net/mlx5e: Make sure the FW max encap size is enough for ipv6 tunnels
[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/vxlan.h>
46 #include "en.h"
47 #include "en_tc.h"
48 #include "eswitch.h"
49 #include "vxlan.h"
50
51 enum {
52         MLX5E_TC_FLOW_ESWITCH   = BIT(0),
53 };
54
55 struct mlx5e_tc_flow {
56         struct rhash_head       node;
57         u64                     cookie;
58         u8                      flags;
59         struct mlx5_flow_handle *rule;
60         struct list_head        encap; /* flows sharing the same encap */
61         struct mlx5_esw_flow_attr *attr;
62 };
63
64 enum {
65         MLX5_HEADER_TYPE_VXLAN = 0x0,
66         MLX5_HEADER_TYPE_NVGRE = 0x1,
67 };
68
69 #define MLX5E_TC_TABLE_NUM_ENTRIES 1024
70 #define MLX5E_TC_TABLE_NUM_GROUPS 4
71
72 static struct mlx5_flow_handle *
73 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
74                       struct mlx5_flow_spec *spec,
75                       u32 action, u32 flow_tag)
76 {
77         struct mlx5_core_dev *dev = priv->mdev;
78         struct mlx5_flow_destination dest = { 0 };
79         struct mlx5_flow_act flow_act = {
80                 .action = action,
81                 .flow_tag = flow_tag,
82                 .encap_id = 0,
83         };
84         struct mlx5_fc *counter = NULL;
85         struct mlx5_flow_handle *rule;
86         bool table_created = false;
87
88         if (action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
89                 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
90                 dest.ft = priv->fs.vlan.ft.t;
91         } else if (action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
92                 counter = mlx5_fc_create(dev, true);
93                 if (IS_ERR(counter))
94                         return ERR_CAST(counter);
95
96                 dest.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
97                 dest.counter = counter;
98         }
99
100         if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
101                 priv->fs.tc.t =
102                         mlx5_create_auto_grouped_flow_table(priv->fs.ns,
103                                                             MLX5E_TC_PRIO,
104                                                             MLX5E_TC_TABLE_NUM_ENTRIES,
105                                                             MLX5E_TC_TABLE_NUM_GROUPS,
106                                                             0, 0);
107                 if (IS_ERR(priv->fs.tc.t)) {
108                         netdev_err(priv->netdev,
109                                    "Failed to create tc offload table\n");
110                         rule = ERR_CAST(priv->fs.tc.t);
111                         goto err_create_ft;
112                 }
113
114                 table_created = true;
115         }
116
117         spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
118         rule = mlx5_add_flow_rules(priv->fs.tc.t, spec, &flow_act, &dest, 1);
119
120         if (IS_ERR(rule))
121                 goto err_add_rule;
122
123         return rule;
124
125 err_add_rule:
126         if (table_created) {
127                 mlx5_destroy_flow_table(priv->fs.tc.t);
128                 priv->fs.tc.t = NULL;
129         }
130 err_create_ft:
131         mlx5_fc_destroy(dev, counter);
132
133         return rule;
134 }
135
136 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
137                                   struct mlx5e_tc_flow *flow)
138 {
139         struct mlx5_fc *counter = NULL;
140
141         if (!IS_ERR(flow->rule)) {
142                 counter = mlx5_flow_rule_counter(flow->rule);
143                 mlx5_del_flow_rules(flow->rule);
144                 mlx5_fc_destroy(priv->mdev, counter);
145         }
146
147         if (!mlx5e_tc_num_filters(priv) && (priv->fs.tc.t)) {
148                 mlx5_destroy_flow_table(priv->fs.tc.t);
149                 priv->fs.tc.t = NULL;
150         }
151 }
152
153 static struct mlx5_flow_handle *
154 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
155                       struct mlx5_flow_spec *spec,
156                       struct mlx5_esw_flow_attr *attr)
157 {
158         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
159         int err;
160
161         err = mlx5_eswitch_add_vlan_action(esw, attr);
162         if (err)
163                 return ERR_PTR(err);
164
165         return mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
166 }
167
168 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
169                                struct mlx5e_tc_flow *flow);
170
171 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
172                                   struct mlx5e_tc_flow *flow)
173 {
174         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
175
176         mlx5_eswitch_del_offloaded_rule(esw, flow->rule, flow->attr);
177
178         mlx5_eswitch_del_vlan_action(esw, flow->attr);
179
180         if (flow->attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP)
181                 mlx5e_detach_encap(priv, flow);
182 }
183
184 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
185                                struct mlx5e_tc_flow *flow)
186 {
187         struct list_head *next = flow->encap.next;
188
189         list_del(&flow->encap);
190         if (list_empty(next)) {
191                 struct mlx5_encap_entry *e;
192
193                 e = list_entry(next, struct mlx5_encap_entry, flows);
194                 if (e->n) {
195                         mlx5_encap_dealloc(priv->mdev, e->encap_id);
196                         neigh_release(e->n);
197                 }
198                 hlist_del_rcu(&e->encap_hlist);
199                 kfree(e);
200         }
201 }
202
203 /* we get here also when setting rule to the FW failed, etc. It means that the
204  * flow rule itself might not exist, but some offloading related to the actions
205  * should be cleaned.
