ethdev: deprecate legacy filter API
[dpdk.git] / drivers / net / mlx5 / mlx5_flow.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11
12 /* Verbs header. */
13 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
14 #ifdef PEDANTIC
15 #pragma GCC diagnostic ignored "-Wpedantic"
16 #endif
17 #include <infiniband/verbs.h>
18 #ifdef PEDANTIC
19 #pragma GCC diagnostic error "-Wpedantic"
20 #endif
21
22 #include <rte_common.h>
23 #include <rte_ether.h>
24 #include <rte_ethdev_driver.h>
25 #include <rte_flow.h>
26 #include <rte_flow_driver.h>
27 #include <rte_malloc.h>
28 #include <rte_ip.h>
29
30 #include "mlx5.h"
31 #include "mlx5_defs.h"
32 #include "mlx5_flow.h"
33 #include "mlx5_glue.h"
34 #include "mlx5_prm.h"
35 #include "mlx5_rxtx.h"
36
37 /* Dev ops structure defined in mlx5.c */
38 extern const struct eth_dev_ops mlx5_dev_ops;
39 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
40
41 /** Device flow drivers. */
42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
43 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
44 #endif
45 extern const struct mlx5_flow_driver_ops mlx5_flow_tcf_drv_ops;
46 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
47
48 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
49
50 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
51         [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
52 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
53         [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
54 #endif
55         [MLX5_FLOW_TYPE_TCF] = &mlx5_flow_tcf_drv_ops,
56         [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
57         [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
58 };
59
60 enum mlx5_expansion {
61         MLX5_EXPANSION_ROOT,
62         MLX5_EXPANSION_ROOT_OUTER,
63         MLX5_EXPANSION_ROOT_ETH_VLAN,
64         MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
65         MLX5_EXPANSION_OUTER_ETH,
66         MLX5_EXPANSION_OUTER_ETH_VLAN,
67         MLX5_EXPANSION_OUTER_VLAN,
68         MLX5_EXPANSION_OUTER_IPV4,
69         MLX5_EXPANSION_OUTER_IPV4_UDP,
70         MLX5_EXPANSION_OUTER_IPV4_TCP,
71         MLX5_EXPANSION_OUTER_IPV6,
72         MLX5_EXPANSION_OUTER_IPV6_UDP,
73         MLX5_EXPANSION_OUTER_IPV6_TCP,
74         MLX5_EXPANSION_VXLAN,
75         MLX5_EXPANSION_VXLAN_GPE,
76         MLX5_EXPANSION_GRE,
77         MLX5_EXPANSION_MPLS,
78         MLX5_EXPANSION_ETH,
79         MLX5_EXPANSION_ETH_VLAN,
80         MLX5_EXPANSION_VLAN,
81         MLX5_EXPANSION_IPV4,
82         MLX5_EXPANSION_IPV4_UDP,
83         MLX5_EXPANSION_IPV4_TCP,
84         MLX5_EXPANSION_IPV6,
85         MLX5_EXPANSION_IPV6_UDP,
86         MLX5_EXPANSION_IPV6_TCP,
87 };
88
89 /** Supported expansion of items. */
90 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
91         [MLX5_EXPANSION_ROOT] = {
92                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
93                                                  MLX5_EXPANSION_IPV4,
94                                                  MLX5_EXPANSION_IPV6),
95                 .type = RTE_FLOW_ITEM_TYPE_END,
96         },
97         [MLX5_EXPANSION_ROOT_OUTER] = {
98                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
99                                                  MLX5_EXPANSION_OUTER_IPV4,
100                                                  MLX5_EXPANSION_OUTER_IPV6),
101                 .type = RTE_FLOW_ITEM_TYPE_END,
102         },
103         [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
104                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
105                 .type = RTE_FLOW_ITEM_TYPE_END,
106         },
107         [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
108                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
109                 .type = RTE_FLOW_ITEM_TYPE_END,
110         },
111         [MLX5_EXPANSION_OUTER_ETH] = {
112                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
113                                                  MLX5_EXPANSION_OUTER_IPV6,
114                                                  MLX5_EXPANSION_MPLS),
115                 .type = RTE_FLOW_ITEM_TYPE_ETH,
116                 .rss_types = 0,
117         },
118         [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
119                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
120                 .type = RTE_FLOW_ITEM_TYPE_ETH,
121                 .rss_types = 0,
122         },
123         [MLX5_EXPANSION_OUTER_VLAN] = {
124                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
125                                                  MLX5_EXPANSION_OUTER_IPV6),
126                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
127         },
128         [MLX5_EXPANSION_OUTER_IPV4] = {
129                 .next = RTE_FLOW_EXPAND_RSS_NEXT
130                         (MLX5_EXPANSION_OUTER_IPV4_UDP,
131                          MLX5_EXPANSION_OUTER_IPV4_TCP,
132                          MLX5_EXPANSION_GRE),
133                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
134                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
135                         ETH_RSS_NONFRAG_IPV4_OTHER,
136         },
137         [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
138                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
139                                                  MLX5_EXPANSION_VXLAN_GPE),
140                 .type = RTE_FLOW_ITEM_TYPE_UDP,
141                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
142         },
143         [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
144                 .type = RTE_FLOW_ITEM_TYPE_TCP,
145                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
146         },
147         [MLX5_EXPANSION_OUTER_IPV6] = {
148                 .next = RTE_FLOW_EXPAND_RSS_NEXT
149                         (MLX5_EXPANSION_OUTER_IPV6_UDP,
150                          MLX5_EXPANSION_OUTER_IPV6_TCP),
151                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
152                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
153                         ETH_RSS_NONFRAG_IPV6_OTHER,
154         },
155         [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
156                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
157                                                  MLX5_EXPANSION_VXLAN_GPE),
158                 .type = RTE_FLOW_ITEM_TYPE_UDP,
159                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
160         },
161         [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
162                 .type = RTE_FLOW_ITEM_TYPE_TCP,
163                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
164         },
165         [MLX5_EXPANSION_VXLAN] = {
166                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
167                 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
168         },
169         [MLX5_EXPANSION_VXLAN_GPE] = {
170                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
171                                                  MLX5_EXPANSION_IPV4,
172                                                  MLX5_EXPANSION_IPV6),
173                 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
174         },
175         [MLX5_EXPANSION_GRE] = {
176                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
177                 .type = RTE_FLOW_ITEM_TYPE_GRE,
178         },
179         [MLX5_EXPANSION_MPLS] = {
180                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
181                                                  MLX5_EXPANSION_IPV6),
182                 .type = RTE_FLOW_ITEM_TYPE_MPLS,
183         },
184         [MLX5_EXPANSION_ETH] = {
185                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
186                                                  MLX5_EXPANSION_IPV6),
187                 .type = RTE_FLOW_ITEM_TYPE_ETH,
188         },
189         [MLX5_EXPANSION_ETH_VLAN] = {
190                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
191                 .type = RTE_FLOW_ITEM_TYPE_ETH,
192         },
193         [MLX5_EXPANSION_VLAN] = {
194                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
195                                                  MLX5_EXPANSION_IPV6),
196                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
197         },
198         [MLX5_EXPANSION_IPV4] = {
199                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
200                                                  MLX5_EXPANSION_IPV4_TCP),
201                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
202                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
203                         ETH_RSS_NONFRAG_IPV4_OTHER,
204         },
205         [MLX5_EXPANSION_IPV4_UDP] = {
206                 .type = RTE_FLOW_ITEM_TYPE_UDP,
207                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
208         },
209         [MLX5_EXPANSION_IPV4_TCP] = {
210                 .type = RTE_FLOW_ITEM_TYPE_TCP,
211                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
212         },
213         [MLX5_EXPANSION_IPV6] = {
214                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
215                                                  MLX5_EXPANSION_IPV6_TCP),
216                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
217                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
218                         ETH_RSS_NONFRAG_IPV6_OTHER,
219         },
220         [MLX5_EXPANSION_IPV6_UDP] = {
221                 .type = RTE_FLOW_ITEM_TYPE_UDP,
222                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
223         },
224         [MLX5_EXPANSION_IPV6_TCP] = {
225                 .type = RTE_FLOW_ITEM_TYPE_TCP,
226                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
227         },
228 };
229
230 static const struct rte_flow_ops mlx5_flow_ops = {
231         .validate = mlx5_flow_validate,
232         .create = mlx5_flow_create,
233         .destroy = mlx5_flow_destroy,
234         .flush = mlx5_flow_flush,
235         .isolate = mlx5_flow_isolate,
236         .query = mlx5_flow_query,
237 };
238
239 /* Convert FDIR request to Generic flow. */
240 struct mlx5_fdir {
241         struct rte_flow_attr attr;
242         struct rte_flow_item items[4];
243         struct rte_flow_item_eth l2;
244         struct rte_flow_item_eth l2_mask;
245         union {
246                 struct rte_flow_item_ipv4 ipv4;
247                 struct rte_flow_item_ipv6 ipv6;
248         } l3;
249         union {
250                 struct rte_flow_item_ipv4 ipv4;
251                 struct rte_flow_item_ipv6 ipv6;
252         } l3_mask;
253         union {
254                 struct rte_flow_item_udp udp;
255                 struct rte_flow_item_tcp tcp;
256         } l4;
257         union {
258                 struct rte_flow_item_udp udp;
259                 struct rte_flow_item_tcp tcp;
260         } l4_mask;
261         struct rte_flow_action actions[2];
262         struct rte_flow_action_queue queue;
263 };
264
265 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
266 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
267         { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
268 };
269
270 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
271 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
272         { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
273         { 9, 10, 11 }, { 12, 13, 14 },
274 };
275
276 /* Tunnel information. */
277 struct mlx5_flow_tunnel_info {
278         uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
279         uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
280 };
281
282 static struct mlx5_flow_tunnel_info tunnels_info[] = {
283         {
284                 .tunnel = MLX5_FLOW_LAYER_VXLAN,
285                 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
286         },
287         {
288                 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
289                 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
290         },
291         {
292                 .tunnel = MLX5_FLOW_LAYER_GRE,
293                 .ptype = RTE_PTYPE_TUNNEL_GRE,
294         },
295         {
296                 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
297                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
298         },
299         {
300                 .tunnel = MLX5_FLOW_LAYER_MPLS,
301                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
302         },
303 };
304
305 /**
306  * Discover the maximum number of priority available.
307  *
308  * @param[in] dev
309  *   Pointer to the Ethernet device structure.
310  *
311  * @return
312  *   number of supported flow priority on success, a negative errno
313  *   value otherwise and rte_errno is set.
314  */
315 int
316 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
317 {
318         struct mlx5_priv *priv = dev->data->dev_private;
319         struct {
320                 struct ibv_flow_attr attr;
321                 struct ibv_flow_spec_eth eth;
322                 struct ibv_flow_spec_action_drop drop;
323         } flow_attr = {
324                 .attr = {
325                         .num_of_specs = 2,
326                         .port = (uint8_t)priv->ibv_port,
327                 },
328                 .eth = {
329                         .type = IBV_FLOW_SPEC_ETH,
330                         .size = sizeof(struct ibv_flow_spec_eth),
331                 },
332                 .drop = {
333                         .size = sizeof(struct ibv_flow_spec_action_drop),
334                         .type = IBV_FLOW_SPEC_ACTION_DROP,
335                 },
336         };
337         struct ibv_flow *flow;
338         struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
339         uint16_t vprio[] = { 8, 16 };
340         int i;
341         int priority = 0;
342
343         if (!drop) {
344                 rte_errno = ENOTSUP;
345                 return -rte_errno;
346         }
347         for (i = 0; i != RTE_DIM(vprio); i++) {
348                 flow_attr.attr.priority = vprio[i] - 1;
349                 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
350                 if (!flow)
351                         break;
352                 claim_zero(mlx5_glue->destroy_flow(flow));
353                 priority = vprio[i];
354         }
355         mlx5_hrxq_drop_release(dev);
356         switch (priority) {
357         case 8:
358                 priority = RTE_DIM(priority_map_3);
359                 break;
360         case 16:
361                 priority = RTE_DIM(priority_map_5);
362                 break;
363         default:
364                 rte_errno = ENOTSUP;
365                 DRV_LOG(ERR,
366                         "port %u verbs maximum priority: %d expected 8/16",
367                         dev->data->port_id, priority);
368                 return -rte_errno;
369         }
370         DRV_LOG(INFO, "port %u flow maximum priority: %d",
371                 dev->data->port_id, priority);
372         return priority;
373 }
374
375 /**
376  * Adjust flow priority based on the highest layer and the request priority.
