dcb8c4993811173781ba56bbecffd6d0f438860a
[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 #include "en/port.h"
56
57 struct mlx5_nic_flow_attr {
58         u32 action;
59         u32 flow_tag;
60         u32 mod_hdr_id;
61         u32 hairpin_tirn;
62         u8 match_level;
63         struct mlx5_flow_table  *hairpin_ft;
64 };
65
66 #define MLX5E_TC_FLOW_BASE (MLX5E_TC_LAST_EXPORTED_BIT + 1)
67
68 enum {
69         MLX5E_TC_FLOW_INGRESS   = MLX5E_TC_INGRESS,
70         MLX5E_TC_FLOW_EGRESS    = MLX5E_TC_EGRESS,
71         MLX5E_TC_FLOW_ESWITCH   = BIT(MLX5E_TC_FLOW_BASE),
72         MLX5E_TC_FLOW_NIC       = BIT(MLX5E_TC_FLOW_BASE + 1),
73         MLX5E_TC_FLOW_OFFLOADED = BIT(MLX5E_TC_FLOW_BASE + 2),
74         MLX5E_TC_FLOW_HAIRPIN   = BIT(MLX5E_TC_FLOW_BASE + 3),
75         MLX5E_TC_FLOW_HAIRPIN_RSS = BIT(MLX5E_TC_FLOW_BASE + 4),
76 };
77
78 #define MLX5E_TC_MAX_SPLITS 1
79
80 struct mlx5e_tc_flow {
81         struct rhash_head       node;
82         struct mlx5e_priv       *priv;
83         u64                     cookie;
84         u8                      flags;
85         struct mlx5_flow_handle *rule[MLX5E_TC_MAX_SPLITS + 1];
86         struct list_head        encap;   /* flows sharing the same encap ID */
87         struct list_head        mod_hdr; /* flows sharing the same mod hdr ID */
88         struct list_head        hairpin; /* flows sharing the same hairpin */
89         union {
90                 struct mlx5_esw_flow_attr esw_attr[0];
91                 struct mlx5_nic_flow_attr nic_attr[0];
92         };
93 };
94
95 struct mlx5e_tc_flow_parse_attr {
96         struct ip_tunnel_info tun_info;
97         struct mlx5_flow_spec spec;
98         int num_mod_hdr_actions;
99         void *mod_hdr_actions;
100         int mirred_ifindex;
101 };
102
103 enum {
104         MLX5_HEADER_TYPE_VXLAN = 0x0,
105         MLX5_HEADER_TYPE_NVGRE = 0x1,
106 };
107
108 #define MLX5E_TC_TABLE_NUM_GROUPS 4
109 #define MLX5E_TC_TABLE_MAX_GROUP_SIZE BIT(16)
110
111 struct mlx5e_hairpin {
112         struct mlx5_hairpin *pair;
113
114         struct mlx5_core_dev *func_mdev;
115         struct mlx5e_priv *func_priv;
116         u32 tdn;
117         u32 tirn;
118
119         int num_channels;
120         struct mlx5e_rqt indir_rqt;
121         u32 indir_tirn[MLX5E_NUM_INDIR_TIRS];
122         struct mlx5e_ttc_table ttc;
123 };
124
125 struct mlx5e_hairpin_entry {
126         /* a node of a hash table which keeps all the  hairpin entries */
127         struct hlist_node hairpin_hlist;
128
129         /* flows sharing the same hairpin */
130         struct list_head flows;
131
132         u16 peer_vhca_id;
133         u8 prio;
134         struct mlx5e_hairpin *hp;
135 };
136
137 struct mod_hdr_key {
138         int num_actions;
139         void *actions;
140 };
141
142 struct mlx5e_mod_hdr_entry {
143         /* a node of a hash table which keeps all the mod_hdr entries */
144         struct hlist_node mod_hdr_hlist;
145
146         /* flows sharing the same mod_hdr entry */
147         struct list_head flows;
148
149         struct mod_hdr_key key;
150
151         u32 mod_hdr_id;
152 };
153
154 #define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)
155
156 static inline u32 hash_mod_hdr_info(struct mod_hdr_key *key)
157 {
158         return jhash(key->actions,
159                      key->num_actions * MLX5_MH_ACT_SZ, 0);
160 }
161
162 static inline int cmp_mod_hdr_info(struct mod_hdr_key *a,
163                                    struct mod_hdr_key *b)
164 {
165         if (a->num_actions != b->num_actions)
166                 return 1;
167
168         return memcmp(a->actions, b->actions, a->num_actions * MLX5_MH_ACT_SZ);
169 }
170
171 static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
172                                 struct mlx5e_tc_flow *flow,
173                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
174 {
175         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
176         int num_actions, actions_size, namespace, err;
177         struct mlx5e_mod_hdr_entry *mh;
178         struct mod_hdr_key key;
179         bool found = false;
180         u32 hash_key;
181
182         num_actions  = parse_attr->num_mod_hdr_actions;
183         actions_size = MLX5_MH_ACT_SZ * num_actions;
184
185         key.actions = parse_attr->mod_hdr_actions;
186         key.num_actions = num_actions;
187
188         hash_key = hash_mod_hdr_info(&key);
189
190         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
191                 namespace = MLX5_FLOW_NAMESPACE_FDB;
192                 hash_for_each_possible(esw->offloads.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         } else {
200                 namespace = MLX5_FLOW_NAMESPACE_KERNEL;
201                 hash_for_each_possible(priv->fs.tc.mod_hdr_tbl, mh,
202                                        mod_hdr_hlist, hash_key) {
203                         if (!cmp_mod_hdr_info(&mh->key, &key)) {
204                                 found = true;
205                                 break;
206                         }
207                 }
208         }
209
210         if (found)
211                 goto attach_flow;
212
213         mh = kzalloc(sizeof(*mh) + actions_size, GFP_KERNEL);
214         if (!mh)
215                 return -ENOMEM;
216
217         mh->key.actions = (void *)mh + sizeof(*mh);
218         memcpy(mh->key.actions, key.actions, actions_size);
219         mh->key.num_actions = num_actions;
220         INIT_LIST_HEAD(&mh->flows);
221
222         err = mlx5_modify_header_alloc(priv->mdev, namespace,
223                                        mh->key.num_actions,
224                                        mh->key.actions,
225                                        &mh->mod_hdr_id);
226         if (err)
227                 goto out_err;
228
229         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
230                 hash_add(esw->offloads.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
231         else
232                 hash_add(priv->fs.tc.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
233
234 attach_flow:
235         list_add(&flow->mod_hdr, &mh->flows);
236         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
237                 flow->esw_attr->mod_hdr_id = mh->mod_hdr_id;
238         else
239                 flow->nic_attr->mod_hdr_id = mh->mod_hdr_id;
240
241         return 0;
242
243 out_err:
244         kfree(mh);
245         return err;
246 }
247
248 static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
249                                  struct mlx5e_tc_flow *flow)
250 {
251         struct list_head *next = flow->mod_hdr.next;
252
253         list_del(&flow->mod_hdr);
254
255         if (list_empty(next)) {
256                 struct mlx5e_mod_hdr_entry *mh;
257
258                 mh = list_entry(next, struct mlx5e_mod_hdr_entry, flows);
259
260                 mlx5_modify_header_dealloc(priv->mdev, mh->mod_hdr_id);
261                 hash_del(&mh->mod_hdr_hlist);
262                 kfree(mh);
263         }
264 }
265
266 static
267 struct mlx5_core_dev *mlx5e_hairpin_get_mdev(struct net *net, int ifindex)
268 {
269         struct net_device *netdev;
270         struct mlx5e_priv *priv;
271
272         netdev = __dev_get_by_index(net, ifindex);
273         priv = netdev_priv(netdev);
274         return priv->mdev;
275 }
276
277 static int mlx5e_hairpin_create_transport(struct mlx5e_hairpin *hp)
278 {
279         u32 in[MLX5_ST_SZ_DW(create_tir_in)] = {0};
280         void *tirc;
281         int err;
282
283         err = mlx5_core_alloc_transport_domain(hp->func_mdev, &hp->tdn);
284         if (err)
285                 goto alloc_tdn_err;
286
287         tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
288
289         MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
290         MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn[0]);
291         MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
292
293         err = mlx5_core_create_tir(hp->func_mdev, in, MLX5_ST_SZ_BYTES(create_tir_in), &hp->tirn);
294         if (err)
295                 goto create_tir_err;
296
297         return 0;
298
299 create_tir_err:
300         mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
301 alloc_tdn_err:
302         return err;
303 }
304
305 static void mlx5e_hairpin_destroy_transport(struct mlx5e_hairpin *hp)
306 {
307         mlx5_core_destroy_tir(hp->func_mdev, hp->tirn);
308         mlx5_core_dealloc_transport_domain(hp->func_mdev, hp->tdn);
309 }
310
311 static void mlx5e_hairpin_fill_rqt_rqns(struct mlx5e_hairpin *hp, void *rqtc)
312 {
313         u32 indirection_rqt[MLX5E_INDIR_RQT_SIZE], rqn;
314         struct mlx5e_priv *priv = hp->func_priv;
315         int i, ix, sz = MLX5E_INDIR_RQT_SIZE;
316
317         mlx5e_build_default_indir_rqt(indirection_rqt, sz,
318                                       hp->num_channels);
319
320         for (i = 0; i < sz; i++) {
321                 ix = i;
322                 if (priv->channels.params.rss_hfunc == ETH_RSS_HASH_XOR)
323                         ix = mlx5e_bits_invert(i, ilog2(sz));
324                 ix = indirection_rqt[ix];
325                 rqn = hp->pair->rqn[ix];
326                 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
327         }
328 }
329
330 static int mlx5e_hairpin_create_indirect_rqt(struct mlx5e_hairpin *hp)
331 {
332         int inlen, err, sz = MLX5E_INDIR_RQT_SIZE;
333         struct mlx5e_priv *priv = hp->func_priv;
334         struct mlx5_core_dev *mdev = priv->mdev;
335         void *rqtc;
336         u32 *in;
337
338         inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
339         in = kvzalloc(inlen, GFP_KERNEL);
340         if (!in)
341                 return -ENOMEM;
342
343         rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
344
345         MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
346         MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
347
348         mlx5e_hairpin_fill_rqt_rqns(hp, rqtc);
349
350         err = mlx5_core_create_rqt(mdev, in, inlen, &hp->indir_rqt.rqtn);
351         if (!err)
352                 hp->indir_rqt.enabled = true;
353
354         kvfree(in);
355         return err;
356 }
357
358 static int mlx5e_hairpin_create_indirect_tirs(struct mlx5e_hairpin *hp)
359 {
360         struct mlx5e_priv *priv = hp->func_priv;
361         u32 in[MLX5_ST_SZ_DW(create_tir_in)];
362         int tt, i, err;
363         void *tirc;
364
365         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
366                 memset(in, 0, MLX5_ST_SZ_BYTES(create_tir_in));
367                 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
368
369                 MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
370                 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
371                 MLX5_SET(tirc, tirc, indirect_table, hp->indir_rqt.rqtn);
372                 mlx5e_build_indir_tir_ctx_hash(&priv->channels.params, tt, tirc, false);
373
374                 err = mlx5_core_create_tir(hp->func_mdev, in,
375                                            MLX5_ST_SZ_BYTES(create_tir_in), &hp->indir_tirn[tt]);
376                 if (err) {
377                         mlx5_core_warn(hp->func_mdev, "create indirect tirs failed, %d\n", err);
378                         goto err_destroy_tirs;
379                 }
380         }
381         return 0;
382
383 err_destroy_tirs:
384         for (i = 0; i < tt; i++)
385                 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[i]);
386         return err;
387 }
388
389 static void mlx5e_hairpin_destroy_indirect_tirs(struct mlx5e_hairpin *hp)
390 {
391         int tt;
392
393         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
394                 mlx5_core_destroy_tir(hp->func_mdev, hp->indir_tirn[tt]);
395 }
396
397 static void mlx5e_hairpin_set_ttc_params(struct mlx5e_hairpin *hp,
398                                          struct ttc_params *ttc_params)
399 {
400         struct mlx5_flow_table_attr *ft_attr = &ttc_params->ft_attr;
401         int tt;
402
403         memset(ttc_params, 0, sizeof(*ttc_params));
404
405         ttc_params->any_tt_tirn = hp->tirn;
406
407         for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
408                 ttc_params->indir_tirn[tt] = hp->indir_tirn[tt];
409
410         ft_attr->max_fte = MLX5E_NUM_TT;
411         ft_attr->level = MLX5E_TC_TTC_FT_LEVEL;
412         ft_attr->prio = MLX5E_TC_PRIO;
413 }
414
415 static int mlx5e_hairpin_rss_init(struct mlx5e_hairpin *hp)
416 {
417         struct mlx5e_priv *priv = hp->func_priv;
418         struct ttc_params ttc_params;
419         int err;
420
421         err = mlx5e_hairpin_create_indirect_rqt(hp);
422         if (err)
423                 return err;
424
425         err = mlx5e_hairpin_create_indirect_tirs(hp);
426         if (err)
427                 goto err_create_indirect_tirs;
428
429         mlx5e_hairpin_set_ttc_params(hp, &ttc_params);
430         err = mlx5e_create_ttc_table(priv, &ttc_params, &hp->ttc);
431         if (err)
432                 goto err_create_ttc_table;
433
434         netdev_dbg(priv->netdev, "add hairpin: using %d channels rss ttc table id %x\n",
435                    hp->num_channels, hp->ttc.ft.t->id);
436
437         return 0;
438
439 err_create_ttc_table:
440         mlx5e_hairpin_destroy_indirect_tirs(hp);
441 err_create_indirect_tirs:
442         mlx5e_destroy_rqt(priv, &hp->indir_rqt);
443
444         return err;
445 }
446
447 static void mlx5e_hairpin_rss_cleanup(struct mlx5e_hairpin *hp)
448 {
449         struct mlx5e_priv *priv = hp->func_priv;
450
451         mlx5e_destroy_ttc_table(priv, &hp->ttc);
452         mlx5e_hairpin_destroy_indirect_tirs(hp);
453         mlx5e_destroy_rqt(priv, &hp->indir_rqt);
454 }
455
456 static struct mlx5e_hairpin *
457 mlx5e_hairpin_create(struct mlx5e_priv *priv, struct mlx5_hairpin_params *params,
458                      int peer_ifindex)
459 {
460         struct mlx5_core_dev *func_mdev, *peer_mdev;
461         struct mlx5e_hairpin *hp;
462         struct mlx5_hairpin *pair;
463         int err;
464
465         hp = kzalloc(sizeof(*hp), GFP_KERNEL);
466         if (!hp)
467                 return ERR_PTR(-ENOMEM);
468
469         func_mdev = priv->mdev;
470         peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
471
472         pair = mlx5_core_hairpin_create(func_mdev, peer_mdev, params);
473         if (IS_ERR(pair)) {
474                 err = PTR_ERR(pair);
475                 goto create_pair_err;
476         }
477         hp->pair = pair;
478         hp->func_mdev = func_mdev;
479         hp->func_priv = priv;
480         hp->num_channels = params->num_channels;
481
482         err = mlx5e_hairpin_create_transport(hp);
483         if (err)
484                 goto create_transport_err;
485
486         if (hp->num_channels > 1) {
487                 err = mlx5e_hairpin_rss_init(hp);
488                 if (err)
489                         goto rss_init_err;
490         }
491
492         return hp;
493
494 rss_init_err:
495         mlx5e_hairpin_destroy_transport(hp);
496 create_transport_err:
497         mlx5_core_hairpin_destroy(hp->pair);
498 create_pair_err:
499         kfree(hp);
500         return ERR_PTR(err);
501 }
502
503 static void mlx5e_hairpin_destroy(struct mlx5e_hairpin *hp)
504 {
505         if (hp->num_channels > 1)
506                 mlx5e_hairpin_rss_cleanup(hp);
507         mlx5e_hairpin_destroy_transport(hp);
508         mlx5_core_hairpin_destroy(hp->pair);
509         kvfree(hp);
510 }
511
512 static inline u32 hash_hairpin_info(u16 peer_vhca_id, u8 prio)
513 {
