net/qede: fix odd number of queues usage in 100G mode
[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_verbs_drv_ops;
46
47 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
48
49 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
50         [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
51 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
52         [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
53 #endif
54         [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
55         [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
56 };
57
58 enum mlx5_expansion {
59         MLX5_EXPANSION_ROOT,
60         MLX5_EXPANSION_ROOT_OUTER,
61         MLX5_EXPANSION_ROOT_ETH_VLAN,
62         MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
63         MLX5_EXPANSION_OUTER_ETH,
64         MLX5_EXPANSION_OUTER_ETH_VLAN,
65         MLX5_EXPANSION_OUTER_VLAN,
66         MLX5_EXPANSION_OUTER_IPV4,
67         MLX5_EXPANSION_OUTER_IPV4_UDP,
68         MLX5_EXPANSION_OUTER_IPV4_TCP,
69         MLX5_EXPANSION_OUTER_IPV6,
70         MLX5_EXPANSION_OUTER_IPV6_UDP,
71         MLX5_EXPANSION_OUTER_IPV6_TCP,
72         MLX5_EXPANSION_VXLAN,
73         MLX5_EXPANSION_VXLAN_GPE,
74         MLX5_EXPANSION_GRE,
75         MLX5_EXPANSION_MPLS,
76         MLX5_EXPANSION_ETH,
77         MLX5_EXPANSION_ETH_VLAN,
78         MLX5_EXPANSION_VLAN,
79         MLX5_EXPANSION_IPV4,
80         MLX5_EXPANSION_IPV4_UDP,
81         MLX5_EXPANSION_IPV4_TCP,
82         MLX5_EXPANSION_IPV6,
83         MLX5_EXPANSION_IPV6_UDP,
84         MLX5_EXPANSION_IPV6_TCP,
85 };
86
87 /** Supported expansion of items. */
88 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
89         [MLX5_EXPANSION_ROOT] = {
90                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
91                                                  MLX5_EXPANSION_IPV4,
92                                                  MLX5_EXPANSION_IPV6),
93                 .type = RTE_FLOW_ITEM_TYPE_END,
94         },
95         [MLX5_EXPANSION_ROOT_OUTER] = {
96                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
97                                                  MLX5_EXPANSION_OUTER_IPV4,
98                                                  MLX5_EXPANSION_OUTER_IPV6),
99                 .type = RTE_FLOW_ITEM_TYPE_END,
100         },
101         [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
102                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
103                 .type = RTE_FLOW_ITEM_TYPE_END,
104         },
105         [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
106                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
107                 .type = RTE_FLOW_ITEM_TYPE_END,
108         },
109         [MLX5_EXPANSION_OUTER_ETH] = {
110                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
111                                                  MLX5_EXPANSION_OUTER_IPV6,
112                                                  MLX5_EXPANSION_MPLS),
113                 .type = RTE_FLOW_ITEM_TYPE_ETH,
114                 .rss_types = 0,
115         },
116         [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
117                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
118                 .type = RTE_FLOW_ITEM_TYPE_ETH,
119                 .rss_types = 0,
120         },
121         [MLX5_EXPANSION_OUTER_VLAN] = {
122                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
123                                                  MLX5_EXPANSION_OUTER_IPV6),
124                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
125         },
126         [MLX5_EXPANSION_OUTER_IPV4] = {
127                 .next = RTE_FLOW_EXPAND_RSS_NEXT
128                         (MLX5_EXPANSION_OUTER_IPV4_UDP,
129                          MLX5_EXPANSION_OUTER_IPV4_TCP,
130                          MLX5_EXPANSION_GRE,
131                          MLX5_EXPANSION_IPV4,
132                          MLX5_EXPANSION_IPV6),
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                          MLX5_EXPANSION_IPV4,
152                          MLX5_EXPANSION_IPV6),
153                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
154                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
155                         ETH_RSS_NONFRAG_IPV6_OTHER,
156         },
157         [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
158                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
159                                                  MLX5_EXPANSION_VXLAN_GPE),
160                 .type = RTE_FLOW_ITEM_TYPE_UDP,
161                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
162         },
163         [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
164                 .type = RTE_FLOW_ITEM_TYPE_TCP,
165                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
166         },
167         [MLX5_EXPANSION_VXLAN] = {
168                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
169                 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
170         },
171         [MLX5_EXPANSION_VXLAN_GPE] = {
172                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
173                                                  MLX5_EXPANSION_IPV4,
174                                                  MLX5_EXPANSION_IPV6),
175                 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
176         },
177         [MLX5_EXPANSION_GRE] = {
178                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
179                 .type = RTE_FLOW_ITEM_TYPE_GRE,
180         },
181         [MLX5_EXPANSION_MPLS] = {
182                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
183                                                  MLX5_EXPANSION_IPV6),
184                 .type = RTE_FLOW_ITEM_TYPE_MPLS,
185         },
186         [MLX5_EXPANSION_ETH] = {
187                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188                                                  MLX5_EXPANSION_IPV6),
189                 .type = RTE_FLOW_ITEM_TYPE_ETH,
190         },
191         [MLX5_EXPANSION_ETH_VLAN] = {
192                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
193                 .type = RTE_FLOW_ITEM_TYPE_ETH,
194         },
195         [MLX5_EXPANSION_VLAN] = {
196                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
197                                                  MLX5_EXPANSION_IPV6),
198                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
199         },
200         [MLX5_EXPANSION_IPV4] = {
201                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
202                                                  MLX5_EXPANSION_IPV4_TCP),
203                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
204                 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
205                         ETH_RSS_NONFRAG_IPV4_OTHER,
206         },
207         [MLX5_EXPANSION_IPV4_UDP] = {
208                 .type = RTE_FLOW_ITEM_TYPE_UDP,
209                 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
210         },
211         [MLX5_EXPANSION_IPV4_TCP] = {
212                 .type = RTE_FLOW_ITEM_TYPE_TCP,
213                 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
214         },
215         [MLX5_EXPANSION_IPV6] = {
216                 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
217                                                  MLX5_EXPANSION_IPV6_TCP),
218                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
219                 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
220                         ETH_RSS_NONFRAG_IPV6_OTHER,
221         },
222         [MLX5_EXPANSION_IPV6_UDP] = {
223                 .type = RTE_FLOW_ITEM_TYPE_UDP,
224                 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
225         },
226         [MLX5_EXPANSION_IPV6_TCP] = {
227                 .type = RTE_FLOW_ITEM_TYPE_TCP,
228                 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
229         },
230 };
231
232 static const struct rte_flow_ops mlx5_flow_ops = {
233         .validate = mlx5_flow_validate,
234         .create = mlx5_flow_create,
235         .destroy = mlx5_flow_destroy,
236         .flush = mlx5_flow_flush,
237         .isolate = mlx5_flow_isolate,
238         .query = mlx5_flow_query,
239 };
240
241 /* Convert FDIR request to Generic flow. */
242 struct mlx5_fdir {
243         struct rte_flow_attr attr;
244         struct rte_flow_item items[4];
245         struct rte_flow_item_eth l2;
246         struct rte_flow_item_eth l2_mask;
247         union {
248                 struct rte_flow_item_ipv4 ipv4;
249                 struct rte_flow_item_ipv6 ipv6;
250         } l3;
251         union {
252                 struct rte_flow_item_ipv4 ipv4;
253                 struct rte_flow_item_ipv6 ipv6;
254         } l3_mask;
255         union {
256                 struct rte_flow_item_udp udp;
257                 struct rte_flow_item_tcp tcp;
258         } l4;
259         union {
260                 struct rte_flow_item_udp udp;
261                 struct rte_flow_item_tcp tcp;
262         } l4_mask;
263         struct rte_flow_action actions[2];
264         struct rte_flow_action_queue queue;
265 };
266
267 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
268 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
269         { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
270 };
271
272 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
273 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
274         { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
275         { 9, 10, 11 }, { 12, 13, 14 },
276 };
277
278 /* Tunnel information. */
279 struct mlx5_flow_tunnel_info {
280         uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
281         uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
282 };
283
284 static struct mlx5_flow_tunnel_info tunnels_info[] = {
285         {
286                 .tunnel = MLX5_FLOW_LAYER_VXLAN,
287                 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
288         },
289         {
290                 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
291                 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
292         },
293         {
294                 .tunnel = MLX5_FLOW_LAYER_GRE,
295                 .ptype = RTE_PTYPE_TUNNEL_GRE,
296         },
297         {
298                 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
299                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
300         },
301         {
302                 .tunnel = MLX5_FLOW_LAYER_MPLS,
303                 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
304         },
305         {
306                 .tunnel = MLX5_FLOW_LAYER_NVGRE,
307                 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
308         },
309         {
310                 .tunnel = MLX5_FLOW_LAYER_IPIP,
311                 .ptype = RTE_PTYPE_TUNNEL_IP,
312         },
313         {
314                 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
315                 .ptype = RTE_PTYPE_TUNNEL_IP,
316         },
317 };
318
319 /**
320  * Discover the maximum number of priority available.
321  *
322  * @param[in] dev
323  *   Pointer to the Ethernet device structure.
324  *
325  * @return
326  *   number of supported flow priority on success, a negative errno
327  *   value otherwise and rte_errno is set.
328  */
329 int
330 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
331 {
332         struct mlx5_priv *priv = dev->data->dev_private;
333         struct {
334                 struct ibv_flow_attr attr;
335                 struct ibv_flow_spec_eth eth;
336                 struct ibv_flow_spec_action_drop drop;
337         } flow_attr = {
338                 .attr = {
339                         .num_of_specs = 2,
340                         .port = (uint8_t)priv->ibv_port,
341                 },
342                 .eth = {
343                         .type = IBV_FLOW_SPEC_ETH,
344                         .size = sizeof(struct ibv_flow_spec_eth),
345                 },
346                 .drop = {
347                         .size = sizeof(struct ibv_flow_spec_action_drop),
348                         .type = IBV_FLOW_SPEC_ACTION_DROP,
349                 },
350         };
351         struct ibv_flow *flow;
352         struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
353         uint16_t vprio[] = { 8, 16 };
354         int i;
355         int priority = 0;
356
357         if (!drop) {
358                 rte_errno = ENOTSUP;
359                 return -rte_errno;
360         }
361         for (i = 0; i != RTE_DIM(vprio); i++) {
362                 flow_attr.attr.priority = vprio[i] - 1;
363                 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
364                 if (!flow)
365                         break;
366                 claim_zero(mlx5_glue->destroy_flow(flow));
367                 priority = vprio[i];
368         }
369         mlx5_hrxq_drop_release(dev);
370         switch (priority) {
371         case 8:
372                 priority = RTE_DIM(priority_map_3);
373                 break;
374         case 16:
375                 priority = RTE_DIM(priority_map_5);
376                 break;
377         default:
378                 rte_errno = ENOTSUP;
379                 DRV_LOG(ERR,
380                         "port %u verbs maximum priority: %d expected 8/16",
381                         dev->data->port_id, priority);
382                 return -rte_errno;
383         }
384         DRV_LOG(INFO, "port %u flow maximum priority: %d",
385                 dev->data->port_id, priority);
386         return priority;
387 }
388
389 /**
390  * Adjust flow priority based on the highest layer and the request priority.
391  *
392  * @param[in] dev
393  *   Pointer to the Ethernet device structure.
394  * @param[in] priority
395  *   The rule base priority.
396  * @param[in] subpriority
397  *   The priority based on the items.
398  *
399  * @return
400  *   The new priority.
401  */
402 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
403                                    uint32_t subpriority)
404 {
405         uint32_t res = 0;
406         struct mlx5_priv *priv = dev->data->dev_private;
407
408         switch (priv->config.flow_prio) {
409         case RTE_DIM(priority_map_3):
410                 res = priority_map_3[priority][subpriority];
411                 break;
412         case RTE_DIM(priority_map_5):
413                 res = priority_map_5[priority][subpriority];
414                 break;
415         }
416         return  res;
417 }
418
419 /**
420  * Verify the @p item specifications (spec, last, mask) are compatible with the
421  * NIC capabilities.
422  *
423  * @param[in] item
424  *   Item specification.
425  * @param[in] mask
426  *   @p item->mask or flow default bit-masks.
427  * @param[in] nic_mask
428  *   Bit-masks covering supported fields by the NIC to compare with user mask.
429  * @param[in] size
430  *   Bit-masks size in bytes.
431  * @param[out] error
432  *   Pointer to error structure.
433  *
434  * @return
435  *   0 on success, a negative errno value otherwise and rte_errno is set.
436  */
437 int
438 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
439                           const uint8_t *mask,
440                           const uint8_t *nic_mask,
441                           unsigned int size,
442                           struct rte_flow_error *error)
443 {
444         unsigned int i;
445
446         assert(nic_mask);
447         for (i = 0; i < size; ++i)
448                 if ((nic_mask[i] | mask[i]) != nic_mask[i])
449                         return rte_flow_error_set(error, ENOTSUP,
450                                                   RTE_FLOW_ERROR_TYPE_ITEM,
451                                                   item,
452                                                   "mask enables non supported"
453                                                   " bits");
454         if (!item->spec && (item->mask || item->last))
455                 return rte_flow_error_set(error, EINVAL,
456                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
457                                           "mask/last without a spec is not"
458                                           " supported");
459         if (item->spec && item->last) {
460                 uint8_t spec[size];
461                 uint8_t last[size];
462                 unsigned int i;
463                 int ret;
464
465                 for (i = 0; i < size; ++i) {
466                         spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
467                         last[i] = ((const uint8_t *)item->last)[i] & mask[i];
468                 }
469                 ret = memcmp(spec, last, size);
470                 if (ret != 0)
471                         return rte_flow_error_set(error, EINVAL,
472                                                   RTE_FLOW_ERROR_TYPE_ITEM,
473                                                   item,
474                                                   "range is not valid");
475         }
476         return 0;
477 }
478
479 /**
480  * Adjust the hash fields according to the @p flow information.
481  *
482  * @param[in] dev_flow.
483  *   Pointer to the mlx5_flow.
484  * @param[in] tunnel
485  *   1 when the hash field is for a tunnel item.
486  * @param[in] layer_types
487  *   ETH_RSS_* types.
488  * @param[in] hash_fields
489  *   Item hash fields.
490  *
491  * @return
492  *   The hash fields that should be used.
