify_ip_header;
1840         LIST_HEAD(actions);
1841         u8 htype, ip_proto;
1842         void *headers_v;
1843         u16 ethertype;
1844         int nkeys, i;
1845
1846         headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
1847         ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
1848
1849         /* for non-IP we only re-write MACs, so we're okay */
1850         if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
1851                 goto out_ok;
1852
1853         modify_ip_header = false;
1854         tcf_exts_to_list(exts, &actions);
1855         list_for_each_entry(a, &actions, list) {
1856                 if (!is_tcf_pedit(a))
1857                         continue;
1858
1859                 nkeys = tcf_pedit_nkeys(a);
1860                 for (i = 0; i < nkeys; i++) {
1861                         htype = tcf_pedit_htype(a, i);
1862                         if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP4 ||
1863                             htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP6) {
1864                                 modify_ip_header = true;
1865                                 break;
1866                         }
1867                 }
1868         }
1869
1870         ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
1871         if (modify_ip_header && ip_proto != IPPROTO_TCP &&
1872             ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
1873                 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
1874                 return false;
1875         }
1876
1877 out_ok:
1878         return true;
1879 }
1880
1881 static bool actions_match_supported(struct mlx5e_priv *priv,
1882                                     struct tcf_exts *exts,
1883                                     struct mlx5e_tc_flow_parse_attr *parse_attr,
1884                                     struct mlx5e_tc_flow *flow)
1885 {
1886         u32 actions;
1887
1888         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1889                 actions = flow->esw_attr->action;
1890         else
1891                 actions = flow->nic_attr->action;
1892
1893         if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
1894                 return modify_header_match_supported(&parse_attr->spec, exts);
1895
1896         return true;
1897 }
1898
1899 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
1900 {
1901         struct mlx5_core_dev *fmdev, *pmdev;
1902         u16 func_id, peer_id;
1903
1904         fmdev = priv->mdev;
1905         pmdev = peer_priv->mdev;
1906
1907         func_id = (u16)((fmdev->pdev->bus->number << 8) | PCI_SLOT(fmdev->pdev->devfn));
1908         peer_id = (u16)((pmdev->pdev->bus->number << 8) | PCI_SLOT(pmdev->pdev->devfn));
1909
1910         return (func_id == peer_id);
1911 }
1912
1913 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
1914                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
1915                                 struct mlx5e_tc_flow *flow)
1916 {
1917         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
1918         const struct tc_action *a;
1919         LIST_HEAD(actions);
1920         int err;
1921
1922         if (!tcf_exts_has_actions(exts))
1923                 return -EINVAL;
1924
1925         attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
1926
1927         tcf_exts_to_list(exts, &actions);
1928         list_for_each_entry(a, &actions, list) {
1929                 if (is_tcf_gact_shot(a)) {
1930                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
1931                         if (MLX5_CAP_FLOWTABLE(priv->mdev,
1932                                                flow_table_properties_nic_receive.flow_counter))
1933                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
1934                         continue;
1935                 }
1936
1937                 if (is_tcf_pedit(a)) {
1938                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_KERNEL,
1939                                                     parse_attr);
1940                         if (err)
1941                                 return err;
1942
1943                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
1944                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1945                         continue;
1946                 }
1947
1948                 if (is_tcf_csum(a)) {
1949                         if (csum_offload_supported(priv, attr->action,
1950                                                    tcf_csum_update_flags(a)))
1951                                 continue;
1952
1953                         return -EOPNOTSUPP;
1954                 }
1955
1956                 if (is_tcf_mirred_egress_redirect(a)) {
1957                         struct net_device *peer_dev = tcf_mirred_dev(a);
1958
1959                         if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
