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