514         return (peer_vhca_id << 16 | prio);
515 }
516
517 static struct mlx5e_hairpin_entry *mlx5e_hairpin_get(struct mlx5e_priv *priv,
518                                                      u16 peer_vhca_id, u8 prio)
519 {
520         struct mlx5e_hairpin_entry *hpe;
521         u32 hash_key = hash_hairpin_info(peer_vhca_id, prio);
522
523         hash_for_each_possible(priv->fs.tc.hairpin_tbl, hpe,
524                                hairpin_hlist, hash_key) {
525                 if (hpe->peer_vhca_id == peer_vhca_id && hpe->prio == prio)
526                         return hpe;
527         }
528
529         return NULL;
530 }
531
532 #define UNKNOWN_MATCH_PRIO 8
533
534 static int mlx5e_hairpin_get_prio(struct mlx5e_priv *priv,
535                                   struct mlx5_flow_spec *spec, u8 *match_prio)
536 {
537         void *headers_c, *headers_v;
538         u8 prio_val, prio_mask = 0;
539         bool vlan_present;
540
541 #ifdef CONFIG_MLX5_CORE_EN_DCB
542         if (priv->dcbx_dp.trust_state != MLX5_QPTS_TRUST_PCP) {
543                 netdev_warn(priv->netdev,
544                             "only PCP trust state supported for hairpin\n");
545                 return -EOPNOTSUPP;
546         }
547 #endif
548         headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, outer_headers);
549         headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
550
551         vlan_present = MLX5_GET(fte_match_set_lyr_2_4, headers_v, cvlan_tag);
552         if (vlan_present) {
553                 prio_mask = MLX5_GET(fte_match_set_lyr_2_4, headers_c, first_prio);
554                 prio_val = MLX5_GET(fte_match_set_lyr_2_4, headers_v, first_prio);
555         }
556
557         if (!vlan_present || !prio_mask) {
558                 prio_val = UNKNOWN_MATCH_PRIO;
559         } else if (prio_mask != 0x7) {
560                 netdev_warn(priv->netdev,
561                             "masked priority match not supported for hairpin\n");
562                 return -EOPNOTSUPP;
563         }
564
565         *match_prio = prio_val;
566         return 0;
567 }
568
569 static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
570                                   struct mlx5e_tc_flow *flow,
571                                   struct mlx5e_tc_flow_parse_attr *parse_attr)
572 {
573         int peer_ifindex = parse_attr->mirred_ifindex;
574         struct mlx5_hairpin_params params;
575         struct mlx5_core_dev *peer_mdev;
576         struct mlx5e_hairpin_entry *hpe;
577         struct mlx5e_hairpin *hp;
578         u64 link_speed64;
579         u32 link_speed;
580         u8 match_prio;
581         u16 peer_id;
582         int err;
583
584         peer_mdev = mlx5e_hairpin_get_mdev(dev_net(priv->netdev), peer_ifindex);
585         if (!MLX5_CAP_GEN(priv->mdev, hairpin) || !MLX5_CAP_GEN(peer_mdev, hairpin)) {
586                 netdev_warn(priv->netdev, "hairpin is not supported\n");
587                 return -EOPNOTSUPP;
588         }
589
590         peer_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
591         err = mlx5e_hairpin_get_prio(priv, &parse_attr->spec, &match_prio);
592         if (err)
593                 return err;
594         hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
595         if (hpe)
596                 goto attach_flow;
597
598         hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
599         if (!hpe)
600                 return -ENOMEM;
601
602         INIT_LIST_HEAD(&hpe->flows);
603         hpe->peer_vhca_id = peer_id;
604         hpe->prio = match_prio;
605
606         params.log_data_size = 15;
607         params.log_data_size = min_t(u8, params.log_data_size,
608                                      MLX5_CAP_GEN(priv->mdev, log_max_hairpin_wq_data_sz));
609         params.log_data_size = max_t(u8, params.log_data_size,
610                                      MLX5_CAP_GEN(priv->mdev, log_min_hairpin_wq_data_sz));
611
612         params.log_num_packets = params.log_data_size -
613                                  MLX5_MPWRQ_MIN_LOG_STRIDE_SZ(priv->mdev);
614         params.log_num_packets = min_t(u8, params.log_num_packets,
615                                        MLX5_CAP_GEN(priv->mdev, log_max_hairpin_num_packets));
616
617         params.q_counter = priv->q_counter;
618         /* set hairpin pair per each 50Gbs share of the link */
619         mlx5e_port_max_linkspeed(priv->mdev, &link_speed);
620         link_speed = max_t(u32, link_speed, 50000);
621         link_speed64 = link_speed;
622         do_div(link_speed64, 50000);
623         params.num_channels = link_speed64;
624
625         hp = mlx5e_hairpin_create(priv, &params, peer_ifindex);
626         if (IS_ERR(hp)) {
627                 err = PTR_ERR(hp);
628                 goto create_hairpin_err;
629         }
630
631         netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
632                    hp->tirn, hp->pair->rqn[0], hp->pair->peer_mdev->priv.name,
633                    hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
634
635         hpe->hp = hp;
636         hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
637                  hash_hairpin_info(peer_id, match_prio));
638
639 attach_flow:
640         if (hpe->hp->num_channels > 1) {
641                 flow->flags |= MLX5E_TC_FLOW_HAIRPIN_RSS;
642                 flow->nic_attr->hairpin_ft = hpe->hp->ttc.ft.t;
643         } else {
644                 flow->nic_attr->hairpin_tirn = hpe->hp->tirn;
645         }
646         list_add(&flow->hairpin, &hpe->flows);
647
648         return 0;
649
650 create_hairpin_err:
651         kfree(hpe);
652         return err;
653 }
654
655 static void mlx5e_hairpin_flow_del(struct mlx5e_priv *priv,
656                                    struct mlx5e_tc_flow *flow)
657 {
658         struct list_head *next = flow->hairpin.next;
659
660         list_del(&flow->hairpin);
661
662         /* no more hairpin flows for us, release the hairpin pair */
663         if (list_empty(next)) {
664                 struct mlx5e_hairpin_entry *hpe;
665
666                 hpe = list_entry(next, struct mlx5e_hairpin_entry, flows);
667
668                 netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
669                            hpe->hp->pair->peer_mdev->priv.name);
670
671                 mlx5e_hairpin_destroy(hpe->hp);
672                 hash_del(&hpe->hairpin_hlist);
673                 kfree(hpe);
674         }
675 }
676
677 static struct mlx5_flow_handle *
678 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
679                       struct mlx5e_tc_flow_parse_attr *parse_attr,
680                       struct mlx5e_tc_flow *flow)
681 {
682         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
683         struct mlx5_core_dev *dev = priv->mdev;
684         struct mlx5_flow_destination dest[2] = {};
685         struct mlx5_flow_act flow_act = {
686                 .action = attr->action,
687                 .has_flow_tag = true,
688                 .flow_tag = attr->flow_tag,
689                 .encap_id = 0,
690         };
691         struct mlx5_fc *counter = NULL;
692         struct mlx5_flow_handle *rule;
693         bool table_created = false;
694         int err, dest_ix = 0;
695
696         if (flow->flags & MLX5E_TC_FLOW_HAIRPIN) {
697                 err = mlx5e_hairpin_flow_add(priv, flow, parse_attr);
698                 if (err) {
699                         rule = ERR_PTR(err);
700                         goto err_add_hairpin_flow;
701                 }
702                 if (flow->flags & MLX5E_TC_FLOW_HAIRPIN_RSS) {
703                         dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
704                         dest[dest_ix].ft = attr->hairpin_ft;
705                 } else {
706                         dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_TIR;
707                         dest[dest_ix].tir_num = attr->hairpin_tirn;
708                 }
709                 dest_ix++;
710         } else if (attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
711                 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
712                 dest[dest_ix].ft = priv->fs.vlan.ft.t;
713                 dest_ix++;
714         }
715
716         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
717                 counter = mlx5_fc_create(dev, true);
718                 if (IS_ERR(counter)) {
719                         rule = ERR_CAST(counter);
720                         goto err_fc_create;
721                 }
722                 dest[dest_ix].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
723                 dest[dest_ix].counter = counter;
724                 dest_ix++;
725         }
726
727         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
728                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
729                 flow_act.modify_id = attr->mod_hdr_id;
730                 kfree(parse_attr->mod_hdr_actions);
731                 if (err) {
732                         rule = ERR_PTR(err);
733                         goto err_create_mod_hdr_id;
734                 }
735         }
736
737         if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
738                 int tc_grp_size, tc_tbl_size;
739                 u32 max_flow_counter;
740
741                 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
742                                     MLX5_CAP_GEN(dev, max_flow_counter_15_0);
743
744                 tc_grp_size = min_t(int, max_flow_counter, MLX5E_TC_TABLE_MAX_GROUP_SIZE);
745
746                 tc_tbl_size = min_t(int, tc_grp_size * MLX5E_TC_TABLE_NUM_GROUPS,
747                                     BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev, log_max_ft_size)));
748
749                 priv->fs.tc.t =
750                         mlx5_create_auto_grouped_flow_table(priv->fs.ns,
751                                                             MLX5E_TC_PRIO,
752                                                             tc_tbl_size,
753                                                             MLX5E_TC_TABLE_NUM_GROUPS,
754                                                             MLX5E_TC_FT_LEVEL, 0);
755                 if (IS_ERR(priv->fs.tc.t)) {
756                         netdev_err(priv->netdev,
757                                    "Failed to create tc offload table\n");
758                         rule = ERR_CAST(priv->fs.tc.t);
759                         goto err_create_ft;
760                 }
761
762                 table_created = true;
763         }
764
765         if (attr->match_level != MLX5_MATCH_NONE)
766                 parse_attr->spec.match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
767
768         rule = mlx5_add_flow_rules(priv->fs.tc.t, &parse_attr->spec,
769                                    &flow_act, dest, dest_ix);
770
771         if (IS_ERR(rule))
772                 goto err_add_rule;
773
774         return rule;
775
776 err_add_rule:
777         if (table_created) {
778                 mlx5_destroy_flow_table(priv->fs.tc.t);
779                 priv->fs.tc.t = NULL;
780         }
781 err_create_ft:
782         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
783                 mlx5e_detach_mod_hdr(priv, flow);
784 err_create_mod_hdr_id:
785         mlx5_fc_destroy(dev, counter);
786 err_fc_create:
787         if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
788                 mlx5e_hairpin_flow_del(priv, flow);
789 err_add_hairpin_flow:
790         return rule;
791 }
792
793 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
794                                   struct mlx5e_tc_flow *flow)
795 {
796         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
797         struct mlx5_fc *counter = NULL;
798
799         counter = mlx5_flow_rule_counter(flow->rule[0]);
800         mlx5_del_flow_rules(flow->rule[0]);
801         mlx5_fc_destroy(priv->mdev, counter);
802
803         if (!mlx5e_tc_num_filters(priv) && priv->fs.tc.t) {
804                 mlx5_destroy_flow_table(priv->fs.tc.t);
805                 priv->fs.tc.t = NULL;
806         }
807
808         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
809                 mlx5e_detach_mod_hdr(priv, flow);
810
811         if (flow->flags & MLX5E_TC_FLOW_HAIRPIN)
812                 mlx5e_hairpin_flow_del(priv, flow);
813 }
814
815 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
816                                struct mlx5e_tc_flow *flow);
817
818 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
819                               struct ip_tunnel_info *tun_info,
820                               struct net_device *mirred_dev,
821                               struct net_device **encap_dev,
822                               struct mlx5e_tc_flow *flow);
823
824 static struct mlx5_flow_handle *
825 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
826                       struct mlx5e_tc_flow_parse_attr *parse_attr,
827                       struct mlx5e_tc_flow *flow)
828 {
829         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
830         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
831         struct net_device *out_dev, *encap_dev = NULL;
832         struct mlx5_flow_handle *rule = NULL;
833         struct mlx5e_rep_priv *rpriv;
834         struct mlx5e_priv *out_priv;
835         int err;
836
837         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP) {
838                 out_dev = __dev_get_by_index(dev_net(priv->netdev),
839                                              attr->parse_attr->mirred_ifindex);
840                 err = mlx5e_attach_encap(priv, &parse_attr->tun_info,
841                                          out_dev, &encap_dev, flow);
842                 if (err) {
843                         rule = ERR_PTR(err);
844                         if (err != -EAGAIN)
845                                 goto err_attach_encap;
846                 }
847                 out_priv = netdev_priv(encap_dev);
848                 rpriv = out_priv->ppriv;
849                 attr->out_rep[attr->out_count] = rpriv->rep;
850                 attr->out_mdev[attr->out_count++] = out_priv->mdev;
851         }
852
853         err = mlx5_eswitch_add_vlan_action(esw, attr);
854         if (err) {
855                 rule = ERR_PTR(err);
856                 goto err_add_vlan;
857         }
858
859         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
860                 err = mlx5e_attach_mod_hdr(priv, flow, parse_attr);
861                 kfree(parse_attr->mod_hdr_actions);
862                 if (err) {
863                         rule = ERR_PTR(err);
864                         goto err_mod_hdr;
865                 }
866         }
867
868         /* we get here if (1) there's no error (rule being null) or when
869          * (2) there's an encap action and we're on -EAGAIN (no valid neigh)
870          */
871         if (rule != ERR_PTR(-EAGAIN)) {
872                 rule = mlx5_eswitch_add_offloaded_rule(esw, &parse_attr->spec, attr);
873                 if (IS_ERR(rule))
874                         goto err_add_rule;
875
876                 if (attr->mirror_count) {
877                         flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, &parse_attr->spec, attr);
878                         if (IS_ERR(flow->rule[1]))
879                                 goto err_fwd_rule;
880                 }
881         }
882         return rule;
883
884 err_fwd_rule:
885         mlx5_eswitch_del_offloaded_rule(esw, rule, attr);
886         rule = flow->rule[1];
887 err_add_rule:
888         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