1960                             same_hw_devs(priv, netdev_priv(peer_dev))) {
1961                                 parse_attr->mirred_ifindex = peer_dev->ifindex;
1962                                 flow->flags |= MLX5E_TC_FLOW_HAIRPIN;
1963                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1964                                                 MLX5_FLOW_CONTEXT_ACTION_COUNT;
1965                         } else {
1966                                 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
1967                                             peer_dev->name);
1968                                 return -EINVAL;
1969                         }
1970                         continue;
1971                 }
1972
1973                 if (is_tcf_skbedit_mark(a)) {
1974                         u32 mark = tcf_skbedit_mark(a);
1975
1976                         if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
1977                                 netdev_warn(priv->netdev, "Bad flow mark - only 16 bit is supported: 0x%x\n",
1978                                             mark);
1979                                 return -EINVAL;
1980                         }
1981
1982                         attr->flow_tag = mark;
1983                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1984                         continue;
1985                 }
1986
1987                 return -EINVAL;
1988         }
1989
1990         if (!actions_match_supported(priv, exts, parse_attr, flow))
1991                 return -EOPNOTSUPP;
1992
1993         return 0;
1994 }
1995
1996 static inline int cmp_encap_info(struct ip_tunnel_key *a,
1997                                  struct ip_tunnel_key *b)
1998 {
1999         return memcmp(a, b, sizeof(*a));
2000 }
2001
2002 static inline int hash_encap_info(struct ip_tunnel_key *key)
2003 {
2004         return jhash(key, sizeof(*key), 0);
2005 }
2006
2007 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
2008                                    struct net_device *mirred_dev,
2009                                    struct net_device **out_dev,
2010                                    struct flowi4 *fl4,
2011                                    struct neighbour **out_n,
2012                                    int *out_ttl)
2013 {
2014         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2015         struct mlx5e_rep_priv *uplink_rpriv;
2016         struct rtable *rt;
2017         struct neighbour *n = NULL;
2018
2019 #if IS_ENABLED(CONFIG_INET)
2020         int ret;
2021
2022         rt = ip_route_output_key(dev_net(mirred_dev), fl4);
2023         ret = PTR_ERR_OR_ZERO(rt);
2024         if (ret)
2025                 return ret;
2026 #else
2027         return -EOPNOTSUPP;
2028 #endif
2029         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2030         /* if the egress device isn't on the same HW e-switch, we use the uplink */
2031         if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
2032                 *out_dev = uplink_rpriv->netdev;
2033         else
2034                 *out_dev = rt->dst.dev;
2035
2036         *out_ttl = ip4_dst_hoplimit(&rt->dst);
2037         n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
2038         ip_rt_put(rt);
2039         if (!n)
2040                 return -ENOMEM;
2041
2042         *out_n = n;
2043         return 0;
2044 }
2045
2046 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
2047                                    struct net_device *mirred_dev,
2048                                    struct net_device **out_dev,
2049                                    struct flowi6 *fl6,
2050                                    struct neighbour **out_n,
2051                                    int *out_ttl)
2052 {
2053         struct neighbour *n = NULL;
2054         struct dst_entry *dst;
2055
2056 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
2057         struct mlx5e_rep_priv *uplink_rpriv;
2058         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2059         int ret;
2060
2061         ret = ipv6_stub->ipv6_dst_lookup(dev_net(mirred_dev), NULL, &dst,
2062                                          fl6);
2063         if (ret < 0)
2064                 return ret;
2065
2066         *out_ttl = ip6_dst_hoplimit(dst);
2067
2068         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2069         /* if the egress device isn't on the same HW e-switch, we use the uplink */
2070         if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
2071                 *out_dev = uplink_rpriv->netdev;
2072         else
2073                 *out_dev = dst->dev;
2074 #else
2075         return -EOPNOTSUPP;
2076 #endif
2077
2078         n = dst_neigh_lookup(dst, &fl6->daddr);
2079         dst_release(dst);
2080         if (!n)
2081                 return -ENOMEM;
2082
2083         *out_n = n;
2084         return 0;
2085 }
2086
2087 static void gen_vxlan_header_ipv4(struct net_device *out_dev,