493  */
494 uint64_t
495 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
496                             int tunnel __rte_unused, uint64_t layer_types,
497                             uint64_t hash_fields)
498 {
499         struct rte_flow *flow = dev_flow->flow;
500 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
501         int rss_request_inner = flow->rss.level >= 2;
502
503         /* Check RSS hash level for tunnel. */
504         if (tunnel && rss_request_inner)
505                 hash_fields |= IBV_RX_HASH_INNER;
506         else if (tunnel || rss_request_inner)
507                 return 0;
508 #endif
509         /* Check if requested layer matches RSS hash fields. */
510         if (!(flow->rss.types & layer_types))
511                 return 0;
512         return hash_fields;
513 }
514
515 /**
516  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
517  * if several tunnel rules are used on this queue, the tunnel ptype will be
518  * cleared.
519  *
520  * @param rxq_ctrl
521  *   Rx queue to update.
522  */
523 static void
524 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
525 {
526         unsigned int i;
527         uint32_t tunnel_ptype = 0;
528
529         /* Look up for the ptype to use. */
530         for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
531                 if (!rxq_ctrl->flow_tunnels_n[i])
532                         continue;
533                 if (!tunnel_ptype) {
534                         tunnel_ptype = tunnels_info[i].ptype;
535                 } else {
536                         tunnel_ptype = 0;
537                         break;
538                 }
539         }
540         rxq_ctrl->rxq.tunnel = tunnel_ptype;
541 }
542
543 /**
544  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
545  * flow.
546  *
547  * @param[in] dev
548  *   Pointer to the Ethernet device structure.
549  * @param[in] dev_flow
550  *   Pointer to device flow structure.
551  */
552 static void
553 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
554 {
555         struct mlx5_priv *priv = dev->data->dev_private;
556         struct rte_flow *flow = dev_flow->flow;
557         const int mark = !!(flow->actions &
558                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
559         const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
560         unsigned int i;
561
562         for (i = 0; i != flow->rss.queue_num; ++i) {
563                 int idx = (*flow->queue)[i];
564                 struct mlx5_rxq_ctrl *rxq_ctrl =
565                         container_of((*priv->rxqs)[idx],
566                                      struct mlx5_rxq_ctrl, rxq);
567
568                 if (mark) {
569                         rxq_ctrl->rxq.mark = 1;
570                         rxq_ctrl->flow_mark_n++;
571                 }
572                 if (tunnel) {
573                         unsigned int j;
574
575                         /* Increase the counter matching the flow. */
576                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
577                                 if ((tunnels_info[j].tunnel &
578                                      dev_flow->layers) ==
579                                     tunnels_info[j].tunnel) {
580                                         rxq_ctrl->flow_tunnels_n[j]++;
581                                         break;
582                                 }
583                         }
584                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
585                 }
586         }
587 }
588
589 /**
590  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
591  *
592  * @param[in] dev
593  *   Pointer to the Ethernet device structure.
594  * @param[in] flow
595  *   Pointer to flow structure.
596  */
597 static void
598 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
599 {
600         struct mlx5_flow *dev_flow;
601
602         LIST_FOREACH(dev_flow, &flow->dev_flows, next)
603                 flow_drv_rxq_flags_set(dev, dev_flow);
604 }
605
606 /**
607  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
608  * device flow if no other flow uses it with the same kind of request.
609  *
610  * @param dev
611  *   Pointer to Ethernet device.
612  * @param[in] dev_flow
613  *   Pointer to the device flow.
614  */
615 static void
616 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
617 {
618         struct mlx5_priv *priv = dev->data->dev_private;
619         struct rte_flow *flow = dev_flow->flow;
620         const int mark = !!(flow->actions &
621                             (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
622         const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
623         unsigned int i;
624
625         assert(dev->data->dev_started);
626         for (i = 0; i != flow->rss.queue_num; ++i) {
627                 int idx = (*flow->queue)[i];
628                 struct mlx5_rxq_ctrl *rxq_ctrl =
629                         container_of((*priv->rxqs)[idx],
630                                      struct mlx5_rxq_ctrl, rxq);
631
632                 if (mark) {
633                         rxq_ctrl->flow_mark_n--;
634                         rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
635                 }
636                 if (tunnel) {
637                         unsigned int j;
638
639                         /* Decrease the counter matching the flow. */
640                         for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
641                                 if ((tunnels_info[j].tunnel &
642                                      dev_flow->layers) ==
643                                     tunnels_info[j].tunnel) {
644                                         rxq_ctrl->flow_tunnels_n[j]--;
645                                         break;
646                                 }
647                         }
648                         flow_rxq_tunnel_ptype_update(rxq_ctrl);
649                 }
650         }
651 }
652
653 /**
654  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
655  * @p flow if no other flow uses it with the same kind of request.
656  *
657  * @param dev
658  *   Pointer to Ethernet device.
659  * @param[in] flow
660  *   Pointer to the flow.
661  */
662 static void
663 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
664 {
665         struct mlx5_flow *dev_flow;
666
667         LIST_FOREACH(dev_flow, &flow->dev_flows, next)
668                 flow_drv_rxq_flags_trim(dev, dev_flow);
669 }
670
671 /**
672  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
673  *
674  * @param dev
675  *   Pointer to Ethernet device.
676  */
677 static void
678 flow_rxq_flags_clear(struct rte_eth_dev *dev)
679 {
680         struct mlx5_priv *priv = dev->data->dev_private;
681         unsigned int i;
682
683         for (i = 0; i != priv->rxqs_n; ++i) {
684                 struct mlx5_rxq_ctrl *rxq_ctrl;
685                 unsigned int j;
686
687                 if (!(*priv->rxqs)[i])
688                         continue;
689                 rxq_ctrl = container_of((*priv->rxqs)[i],
690                                         struct mlx5_rxq_ctrl, rxq);
691                 rxq_ctrl->flow_mark_n = 0;
692                 rxq_ctrl->rxq.mark = 0;
693                 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
694                         rxq_ctrl->flow_tunnels_n[j] = 0;
695                 rxq_ctrl->rxq.tunnel = 0;
696         }
697 }
698
699 /*
700  * return a pointer to the desired action in the list of actions.
701  *
702  * @param[in] actions
703  *   The list of actions to search the action in.
704  * @param[in] action
705  *   The action to find.
706  *
707  * @return
708  *   Pointer to the action in the list, if found. NULL otherwise.
709  */
710 const struct rte_flow_action *
711 mlx5_flow_find_action(const struct rte_flow_action *actions,
712                       enum rte_flow_action_type action)
713 {
714         if (actions == NULL)
715                 return NULL;
716         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
717                 if (actions->type == action)
718                         return actions;
719         return NULL;
720 }
721
722 /*
723  * Validate the flag action.
724  *
725  * @param[in] action_flags
726  *   Bit-fields that holds the actions detected until now.
727  * @param[in] attr
728  *   Attributes of flow that includes this action.
729  * @param[out] error
730  *   Pointer to error structure.
731  *
732  * @return
733  *   0 on success, a negative errno value otherwise and rte_errno is set.
734  */
735 int
736 mlx5_flow_validate_action_flag(uint64_t action_flags,
737                                const struct rte_flow_attr *attr,
738                                struct rte_flow_error *error)
739 {
740
741         if (action_flags & MLX5_FLOW_ACTION_DROP)
742                 return rte_flow_error_set(error, EINVAL,
743                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
744                                           "can't drop and flag in same flow");
745         if (action_flags & MLX5_FLOW_ACTION_MARK)
746                 return rte_flow_error_set(error, EINVAL,
747                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
748                                           "can't mark and flag in same flow");
749         if (action_flags & MLX5_FLOW_ACTION_FLAG)
750                 return rte_flow_error_set(error, EINVAL,
751                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
752                                           "can't have 2 flag"
753                                           " actions in same flow");
754         if (attr->egress)
755                 return rte_flow_error_set(error, ENOTSUP,
756                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
757                                           "flag action not supported for "
758                                           "egress");
759         return 0;
760 }
761
762 /*
763  * Validate the mark action.
764  *
765  * @param[in] action
766  *   Pointer to the queue action.
767  * @param[in] action_flags
768  *   Bit-fields that holds the actions detected until now.
769  * @param[in] attr
770  *   Attributes of flow that includes this action.
771  * @param[out] error
772  *   Pointer to error structure.
773  *
774  * @return
775  *   0 on success, a negative errno value otherwise and rte_errno is set.
776  */
777 int
778 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
779                                uint64_t action_flags,
780                                const struct rte_flow_attr *attr,
781                                struct rte_flow_error *error)
782 {
783         const struct rte_flow_action_mark *mark = action->conf;
784
785         if (!mark)
786                 return rte_flow_error_set(error, EINVAL,
787                                           RTE_FLOW_ERROR_TYPE_ACTION,
788                                           action,
789                                           "configuration cannot be null");
790         if (mark->id >= MLX5_FLOW_MARK_MAX)
791                 return rte_flow_error_set(error, EINVAL,
792                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
793                                           &mark->id,
794                                           "mark id must in 0 <= id < "
795                                           RTE_STR(MLX5_FLOW_MARK_MAX));
796         if (action_flags & MLX5_FLOW_ACTION_DROP)
797                 return rte_flow_error_set(error, EINVAL,
798                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
799                                           "can't drop and mark in same flow");
800         if (action_flags & MLX5_FLOW_ACTION_FLAG)
801                 return rte_flow_error_set(error, EINVAL,
802                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
803                                           "can't flag and mark in same flow");
804         if (action_flags & MLX5_FLOW_ACTION_MARK)
805                 return rte_flow_error_set(error, EINVAL,
806                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
807                                           "can't have 2 mark actions in same"
808                                           " flow");
809         if (attr->egress)
810                 return rte_flow_error_set(error, ENOTSUP,
811                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
812                                           "mark action not supported for "
813                                           "egress");
814         return 0;
815 }
816
817 /*
818  * Validate the drop action.
819  *
820  * @param[in] action_flags
821  *   Bit-fields that holds the actions detected until now.
822  * @param[in] attr
823  *   Attributes of flow that includes this action.
824  * @param[out] error
825  *   Pointer to error structure.
826  *
827  * @return
828  *   0 on success, a negative errno value otherwise and rte_errno is set.
829  */
830 int
831 mlx5_flow_validate_action_drop(uint64_t action_flags,
832                                const struct rte_flow_attr *attr,
833                                struct rte_flow_error *error)
834 {
835         if (action_flags & MLX5_FLOW_ACTION_FLAG)
836                 return rte_flow_error_set(error, EINVAL,
837                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
838                                           "can't drop and flag in same flow");
839         if (action_flags & MLX5_FLOW_ACTION_MARK)
840                 return rte_flow_error_set(error, EINVAL,
841                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
842                                           "can't drop and mark in same flow");
843         if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
844                             MLX5_FLOW_FATE_ESWITCH_ACTIONS))
845                 return rte_flow_error_set(error, EINVAL,
846                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
847                                           "can't have 2 fate actions in"
848                                           " same flow");
849         if (attr->egress)
850                 return rte_flow_error_set(error, ENOTSUP,
851                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
852                                           "drop action not supported for "
853                                           "egress");
854         return 0;
855 }
856
857 /*
858  * Validate the queue action.
859  *
860  * @param[in] action
861  *   Pointer to the queue action.
862  * @param[in] action_flags
863  *   Bit-fields that holds the actions detected until now.
864  * @param[in] dev
865  *   Pointer to the Ethernet device structure.
866  * @param[in] attr
867  *   Attributes of flow that includes this action.
868  * @param[out] error
869  *   Pointer to error structure.
870  *
871  * @return
872  *   0 on success, a negative errno value otherwise and rte_errno is set.
873  */
874 int
875 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
876                                 uint64_t action_flags,
877                                 struct rte_eth_dev *dev,
878                                 const struct rte_flow_attr *attr,
879                                 struct rte_flow_error *error)
880 {
881         struct mlx5_priv *priv = dev->data->dev_private;
882         const struct rte_flow_action_queue *queue = action->conf;
883
884         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
885                 return rte_flow_error_set(error, EINVAL,
886                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
887                                           "can't have 2 fate actions in"
888                                           " same flow");
889         if (!priv->rxqs_n)
890                 return rte_flow_error_set(error, EINVAL,
891                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
892                                           NULL, "No Rx queues configured");
893         if (queue->index >= priv->rxqs_n)
894                 return rte_flow_error_set(error, EINVAL,
895                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
896                                           &queue->index,
897                                           "queue index out of range");
898         if (!(*priv->rxqs)[queue->index])
899                 return rte_flow_error_set(error, EINVAL,
900                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
901                                           &queue->index,
902                                           "queue is not configured");
903         if (attr->egress)
904                 return rte_flow_error_set(error, ENOTSUP,
905                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
906                                           "queue action not supported for "
907                                           "egress");
908         return 0;
909 }
910
911 /*
912  * Validate the rss action.
913  *
914  * @param[in] action
915  *   Pointer to the queue action.
916  * @param[in] action_flags
917  *   Bit-fields that holds the actions detected until now.
918  * @param[in] dev
919  *   Pointer to the Ethernet device structure.
920  * @param[in] attr
921  *   Attributes of flow that includes this action.
922  * @param[in] item_flags
923  *   Items that were detected.
924  * @param[out] error
925  *   Pointer to error structure.
926  *
927  * @return
928  *   0 on success, a negative errno value otherwise and rte_errno is set.