889                 mlx5e_detach_mod_hdr(priv, flow);
890 err_mod_hdr:
891         mlx5_eswitch_del_vlan_action(esw, attr);
892 err_add_vlan:
893         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP)
894                 mlx5e_detach_encap(priv, flow);
895 err_attach_encap:
896         return rule;
897 }
898
899 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
900                                   struct mlx5e_tc_flow *flow)
901 {
902         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
903         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
904
905         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
906                 flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
907                 if (attr->mirror_count)
908                         mlx5_eswitch_del_offloaded_rule(esw, flow->rule[1], attr);
909                 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
910         }
911
912         mlx5_eswitch_del_vlan_action(esw, attr);
913
914         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP) {
915                 mlx5e_detach_encap(priv, flow);
916                 kvfree(attr->parse_attr);
917         }
918
919         if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
920                 mlx5e_detach_mod_hdr(priv, flow);
921 }
922
923 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,
924                               struct mlx5e_encap_entry *e)
925 {
926         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
927         struct mlx5_esw_flow_attr *esw_attr;
928         struct mlx5e_tc_flow *flow;
929         int err;
930
931         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
932                                e->encap_size, e->encap_header,
933                                &e->encap_id);
934         if (err) {
935                 mlx5_core_warn(priv->mdev, "Failed to offload cached encapsulation header, %d\n",
936                                err);
937                 return;
938         }
939         e->flags |= MLX5_ENCAP_ENTRY_VALID;
940         mlx5e_rep_queue_neigh_stats_work(priv);
941
942         list_for_each_entry(flow, &e->flows, encap) {
943                 esw_attr = flow->esw_attr;
944                 esw_attr->encap_id = e->encap_id;
945                 flow->rule[0] = mlx5_eswitch_add_offloaded_rule(esw, &esw_attr->parse_attr->spec, esw_attr);
946                 if (IS_ERR(flow->rule[0])) {
947                         err = PTR_ERR(flow->rule[0]);
948                         mlx5_core_warn(priv->mdev, "Failed to update cached encapsulation flow, %d\n",
949                                        err);
950                         continue;
951                 }
952
953                 if (esw_attr->mirror_count) {
954                         flow->rule[1] = mlx5_eswitch_add_fwd_rule(esw, &esw_attr->parse_attr->spec, esw_attr);
955                         if (IS_ERR(flow->rule[1])) {
956                                 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], esw_attr);
957                                 err = PTR_ERR(flow->rule[1]);
958                                 mlx5_core_warn(priv->mdev, "Failed to update cached mirror flow, %d\n",
959                                                err);
960                                 continue;
961                         }
962                 }
963
964                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
965         }
966 }
967
968 void mlx5e_tc_encap_flows_del(struct mlx5e_priv *priv,
969                               struct mlx5e_encap_entry *e)
970 {
971         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
972         struct mlx5e_tc_flow *flow;
973
974         list_for_each_entry(flow, &e->flows, encap) {
975                 if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
976                         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
977
978                         flow->flags &= ~MLX5E_TC_FLOW_OFFLOADED;
979                         if (attr->mirror_count)
980                                 mlx5_eswitch_del_offloaded_rule(esw, flow->rule[1], attr);
981                         mlx5_eswitch_del_offloaded_rule(esw, flow->rule[0], attr);
982                 }
983         }
984
985         if (e->flags & MLX5_ENCAP_ENTRY_VALID) {
986                 e->flags &= ~MLX5_ENCAP_ENTRY_VALID;
987                 mlx5_encap_dealloc(priv->mdev, e->encap_id);
988         }
989 }
990
991 void mlx5e_tc_update_neigh_used_value(struct mlx5e_neigh_hash_entry *nhe)
992 {
993         struct mlx5e_neigh *m_neigh = &nhe->m_neigh;
994         u64 bytes, packets, lastuse = 0;
995         struct mlx5e_tc_flow *flow;
996         struct mlx5e_encap_entry *e;
997         struct mlx5_fc *counter;
998         struct neigh_table *tbl;
999         bool neigh_used = false;
1000         struct neighbour *n;
1001
1002         if (m_neigh->family == AF_INET)
1003                 tbl = &arp_tbl;
1004 #if IS_ENABLED(CONFIG_IPV6)
1005         else if (m_neigh->family == AF_INET6)
1006                 tbl = &nd_tbl;
1007 #endif
1008         else
1009                 return;
1010
1011         list_for_each_entry(e, &nhe->encap_list, encap_list) {
1012                 if (!(e->flags & MLX5_ENCAP_ENTRY_VALID))
1013                         continue;
1014                 list_for_each_entry(flow, &e->flows, encap) {
1015                         if (flow->flags & MLX5E_TC_FLOW_OFFLOADED) {
1016                                 counter = mlx5_flow_rule_counter(flow->rule[0]);
1017                                 mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
1018                                 if (time_after((unsigned long)lastuse, nhe->reported_lastuse)) {
1019                                         neigh_used = true;
1020                                         break;
1021                                 }
1022                         }
1023                 }
1024                 if (neigh_used)
1025                         break;
1026         }
1027
1028         if (neigh_used) {
1029                 nhe->reported_lastuse = jiffies;
1030
1031                 /* find the relevant neigh according to the cached device and
1032                  * dst ip pair
1033                  */
1034                 n = neigh_lookup(tbl, &m_neigh->dst_ip, m_neigh->dev);
1035                 if (!n)
1036                         return;
1037
1038                 neigh_event_send(n, NULL);
1039                 neigh_release(n);
1040         }
1041 }
1042
1043 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
1044                                struct mlx5e_tc_flow *flow)
1045 {
1046         struct list_head *next = flow->encap.next;
1047
1048         list_del(&flow->encap);
1049         if (list_empty(next)) {
1050                 struct mlx5e_encap_entry *e;
1051
1052                 e = list_entry(next, struct mlx5e_encap_entry, flows);
1053                 mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
1054
1055                 if (e->flags & MLX5_ENCAP_ENTRY_VALID)
1056                         mlx5_encap_dealloc(priv->mdev, e->encap_id);
1057
1058                 hash_del_rcu(&e->encap_hlist);
1059                 kfree(e->encap_header);
1060                 kfree(e);
1061         }
1062 }
1063
1064 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
1065                               struct mlx5e_tc_flow *flow)
1066 {
1067         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1068                 mlx5e_tc_del_fdb_flow(priv, flow);
1069         else
1070                 mlx5e_tc_del_nic_flow(priv, flow);
1071 }
1072
1073 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
1074                              struct tc_cls_flower_offload *f)
1075 {
1076         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1077                                        outer_headers);
1078         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1079                                        outer_headers);
1080         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1081                                     misc_parameters);
1082         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1083                                     misc_parameters);
1084
1085         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
1086         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
1087
1088         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
1089                 struct flow_dissector_key_keyid *key =
1090                         skb_flow_dissector_target(f->dissector,
1091                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
1092                                                   f->key);
1093                 struct flow_dissector_key_keyid *mask =
1094                         skb_flow_dissector_target(f->dissector,
1095                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
1096                                                   f->mask);
1097                 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
1098                          be32_to_cpu(mask->keyid));
1099                 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
1100                          be32_to_cpu(key->keyid));
1101         }
1102 }
1103
1104 static int parse_tunnel_attr(struct mlx5e_priv *priv,
1105                              struct mlx5_flow_spec *spec,
1106                              struct tc_cls_flower_offload *f)
1107 {
1108         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1109                                        outer_headers);
1110         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1111                                        outer_headers);
1112
1113         struct flow_dissector_key_control *enc_control =
1114                 skb_flow_dissector_target(f->dissector,
1115                                           FLOW_DISSECTOR_KEY_ENC_CONTROL,
1116                                           f->key);
1117
1118         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
1119                 struct flow_dissector_key_ports *key =
1120                         skb_flow_dissector_target(f->dissector,
1121                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
1122                                                   f->key);
1123                 struct flow_dissector_key_ports *mask =
1124                         skb_flow_dissector_target(f->dissector,
1125                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
1126                                                   f->mask);
1127                 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1128                 struct mlx5e_rep_priv *uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
1129                 struct net_device *up_dev = uplink_rpriv->netdev;
1130                 struct mlx5e_priv *up_priv = netdev_priv(up_dev);
1131
1132                 /* Full udp dst port must be given */
1133                 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
1134                         goto vxlan_match_offload_err;
1135
1136                 if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->dst)) &&
1137                     MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
1138                         parse_vxlan_attr(spec, f);
1139                 else {
1140                         netdev_warn(priv->netdev,
1141                                     "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
1142                         return -EOPNOTSUPP;
1143                 }
1144
1145                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1146                          udp_dport, ntohs(mask->dst));
1147                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1148                          udp_dport, ntohs(key->dst));
1149
1150                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1151                          udp_sport, ntohs(mask->src));
1152                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1153                          udp_sport, ntohs(key->src));
1154         } else { /* udp dst port must be given */
1155 vxlan_match_offload_err:
1156                 netdev_warn(priv->netdev,
1157                             "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
1158                 return -EOPNOTSUPP;
1159         }
1160
1161         if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1162                 struct flow_dissector_key_ipv4_addrs *key =
1163                         skb_flow_dissector_target(f->dissector,
1164                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1165                                                   f->key);
1166                 struct flow_dissector_key_ipv4_addrs *mask =
1167                         skb_flow_dissector_target(f->dissector,
1168                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
1169                                                   f->mask);
1170                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1171                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
1172                          ntohl(mask->src));
1173                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1174                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
1175                          ntohl(key->src));
1176
1177                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1178                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1179                          ntohl(mask->dst));
1180                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1181                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
1182                          ntohl(key->dst));
1183
1184                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1185                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
1186         } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1187                 struct flow_dissector_key_ipv6_addrs *key =
1188                         skb_flow_dissector_target(f->dissector,
1189                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1190                                                   f->key);
1191                 struct flow_dissector_key_ipv6_addrs *mask =
1192                         skb_flow_dissector_target(f->dissector,
1193                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
1194                                                   f->mask);
1195
1196                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1197                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1198                        &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1199                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1200                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1201                        &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1202
1203                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1204                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1205                        &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1206                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1207                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1208                        &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
1209
1210                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
1211                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
1212         }
1213
1214         /* Enforce DMAC when offloading incoming tunneled flows.