2088                                   char buf[], int encap_size,
2089                                   unsigned char h_dest[ETH_ALEN],
2090                                   int ttl,
2091                                   __be32 daddr,
2092                                   __be32 saddr,
2093                                   __be16 udp_dst_port,
2094                                   __be32 vx_vni)
2095 {
2096         struct ethhdr *eth = (struct ethhdr *)buf;
2097         struct iphdr  *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
2098         struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
2099         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2100
2101         memset(buf, 0, encap_size);
2102
2103         ether_addr_copy(eth->h_dest, h_dest);
2104         ether_addr_copy(eth->h_source, out_dev->dev_addr);
2105         eth->h_proto = htons(ETH_P_IP);
2106
2107         ip->daddr = daddr;
2108         ip->saddr = saddr;
2109
2110         ip->ttl = ttl;
2111         ip->protocol = IPPROTO_UDP;
2112         ip->version = 0x4;
2113         ip->ihl = 0x5;
2114
2115         udp->dest = udp_dst_port;
2116         vxh->vx_flags = VXLAN_HF_VNI;
2117         vxh->vx_vni = vxlan_vni_field(vx_vni);
2118 }
2119
2120 static void gen_vxlan_header_ipv6(struct net_device *out_dev,
2121                                   char buf[], int encap_size,
2122                                   unsigned char h_dest[ETH_ALEN],
2123                                   int ttl,
2124                                   struct in6_addr *daddr,
2125                                   struct in6_addr *saddr,
2126                                   __be16 udp_dst_port,
2127                                   __be32 vx_vni)
2128 {
2129         struct ethhdr *eth = (struct ethhdr *)buf;
2130         struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
2131         struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
2132         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2133
2134         memset(buf, 0, encap_size);
2135
2136         ether_addr_copy(eth->h_dest, h_dest);
2137         ether_addr_copy(eth->h_source, out_dev->dev_addr);
2138         eth->h_proto = htons(ETH_P_IPV6);
2139
2140         ip6_flow_hdr(ip6h, 0, 0);
2141         /* the HW fills up ipv6 payload len */
2142         ip6h->nexthdr     = IPPROTO_UDP;
2143         ip6h->hop_limit   = ttl;
2144         ip6h->daddr       = *daddr;
2145         ip6h->saddr       = *saddr;
2146
2147         udp->dest = udp_dst_port;
2148         vxh->vx_flags = VXLAN_HF_VNI;
2149         vxh->vx_vni = vxlan_vni_field(vx_vni);
2150 }
2151
2152 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
2153                                           struct net_device *mirred_dev,
2154                                           struct mlx5e_encap_entry *e)
2155 {
2156         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2157         int ipv4_encap_size = ETH_HLEN + sizeof(struct iphdr) + VXLAN_HLEN;
2158         struct ip_tunnel_key *tun_key = &e->tun_info.key;
2159         struct net_device *out_dev;
2160         struct neighbour *n = NULL;
2161         struct flowi4 fl4 = {};
2162         char *encap_header;
2163         int ttl, err;
2164         u8 nud_state;
2165
2166         if (max_encap_size < ipv4_encap_size) {
2167                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2168                                ipv4_encap_size, max_encap_size);
2169                 return -EOPNOTSUPP;
2170         }
2171
2172         encap_header = kzalloc(ipv4_encap_size, GFP_KERNEL);
2173         if (!encap_header)
2174                 return -ENOMEM;
2175
2176         switch (e->tunnel_type) {
2177         case MLX5_HEADER_TYPE_VXLAN:
2178                 fl4.flowi4_proto = IPPROTO_UDP;
2179                 fl4.fl4_dport = tun_key->tp_dst;
2180                 break;
2181         default:
2182                 err = -EOPNOTSUPP;
2183                 goto free_encap;
2184         }
2185         fl4.flowi4_tos = tun_key->tos;
2186         fl4.daddr = tun_key->u.ipv4.dst;
2187         fl4.saddr = tun_key->u.ipv4.src;
2188
2189         err = mlx5e_route_lookup_ipv4(priv, mirred_dev, &out_dev,
2190                                       &fl4, &n, &ttl);
2191         if (err)
2192                 goto free_encap;
2193
2194         /* used by mlx5e_detach_encap to lookup a neigh hash table
2195          * entry in the neigh hash table when a user deletes a rule
2196          */
2197         e->m_neigh.dev = n->dev;
2198         e->m_neigh.family = n->ops->family;
2199         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2200         e->out_dev = out_dev;
2201
2202         /* It's importent to add the neigh to the hash table before checking
2203          * the neigh validity state. So if we'll get a notification, in case the
2204          * neigh changes it's validity state, we would find the relevant neigh
2205          * in the hash.