206  */
207 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
208                               struct mlx5e_tc_flow *flow)
209 {
210         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
211                 mlx5e_tc_del_fdb_flow(priv, flow);
212         else
213                 mlx5e_tc_del_nic_flow(priv, flow);
214 }
215
216 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
217                              struct tc_cls_flower_offload *f)
218 {
219         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
220                                        outer_headers);
221         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
222                                        outer_headers);
223         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
224                                     misc_parameters);
225         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
226                                     misc_parameters);
227
228         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
229         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
230
231         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
232                 struct flow_dissector_key_keyid *key =
233                         skb_flow_dissector_target(f->dissector,
234                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
235                                                   f->key);
236                 struct flow_dissector_key_keyid *mask =
237                         skb_flow_dissector_target(f->dissector,
238                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
239                                                   f->mask);
240                 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
241                          be32_to_cpu(mask->keyid));
242                 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
243                          be32_to_cpu(key->keyid));
244         }
245 }
246
247 static int parse_tunnel_attr(struct mlx5e_priv *priv,
248                              struct mlx5_flow_spec *spec,
249                              struct tc_cls_flower_offload *f)
250 {
251         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
252                                        outer_headers);
253         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
254                                        outer_headers);
255
256         struct flow_dissector_key_control *enc_control =
257                 skb_flow_dissector_target(f->dissector,
258                                           FLOW_DISSECTOR_KEY_ENC_CONTROL,
259                                           f->key);
260
261         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
262                 struct flow_dissector_key_ports *key =
263                         skb_flow_dissector_target(f->dissector,
264                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
265                                                   f->key);
266                 struct flow_dissector_key_ports *mask =
267                         skb_flow_dissector_target(f->dissector,
268                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
269                                                   f->mask);
270                 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
271                 struct net_device *up_dev = mlx5_eswitch_get_uplink_netdev(esw);
272                 struct mlx5e_priv *up_priv = netdev_priv(up_dev);
273
274                 /* Full udp dst port must be given */
275                 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
276                         goto vxlan_match_offload_err;
277
278                 if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->dst)) &&
279                     MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
280                         parse_vxlan_attr(spec, f);
281                 else {
282                         netdev_warn(priv->netdev,
283                                     "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
284                         return -EOPNOTSUPP;
285                 }
286
287                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
288                          udp_dport, ntohs(mask->dst));
289                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
290                          udp_dport, ntohs(key->dst));
291
292                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
293                          udp_sport, ntohs(mask->src));
294                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
295                          udp_sport, ntohs(key->src));
296         } else { /* udp dst port must be given */
297 vxlan_match_offload_err:
298                 netdev_warn(priv->netdev,
299                             "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
300                 return -EOPNOTSUPP;
301         }
302
303         if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
304                 struct flow_dissector_key_ipv4_addrs *key =
305                         skb_flow_dissector_target(f->dissector,
306                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
307                                                   f->key);
308                 struct flow_dissector_key_ipv4_addrs *mask =
309                         skb_flow_dissector_target(f->dissector,
310                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
311                                                   f->mask);
312                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
313                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
314                          ntohl(mask->src));
315                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
316                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
317                          ntohl(key->src));
318
319                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
320                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
321                          ntohl(mask->dst));
322                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
323                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
324                          ntohl(key->dst));
325
326                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
327                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
328         } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
329                 struct flow_dissector_key_ipv6_addrs *key =
330                         skb_flow_dissector_target(f->dissector,
331                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
332                                                   f->key);
333                 struct flow_dissector_key_ipv6_addrs *mask =
334                         skb_flow_dissector_target(f->dissector,
335                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
336                                                   f->mask);
337
338                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
339                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
340                        &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
341                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
342                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
343                        &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
344
345                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
346                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
347                        &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
348                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
349                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
350                        &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
351
352                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
353                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
354         }
355
356         /* Enforce DMAC when offloading incoming tunneled flows.
357          * Flow counters require a match on the DMAC.