377  *
378  * @param[in] dev
379  *   Pointer to the Ethernet device structure.
380  * @param[in] priority
381  *   The rule base priority.
382  * @param[in] subpriority
383  *   The priority based on the items.
384  *
385  * @return
386  *   The new priority.
387  */
388 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
389                                    uint32_t subpriority)
390 {
391         uint32_t res = 0;
392         struct mlx5_priv *priv = dev->data->dev_private;
393
394         switch (priv->config.flow_prio) {
395         case RTE_DIM(priority_map_3):
396                 res = priority_map_3[priority][subpriority];
397                 break;
398         case RTE_DIM(priority_map_5):
399                 res = priority_map_5[priority][subpriority];
400                 break;
401         }
402         return  res;
403 }
404
405 /**
406  * Verify the @p item specifications (spec, last, mask) are compatible with the
407  * NIC capabilities.
408  *
409  * @param[in] item
410  *   Item specification.
411  * @param[in] mask
412  *   @p item->mask or flow default bit-masks.
413  * @param[in] nic_mask
414  *   Bit-masks covering supported fields by the NIC to compare with user mask.
415  * @param[in] size
416  *   Bit-masks size in bytes.
417  * @param[out] error
418  *   Pointer to error structure.
419  *
420  * @return
421  *   0 on success, a negative errno value otherwise and rte_errno is set.
422  */
423 int
424 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
425                           const uint8_t *mask,
426                           const uint8_t *nic_mask,
427                           unsigned int size,
428                           struct rte_flow_error *error)
429 {
430         unsigned int i;
431
432         assert(nic_mask);
433         for (i = 0; i < size; ++i)
434                 if ((nic_mask[i] | mask[i]) != nic_mask[i])
435                         return rte_flow_error_set(error, ENOTSUP,
436                                                   RTE_FLOW_ERROR_TYPE_ITEM,
437                                                   item,
438                                                   "mask enables non supported"
439                                                   " bits");
440         if (!item->spec && (item->mask || item->last))
441                 return rte_flow_error_set(error, EINVAL,
442                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
443                                           "mask/last without a spec is not"
444                                           " supported");
445         if (item->spec && item->last) {
446                 uint8_t spec[size];
447                 uint8_t last[size];
448                 unsigned int i;
449                 int ret;
450
451                 for (i = 0; i < size; ++i) {
452                         spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
453                         last[i] = ((const uint8_t *)item->last)[i] & mask[i];
454                 }
455                 ret = memcmp(spec, last, size);
456                 if (ret != 0)
457                         return rte_flow_error_set(error, EINVAL,
458                                                   RTE_FLOW_ERROR_TYPE_ITEM,
459                                                   item,
460                                                   "range is not valid");
461         }
462         return 0;
463 }
464
465 /**
466  * Adjust the hash fields according to the @p flow information.
467  *
468  * @param[in] dev_flow.
469  *   Pointer to the mlx5_flow.
470  * @param[in] tunnel
471  *   1 when the hash field is for a tunnel item.
472  * @param[in] layer_types
473  *   ETH_RSS_* types.
474  * @param[in] hash_fields
475  *   Item hash fields.
476  *
477  * @return
478  *   The hash fileds that should be used.
479  */
480 uint64_t
481 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
482                             int tunnel __rte_unused, uint64_t layer_types,
483                             uint64_t hash_fields)
484 {
485         struct rte_flow *flow = dev_flow->flow;
486 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
487         int rss_request_inner = flow->rss.level >= 2;
488
489         /* Check RSS hash level for tunnel. */
490         if (tunnel && rss_request_inner)
491                 hash_fields |= IBV_RX_HASH_INNER;
492         else if (tunnel || rss_request_inner)
493                 return 0;
494 #endif
495         /* Check if requested layer matches RSS hash fields. */
496         if (!(flow->rss.types & layer_types))
497                 return 0;
498         return hash_fields;
499 }
500
501 /**
502  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
503  * if several tunnel rules are used on this queue, the tunnel ptype will be
504  * cleared.
505  *
506  * @param rxq_ctrl
507  *   Rx queue to update.
508  */
509 static void
510 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
511 {
512         unsigned int i;
513         uint32_t tunnel_ptype = 0;
514
515         /* Look up for the ptype to use. */
516         for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
517                 if (!rxq_ctrl->flow_tunnels_n[i])
518                         continue;
519                 if (!tunnel_ptype) {
520                         tunnel_ptype = tunnels_info[i].ptype;
521                 } else {
522                         tunnel_ptype = 0;
523                         break;
524                 }
525         }
526         rxq_ctrl->rxq.tunnel = tunnel_ptype;
527 }
528
529 /**
530  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
531  * flow.
532  *
533  * @param[in] dev
534  *   Pointer to the Ethernet device structure.
535  * @param[in] dev_flow
536  *   Pointer to device flow structure.
537  */
538 static void
539 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
540 {
541         struct mlx5_priv *priv = dev->data->dev_private;
542         struct rte_flow *flow = dev_flow->flow;
543         const int mark = !!(flow->actions &
544                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
545         const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
546         unsigned int i;
547
548         for (i = 0; i != flow->rss.queue_num; ++i) {
549                 int idx = (*flow->queue)[i];
550                 struct mlx5_rxq_ctrl *rxq_ctrl =
551                         container_of((*priv->rxqs)[idx],
552                                      struct mlx5_rxq_ctrl, rxq);
553
554                 if (mark) {
555                         rxq_ctrl->rxq.mark = 1;
556                         rxq_ctrl->flow_mark_n++;
557                 }
558                 if (tunnel) {
559                         unsigned int j;
560
561                         /* Increase the counter matching the flow. */
562                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
563                                 if ((tunnels_info[j].tunnel &
564                                      dev_flow->layers) ==
565                                     tunnels_info[j].tunnel) {
566                                         rxq_ctrl->flow_tunnels_n[j]++;
567                                         break;
568                                 }
569                         }
570                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
571                 }
572         }
573 }
574
575 /**
576  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
577  *
578  * @param[in] dev
579  *   Pointer to the Ethernet device structure.
580  * @param[in] flow
581  *   Pointer to flow structure.
582  */
583 static void
584 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
585 {
586         struct mlx5_flow *dev_flow;
587
588         LIST_FOREACH(dev_flow, &flow->dev_flows, next)
589                 flow_drv_rxq_flags_set(dev, dev_flow);
590 }
591
592 /**
593  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
594  * device flow if no other flow uses it with the same kind of request.
595  *
596  * @param dev
597  *   Pointer to Ethernet device.
598  * @param[in] dev_flow
599  *   Pointer to the device flow.
600  */
601 static void
602 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
603 {
604         struct mlx5_priv *priv = dev->data->dev_private;
605         struct rte_flow *flow = dev_flow->flow;
606         const int mark = !!(flow->actions &
607                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
608         const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
609         unsigned int i;
610
611         assert(dev->data->dev_started);
612         for (i = 0; i != flow->rss.queue_num; ++i) {
613                 int idx = (*flow->queue)[i];
614                 struct mlx5_rxq_ctrl *rxq_ctrl =
615                         container_of((*priv->rxqs)[idx],
616                                      struct mlx5_rxq_ctrl, rxq);
617
618                 if (mark) {
619                         rxq_ctrl->flow_mark_n--;
620                         rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
621                 }
622                 if (tunnel) {
623                         unsigned int j;
624
625                         /* Decrease the counter matching the flow. */
626                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
627                                 if ((tunnels_info[j].tunnel &
628                                      dev_flow->layers) ==
629                                     tunnels_info[j].tunnel) {
630                                         rxq_ctrl->flow_tunnels_n[j]--;
631                                         break;
632                                 }
633                         }
634                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
635                 }
636         }
637 }
638
639 /**
640  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
641  * @p flow if no other flow uses it with the same kind of request.
642  *
643  * @param dev
644  *   Pointer to Ethernet device.
645  * @param[in] flow
646  *   Pointer to the flow.
647  */
648 static void
649 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
650 {
651         struct mlx5_flow *dev_flow;
652
653         LIST_FOREACH(dev_flow, &flow->dev_flows, next)
654                 flow_drv_rxq_flags_trim(dev, dev_flow);
655 }
656
657 /**
658  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
659  *
660  * @param dev
661  *   Pointer to Ethernet device.
662  */
663 static void
664 flow_rxq_flags_clear(struct rte_eth_dev *dev)
665 {
666         struct mlx5_priv *priv = dev->data->dev_private;
667         unsigned int i;
668
669         for (i = 0; i != priv->rxqs_n; ++i) {
670                 struct mlx5_rxq_ctrl *rxq_ctrl;
671                 unsigned int j;
672
673                 if (!(*priv->rxqs)[i])
674                         continue;
675                 rxq_ctrl = container_of((*priv->rxqs)[i],
676                                         struct mlx5_rxq_ctrl, rxq);
677                 rxq_ctrl->flow_mark_n = 0;
678                 rxq_ctrl->rxq.mark = 0;
679                 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
680                         rxq_ctrl->flow_tunnels_n[j] = 0;
681                 rxq_ctrl->rxq.tunnel = 0;
682         }
683 }
684
685 /*
686  * Validate the flag action.
687  *
688  * @param[in] action_flags
689  *   Bit-fields that holds the actions detected until now.
690  * @param[in] attr
691  *   Attributes of flow that includes this action.
692  * @param[out] error
693  *   Pointer to error structure.
694  *
695  * @return
696  *   0 on success, a negative errno value otherwise and rte_errno is set.
697  */
698 int
699 mlx5_flow_validate_action_flag(uint64_t action_flags,
700                                const struct rte_flow_attr *attr,
701                                struct rte_flow_error *error)
702 {
703
704         if (action_flags & MLX5_FLOW_ACTION_DROP)
705                 return rte_flow_error_set(error, EINVAL,
706                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
707                                           "can't drop and flag in same flow");
708         if (action_flags & MLX5_FLOW_ACTION_MARK)
709                 return rte_flow_error_set(error, EINVAL,
710                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
711                                           "can't mark and flag in same flow");
712         if (action_flags & MLX5_FLOW_ACTION_FLAG)
713                 return rte_flow_error_set(error, EINVAL,
714                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
715                                           "can't have 2 flag"
716                                           " actions in same flow");
717         if (attr->egress)
718                 return rte_flow_error_set(error, ENOTSUP,
719                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
720                                           "flag action not supported for "
721                                           "egress");
722         return 0;
723 }
724
725 /*
726  * Validate the mark action.
727  *
728  * @param[in] action
729  *   Pointer to the queue action.
730  * @param[in] action_flags
731  *   Bit-fields that holds the actions detected until now.
732  * @param[in] attr
733  *   Attributes of flow that includes this action.
734  * @param[out] error
735  *   Pointer to error structure.
736  *
737  * @return
738  *   0 on success, a negative errno value otherwise and rte_errno is set.
739  */
740 int
741 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
742                                uint64_t action_flags,
743                                const struct rte_flow_attr *attr,
744                                struct rte_flow_error *error)
745 {
746         const struct rte_flow_action_mark *mark = action->conf;
747
748         if (!mark)
749                 return rte_flow_error_set(error, EINVAL,
750                                           RTE_FLOW_ERROR_TYPE_ACTION,
751                                           action,
752                                           "configuration cannot be null");
753         if (mark->id >= MLX5_FLOW_MARK_MAX)
754                 return rte_flow_error_set(error, EINVAL,
755                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
756                                           &mark->id,
757                                           "mark id must in 0 <= id < "
758                                           RTE_STR(MLX5_FLOW_MARK_MAX));
759         if (action_flags & MLX5_FLOW_ACTION_DROP)
760                 return rte_flow_error_set(error, EINVAL,
761                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
762                                           "can't drop and mark in same flow");
763         if (action_flags & MLX5_FLOW_ACTION_FLAG)
764                 return rte_flow_error_set(error, EINVAL,
765                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
766                                           "can't flag and mark in same flow");
767         if (action_flags & MLX5_FLOW_ACTION_MARK)
768                 return rte_flow_error_set(error, EINVAL,
769                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
770                                           "can't have 2 mark actions in same"
771                                           " flow");
772         if (attr->egress)
773                 return rte_flow_error_set(error, ENOTSUP,
774                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
775                                           "mark action not supported for "
776                                           "egress");
777         return 0;
778 }
779
780 /*
781  * Validate the drop action.