2206          */
2207         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2208         if (err)
2209                 goto free_encap;
2210
2211         read_lock_bh(&n->lock);
2212         nud_state = n->nud_state;
2213         ether_addr_copy(e->h_dest, n->ha);
2214         read_unlock_bh(&n->lock);
2215
2216         switch (e->tunnel_type) {
2217         case MLX5_HEADER_TYPE_VXLAN:
2218                 gen_vxlan_header_ipv4(out_dev, encap_header,
2219                                       ipv4_encap_size, e->h_dest, ttl,
2220                                       fl4.daddr,
2221                                       fl4.saddr, tun_key->tp_dst,
2222                                       tunnel_id_to_key32(tun_key->tun_id));
2223                 break;
2224         default:
2225                 err = -EOPNOTSUPP;
2226                 goto destroy_neigh_entry;
2227         }
2228         e->encap_size = ipv4_encap_size;
2229         e->encap_header = encap_header;
2230
2231         if (!(nud_state & NUD_VALID)) {
2232                 neigh_event_send(n, NULL);
2233                 err = -EAGAIN;
2234                 goto out;
2235         }
2236
2237         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
2238                                ipv4_encap_size, encap_header, &e->encap_id);
2239         if (err)
2240                 goto destroy_neigh_entry;
2241
2242         e->flags |= MLX5_ENCAP_ENTRY_VALID;
2243         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2244         neigh_release(n);
2245         return err;
2246
2247 destroy_neigh_entry:
2248         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2249 free_encap:
2250         kfree(encap_header);
2251 out:
2252         if (n)
2253                 neigh_release(n);
2254         return err;
2255 }
2256
2257 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
2258                                           struct net_device *mirred_dev,
2259                                           struct mlx5e_encap_entry *e)
2260 {
2261         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2262         int ipv6_encap_size = ETH_HLEN + sizeof(struct ipv6hdr) + VXLAN_HLEN;
2263         struct ip_tunnel_key *tun_key = &e->tun_info.key;
2264         struct net_device *out_dev;
2265         struct neighbour *n = NULL;
2266         struct flowi6 fl6 = {};
2267         char *encap_header;
2268         int err, ttl = 0;
2269         u8 nud_state;
2270
2271         if (max_encap_size < ipv6_encap_size) {
2272                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2273                                ipv6_encap_size, max_encap_size);
2274                 return -EOPNOTSUPP;
2275         }
2276
2277         encap_header = kzalloc(ipv6_encap_size, GFP_KERNEL);
2278         if (!encap_header)
2279                 return -ENOMEM;
2280
2281         switch (e->tunnel_type) {
2282         case MLX5_HEADER_TYPE_VXLAN:
2283                 fl6.flowi6_proto = IPPROTO_UDP;
2284                 fl6.fl6_dport = tun_key->tp_dst;
2285                 break;
2286         default:
2287                 err = -EOPNOTSUPP;
2288                 goto free_encap;
2289         }
2290
2291         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
2292         fl6.daddr = tun_key->u.ipv6.dst;
2293         fl6.saddr = tun_key->u.ipv6.src;
2294
2295         err = mlx5e_route_lookup_ipv6(priv, mirred_dev, &out_dev,
2296                                       &fl6, &n, &ttl);
2297         if (err)
2298                 goto free_encap;
2299
2300         /* used by mlx5e_detach_encap to lookup a neigh hash table
2301          * entry in the neigh hash table when a user deletes a rule
2302          */
2303         e->m_neigh.dev = n->dev;
2304         e->m_neigh.family = n->ops->family;
2305         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2306         e->out_dev = out_dev;
2307
2308         /* It's importent to add the neigh to the hash table before checking
2309          * the neigh validity state. So if we'll get a notification, in case the
2310          * neigh changes it's validity state, we would find the relevant neigh
2311          * in the hash.