358          */
359         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
360         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
361         ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
362                                      dmac_47_16), priv->netdev->dev_addr);
363
364         /* let software handle IP fragments */
365         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
366         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
367
368         return 0;
369 }
370
371 static int __parse_cls_flower(struct mlx5e_priv *priv,
372                               struct mlx5_flow_spec *spec,
373                               struct tc_cls_flower_offload *f,
374                               u8 *min_inline)
375 {
376         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
377                                        outer_headers);
378         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
379                                        outer_headers);
380         u16 addr_type = 0;
381         u8 ip_proto = 0;
382
383         *min_inline = MLX5_INLINE_MODE_L2;
384
385         if (f->dissector->used_keys &
386             ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
387               BIT(FLOW_DISSECTOR_KEY_BASIC) |
388               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
389               BIT(FLOW_DISSECTOR_KEY_VLAN) |
390               BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
391               BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
392               BIT(FLOW_DISSECTOR_KEY_PORTS) |
393               BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
394               BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
395               BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
396               BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
397               BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL))) {
398                 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
399                             f->dissector->used_keys);
400                 return -EOPNOTSUPP;
401         }
402
403         if ((dissector_uses_key(f->dissector,
404                                 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
405              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
406              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
407             dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
408                 struct flow_dissector_key_control *key =
409                         skb_flow_dissector_target(f->dissector,
410                                                   FLOW_DISSECTOR_KEY_ENC_CONTROL,
411                                                   f->key);
412                 switch (key->addr_type) {
413                 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
414                 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
415                         if (parse_tunnel_attr(priv, spec, f))
416                                 return -EOPNOTSUPP;
417                         break;
418                 default:
419                         return -EOPNOTSUPP;
420                 }
421
422                 /* In decap flow, header pointers should point to the inner
423                  * headers, outer header were already set by parse_tunnel_attr
424                  */
425                 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
426                                          inner_headers);
427                 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
428                                          inner_headers);
429         }
430
431         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
432                 struct flow_dissector_key_control *key =
433                         skb_flow_dissector_target(f->dissector,
434                                                   FLOW_DISSECTOR_KEY_CONTROL,
435                                                   f->key);
436
437                 struct flow_dissector_key_control *mask =
438                         skb_flow_dissector_target(f->dissector,
439                                                   FLOW_DISSECTOR_KEY_CONTROL,
440                                                   f->mask);
441                 addr_type = key->addr_type;
442
443                 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
444                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
445                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
446                                  key->flags & FLOW_DIS_IS_FRAGMENT);
447
448                         /* the HW doesn't need L3 inline to match on frag=no */
449                         if (key->flags & FLOW_DIS_IS_FRAGMENT)
450                                 *min_inline = MLX5_INLINE_MODE_IP;
451                 }
452         }
453
454         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
455                 struct flow_dissector_key_basic *key =
456                         skb_flow_dissector_target(f->dissector,
457                                                   FLOW_DISSECTOR_KEY_BASIC,
458                                                   f->key);
459                 struct flow_dissector_key_basic *mask =
460                         skb_flow_dissector_target(f->dissector,
461                                                   FLOW_DISSECTOR_KEY_BASIC,
462                                                   f->mask);
463                 ip_proto = key->ip_proto;
464
465                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
466                          ntohs(mask->n_proto));
467                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
468                          ntohs(key->n_proto));
469
470                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
471                          mask->ip_proto);
472                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
473                          key->ip_proto);
474
475                 if (mask->ip_proto)
476                         *min_inline = MLX5_INLINE_MODE_IP;
477         }
478
479         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
480                 struct flow_dissector_key_eth_addrs *key =
481                         skb_flow_dissector_target(f->dissector,
482                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
483                                                   f->key);
484                 struct flow_dissector_key_eth_addrs *mask =
485                         skb_flow_dissector_target(f->dissector,
486                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
487                                                   f->mask);
488
489                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