929  */
930 int
931 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
932                               uint64_t action_flags,
933                               struct rte_eth_dev *dev,
934                               const struct rte_flow_attr *attr,
935                               uint64_t item_flags,
936                               struct rte_flow_error *error)
937 {
938         struct mlx5_priv *priv = dev->data->dev_private;
939         const struct rte_flow_action_rss *rss = action->conf;
940         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
941         unsigned int i;
942
943         if (action_flags & MLX5_FLOW_FATE_ACTIONS)
944                 return rte_flow_error_set(error, EINVAL,
945                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
946                                           "can't have 2 fate actions"
947                                           " in same flow");
948         if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
949             rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
950                 return rte_flow_error_set(error, ENOTSUP,
951                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
952                                           &rss->func,
953                                           "RSS hash function not supported");
954 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
955         if (rss->level > 2)
956 #else
957         if (rss->level > 1)
958 #endif
959                 return rte_flow_error_set(error, ENOTSUP,
960                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
961                                           &rss->level,
962                                           "tunnel RSS is not supported");
963         /* allow RSS key_len 0 in case of NULL (default) RSS key. */
964         if (rss->key_len == 0 && rss->key != NULL)
965                 return rte_flow_error_set(error, ENOTSUP,
966                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
967                                           &rss->key_len,
968                                           "RSS hash key length 0");
969         if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
970                 return rte_flow_error_set(error, ENOTSUP,
971                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
972                                           &rss->key_len,
973                                           "RSS hash key too small");
974         if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
975                 return rte_flow_error_set(error, ENOTSUP,
976                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
977                                           &rss->key_len,
978                                           "RSS hash key too large");
979         if (rss->queue_num > priv->config.ind_table_max_size)
980                 return rte_flow_error_set(error, ENOTSUP,
981                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
982                                           &rss->queue_num,
983                                           "number of queues too large");
984         if (rss->types & MLX5_RSS_HF_MASK)
985                 return rte_flow_error_set(error, ENOTSUP,
986                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
987                                           &rss->types,
988                                           "some RSS protocols are not"
989                                           " supported");
990         if (!priv->rxqs_n)
991                 return rte_flow_error_set(error, EINVAL,
992                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
993                                           NULL, "No Rx queues configured");
994         if (!rss->queue_num)
995                 return rte_flow_error_set(error, EINVAL,
996                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF,
997                                           NULL, "No queues configured");
998         for (i = 0; i != rss->queue_num; ++i) {
999                 if (!(*priv->rxqs)[rss->queue[i]])
1000                         return rte_flow_error_set
1001                                 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1002                                  &rss->queue[i], "queue is not configured");
1003         }
1004         if (attr->egress)
1005                 return rte_flow_error_set(error, ENOTSUP,
1006                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1007                                           "rss action not supported for "
1008                                           "egress");
1009         if (rss->level > 1 &&  !tunnel)
1010                 return rte_flow_error_set(error, EINVAL,
1011                                           RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1012                                           "inner RSS is not supported for "
1013                                           "non-tunnel flows");
1014         return 0;
1015 }
1016
1017 /*
1018  * Validate the count action.
1019  *
1020  * @param[in] dev
1021  *   Pointer to the Ethernet device structure.
1022  * @param[in] attr
1023  *   Attributes of flow that includes this action.
1024  * @param[out] error
1025  *   Pointer to error structure.
1026  *
1027  * @return
1028  *   0 on success, a negative errno value otherwise and rte_errno is set.
1029  */
1030 int
1031 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1032                                 const struct rte_flow_attr *attr,
1033                                 struct rte_flow_error *error)
1034 {
1035         if (attr->egress)
1036                 return rte_flow_error_set(error, ENOTSUP,
1037                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1038                                           "count action not supported for "
1039                                           "egress");
1040         return 0;
1041 }
1042
1043 /**
1044  * Verify the @p attributes will be correctly understood by the NIC and store
1045  * them in the @p flow if everything is correct.
1046  *
1047  * @param[in] dev
1048  *   Pointer to the Ethernet device structure.
1049  * @param[in] attributes
1050  *   Pointer to flow attributes
1051  * @param[out] error
1052  *   Pointer to error structure.
1053  *
1054  * @return
1055  *   0 on success, a negative errno value otherwise and rte_errno is set.
1056  */
1057 int
1058 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1059                               const struct rte_flow_attr *attributes,
1060                               struct rte_flow_error *error)
1061 {
1062         struct mlx5_priv *priv = dev->data->dev_private;
1063         uint32_t priority_max = priv->config.flow_prio - 1;
1064
1065         if (attributes->group)
1066                 return rte_flow_error_set(error, ENOTSUP,
1067                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1068                                           NULL, "groups is not supported");
1069         if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1070             attributes->priority >= priority_max)
1071                 return rte_flow_error_set(error, ENOTSUP,
1072                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1073                                           NULL, "priority out of range");
1074         if (attributes->egress)
1075                 return rte_flow_error_set(error, ENOTSUP,
1076                                           RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1077                                           "egress is not supported");
1078         if (attributes->transfer && !priv->config.dv_esw_en)
1079                 return rte_flow_error_set(error, ENOTSUP,
1080                                           RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1081                                           NULL, "transfer is not supported");
1082         if (!attributes->ingress)
1083                 return rte_flow_error_set(error, EINVAL,
1084                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1085                                           NULL,
1086                                           "ingress attribute is mandatory");
1087         return 0;
1088 }
1089
1090 /**
1091  * Validate ICMP6 item.
1092  *
1093  * @param[in] item
1094  *   Item specification.
1095  * @param[in] item_flags
1096  *   Bit-fields that holds the items detected until now.
1097  * @param[out] error
1098  *   Pointer to error structure.
1099  *
1100  * @return
1101  *   0 on success, a negative errno value otherwise and rte_errno is set.
1102  */
1103 int
1104 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1105                                uint64_t item_flags,
1106                                uint8_t target_protocol,
1107                                struct rte_flow_error *error)
1108 {
1109         const struct rte_flow_item_icmp6 *mask = item->mask;
1110         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1111         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1112                                       MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1113         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1114                                       MLX5_FLOW_LAYER_OUTER_L4;
1115         int ret;
1116
1117         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1118                 return rte_flow_error_set(error, EINVAL,
1119                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1120                                           "protocol filtering not compatible"
1121                                           " with ICMP6 layer");
1122         if (!(item_flags & l3m))
1123                 return rte_flow_error_set(error, EINVAL,
1124                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1125                                           "IPv6 is mandatory to filter on"
1126                                           " ICMP6");
1127         if (item_flags & l4m)
1128                 return rte_flow_error_set(error, EINVAL,
1129                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1130                                           "multiple L4 layers not supported");
1131         if (!mask)
1132                 mask = &rte_flow_item_icmp6_mask;
1133         ret = mlx5_flow_item_acceptable
1134                 (item, (const uint8_t *)mask,
1135                  (const uint8_t *)&rte_flow_item_icmp6_mask,
1136                  sizeof(struct rte_flow_item_icmp6), error);
1137         if (ret < 0)
1138                 return ret;
1139         return 0;
1140 }
1141
1142 /**
1143  * Validate ICMP item.
1144  *
1145  * @param[in] item
1146  *   Item specification.
1147  * @param[in] item_flags
1148  *   Bit-fields that holds the items detected until now.
1149  * @param[out] error
1150  *   Pointer to error structure.
1151  *
1152  * @return
1153  *   0 on success, a negative errno value otherwise and rte_errno is set.
1154  */
1155 int
1156 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1157                              uint64_t item_flags,
1158                              uint8_t target_protocol,
1159                              struct rte_flow_error *error)
1160 {
1161         const struct rte_flow_item_icmp *mask = item->mask;
1162         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1163         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1164                                       MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1165         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1166                                       MLX5_FLOW_LAYER_OUTER_L4;
1167         int ret;
1168
1169         if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1170                 return rte_flow_error_set(error, EINVAL,
1171                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1172                                           "protocol filtering not compatible"
1173                                           " with ICMP layer");
1174         if (!(item_flags & l3m))
1175                 return rte_flow_error_set(error, EINVAL,
1176                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1177                                           "IPv4 is mandatory to filter"
1178                                           " on ICMP");
1179         if (item_flags & l4m)
1180                 return rte_flow_error_set(error, EINVAL,
1181                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1182                                           "multiple L4 layers not supported");
1183         if (!mask)
1184                 mask = &rte_flow_item_icmp_mask;
1185         ret = mlx5_flow_item_acceptable
1186                 (item, (const uint8_t *)mask,
1187                  (const uint8_t *)&rte_flow_item_icmp_mask,
1188                  sizeof(struct rte_flow_item_icmp), error);
1189         if (ret < 0)
1190                 return ret;
1191         return 0;
1192 }
1193
1194 /**
1195  * Validate Ethernet item.
1196  *
1197  * @param[in] item
1198  *   Item specification.
1199  * @param[in] item_flags
1200  *   Bit-fields that holds the items detected until now.
1201  * @param[out] error
1202  *   Pointer to error structure.
1203  *
1204  * @return
1205  *   0 on success, a negative errno value otherwise and rte_errno is set.
1206  */
1207 int
1208 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1209                             uint64_t item_flags,
1210                             struct rte_flow_error *error)
1211 {
1212         const struct rte_flow_item_eth *mask = item->mask;
1213         const struct rte_flow_item_eth nic_mask = {
1214                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1215                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1216                 .type = RTE_BE16(0xffff),
1217         };
1218         int ret;
1219         int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1220         const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1221                                        MLX5_FLOW_LAYER_OUTER_L2;
1222
1223         if (item_flags & ethm)
1224                 return rte_flow_error_set(error, ENOTSUP,
1225                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1226                                           "multiple L2 layers not supported");
1227         if (!mask)
1228                 mask = &rte_flow_item_eth_mask;
1229         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1230                                         (const uint8_t *)&nic_mask,
1231                                         sizeof(struct rte_flow_item_eth),
1232                                         error);
1233         return ret;
1234 }
1235
1236 /**
1237  * Validate VLAN item.
1238  *
1239  * @param[in] item
1240  *   Item specification.
1241  * @param[in] item_flags
1242  *   Bit-fields that holds the items detected until now.
1243  * @param[in] dev
1244  *   Ethernet device flow is being created on.
1245  * @param[out] error
1246  *   Pointer to error structure.
1247  *
1248  * @return
1249  *   0 on success, a negative errno value otherwise and rte_errno is set.
1250  */
1251 int
1252 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1253                              uint64_t item_flags,
1254                              struct rte_eth_dev *dev,
1255                              struct rte_flow_error *error)
1256 {
1257         const struct rte_flow_item_vlan *spec = item->spec;
1258         const struct rte_flow_item_vlan *mask = item->mask;
1259         const struct rte_flow_item_vlan nic_mask = {
1260                 .tci = RTE_BE16(UINT16_MAX),
1261                 .inner_type = RTE_BE16(UINT16_MAX),
1262         };
1263         uint16_t vlan_tag = 0;
1264         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1265         int ret;
1266         const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1267                                         MLX5_FLOW_LAYER_INNER_L4) :
1268                                        (MLX5_FLOW_LAYER_OUTER_L3 |
1269                                         MLX5_FLOW_LAYER_OUTER_L4);
1270         const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1271                                         MLX5_FLOW_LAYER_OUTER_VLAN;
1272
1273         if (item_flags & vlanm)
1274                 return rte_flow_error_set(error, EINVAL,
1275                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1276                                           "multiple VLAN layers not supported");
1277         else if ((item_flags & l34m) != 0)
1278                 return rte_flow_error_set(error, EINVAL,
1279                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1280                                           "L2 layer cannot follow L3/L4 layer");
1281         if (!mask)
1282                 mask = &rte_flow_item_vlan_mask;
1283         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1284                                         (const uint8_t *)&nic_mask,
1285                                         sizeof(struct rte_flow_item_vlan),
1286                                         error);
1287         if (ret)
1288                 return ret;
1289         if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1290                 struct mlx5_priv *priv = dev->data->dev_private;
1291
1292                 if (priv->vmwa_context) {
1293                         /*
1294                          * Non-NULL context means we have a virtual machine
1295                          * and SR-IOV enabled, we have to create VLAN interface
1296                          * to make hypervisor to setup E-Switch vport
1297                          * context correctly. We avoid creating the multiple
1298                          * VLAN interfaces, so we cannot support VLAN tag mask.
1299                          */
1300                         return rte_flow_error_set(error, EINVAL,
1301                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1302                                                   item,
1303                                                   "VLAN tag mask is not"
1304                                                   " supported in virtual"
1305                                                   " environment");
1306                 }
1307         }
1308         if (spec) {
1309                 vlan_tag = spec->tci;
1310                 vlan_tag &= mask->tci;
1311         }
1312         /*
1313          * From verbs perspective an empty VLAN is equivalent
1314          * to a packet without VLAN layer.
1315          */
1316         if (!vlan_tag)
1317                 return rte_flow_error_set(error, EINVAL,
1318                                           RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1319                                           item->spec,
1320                                           "VLAN cannot be empty");
1321         return 0;
1322 }
1323
1324 /**
1325  * Validate IPV4 item.
1326  *
1327  * @param[in] item
1328  *   Item specification.
1329  * @param[in] item_flags
1330  *   Bit-fields that holds the items detected until now.
1331  * @param[in] acc_mask
1332  *   Acceptable mask, if NULL default internal default mask
1333  *   will be used to check whether item fields are supported.
1334  * @param[out] error
1335  *   Pointer to error structure.
1336  *
1337  * @return
1338  *   0 on success, a negative errno value otherwise and rte_errno is set.
1339  */
1340 int
1341 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1342                              uint64_t item_flags,
1343                              const struct rte_flow_item_ipv4 *acc_mask,
1344                              struct rte_flow_error *error)
1345 {
1346         const struct rte_flow_item_ipv4 *mask = item->mask;
1347         const struct rte_flow_item_ipv4 *spec = item->spec;
1348         const struct rte_flow_item_ipv4 nic_mask = {
1349                 .hdr = {
1350                         .src_addr = RTE_BE32(0xffffffff),
1351                         .dst_addr = RTE_BE32(0xffffffff),
1352                         .type_of_service = 0xff,
1353                         .next_proto_id = 0xff,
1354                 },
1355         };
1356         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1357         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1358                                       MLX5_FLOW_LAYER_OUTER_L3;
1359         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1360                                       MLX5_FLOW_LAYER_OUTER_L4;
1361         int ret;
1362         uint8_t next_proto = 0xFF;
1363
1364         if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1365                 if (mask && spec)
1366                         next_proto = mask->hdr.next_proto_id &
1367                                      spec->hdr.next_proto_id;
1368                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1369                         return rte_flow_error_set(error, EINVAL,
1370                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1371                                                   item,
1372                                                   "multiple tunnel "
1373                                                   "not supported");
1374         }
1375         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1376                 return rte_flow_error_set(error, EINVAL,
1377                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1378                                           "wrong tunnel type - IPv6 specified "
1379                                           "but IPv4 item provided");
1380         if (item_flags & l3m)
1381                 return rte_flow_error_set(error, ENOTSUP,
1382                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1383                                           "multiple L3 layers not supported");
1384         else if (item_flags & l4m)
1385                 return rte_flow_error_set(error, EINVAL,
1386                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1387                                           "L3 cannot follow an L4 layer.");
1388         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1389                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1390                 return rte_flow_error_set(error, EINVAL,
1391                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1392                                           "L3 cannot follow an NVGRE layer.");
1393         if (!mask)
1394                 mask = &rte_flow_item_ipv4_mask;
1395         else if (mask->hdr.next_proto_id != 0 &&
1396                  mask->hdr.next_proto_id != 0xff)
1397                 return rte_flow_error_set(error, EINVAL,
1398                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1399                                           "partial mask is not supported"
1400                                           " for protocol");
1401         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1402                                         acc_mask ? (const uint8_t *)acc_mask
1403                                                  : (const uint8_t *)&nic_mask,
1404                                         sizeof(struct rte_flow_item_ipv4),
1405                                         error);
1406         if (ret < 0)
1407                 return ret;
1408         return 0;
1409 }
1410
1411 /**
1412  * Validate IPV6 item.