2312          */
2313         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2314         if (err)
2315                 goto free_encap;
2316
2317         read_lock_bh(&n->lock);
2318         nud_state = n->nud_state;
2319         ether_addr_copy(e->h_dest, n->ha);
2320         read_unlock_bh(&n->lock);
2321
2322         switch (e->tunnel_type) {
2323         case MLX5_HEADER_TYPE_VXLAN:
2324                 gen_vxlan_header_ipv6(out_dev, encap_header,
2325                                       ipv6_encap_size, e->h_dest, ttl,
2326                                       &fl6.daddr,
2327                                       &fl6.saddr, tun_key->tp_dst,
2328                                       tunnel_id_to_key32(tun_key->tun_id));
2329                 break;
2330         default:
2331                 err = -EOPNOTSUPP;
2332                 goto destroy_neigh_entry;
2333         }
2334
2335         e->encap_size = ipv6_encap_size;
2336         e->encap_header = encap_header;
2337
2338         if (!(nud_state & NUD_VALID)) {
2339                 neigh_event_send(n, NULL);
2340                 err = -EAGAIN;
2341                 goto out;
2342         }
2343
2344         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
2345                                ipv6_encap_size, encap_header, &e->encap_id);
2346         if (err)
2347                 goto destroy_neigh_entry;
2348
2349         e->flags |= MLX5_ENCAP_ENTRY_VALID;
2350         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2351         neigh_release(n);
2352         return err;
2353
2354 destroy_neigh_entry:
2355         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2356 free_encap:
2357         kfree(encap_header);
2358 out:
2359         if (n)
2360                 neigh_release(n);
2361         return err;
2362 }
2363
2364 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
2365                               struct ip_tunnel_info *tun_info,
2366                               struct net_device *mirred_dev,
2367                               struct net_device **encap_dev,
2368                               struct mlx5e_tc_flow *flow)
2369 {
2370         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2371         struct mlx5e_rep_priv *uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw,
2372                                                                            REP_ETH);
2373         struct net_device *up_dev = uplink_rpriv->netdev;
2374         unsigned short family = ip_tunnel_info_af(tun_info);
2375         struct mlx5e_priv *up_priv = netdev_priv(up_dev);
2376         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2377         struct ip_tunnel_key *key = &tun_info->key;
2378         struct mlx5e_encap_entry *e;
2379         int tunnel_type, err = 0;
2380         uintptr_t hash_key;
2381         bool found = false;
2382
2383         /* udp dst port must be set */
2384         if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
2385                 goto vxlan_encap_offload_err;
2386
2387         /* setting udp src port isn't supported */
2388         if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
2389 vxlan_encap_offload_err:
2390                 netdev_warn(priv->netdev,
2391                             "must set udp dst port and not set udp src port\n");
2392                 return -EOPNOTSUPP;
2393         }
2394
2395         if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->tp_dst)) &&
2396             MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
2397                 tunnel_type = MLX5_HEADER_TYPE_VXLAN;
2398         } else {
2399                 netdev_warn(priv->netdev,
2400                             "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
2401                 return -EOPNOTSUPP;
2402         }
2403
2404         hash_key = hash_encap_info(key);
2405
2406         hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
2407                                    encap_hlist, hash_key) {
2408                 if (!cmp_encap_info(&e->tun_info.key, key)) {
2409                         found = true;
2410                         break;
2411                 }
2412         }
2413
2414         /* must verify if encap is valid or not */
2415         if (found)
2416                 goto attach_flow;
2417
2418         e = kzalloc(sizeof(*e), GFP_KERNEL);
2419         if (!e)
2420                 return -ENOMEM;
2421
2422         e->tun_info = *tun_info;
2423         e->tunnel_type = tunnel_type;
2424         INIT_LIST_HEAD(&e->flows);
2425
2426         if (family == AF_INET)
2427                 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e);
2428         else if (family == AF_INET6)
2429                 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e);
2430
2431         if (err && err != -EAGAIN)
2432                 goto out_err;
2433
2434         hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
2435
2436 attach_flow:
2437         list_add(&flow->encap, &e->flows);
2438         *encap_dev = e->out_dev;
2439         if (e->flags & MLX5_ENCAP_ENTRY_VALID)
2440                 attr->encap_id = e->encap_id;
2441         else
2442                 err = -EAGAIN;
2443
2444         return err;
2445
2446 out_err:
2447         kfree(e);
2448         return err;
2449 }
2450
2451 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
2452                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
2453                                 struct mlx5e_tc_flow *flow)