490                                              dmac_47_16),
491                                 mask->dst);
492                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
493                                              dmac_47_16),
494                                 key->dst);
495
496                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
497                                              smac_47_16),
498                                 mask->src);
499                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
500                                              smac_47_16),
501                                 key->src);
502         }
503
504         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
505                 struct flow_dissector_key_vlan *key =
506                         skb_flow_dissector_target(f->dissector,
507                                                   FLOW_DISSECTOR_KEY_VLAN,
508                                                   f->key);
509                 struct flow_dissector_key_vlan *mask =
510                         skb_flow_dissector_target(f->dissector,
511                                                   FLOW_DISSECTOR_KEY_VLAN,
512                                                   f->mask);
513                 if (mask->vlan_id || mask->vlan_priority) {
514                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
515                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
516
517                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
518                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
519
520                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
521                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
522                 }
523         }
524
525         if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
526                 struct flow_dissector_key_ipv4_addrs *key =
527                         skb_flow_dissector_target(f->dissector,
528                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
529                                                   f->key);
530                 struct flow_dissector_key_ipv4_addrs *mask =
531                         skb_flow_dissector_target(f->dissector,
532                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
533                                                   f->mask);
534
535                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
536                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
537                        &mask->src, sizeof(mask->src));
538                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
539                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
540                        &key->src, sizeof(key->src));
541                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
542                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
543                        &mask->dst, sizeof(mask->dst));
544                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
545                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
546                        &key->dst, sizeof(key->dst));
547
548                 if (mask->src || mask->dst)
549                         *min_inline = MLX5_INLINE_MODE_IP;
550         }
551
552         if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
553                 struct flow_dissector_key_ipv6_addrs *key =
554                         skb_flow_dissector_target(f->dissector,
555                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
556                                                   f->key);
557                 struct flow_dissector_key_ipv6_addrs *mask =
558                         skb_flow_dissector_target(f->dissector,
559                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
560                                                   f->mask);
561
562                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
563                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
564                        &mask->src, sizeof(mask->src));
565                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
566                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
567                        &key->src, sizeof(key->src));
568
569                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
570                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
571                        &mask->dst, sizeof(mask->dst));
572                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
573                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
574                        &key->dst, sizeof(key->dst));
575
576                 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
577                     ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
578                         *min_inline = MLX5_INLINE_MODE_IP;
579         }
580
581         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
582                 struct flow_dissector_key_ports *key =
583                         skb_flow_dissector_target(f->dissector,
584                                                   FLOW_DISSECTOR_KEY_PORTS,
585                                                   f->key);
586                 struct flow_dissector_key_ports *mask =
587                         skb_flow_dissector_target(f->dissector,
588                                                   FLOW_DISSECTOR_KEY_PORTS,
589                                                   f->mask);
590                 switch (ip_proto) {
591                 case IPPROTO_TCP:
592                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
593                                  tcp_sport, ntohs(mask->src));
594                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
595                                  tcp_sport, ntohs(key->src));
596
597                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
598                                  tcp_dport, ntohs(mask->dst));
599                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
600                                  tcp_dport, ntohs(key->dst));
601                         break;
602
603                 case IPPROTO_UDP:
604                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
605                                  udp_sport, ntohs(mask->src));
606                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
607                                  udp_sport, ntohs(key->src));
608
609                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
610                                  udp_dport, ntohs(mask->dst));