782  *
783  * @param[in] action_flags
784  *   Bit-fields that holds the actions detected until now.
785  * @param[in] attr
786  *   Attributes of flow that includes this action.
787  * @param[out] error
788  *   Pointer to error structure.
789  *
790  * @return
791  *   0 on success, a negative errno value otherwise and rte_errno is set.
792  */
793 int
794 mlx5_flow_validate_action_drop(uint64_t action_flags,
795                                const struct rte_flow_attr *attr,
796                                struct rte_flow_error *error)
797 {
798         if (action_flags & MLX5_FLOW_ACTION_FLAG)
799                 return rte_flow_error_set(error, EINVAL,
800                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
801                                           "can't drop and flag in same flow");
802         if (action_flags & MLX5_FLOW_ACTION_MARK)
803                 return rte_flow_error_set(error, EINVAL,
804                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
805                                           "can't drop and mark in same flow");
806         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
807                 return rte_flow_error_set(error, EINVAL,
808                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
809                                           "can't have 2 fate actions in"
810                                           " same flow");
811         if (attr->egress)
812                 return rte_flow_error_set(error, ENOTSUP,
813                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
814                                           "drop action not supported for "
815                                           "egress");
816         return 0;
817 }
818
819 /*
820  * Validate the queue action.
821  *
822  * @param[in] action
823  *   Pointer to the queue action.
824  * @param[in] action_flags
825  *   Bit-fields that holds the actions detected until now.
826  * @param[in] dev
827  *   Pointer to the Ethernet device structure.
828  * @param[in] attr
829  *   Attributes of flow that includes this action.
830  * @param[out] error
831  *   Pointer to error structure.
832  *
833  * @return
834  *   0 on success, a negative errno value otherwise and rte_errno is set.
835  */
836 int
837 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
838                                 uint64_t action_flags,
839                                 struct rte_eth_dev *dev,
840                                 const struct rte_flow_attr *attr,
841                                 struct rte_flow_error *error)
842 {
843         struct mlx5_priv *priv = dev->data->dev_private;
844         const struct rte_flow_action_queue *queue = action->conf;
845
846         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
847                 return rte_flow_error_set(error, EINVAL,
848                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
849                                           "can't have 2 fate actions in"
850                                           " same flow");
851         if (!priv->rxqs_n)
852                 return rte_flow_error_set(error, EINVAL,
853                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
854                                           NULL, "No Rx queues configured");
855         if (queue->index >= priv->rxqs_n)
856                 return rte_flow_error_set(error, EINVAL,
857                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
858                                           &queue->index,
859                                           "queue index out of range");
860         if (!(*priv->rxqs)[queue->index])
861                 return rte_flow_error_set(error, EINVAL,
862                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
863                                           &queue->index,
864                                           "queue is not configured");
865         if (attr->egress)
866                 return rte_flow_error_set(error, ENOTSUP,
867                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
868                                           "queue action not supported for "
869                                           "egress");
870         return 0;
871 }
872
873 /*
874  * Validate the rss action.
875  *
876  * @param[in] action
877  *   Pointer to the queue action.
878  * @param[in] action_flags
879  *   Bit-fields that holds the actions detected until now.
880  * @param[in] dev
881  *   Pointer to the Ethernet device structure.
882  * @param[in] attr
883  *   Attributes of flow that includes this action.
884  * @param[in] item_flags
885  *   Items that were detected.
886  * @param[out] error
887  *   Pointer to error structure.
888  *
889  * @return
890  *   0 on success, a negative errno value otherwise and rte_errno is set.
891  */
892 int
893 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
894                               uint64_t action_flags,
895                               struct rte_eth_dev *dev,
896                               const struct rte_flow_attr *attr,
897                               uint64_t item_flags,
898                               struct rte_flow_error *error)
899 {
900         struct mlx5_priv *priv = dev->data->dev_private;
901         const struct rte_flow_action_rss *rss = action->conf;
902         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
903         unsigned int i;
904
905         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
906                 return rte_flow_error_set(error, EINVAL,
907                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
908                                           "can't have 2 fate actions"
909                                           " in same flow");
910         if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
911             rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
912                 return rte_flow_error_set(error, ENOTSUP,
913                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
914                                           &rss->func,
915                                           "RSS hash function not supported");
916 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
917         if (rss->level > 2)
918 #else
919         if (rss->level > 1)
920 #endif
921                 return rte_flow_error_set(error, ENOTSUP,
922                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
923                                           &rss->level,
924                                           "tunnel RSS is not supported");
925         /* allow RSS key_len 0 in case of NULL (default) RSS key. */
926         if (rss->key_len == 0 && rss->key != NULL)
927                 return rte_flow_error_set(error, ENOTSUP,
928                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
929                                           &rss->key_len,
930                                           "RSS hash key length 0");
931         if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
932                 return rte_flow_error_set(error, ENOTSUP,
933                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
934                                           &rss->key_len,
935                                           "RSS hash key too small");
936         if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
937                 return rte_flow_error_set(error, ENOTSUP,
938                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
939                                           &rss->key_len,
940                                           "RSS hash key too large");
941         if (rss->queue_num > priv->config.ind_table_max_size)
942                 return rte_flow_error_set(error, ENOTSUP,
943                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
944                                           &rss->queue_num,
945                                           "number of queues too large");
946         if (rss->types & MLX5_RSS_HF_MASK)
947                 return rte_flow_error_set(error, ENOTSUP,
948                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
949                                           &rss->types,
950                                           "some RSS protocols are not"
951                                           " supported");
952         if (!priv->rxqs_n)
953                 return rte_flow_error_set(error, EINVAL,
954                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
955                                           NULL, "No Rx queues configured");
956         if (!rss->queue_num)
957                 return rte_flow_error_set(error, EINVAL,
958                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
959                                           NULL, "No queues configured");
960         for (i = 0; i != rss->queue_num; ++i) {
961                 if (!(*priv->rxqs)[rss->queue[i]])
962                         return rte_flow_error_set
963                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
964                                  &rss->queue[i], "queue is not configured");
965         }
966         if (attr->egress)
967                 return rte_flow_error_set(error, ENOTSUP,
968                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
969                                           "rss action not supported for "
970                                           "egress");
971         if (rss->level > 1 &&  !tunnel)
972                 return rte_flow_error_set(error, EINVAL,
973                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
974                                           "inner RSS is not supported for "
975                                           "non-tunnel flows");
976         return 0;
977 }
978
979 /*
980  * Validate the count action.
981  *
982  * @param[in] dev
983  *   Pointer to the Ethernet device structure.
984  * @param[in] attr
985  *   Attributes of flow that includes this action.
986  * @param[out] error
987  *   Pointer to error structure.
988  *
989  * @return
990  *   0 on success, a negative errno value otherwise and rte_errno is set.
991  */
992 int
993 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
994                                 const struct rte_flow_attr *attr,
995                                 struct rte_flow_error *error)
996 {
997         if (attr->egress)
998                 return rte_flow_error_set(error, ENOTSUP,
999                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1000                                           "count action not supported for "
1001                                           "egress");
1002         return 0;
1003 }
1004
1005 /**
1006  * Verify the @p attributes will be correctly understood by the NIC and store
1007  * them in the @p flow if everything is correct.
1008  *
1009  * @param[in] dev
1010  *   Pointer to the Ethernet device structure.
1011  * @param[in] attributes
1012  *   Pointer to flow attributes
1013  * @param[out] error
1014  *   Pointer to error structure.
1015  *
1016  * @return
1017  *   0 on success, a negative errno value otherwise and rte_errno is set.
1018  */
1019 int
1020 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1021                               const struct rte_flow_attr *attributes,
1022                               struct rte_flow_error *error)
1023 {
1024         struct mlx5_priv *priv = dev->data->dev_private;
1025         uint32_t priority_max = priv->config.flow_prio - 1;
1026
1027         if (attributes->group)
1028                 return rte_flow_error_set(error, ENOTSUP,
1029                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1030                                           NULL, "groups is not supported");
1031         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1032             attributes->priority >= priority_max)
1033                 return rte_flow_error_set(error, ENOTSUP,
1034                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1035                                           NULL, "priority out of range");
1036         if (attributes->egress)
1037                 return rte_flow_error_set(error, ENOTSUP,
1038                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1039                                           "egress is not supported");
1040         if (attributes->transfer)
1041                 return rte_flow_error_set(error, ENOTSUP,
1042                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1043                                           NULL, "transfer is not supported");
1044         if (!attributes->ingress)
1045                 return rte_flow_error_set(error, EINVAL,
1046                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1047                                           NULL,
1048                                           "ingress attribute is mandatory");
1049         return 0;
1050 }
1051
1052 /**
1053  * Validate Ethernet item.
1054  *
1055  * @param[in] item
1056  *   Item specification.
1057  * @param[in] item_flags
1058  *   Bit-fields that holds the items detected until now.
1059  * @param[out] error
1060  *   Pointer to error structure.
1061  *
1062  * @return
1063  *   0 on success, a negative errno value otherwise and rte_errno is set.
1064  */
1065 int
1066 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1067                             uint64_t item_flags,
1068                             struct rte_flow_error *error)
1069 {
1070         const struct rte_flow_item_eth *mask = item->mask;
1071         const struct rte_flow_item_eth nic_mask = {
1072                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1073                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1074                 .type = RTE_BE16(0xffff),
1075         };
1076         int ret;
1077         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1078         const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1079                                        MLX5_FLOW_LAYER_OUTER_L2;
1080
1081         if (item_flags & ethm)
1082                 return rte_flow_error_set(error, ENOTSUP,
1083                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1084                                           "multiple L2 layers not supported");
1085         if (!mask)
1086                 mask = &rte_flow_item_eth_mask;
1087         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1088                                         (const uint8_t *)&nic_mask,
1089                                         sizeof(struct rte_flow_item_eth),
1090                                         error);
1091         return ret;
1092 }
1093
1094 /**
1095  * Validate VLAN item.
1096  *
1097  * @param[in] item
1098  *   Item specification.
1099  * @param[in] item_flags
1100  *   Bit-fields that holds the items detected until now.
1101  * @param[out] error
1102  *   Pointer to error structure.
1103  *
1104  * @return
1105  *   0 on success, a negative errno value otherwise and rte_errno is set.
1106  */
1107 int
1108 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1109                              uint64_t item_flags,
1110                              struct rte_flow_error *error)
1111 {
1112         const struct rte_flow_item_vlan *spec = item->spec;
1113         const struct rte_flow_item_vlan *mask = item->mask;
1114         const struct rte_flow_item_vlan nic_mask = {
1115                 .tci = RTE_BE16(0x0fff),
1116                 .inner_type = RTE_BE16(0xffff),
1117         };
1118         uint16_t vlan_tag = 0;
1119         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1120         int ret;
1121         const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1122                                         MLX5_FLOW_LAYER_INNER_L4) :
1123                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1124                                         MLX5_FLOW_LAYER_OUTER_L4);
1125         const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1126                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1127
1128         if (item_flags & vlanm)
1129                 return rte_flow_error_set(error, EINVAL,
1130                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1131                                           "multiple VLAN layers not supported");
1132         else if ((item_flags & l34m) != 0)
1133                 return rte_flow_error_set(error, EINVAL,
1134                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1135                                           "L2 layer cannot follow L3/L4 layer");
1136         if (!mask)
1137                 mask = &rte_flow_item_vlan_mask;
1138         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1139                                         (const uint8_t *)&nic_mask,
1140                                         sizeof(struct rte_flow_item_vlan),
1141                                         error);
1142         if (ret)
1143                 return ret;
1144         if (spec) {
1145                 vlan_tag = spec->tci;
1146                 vlan_tag &= mask->tci;
1147         }
1148         /*
1149          * From verbs perspective an empty VLAN is equivalent
1150          * to a packet without VLAN layer.