1413  *
1414  * @param[in] item
1415  *   Item specification.
1416  * @param[in] item_flags
1417  *   Bit-fields that holds the items detected until now.
1418  * @param[in] acc_mask
1419  *   Acceptable mask, if NULL default internal default mask
1420  *   will be used to check whether item fields are supported.
1421  * @param[out] error
1422  *   Pointer to error structure.
1423  *
1424  * @return
1425  *   0 on success, a negative errno value otherwise and rte_errno is set.
1426  */
1427 int
1428 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1429                              uint64_t item_flags,
1430                              const struct rte_flow_item_ipv6 *acc_mask,
1431                              struct rte_flow_error *error)
1432 {
1433         const struct rte_flow_item_ipv6 *mask = item->mask;
1434         const struct rte_flow_item_ipv6 *spec = item->spec;
1435         const struct rte_flow_item_ipv6 nic_mask = {
1436                 .hdr = {
1437                         .src_addr =
1438                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1439                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1440                         .dst_addr =
1441                                 "\xff\xff\xff\xff\xff\xff\xff\xff"
1442                                 "\xff\xff\xff\xff\xff\xff\xff\xff",
1443                         .vtc_flow = RTE_BE32(0xffffffff),
1444                         .proto = 0xff,
1445                         .hop_limits = 0xff,
1446                 },
1447         };
1448         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1449         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1450                                       MLX5_FLOW_LAYER_OUTER_L3;
1451         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1452                                       MLX5_FLOW_LAYER_OUTER_L4;
1453         int ret;
1454         uint8_t next_proto = 0xFF;
1455
1456         if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1457                 if (mask && spec)
1458                         next_proto = mask->hdr.proto & spec->hdr.proto;
1459                 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1460                         return rte_flow_error_set(error, EINVAL,
1461                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1462                                                   item,
1463                                                   "multiple tunnel "
1464                                                   "not supported");
1465         }
1466         if (item_flags & MLX5_FLOW_LAYER_IPIP)
1467                 return rte_flow_error_set(error, EINVAL,
1468                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1469                                           "wrong tunnel type - IPv4 specified "
1470                                           "but IPv6 item provided");
1471         if (item_flags & l3m)
1472                 return rte_flow_error_set(error, ENOTSUP,
1473                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1474                                           "multiple L3 layers not supported");
1475         else if (item_flags & l4m)
1476                 return rte_flow_error_set(error, EINVAL,
1477                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1478                                           "L3 cannot follow an L4 layer.");
1479         else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1480                   !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1481                 return rte_flow_error_set(error, EINVAL,
1482                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1483                                           "L3 cannot follow an NVGRE layer.");
1484         if (!mask)
1485                 mask = &rte_flow_item_ipv6_mask;
1486         ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1487                                         acc_mask ? (const uint8_t *)acc_mask
1488                                                  : (const uint8_t *)&nic_mask,
1489                                         sizeof(struct rte_flow_item_ipv6),
1490                                         error);
1491         if (ret < 0)
1492                 return ret;
1493         return 0;
1494 }
1495
1496 /**
1497  * Validate UDP item.
1498  *
1499  * @param[in] item
1500  *   Item specification.
1501  * @param[in] item_flags
1502  *   Bit-fields that holds the items detected until now.
1503  * @param[in] target_protocol
1504  *   The next protocol in the previous item.
1505  * @param[in] flow_mask
1506  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1507  * @param[out] error
1508  *   Pointer to error structure.
1509  *
1510  * @return
1511  *   0 on success, a negative errno value otherwise and rte_errno is set.
1512  */
1513 int
1514 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1515                             uint64_t item_flags,
1516                             uint8_t target_protocol,
1517                             struct rte_flow_error *error)
1518 {
1519         const struct rte_flow_item_udp *mask = item->mask;
1520         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1521         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1522                                       MLX5_FLOW_LAYER_OUTER_L3;
1523         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1524                                       MLX5_FLOW_LAYER_OUTER_L4;
1525         int ret;
1526
1527         if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1528                 return rte_flow_error_set(error, EINVAL,
1529                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1530                                           "protocol filtering not compatible"
1531                                           " with UDP layer");
1532         if (!(item_flags & l3m))
1533                 return rte_flow_error_set(error, EINVAL,
1534                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1535                                           "L3 is mandatory to filter on L4");
1536         if (item_flags & l4m)
1537                 return rte_flow_error_set(error, EINVAL,
1538                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1539                                           "multiple L4 layers not supported");
1540         if (!mask)
1541                 mask = &rte_flow_item_udp_mask;
1542         ret = mlx5_flow_item_acceptable
1543                 (item, (const uint8_t *)mask,
1544                  (const uint8_t *)&rte_flow_item_udp_mask,
1545                  sizeof(struct rte_flow_item_udp), error);
1546         if (ret < 0)
1547                 return ret;
1548         return 0;
1549 }
1550
1551 /**
1552  * Validate TCP item.
1553  *
1554  * @param[in] item
1555  *   Item specification.
1556  * @param[in] item_flags
1557  *   Bit-fields that holds the items detected until now.
1558  * @param[in] target_protocol
1559  *   The next protocol in the previous item.
1560  * @param[out] error
1561  *   Pointer to error structure.
1562  *
1563  * @return
1564  *   0 on success, a negative errno value otherwise and rte_errno is set.
1565  */
1566 int
1567 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1568                             uint64_t item_flags,
1569                             uint8_t target_protocol,
1570                             const struct rte_flow_item_tcp *flow_mask,
1571                             struct rte_flow_error *error)
1572 {
1573         const struct rte_flow_item_tcp *mask = item->mask;
1574         const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1575         const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1576                                       MLX5_FLOW_LAYER_OUTER_L3;
1577         const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1578                                       MLX5_FLOW_LAYER_OUTER_L4;
1579         int ret;
1580
1581         assert(flow_mask);
1582         if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1583                 return rte_flow_error_set(error, EINVAL,
1584                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1585                                           "protocol filtering not compatible"
1586                                           " with TCP layer");
1587         if (!(item_flags & l3m))
1588                 return rte_flow_error_set(error, EINVAL,
1589                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1590                                           "L3 is mandatory to filter on L4");
1591         if (item_flags & l4m)
1592                 return rte_flow_error_set(error, EINVAL,
1593                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1594                                           "multiple L4 layers not supported");
1595         if (!mask)
1596                 mask = &rte_flow_item_tcp_mask;
1597         ret = mlx5_flow_item_acceptable
1598                 (item, (const uint8_t *)mask,
1599                  (const uint8_t *)flow_mask,
1600                  sizeof(struct rte_flow_item_tcp), error);
1601         if (ret < 0)
1602                 return ret;
1603         return 0;
1604 }
1605
1606 /**
1607  * Validate VXLAN item.
1608  *
1609  * @param[in] item
1610  *   Item specification.
1611  * @param[in] item_flags
1612  *   Bit-fields that holds the items detected until now.
1613  * @param[in] target_protocol
1614  *   The next protocol in the previous item.
1615  * @param[out] error
1616  *   Pointer to error structure.
1617  *
1618  * @return
1619  *   0 on success, a negative errno value otherwise and rte_errno is set.
1620  */
1621 int
1622 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1623                               uint64_t item_flags,
1624                               struct rte_flow_error *error)
1625 {
1626         const struct rte_flow_item_vxlan *spec = item->spec;
1627         const struct rte_flow_item_vxlan *mask = item->mask;
1628         int ret;
1629         union vni {
1630                 uint32_t vlan_id;
1631                 uint8_t vni[4];
1632         } id = { .vlan_id = 0, };
1633         uint32_t vlan_id = 0;
1634
1635
1636         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1637                 return rte_flow_error_set(error, ENOTSUP,
1638                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1639                                           "multiple tunnel layers not"
1640                                           " supported");
1641         /*
1642          * Verify only UDPv4 is present as defined in
1643          * https://tools.ietf.org/html/rfc7348
1644          */
1645         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1646                 return rte_flow_error_set(error, EINVAL,
1647                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1648                                           "no outer UDP layer found");
1649         if (!mask)
1650                 mask = &rte_flow_item_vxlan_mask;
1651         ret = mlx5_flow_item_acceptable
1652                 (item, (const uint8_t *)mask,
1653                  (const uint8_t *)&rte_flow_item_vxlan_mask,
1654                  sizeof(struct rte_flow_item_vxlan),
1655                  error);
1656         if (ret < 0)
1657                 return ret;
1658         if (spec) {
1659                 memcpy(&id.vni[1], spec->vni, 3);
1660                 vlan_id = id.vlan_id;
1661                 memcpy(&id.vni[1], mask->vni, 3);
1662                 vlan_id &= id.vlan_id;
1663         }
1664         /*
1665          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1666          * only this layer is defined in the Verbs specification it is
1667          * interpreted as wildcard and all packets will match this
1668          * rule, if it follows a full stack layer (ex: eth / ipv4 /
1669          * udp), all packets matching the layers before will also
1670          * match this rule.  To avoid such situation, VNI 0 is
1671          * currently refused.
1672          */
1673         if (!vlan_id)
1674                 return rte_flow_error_set(error, ENOTSUP,
1675                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1676                                           "VXLAN vni cannot be 0");
1677         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1678                 return rte_flow_error_set(error, ENOTSUP,
1679                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1680                                           "VXLAN tunnel must be fully defined");
1681         return 0;
1682 }
1683
1684 /**
1685  * Validate VXLAN_GPE item.
1686  *
1687  * @param[in] item
1688  *   Item specification.
1689  * @param[in] item_flags
1690  *   Bit-fields that holds the items detected until now.
1691  * @param[in] priv
1692  *   Pointer to the private data structure.
1693  * @param[in] target_protocol
1694  *   The next protocol in the previous item.
1695  * @param[out] error
1696  *   Pointer to error structure.
1697  *
1698  * @return
1699  *   0 on success, a negative errno value otherwise and rte_errno is set.
1700  */
1701 int
1702 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1703                                   uint64_t item_flags,
1704                                   struct rte_eth_dev *dev,
1705                                   struct rte_flow_error *error)
1706 {
1707         struct mlx5_priv *priv = dev->data->dev_private;
1708         const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1709         const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1710         int ret;
1711         union vni {
1712                 uint32_t vlan_id;
1713                 uint8_t vni[4];
1714         } id = { .vlan_id = 0, };
1715         uint32_t vlan_id = 0;
1716
1717         if (!priv->config.l3_vxlan_en)
1718                 return rte_flow_error_set(error, ENOTSUP,
1719                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1720                                           "L3 VXLAN is not enabled by device"
1721                                           " parameter and/or not configured in"
1722                                           " firmware");
1723         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1724                 return rte_flow_error_set(error, ENOTSUP,
1725                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1726                                           "multiple tunnel layers not"
1727                                           " supported");
1728         /*
1729          * Verify only UDPv4 is present as defined in
1730          * https://tools.ietf.org/html/rfc7348
1731          */
1732         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1733                 return rte_flow_error_set(error, EINVAL,
1734                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1735                                           "no outer UDP layer found");
1736         if (!mask)
1737                 mask = &rte_flow_item_vxlan_gpe_mask;
1738         ret = mlx5_flow_item_acceptable
1739                 (item, (const uint8_t *)mask,
1740                  (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1741                  sizeof(struct rte_flow_item_vxlan_gpe),
1742                  error);
1743         if (ret < 0)
1744                 return ret;
1745         if (spec) {
1746                 if (spec->protocol)
1747                         return rte_flow_error_set(error, ENOTSUP,
1748                                                   RTE_FLOW_ERROR_TYPE_ITEM,
1749                                                   item,
1750                                                   "VxLAN-GPE protocol"
1751                                                   " not supported");
1752                 memcpy(&id.vni[1], spec->vni, 3);
1753                 vlan_id = id.vlan_id;
1754                 memcpy(&id.vni[1], mask->vni, 3);
1755                 vlan_id &= id.vlan_id;
1756         }
1757         /*
1758          * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1759          * layer is defined in the Verbs specification it is interpreted as
1760          * wildcard and all packets will match this rule, if it follows a full
1761          * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1762          * before will also match this rule.  To avoid such situation, VNI 0
1763          * is currently refused.
1764          */
1765         if (!vlan_id)
1766                 return rte_flow_error_set(error, ENOTSUP,
1767                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1768                                           "VXLAN-GPE vni cannot be 0");
1769         if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1770                 return rte_flow_error_set(error, ENOTSUP,
1771                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1772                                           "VXLAN-GPE tunnel must be fully"
1773                                           " defined");
1774         return 0;
1775 }
1776 /**
1777  * Validate GRE Key item.
1778  *
1779  * @param[in] item
1780  *   Item specification.
1781  * @param[in] item_flags
1782  *   Bit flags to mark detected items.
1783  * @param[in] gre_item
1784  *   Pointer to gre_item
1785  * @param[out] error
1786  *   Pointer to error structure.
1787  *
1788  * @return
1789  *   0 on success, a negative errno value otherwise and rte_errno is set.