1215          * Flow counters require a match on the DMAC.
1216          */
1217         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
1218         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
1219         ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1220                                      dmac_47_16), priv->netdev->dev_addr);
1221
1222         /* let software handle IP fragments */
1223         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1224         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
1225
1226         return 0;
1227 }
1228
1229 static int __parse_cls_flower(struct mlx5e_priv *priv,
1230                               struct mlx5_flow_spec *spec,
1231                               struct tc_cls_flower_offload *f,
1232                               u8 *match_level)
1233 {
1234         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1235                                        outer_headers);
1236         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1237                                        outer_headers);
1238         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1239                                     misc_parameters);
1240         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1241                                     misc_parameters);
1242         u16 addr_type = 0;
1243         u8 ip_proto = 0;
1244
1245         *match_level = MLX5_MATCH_NONE;
1246
1247         if (f->dissector->used_keys &
1248             ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
1249               BIT(FLOW_DISSECTOR_KEY_BASIC) |
1250               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
1251               BIT(FLOW_DISSECTOR_KEY_VLAN) |
1252               BIT(FLOW_DISSECTOR_KEY_CVLAN) |
1253               BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
1254               BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
1255               BIT(FLOW_DISSECTOR_KEY_PORTS) |
1256               BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
1257               BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
1258               BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
1259               BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
1260               BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
1261               BIT(FLOW_DISSECTOR_KEY_TCP) |
1262               BIT(FLOW_DISSECTOR_KEY_IP))) {
1263                 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
1264                             f->dissector->used_keys);
1265                 return -EOPNOTSUPP;
1266         }
1267
1268         if ((dissector_uses_key(f->dissector,
1269                                 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
1270              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
1271              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
1272             dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
1273                 struct flow_dissector_key_control *key =
1274                         skb_flow_dissector_target(f->dissector,
1275                                                   FLOW_DISSECTOR_KEY_ENC_CONTROL,
1276                                                   f->key);
1277                 switch (key->addr_type) {
1278                 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
1279                 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
1280                         if (parse_tunnel_attr(priv, spec, f))
1281                                 return -EOPNOTSUPP;
1282                         break;
1283                 default:
1284                         return -EOPNOTSUPP;
1285                 }
1286
1287                 /* In decap flow, header pointers should point to the inner
1288                  * headers, outer header were already set by parse_tunnel_attr
1289                  */
1290                 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1291                                          inner_headers);
1292                 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1293                                          inner_headers);
1294         }
1295
1296         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
1297                 struct flow_dissector_key_eth_addrs *key =
1298                         skb_flow_dissector_target(f->dissector,
1299                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
1300                                                   f->key);
1301                 struct flow_dissector_key_eth_addrs *mask =
1302                         skb_flow_dissector_target(f->dissector,
1303                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
1304                                                   f->mask);
1305
1306                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1307                                              dmac_47_16),
1308                                 mask->dst);
1309                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1310                                              dmac_47_16),
1311                                 key->dst);
1312
1313                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1314                                              smac_47_16),
1315                                 mask->src);
1316                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1317                                              smac_47_16),
1318                                 key->src);
1319
1320                 if (!is_zero_ether_addr(mask->src) || !is_zero_ether_addr(mask->dst))
1321                         *match_level = MLX5_MATCH_L2;
1322         }
1323
1324         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
1325                 struct flow_dissector_key_vlan *key =
1326                         skb_flow_dissector_target(f->dissector,
1327                                                   FLOW_DISSECTOR_KEY_VLAN,
1328                                                   f->key);
1329                 struct flow_dissector_key_vlan *mask =
1330                         skb_flow_dissector_target(f->dissector,
1331                                                   FLOW_DISSECTOR_KEY_VLAN,
1332                                                   f->mask);
1333                 if (mask->vlan_id || mask->vlan_priority || mask->vlan_tpid) {
1334                         if (key->vlan_tpid == htons(ETH_P_8021AD)) {
1335                                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1336                                          svlan_tag, 1);
1337                                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1338                                          svlan_tag, 1);
1339                         } else {
1340                                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1341                                          cvlan_tag, 1);
1342                                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1343                                          cvlan_tag, 1);
1344                         }
1345
1346                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
1347                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
1348
1349                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
1350                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
1351
1352                         *match_level = MLX5_MATCH_L2;
1353                 }
1354         }
1355
1356         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CVLAN)) {
1357                 struct flow_dissector_key_vlan *key =
1358                         skb_flow_dissector_target(f->dissector,
1359                                                   FLOW_DISSECTOR_KEY_CVLAN,
1360                                                   f->key);
1361                 struct flow_dissector_key_vlan *mask =
1362                         skb_flow_dissector_target(f->dissector,
1363                                                   FLOW_DISSECTOR_KEY_CVLAN,
1364                                                   f->mask);
1365                 if (mask->vlan_id || mask->vlan_priority || mask->vlan_tpid) {
1366                         if (key->vlan_tpid == htons(ETH_P_8021AD)) {
1367                                 MLX5_SET(fte_match_set_misc, misc_c,
1368                                          outer_second_svlan_tag, 1);
1369                                 MLX5_SET(fte_match_set_misc, misc_v,
1370                                          outer_second_svlan_tag, 1);
1371                         } else {
1372                                 MLX5_SET(fte_match_set_misc, misc_c,
1373                                          outer_second_cvlan_tag, 1);
1374                                 MLX5_SET(fte_match_set_misc, misc_v,
1375                                          outer_second_cvlan_tag, 1);
1376                         }
1377
1378                         MLX5_SET(fte_match_set_misc, misc_c, outer_second_vid,
1379                                  mask->vlan_id);
1380                         MLX5_SET(fte_match_set_misc, misc_v, outer_second_vid,
1381                                  key->vlan_id);
1382                         MLX5_SET(fte_match_set_misc, misc_c, outer_second_prio,
1383                                  mask->vlan_priority);
1384                         MLX5_SET(fte_match_set_misc, misc_v, outer_second_prio,
1385                                  key->vlan_priority);
1386
1387                         *match_level = MLX5_MATCH_L2;
1388                 }
1389         }
1390
1391         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1392                 struct flow_dissector_key_basic *key =
1393                         skb_flow_dissector_target(f->dissector,
1394                                                   FLOW_DISSECTOR_KEY_BASIC,
1395                                                   f->key);
1396                 struct flow_dissector_key_basic *mask =
1397                         skb_flow_dissector_target(f->dissector,
1398                                                   FLOW_DISSECTOR_KEY_BASIC,
1399                                                   f->mask);
1400                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
1401                          ntohs(mask->n_proto));
1402                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
1403                          ntohs(key->n_proto));
1404
1405                 if (mask->n_proto)
1406                         *match_level = MLX5_MATCH_L2;
1407         }
1408
1409         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
1410                 struct flow_dissector_key_control *key =
1411                         skb_flow_dissector_target(f->dissector,
1412                                                   FLOW_DISSECTOR_KEY_CONTROL,
1413                                                   f->key);
1414
1415                 struct flow_dissector_key_control *mask =
1416                         skb_flow_dissector_target(f->dissector,
1417                                                   FLOW_DISSECTOR_KEY_CONTROL,
1418                                                   f->mask);
1419                 addr_type = key->addr_type;
1420
1421                 /* the HW doesn't support frag first/later */
1422                 if (mask->flags & FLOW_DIS_FIRST_FRAG)
1423                         return -EOPNOTSUPP;
1424
1425                 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
1426                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
1427                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
1428                                  key->flags & FLOW_DIS_IS_FRAGMENT);
1429
1430                         /* the HW doesn't need L3 inline to match on frag=no */
1431                         if (!(key->flags & FLOW_DIS_IS_FRAGMENT))
1432                                 *match_level = MLX5_INLINE_MODE_L2;
1433         /* ***  L2 attributes parsing up to here *** */
1434                         else
1435                                 *match_level = MLX5_INLINE_MODE_IP;
1436                 }
1437         }
1438
1439         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
1440                 struct flow_dissector_key_basic *key =
1441                         skb_flow_dissector_target(f->dissector,
1442                                                   FLOW_DISSECTOR_KEY_BASIC,
1443                                                   f->key);
1444                 struct flow_dissector_key_basic *mask =
1445                         skb_flow_dissector_target(f->dissector,
1446                                                   FLOW_DISSECTOR_KEY_BASIC,
1447                                                   f->mask);
1448                 ip_proto = key->ip_proto;
1449
1450                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
1451                          mask->ip_proto);
1452                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
1453                          key->ip_proto);
1454
1455                 if (mask->ip_proto)
1456                         *match_level = MLX5_MATCH_L3;
1457         }
1458
1459         if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1460                 struct flow_dissector_key_ipv4_addrs *key =
1461                         skb_flow_dissector_target(f->dissector,
1462                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1463                                                   f->key);
1464                 struct flow_dissector_key_ipv4_addrs *mask =
1465                         skb_flow_dissector_target(f->dissector,
1466                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
1467                                                   f->mask);
1468
1469                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1470                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
1471                        &mask->src, sizeof(mask->src));
1472                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1473                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
1474                        &key->src, sizeof(key->src));
1475                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1476                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1477                        &mask->dst, sizeof(mask->dst));
1478                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1479                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
1480                        &key->dst, sizeof(key->dst));
1481
1482                 if (mask->src || mask->dst)
1483                         *match_level = MLX5_MATCH_L3;
1484         }
1485
1486         if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1487                 struct flow_dissector_key_ipv6_addrs *key =
1488                         skb_flow_dissector_target(f->dissector,
1489                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1490                                                   f->key);