1151          */
1152         if (!vlan_tag)
1153                 return rte_flow_error_set(error, EINVAL,
1154                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1155                                           item->spec,
1156                                           "VLAN cannot be empty");
1157         return 0;
1158 }
1159
1160 /**
1161  * Validate IPV4 item.
1162  *
1163  * @param[in] item
1164  *   Item specification.
1165  * @param[in] item_flags
1166  *   Bit-fields that holds the items detected until now.
1167  * @param[in] acc_mask
1168  *   Acceptable mask, if NULL default internal default mask
1169  *   will be used to check whether item fields are supported.
1170  * @param[out] error
1171  *   Pointer to error structure.
1172  *
1173  * @return
1174  *   0 on success, a negative errno value otherwise and rte_errno is set.
1175  */
1176 int
1177 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1178                              uint64_t item_flags,
1179                              const struct rte_flow_item_ipv4 *acc_mask,
1180                              struct rte_flow_error *error)
1181 {
1182         const struct rte_flow_item_ipv4 *mask = item->mask;
1183         const struct rte_flow_item_ipv4 nic_mask = {
1184                 .hdr = {
1185                         .src_addr = RTE_BE32(0xffffffff),
1186                         .dst_addr = RTE_BE32(0xffffffff),
1187                         .type_of_service = 0xff,
1188                         .next_proto_id = 0xff,
1189                 },
1190         };
1191         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1192         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1193                                       MLX5_FLOW_LAYER_OUTER_L3;
1194         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1195                                       MLX5_FLOW_LAYER_OUTER_L4;
1196         int ret;
1197
1198         if (item_flags & l3m)
1199                 return rte_flow_error_set(error, ENOTSUP,
1200                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1201                                           "multiple L3 layers not supported");
1202         else if (item_flags & l4m)
1203                 return rte_flow_error_set(error, EINVAL,
1204                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1205                                           "L3 cannot follow an L4 layer.");
1206         if (!mask)
1207                 mask = &rte_flow_item_ipv4_mask;
1208         else if (mask->hdr.next_proto_id != 0 &&
1209                  mask->hdr.next_proto_id != 0xff)
1210                 return rte_flow_error_set(error, EINVAL,
1211                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1212                                           "partial mask is not supported"
1213                                           " for protocol");
1214         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1215                                         acc_mask ? (const uint8_t *)acc_mask
1216                                                  : (const uint8_t *)&nic_mask,
1217                                         sizeof(struct rte_flow_item_ipv4),
1218                                         error);
1219         if (ret < 0)
1220                 return ret;
1221         return 0;
1222 }
1223
1224 /**
1225  * Validate IPV6 item.
1226  *
1227  * @param[in] item
1228  *   Item specification.
1229  * @param[in] item_flags
1230  *   Bit-fields that holds the items detected until now.
1231  * @param[in] acc_mask
1232  *   Acceptable mask, if NULL default internal default mask
1233  *   will be used to check whether item fields are supported.
1234  * @param[out] error
1235  *   Pointer to error structure.
1236  *
1237  * @return
1238  *   0 on success, a negative errno value otherwise and rte_errno is set.
1239  */
1240 int
1241 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1242                              uint64_t item_flags,
1243                              const struct rte_flow_item_ipv6 *acc_mask,
1244                              struct rte_flow_error *error)
1245 {
1246         const struct rte_flow_item_ipv6 *mask = item->mask;
1247         const struct rte_flow_item_ipv6 nic_mask = {
1248                 .hdr = {
1249                         .src_addr =
1250                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1251                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1252                         .dst_addr =
1253                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1254                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1255                         .vtc_flow = RTE_BE32(0xffffffff),
1256                         .proto = 0xff,
1257                         .hop_limits = 0xff,
1258                 },
1259         };
1260         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1261         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1262                                       MLX5_FLOW_LAYER_OUTER_L3;
1263         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1264                                       MLX5_FLOW_LAYER_OUTER_L4;
1265         int ret;
1266
1267         if (item_flags & l3m)
1268                 return rte_flow_error_set(error, ENOTSUP,
1269                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1270                                           "multiple L3 layers not supported");
1271         else if (item_flags & l4m)
1272                 return rte_flow_error_set(error, EINVAL,
1273                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1274                                           "L3 cannot follow an L4 layer.");
1275         if (!mask)
1276                 mask = &rte_flow_item_ipv6_mask;
1277         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1278                                         acc_mask ? (const uint8_t *)acc_mask
1279                                                  : (const uint8_t *)&nic_mask,
1280                                         sizeof(struct rte_flow_item_ipv6),
1281                                         error);
1282         if (ret < 0)
1283                 return ret;
1284         return 0;
1285 }
1286
1287 /**
1288  * Validate UDP item.
1289  *
1290  * @param[in] item
1291  *   Item specification.
1292  * @param[in] item_flags
1293  *   Bit-fields that holds the items detected until now.
1294  * @param[in] target_protocol
1295  *   The next protocol in the previous item.
1296  * @param[in] flow_mask
1297  *   mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1298  * @param[out] error
1299  *   Pointer to error structure.
1300  *
1301  * @return
1302  *   0 on success, a negative errno value otherwise and rte_errno is set.
1303  */
1304 int
1305 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1306                             uint64_t item_flags,
1307                             uint8_t target_protocol,
1308                             struct rte_flow_error *error)
1309 {
1310         const struct rte_flow_item_udp *mask = item->mask;
1311         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1312         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1313                                       MLX5_FLOW_LAYER_OUTER_L3;
1314         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1315                                       MLX5_FLOW_LAYER_OUTER_L4;
1316         int ret;
1317
1318         if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1319                 return rte_flow_error_set(error, EINVAL,
1320                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1321                                           "protocol filtering not compatible"
1322                                           " with UDP layer");
1323         if (!(item_flags & l3m))
1324                 return rte_flow_error_set(error, EINVAL,
1325                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1326                                           "L3 is mandatory to filter on L4");
1327         if (item_flags & l4m)
1328                 return rte_flow_error_set(error, EINVAL,
1329                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1330                                           "multiple L4 layers not supported");
1331         if (!mask)
1332                 mask = &rte_flow_item_udp_mask;
1333         ret = mlx5_flow_item_acceptable
1334                 (item, (const uint8_t *)mask,
1335                  (const uint8_t *)&rte_flow_item_udp_mask,
1336                  sizeof(struct rte_flow_item_udp), error);
1337         if (ret < 0)
1338                 return ret;
1339         return 0;
1340 }
1341
1342 /**
1343  * Validate TCP item.
1344  *
1345  * @param[in] item
1346  *   Item specification.
1347  * @param[in] item_flags
1348  *   Bit-fields that holds the items detected until now.
1349  * @param[in] target_protocol
1350  *   The next protocol in the previous item.
1351  * @param[out] error
1352  *   Pointer to error structure.
1353  *
1354  * @return
1355  *   0 on success, a negative errno value otherwise and rte_errno is set.
1356  */
1357 int
1358 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1359                             uint64_t item_flags,
1360                             uint8_t target_protocol,
1361                             const struct rte_flow_item_tcp *flow_mask,
1362                             struct rte_flow_error *error)
1363 {
1364         const struct rte_flow_item_tcp *mask = item->mask;
1365         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1366         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1367                                       MLX5_FLOW_LAYER_OUTER_L3;
1368         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1369                                       MLX5_FLOW_LAYER_OUTER_L4;
1370         int ret;
1371
1372         assert(flow_mask);
1373         if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1374                 return rte_flow_error_set(error, EINVAL,
1375                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1376                                           "protocol filtering not compatible"
1377                                           " with TCP layer");
1378         if (!(item_flags & l3m))
1379                 return rte_flow_error_set(error, EINVAL,
1380                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1381                                           "L3 is mandatory to filter on L4");
1382         if (item_flags & l4m)
1383                 return rte_flow_error_set(error, EINVAL,
1384                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1385                                           "multiple L4 layers not supported");
1386         if (!mask)
1387                 mask = &rte_flow_item_tcp_mask;
1388         ret = mlx5_flow_item_acceptable
1389                 (item, (const uint8_t *)mask,
1390                  (const uint8_t *)flow_mask,
1391                  sizeof(struct rte_flow_item_tcp), error);
1392         if (ret < 0)
1393                 return ret;
1394         return 0;
1395 }
1396
1397 /**
1398  * Validate VXLAN item.
1399  *
1400  * @param[in] item
1401  *   Item specification.
1402  * @param[in] item_flags
1403  *   Bit-fields that holds the items detected until now.
1404  * @param[in] target_protocol
1405  *   The next protocol in the previous item.
1406  * @param[out] error
1407  *   Pointer to error structure.
1408  *
1409  * @return
1410  *   0 on success, a negative errno value otherwise and rte_errno is set.
1411  */
1412 int
1413 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1414                               uint64_t item_flags,
1415                               struct rte_flow_error *error)
1416 {
1417         const struct rte_flow_item_vxlan *spec = item->spec;
1418         const struct rte_flow_item_vxlan *mask = item->mask;
1419         int ret;
1420         union vni {
1421                 uint32_t vlan_id;
1422                 uint8_t vni[4];
1423         } id = { .vlan_id = 0, };
1424         uint32_t vlan_id = 0;
1425
1426
1427         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1428                 return rte_flow_error_set(error, ENOTSUP,
1429                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1430                                           "multiple tunnel layers not"
1431                                           " supported");
1432         /*
1433          * Verify only UDPv4 is present as defined in
1434          * https://tools.ietf.org/html/rfc7348
1435          */
1436         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1437                 return rte_flow_error_set(error, EINVAL,
1438                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1439                                           "no outer UDP layer found");
1440         if (!mask)
1441                 mask = &rte_flow_item_vxlan_mask;
1442         ret = mlx5_flow_item_acceptable
1443                 (item, (const uint8_t *)mask,
1444                  (const uint8_t *)&rte_flow_item_vxlan_mask,
1445                  sizeof(struct rte_flow_item_vxlan),
1446                  error);
1447         if (ret < 0)
1448                 return ret;
1449         if (spec) {
1450                 memcpy(&id.vni[1], spec->vni, 3);
1451                 vlan_id = id.vlan_id;
1452                 memcpy(&id.vni[1], mask->vni, 3);
1453                 vlan_id &= id.vlan_id;
1454         }
1455         /*
1456          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1457          * only this layer is defined in the Verbs specification it is
1458          * interpreted as wildcard and all packets will match this
1459          * rule, if it follows a full stack layer (ex: eth / ipv4 /
1460          * udp), all packets matching the layers before will also
1461          * match this rule.  To avoid such situation, VNI 0 is
1462          * currently refused.
1463          */
1464         if (!vlan_id)
1465                 return rte_flow_error_set(error, ENOTSUP,
1466                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1467                                           "VXLAN vni cannot be 0");
1468         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1469                 return rte_flow_error_set(error, ENOTSUP,
1470                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1471                                           "VXLAN tunnel must be fully defined");
1472         return 0;
1473 }
1474
1475 /**
1476  * Validate VXLAN_GPE item.
1477  *
1478  * @param[in] item
1479  *   Item specification.
1480  * @param[in] item_flags
1481  *   Bit-fields that holds the items detected until now.
1482  * @param[in] priv
1483  *   Pointer to the private data structure.
1484  * @param[in] target_protocol
1485  *   The next protocol in the previous item.
1486  * @param[out] error
1487  *   Pointer to error structure.
1488  *
1489  * @return
1490  *   0 on success, a negative errno value otherwise and rte_errno is set.