1790  */
1791 int
1792 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1793                                 uint64_t item_flags,
1794                                 const struct rte_flow_item *gre_item,
1795                                 struct rte_flow_error *error)
1796 {
1797         const rte_be32_t *mask = item->mask;
1798         int ret = 0;
1799         rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1800         const struct rte_flow_item_gre *gre_spec = gre_item->spec;
1801         const struct rte_flow_item_gre *gre_mask = gre_item->mask;
1802
1803         if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1804                 return rte_flow_error_set(error, ENOTSUP,
1805                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1806                                           "Multiple GRE key not support");
1807         if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1808                 return rte_flow_error_set(error, ENOTSUP,
1809                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1810                                           "No preceding GRE header");
1811         if (item_flags & MLX5_FLOW_LAYER_INNER)
1812                 return rte_flow_error_set(error, ENOTSUP,
1813                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1814                                           "GRE key following a wrong item");
1815         if (!gre_mask)
1816                 gre_mask = &rte_flow_item_gre_mask;
1817         if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
1818                          !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
1819                 return rte_flow_error_set(error, EINVAL,
1820                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1821                                           "Key bit must be on");
1822
1823         if (!mask)
1824                 mask = &gre_key_default_mask;
1825         ret = mlx5_flow_item_acceptable
1826                 (item, (const uint8_t *)mask,
1827                  (const uint8_t *)&gre_key_default_mask,
1828                  sizeof(rte_be32_t), error);
1829         return ret;
1830 }
1831
1832 /**
1833  * Validate GRE item.
1834  *
1835  * @param[in] item
1836  *   Item specification.
1837  * @param[in] item_flags
1838  *   Bit flags to mark detected items.
1839  * @param[in] target_protocol
1840  *   The next protocol in the previous item.
1841  * @param[out] error
1842  *   Pointer to error structure.
1843  *
1844  * @return
1845  *   0 on success, a negative errno value otherwise and rte_errno is set.
1846  */
1847 int
1848 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1849                             uint64_t item_flags,
1850                             uint8_t target_protocol,
1851                             struct rte_flow_error *error)
1852 {
1853         const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1854         const struct rte_flow_item_gre *mask = item->mask;
1855         int ret;
1856         const struct rte_flow_item_gre nic_mask = {
1857                 .c_rsvd0_ver = RTE_BE16(0xB000),
1858                 .protocol = RTE_BE16(UINT16_MAX),
1859         };
1860
1861         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1862                 return rte_flow_error_set(error, EINVAL,
1863                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1864                                           "protocol filtering not compatible"
1865                                           " with this GRE layer");
1866         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1867                 return rte_flow_error_set(error, ENOTSUP,
1868                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1869                                           "multiple tunnel layers not"
1870                                           " supported");
1871         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1872                 return rte_flow_error_set(error, ENOTSUP,
1873                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1874                                           "L3 Layer is missing");
1875         if (!mask)
1876                 mask = &rte_flow_item_gre_mask;
1877         ret = mlx5_flow_item_acceptable
1878                 (item, (const uint8_t *)mask,
1879                  (const uint8_t *)&nic_mask,
1880                  sizeof(struct rte_flow_item_gre), error);
1881         if (ret < 0)
1882                 return ret;
1883 #ifndef HAVE_MLX5DV_DR
1884 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1885         if (spec && (spec->protocol & mask->protocol))
1886                 return rte_flow_error_set(error, ENOTSUP,
1887                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1888                                           "without MPLS support the"
1889                                           " specification cannot be used for"
1890                                           " filtering");
1891 #endif
1892 #endif
1893         return 0;
1894 }
1895
1896 /**
1897  * Validate MPLS item.
1898  *
1899  * @param[in] dev
1900  *   Pointer to the rte_eth_dev structure.
1901  * @param[in] item
1902  *   Item specification.
1903  * @param[in] item_flags
1904  *   Bit-fields that holds the items detected until now.
1905  * @param[in] prev_layer
1906  *   The protocol layer indicated in previous item.
1907  * @param[out] error
1908  *   Pointer to error structure.
1909  *
1910  * @return
1911  *   0 on success, a negative errno value otherwise and rte_errno is set.
1912  */
1913 int
1914 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
1915                              const struct rte_flow_item *item __rte_unused,
1916                              uint64_t item_flags __rte_unused,
1917                              uint64_t prev_layer __rte_unused,
1918                              struct rte_flow_error *error)
1919 {
1920 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1921         const struct rte_flow_item_mpls *mask = item->mask;
1922         struct mlx5_priv *priv = dev->data->dev_private;
1923         int ret;
1924
1925         if (!priv->config.mpls_en)
1926                 return rte_flow_error_set(error, ENOTSUP,
1927                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1928                                           "MPLS not supported or"
1929                                           " disabled in firmware"
1930                                           " configuration.");
1931         /* MPLS over IP, UDP, GRE is allowed */
1932         if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
1933                             MLX5_FLOW_LAYER_OUTER_L4_UDP |
1934                             MLX5_FLOW_LAYER_GRE)))
1935                 return rte_flow_error_set(error, EINVAL,
1936                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1937                                           "protocol filtering not compatible"
1938                                           " with MPLS layer");
1939         /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
1940         if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1941             !(item_flags & MLX5_FLOW_LAYER_GRE))
1942                 return rte_flow_error_set(error, ENOTSUP,
1943                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1944                                           "multiple tunnel layers not"
1945                                           " supported");
1946         if (!mask)
1947                 mask = &rte_flow_item_mpls_mask;
1948         ret = mlx5_flow_item_acceptable
1949                 (item, (const uint8_t *)mask,
1950                  (const uint8_t *)&rte_flow_item_mpls_mask,
1951                  sizeof(struct rte_flow_item_mpls), error);
1952         if (ret < 0)
1953                 return ret;
1954         return 0;
1955 #endif
1956         return rte_flow_error_set(error, ENOTSUP,
1957                                   RTE_FLOW_ERROR_TYPE_ITEM, item,
1958                                   "MPLS is not supported by Verbs, please"
1959                                   " update.");
1960 }
1961
1962 /**
1963  * Validate NVGRE item.
1964  *
1965  * @param[in] item
1966  *   Item specification.
1967  * @param[in] item_flags
1968  *   Bit flags to mark detected items.
1969  * @param[in] target_protocol
1970  *   The next protocol in the previous item.
1971  * @param[out] error
1972  *   Pointer to error structure.
1973  *
1974  * @return
1975  *   0 on success, a negative errno value otherwise and rte_errno is set.
1976  */
1977 int
1978 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
1979                               uint64_t item_flags,
1980                               uint8_t target_protocol,
1981                               struct rte_flow_error *error)
1982 {
1983         const struct rte_flow_item_nvgre *mask = item->mask;
1984         int ret;
1985
1986         if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1987                 return rte_flow_error_set(error, EINVAL,
1988                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1989                                           "protocol filtering not compatible"
1990                                           " with this GRE layer");
1991         if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1992                 return rte_flow_error_set(error, ENOTSUP,
1993                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1994                                           "multiple tunnel layers not"
1995                                           " supported");
1996         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1997                 return rte_flow_error_set(error, ENOTSUP,
1998                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1999                                           "L3 Layer is missing");
2000         if (!mask)
2001                 mask = &rte_flow_item_nvgre_mask;
2002         ret = mlx5_flow_item_acceptable
2003                 (item, (const uint8_t *)mask,
2004                  (const uint8_t *)&rte_flow_item_nvgre_mask,
2005                  sizeof(struct rte_flow_item_nvgre), error);
2006         if (ret < 0)
2007                 return ret;
2008         return 0;
2009 }
2010
2011 static int
2012 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2013                    const struct rte_flow_attr *attr __rte_unused,
2014                    const struct rte_flow_item items[] __rte_unused,
2015                    const struct rte_flow_action actions[] __rte_unused,
2016                    bool external __rte_unused,
2017                    struct rte_flow_error *error)
2018 {
2019         return rte_flow_error_set(error, ENOTSUP,
2020                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2021 }
2022
2023 static struct mlx5_flow *
2024 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2025                   const struct rte_flow_item items[] __rte_unused,
2026                   const struct rte_flow_action actions[] __rte_unused,
2027                   struct rte_flow_error *error)
2028 {
2029         rte_flow_error_set(error, ENOTSUP,
2030                            RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2031         return NULL;
2032 }
2033
2034 static int
2035 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2036                     struct mlx5_flow *dev_flow __rte_unused,
2037                     const struct rte_flow_attr *attr __rte_unused,
2038                     const struct rte_flow_item items[] __rte_unused,
2039                     const struct rte_flow_action actions[] __rte_unused,
2040                     struct rte_flow_error *error)
2041 {
2042         return rte_flow_error_set(error, ENOTSUP,
2043                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2044 }
2045
2046 static int
2047 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2048                 struct rte_flow *flow __rte_unused,
2049                 struct rte_flow_error *error)
2050 {
2051         return rte_flow_error_set(error, ENOTSUP,
2052                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2053 }
2054
2055 static void
2056 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2057                  struct rte_flow *flow __rte_unused)
2058 {
2059 }
2060
2061 static void
2062 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2063                   struct rte_flow *flow __rte_unused)
2064 {
2065 }
2066
2067 static int
2068 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2069                 struct rte_flow *flow __rte_unused,
2070                 const struct rte_flow_action *actions __rte_unused,
2071                 void *data __rte_unused,
2072                 struct rte_flow_error *error)
2073 {
2074         return rte_flow_error_set(error, ENOTSUP,
2075                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2076 }
2077
2078 /* Void driver to protect from null pointer reference. */
2079 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2080         .validate = flow_null_validate,
2081         .prepare = flow_null_prepare,
2082         .translate = flow_null_translate,
2083         .apply = flow_null_apply,
2084         .remove = flow_null_remove,
2085         .destroy = flow_null_destroy,
2086         .query = flow_null_query,
2087 };
2088
2089 /**
2090  * Select flow driver type according to flow attributes and device
2091  * configuration.
2092  *
2093  * @param[in] dev
2094  *   Pointer to the dev structure.
2095  * @param[in] attr
2096  *   Pointer to the flow attributes.
2097  *
2098  * @return
2099  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2100  */
2101 static enum mlx5_flow_drv_type
2102 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2103 {
2104         struct mlx5_priv *priv = dev->data->dev_private;
2105         enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2106
2107         if (attr->transfer && priv->config.dv_esw_en)
2108                 type = MLX5_FLOW_TYPE_DV;
2109         if (!attr->transfer)
2110                 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2111                                                  MLX5_FLOW_TYPE_VERBS;
2112         return type;
2113 }
2114
2115 #define flow_get_drv_ops(type) flow_drv_ops[type]
2116
2117 /**
2118  * Flow driver validation API. This abstracts calling driver specific functions.
2119  * The type of flow driver is determined according to flow attributes.
2120  *
2121  * @param[in] dev
2122  *   Pointer to the dev structure.
2123  * @param[in] attr
2124  *   Pointer to the flow attributes.
2125  * @param[in] items
2126  *   Pointer to the list of items.
2127  * @param[in] actions
2128  *   Pointer to the list of actions.
2129  * @param[in] external
2130  *   This flow rule is created by request external to PMD.
2131  * @param[out] error
2132  *   Pointer to the error structure.
2133  *
2134  * @return
2135  *   0 on success, a negative errno value otherwise and rte_errno is set.
2136  */
2137 static inline int
2138 flow_drv_validate(struct rte_eth_dev *dev,
2139                   const struct rte_flow_attr *attr,
2140                   const struct rte_flow_item items[],
2141                   const struct rte_flow_action actions[],
2142                   bool external, struct rte_flow_error *error)
2143 {
2144         const struct mlx5_flow_driver_ops *fops;
2145         enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2146
2147         fops = flow_get_drv_ops(type);
2148         return fops->validate(dev, attr, items, actions, external, error);
2149 }
2150
2151 /**
2152  * Flow driver preparation API. This abstracts calling driver specific
2153  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2154  * calculates the size of memory required for device flow, allocates the memory,
2155  * initializes the device flow and returns the pointer.
2156  *
2157  * @note
2158  *   This function initializes device flow structure such as dv or verbs in
2159  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2160  *   rest. For example, adding returning device flow to flow->dev_flow list and
2161  *   setting backward reference to the flow should be done out of this function.
2162  *   layers field is not filled either.
2163  *
2164  * @param[in] attr
2165  *   Pointer to the flow attributes.
2166  * @param[in] items
2167  *   Pointer to the list of items.
2168  * @param[in] actions
2169  *   Pointer to the list of actions.
2170  * @param[out] error
2171  *   Pointer to the error structure.
2172  *
2173  * @return
2174  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2175  */
2176 static inline struct mlx5_flow *
2177 flow_drv_prepare(const struct rte_flow *flow,
2178                  const struct rte_flow_attr *attr,
2179                  const struct rte_flow_item items[],
2180                  const struct rte_flow_action actions[],
2181                  struct rte_flow_error *error)
2182 {
2183         const struct mlx5_flow_driver_ops *fops;
2184         enum mlx5_flow_drv_type type = flow->drv_type;
2185
2186         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2187         fops = flow_get_drv_ops(type);
2188         return fops->prepare(attr, items, actions, error);
2189 }
2190
2191 /**
2192  * Flow driver translation API. This abstracts calling driver specific
2193  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2194  * translates a generic flow into a driver flow. flow_drv_prepare() must
2195  * precede.
2196  *
2197  * @note
2198  *   dev_flow->layers could be filled as a result of parsing during translation
2199  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2200  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2201  *   flow->actions could be overwritten even though all the expanded dev_flows
2202  *   have the same actions.
2203  *
2204  * @param[in] dev
2205  *   Pointer to the rte dev structure.
2206  * @param[in, out] dev_flow
2207  *   Pointer to the mlx5 flow.
2208  * @param[in] attr
2209  *   Pointer to the flow attributes.
2210  * @param[in] items
2211  *   Pointer to the list of items.
2212  * @param[in] actions
2213  *   Pointer to the list of actions.
2214  * @param[out] error
2215  *   Pointer to the error structure.
2216  *
2217  * @return
2218  *   0 on success, a negative errno value otherwise and rte_errno is set.
2219  */
2220 static inline int
2221 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2222                    const struct rte_flow_attr *attr,
2223                    const struct rte_flow_item items[],
2224                    const struct rte_flow_action actions[],
2225                    struct rte_flow_error *error)
2226 {
2227         const struct mlx5_flow_driver_ops *fops;
2228         enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2229
2230         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2231         fops = flow_get_drv_ops(type);
2232         return fops->translate(dev, dev_flow, attr, items, actions, error);
2233 }
2234
2235 /**
2236  * Flow driver apply API. This abstracts calling driver specific functions.