1491                 struct flow_dissector_key_ipv6_addrs *mask =
1492                         skb_flow_dissector_target(f->dissector,
1493                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
1494                                                   f->mask);
1495
1496                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1497                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1498                        &mask->src, sizeof(mask->src));
1499                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1500                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
1501                        &key->src, sizeof(key->src));
1502
1503                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
1504                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1505                        &mask->dst, sizeof(mask->dst));
1506                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
1507                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
1508                        &key->dst, sizeof(key->dst));
1509
1510                 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
1511                     ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
1512                         *match_level = MLX5_MATCH_L3;
1513         }
1514
1515         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_IP)) {
1516                 struct flow_dissector_key_ip *key =
1517                         skb_flow_dissector_target(f->dissector,
1518                                                   FLOW_DISSECTOR_KEY_IP,
1519                                                   f->key);
1520                 struct flow_dissector_key_ip *mask =
1521                         skb_flow_dissector_target(f->dissector,
1522                                                   FLOW_DISSECTOR_KEY_IP,
1523                                                   f->mask);
1524
1525                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_ecn, mask->tos & 0x3);
1526                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, key->tos & 0x3);
1527
1528                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_dscp, mask->tos >> 2);
1529                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, key->tos  >> 2);
1530
1531                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ttl_hoplimit, mask->ttl);
1532                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ttl_hoplimit, key->ttl);
1533
1534                 if (mask->ttl &&
1535                     !MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev,
1536                                                 ft_field_support.outer_ipv4_ttl))
1537                         return -EOPNOTSUPP;
1538
1539                 if (mask->tos || mask->ttl)
1540                         *match_level = MLX5_MATCH_L3;
1541         }
1542
1543         /* ***  L3 attributes parsing up to here *** */
1544
1545         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
1546                 struct flow_dissector_key_ports *key =
1547                         skb_flow_dissector_target(f->dissector,
1548                                                   FLOW_DISSECTOR_KEY_PORTS,
1549                                                   f->key);
1550                 struct flow_dissector_key_ports *mask =
1551                         skb_flow_dissector_target(f->dissector,
1552                                                   FLOW_DISSECTOR_KEY_PORTS,
1553                                                   f->mask);
1554                 switch (ip_proto) {
1555                 case IPPROTO_TCP:
1556                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1557                                  tcp_sport, ntohs(mask->src));
1558                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1559                                  tcp_sport, ntohs(key->src));
1560
1561                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1562                                  tcp_dport, ntohs(mask->dst));
1563                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1564                                  tcp_dport, ntohs(key->dst));
1565                         break;
1566
1567                 case IPPROTO_UDP:
1568                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1569                                  udp_sport, ntohs(mask->src));
1570                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1571                                  udp_sport, ntohs(key->src));
1572
1573                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
1574                                  udp_dport, ntohs(mask->dst));
1575                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
1576                                  udp_dport, ntohs(key->dst));
1577                         break;
1578                 default:
1579                         netdev_err(priv->netdev,
1580                                    "Only UDP and TCP transport are supported\n");
1581                         return -EINVAL;
1582                 }
1583
1584                 if (mask->src || mask->dst)
1585                         *match_level = MLX5_MATCH_L4;
1586         }
1587
1588         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_TCP)) {
1589                 struct flow_dissector_key_tcp *key =
1590                         skb_flow_dissector_target(f->dissector,
1591                                                   FLOW_DISSECTOR_KEY_TCP,
1592                                                   f->key);
1593                 struct flow_dissector_key_tcp *mask =
1594                         skb_flow_dissector_target(f->dissector,
1595                                                   FLOW_DISSECTOR_KEY_TCP,
1596                                                   f->mask);
1597
1598                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_flags,
1599                          ntohs(mask->flags));
1600                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
1601                          ntohs(key->flags));
1602
1603                 if (mask->flags)
1604                         *match_level = MLX5_MATCH_L4;
1605         }
1606
1607         return 0;
1608 }
1609
1610 static int parse_cls_flower(struct mlx5e_priv *priv,
1611                             struct mlx5e_tc_flow *flow,
1612                             struct mlx5_flow_spec *spec,
1613                             struct tc_cls_flower_offload *f)
1614 {
1615         struct mlx5_core_dev *dev = priv->mdev;
1616         struct mlx5_eswitch *esw = dev->priv.eswitch;
1617         struct mlx5e_rep_priv *rpriv = priv->ppriv;
1618         struct mlx5_eswitch_rep *rep;
1619         u8 match_level;
1620         int err;
1621
1622         err = __parse_cls_flower(priv, spec, f, &match_level);
1623
1624         if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH)) {
1625                 rep = rpriv->rep;
1626                 if (rep->vport != FDB_UPLINK_VPORT &&
1627                     (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
1628                     esw->offloads.inline_mode < match_level)) {
1629                         netdev_warn(priv->netdev,
1630                                     "Flow is not offloaded due to min inline setting, required %d actual %d\n",
1631                                     match_level, esw->offloads.inline_mode);
1632                         return -EOPNOTSUPP;
1633                 }
1634         }
1635
1636         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
1637                 flow->esw_attr->match_level = match_level;
1638         else
1639                 flow->nic_attr->match_level = match_level;
1640
1641         return err;
1642 }
1643
1644 struct pedit_headers {
1645         struct ethhdr  eth;
1646         struct iphdr   ip4;
1647         struct ipv6hdr ip6;
1648         struct tcphdr  tcp;
1649         struct udphdr  udp;
1650 };
1651
1652 static int pedit_header_offsets[] = {
1653         [TCA_PEDIT_KEY_EX_HDR_TYPE_ETH] = offsetof(struct pedit_headers, eth),
1654         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP4] = offsetof(struct pedit_headers, ip4),
1655         [TCA_PEDIT_KEY_EX_HDR_TYPE_IP6] = offsetof(struct pedit_headers, ip6),
1656         [TCA_PEDIT_KEY_EX_HDR_TYPE_TCP] = offsetof(struct pedit_headers, tcp),
1657         [TCA_PEDIT_KEY_EX_HDR_TYPE_UDP] = offsetof(struct pedit_headers, udp),
1658 };
1659
1660 #define pedit_header(_ph, _htype) ((void *)(_ph) + pedit_header_offsets[_htype])
1661
1662 static int set_pedit_val(u8 hdr_type, u32 mask, u32 val, u32 offset,
1663                          struct pedit_headers *masks,
1664                          struct pedit_headers *vals)
1665 {
1666         u32 *curr_pmask, *curr_pval;
1667
1668         if (hdr_type >= __PEDIT_HDR_TYPE_MAX)
1669                 goto out_err;
1670
1671         curr_pmask = (u32 *)(pedit_header(masks, hdr_type) + offset);
1672         curr_pval  = (u32 *)(pedit_header(vals, hdr_type) + offset);
1673
1674         if (*curr_pmask & mask)  /* disallow acting twice on the same location */
1675                 goto out_err;
1676
1677         *curr_pmask |= mask;
1678         *curr_pval  |= (val & mask);
1679
1680         return 0;
1681
1682 out_err:
1683         return -EOPNOTSUPP;
1684 }
1685
1686 struct mlx5_fields {
1687         u8  field;
1688         u8  size;
1689         u32 offset;
1690 };
1691
1692 #define OFFLOAD(fw_field, size, field, off) \
1693                 {MLX5_ACTION_IN_FIELD_OUT_ ## fw_field, size, offsetof(struct pedit_headers, field) + (off)}
1694
1695 static struct mlx5_fields fields[] = {
1696         OFFLOAD(DMAC_47_16, 4, eth.h_dest[0], 0),
1697         OFFLOAD(DMAC_15_0,  2, eth.h_dest[4], 0),
1698         OFFLOAD(SMAC_47_16, 4, eth.h_source[0], 0),
1699         OFFLOAD(SMAC_15_0,  2, eth.h_source[4], 0),
1700         OFFLOAD(ETHERTYPE,  2, eth.h_proto, 0),
1701
1702         OFFLOAD(IP_TTL, 1, ip4.ttl,   0),
1703         OFFLOAD(SIPV4,  4, ip4.saddr, 0),
1704         OFFLOAD(DIPV4,  4, ip4.daddr, 0),
1705
1706         OFFLOAD(SIPV6_127_96, 4, ip6.saddr.s6_addr32[0], 0),
1707         OFFLOAD(SIPV6_95_64,  4, ip6.saddr.s6_addr32[1], 0),
1708         OFFLOAD(SIPV6_63_32,  4, ip6.saddr.s6_addr32[2], 0),
1709         OFFLOAD(SIPV6_31_0,   4, ip6.saddr.s6_addr32[3], 0),
1710         OFFLOAD(DIPV6_127_96, 4, ip6.daddr.s6_addr32[0], 0),
1711         OFFLOAD(DIPV6_95_64,  4, ip6.daddr.s6_addr32[1], 0),
1712         OFFLOAD(DIPV6_63_32,  4, ip6.daddr.s6_addr32[2], 0),
1713         OFFLOAD(DIPV6_31_0,   4, ip6.daddr.s6_addr32[3], 0),
1714         OFFLOAD(IPV6_HOPLIMIT, 1, ip6.hop_limit, 0),
1715
1716         OFFLOAD(TCP_SPORT, 2, tcp.source,  0),
1717         OFFLOAD(TCP_DPORT, 2, tcp.dest,    0),
1718         OFFLOAD(TCP_FLAGS, 1, tcp.ack_seq, 5),
1719
1720         OFFLOAD(UDP_SPORT, 2, udp.source, 0),
1721         OFFLOAD(UDP_DPORT, 2, udp.dest,   0),
1722 };
1723
1724 /* On input attr->num_mod_hdr_actions tells how many HW actions can be parsed at
1725  * max from the SW pedit action. On success, it says how many HW actions were
1726  * actually parsed.
1727  */
1728 static int offload_pedit_fields(struct pedit_headers *masks,
1729                                 struct pedit_headers *vals,
1730                                 struct mlx5e_tc_flow_parse_attr *parse_attr)
1731 {
1732         struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
1733         int i, action_size, nactions, max_actions, first, last, next_z;
1734         void *s_masks_p, *a_masks_p, *vals_p;
1735         struct mlx5_fields *f;
1736         u8 cmd, field_bsize;
1737         u32 s_mask, a_mask;
1738         unsigned long mask;
1739         __be32 mask_be32;
1740         __be16 mask_be16;
1741         void *action;
1742
1743         set_masks = &masks[TCA_PEDIT_KEY_EX_CMD_SET];
1744         add_masks = &masks[TCA_PEDIT_KEY_EX_CMD_ADD];
1745         set_vals = &vals[TCA_PEDIT_KEY_EX_CMD_SET];
1746         add_vals = &vals[TCA_PEDIT_KEY_EX_CMD_ADD];
1747
1748         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1749         action = parse_attr->mod_hdr_actions;
1750         max_actions = parse_attr->num_mod_hdr_actions;
1751         nactions = 0;
1752
1753         for (i = 0; i < ARRAY_SIZE(fields); i++) {
1754                 f = &fields[i];
1755                 /* avoid seeing bits set from previous iterations */
1756                 s_mask = 0;
1757                 a_mask = 0;
1758
1759                 s_masks_p = (void *)set_masks + f->offset;
1760                 a_masks_p = (void *)add_masks + f->offset;
1761
1762                 memcpy(&s_mask, s_masks_p, f->size);
1763                 memcpy(&a_mask, a_masks_p, f->size);
1764
1765                 if (!s_mask && !a_mask) /* nothing to offload here */
1766                         continue;
1767
1768                 if (s_mask && a_mask) {
1769                         printk(KERN_WARNING "mlx5: can't set and add to the same HW field (%x)\n", f->field);
1770                         return -EOPNOTSUPP;
1771                 }
1772
1773                 if (nactions == max_actions) {
1774                         printk(KERN_WARNING "mlx5: parsed %d pedit actions, can't do more\n", nactions);
1775                         return -EOPNOTSUPP;
1776                 }
1777
1778                 if (s_mask) {
1779                         cmd  = MLX5_ACTION_TYPE_SET;
1780                         mask = s_mask;
1781                         vals_p = (void *)set_vals + f->offset;
1782                         /* clear to denote we consumed this field */
1783                         memset(s_masks_p, 0, f->size);
1784                 } else {
1785                         cmd  = MLX5_ACTION_TYPE_ADD;
1786                         mask = a_mask;
1787                         vals_p = (void *)add_vals + f->offset;
1788                         /* clear to denote we consumed this field */
1789                         memset(a_masks_p, 0, f->size);
1790                 }
1791
1792                 field_bsize = f->size * BITS_PER_BYTE;
1793
1794                 if (field_bsize == 32) {
1795                         mask_be32 = *(__be32 *)&mask;
1796                         mask = (__force unsigned long)cpu_to_le32(be32_to_cpu(mask_be32));
1797                 } else if (field_bsize == 16) {
1798                         mask_be16 = *(__be16 *)&mask;
1799                         mask = (__force unsigned long)cpu_to_le16(be16_to_cpu(mask_be16));
1800                 }
1801
1802                 first = find_first_bit(&mask, field_bsize);
1803                 next_z = find_next_zero_bit(&mask, field_bsize, first);
1804                 last  = find_last_bit(&mask, field_bsize);
1805                 if (first < next_z && next_z < last) {
1806                         printk(KERN_WARNING "mlx5: rewrite of few sub-fields (mask %lx) isn't offloaded\n",
1807                                mask);
1808                         return -EOPNOTSUPP;
1809                 }
1810
1811                 MLX5_SET(set_action_in, action, action_type, cmd);
1812                 MLX5_SET(set_action_in, action, field, f->field);