1491  */
1492 int
1493 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1494                                   uint64_t item_flags,
1495                                   struct rte_eth_dev *dev,
1496                                   struct rte_flow_error *error)
1497 {
1498         struct mlx5_priv *priv = dev->data->dev_private;
1499         const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1500         const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1501         int ret;
1502         union vni {
1503                 uint32_t vlan_id;
1504                 uint8_t vni[4];
1505         } id = { .vlan_id = 0, };
1506         uint32_t vlan_id = 0;
1507
1508         if (!priv->config.l3_vxlan_en)
1509                 return rte_flow_error_set(error, ENOTSUP,
1510                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1511                                           "L3 VXLAN is not enabled by device"
1512                                           " parameter and/or not configured in"
1513                                           " firmware");
1514         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1515                 return rte_flow_error_set(error, ENOTSUP,
1516                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1517                                           "multiple tunnel layers not"
1518                                           " supported");
1519         /*
1520          * Verify only UDPv4 is present as defined in
1521          * https://tools.ietf.org/html/rfc7348
1522          */
1523         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1524                 return rte_flow_error_set(error, EINVAL,
1525                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1526                                           "no outer UDP layer found");
1527         if (!mask)
1528                 mask = &rte_flow_item_vxlan_gpe_mask;
1529         ret = mlx5_flow_item_acceptable
1530                 (item, (const uint8_t *)mask,
1531                  (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1532                  sizeof(struct rte_flow_item_vxlan_gpe),
1533                  error);
1534         if (ret < 0)
1535                 return ret;
1536         if (spec) {
1537                 if (spec->protocol)
1538                         return rte_flow_error_set(error, ENOTSUP,
1539                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1540                                                   item,
1541                                                   "VxLAN-GPE protocol"
1542                                                   " not supported");
1543                 memcpy(&id.vni[1], spec->vni, 3);
1544                 vlan_id = id.vlan_id;
1545                 memcpy(&id.vni[1], mask->vni, 3);
1546                 vlan_id &= id.vlan_id;
1547         }
1548         /*
1549          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1550          * layer is defined in the Verbs specification it is interpreted as
1551          * wildcard and all packets will match this rule, if it follows a full
1552          * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1553          * before will also match this rule.  To avoid such situation, VNI 0
1554          * is currently refused.
1555          */
1556         if (!vlan_id)
1557                 return rte_flow_error_set(error, ENOTSUP,
1558                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1559                                           "VXLAN-GPE vni cannot be 0");
1560         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1561                 return rte_flow_error_set(error, ENOTSUP,
1562                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1563                                           "VXLAN-GPE tunnel must be fully"
1564                                           " defined");
1565         return 0;
1566 }
1567
1568 /**
1569  * Validate GRE item.
1570  *
1571  * @param[in] item
1572  *   Item specification.
1573  * @param[in] item_flags
1574  *   Bit flags to mark detected items.
1575  * @param[in] target_protocol
1576  *   The next protocol in the previous item.
1577  * @param[out] error
1578  *   Pointer to error structure.
1579  *
1580  * @return
1581  *   0 on success, a negative errno value otherwise and rte_errno is set.
1582  */
1583 int
1584 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1585                             uint64_t item_flags,
1586                             uint8_t target_protocol,
1587                             struct rte_flow_error *error)
1588 {
1589         const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1590         const struct rte_flow_item_gre *mask = item->mask;
1591         int ret;
1592
1593         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1594                 return rte_flow_error_set(error, EINVAL,
1595                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1596                                           "protocol filtering not compatible"
1597                                           " with this GRE layer");
1598         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1599                 return rte_flow_error_set(error, ENOTSUP,
1600                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1601                                           "multiple tunnel layers not"
1602                                           " supported");
1603         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1604                 return rte_flow_error_set(error, ENOTSUP,
1605                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1606                                           "L3 Layer is missing");
1607         if (!mask)
1608                 mask = &rte_flow_item_gre_mask;
1609         ret = mlx5_flow_item_acceptable
1610                 (item, (const uint8_t *)mask,
1611                  (const uint8_t *)&rte_flow_item_gre_mask,
1612                  sizeof(struct rte_flow_item_gre), error);
1613         if (ret < 0)
1614                 return ret;
1615 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1616         if (spec && (spec->protocol & mask->protocol))
1617                 return rte_flow_error_set(error, ENOTSUP,
1618                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1619                                           "without MPLS support the"
1620                                           " specification cannot be used for"
1621                                           " filtering");
1622 #endif
1623         return 0;
1624 }
1625
1626 /**
1627  * Validate MPLS item.
1628  *
1629  * @param[in] dev
1630  *   Pointer to the rte_eth_dev structure.
1631  * @param[in] item
1632  *   Item specification.
1633  * @param[in] item_flags
1634  *   Bit-fields that holds the items detected until now.
1635  * @param[in] prev_layer
1636  *   The protocol layer indicated in previous item.
1637  * @param[out] error
1638  *   Pointer to error structure.
1639  *
1640  * @return
1641  *   0 on success, a negative errno value otherwise and rte_errno is set.
1642  */
1643 int
1644 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
1645                              const struct rte_flow_item *item __rte_unused,
1646                              uint64_t item_flags __rte_unused,
1647                              uint64_t prev_layer __rte_unused,
1648                              struct rte_flow_error *error)
1649 {
1650 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1651         const struct rte_flow_item_mpls *mask = item->mask;
1652         struct mlx5_priv *priv = dev->data->dev_private;
1653         int ret;
1654
1655         if (!priv->config.mpls_en)
1656                 return rte_flow_error_set(error, ENOTSUP,
1657                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1658                                           "MPLS not supported or"
1659                                           " disabled in firmware"
1660                                           " configuration.");
1661         /* MPLS over IP, UDP, GRE is allowed */
1662         if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
1663                             MLX5_FLOW_LAYER_OUTER_L4_UDP |
1664                             MLX5_FLOW_LAYER_GRE)))
1665                 return rte_flow_error_set(error, EINVAL,
1666                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1667                                           "protocol filtering not compatible"
1668                                           " with MPLS layer");
1669         /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
1670         if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1671             !(item_flags & MLX5_FLOW_LAYER_GRE))
1672                 return rte_flow_error_set(error, ENOTSUP,
1673                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1674                                           "multiple tunnel layers not"
1675                                           " supported");
1676         if (!mask)
1677                 mask = &rte_flow_item_mpls_mask;
1678         ret = mlx5_flow_item_acceptable
1679                 (item, (const uint8_t *)mask,
1680                  (const uint8_t *)&rte_flow_item_mpls_mask,
1681                  sizeof(struct rte_flow_item_mpls), error);
1682         if (ret < 0)
1683                 return ret;
1684         return 0;
1685 #endif
1686         return rte_flow_error_set(error, ENOTSUP,
1687                                   RTE_FLOW_ERROR_TYPE_ITEM, item,
1688                                   "MPLS is not supported by Verbs, please"
1689                                   " update.");
1690 }
1691
1692 static int
1693 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1694                    const struct rte_flow_attr *attr __rte_unused,
1695                    const struct rte_flow_item items[] __rte_unused,
1696                    const struct rte_flow_action actions[] __rte_unused,
1697                    struct rte_flow_error *error __rte_unused)
1698 {
1699         rte_errno = ENOTSUP;
1700         return -rte_errno;
1701 }
1702
1703 static struct mlx5_flow *
1704 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1705                   const struct rte_flow_item items[] __rte_unused,
1706                   const struct rte_flow_action actions[] __rte_unused,
1707                   struct rte_flow_error *error __rte_unused)
1708 {
1709         rte_errno = ENOTSUP;
1710         return NULL;
1711 }
1712
1713 static int
1714 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1715                     struct mlx5_flow *dev_flow __rte_unused,
1716                     const struct rte_flow_attr *attr __rte_unused,
1717                     const struct rte_flow_item items[] __rte_unused,
1718                     const struct rte_flow_action actions[] __rte_unused,
1719                     struct rte_flow_error *error __rte_unused)
1720 {
1721         rte_errno = ENOTSUP;
1722         return -rte_errno;
1723 }
1724
1725 static int
1726 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1727                 struct rte_flow *flow __rte_unused,
1728                 struct rte_flow_error *error __rte_unused)
1729 {
1730         rte_errno = ENOTSUP;
1731         return -rte_errno;
1732 }
1733
1734 static void
1735 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1736                  struct rte_flow *flow __rte_unused)
1737 {
1738 }
1739
1740 static void
1741 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1742                   struct rte_flow *flow __rte_unused)
1743 {
1744 }
1745
1746 static int
1747 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1748                 struct rte_flow *flow __rte_unused,
1749                 const struct rte_flow_action *actions __rte_unused,
1750                 void *data __rte_unused,
1751                 struct rte_flow_error *error __rte_unused)
1752 {
1753         rte_errno = ENOTSUP;
1754         return -rte_errno;
1755 }
1756
1757 /* Void driver to protect from null pointer reference. */
1758 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1759         .validate = flow_null_validate,
1760         .prepare = flow_null_prepare,
1761         .translate = flow_null_translate,
1762         .apply = flow_null_apply,
1763         .remove = flow_null_remove,
1764         .destroy = flow_null_destroy,
1765         .query = flow_null_query,
1766 };
1767
1768 /**
1769  * Select flow driver type according to flow attributes and device
1770  * configuration.
1771  *
1772  * @param[in] dev
1773  *   Pointer to the dev structure.
1774  * @param[in] attr
1775  *   Pointer to the flow attributes.
1776  *
1777  * @return
1778  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1779  */
1780 static enum mlx5_flow_drv_type
1781 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1782 {
1783         struct mlx5_priv *priv = dev->data->dev_private;
1784         enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1785
1786         if (attr->transfer)
1787                 type = MLX5_FLOW_TYPE_TCF;
1788         else
1789                 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1790                                                  MLX5_FLOW_TYPE_VERBS;
1791         return type;
1792 }
1793
1794 #define flow_get_drv_ops(type) flow_drv_ops[type]
1795
1796 /**
1797  * Flow driver validation API. This abstracts calling driver specific functions.
1798  * The type of flow driver is determined according to flow attributes.
1799  *
1800  * @param[in] dev
1801  *   Pointer to the dev structure.
1802  * @param[in] attr
1803  *   Pointer to the flow attributes.
1804  * @param[in] items
1805  *   Pointer to the list of items.
1806  * @param[in] actions
1807  *   Pointer to the list of actions.
1808  * @param[out] error
1809  *   Pointer to the error structure.
1810  *
1811  * @return
1812  *   0 on success, a negative errno value otherwise and rte_errno is set.
1813  */
1814 static inline int
1815 flow_drv_validate(struct rte_eth_dev *dev,
1816                   const struct rte_flow_attr *attr,
1817                   const struct rte_flow_item items[],
1818                   const struct rte_flow_action actions[],
1819                   struct rte_flow_error *error)
1820 {
1821         const struct mlx5_flow_driver_ops *fops;
1822         enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1823
1824         fops = flow_get_drv_ops(type);
1825         return fops->validate(dev, attr, items, actions, error);
1826 }
1827
1828 /**
1829  * Flow driver preparation API. This abstracts calling driver specific
1830  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1831  * calculates the size of memory required for device flow, allocates the memory,
1832  * initializes the device flow and returns the pointer.
1833  *
1834  * @note
1835  *   This function initializes device flow structure such as dv, tcf or verbs in
1836  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
1837  *   rest. For example, adding returning device flow to flow->dev_flow list and
1838  *   setting backward reference to the flow should be done out of this function.
1839  *   layers field is not filled either.
1840  *
1841  * @param[in] attr
1842  *   Pointer to the flow attributes.
1843  * @param[in] items
1844  *   Pointer to the list of items.
1845  * @param[in] actions
1846  *   Pointer to the list of actions.
1847  * @param[out] error
1848  *   Pointer to the error structure.
1849  *
1850  * @return
1851  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
1852  */
1853 static inline struct mlx5_flow *
1854 flow_drv_prepare(const struct rte_flow *flow,
1855                  const struct rte_flow_attr *attr,
1856                  const struct rte_flow_item items[],
1857                  const struct rte_flow_action actions[],
1858                  struct rte_flow_error *error)
1859 {
1860         const struct mlx5_flow_driver_ops *fops;
1861         enum mlx5_flow_drv_type type = flow->drv_type;
1862
1863         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1864         fops = flow_get_drv_ops(type);
1865         return fops->prepare(attr, items, actions, error);
1866 }
1867
1868 /**
1869  * Flow driver translation API. This abstracts calling driver specific
1870  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1871  * translates a generic flow into a driver flow. flow_drv_prepare() must
1872  * precede.