2237  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2238  * translated driver flows on to device. flow_drv_translate() must precede.
2239  *
2240  * @param[in] dev
2241  *   Pointer to Ethernet device structure.
2242  * @param[in, out] flow
2243  *   Pointer to flow structure.
2244  * @param[out] error
2245  *   Pointer to error structure.
2246  *
2247  * @return
2248  *   0 on success, a negative errno value otherwise and rte_errno is set.
2249  */
2250 static inline int
2251 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2252                struct rte_flow_error *error)
2253 {
2254         const struct mlx5_flow_driver_ops *fops;
2255         enum mlx5_flow_drv_type type = flow->drv_type;
2256
2257         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2258         fops = flow_get_drv_ops(type);
2259         return fops->apply(dev, flow, error);
2260 }
2261
2262 /**
2263  * Flow driver remove API. This abstracts calling driver specific functions.
2264  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2265  * on device. All the resources of the flow should be freed by calling
2266  * flow_drv_destroy().
2267  *
2268  * @param[in] dev
2269  *   Pointer to Ethernet device.
2270  * @param[in, out] flow
2271  *   Pointer to flow structure.
2272  */
2273 static inline void
2274 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2275 {
2276         const struct mlx5_flow_driver_ops *fops;
2277         enum mlx5_flow_drv_type type = flow->drv_type;
2278
2279         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2280         fops = flow_get_drv_ops(type);
2281         fops->remove(dev, flow);
2282 }
2283
2284 /**
2285  * Flow driver destroy API. This abstracts calling driver specific functions.
2286  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2287  * on device and releases resources of the flow.
2288  *
2289  * @param[in] dev
2290  *   Pointer to Ethernet device.
2291  * @param[in, out] flow
2292  *   Pointer to flow structure.
2293  */
2294 static inline void
2295 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2296 {
2297         const struct mlx5_flow_driver_ops *fops;
2298         enum mlx5_flow_drv_type type = flow->drv_type;
2299
2300         assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2301         fops = flow_get_drv_ops(type);
2302         fops->destroy(dev, flow);
2303 }
2304
2305 /**
2306  * Validate a flow supported by the NIC.
2307  *
2308  * @see rte_flow_validate()
2309  * @see rte_flow_ops
2310  */
2311 int
2312 mlx5_flow_validate(struct rte_eth_dev *dev,
2313                    const struct rte_flow_attr *attr,
2314                    const struct rte_flow_item items[],
2315                    const struct rte_flow_action actions[],
2316                    struct rte_flow_error *error)
2317 {
2318         int ret;
2319
2320         ret = flow_drv_validate(dev, attr, items, actions, true, error);
2321         if (ret < 0)
2322                 return ret;
2323         return 0;
2324 }
2325
2326 /**
2327  * Get RSS action from the action list.
2328  *
2329  * @param[in] actions
2330  *   Pointer to the list of actions.
2331  *
2332  * @return
2333  *   Pointer to the RSS action if exist, else return NULL.
2334  */
2335 static const struct rte_flow_action_rss*
2336 flow_get_rss_action(const struct rte_flow_action actions[])
2337 {
2338         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2339                 switch (actions->type) {
2340                 case RTE_FLOW_ACTION_TYPE_RSS:
2341                         return (const struct rte_flow_action_rss *)
2342                                actions->conf;
2343                 default:
2344                         break;
2345                 }
2346         }
2347         return NULL;
2348 }
2349
2350 static unsigned int
2351 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2352 {
2353         const struct rte_flow_item *item;
2354         unsigned int has_vlan = 0;
2355
2356         for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2357                 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2358                         has_vlan = 1;
2359                         break;
2360                 }
2361         }
2362         if (has_vlan)
2363                 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2364                                        MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2365         return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2366                                MLX5_EXPANSION_ROOT_OUTER;
2367 }
2368
2369 /**
2370  * Create a flow and add it to @p list.
2371  *
2372  * @param dev
2373  *   Pointer to Ethernet device.
2374  * @param list
2375  *   Pointer to a TAILQ flow list.
2376  * @param[in] attr
2377  *   Flow rule attributes.
2378  * @param[in] items
2379  *   Pattern specification (list terminated by the END pattern item).
2380  * @param[in] actions
2381  *   Associated actions (list terminated by the END action).
2382  * @param[in] external
2383  *   This flow rule is created by request external to PMD.
2384  * @param[out] error
2385  *   Perform verbose error reporting if not NULL.
2386  *
2387  * @return
2388  *   A flow on success, NULL otherwise and rte_errno is set.
2389  */
2390 static struct rte_flow *
2391 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2392                  const struct rte_flow_attr *attr,
2393                  const struct rte_flow_item items[],
2394                  const struct rte_flow_action actions[],
2395                  bool external, struct rte_flow_error *error)
2396 {
2397         struct rte_flow *flow = NULL;
2398         struct mlx5_flow *dev_flow;
2399         const struct rte_flow_action_rss *rss;
2400         union {
2401                 struct rte_flow_expand_rss buf;
2402                 uint8_t buffer[2048];
2403         } expand_buffer;
2404         struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2405         int ret;
2406         uint32_t i;
2407         uint32_t flow_size;
2408
2409         ret = flow_drv_validate(dev, attr, items, actions, external, error);
2410         if (ret < 0)
2411                 return NULL;
2412         flow_size = sizeof(struct rte_flow);
2413         rss = flow_get_rss_action(actions);
2414         if (rss)
2415                 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2416                                             sizeof(void *));
2417         else
2418                 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2419         flow = rte_calloc(__func__, 1, flow_size, 0);
2420         if (!flow) {
2421                 rte_errno = ENOMEM;
2422                 return NULL;
2423         }
2424         flow->drv_type = flow_get_drv_type(dev, attr);
2425         flow->ingress = attr->ingress;
2426         flow->transfer = attr->transfer;
2427         assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2428                flow->drv_type < MLX5_FLOW_TYPE_MAX);
2429         flow->queue = (void *)(flow + 1);
2430         LIST_INIT(&flow->dev_flows);
2431         if (rss && rss->types) {
2432                 unsigned int graph_root;
2433
2434                 graph_root = find_graph_root(items, rss->level);
2435                 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2436                                           items, rss->types,
2437                                           mlx5_support_expansion,
2438                                           graph_root);
2439                 assert(ret > 0 &&
2440                        (unsigned int)ret < sizeof(expand_buffer.buffer));
2441         } else {
2442                 buf->entries = 1;
2443                 buf->entry[0].pattern = (void *)(uintptr_t)items;
2444         }
2445         for (i = 0; i < buf->entries; ++i) {
2446                 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2447                                             actions, error);
2448                 if (!dev_flow)
2449                         goto error;
2450                 dev_flow->flow = flow;
2451                 dev_flow->external = external;
2452                 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2453                 ret = flow_drv_translate(dev, dev_flow, attr,
2454                                          buf->entry[i].pattern,
2455                                          actions, error);
2456                 if (ret < 0)
2457                         goto error;
2458         }
2459         if (dev->data->dev_started) {
2460                 ret = flow_drv_apply(dev, flow, error);
2461                 if (ret < 0)
2462                         goto error;
2463         }
2464         TAILQ_INSERT_TAIL(list, flow, next);
2465         flow_rxq_flags_set(dev, flow);
2466         return flow;
2467 error:
2468         ret = rte_errno; /* Save rte_errno before cleanup. */
2469         assert(flow);
2470         flow_drv_destroy(dev, flow);
2471         rte_free(flow);
2472         rte_errno = ret; /* Restore rte_errno. */
2473         return NULL;
2474 }
2475
2476 /**
2477  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
2478  * incoming packets to table 1.
2479  *
2480  * Other flow rules, requested for group n, will be created in
2481  * e-switch table n+1.
2482  * Jump action to e-switch group n will be created to group n+1.
2483  *
2484  * Used when working in switchdev mode, to utilise advantages of table 1
2485  * and above.
2486  *
2487  * @param dev
2488  *   Pointer to Ethernet device.
2489  *
2490  * @return
2491  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
2492  */
2493 struct rte_flow *
2494 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
2495 {
2496         const struct rte_flow_attr attr = {
2497                 .group = 0,
2498                 .priority = 0,
2499                 .ingress = 1,
2500                 .egress = 0,
2501                 .transfer = 1,
2502         };
2503         const struct rte_flow_item pattern = {
2504                 .type = RTE_FLOW_ITEM_TYPE_END,
2505         };
2506         struct rte_flow_action_jump jump = {
2507                 .group = 1,
2508         };
2509         const struct rte_flow_action actions[] = {
2510                 {
2511                         .type = RTE_FLOW_ACTION_TYPE_JUMP,
2512                         .conf = &jump,
2513                 },
2514                 {
2515                         .type = RTE_FLOW_ACTION_TYPE_END,
2516                 },
2517         };
2518         struct mlx5_priv *priv = dev->data->dev_private;
2519         struct rte_flow_error error;
2520
2521         return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
2522                                 actions, false, &error);
2523 }
2524
2525 /**
2526  * Create a flow.
2527  *
2528  * @see rte_flow_create()
2529  * @see rte_flow_ops
2530  */
2531 struct rte_flow *
2532 mlx5_flow_create(struct rte_eth_dev *dev,
2533                  const struct rte_flow_attr *attr,
2534                  const struct rte_flow_item items[],
2535                  const struct rte_flow_action actions[],
2536                  struct rte_flow_error *error)
2537 {
2538         struct mlx5_priv *priv = dev->data->dev_private;
2539
2540         return flow_list_create(dev, &priv->flows,
2541                                 attr, items, actions, true, error);
2542 }
2543
2544 /**
2545  * Destroy a flow in a list.
2546  *
2547  * @param dev
2548  *   Pointer to Ethernet device.
2549  * @param list
2550  *   Pointer to a TAILQ flow list.
2551  * @param[in] flow
2552  *   Flow to destroy.
2553  */
2554 static void
2555 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2556                   struct rte_flow *flow)
2557 {
2558         /*
2559          * Update RX queue flags only if port is started, otherwise it is
2560          * already clean.
2561          */
2562         if (dev->data->dev_started)
2563                 flow_rxq_flags_trim(dev, flow);
2564         flow_drv_destroy(dev, flow);
2565         TAILQ_REMOVE(list, flow, next);
2566         rte_free(flow->fdir);
2567         rte_free(flow);
2568 }
2569
2570 /**
2571  * Destroy all flows.
2572  *
2573  * @param dev
2574  *   Pointer to Ethernet device.
2575  * @param list
2576  *   Pointer to a TAILQ flow list.
2577  */
2578 void
2579 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2580 {
2581         while (!TAILQ_EMPTY(list)) {
2582                 struct rte_flow *flow;
2583
2584                 flow = TAILQ_FIRST(list);
2585                 flow_list_destroy(dev, list, flow);
2586         }
2587 }
2588
2589 /**
2590  * Remove all flows.
2591  *
2592  * @param dev
2593  *   Pointer to Ethernet device.
2594  * @param list
2595  *   Pointer to a TAILQ flow list.
2596  */
2597 void
2598 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2599 {
2600         struct rte_flow *flow;
2601
2602         TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2603                 flow_drv_remove(dev, flow);
2604         flow_rxq_flags_clear(dev);
2605 }
2606
2607 /**
2608  * Add all flows.
2609  *
2610  * @param dev
2611  *   Pointer to Ethernet device.
2612  * @param list
2613  *   Pointer to a TAILQ flow list.
2614  *
2615  * @return
2616  *   0 on success, a negative errno value otherwise and rte_errno is set.
2617  */
2618 int
2619 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2620 {
2621         struct rte_flow *flow;
2622         struct rte_flow_error error;
2623         int ret = 0;
2624
2625         TAILQ_FOREACH(flow, list, next) {
2626                 ret = flow_drv_apply(dev, flow, &error);
2627                 if (ret < 0)
2628                         goto error;
2629                 flow_rxq_flags_set(dev, flow);
2630         }
2631         return 0;
2632 error:
2633         ret = rte_errno; /* Save rte_errno before cleanup. */
2634         mlx5_flow_stop(dev, list);
2635         rte_errno = ret; /* Restore rte_errno. */
2636         return -rte_errno;
2637 }
2638
2639 /**
2640  * Verify the flow list is empty
2641  *
2642  * @param dev
2643  *  Pointer to Ethernet device.
2644  *
2645  * @return the number of flows not released.
2646  */
2647 int
2648 mlx5_flow_verify(struct rte_eth_dev *dev)
2649 {
2650         struct mlx5_priv *priv = dev->data->dev_private;
2651         struct rte_flow *flow;
2652         int ret = 0;
2653
2654         TAILQ_FOREACH(flow, &priv->flows, next) {
2655                 DRV_LOG(DEBUG, "port %u flow %p still referenced",
2656                         dev->data->port_id, (void *)flow);
2657                 ++ret;
2658         }
2659         return ret;
2660 }
2661
2662 /**
2663  * Enable a control flow configured from the control plane.
2664  *
2665  * @param dev
2666  *   Pointer to Ethernet device.
2667  * @param eth_spec
2668  *   An Ethernet flow spec to apply.
2669  * @param eth_mask
2670  *   An Ethernet flow mask to apply.
2671  * @param vlan_spec
2672  *   A VLAN flow spec to apply.
2673  * @param vlan_mask
2674  *   A VLAN flow mask to apply.
2675  *
2676  * @return
2677  *   0 on success, a negative errno value otherwise and rte_errno is set.