611                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
612                                  udp_dport, ntohs(key->dst));
613                         break;
614                 default:
615                         netdev_err(priv->netdev,
616                                    "Only UDP and TCP transport are supported\n");
617                         return -EINVAL;
618                 }
619
620                 if (mask->src || mask->dst)
621                         *min_inline = MLX5_INLINE_MODE_TCP_UDP;
622         }
623
624         return 0;
625 }
626
627 static int parse_cls_flower(struct mlx5e_priv *priv,
628                             struct mlx5e_tc_flow *flow,
629                             struct mlx5_flow_spec *spec,
630                             struct tc_cls_flower_offload *f)
631 {
632         struct mlx5_core_dev *dev = priv->mdev;
633         struct mlx5_eswitch *esw = dev->priv.eswitch;
634         struct mlx5_eswitch_rep *rep = priv->ppriv;
635         u8 min_inline;
636         int err;
637
638         err = __parse_cls_flower(priv, spec, f, &min_inline);
639
640         if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH) &&
641             rep->vport != FDB_UPLINK_VPORT) {
642                 if (esw->offloads.inline_mode != MLX5_INLINE_MODE_NONE &&
643                     esw->offloads.inline_mode < min_inline) {
644                         netdev_warn(priv->netdev,
645                                     "Flow is not offloaded due to min inline setting, required %d actual %d\n",
646                                     min_inline, esw->offloads.inline_mode);
647                         return -EOPNOTSUPP;
648                 }
649         }
650
651         return err;
652 }
653
654 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
655                                 u32 *action, u32 *flow_tag)
656 {
657         const struct tc_action *a;
658         LIST_HEAD(actions);
659
660         if (tc_no_actions(exts))
661                 return -EINVAL;
662
663         *flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
664         *action = 0;
665
666         tcf_exts_to_list(exts, &actions);
667         list_for_each_entry(a, &actions, list) {
668                 /* Only support a single action per rule */
669                 if (*action)
670                         return -EINVAL;
671
672                 if (is_tcf_gact_shot(a)) {
673                         *action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
674                         if (MLX5_CAP_FLOWTABLE(priv->mdev,
675                                                flow_table_properties_nic_receive.flow_counter))
676                                 *action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
677                         continue;
678                 }
679
680                 if (is_tcf_skbedit_mark(a)) {
681                         u32 mark = tcf_skbedit_mark(a);
682
683                         if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
684                                 netdev_warn(priv->netdev, "Bad flow mark - only 16 bit is supported: 0x%x\n",
685                                             mark);
686                                 return -EINVAL;
687                         }
688
689                         *flow_tag = mark;
690                         *action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
691                         continue;
692                 }
693
694                 return -EINVAL;
695         }
696
697         return 0;
698 }
699
700 static inline int cmp_encap_info(struct ip_tunnel_key *a,
701                                  struct ip_tunnel_key *b)
702 {
703         return memcmp(a, b, sizeof(*a));
704 }
705
706 static inline int hash_encap_info(struct ip_tunnel_key *key)
707 {
708         return jhash(key, sizeof(*key), 0);
709 }
710
711 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
712                                    struct net_device *mirred_dev,
713                                    struct net_device **out_dev,
714                                    struct flowi4 *fl4,
715                                    struct neighbour **out_n,
716                                    int *out_ttl)
717 {
718         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
719         struct rtable *rt;
720         struct neighbour *n = NULL;
721
722 #if IS_ENABLED(CONFIG_INET)
723         int ret;
724
725         rt = ip_route_output_key(dev_net(mirred_dev), fl4);
726         ret = PTR_ERR_OR_ZERO(rt);
727         if (ret)
728                 return ret;
729 #else
730         return -EOPNOTSUPP;
731 #endif
732         /* if the egress device isn't on the same HW e-switch, we use the uplink */
733         if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
734                 *out_dev = mlx5_eswitch_get_uplink_netdev(esw);
735         else
736                 *out_dev = rt->dst.dev;
737
738         *out_ttl = ip4_dst_hoplimit(&rt->dst);
739         n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
740         ip_rt_put(rt);
741         if (!n)
742                 return -ENOMEM;
743
744         *out_n = n;
745         return 0;
746 }
747
748 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
749                                    struct net_device *mirred_dev,
750                                    struct net_device **out_dev,
751                                    struct flowi6 *fl6,
752                                    struct neighbour **out_n,
753                                    int *out_ttl)
754 {
755         struct neighbour *n = NULL;
756         struct dst_entry *dst;
757
758 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
759         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
760         int ret;
761
762         dst = ip6_route_output(dev_net(mirred_dev), NULL, fl6);
763         ret = dst->error;
764         if (ret) {
765                 dst_release(dst);
766                 return ret;
767         }
768
769         *out_ttl = ip6_dst_hoplimit(dst);
770
771         /* if the egress device isn't on the same HW e-switch, we use the uplink */
772         if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
773                 *out_dev = mlx5_eswitch_get_uplink_netdev(esw);
774         else
775                 *out_dev = dst->dev;
776 #else
777         return -EOPNOTSUPP;
778 #endif
779
780         n = dst_neigh_lookup(dst, &fl6->daddr);
781         dst_release(dst);
782         if (!n)
783                 return -ENOMEM;
784
785         *out_n = n;
786         return 0;
787 }
788
789 static void gen_vxlan_header_ipv4(struct net_device *out_dev,
790                                   char buf[], int encap_size,
791                                   unsigned char h_dest[ETH_ALEN],
792                                   int ttl,
793                                   __be32 daddr,