1873  *
1874  * @note
1875  *   dev_flow->layers could be filled as a result of parsing during translation
1876  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
1877  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
1878  *   flow->actions could be overwritten even though all the expanded dev_flows
1879  *   have the same actions.
1880  *
1881  * @param[in] dev
1882  *   Pointer to the rte dev structure.
1883  * @param[in, out] dev_flow
1884  *   Pointer to the mlx5 flow.
1885  * @param[in] attr
1886  *   Pointer to the flow attributes.
1887  * @param[in] items
1888  *   Pointer to the list of items.
1889  * @param[in] actions
1890  *   Pointer to the list of actions.
1891  * @param[out] error
1892  *   Pointer to the error structure.
1893  *
1894  * @return
1895  *   0 on success, a negative errno value otherwise and rte_errno is set.
1896  */
1897 static inline int
1898 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1899                    const struct rte_flow_attr *attr,
1900                    const struct rte_flow_item items[],
1901                    const struct rte_flow_action actions[],
1902                    struct rte_flow_error *error)
1903 {
1904         const struct mlx5_flow_driver_ops *fops;
1905         enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1906
1907         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1908         fops = flow_get_drv_ops(type);
1909         return fops->translate(dev, dev_flow, attr, items, actions, error);
1910 }
1911
1912 /**
1913  * Flow driver apply API. This abstracts calling driver specific functions.
1914  * Parent flow (rte_flow) should have driver type (drv_type). It applies
1915  * translated driver flows on to device. flow_drv_translate() must precede.
1916  *
1917  * @param[in] dev
1918  *   Pointer to Ethernet device structure.
1919  * @param[in, out] flow
1920  *   Pointer to flow structure.
1921  * @param[out] error
1922  *   Pointer to error structure.
1923  *
1924  * @return
1925  *   0 on success, a negative errno value otherwise and rte_errno is set.
1926  */
1927 static inline int
1928 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1929                struct rte_flow_error *error)
1930 {
1931         const struct mlx5_flow_driver_ops *fops;
1932         enum mlx5_flow_drv_type type = flow->drv_type;
1933
1934         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1935         fops = flow_get_drv_ops(type);
1936         return fops->apply(dev, flow, error);
1937 }
1938
1939 /**
1940  * Flow driver remove API. This abstracts calling driver specific functions.
1941  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1942  * on device. All the resources of the flow should be freed by calling
1943  * flow_drv_destroy().
1944  *
1945  * @param[in] dev
1946  *   Pointer to Ethernet device.
1947  * @param[in, out] flow
1948  *   Pointer to flow structure.
1949  */
1950 static inline void
1951 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1952 {
1953         const struct mlx5_flow_driver_ops *fops;
1954         enum mlx5_flow_drv_type type = flow->drv_type;
1955
1956         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1957         fops = flow_get_drv_ops(type);
1958         fops->remove(dev, flow);
1959 }
1960
1961 /**
1962  * Flow driver destroy API. This abstracts calling driver specific functions.
1963  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1964  * on device and releases resources of the flow.
1965  *
1966  * @param[in] dev
1967  *   Pointer to Ethernet device.
1968  * @param[in, out] flow
1969  *   Pointer to flow structure.
1970  */
1971 static inline void
1972 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1973 {
1974         const struct mlx5_flow_driver_ops *fops;
1975         enum mlx5_flow_drv_type type = flow->drv_type;
1976
1977         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1978         fops = flow_get_drv_ops(type);
1979         fops->destroy(dev, flow);
1980 }
1981
1982 /**
1983  * Validate a flow supported by the NIC.
1984  *
1985  * @see rte_flow_validate()
1986  * @see rte_flow_ops
1987  */
1988 int
1989 mlx5_flow_validate(struct rte_eth_dev *dev,
1990                    const struct rte_flow_attr *attr,
1991                    const struct rte_flow_item items[],
1992                    const struct rte_flow_action actions[],
1993                    struct rte_flow_error *error)
1994 {
1995         int ret;
1996
1997         ret = flow_drv_validate(dev, attr, items, actions, error);
1998         if (ret < 0)
1999                 return ret;
2000         return 0;
2001 }
2002
2003 /**
2004  * Get RSS action from the action list.
2005  *
2006  * @param[in] actions
2007  *   Pointer to the list of actions.
2008  *
2009  * @return
2010  *   Pointer to the RSS action if exist, else return NULL.
2011  */
2012 static const struct rte_flow_action_rss*
2013 flow_get_rss_action(const struct rte_flow_action actions[])
2014 {
2015         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2016                 switch (actions->type) {
2017                 case RTE_FLOW_ACTION_TYPE_RSS:
2018                         return (const struct rte_flow_action_rss *)
2019                                actions->conf;
2020                 default:
2021                         break;
2022                 }
2023         }
2024         return NULL;
2025 }
2026
2027 static unsigned int
2028 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2029 {
2030         const struct rte_flow_item *item;
2031         unsigned int has_vlan = 0;
2032
2033         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2034                 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2035                         has_vlan = 1;
2036                         break;
2037                 }
2038         }
2039         if (has_vlan)
2040                 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2041                                        MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2042         return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2043                                MLX5_EXPANSION_ROOT_OUTER;
2044 }
2045
2046 /**
2047  * Create a flow and add it to @p list.
2048  *
2049  * @param dev
2050  *   Pointer to Ethernet device.
2051  * @param list
2052  *   Pointer to a TAILQ flow list.
2053  * @param[in] attr
2054  *   Flow rule attributes.
2055  * @param[in] items
2056  *   Pattern specification (list terminated by the END pattern item).
2057  * @param[in] actions
2058  *   Associated actions (list terminated by the END action).
2059  * @param[out] error
2060  *   Perform verbose error reporting if not NULL.
2061  *
2062  * @return
2063  *   A flow on success, NULL otherwise and rte_errno is set.
2064  */
2065 static struct rte_flow *
2066 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2067                  const struct rte_flow_attr *attr,
2068                  const struct rte_flow_item items[],
2069                  const struct rte_flow_action actions[],
2070                  struct rte_flow_error *error)
2071 {
2072         struct rte_flow *flow = NULL;
2073         struct mlx5_flow *dev_flow;
2074         const struct rte_flow_action_rss *rss;
2075         union {
2076                 struct rte_flow_expand_rss buf;
2077                 uint8_t buffer[2048];
2078         } expand_buffer;
2079         struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2080         int ret;
2081         uint32_t i;
2082         uint32_t flow_size;
2083
2084         ret = flow_drv_validate(dev, attr, items, actions, error);
2085         if (ret < 0)
2086                 return NULL;
2087         flow_size = sizeof(struct rte_flow);
2088         rss = flow_get_rss_action(actions);
2089         if (rss)
2090                 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2091                                             sizeof(void *));
2092         else
2093                 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2094         flow = rte_calloc(__func__, 1, flow_size, 0);
2095         flow->drv_type = flow_get_drv_type(dev, attr);
2096         flow->ingress = attr->ingress;
2097         assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2098                flow->drv_type < MLX5_FLOW_TYPE_MAX);
2099         flow->queue = (void *)(flow + 1);
2100         LIST_INIT(&flow->dev_flows);
2101         if (rss && rss->types) {
2102                 unsigned int graph_root;
2103
2104                 graph_root = find_graph_root(items, rss->level);
2105                 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2106                                           items, rss->types,
2107                                           mlx5_support_expansion,
2108                                           graph_root);
2109                 assert(ret > 0 &&
2110                        (unsigned int)ret < sizeof(expand_buffer.buffer));
2111         } else {
2112                 buf->entries = 1;
2113                 buf->entry[0].pattern = (void *)(uintptr_t)items;
2114         }
2115         for (i = 0; i < buf->entries; ++i) {
2116                 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2117                                             actions, error);
2118                 if (!dev_flow)
2119                         goto error;
2120                 dev_flow->flow = flow;
2121                 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2122                 ret = flow_drv_translate(dev, dev_flow, attr,
2123                                          buf->entry[i].pattern,
2124                                          actions, error);
2125                 if (ret < 0)
2126                         goto error;
2127         }
2128         if (dev->data->dev_started) {
2129                 ret = flow_drv_apply(dev, flow, error);
2130                 if (ret < 0)
2131                         goto error;
2132         }
2133         TAILQ_INSERT_TAIL(list, flow, next);
2134         flow_rxq_flags_set(dev, flow);
2135         return flow;
2136 error:
2137         ret = rte_errno; /* Save rte_errno before cleanup. */
2138         assert(flow);
2139         flow_drv_destroy(dev, flow);
2140         rte_free(flow);
2141         rte_errno = ret; /* Restore rte_errno. */
2142         return NULL;
2143 }
2144
2145 /**
2146  * Create a flow.
2147  *
2148  * @see rte_flow_create()
2149  * @see rte_flow_ops
2150  */
2151 struct rte_flow *
2152 mlx5_flow_create(struct rte_eth_dev *dev,
2153                  const struct rte_flow_attr *attr,
2154                  const struct rte_flow_item items[],
2155                  const struct rte_flow_action actions[],
2156                  struct rte_flow_error *error)
2157 {
2158         struct mlx5_priv *priv = (struct mlx5_priv *)dev->data->dev_private;
2159
2160         return flow_list_create(dev, &priv->flows,
2161                                 attr, items, actions, error);
2162 }
2163
2164 /**
2165  * Destroy a flow in a list.
2166  *
2167  * @param dev
2168  *   Pointer to Ethernet device.
2169  * @param list
2170  *   Pointer to a TAILQ flow list.
2171  * @param[in] flow
2172  *   Flow to destroy.
2173  */
2174 static void
2175 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2176                   struct rte_flow *flow)
2177 {
2178         /*
2179          * Update RX queue flags only if port is started, otherwise it is
2180          * already clean.
2181          */
2182         if (dev->data->dev_started)
2183                 flow_rxq_flags_trim(dev, flow);
2184         flow_drv_destroy(dev, flow);
2185         TAILQ_REMOVE(list, flow, next);
2186         rte_free(flow->fdir);
2187         rte_free(flow);
2188 }
2189
2190 /**
2191  * Destroy all flows.
2192  *
2193  * @param dev
2194  *   Pointer to Ethernet device.
2195  * @param list
2196  *   Pointer to a TAILQ flow list.
2197  */
2198 void
2199 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2200 {
2201         while (!TAILQ_EMPTY(list)) {
2202                 struct rte_flow *flow;
2203
2204                 flow = TAILQ_FIRST(list);
2205                 flow_list_destroy(dev, list, flow);
2206         }
2207 }
2208
2209 /**
2210  * Remove all flows.
2211  *
2212  * @param dev
2213  *   Pointer to Ethernet device.
2214  * @param list
2215  *   Pointer to a TAILQ flow list.
2216  */
2217 void
2218 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2219 {
2220         struct rte_flow *flow;
2221
2222         TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2223                 flow_drv_remove(dev, flow);
2224         flow_rxq_flags_clear(dev);
2225 }
2226
2227 /**
2228  * Add all flows.
2229  *
2230  * @param dev
2231  *   Pointer to Ethernet device.
2232  * @param list
2233  *   Pointer to a TAILQ flow list.
2234  *
2235  * @return
2236  *   0 on success, a negative errno value otherwise and rte_errno is set.
2237  */
2238 int
2239 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2240 {
2241         struct rte_flow *flow;
2242         struct rte_flow_error error;
2243         int ret = 0;
2244
2245         TAILQ_FOREACH(flow, list, next) {
2246                 ret = flow_drv_apply(dev, flow, &error);
2247                 if (ret < 0)
2248                         goto error;
2249                 flow_rxq_flags_set(dev, flow);
2250         }
2251         return 0;
2252 error:
2253         ret = rte_errno; /* Save rte_errno before cleanup. */
2254         mlx5_flow_stop(dev, list);
2255         rte_errno = ret; /* Restore rte_errno. */
2256         return -rte_errno;
2257 }
2258
2259 /**
2260  * Verify the flow list is empty
2261  *
2262  * @param dev
2263  *  Pointer to Ethernet device.