2678  */
2679 int
2680 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2681                     struct rte_flow_item_eth *eth_spec,
2682                     struct rte_flow_item_eth *eth_mask,
2683                     struct rte_flow_item_vlan *vlan_spec,
2684                     struct rte_flow_item_vlan *vlan_mask)
2685 {
2686         struct mlx5_priv *priv = dev->data->dev_private;
2687         const struct rte_flow_attr attr = {
2688                 .ingress = 1,
2689                 .priority = MLX5_FLOW_PRIO_RSVD,
2690         };
2691         struct rte_flow_item items[] = {
2692                 {
2693                         .type = RTE_FLOW_ITEM_TYPE_ETH,
2694                         .spec = eth_spec,
2695                         .last = NULL,
2696                         .mask = eth_mask,
2697                 },
2698                 {
2699                         .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2700                                               RTE_FLOW_ITEM_TYPE_END,
2701                         .spec = vlan_spec,
2702                         .last = NULL,
2703                         .mask = vlan_mask,
2704                 },
2705                 {
2706                         .type = RTE_FLOW_ITEM_TYPE_END,
2707                 },
2708         };
2709         uint16_t queue[priv->reta_idx_n];
2710         struct rte_flow_action_rss action_rss = {
2711                 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2712                 .level = 0,
2713                 .types = priv->rss_conf.rss_hf,
2714                 .key_len = priv->rss_conf.rss_key_len,
2715                 .queue_num = priv->reta_idx_n,
2716                 .key = priv->rss_conf.rss_key,
2717                 .queue = queue,
2718         };
2719         struct rte_flow_action actions[] = {
2720                 {
2721                         .type = RTE_FLOW_ACTION_TYPE_RSS,
2722                         .conf = &action_rss,
2723                 },
2724                 {
2725                         .type = RTE_FLOW_ACTION_TYPE_END,
2726                 },
2727         };
2728         struct rte_flow *flow;
2729         struct rte_flow_error error;
2730         unsigned int i;
2731
2732         if (!priv->reta_idx_n || !priv->rxqs_n) {
2733                 return 0;
2734         }
2735         for (i = 0; i != priv->reta_idx_n; ++i)
2736                 queue[i] = (*priv->reta_idx)[i];
2737         flow = flow_list_create(dev, &priv->ctrl_flows,
2738                                 &attr, items, actions, false, &error);
2739         if (!flow)
2740                 return -rte_errno;
2741         return 0;
2742 }
2743
2744 /**
2745  * Enable a flow control configured from the control plane.
2746  *
2747  * @param dev
2748  *   Pointer to Ethernet device.
2749  * @param eth_spec
2750  *   An Ethernet flow spec to apply.
2751  * @param eth_mask
2752  *   An Ethernet flow mask to apply.
2753  *
2754  * @return
2755  *   0 on success, a negative errno value otherwise and rte_errno is set.
2756  */
2757 int
2758 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2759                struct rte_flow_item_eth *eth_spec,
2760                struct rte_flow_item_eth *eth_mask)
2761 {
2762         return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2763 }
2764
2765 /**
2766  * Destroy a flow.
2767  *
2768  * @see rte_flow_destroy()
2769  * @see rte_flow_ops
2770  */
2771 int
2772 mlx5_flow_destroy(struct rte_eth_dev *dev,
2773                   struct rte_flow *flow,
2774                   struct rte_flow_error *error __rte_unused)
2775 {
2776         struct mlx5_priv *priv = dev->data->dev_private;
2777
2778         flow_list_destroy(dev, &priv->flows, flow);
2779         return 0;
2780 }
2781
2782 /**
2783  * Destroy all flows.
2784  *
2785  * @see rte_flow_flush()
2786  * @see rte_flow_ops
2787  */
2788 int
2789 mlx5_flow_flush(struct rte_eth_dev *dev,
2790                 struct rte_flow_error *error __rte_unused)
2791 {
2792         struct mlx5_priv *priv = dev->data->dev_private;
2793
2794         mlx5_flow_list_flush(dev, &priv->flows);
2795         return 0;
2796 }
2797
2798 /**
2799  * Isolated mode.
2800  *
2801  * @see rte_flow_isolate()
2802  * @see rte_flow_ops
2803  */
2804 int
2805 mlx5_flow_isolate(struct rte_eth_dev *dev,
2806                   int enable,
2807                   struct rte_flow_error *error)
2808 {
2809         struct mlx5_priv *priv = dev->data->dev_private;
2810
2811         if (dev->data->dev_started) {
2812                 rte_flow_error_set(error, EBUSY,
2813                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2814                                    NULL,
2815                                    "port must be stopped first");
2816                 return -rte_errno;
2817         }
2818         priv->isolated = !!enable;
2819         if (enable)
2820                 dev->dev_ops = &mlx5_dev_ops_isolate;
2821         else
2822                 dev->dev_ops = &mlx5_dev_ops;
2823         return 0;
2824 }
2825
2826 /**
2827  * Query a flow.
2828  *
2829  * @see rte_flow_query()
2830  * @see rte_flow_ops
2831  */
2832 static int
2833 flow_drv_query(struct rte_eth_dev *dev,
2834                struct rte_flow *flow,
2835                const struct rte_flow_action *actions,
2836                void *data,
2837                struct rte_flow_error *error)
2838 {
2839         const struct mlx5_flow_driver_ops *fops;
2840         enum mlx5_flow_drv_type ftype = flow->drv_type;
2841
2842         assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2843         fops = flow_get_drv_ops(ftype);
2844
2845         return fops->query(dev, flow, actions, data, error);
2846 }
2847
2848 /**
2849  * Query a flow.
2850  *
2851  * @see rte_flow_query()
2852  * @see rte_flow_ops
2853  */
2854 int
2855 mlx5_flow_query(struct rte_eth_dev *dev,
2856                 struct rte_flow *flow,
2857                 const struct rte_flow_action *actions,
2858                 void *data,
2859                 struct rte_flow_error *error)
2860 {
2861         int ret;
2862
2863         ret = flow_drv_query(dev, flow, actions, data, error);
2864         if (ret < 0)
2865                 return ret;
2866         return 0;
2867 }
2868
2869 /**
2870  * Convert a flow director filter to a generic flow.
2871  *
2872  * @param dev
2873  *   Pointer to Ethernet device.
2874  * @param fdir_filter
2875  *   Flow director filter to add.
2876  * @param attributes
2877  *   Generic flow parameters structure.
2878  *
2879  * @return
2880  *   0 on success, a negative errno value otherwise and rte_errno is set.
2881  */
2882 static int
2883 flow_fdir_filter_convert(struct rte_eth_dev *dev,
2884                          const struct rte_eth_fdir_filter *fdir_filter,
2885                          struct mlx5_fdir *attributes)
2886 {
2887         struct mlx5_priv *priv = dev->data->dev_private;
2888         const struct rte_eth_fdir_input *input = &fdir_filter->input;
2889         const struct rte_eth_fdir_masks *mask =
2890                 &dev->data->dev_conf.fdir_conf.mask;
2891
2892         /* Validate queue number. */
2893         if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2894                 DRV_LOG(ERR, "port %u invalid queue number %d",
2895                         dev->data->port_id, fdir_filter->action.rx_queue);
2896                 rte_errno = EINVAL;
2897                 return -rte_errno;
2898         }
2899         attributes->attr.ingress = 1;
2900         attributes->items[0] = (struct rte_flow_item) {
2901                 .type = RTE_FLOW_ITEM_TYPE_ETH,
2902                 .spec = &attributes->l2,
2903                 .mask = &attributes->l2_mask,
2904         };
2905         switch (fdir_filter->action.behavior) {
2906         case RTE_ETH_FDIR_ACCEPT:
2907                 attributes->actions[0] = (struct rte_flow_action){
2908                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
2909                         .conf = &attributes->queue,
2910                 };
2911                 break;
2912         case RTE_ETH_FDIR_REJECT:
2913                 attributes->actions[0] = (struct rte_flow_action){
2914                         .type = RTE_FLOW_ACTION_TYPE_DROP,
2915                 };
2916                 break;
2917         default:
2918                 DRV_LOG(ERR, "port %u invalid behavior %d",
2919                         dev->data->port_id,
2920                         fdir_filter->action.behavior);
2921                 rte_errno = ENOTSUP;
2922                 return -rte_errno;
2923         }
2924         attributes->queue.index = fdir_filter->action.rx_queue;
2925         /* Handle L3. */
2926         switch (fdir_filter->input.flow_type) {
2927         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2928         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2929         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2930                 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
2931                         .src_addr = input->flow.ip4_flow.src_ip,
2932                         .dst_addr = input->flow.ip4_flow.dst_ip,
2933                         .time_to_live = input->flow.ip4_flow.ttl,
2934                         .type_of_service = input->flow.ip4_flow.tos,
2935                 };
2936                 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
2937                         .src_addr = mask->ipv4_mask.src_ip,
2938                         .dst_addr = mask->ipv4_mask.dst_ip,
2939                         .time_to_live = mask->ipv4_mask.ttl,
2940                         .type_of_service = mask->ipv4_mask.tos,
2941                         .next_proto_id = mask->ipv4_mask.proto,
2942                 };
2943                 attributes->items[1] = (struct rte_flow_item){
2944                         .type = RTE_FLOW_ITEM_TYPE_IPV4,
2945                         .spec = &attributes->l3,
2946                         .mask = &attributes->l3_mask,
2947                 };
2948                 break;
2949         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2950         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2951         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2952                 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
2953                         .hop_limits = input->flow.ipv6_flow.hop_limits,
2954                         .proto = input->flow.ipv6_flow.proto,
2955                 };
2956
2957                 memcpy(attributes->l3.ipv6.hdr.src_addr,
2958                        input->flow.ipv6_flow.src_ip,
2959                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2960                 memcpy(attributes->l3.ipv6.hdr.dst_addr,
2961                        input->flow.ipv6_flow.dst_ip,
2962                        RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2963                 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2964                        mask->ipv6_mask.src_ip,
2965                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2966                 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2967                        mask->ipv6_mask.dst_ip,
2968                        RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2969                 attributes->items[1] = (struct rte_flow_item){
2970                         .type = RTE_FLOW_ITEM_TYPE_IPV6,
2971                         .spec = &attributes->l3,
2972                         .mask = &attributes->l3_mask,
2973                 };
2974                 break;
2975         default:
2976                 DRV_LOG(ERR, "port %u invalid flow type%d",
2977                         dev->data->port_id, fdir_filter->input.flow_type);
2978                 rte_errno = ENOTSUP;
2979                 return -rte_errno;
2980         }
2981         /* Handle L4. */
2982         switch (fdir_filter->input.flow_type) {
2983         case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2984                 attributes->l4.udp.hdr = (struct rte_udp_hdr){
2985                         .src_port = input->flow.udp4_flow.src_port,
2986                         .dst_port = input->flow.udp4_flow.dst_port,
2987                 };
2988                 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
2989                         .src_port = mask->src_port_mask,
2990                         .dst_port = mask->dst_port_mask,
2991                 };
2992                 attributes->items[2] = (struct rte_flow_item){
2993                         .type = RTE_FLOW_ITEM_TYPE_UDP,
2994                         .spec = &attributes->l4,
2995                         .mask = &attributes->l4_mask,
2996                 };
2997                 break;
2998         case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2999                 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
3000                         .src_port = input->flow.tcp4_flow.src_port,
3001                         .dst_port = input->flow.tcp4_flow.dst_port,
3002                 };
3003                 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
3004                         .src_port = mask->src_port_mask,
3005                         .dst_port = mask->dst_port_mask,
3006                 };
3007                 attributes->items[2] = (struct rte_flow_item){
3008                         .type = RTE_FLOW_ITEM_TYPE_TCP,
3009                         .spec = &attributes->l4,
3010                         .mask = &attributes->l4_mask,
3011                 };
3012                 break;
3013         case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
3014                 attributes->l4.udp.hdr = (struct rte_udp_hdr){
3015                         .src_port = input->flow.udp6_flow.src_port,
3016                         .dst_port = input->flow.udp6_flow.dst_port,
3017                 };
3018                 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
3019                         .src_port = mask->src_port_mask,
3020                         .dst_port = mask->dst_port_mask,
3021                 };
3022                 attributes->items[2] = (struct rte_flow_item){
3023                         .type = RTE_FLOW_ITEM_TYPE_UDP,
3024                         .spec = &attributes->l4,
3025                         .mask = &attributes->l4_mask,
3026                 };
3027                 break;
3028         case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
3029                 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
3030                         .src_port = input->flow.tcp6_flow.src_port,
3031                         .dst_port = input->flow.tcp6_flow.dst_port,
3032                 };
3033                 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
3034                         .src_port = mask->src_port_mask,
3035                         .dst_port = mask->dst_port_mask,
3036                 };
3037                 attributes->items[2] = (struct rte_flow_item){
3038                         .type = RTE_FLOW_ITEM_TYPE_TCP,
3039                         .spec = &attributes->l4,
3040                         .mask = &attributes->l4_mask,
3041                 };
3042                 break;
3043         case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
3044         case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
3045                 break;
3046         default:
3047                 DRV_LOG(ERR, "port %u invalid flow type%d",
3048                         dev->data->port_id, fdir_filter->input.flow_type);
3049                 rte_errno = ENOTSUP;
3050                 return -rte_errno;
3051         }
3052         return 0;
3053 }
3054
3055 #define FLOW_FDIR_CMP(f1, f2, fld) \
3056         memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
3057
3058 /**
3059  * Compare two FDIR flows. If items and actions are identical, the two flows are
3060  * regarded as same.
3061  *
3062  * @param dev
3063  *   Pointer to Ethernet device.
3064  * @param f1
3065  *   FDIR flow to compare.
3066  * @param f2
3067  *   FDIR flow to compare.
3068  *
3069  * @return
3070  *   Zero on match, 1 otherwise.
3071  */
3072 static int
3073 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
3074 {
3075         if (FLOW_FDIR_CMP(f1, f2, attr) ||
3076             FLOW_FDIR_CMP(f1, f2, l2) ||
3077             FLOW_FDIR_CMP(f1, f2, l2_mask) ||
3078             FLOW_FDIR_CMP(f1, f2, l3) ||
3079             FLOW_FDIR_CMP(f1, f2, l3_mask) ||
3080             FLOW_FDIR_CMP(f1, f2, l4) ||
3081             FLOW_FDIR_CMP(f1, f2, l4_mask) ||
3082             FLOW_FDIR_CMP(f1, f2, actions[0].type))
3083                 return 1;
3084         if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
3085             FLOW_FDIR_CMP(f1, f2, queue))
3086                 return 1;
3087         return 0;
3088 }
3089
3090 /**
3091  * Search device flow list to find out a matched FDIR flow.
3092  *
3093  * @param dev
3094  *   Pointer to Ethernet device.
3095  * @param fdir_flow
3096  *   FDIR flow to lookup.
3097  *
3098  * @return
3099  *   Pointer of flow if found, NULL otherwise.