1813
1814                 if (cmd == MLX5_ACTION_TYPE_SET) {
1815                         MLX5_SET(set_action_in, action, offset, first);
1816                         /* length is num of bits to be written, zero means length of 32 */
1817                         MLX5_SET(set_action_in, action, length, (last - first + 1));
1818                 }
1819
1820                 if (field_bsize == 32)
1821                         MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p) >> first);
1822                 else if (field_bsize == 16)
1823                         MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p) >> first);
1824                 else if (field_bsize == 8)
1825                         MLX5_SET(set_action_in, action, data, *(u8 *)vals_p >> first);
1826
1827                 action += action_size;
1828                 nactions++;
1829         }
1830
1831         parse_attr->num_mod_hdr_actions = nactions;
1832         return 0;
1833 }
1834
1835 static int alloc_mod_hdr_actions(struct mlx5e_priv *priv,
1836                                  const struct tc_action *a, int namespace,
1837                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1838 {
1839         int nkeys, action_size, max_actions;
1840
1841         nkeys = tcf_pedit_nkeys(a);
1842         action_size = MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto);
1843
1844         if (namespace == MLX5_FLOW_NAMESPACE_FDB) /* FDB offloading */
1845                 max_actions = MLX5_CAP_ESW_FLOWTABLE_FDB(priv->mdev, max_modify_header_actions);
1846         else /* namespace is MLX5_FLOW_NAMESPACE_KERNEL - NIC offloading */
1847                 max_actions = MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, max_modify_header_actions);
1848
1849         /* can get up to crazingly 16 HW actions in 32 bits pedit SW key */
1850         max_actions = min(max_actions, nkeys * 16);
1851
1852         parse_attr->mod_hdr_actions = kcalloc(max_actions, action_size, GFP_KERNEL);
1853         if (!parse_attr->mod_hdr_actions)
1854                 return -ENOMEM;
1855
1856         parse_attr->num_mod_hdr_actions = max_actions;
1857         return 0;
1858 }
1859
1860 static const struct pedit_headers zero_masks = {};
1861
1862 static int parse_tc_pedit_action(struct mlx5e_priv *priv,
1863                                  const struct tc_action *a, int namespace,
1864                                  struct mlx5e_tc_flow_parse_attr *parse_attr)
1865 {
1866         struct pedit_headers masks[__PEDIT_CMD_MAX], vals[__PEDIT_CMD_MAX], *cmd_masks;
1867         int nkeys, i, err = -EOPNOTSUPP;
1868         u32 mask, val, offset;
1869         u8 cmd, htype;
1870
1871         nkeys = tcf_pedit_nkeys(a);
1872
1873         memset(masks, 0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1874         memset(vals,  0, sizeof(struct pedit_headers) * __PEDIT_CMD_MAX);
1875
1876         for (i = 0; i < nkeys; i++) {
1877                 htype = tcf_pedit_htype(a, i);
1878                 cmd = tcf_pedit_cmd(a, i);
1879                 err = -EOPNOTSUPP; /* can't be all optimistic */
1880
1881                 if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK) {
1882                         netdev_warn(priv->netdev, "legacy pedit isn't offloaded\n");
1883                         goto out_err;
1884                 }
1885
1886                 if (cmd != TCA_PEDIT_KEY_EX_CMD_SET && cmd != TCA_PEDIT_KEY_EX_CMD_ADD) {
1887                         netdev_warn(priv->netdev, "pedit cmd %d isn't offloaded\n", cmd);
1888                         goto out_err;
1889                 }
1890
1891                 mask = tcf_pedit_mask(a, i);
1892                 val = tcf_pedit_val(a, i);
1893                 offset = tcf_pedit_offset(a, i);
1894
1895                 err = set_pedit_val(htype, ~mask, val, offset, &masks[cmd], &vals[cmd]);
1896                 if (err)
1897                         goto out_err;
1898         }
1899
1900         err = alloc_mod_hdr_actions(priv, a, namespace, parse_attr);
1901         if (err)
1902                 goto out_err;
1903
1904         err = offload_pedit_fields(masks, vals, parse_attr);
1905         if (err < 0)
1906                 goto out_dealloc_parsed_actions;
1907
1908         for (cmd = 0; cmd < __PEDIT_CMD_MAX; cmd++) {
1909                 cmd_masks = &masks[cmd];
1910                 if (memcmp(cmd_masks, &zero_masks, sizeof(zero_masks))) {
1911                         netdev_warn(priv->netdev, "attempt to offload an unsupported field (cmd %d)\n", cmd);
1912                         print_hex_dump(KERN_WARNING, "mask: ", DUMP_PREFIX_ADDRESS,
1913                                        16, 1, cmd_masks, sizeof(zero_masks), true);
1914                         err = -EOPNOTSUPP;
1915                         goto out_dealloc_parsed_actions;
1916                 }
1917         }
1918
1919         return 0;
1920
1921 out_dealloc_parsed_actions:
1922         kfree(parse_attr->mod_hdr_actions);
1923 out_err:
1924         return err;
1925 }
1926
1927 static bool csum_offload_supported(struct mlx5e_priv *priv, u32 action, u32 update_flags)
1928 {
1929         u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
1930                          TCA_CSUM_UPDATE_FLAG_UDP;
1931
1932         /*  The HW recalcs checksums only if re-writing headers */
1933         if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
1934                 netdev_warn(priv->netdev,
1935                             "TC csum action is only offloaded with pedit\n");
1936                 return false;
1937         }
1938
1939         if (update_flags & ~prot_flags) {
1940                 netdev_warn(priv->netdev,
1941                             "can't offload TC csum action for some header/s - flags %#x\n",
1942                             update_flags);
1943                 return false;
1944         }
1945
1946         return true;
1947 }
1948
1949 static bool modify_header_match_supported(struct mlx5_flow_spec *spec,
1950                                           struct tcf_exts *exts)
1951 {
1952         const struct tc_action *a;
1953         bool modify_ip_header;
1954         LIST_HEAD(actions);
1955         u8 htype, ip_proto;
1956         void *headers_v;
1957         u16 ethertype;
1958         int nkeys, i;
1959
1960         headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value, outer_headers);
1961         ethertype = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ethertype);
1962
1963         /* for non-IP we only re-write MACs, so we're okay */
1964         if (ethertype != ETH_P_IP && ethertype != ETH_P_IPV6)
1965                 goto out_ok;
1966
1967         modify_ip_header = false;
1968         tcf_exts_to_list(exts, &actions);
1969         list_for_each_entry(a, &actions, list) {
1970                 if (!is_tcf_pedit(a))
1971                         continue;
1972
1973                 nkeys = tcf_pedit_nkeys(a);
1974                 for (i = 0; i < nkeys; i++) {
1975                         htype = tcf_pedit_htype(a, i);
1976                         if (htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP4 ||
1977                             htype == TCA_PEDIT_KEY_EX_HDR_TYPE_IP6) {
1978                                 modify_ip_header = true;
1979                                 break;
1980                         }
1981                 }
1982         }
1983
1984         ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
1985         if (modify_ip_header && ip_proto != IPPROTO_TCP &&
1986             ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
1987                 pr_info("can't offload re-write of ip proto %d\n", ip_proto);
1988                 return false;
1989         }
1990
1991 out_ok:
1992         return true;
1993 }
1994
1995 static bool actions_match_supported(struct mlx5e_priv *priv,
1996                                     struct tcf_exts *exts,
1997                                     struct mlx5e_tc_flow_parse_attr *parse_attr,
1998                                     struct mlx5e_tc_flow *flow)
1999 {
2000         u32 actions;
2001
2002         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
2003                 actions = flow->esw_attr->action;
2004         else
2005                 actions = flow->nic_attr->action;
2006
2007         if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
2008                 return modify_header_match_supported(&parse_attr->spec, exts);
2009
2010         return true;
2011 }
2012
2013 static bool same_hw_devs(struct mlx5e_priv *priv, struct mlx5e_priv *peer_priv)
2014 {
2015         struct mlx5_core_dev *fmdev, *pmdev;
2016         u16 func_id, peer_id;
2017
2018         fmdev = priv->mdev;
2019         pmdev = peer_priv->mdev;
2020
2021         func_id = (u16)((fmdev->pdev->bus->number << 8) | PCI_SLOT(fmdev->pdev->devfn));
2022         peer_id = (u16)((pmdev->pdev->bus->number << 8) | PCI_SLOT(pmdev->pdev->devfn));
2023
2024         return (func_id == peer_id);
2025 }
2026
2027 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
2028                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
2029                                 struct mlx5e_tc_flow *flow)
2030 {
2031         struct mlx5_nic_flow_attr *attr = flow->nic_attr;
2032         const struct tc_action *a;
2033         LIST_HEAD(actions);
2034         u32 action = 0;
2035         int err;
2036
2037         if (!tcf_exts_has_actions(exts))
2038                 return -EINVAL;
2039
2040         attr->flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
2041
2042         tcf_exts_to_list(exts, &actions);
2043         list_for_each_entry(a, &actions, list) {
2044                 if (is_tcf_gact_shot(a)) {
2045                         action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
2046                         if (MLX5_CAP_FLOWTABLE(priv->mdev,
2047                                                flow_table_properties_nic_receive.flow_counter))
2048                                 action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
2049                         continue;
2050                 }
2051
2052                 if (is_tcf_pedit(a)) {
2053                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_KERNEL,
2054                                                     parse_attr);
2055                         if (err)
2056                                 return err;
2057
2058                         action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
2059                                   MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2060                         continue;
2061                 }
2062
2063                 if (is_tcf_csum(a)) {
2064                         if (csum_offload_supported(priv, action,
2065                                                    tcf_csum_update_flags(a)))
2066                                 continue;
2067
2068                         return -EOPNOTSUPP;
2069                 }
2070
2071                 if (is_tcf_mirred_egress_redirect(a)) {
2072                         struct net_device *peer_dev = tcf_mirred_dev(a);
2073
2074                         if (priv->netdev->netdev_ops == peer_dev->netdev_ops &&
2075                             same_hw_devs(priv, netdev_priv(peer_dev))) {
2076                                 parse_attr->mirred_ifindex = peer_dev->ifindex;
2077                                 flow->flags |= MLX5E_TC_FLOW_HAIRPIN;
2078                                 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2079                                           MLX5_FLOW_CONTEXT_ACTION_COUNT;
2080                         } else {
2081                                 netdev_warn(priv->netdev, "device %s not on same HW, can't offload\n",
2082                                             peer_dev->name);
2083                                 return -EINVAL;
2084                         }
2085                         continue;
2086                 }
2087
2088                 if (is_tcf_skbedit_mark(a)) {
2089                         u32 mark = tcf_skbedit_mark(a);
2090
2091                         if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
2092                                 netdev_warn(priv->netdev, "Bad flow mark - only 16 bit is supported: 0x%x\n",
2093                                             mark);
2094                                 return -EINVAL;
2095                         }
2096
2097                         attr->flow_tag = mark;
2098                         action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2099                         continue;
2100                 }
2101
2102                 return -EINVAL;
2103         }
2104
2105         attr->action = action;
2106         if (!actions_match_supported(priv, exts, parse_attr, flow))
2107                 return -EOPNOTSUPP;
2108
2109         return 0;
2110 }
2111
2112 static inline int cmp_encap_info(struct ip_tunnel_key *a,
2113                                  struct ip_tunnel_key *b)
2114 {
2115         return memcmp(a, b, sizeof(*a));
2116 }
2117
2118 static inline int hash_encap_info(struct ip_tunnel_key *key)
2119 {
2120         return jhash(key, sizeof(*key), 0);
2121 }
2122
2123 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
2124                                    struct net_device *mirred_dev,
2125                                    struct net_device **out_dev,
2126                                    struct flowi4 *fl4,
2127                                    struct neighbour **out_n,
2128                                    int *out_ttl)
2129 {
2130         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2131         struct mlx5e_rep_priv *uplink_rpriv;
2132         struct rtable *rt;
2133         struct neighbour *n = NULL;
2134
2135 #if IS_ENABLED(CONFIG_INET)
2136         int ret;
2137
2138         rt = ip_route_output_key(dev_net(mirred_dev), fl4);
2139         ret = PTR_ERR_OR_ZERO(rt);
2140         if (ret)
2141                 return ret;
2142 #else
2143         return -EOPNOTSUPP;
2144 #endif
2145         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2146         /* if the egress device isn't on the same HW e-switch, we use the uplink */
2147         if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
2148                 *out_dev = uplink_rpriv->netdev;
2149         else
2150                 *out_dev = rt->dst.dev;
2151
2152         *out_ttl = ip4_dst_hoplimit(&rt->dst);
2153         n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
2154         ip_rt_put(rt);
2155         if (!n)
2156                 return -ENOMEM;
2157
2158         *out_n = n;
2159         return 0;
2160 }
2161
2162 static bool is_merged_eswitch_dev(struct mlx5e_priv *priv,
2163                                   struct net_device *peer_netdev)
2164 {
2165         struct mlx5e_priv *peer_priv;
2166
2167         peer_priv = netdev_priv(peer_netdev);
2168
2169         return (MLX5_CAP_ESW(priv->mdev, merged_eswitch) &&
2170                 (priv->netdev->netdev_ops == peer_netdev->netdev_ops) &&
2171                 same_hw_devs(priv, peer_priv) &&
2172                 MLX5_VPORT_MANAGER(peer_priv->mdev) &&