2264  *
2265  * @return the number of flows not released.
2266  */
2267 int
2268 mlx5_flow_verify(struct rte_eth_dev *dev)
2269 {
2270         struct mlx5_priv *priv = dev->data->dev_private;
2271         struct rte_flow *flow;
2272         int ret = 0;
2273
2274         TAILQ_FOREACH(flow, &priv->flows, next) {
2275                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
2276                         dev->data->port_id, (void *)flow);
2277                 ++ret;
2278         }
2279         return ret;
2280 }
2281
2282 /**
2283  * Enable a control flow configured from the control plane.
2284  *
2285  * @param dev
2286  *   Pointer to Ethernet device.
2287  * @param eth_spec
2288  *   An Ethernet flow spec to apply.
2289  * @param eth_mask
2290  *   An Ethernet flow mask to apply.
2291  * @param vlan_spec
2292  *   A VLAN flow spec to apply.
2293  * @param vlan_mask
2294  *   A VLAN flow mask to apply.
2295  *
2296  * @return
2297  *   0 on success, a negative errno value otherwise and rte_errno is set.
2298  */
2299 int
2300 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2301                     struct rte_flow_item_eth *eth_spec,
2302                     struct rte_flow_item_eth *eth_mask,
2303                     struct rte_flow_item_vlan *vlan_spec,
2304                     struct rte_flow_item_vlan *vlan_mask)
2305 {
2306         struct mlx5_priv *priv = dev->data->dev_private;
2307         const struct rte_flow_attr attr = {
2308                 .ingress = 1,
2309                 .priority = MLX5_FLOW_PRIO_RSVD,
2310         };
2311         struct rte_flow_item items[] = {
2312                 {
2313                         .type = RTE_FLOW_ITEM_TYPE_ETH,
2314                         .spec = eth_spec,
2315                         .last = NULL,
2316                         .mask = eth_mask,
2317                 },
2318                 {
2319                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2320                                               RTE_FLOW_ITEM_TYPE_END,
2321                         .spec = vlan_spec,
2322                         .last = NULL,
2323                         .mask = vlan_mask,
2324                 },
2325                 {
2326                         .type = RTE_FLOW_ITEM_TYPE_END,
2327                 },
2328         };
2329         uint16_t queue[priv->reta_idx_n];
2330         struct rte_flow_action_rss action_rss = {
2331                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2332                 .level = 0,
2333                 .types = priv->rss_conf.rss_hf,
2334                 .key_len = priv->rss_conf.rss_key_len,
2335                 .queue_num = priv->reta_idx_n,
2336                 .key = priv->rss_conf.rss_key,
2337                 .queue = queue,
2338         };
2339         struct rte_flow_action actions[] = {
2340                 {
2341                         .type = RTE_FLOW_ACTION_TYPE_RSS,
2342                         .conf = &action_rss,
2343                 },
2344                 {
2345                         .type = RTE_FLOW_ACTION_TYPE_END,
2346                 },
2347         };
2348         struct rte_flow *flow;
2349         struct rte_flow_error error;
2350         unsigned int i;
2351
2352         if (!priv->reta_idx_n || !priv->rxqs_n) {
2353                 return 0;
2354         }
2355         for (i = 0; i != priv->reta_idx_n; ++i)
2356                 queue[i] = (*priv->reta_idx)[i];
2357         flow = flow_list_create(dev, &priv->ctrl_flows,
2358                                 &attr, items, actions, &error);
2359         if (!flow)
2360                 return -rte_errno;
2361         return 0;
2362 }
2363
2364 /**
2365  * Enable a flow control configured from the control plane.
2366  *
2367  * @param dev
2368  *   Pointer to Ethernet device.
2369  * @param eth_spec
2370  *   An Ethernet flow spec to apply.
2371  * @param eth_mask
2372  *   An Ethernet flow mask to apply.
2373  *
2374  * @return
2375  *   0 on success, a negative errno value otherwise and rte_errno is set.
2376  */
2377 int
2378 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2379                struct rte_flow_item_eth *eth_spec,
2380                struct rte_flow_item_eth *eth_mask)
2381 {
2382         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2383 }
2384
2385 /**
2386  * Destroy a flow.
2387  *
2388  * @see rte_flow_destroy()
2389  * @see rte_flow_ops
2390  */
2391 int
2392 mlx5_flow_destroy(struct rte_eth_dev *dev,
2393                   struct rte_flow *flow,
2394                   struct rte_flow_error *error __rte_unused)
2395 {
2396         struct mlx5_priv *priv = dev->data->dev_private;
2397
2398         flow_list_destroy(dev, &priv->flows, flow);
2399         return 0;
2400 }
2401
2402 /**
2403  * Destroy all flows.
2404  *
2405  * @see rte_flow_flush()
2406  * @see rte_flow_ops
2407  */
2408 int
2409 mlx5_flow_flush(struct rte_eth_dev *dev,
2410                 struct rte_flow_error *error __rte_unused)
2411 {
2412         struct mlx5_priv *priv = dev->data->dev_private;
2413
2414         mlx5_flow_list_flush(dev, &priv->flows);
2415         return 0;
2416 }
2417
2418 /**
2419  * Isolated mode.
2420  *
2421  * @see rte_flow_isolate()
2422  * @see rte_flow_ops
2423  */
2424 int
2425 mlx5_flow_isolate(struct rte_eth_dev *dev,
2426                   int enable,
2427                   struct rte_flow_error *error)
2428 {
2429         struct mlx5_priv *priv = dev->data->dev_private;
2430
2431         if (dev->data->dev_started) {
2432                 rte_flow_error_set(error, EBUSY,
2433                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2434                                    NULL,
2435                                    "port must be stopped first");
2436                 return -rte_errno;
2437         }
2438         priv->isolated = !!enable;
2439         if (enable)
2440                 dev->dev_ops = &mlx5_dev_ops_isolate;
2441         else
2442                 dev->dev_ops = &mlx5_dev_ops;
2443         return 0;
2444 }
2445
2446 /**
2447  * Query a flow.
2448  *
2449  * @see rte_flow_query()
2450  * @see rte_flow_ops
2451  */
2452 static int
2453 flow_drv_query(struct rte_eth_dev *dev,
2454                struct rte_flow *flow,
2455                const struct rte_flow_action *actions,
2456                void *data,
2457                struct rte_flow_error *error)
2458 {
2459         const struct mlx5_flow_driver_ops *fops;
2460         enum mlx5_flow_drv_type ftype = flow->drv_type;
2461
2462         assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2463         fops = flow_get_drv_ops(ftype);
2464
2465         return fops->query(dev, flow, actions, data, error);
2466 }
2467
2468 /**
2469  * Query a flow.
2470  *
2471  * @see rte_flow_query()
2472  * @see rte_flow_ops
2473  */
2474 int
2475 mlx5_flow_query(struct rte_eth_dev *dev,
2476                 struct rte_flow *flow,
2477                 const struct rte_flow_action *actions,
2478                 void *data,
2479                 struct rte_flow_error *error)
2480 {
2481         int ret;
2482
2483         ret = flow_drv_query(dev, flow, actions, data, error);
2484         if (ret < 0)
2485                 return ret;
2486         return 0;
2487 }
2488
2489 /**
2490  * Convert a flow director filter to a generic flow.
2491  *
2492  * @param dev
2493  *   Pointer to Ethernet device.
2494  * @param fdir_filter
2495  *   Flow director filter to add.
2496  * @param attributes
2497  *   Generic flow parameters structure.
2498  *
2499  * @return
2500  *   0 on success, a negative errno value otherwise and rte_errno is set.
2501  */
2502 static int
2503 flow_fdir_filter_convert(struct rte_eth_dev *dev,
2504                          const struct rte_eth_fdir_filter *fdir_filter,
2505                          struct mlx5_fdir *attributes)
2506 {
2507         struct mlx5_priv *priv = dev->data->dev_private;
2508         const struct rte_eth_fdir_input *input = &fdir_filter->input;
2509         const struct rte_eth_fdir_masks *mask =
2510                 &dev->data->dev_conf.fdir_conf.mask;
2511
2512         /* Validate queue number. */
2513         if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2514                 DRV_LOG(ERR, "port %u invalid queue number %d",
2515                         dev->data->port_id, fdir_filter->action.rx_queue);
2516                 rte_errno = EINVAL;
2517                 return -rte_errno;
2518         }
2519         attributes->attr.ingress = 1;
2520         attributes->items[0] = (struct rte_flow_item) {
2521                 .type = RTE_FLOW_ITEM_TYPE_ETH,
2522                 .spec = &attributes->l2,
2523                 .mask = &attributes->l2_mask,
2524         };
2525         switch (fdir_filter->action.behavior) {
2526         case RTE_ETH_FDIR_ACCEPT:
2527                 attributes->actions[0] = (struct rte_flow_action){
2528                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
2529                         .conf = &attributes->queue,
2530                 };
2531                 break;
2532         case RTE_ETH_FDIR_REJECT:
2533                 attributes->actions[0] = (struct rte_flow_action){
2534                         .type = RTE_FLOW_ACTION_TYPE_DROP,
2535                 };
2536                 break;
2537         default:
2538                 DRV_LOG(ERR, "port %u invalid behavior %d",
2539                         dev->data->port_id,
2540                         fdir_filter->action.behavior);
2541                 rte_errno = ENOTSUP;
2542                 return -rte_errno;
2543         }
2544         attributes->queue.index = fdir_filter->action.rx_queue;
2545         /* Handle L3. */
2546         switch (fdir_filter->input.flow_type) {
2547         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2548         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2549         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2550                 attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2551                         .src_addr = input->flow.ip4_flow.src_ip,
2552                         .dst_addr = input->flow.ip4_flow.dst_ip,
2553                         .time_to_live = input->flow.ip4_flow.ttl,
2554                         .type_of_service = input->flow.ip4_flow.tos,
2555                 };
2556                 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2557                         .src_addr = mask->ipv4_mask.src_ip,
2558                         .dst_addr = mask->ipv4_mask.dst_ip,
2559                         .time_to_live = mask->ipv4_mask.ttl,
2560                         .type_of_service = mask->ipv4_mask.tos,
2561                         .next_proto_id = mask->ipv4_mask.proto,
2562                 };
2563                 attributes->items[1] = (struct rte_flow_item){
2564                         .type = RTE_FLOW_ITEM_TYPE_IPV4,
2565                         .spec = &attributes->l3,
2566                         .mask = &attributes->l3_mask,
2567                 };
2568                 break;
2569         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2570         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2571         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2572                 attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2573                         .hop_limits = input->flow.ipv6_flow.hop_limits,
2574                         .proto = input->flow.ipv6_flow.proto,
2575                 };
2576
2577                 memcpy(attributes->l3.ipv6.hdr.src_addr,
2578                        input->flow.ipv6_flow.src_ip,
2579                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2580                 memcpy(attributes->l3.ipv6.hdr.dst_addr,
2581                        input->flow.ipv6_flow.dst_ip,
2582                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2583                 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2584                        mask->ipv6_mask.src_ip,
2585                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2586                 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2587                        mask->ipv6_mask.dst_ip,
2588                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2589                 attributes->items[1] = (struct rte_flow_item){
2590                         .type = RTE_FLOW_ITEM_TYPE_IPV6,
2591                         .spec = &attributes->l3,
2592                         .mask = &attributes->l3_mask,
2593                 };
2594                 break;
2595         default:
2596                 DRV_LOG(ERR, "port %u invalid flow type%d",
2597                         dev->data->port_id, fdir_filter->input.flow_type);
2598                 rte_errno = ENOTSUP;
2599                 return -rte_errno;
2600         }
2601         /* Handle L4. */
2602         switch (fdir_filter->input.flow_type) {
2603         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2604                 attributes->l4.udp.hdr = (struct udp_hdr){
2605                         .src_port = input->flow.udp4_flow.src_port,
2606                         .dst_port = input->flow.udp4_flow.dst_port,
2607                 };
2608                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2609                         .src_port = mask->src_port_mask,
2610                         .dst_port = mask->dst_port_mask,
2611                 };
2612                 attributes->items[2] = (struct rte_flow_item){
2613                         .type = RTE_FLOW_ITEM_TYPE_UDP,
2614                         .spec = &attributes->l4,
2615                         .mask = &attributes->l4_mask,
2616                 };
2617                 break;
2618         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2619                 attributes->l4.tcp.hdr = (struct tcp_hdr){
2620                         .src_port = input->flow.tcp4_flow.src_port,
2621                         .dst_port = input->flow.tcp4_flow.dst_port,
2622                 };
2623                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2624                         .src_port = mask->src_port_mask,
2625                         .dst_port = mask->dst_port_mask,
2626                 };
2627                 attributes->items[2] = (struct rte_flow_item){
2628                         .type = RTE_FLOW_ITEM_TYPE_TCP,
2629                         .spec = &attributes->l4,
2630                         .mask = &attributes->l4_mask,
2631                 };
2632                 break;
2633         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2634                 attributes->l4.udp.hdr = (struct udp_hdr){
2635                         .src_port = input->flow.udp6_flow.src_port,
2636                         .dst_port = input->flow.udp6_flow.dst_port,
2637                 };
2638                 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2639                         .src_port = mask->src_port_mask,
2640                         .dst_port = mask->dst_port_mask,
2641                 };
2642                 attributes->items[2] = (struct rte_flow_item){
2643                         .type = RTE_FLOW_ITEM_TYPE_UDP,
2644                         .spec = &attributes->l4,
2645                         .mask = &attributes->l4_mask,
2646                 };
2647                 break;
2648         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2649                 attributes->l4.tcp.hdr = (struct tcp_hdr){
2650                         .src_port = input->flow.tcp6_flow.src_port,
2651                         .dst_port = input->flow.tcp6_flow.dst_port,
2652                 };
2653                 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2654                         .src_port = mask->src_port_mask,
2655                         .dst_port = mask->dst_port_mask,
2656                 };
2657                 attributes->items[2] = (struct rte_flow_item){
2658                         .type = RTE_FLOW_ITEM_TYPE_TCP,
2659                         .spec = &attributes->l4,
2660                         .mask = &attributes->l4_mask,
2661                 };
2662                 break;
2663         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2664         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2665                 break;
2666         default:
2667                 DRV_LOG(ERR, "port %u invalid flow type%d",
2668                         dev->data->port_id, fdir_filter->input.flow_type);
2669                 rte_errno = ENOTSUP;
2670                 return -rte_errno;
2671         }
2672         return 0;
2673 }
2674
2675 #define FLOW_FDIR_CMP(f1, f2, fld) \
2676         memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
2677
2678 /**
2679  * Compare two FDIR flows. If items and actions are identical, the two flows are
2680  * regarded as same.