3100  */
3101 static struct rte_flow *
3102 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
3103 {
3104         struct mlx5_priv *priv = dev->data->dev_private;
3105         struct rte_flow *flow = NULL;
3106
3107         assert(fdir_flow);
3108         TAILQ_FOREACH(flow, &priv->flows, next) {
3109                 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
3110                         DRV_LOG(DEBUG, "port %u found FDIR flow %p",
3111                                 dev->data->port_id, (void *)flow);
3112                         break;
3113                 }
3114         }
3115         return flow;
3116 }
3117
3118 /**
3119  * Add new flow director filter and store it in list.
3120  *
3121  * @param dev
3122  *   Pointer to Ethernet device.
3123  * @param fdir_filter
3124  *   Flow director filter to add.
3125  *
3126  * @return
3127  *   0 on success, a negative errno value otherwise and rte_errno is set.
3128  */
3129 static int
3130 flow_fdir_filter_add(struct rte_eth_dev *dev,
3131                      const struct rte_eth_fdir_filter *fdir_filter)
3132 {
3133         struct mlx5_priv *priv = dev->data->dev_private;
3134         struct mlx5_fdir *fdir_flow;
3135         struct rte_flow *flow;
3136         int ret;
3137
3138         fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
3139         if (!fdir_flow) {
3140                 rte_errno = ENOMEM;
3141                 return -rte_errno;
3142         }
3143         ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
3144         if (ret)
3145                 goto error;
3146         flow = flow_fdir_filter_lookup(dev, fdir_flow);
3147         if (flow) {
3148                 rte_errno = EEXIST;
3149                 goto error;
3150         }
3151         flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
3152                                 fdir_flow->items, fdir_flow->actions, true,
3153                                 NULL);
3154         if (!flow)
3155                 goto error;
3156         assert(!flow->fdir);
3157         flow->fdir = fdir_flow;
3158         DRV_LOG(DEBUG, "port %u created FDIR flow %p",
3159                 dev->data->port_id, (void *)flow);
3160         return 0;
3161 error:
3162         rte_free(fdir_flow);
3163         return -rte_errno;
3164 }
3165
3166 /**
3167  * Delete specific filter.
3168  *
3169  * @param dev
3170  *   Pointer to Ethernet device.
3171  * @param fdir_filter
3172  *   Filter to be deleted.
3173  *
3174  * @return
3175  *   0 on success, a negative errno value otherwise and rte_errno is set.
3176  */
3177 static int
3178 flow_fdir_filter_delete(struct rte_eth_dev *dev,
3179                         const struct rte_eth_fdir_filter *fdir_filter)
3180 {
3181         struct mlx5_priv *priv = dev->data->dev_private;
3182         struct rte_flow *flow;
3183         struct mlx5_fdir fdir_flow = {
3184                 .attr.group = 0,
3185         };
3186         int ret;
3187
3188         ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
3189         if (ret)
3190                 return -rte_errno;
3191         flow = flow_fdir_filter_lookup(dev, &fdir_flow);
3192         if (!flow) {
3193                 rte_errno = ENOENT;
3194                 return -rte_errno;
3195         }
3196         flow_list_destroy(dev, &priv->flows, flow);
3197         DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
3198                 dev->data->port_id, (void *)flow);
3199         return 0;
3200 }
3201
3202 /**
3203  * Update queue for specific filter.
3204  *
3205  * @param dev
3206  *   Pointer to Ethernet device.
3207  * @param fdir_filter
3208  *   Filter to be updated.
3209  *
3210  * @return
3211  *   0 on success, a negative errno value otherwise and rte_errno is set.
3212  */
3213 static int
3214 flow_fdir_filter_update(struct rte_eth_dev *dev,
3215                         const struct rte_eth_fdir_filter *fdir_filter)
3216 {
3217         int ret;
3218
3219         ret = flow_fdir_filter_delete(dev, fdir_filter);
3220         if (ret)
3221                 return ret;
3222         return flow_fdir_filter_add(dev, fdir_filter);
3223 }
3224
3225 /**
3226  * Flush all filters.
3227  *
3228  * @param dev
3229  *   Pointer to Ethernet device.
3230  */
3231 static void
3232 flow_fdir_filter_flush(struct rte_eth_dev *dev)
3233 {
3234         struct mlx5_priv *priv = dev->data->dev_private;
3235
3236         mlx5_flow_list_flush(dev, &priv->flows);
3237 }
3238
3239 /**
3240  * Get flow director information.
3241  *
3242  * @param dev
3243  *   Pointer to Ethernet device.
3244  * @param[out] fdir_info
3245  *   Resulting flow director information.
3246  */
3247 static void
3248 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
3249 {
3250         struct rte_eth_fdir_masks *mask =
3251                 &dev->data->dev_conf.fdir_conf.mask;
3252
3253         fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
3254         fdir_info->guarant_spc = 0;
3255         rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
3256         fdir_info->max_flexpayload = 0;
3257         fdir_info->flow_types_mask[0] = 0;
3258         fdir_info->flex_payload_unit = 0;
3259         fdir_info->max_flex_payload_segment_num = 0;
3260         fdir_info->flex_payload_limit = 0;
3261         memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
3262 }
3263
3264 /**
3265  * Deal with flow director operations.
3266  *
3267  * @param dev
3268  *   Pointer to Ethernet device.
3269  * @param filter_op
3270  *   Operation to perform.
3271  * @param arg
3272  *   Pointer to operation-specific structure.
3273  *
3274  * @return
3275  *   0 on success, a negative errno value otherwise and rte_errno is set.
3276  */
3277 static int
3278 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
3279                     void *arg)
3280 {
3281         enum rte_fdir_mode fdir_mode =
3282                 dev->data->dev_conf.fdir_conf.mode;
3283
3284         if (filter_op == RTE_ETH_FILTER_NOP)
3285                 return 0;
3286         if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
3287             fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
3288                 DRV_LOG(ERR, "port %u flow director mode %d not supported",
3289                         dev->data->port_id, fdir_mode);
3290                 rte_errno = EINVAL;
3291                 return -rte_errno;
3292         }
3293         switch (filter_op) {
3294         case RTE_ETH_FILTER_ADD:
3295                 return flow_fdir_filter_add(dev, arg);
3296         case RTE_ETH_FILTER_UPDATE:
3297                 return flow_fdir_filter_update(dev, arg);
3298         case RTE_ETH_FILTER_DELETE:
3299                 return flow_fdir_filter_delete(dev, arg);
3300         case RTE_ETH_FILTER_FLUSH:
3301                 flow_fdir_filter_flush(dev);
3302                 break;
3303         case RTE_ETH_FILTER_INFO:
3304                 flow_fdir_info_get(dev, arg);
3305                 break;
3306         default:
3307                 DRV_LOG(DEBUG, "port %u unknown operation %u",
3308                         dev->data->port_id, filter_op);
3309                 rte_errno = EINVAL;
3310                 return -rte_errno;
3311         }
3312         return 0;
3313 }
3314
3315 /**
3316  * Manage filter operations.
3317  *
3318  * @param dev
3319  *   Pointer to Ethernet device structure.
3320  * @param filter_type
3321  *   Filter type.
3322  * @param filter_op
3323  *   Operation to perform.
3324  * @param arg
3325  *   Pointer to operation-specific structure.
3326  *
3327  * @return
3328  *   0 on success, a negative errno value otherwise and rte_errno is set.
3329  */
3330 int
3331 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
3332                      enum rte_filter_type filter_type,
3333                      enum rte_filter_op filter_op,
3334                      void *arg)
3335 {
3336         switch (filter_type) {
3337         case RTE_ETH_FILTER_GENERIC:
3338                 if (filter_op != RTE_ETH_FILTER_GET) {
3339                         rte_errno = EINVAL;
3340                         return -rte_errno;
3341                 }
3342                 *(const void **)arg = &mlx5_flow_ops;
3343                 return 0;
3344         case RTE_ETH_FILTER_FDIR:
3345                 return flow_fdir_ctrl_func(dev, filter_op, arg);
3346         default:
3347                 DRV_LOG(ERR, "port %u filter type (%d) not supported",
3348                         dev->data->port_id, filter_type);
3349                 rte_errno = ENOTSUP;
3350                 return -rte_errno;
3351         }
3352         return 0;
3353 }
3354
3355 #define MLX5_POOL_QUERY_FREQ_US 1000000
3356
3357 /**
3358  * Set the periodic procedure for triggering asynchronous batch queries for all
3359  * the counter pools.
3360  *
3361  * @param[in] sh
3362  *   Pointer to mlx5_ibv_shared object.
3363  */
3364 void
3365 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
3366 {
3367         struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
3368         uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
3369         uint32_t us;
3370
3371         cont = MLX5_CNT_CONTAINER(sh, 1, 0);
3372         pools_n += rte_atomic16_read(&cont->n_valid);
3373         us = MLX5_POOL_QUERY_FREQ_US / pools_n;
3374         DRV_LOG(DEBUG, "Set alarm for %u pools each %u us\n", pools_n, us);
3375         if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
3376                 sh->cmng.query_thread_on = 0;
3377                 DRV_LOG(ERR, "Cannot reinitialize query alarm\n");
3378         } else {
3379                 sh->cmng.query_thread_on = 1;
3380         }
3381 }
3382
3383 /**
3384  * The periodic procedure for triggering asynchronous batch queries for all the
3385  * counter pools. This function is probably called by the host thread.
3386  *
3387  * @param[in] arg
3388  *   The parameter for the alarm process.
3389  */
3390 void
3391 mlx5_flow_query_alarm(void *arg)
3392 {
3393         struct mlx5_ibv_shared *sh = arg;
3394         struct mlx5_devx_obj *dcs;
3395         uint16_t offset;
3396         int ret;
3397         uint8_t batch = sh->cmng.batch;
3398         uint16_t pool_index = sh->cmng.pool_index;
3399         struct mlx5_pools_container *cont;
3400         struct mlx5_pools_container *mcont;
3401         struct mlx5_flow_counter_pool *pool;
3402
3403         if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
3404                 goto set_alarm;
3405 next_container:
3406         cont = MLX5_CNT_CONTAINER(sh, batch, 1);
3407         mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
3408         /* Check if resize was done and need to flip a container. */
3409         if (cont != mcont) {
3410                 if (cont->pools) {
3411                         /* Clean the old container. */
3412                         rte_free(cont->pools);
3413                         memset(cont, 0, sizeof(*cont));
3414                 }
3415                 rte_cio_wmb();
3416                  /* Flip the host container. */
3417                 sh->cmng.mhi[batch] ^= (uint8_t)2;
3418                 cont = mcont;
3419         }
3420         if (!cont->pools) {
3421                 /* 2 empty containers case is unexpected. */
3422                 if (unlikely(batch != sh->cmng.batch))
3423                         goto set_alarm;
3424                 batch ^= 0x1;
3425                 pool_index = 0;
3426                 goto next_container;
3427         }
3428         pool = cont->pools[pool_index];
3429         if (pool->raw_hw)
3430                 /* There is a pool query in progress. */
3431                 goto set_alarm;
3432         pool->raw_hw =
3433                 LIST_FIRST(&sh->cmng.free_stat_raws);
3434         if (!pool->raw_hw)
3435                 /* No free counter statistics raw memory. */
3436                 goto set_alarm;
3437         dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
3438                                                               (&pool->a64_dcs);
3439         offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
3440         ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
3441                                                offset, NULL, NULL,
3442                                                pool->raw_hw->mem_mng->dm->id,
3443                                                (void *)(uintptr_t)
3444                                                (pool->raw_hw->data + offset),
3445                                                sh->devx_comp,
3446                                                (uint64_t)(uintptr_t)pool);
3447         if (ret) {
3448                 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
3449                         " %d\n", pool->min_dcs->id);
3450                 pool->raw_hw = NULL;
3451                 goto set_alarm;
3452         }
3453         pool->raw_hw->min_dcs_id = dcs->id;
3454         LIST_REMOVE(pool->raw_hw, next);
3455         sh->cmng.pending_queries++;
3456         pool_index++;
3457         if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
3458                 batch ^= 0x1;
3459                 pool_index = 0;
3460         }
3461 set_alarm:
3462         sh->cmng.batch = batch;
3463         sh->cmng.pool_index = pool_index;
3464         mlx5_set_query_alarm(sh);
3465 }
3466
3467 /**
3468  * Handler for the HW respond about ready values from an asynchronous batch
3469  * query. This function is probably called by the host thread.
3470  *
3471  * @param[in] sh
3472  *   The pointer to the shared IB device context.
3473  * @param[in] async_id
3474  *   The Devx async ID.
3475  * @param[in] status
3476  *   The status of the completion.
3477  */
3478 void
3479 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
3480                                   uint64_t async_id, int status)
3481 {
3482         struct mlx5_flow_counter_pool *pool =
3483                 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
3484         struct mlx5_counter_stats_raw *raw_to_free;
3485
3486         if (unlikely(status)) {
3487                 raw_to_free = pool->raw_hw;
3488         } else {
3489                 raw_to_free = pool->raw;
3490                 rte_spinlock_lock(&pool->sl);
3491                 pool->raw = pool->raw_hw;
3492                 rte_spinlock_unlock(&pool->sl);
3493                 rte_atomic64_add(&pool->query_gen, 1);
3494                 /* Be sure the new raw counters data is updated in memory. */
3495                 rte_cio_wmb();
3496         }
3497         LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
3498         pool->raw_hw = NULL;
3499         sh->cmng.pending_queries--;
3500 }
3501
3502 /**
3503  * Translate the rte_flow group index to HW table value.
3504  *
3505  * @param[in] attributes
3506  *   Pointer to flow attributes
3507  * @param[in] external
3508  *   Value is part of flow rule created by request external to PMD.
3509  * @param[in] group
3510  *   rte_flow group index value.
3511  * @param[out] table
3512  *   HW table value.
3513  * @param[out] error
3514  *   Pointer to error structure.
3515  *
3516  * @return
3517  *   0 on success, a negative errno value otherwise and rte_errno is set.
3518  */
3519 int
3520 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
3521                          uint32_t group, uint32_t *table,
3522                          struct rte_flow_error *error)
3523 {
3524         if (attributes->transfer && external) {
3525                 if (group == UINT32_MAX)
3526                         return rte_flow_error_set
3527                                                 (error, EINVAL,
3528                                                  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
3529                                                  NULL,
3530                                                  "group index not supported");
3531                 *table = group + 1;
3532         } else {
3533                 *table = group;
3534         }
3535         return 0;
3536 }