2173                 (peer_priv->mdev->priv.eswitch->mode == SRIOV_OFFLOADS));
2174 }
2175
2176 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
2177                                    struct net_device *mirred_dev,
2178                                    struct net_device **out_dev,
2179                                    struct flowi6 *fl6,
2180                                    struct neighbour **out_n,
2181                                    int *out_ttl)
2182 {
2183         struct neighbour *n = NULL;
2184         struct dst_entry *dst;
2185
2186 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
2187         struct mlx5e_rep_priv *uplink_rpriv;
2188         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2189         int ret;
2190
2191         ret = ipv6_stub->ipv6_dst_lookup(dev_net(mirred_dev), NULL, &dst,
2192                                          fl6);
2193         if (ret < 0)
2194                 return ret;
2195
2196         *out_ttl = ip6_dst_hoplimit(dst);
2197
2198         uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2199         /* if the egress device isn't on the same HW e-switch, we use the uplink */
2200         if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
2201                 *out_dev = uplink_rpriv->netdev;
2202         else
2203                 *out_dev = dst->dev;
2204 #else
2205         return -EOPNOTSUPP;
2206 #endif
2207
2208         n = dst_neigh_lookup(dst, &fl6->daddr);
2209         dst_release(dst);
2210         if (!n)
2211                 return -ENOMEM;
2212
2213         *out_n = n;
2214         return 0;
2215 }
2216
2217 static void gen_vxlan_header_ipv4(struct net_device *out_dev,
2218                                   char buf[], int encap_size,
2219                                   unsigned char h_dest[ETH_ALEN],
2220                                   int ttl,
2221                                   __be32 daddr,
2222                                   __be32 saddr,
2223                                   __be16 udp_dst_port,
2224                                   __be32 vx_vni)
2225 {
2226         struct ethhdr *eth = (struct ethhdr *)buf;
2227         struct iphdr  *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
2228         struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
2229         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2230
2231         memset(buf, 0, encap_size);
2232
2233         ether_addr_copy(eth->h_dest, h_dest);
2234         ether_addr_copy(eth->h_source, out_dev->dev_addr);
2235         eth->h_proto = htons(ETH_P_IP);
2236
2237         ip->daddr = daddr;
2238         ip->saddr = saddr;
2239
2240         ip->ttl = ttl;
2241         ip->protocol = IPPROTO_UDP;
2242         ip->version = 0x4;
2243         ip->ihl = 0x5;
2244
2245         udp->dest = udp_dst_port;
2246         vxh->vx_flags = VXLAN_HF_VNI;
2247         vxh->vx_vni = vxlan_vni_field(vx_vni);
2248 }
2249
2250 static void gen_vxlan_header_ipv6(struct net_device *out_dev,
2251                                   char buf[], int encap_size,
2252                                   unsigned char h_dest[ETH_ALEN],
2253                                   int ttl,
2254                                   struct in6_addr *daddr,
2255                                   struct in6_addr *saddr,
2256                                   __be16 udp_dst_port,
2257                                   __be32 vx_vni)
2258 {
2259         struct ethhdr *eth = (struct ethhdr *)buf;
2260         struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
2261         struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
2262         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
2263
2264         memset(buf, 0, encap_size);
2265
2266         ether_addr_copy(eth->h_dest, h_dest);
2267         ether_addr_copy(eth->h_source, out_dev->dev_addr);
2268         eth->h_proto = htons(ETH_P_IPV6);
2269
2270         ip6_flow_hdr(ip6h, 0, 0);
2271         /* the HW fills up ipv6 payload len */
2272         ip6h->nexthdr     = IPPROTO_UDP;
2273         ip6h->hop_limit   = ttl;
2274         ip6h->daddr       = *daddr;
2275         ip6h->saddr       = *saddr;
2276
2277         udp->dest = udp_dst_port;
2278         vxh->vx_flags = VXLAN_HF_VNI;
2279         vxh->vx_vni = vxlan_vni_field(vx_vni);
2280 }
2281
2282 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
2283                                           struct net_device *mirred_dev,
2284                                           struct mlx5e_encap_entry *e)
2285 {
2286         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2287         int ipv4_encap_size = ETH_HLEN + sizeof(struct iphdr) + VXLAN_HLEN;
2288         struct ip_tunnel_key *tun_key = &e->tun_info.key;
2289         struct net_device *out_dev;
2290         struct neighbour *n = NULL;
2291         struct flowi4 fl4 = {};
2292         char *encap_header;
2293         int ttl, err;
2294         u8 nud_state;
2295
2296         if (max_encap_size < ipv4_encap_size) {
2297                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2298                                ipv4_encap_size, max_encap_size);
2299                 return -EOPNOTSUPP;
2300         }
2301
2302         encap_header = kzalloc(ipv4_encap_size, GFP_KERNEL);
2303         if (!encap_header)
2304                 return -ENOMEM;
2305
2306         switch (e->tunnel_type) {
2307         case MLX5_HEADER_TYPE_VXLAN:
2308                 fl4.flowi4_proto = IPPROTO_UDP;
2309                 fl4.fl4_dport = tun_key->tp_dst;
2310                 break;
2311         default:
2312                 err = -EOPNOTSUPP;
2313                 goto free_encap;
2314         }
2315         fl4.flowi4_tos = tun_key->tos;
2316         fl4.daddr = tun_key->u.ipv4.dst;
2317         fl4.saddr = tun_key->u.ipv4.src;
2318
2319         err = mlx5e_route_lookup_ipv4(priv, mirred_dev, &out_dev,
2320                                       &fl4, &n, &ttl);
2321         if (err)
2322                 goto free_encap;
2323
2324         /* used by mlx5e_detach_encap to lookup a neigh hash table
2325          * entry in the neigh hash table when a user deletes a rule
2326          */
2327         e->m_neigh.dev = n->dev;
2328         e->m_neigh.family = n->ops->family;
2329         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2330         e->out_dev = out_dev;
2331
2332         /* It's importent to add the neigh to the hash table before checking
2333          * the neigh validity state. So if we'll get a notification, in case the
2334          * neigh changes it's validity state, we would find the relevant neigh
2335          * in the hash.
2336          */
2337         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2338         if (err)
2339                 goto free_encap;
2340
2341         read_lock_bh(&n->lock);
2342         nud_state = n->nud_state;
2343         ether_addr_copy(e->h_dest, n->ha);
2344         read_unlock_bh(&n->lock);
2345
2346         switch (e->tunnel_type) {
2347         case MLX5_HEADER_TYPE_VXLAN:
2348                 gen_vxlan_header_ipv4(out_dev, encap_header,
2349                                       ipv4_encap_size, e->h_dest, ttl,
2350                                       fl4.daddr,
2351                                       fl4.saddr, tun_key->tp_dst,
2352                                       tunnel_id_to_key32(tun_key->tun_id));
2353                 break;
2354         default:
2355                 err = -EOPNOTSUPP;
2356                 goto destroy_neigh_entry;
2357         }
2358         e->encap_size = ipv4_encap_size;
2359         e->encap_header = encap_header;
2360
2361         if (!(nud_state & NUD_VALID)) {
2362                 neigh_event_send(n, NULL);
2363                 err = -EAGAIN;
2364                 goto out;
2365         }
2366
2367         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
2368                                ipv4_encap_size, encap_header, &e->encap_id);
2369         if (err)
2370                 goto destroy_neigh_entry;
2371
2372         e->flags |= MLX5_ENCAP_ENTRY_VALID;
2373         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2374         neigh_release(n);
2375         return err;
2376
2377 destroy_neigh_entry:
2378         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2379 free_encap:
2380         kfree(encap_header);
2381 out:
2382         if (n)
2383                 neigh_release(n);
2384         return err;
2385 }
2386
2387 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
2388                                           struct net_device *mirred_dev,
2389                                           struct mlx5e_encap_entry *e)
2390 {
2391         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
2392         int ipv6_encap_size = ETH_HLEN + sizeof(struct ipv6hdr) + VXLAN_HLEN;
2393         struct ip_tunnel_key *tun_key = &e->tun_info.key;
2394         struct net_device *out_dev;
2395         struct neighbour *n = NULL;
2396         struct flowi6 fl6 = {};
2397         char *encap_header;
2398         int err, ttl = 0;
2399         u8 nud_state;
2400
2401         if (max_encap_size < ipv6_encap_size) {
2402                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
2403                                ipv6_encap_size, max_encap_size);
2404                 return -EOPNOTSUPP;
2405         }
2406
2407         encap_header = kzalloc(ipv6_encap_size, GFP_KERNEL);
2408         if (!encap_header)
2409                 return -ENOMEM;
2410
2411         switch (e->tunnel_type) {
2412         case MLX5_HEADER_TYPE_VXLAN:
2413                 fl6.flowi6_proto = IPPROTO_UDP;
2414                 fl6.fl6_dport = tun_key->tp_dst;
2415                 break;
2416         default:
2417                 err = -EOPNOTSUPP;
2418                 goto free_encap;
2419         }
2420
2421         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
2422         fl6.daddr = tun_key->u.ipv6.dst;
2423         fl6.saddr = tun_key->u.ipv6.src;
2424
2425         err = mlx5e_route_lookup_ipv6(priv, mirred_dev, &out_dev,
2426                                       &fl6, &n, &ttl);
2427         if (err)
2428                 goto free_encap;
2429
2430         /* used by mlx5e_detach_encap to lookup a neigh hash table
2431          * entry in the neigh hash table when a user deletes a rule
2432          */
2433         e->m_neigh.dev = n->dev;
2434         e->m_neigh.family = n->ops->family;
2435         memcpy(&e->m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
2436         e->out_dev = out_dev;
2437
2438         /* It's importent to add the neigh to the hash table before checking
2439          * the neigh validity state. So if we'll get a notification, in case the
2440          * neigh changes it's validity state, we would find the relevant neigh
2441          * in the hash.
2442          */
2443         err = mlx5e_rep_encap_entry_attach(netdev_priv(out_dev), e);
2444         if (err)
2445                 goto free_encap;
2446
2447         read_lock_bh(&n->lock);
2448         nud_state = n->nud_state;
2449         ether_addr_copy(e->h_dest, n->ha);
2450         read_unlock_bh(&n->lock);
2451
2452         switch (e->tunnel_type) {
2453         case MLX5_HEADER_TYPE_VXLAN:
2454                 gen_vxlan_header_ipv6(out_dev, encap_header,
2455                                       ipv6_encap_size, e->h_dest, ttl,
2456                                       &fl6.daddr,
2457                                       &fl6.saddr, tun_key->tp_dst,
2458                                       tunnel_id_to_key32(tun_key->tun_id));
2459                 break;
2460         default:
2461                 err = -EOPNOTSUPP;
2462                 goto destroy_neigh_entry;
2463         }
2464
2465         e->encap_size = ipv6_encap_size;
2466         e->encap_header = encap_header;
2467
2468         if (!(nud_state & NUD_VALID)) {
2469                 neigh_event_send(n, NULL);
2470                 err = -EAGAIN;
2471                 goto out;
2472         }
2473
2474         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
2475                                ipv6_encap_size, encap_header, &e->encap_id);
2476         if (err)
2477                 goto destroy_neigh_entry;
2478
2479         e->flags |= MLX5_ENCAP_ENTRY_VALID;
2480         mlx5e_rep_queue_neigh_stats_work(netdev_priv(out_dev));
2481         neigh_release(n);
2482         return err;
2483
2484 destroy_neigh_entry:
2485         mlx5e_rep_encap_entry_detach(netdev_priv(e->out_dev), e);
2486 free_encap:
2487         kfree(encap_header);
2488 out:
2489         if (n)
2490                 neigh_release(n);
2491         return err;
2492 }
2493
2494 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
2495                               struct ip_tunnel_info *tun_info,
2496                               struct net_device *mirred_dev,
2497                               struct net_device **encap_dev,
2498                               struct mlx5e_tc_flow *flow)
2499 {
2500         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2501         struct mlx5e_rep_priv *uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw,
2502                                                                            REP_ETH);
2503         struct net_device *up_dev = uplink_rpriv->netdev;
2504         unsigned short family = ip_tunnel_info_af(tun_info);
2505         struct mlx5e_priv *up_priv = netdev_priv(up_dev);
2506         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2507         struct ip_tunnel_key *key = &tun_info->key;
2508         struct mlx5e_encap_entry *e;
2509         int tunnel_type, err = 0;
2510         uintptr_t hash_key;
2511         bool found = false;
2512
2513         /* udp dst port must be set */
2514         if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
2515                 goto vxlan_encap_offload_err;
2516
2517         /* setting udp src port isn't supported */
2518         if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
2519 vxlan_encap_offload_err:
2520                 netdev_warn(priv->netdev,
2521                             "must set udp dst port and not set udp src port\n");
2522                 return -EOPNOTSUPP;
2523         }
2524
2525         if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->tp_dst)) &&
2526             MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
2527                 tunnel_type = MLX5_HEADER_TYPE_VXLAN;
2528         } else {
2529                 netdev_warn(priv->netdev,
2530                             "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
2531                 return -EOPNOTSUPP;
2532         }
2533
2534         hash_key = hash_encap_info(key);
2535
2536         hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
2537                                    encap_hlist, hash_key) {
2538                 if (!cmp_encap_info(&e->tun_info.key, key)) {
2539                         found = true;
2540                         break;
2541                 }
2542         }
2543
2544         /* must verify if encap is valid or not */
2545         if (found)
2546                 goto attach_flow;
2547
2548         e = kzalloc(sizeof(*e), GFP_KERNEL);
2549         if (!e)
2550                 return -ENOMEM;
2551