794                                   __be32 saddr,
795                                   __be16 udp_dst_port,
796                                   __be32 vx_vni)
797 {
798         struct ethhdr *eth = (struct ethhdr *)buf;
799         struct iphdr  *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
800         struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
801         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
802
803         memset(buf, 0, encap_size);
804
805         ether_addr_copy(eth->h_dest, h_dest);
806         ether_addr_copy(eth->h_source, out_dev->dev_addr);
807         eth->h_proto = htons(ETH_P_IP);
808
809         ip->daddr = daddr;
810         ip->saddr = saddr;
811
812         ip->ttl = ttl;
813         ip->protocol = IPPROTO_UDP;
814         ip->version = 0x4;
815         ip->ihl = 0x5;
816
817         udp->dest = udp_dst_port;
818         vxh->vx_flags = VXLAN_HF_VNI;
819         vxh->vx_vni = vxlan_vni_field(vx_vni);
820 }
821
822 static void gen_vxlan_header_ipv6(struct net_device *out_dev,
823                                   char buf[], int encap_size,
824                                   unsigned char h_dest[ETH_ALEN],
825                                   int ttl,
826                                   struct in6_addr *daddr,
827                                   struct in6_addr *saddr,
828                                   __be16 udp_dst_port,
829                                   __be32 vx_vni)
830 {
831         struct ethhdr *eth = (struct ethhdr *)buf;
832         struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
833         struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
834         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
835
836         memset(buf, 0, encap_size);
837
838         ether_addr_copy(eth->h_dest, h_dest);
839         ether_addr_copy(eth->h_source, out_dev->dev_addr);
840         eth->h_proto = htons(ETH_P_IPV6);
841
842         ip6_flow_hdr(ip6h, 0, 0);
843         /* the HW fills up ipv6 payload len */
844         ip6h->nexthdr     = IPPROTO_UDP;
845         ip6h->hop_limit   = ttl;
846         ip6h->daddr       = *daddr;
847         ip6h->saddr       = *saddr;
848
849         udp->dest = udp_dst_port;
850         vxh->vx_flags = VXLAN_HF_VNI;
851         vxh->vx_vni = vxlan_vni_field(vx_vni);
852 }
853
854 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
855                                           struct net_device *mirred_dev,
856                                           struct mlx5_encap_entry *e,
857                                           struct net_device **out_dev)
858 {
859         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
860         int ipv4_encap_size = ETH_HLEN + sizeof(struct iphdr) + VXLAN_HLEN;
861         struct ip_tunnel_key *tun_key = &e->tun_info.key;
862         struct neighbour *n = NULL;
863         struct flowi4 fl4 = {};
864         char *encap_header;
865         int ttl, err;
866
867         if (max_encap_size < ipv4_encap_size) {
868                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
869                                ipv4_encap_size, max_encap_size);
870                 return -EOPNOTSUPP;
871         }
872
873         encap_header = kzalloc(ipv4_encap_size, GFP_KERNEL);
874         if (!encap_header)
875                 return -ENOMEM;
876
877         switch (e->tunnel_type) {
878         case MLX5_HEADER_TYPE_VXLAN:
879                 fl4.flowi4_proto = IPPROTO_UDP;
880                 fl4.fl4_dport = tun_key->tp_dst;
881                 break;
882         default:
883                 err = -EOPNOTSUPP;
884                 goto out;
885         }
886         fl4.flowi4_tos = tun_key->tos;
887         fl4.daddr = tun_key->u.ipv4.dst;
888         fl4.saddr = tun_key->u.ipv4.src;
889
890         err = mlx5e_route_lookup_ipv4(priv, mirred_dev, out_dev,
891                                       &fl4, &n, &ttl);
892         if (err)
893                 goto out;
894
895         if (!(n->nud_state & NUD_VALID)) {
896                 pr_warn("%s: can't offload, neighbour to %pI4 invalid\n", __func__, &fl4.daddr);
897                 err = -EOPNOTSUPP;
898                 goto out;
899         }
900
901         e->n = n;
902         e->out_dev = *out_dev;
903
904         neigh_ha_snapshot(e->h_dest, n, *out_dev);
905
906         switch (e->tunnel_type) {
907         case MLX5_HEADER_TYPE_VXLAN:
908                 gen_vxlan_header_ipv4(*out_dev, encap_header,
909                                       ipv4_encap_size, e->h_dest, ttl,
910                                       fl4.daddr,
911                                       fl4.saddr, tun_key->tp_dst,
912                                       tunnel_id_to_key32(tun_key->tun_id));
913                 break;
914         default:
915                 err = -EOPNOTSUPP;
916                 goto out;
917         }
918
919         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
920                                ipv4_encap_size, encap_header, &e->encap_id);
921 out:
922         if (err && n)
923                 neigh_release(n);
924         kfree(encap_header);
925         return err;
926 }
927
928 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
929                                           struct net_device *mirred_dev,
930                                           struct mlx5_encap_entry *e,
931                                           struct net_device **out_dev)
932
933 {
934         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
935         int ipv6_encap_size = ETH_HLEN + sizeof(struct ipv6hdr) + VXLAN_HLEN;
936         struct ip_tunnel_key *tun_key = &e->tun_info.key;
937         struct neighbour *n = NULL;
938         struct flowi6 fl6 = {};
939         char *encap_header;
940         int err, ttl = 0;
941
942         if (max_encap_size < ipv6_encap_size) {
943                 mlx5_core_warn(priv->mdev, "encap size %d too big, max supported is %d\n",
944                                ipv6_encap_size, max_encap_size);
945                 return -EOPNOTSUPP;
946         }
947
948         encap_header = kzalloc(ipv6_encap_size, GFP_KERNEL);
949         if (!encap_header)
950                 return -ENOMEM;
951
952         switch (e->tunnel_type) {
953         case MLX5_HEADER_TYPE_VXLAN:
954                 fl6.flowi6_proto = IPPROTO_UDP;
955                 fl6.fl6_dport = tun_key->tp_dst;
956                 break;
957         default:
958                 err = -EOPNOTSUPP;
959                 goto out;
960         }
961
962         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
963         fl6.daddr = tun_key->u.ipv6.dst;
964         fl6.saddr = tun_key->u.ipv6.src;
965
966         err = mlx5e_route_lookup_ipv6(priv, mirred_dev, out_dev,
967                                       &fl6, &n, &ttl);