2681  *
2682  * @param dev
2683  *   Pointer to Ethernet device.
2684  * @param f1
2685  *   FDIR flow to compare.
2686  * @param f2
2687  *   FDIR flow to compare.
2688  *
2689  * @return
2690  *   Zero on match, 1 otherwise.
2691  */
2692 static int
2693 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
2694 {
2695         if (FLOW_FDIR_CMP(f1, f2, attr) ||
2696             FLOW_FDIR_CMP(f1, f2, l2) ||
2697             FLOW_FDIR_CMP(f1, f2, l2_mask) ||
2698             FLOW_FDIR_CMP(f1, f2, l3) ||
2699             FLOW_FDIR_CMP(f1, f2, l3_mask) ||
2700             FLOW_FDIR_CMP(f1, f2, l4) ||
2701             FLOW_FDIR_CMP(f1, f2, l4_mask) ||
2702             FLOW_FDIR_CMP(f1, f2, actions[0].type))
2703                 return 1;
2704         if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
2705             FLOW_FDIR_CMP(f1, f2, queue))
2706                 return 1;
2707         return 0;
2708 }
2709
2710 /**
2711  * Search device flow list to find out a matched FDIR flow.
2712  *
2713  * @param dev
2714  *   Pointer to Ethernet device.
2715  * @param fdir_flow
2716  *   FDIR flow to lookup.
2717  *
2718  * @return
2719  *   Pointer of flow if found, NULL otherwise.
2720  */
2721 static struct rte_flow *
2722 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
2723 {
2724         struct mlx5_priv *priv = dev->data->dev_private;
2725         struct rte_flow *flow = NULL;
2726
2727         assert(fdir_flow);
2728         TAILQ_FOREACH(flow, &priv->flows, next) {
2729                 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
2730                         DRV_LOG(DEBUG, "port %u found FDIR flow %p",
2731                                 dev->data->port_id, (void *)flow);
2732                         break;
2733                 }
2734         }
2735         return flow;
2736 }
2737
2738 /**
2739  * Add new flow director filter and store it in list.
2740  *
2741  * @param dev
2742  *   Pointer to Ethernet device.
2743  * @param fdir_filter
2744  *   Flow director filter to add.
2745  *
2746  * @return
2747  *   0 on success, a negative errno value otherwise and rte_errno is set.
2748  */
2749 static int
2750 flow_fdir_filter_add(struct rte_eth_dev *dev,
2751                      const struct rte_eth_fdir_filter *fdir_filter)
2752 {
2753         struct mlx5_priv *priv = dev->data->dev_private;
2754         struct mlx5_fdir *fdir_flow;
2755         struct rte_flow *flow;
2756         int ret;
2757
2758         fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
2759         if (!fdir_flow) {
2760                 rte_errno = ENOMEM;
2761                 return -rte_errno;
2762         }
2763         ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
2764         if (ret)
2765                 goto error;
2766         flow = flow_fdir_filter_lookup(dev, fdir_flow);
2767         if (flow) {
2768                 rte_errno = EEXIST;
2769                 goto error;
2770         }
2771         flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
2772                                 fdir_flow->items, fdir_flow->actions, NULL);
2773         if (!flow)
2774                 goto error;
2775         assert(!flow->fdir);
2776         flow->fdir = fdir_flow;
2777         DRV_LOG(DEBUG, "port %u created FDIR flow %p",
2778                 dev->data->port_id, (void *)flow);
2779         return 0;
2780 error:
2781         rte_free(fdir_flow);
2782         return -rte_errno;
2783 }
2784
2785 /**
2786  * Delete specific filter.
2787  *
2788  * @param dev
2789  *   Pointer to Ethernet device.
2790  * @param fdir_filter
2791  *   Filter to be deleted.
2792  *
2793  * @return
2794  *   0 on success, a negative errno value otherwise and rte_errno is set.
2795  */
2796 static int
2797 flow_fdir_filter_delete(struct rte_eth_dev *dev,
2798                         const struct rte_eth_fdir_filter *fdir_filter)
2799 {
2800         struct mlx5_priv *priv = dev->data->dev_private;
2801         struct rte_flow *flow;
2802         struct mlx5_fdir fdir_flow = {
2803                 .attr.group = 0,
2804         };
2805         int ret;
2806
2807         ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
2808         if (ret)
2809                 return -rte_errno;
2810         flow = flow_fdir_filter_lookup(dev, &fdir_flow);
2811         if (!flow) {
2812                 rte_errno = ENOENT;
2813                 return -rte_errno;
2814         }
2815         flow_list_destroy(dev, &priv->flows, flow);
2816         DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
2817                 dev->data->port_id, (void *)flow);
2818         return 0;
2819 }
2820
2821 /**
2822  * Update queue for specific filter.
2823  *
2824  * @param dev
2825  *   Pointer to Ethernet device.
2826  * @param fdir_filter
2827  *   Filter to be updated.
2828  *
2829  * @return
2830  *   0 on success, a negative errno value otherwise and rte_errno is set.
2831  */
2832 static int
2833 flow_fdir_filter_update(struct rte_eth_dev *dev,
2834                         const struct rte_eth_fdir_filter *fdir_filter)
2835 {
2836         int ret;
2837
2838         ret = flow_fdir_filter_delete(dev, fdir_filter);
2839         if (ret)
2840                 return ret;
2841         return flow_fdir_filter_add(dev, fdir_filter);
2842 }
2843
2844 /**
2845  * Flush all filters.
2846  *
2847  * @param dev
2848  *   Pointer to Ethernet device.
2849  */
2850 static void
2851 flow_fdir_filter_flush(struct rte_eth_dev *dev)
2852 {
2853         struct mlx5_priv *priv = dev->data->dev_private;
2854
2855         mlx5_flow_list_flush(dev, &priv->flows);
2856 }
2857
2858 /**
2859  * Get flow director information.
2860  *
2861  * @param dev
2862  *   Pointer to Ethernet device.
2863  * @param[out] fdir_info
2864  *   Resulting flow director information.
2865  */
2866 static void
2867 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2868 {
2869         struct rte_eth_fdir_masks *mask =
2870                 &dev->data->dev_conf.fdir_conf.mask;
2871
2872         fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2873         fdir_info->guarant_spc = 0;
2874         rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2875         fdir_info->max_flexpayload = 0;
2876         fdir_info->flow_types_mask[0] = 0;
2877         fdir_info->flex_payload_unit = 0;
2878         fdir_info->max_flex_payload_segment_num = 0;
2879         fdir_info->flex_payload_limit = 0;
2880         memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2881 }
2882
2883 /**
2884  * Deal with flow director operations.
2885  *
2886  * @param dev
2887  *   Pointer to Ethernet device.
2888  * @param filter_op
2889  *   Operation to perform.
2890  * @param arg
2891  *   Pointer to operation-specific structure.
2892  *
2893  * @return
2894  *   0 on success, a negative errno value otherwise and rte_errno is set.
2895  */
2896 static int
2897 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2898                     void *arg)
2899 {
2900         enum rte_fdir_mode fdir_mode =
2901                 dev->data->dev_conf.fdir_conf.mode;
2902
2903         if (filter_op == RTE_ETH_FILTER_NOP)
2904                 return 0;
2905         if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2906             fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2907                 DRV_LOG(ERR, "port %u flow director mode %d not supported",
2908                         dev->data->port_id, fdir_mode);
2909                 rte_errno = EINVAL;
2910                 return -rte_errno;
2911         }
2912         switch (filter_op) {
2913         case RTE_ETH_FILTER_ADD:
2914                 return flow_fdir_filter_add(dev, arg);
2915         case RTE_ETH_FILTER_UPDATE:
2916                 return flow_fdir_filter_update(dev, arg);
2917         case RTE_ETH_FILTER_DELETE:
2918                 return flow_fdir_filter_delete(dev, arg);
2919         case RTE_ETH_FILTER_FLUSH:
2920                 flow_fdir_filter_flush(dev);
2921                 break;
2922         case RTE_ETH_FILTER_INFO:
2923                 flow_fdir_info_get(dev, arg);
2924                 break;
2925         default:
2926                 DRV_LOG(DEBUG, "port %u unknown operation %u",
2927                         dev->data->port_id, filter_op);
2928                 rte_errno = EINVAL;
2929                 return -rte_errno;
2930         }
2931         return 0;
2932 }
2933
2934 /**
2935  * Manage filter operations.
2936  *
2937  * @param dev
2938  *   Pointer to Ethernet device structure.
2939  * @param filter_type
2940  *   Filter type.
2941  * @param filter_op
2942  *   Operation to perform.
2943  * @param arg
2944  *   Pointer to operation-specific structure.
2945  *
2946  * @return
2947  *   0 on success, a negative errno value otherwise and rte_errno is set.
2948  */
2949 int
2950 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2951                      enum rte_filter_type filter_type,
2952                      enum rte_filter_op filter_op,
2953                      void *arg)
2954 {
2955         switch (filter_type) {
2956         case RTE_ETH_FILTER_GENERIC:
2957                 if (filter_op != RTE_ETH_FILTER_GET) {
2958                         rte_errno = EINVAL;
2959                         return -rte_errno;
2960                 }
2961                 *(const void **)arg = &mlx5_flow_ops;
2962                 return 0;
2963         case RTE_ETH_FILTER_FDIR:
2964                 return flow_fdir_ctrl_func(dev, filter_op, arg);
2965         default:
2966                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
2967                         dev->data->port_id, filter_type);
2968                 rte_errno = ENOTSUP;
2969                 return -rte_errno;
2970         }
2971         return 0;
2972 }