2552         e->tun_info = *tun_info;
2553         e->tunnel_type = tunnel_type;
2554         INIT_LIST_HEAD(&e->flows);
2555
2556         if (family == AF_INET)
2557                 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e);
2558         else if (family == AF_INET6)
2559                 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e);
2560
2561         if (err && err != -EAGAIN)
2562                 goto out_err;
2563
2564         hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
2565
2566 attach_flow:
2567         list_add(&flow->encap, &e->flows);
2568         *encap_dev = e->out_dev;
2569         if (e->flags & MLX5_ENCAP_ENTRY_VALID)
2570                 attr->encap_id = e->encap_id;
2571         else
2572                 err = -EAGAIN;
2573
2574         return err;
2575
2576 out_err:
2577         kfree(e);
2578         return err;
2579 }
2580
2581 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
2582                                 struct mlx5e_tc_flow_parse_attr *parse_attr,
2583                                 struct mlx5e_tc_flow *flow)
2584 {
2585         struct mlx5_esw_flow_attr *attr = flow->esw_attr;
2586         struct mlx5e_rep_priv *rpriv = priv->ppriv;
2587         struct ip_tunnel_info *info = NULL;
2588         const struct tc_action *a;
2589         LIST_HEAD(actions);
2590         bool encap = false;
2591         u32 action = 0;
2592
2593         if (!tcf_exts_has_actions(exts))
2594                 return -EINVAL;
2595
2596         attr->in_rep = rpriv->rep;
2597         attr->in_mdev = priv->mdev;
2598
2599         tcf_exts_to_list(exts, &actions);
2600         list_for_each_entry(a, &actions, list) {
2601                 if (is_tcf_gact_shot(a)) {
2602                         action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
2603                                   MLX5_FLOW_CONTEXT_ACTION_COUNT;
2604                         continue;
2605                 }
2606
2607                 if (is_tcf_pedit(a)) {
2608                         int err;
2609
2610                         err = parse_tc_pedit_action(priv, a, MLX5_FLOW_NAMESPACE_FDB,
2611                                                     parse_attr);
2612                         if (err)
2613                                 return err;
2614
2615                         action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
2616                         attr->mirror_count = attr->out_count;
2617                         continue;
2618                 }
2619
2620                 if (is_tcf_csum(a)) {
2621                         if (csum_offload_supported(priv, action,
2622                                                    tcf_csum_update_flags(a)))
2623                                 continue;
2624
2625                         return -EOPNOTSUPP;
2626                 }
2627
2628                 if (is_tcf_mirred_egress_redirect(a) || is_tcf_mirred_egress_mirror(a)) {
2629                         struct mlx5e_priv *out_priv;
2630                         struct net_device *out_dev;
2631
2632                         out_dev = tcf_mirred_dev(a);
2633
2634                         if (attr->out_count >= MLX5_MAX_FLOW_FWD_VPORTS) {
2635                                 pr_err("can't support more than %d output ports, can't offload forwarding\n",
2636                                        attr->out_count);
2637                                 return -EOPNOTSUPP;
2638                         }
2639
2640                         if (switchdev_port_same_parent_id(priv->netdev,
2641                                                           out_dev) ||
2642                             is_merged_eswitch_dev(priv, out_dev)) {
2643                                 action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2644                                           MLX5_FLOW_CONTEXT_ACTION_COUNT;
2645                                 out_priv = netdev_priv(out_dev);
2646                                 rpriv = out_priv->ppriv;
2647                                 attr->out_rep[attr->out_count] = rpriv->rep;
2648                                 attr->out_mdev[attr->out_count++] = out_priv->mdev;
2649                         } else if (encap) {
2650                                 parse_attr->mirred_ifindex = out_dev->ifindex;
2651                                 parse_attr->tun_info = *info;
2652                                 attr->parse_attr = parse_attr;
2653                                 action |= MLX5_FLOW_CONTEXT_ACTION_ENCAP |
2654                                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
2655                                           MLX5_FLOW_CONTEXT_ACTION_COUNT;
2656                                 /* attr->out_rep is resolved when we handle encap */
2657                         } else {
2658                                 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
2659                                        priv->netdev->name, out_dev->name);
2660                                 return -EINVAL;
2661                         }
2662                         continue;
2663                 }
2664
2665                 if (is_tcf_tunnel_set(a)) {
2666                         info = tcf_tunnel_info(a);
2667                         if (info)
2668                                 encap = true;
2669                         else
2670                                 return -EOPNOTSUPP;
2671                         attr->mirror_count = attr->out_count;
2672                         continue;
2673                 }
2674
2675                 if (is_tcf_vlan(a)) {
2676                         if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
2677                                 action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
2678                         } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
2679                                 action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
2680                                 attr->vlan_vid = tcf_vlan_push_vid(a);
2681                                 if (mlx5_eswitch_vlan_actions_supported(priv->mdev)) {
2682                                         attr->vlan_prio = tcf_vlan_push_prio(a);
2683                                         attr->vlan_proto = tcf_vlan_push_proto(a);
2684                                         if (!attr->vlan_proto)
2685                                                 attr->vlan_proto = htons(ETH_P_8021Q);
2686                                 } else if (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q) ||
2687                                            tcf_vlan_push_prio(a)) {
2688                                         return -EOPNOTSUPP;
2689                                 }
2690                         } else { /* action is TCA_VLAN_ACT_MODIFY */
2691                                 return -EOPNOTSUPP;
2692                         }
2693                         attr->mirror_count = attr->out_count;
2694                         continue;
2695                 }
2696
2697                 if (is_tcf_tunnel_release(a)) {
2698                         action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
2699                         continue;
2700                 }
2701
2702                 return -EINVAL;
2703         }
2704
2705         attr->action = action;
2706         if (!actions_match_supported(priv, exts, parse_attr, flow))
2707                 return -EOPNOTSUPP;
2708
2709         if (attr->out_count > 1 && !mlx5_esw_has_fwd_fdb(priv->mdev)) {
2710                 netdev_warn_once(priv->netdev, "current firmware doesn't support split rule for port mirroring\n");
2711                 return -EOPNOTSUPP;
2712         }
2713
2714         return 0;
2715 }
2716
2717 static void get_flags(int flags, u8 *flow_flags)
2718 {
2719         u8 __flow_flags = 0;
2720
2721         if (flags & MLX5E_TC_INGRESS)
2722                 __flow_flags |= MLX5E_TC_FLOW_INGRESS;
2723         if (flags & MLX5E_TC_EGRESS)
2724                 __flow_flags |= MLX5E_TC_FLOW_EGRESS;
2725
2726         *flow_flags = __flow_flags;
2727 }
2728
2729 static const struct rhashtable_params tc_ht_params = {
2730         .head_offset = offsetof(struct mlx5e_tc_flow, node),
2731         .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
2732         .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
2733         .automatic_shrinking = true,
2734 };
2735
2736 static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv)
2737 {
2738         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2739         struct mlx5e_rep_priv *uplink_rpriv;
2740
2741         if (MLX5_VPORT_MANAGER(priv->mdev) && esw->mode == SRIOV_OFFLOADS) {
2742                 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
2743                 return &uplink_rpriv->tc_ht;
2744         } else
2745                 return &priv->fs.tc.ht;
2746 }
2747
2748 int mlx5e_configure_flower(struct mlx5e_priv *priv,
2749                            struct tc_cls_flower_offload *f, int flags)
2750 {
2751         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
2752         struct mlx5e_tc_flow_parse_attr *parse_attr;
2753         struct rhashtable *tc_ht = get_tc_ht(priv);
2754         struct mlx5e_tc_flow *flow;
2755         int attr_size, err = 0;
2756         u8 flow_flags = 0;
2757
2758         get_flags(flags, &flow_flags);
2759
2760         flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
2761         if (flow) {
2762                 netdev_warn_once(priv->netdev, "flow cookie %lx already exists, ignoring\n", f->cookie);
2763                 return 0;
2764         }
2765
2766         if (esw && esw->mode == SRIOV_OFFLOADS) {
2767                 flow_flags |= MLX5E_TC_FLOW_ESWITCH;
2768                 attr_size  = sizeof(struct mlx5_esw_flow_attr);
2769         } else {
2770                 flow_flags |= MLX5E_TC_FLOW_NIC;
2771                 attr_size  = sizeof(struct mlx5_nic_flow_attr);
2772         }
2773
2774         flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
2775         parse_attr = kvzalloc(sizeof(*parse_attr), GFP_KERNEL);
2776         if (!parse_attr || !flow) {
2777                 err = -ENOMEM;
2778                 goto err_free;
2779         }
2780
2781         flow->cookie = f->cookie;
2782         flow->flags = flow_flags;
2783         flow->priv = priv;
2784
2785         err = parse_cls_flower(priv, flow, &parse_attr->spec, f);
2786         if (err < 0)
2787                 goto err_free;
2788
2789         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
2790                 err = parse_tc_fdb_actions(priv, f->exts, parse_attr, flow);
2791                 if (err < 0)
2792                         goto err_free;
2793                 flow->rule[0] = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow);
2794         } else {
2795                 err = parse_tc_nic_actions(priv, f->exts, parse_attr, flow);
2796                 if (err < 0)
2797                         goto err_free;
2798                 flow->rule[0] = mlx5e_tc_add_nic_flow(priv, parse_attr, flow);
2799         }
2800
2801         if (IS_ERR(flow->rule[0])) {
2802                 err = PTR_ERR(flow->rule[0]);
2803                 if (err != -EAGAIN)
2804                         goto err_free;
2805         }
2806
2807         if (err != -EAGAIN)
2808                 flow->flags |= MLX5E_TC_FLOW_OFFLOADED;
2809
2810         if (!(flow->flags & MLX5E_TC_FLOW_ESWITCH) ||
2811             !(flow->esw_attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP))
2812                 kvfree(parse_attr);
2813
2814         err = rhashtable_insert_fast(tc_ht, &flow->node, tc_ht_params);
2815         if (err) {
2816                 mlx5e_tc_del_flow(priv, flow);
2817                 kfree(flow);
2818         }
2819
2820         return err;
2821
2822 err_free:
2823         kvfree(parse_attr);
2824         kfree(flow);
2825         return err;
2826 }
2827
2828 #define DIRECTION_MASK (MLX5E_TC_INGRESS | MLX5E_TC_EGRESS)
2829 #define FLOW_DIRECTION_MASK (MLX5E_TC_FLOW_INGRESS | MLX5E_TC_FLOW_EGRESS)
2830
2831 static bool same_flow_direction(struct mlx5e_tc_flow *flow, int flags)
2832 {
2833         if ((flow->flags & FLOW_DIRECTION_MASK) == (flags & DIRECTION_MASK))
2834                 return true;
2835
2836         return false;
2837 }
2838
2839 int mlx5e_delete_flower(struct mlx5e_priv *priv,
2840                         struct tc_cls_flower_offload *f, int flags)
2841 {
2842         struct rhashtable *tc_ht = get_tc_ht(priv);
2843         struct mlx5e_tc_flow *flow;
2844
2845         flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
2846         if (!flow || !same_flow_direction(flow, flags))
2847                 return -EINVAL;
2848
2849         rhashtable_remove_fast(tc_ht, &flow->node, tc_ht_params);
2850
2851         mlx5e_tc_del_flow(priv, flow);
2852
2853         kfree(flow);
2854
2855         return 0;
2856 }
2857
2858 int mlx5e_stats_flower(struct mlx5e_priv *priv,
2859                        struct tc_cls_flower_offload *f, int flags)
2860 {
2861         struct rhashtable *tc_ht = get_tc_ht(priv);
2862         struct mlx5e_tc_flow *flow;
2863         struct mlx5_fc *counter;
2864         u64 bytes;
2865         u64 packets;
2866         u64 lastuse;
2867
2868         flow = rhashtable_lookup_fast(tc_ht, &f->cookie, tc_ht_params);
2869         if (!flow || !same_flow_direction(flow, flags))
2870                 return -EINVAL;
2871
2872         if (!(flow->flags & MLX5E_TC_FLOW_OFFLOADED))
2873                 return 0;
2874
2875         counter = mlx5_flow_rule_counter(flow->rule[0]);
2876         if (!counter)
2877                 return 0;
2878
2879         mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
2880
2881         tcf_exts_stats_update(f->exts, bytes, packets, lastuse);
2882
2883         return 0;
2884 }
2885
2886 int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
2887 {
2888         struct mlx5e_tc_table *tc = &priv->fs.tc;
2889
2890         hash_init(tc->mod_hdr_tbl);
2891         hash_init(tc->hairpin_tbl);
2892
2893         return rhashtable_init(&tc->ht, &tc_ht_params);
2894 }
2895
2896 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
2897 {
2898         struct mlx5e_tc_flow *flow = ptr;
2899         struct mlx5e_priv *priv = flow->priv;
2900
2901         mlx5e_tc_del_flow(priv, flow);
2902         kfree(flow);
2903 }
2904
2905 void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
2906 {
2907         struct mlx5e_tc_table *tc = &priv->fs.tc;
2908
2909         rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, NULL);
2910
2911         if (!IS_ERR_OR_NULL(tc->t)) {
2912                 mlx5_destroy_flow_table(tc->t);
2913                 tc->t = NULL;
2914         }
2915 }
2916
2917 int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
2918 {
2919         return rhashtable_init(tc_ht, &tc_ht_params);
2920 }
2921
2922 void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
2923 {
2924         rhashtable_free_and_destroy(tc_ht, _mlx5e_tc_del_flow, NULL);
2925 }
2926
2927 int mlx5e_tc_num_filters(struct mlx5e_priv *priv)
2928 {
2929         struct rhashtable *tc_ht = get_tc_ht(priv);
2930
2931         return atomic_read(&tc_ht->nelems);
2932 }