2454 {
2455         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2456         struct mlx5e_rep_priv *rpriv = priv->ppriv;
2457         struct ip_tunnel_info *info = NULL;
2458         const struct tc_action *a;
2459         LIST_HEAD(actions);
2460         bool encap = false;
2461
2462         if (!tcf_exts_has_actions(exts))
2463                 return -EINVAL;
2464
2465         attr->in_rep = rpriv->rep;
2466
2467         tcf_exts_to_list(exts, &actions);
2468         list_for_each_entry(a, &actions, list) {
2469                 if (is_tcf_gact_shot(a)) {
2470                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
2471                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
2472                         continue;
2473                 }
2474
2475                 if (is_tcf_pedit(a)) {
2476                         int err;
2477
2478                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_FDB,
2479                                                     parse_attr);
2480                         if (err)
2481                                 return err;
2482
2483                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2484                         continue;
2485                 }
2486
2487                 if (is_tcf_csum(a)) {
2488                         if (csum_offload_supported(priv, attr->action,
2489                                                    tcf_csum_update_flags(a)))
2490                                 continue;
2491
2492                         return -EOPNOTSUPP;
2493                 }
2494
2495                 if (is_tcf_mirred_egress_redirect(a)) {
2496                         struct net_device *out_dev;
2497                         struct mlx5e_priv *out_priv;
2498
2499                         out_dev = tcf_mirred_dev(a);
2500
2501                         if (switchdev_port_same_parent_id(priv->netdev,
2502                                                           out_dev)) {
2503                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2504                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
2505                                 out_priv = netdev_priv(out_dev);
2506                                 rpriv = out_priv->ppriv;
2507                                 attr->out_rep = rpriv->rep;
2508                         } else if (encap) {
2509                                 parse_attr->mirred_ifindex = out_dev->ifindex;
2510                                 parse_attr->tun_info = *info;
2511                                 attr->parse_attr = parse_attr;
2512                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_ENCAP |
2513                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2514                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
2515                                 /* attr->out_rep is resolved when we handle encap */
2516                         } else {
2517                                 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
2518                                        priv->netdev->name, out_dev->name);
2519                                 return -EINVAL;
2520                         }
2521                         continue;
2522                 }
2523
2524                 if (is_tcf_tunnel_set(a)) {
2525                         info = tcf_tunnel_info(a);
2526                         if (info)
2527                                 encap = true;
2528                         else
2529                                 return -EOPNOTSUPP;
2530                         continue;
2531                 }
2532
2533                 if (is_tcf_vlan(a)) {
2534                         if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
2535                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2536                         } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
2537                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
2538                                 attr->vlan_vid = tcf_vlan_push_vid(a);
2539                                 if (mlx5_eswitch_vlan_actions_supported(priv->mdev)) {
2540                                         attr->vlan_prio = tcf_vlan_push_prio(a);
2541                                         attr->vlan_proto = tcf_vlan_push_proto(a);
2542                                         if (!attr->vlan_proto)
2543                                                 attr->vlan_proto = htons(ETH_P_8021Q);
2544                                 } else if (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q) ||
2545                                            tcf_vlan_push_prio(a)) {
2546                                         return -EOPNOTSUPP;
2547                                 }
2548                         } else { /* action is TCA_VLAN_ACT_MODIFY */
2549                                 return -EOPNOTSUPP;
2550                         }
2551                         continue;
2552                 }
2553
2554                 if (is_tcf_tunnel_release(a)) {
2555                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
2556                         continue;
2557                 }
2558
2559                 return -EINVAL;
2560         }
2561
2562         if (!actions_match_supported(priv, exts, parse_attr, flow))
2563                 return -EOPNOTSUPP;
2564
2565         return 0;
2566 }
2567
2568 int mlx5e_configure_flower(struct mlx5e_priv *priv,