968         if (err)
969                 goto out;
970
971         if (!(n->nud_state & NUD_VALID)) {
972                 pr_warn("%s: can't offload, neighbour to %pI6 invalid\n", __func__, &fl6.daddr);
973                 err = -EOPNOTSUPP;
974                 goto out;
975         }
976
977         e->n = n;
978         e->out_dev = *out_dev;
979
980         neigh_ha_snapshot(e->h_dest, n, *out_dev);
981
982         switch (e->tunnel_type) {
983         case MLX5_HEADER_TYPE_VXLAN:
984                 gen_vxlan_header_ipv6(*out_dev, encap_header,
985                                       ipv6_encap_size, e->h_dest, ttl,
986                                       &fl6.daddr,
987                                       &fl6.saddr, tun_key->tp_dst,
988                                       tunnel_id_to_key32(tun_key->tun_id));
989                 break;
990         default:
991                 err = -EOPNOTSUPP;
992                 goto out;
993         }
994
995         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
996                                ipv6_encap_size, encap_header, &e->encap_id);
997 out:
998         if (err && n)
999                 neigh_release(n);
1000         kfree(encap_header);
1001         return err;
1002 }
1003
1004 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
1005                               struct ip_tunnel_info *tun_info,
1006                               struct net_device *mirred_dev,
1007                               struct mlx5_esw_flow_attr *attr)
1008 {
1009         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1010         struct net_device *up_dev = mlx5_eswitch_get_uplink_netdev(esw);
1011         struct mlx5e_priv *up_priv = netdev_priv(up_dev);
1012         unsigned short family = ip_tunnel_info_af(tun_info);
1013         struct ip_tunnel_key *key = &tun_info->key;
1014         struct mlx5_encap_entry *e;
1015         struct net_device *out_dev;
1016         int tunnel_type, err = -EOPNOTSUPP;
1017         uintptr_t hash_key;
1018         bool found = false;
1019
1020         /* udp dst port must be set */
1021         if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
1022                 goto vxlan_encap_offload_err;
1023
1024         /* setting udp src port isn't supported */
1025         if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
1026 vxlan_encap_offload_err:
1027                 netdev_warn(priv->netdev,
1028                             "must set udp dst port and not set udp src port\n");
1029                 return -EOPNOTSUPP;
1030         }
1031
1032         if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->tp_dst)) &&
1033             MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
1034                 tunnel_type = MLX5_HEADER_TYPE_VXLAN;
1035         } else {
1036                 netdev_warn(priv->netdev,
1037                             "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
1038                 return -EOPNOTSUPP;
1039         }
1040
1041         hash_key = hash_encap_info(key);
1042
1043         hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
1044                                    encap_hlist, hash_key) {
1045                 if (!cmp_encap_info(&e->tun_info.key, key)) {
1046                         found = true;
1047                         break;
1048                 }
1049         }
1050
1051         if (found) {
1052                 attr->encap = e;
1053                 return 0;
1054         }
1055
1056         e = kzalloc(sizeof(*e), GFP_KERNEL);
1057         if (!e)
1058                 return -ENOMEM;
1059
1060         e->tun_info = *tun_info;
1061         e->tunnel_type = tunnel_type;
1062         INIT_LIST_HEAD(&e->flows);
1063
1064         if (family == AF_INET)
1065                 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e, &out_dev);
1066         else if (family == AF_INET6)
1067                 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e, &out_dev);
1068
1069         if (err)
1070                 goto out_err;
1071
1072         attr->encap = e;
1073         hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
1074
1075         return err;
1076
1077 out_err:
1078         kfree(e);
1079         return err;
1080 }
1081
1082 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
1083                                 struct mlx5e_tc_flow *flow)
1084 {
1085         struct mlx5_esw_flow_attr *attr = flow->attr;
1086         struct ip_tunnel_info *info = NULL;
1087         const struct tc_action *a;
1088         LIST_HEAD(actions);
1089         bool encap = false;
1090         int err;
1091
1092         if (tc_no_actions(exts))
1093                 return -EINVAL;
1094
1095         memset(attr, 0, sizeof(*attr));
1096         attr->in_rep = priv->ppriv;
1097
1098         tcf_exts_to_list(exts, &actions);
1099         list_for_each_entry(a, &actions, list) {
1100                 if (is_tcf_gact_shot(a)) {
1101                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
1102                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1103                         continue;
1104                 }
1105
1106                 if (is_tcf_mirred_egress_redirect(a)) {
1107                         int ifindex = tcf_mirred_ifindex(a);
1108                         struct net_device *out_dev;
1109                         struct mlx5e_priv *out_priv;
1110
1111                         out_dev = __dev_get_by_index(dev_net(priv->netdev), ifindex);
1112
1113                         if (switchdev_port_same_parent_id(priv->netdev,
1114                                                           out_dev)) {
1115                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1116                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1117                                 out_priv = netdev_priv(out_dev);
1118                                 attr->out_rep = out_priv->ppriv;
1119                         } else if (encap) {
1120                                 err = mlx5e_attach_encap(priv, info,
1121                                                          out_dev, attr);
1122                                 if (err)
1123                                         return err;
1124                                 list_add(&flow->encap, &attr->encap->flows);
1125                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_ENCAP |
1126                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1127                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1128                                 out_priv = netdev_priv(attr->encap->out_dev);
1129                                 attr->out_rep = out_priv->ppriv;
1130                         } else {
1131                                 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
1132                                        priv->netdev->name, out_dev->name);
1133                                 return -EINVAL;
1134                         }
1135                         continue;
1136                 }
1137
1138                 if (is_tcf_tunnel_set(a)) {
1139                         info = tcf_tunnel_info(a);
1140                         if (info)
1141                                 encap = true;
1142                         else
1143                                 return -EOPNOTSUPP;
1144                         continue;
1145                 }
1146
1147                 if (is_tcf_vlan(a)) {
1148                         if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
1149                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
1150                         } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
1151                                 if (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q))
1152                                         return -EOPNOTSUPP;
1153
1154                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
1155                                 attr->vlan = tcf_vlan_push_vid(a);
1156                         } else { /* action is TCA_VLAN_ACT_MODIFY */
1157                                 return -EOPNOTSUPP;
1158                         }
1159                         continue;
1160                 }
1161
1162                 if (is_tcf_tunnel_release(a)) {
1163                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
1164                         continue;
1165                 }
1166
1167                 return -EINVAL;
1168         }
1169         return 0;
1170 }
1171
1172 int mlx5e_configure_flower(struct mlx5e_priv *priv, __be16 protocol,
1173                            struct tc_cls_flower_offload *f)
1174 {
1175         struct mlx5e_tc_table *tc = &priv->fs.tc;
1176         int err, attr_size = 0;
1177         u32 flow_tag, action;
1178         struct mlx5e_tc_flow *flow;
1179         struct mlx5_flow_spec *spec;
1180         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1181         u8 flow_flags = 0;
1182
1183         if (esw && esw->mode == SRIOV_OFFLOADS) {
1184                 flow_flags = MLX5E_TC_FLOW_ESWITCH;
1185                 attr_size  = sizeof(struct mlx5_esw_flow_attr);
1186         }
1187
1188         flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
1189         spec = mlx5_vzalloc(sizeof(*spec));
1190         if (!spec || !flow) {
1191                 err = -ENOMEM;
1192                 goto err_free;
1193         }
1194
1195         flow->cookie = f->cookie;
1196         flow->flags = flow_flags;
1197
1198         err = parse_cls_flower(priv, flow, spec, f);
1199         if (err < 0)
1200                 goto err_free;
1201
1202         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
1203                 flow->attr  = (struct mlx5_esw_flow_attr *)(flow + 1);
1204                 err = parse_tc_fdb_actions(priv, f->exts, flow);
1205                 if (err < 0)
1206                         goto err_free;
1207                 flow->rule = mlx5e_tc_add_fdb_flow(priv, spec, flow->attr);
1208         } else {
1209                 err = parse_tc_nic_actions(priv, f->exts, &action, &flow_tag);
1210                 if (err < 0)
1211                         goto err_free;
1212                 flow->rule = mlx5e_tc_add_nic_flow(priv, spec, action, flow_tag);
1213         }
1214
1215         if (IS_ERR(flow->rule)) {
1216                 err = PTR_ERR(flow->rule);
1217                 goto err_del_rule;
1218         }
1219
1220         err = rhashtable_insert_fast(&tc->ht, &flow->node,
1221                                      tc->ht_params);
1222         if (err)
1223                 goto err_del_rule;
1224
1225         goto out;
1226
1227 err_del_rule:
1228         mlx5e_tc_del_flow(priv, flow);
1229
1230 err_free:
1231         kfree(flow);
1232 out:
1233         kvfree(spec);
1234         return err;
1235 }
1236
1237 int mlx5e_delete_flower(struct mlx5e_priv *priv,
1238                         struct tc_cls_flower_offload *f)
1239 {
1240         struct mlx5e_tc_flow *flow;
1241         struct mlx5e_tc_table *tc = &priv->fs.tc;
1242
1243         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
1244                                       tc->ht_params);
1245         if (!flow)
1246                 return -EINVAL;
1247
1248         rhashtable_remove_fast(&tc->ht, &flow->node, tc->ht_params);
1249
1250         mlx5e_tc_del_flow(priv, flow);
1251
1252
1253         kfree(flow);
1254
1255         return 0;
1256 }
1257
1258 int mlx5e_stats_flower(struct mlx5e_priv *priv,
1259                        struct tc_cls_flower_offload *f)
1260 {
1261         struct mlx5e_tc_table *tc = &priv->fs.tc;
1262         struct mlx5e_tc_flow *flow;
1263         struct tc_action *a;
1264         struct mlx5_fc *counter;
1265         LIST_HEAD(actions);
1266         u64 bytes;
1267         u64 packets;
1268         u64 lastuse;
1269
1270         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
1271                                       tc->ht_params);
1272         if (!flow)
1273                 return -EINVAL;
1274
1275         counter = mlx5_flow_rule_counter(flow->rule);
1276         if (!counter)
1277                 return 0;
1278
1279         mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
1280
1281         preempt_disable();
1282
1283         tcf_exts_to_list(f->exts, &actions);
1284         list_for_each_entry(a, &actions, list)
1285                 tcf_action_stats_update(a, bytes, packets, lastuse);
1286
1287         preempt_enable();
1288
1289         return 0;
1290 }
1291
1292 static const struct rhashtable_params mlx5e_tc_flow_ht_params = {
1293         .head_offset = offsetof(struct mlx5e_tc_flow, node),
1294         .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
1295         .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
1296         .automatic_shrinking = true,
1297 };
1298
1299 int mlx5e_tc_init(struct mlx5e_priv *priv)
1300 {
1301         struct mlx5e_tc_table *tc = &priv->fs.tc;
1302
1303         tc->ht_params = mlx5e_tc_flow_ht_params;
1304         return rhashtable_init(&tc->ht, &tc->ht_params);
1305 }
1306
1307 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
1308 {
1309         struct mlx5e_tc_flow *flow = ptr;
1310         struct mlx5e_priv *priv = arg;
1311
1312         mlx5e_tc_del_flow(priv, flow);
1313         kfree(flow);
1314 }
1315
1316 void mlx5e_tc_cleanup(struct mlx5e_priv *priv)
1317 {
1318         struct mlx5e_tc_table *tc = &priv->fs.tc;
1319
1320         rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, priv);
1321
1322         if (!IS_ERR_OR_NULL(tc->t)) {
1323                 mlx5_destroy_flow_table(tc->t);
1324                 tc->t = NULL;
1325         }
1326 }