2569                            struct tc_cls_flower_offload *f)
2570 {
2571         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2572         struct mlx5e_tc_flow_parse_attr *parse_attr;
2573         struct mlx5e_tc_table *tc = &priv->fs.tc;
2574         struct mlx5e_tc_flow *flow;
2575         int attr_size, err = 0;
2576         u8 flow_flags = 0;
2577
2578         if (esw && esw->mode == SRIOV_OFFLOADS) {
2579                 flow_flags = MLX5E_TC_FLOW_ESWITCH;
2580                 attr_size  = sizeof(struct mlx5_esw_flow_attr);
2581         } else {
2582                 flow_flags = MLX5E_TC_FLOW_NIC;
2583                 attr_size  = sizeof(struct mlx5_nic_flow_attr);
2584         }
2585
2586         flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
2587         parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
2588         if (!parse_attr || !flow) {
2589                 err = -ENOMEM;
2590                 goto err_free;
2591         }
2592
2593         flow->cookie = f->cookie;
2594         flow->flags = flow_flags;
2595
2596         err = parse_cls_flower(priv, flow, &parse_attr->spec, f);
2597         if (err < 0)
2598                 goto err_free;
2599
2600         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
2601                 err = parse_tc_fdb_actions(priv, f->exts, parse_attr, flow);
2602                 if (err < 0)
2603                         goto err_free;
2604                 flow->rule = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow);
2605         } else {
2606                 err = parse_tc_nic_actions(priv, f->exts, parse_attr, flow);
2607                 if (err < 0)
2608                         goto err_free;
2609                 flow->rule = mlx5e_tc_add_nic_flow(priv, parse_attr, flow);
2610         }
2611
2612         if (IS_ERR(flow->rule)) {
2613                 err = PTR_ERR(flow->rule);
2614                 if (err != -EAGAIN)
2615                         goto err_free;
2616         }
2617
2618         if (err != -EAGAIN)
2619                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
2620
2621         if (!(flow->flags & MLX5E_TC_FLOW_ESWITCH) ||
2622             !(flow->esw_attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP))
2623                 kvfree(parse_attr);
2624
2625         err = rhashtable_insert_fast(&tc->ht, &flow->node,
2626                                      tc->ht_params);
2627         if (err) {
2628                 mlx5e_tc_del_flow(priv, flow);
2629                 kfree(flow);
2630         }
2631
2632         return err;
2633
2634 err_free:
2635         kvfree(parse_attr);
2636         kfree(flow);
2637         return err;
2638 }
2639
2640 int mlx5e_delete_flower(struct mlx5e_priv *priv,
2641                         struct tc_cls_flower_offload *f)
2642 {
2643         struct mlx5e_tc_flow *flow;
2644         struct mlx5e_tc_table *tc = &priv->fs.tc;
2645
2646         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
2647                                       tc->ht_params);
2648         if (!flow)
2649                 return -EINVAL;
2650
2651         rhashtable_remove_fast(&tc->ht, &flow->node, tc->ht_params);
2652
2653         mlx5e_tc_del_flow(priv, flow);
2654
2655         kfree(flow);
2656
2657         return 0;
2658 }
2659
2660 int mlx5e_stats_flower(struct mlx5e_priv *priv,
2661                        struct tc_cls_flower_offload *f)
2662 {
2663         struct mlx5e_tc_table *tc = &priv->fs.tc;
2664         struct mlx5e_tc_flow *flow;
2665         struct mlx5_fc *counter;
2666         u64 bytes;
2667         u64 packets;
2668         u64 lastuse;
2669
2670         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
2671                                       tc->ht_params);
2672         if (!flow)
2673                 return -EINVAL;
2674
2675         if (!(flow->flags & MLX5E_TC_FLOW_OFFLOADED))
2676                 return 0;
2677
2678         counter = mlx5_flow_rule_counter(flow->rule);
2679         if (!counter)
2680                 return 0;
2681
2682         mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
2683
2684         tcf_exts_stats_update(f->exts, bytes, packets, lastuse);
2685
2686         return 0;
2687 }
2688
2689 static const struct rhashtable_params mlx5e_tc_flow_ht_params = {
2690         .head_offset = offsetof(struct mlx5e_tc_flow, node),
2691         .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
2692         .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
2693         .automatic_shrinking = true,
2694 };
2695
2696 int mlx5e_tc_init(struct mlx5e_priv *priv)
2697 {
2698         struct mlx5e_tc_table *tc = &priv->fs.tc;
2699
2700         hash_init(tc->mod_hdr_tbl);
2701         hash_init(tc->hairpin_tbl);
2702
2703         tc->ht_params = mlx5e_tc_flow_ht_params;
2704         return rhashtable_init(&tc->ht, &tc->ht_params);
2705 }
2706
2707 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
2708 {
2709         struct mlx5e_tc_flow *flow = ptr;
2710         struct mlx5e_priv *priv = arg;
2711
2712         mlx5e_tc_del_flow(priv, flow);
2713         kfree(flow);
2714 }
2715
2716 void mlx5e_tc_cleanup(struct mlx5e_priv *priv)
2717 {
2718         struct mlx5e_tc_table *tc = &priv->fs.tc;
2719
2720         rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, priv);
2721
2722         if (!IS_ERR_OR_NULL(tc->t)) {
2723                 mlx5_destroy_flow_table(tc->t);
2724                 tc->t = NULL;
2725         }
2726 }