xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision c1cfb132fa864cb6d7f57c3376d64513b74987b9)
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_GRE | 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 {
319 		struct ibv_flow_attr attr;
320 		struct ibv_flow_spec_eth eth;
321 		struct ibv_flow_spec_action_drop drop;
322 	} flow_attr = {
323 		.attr = {
324 			.num_of_specs = 2,
325 		},
326 		.eth = {
327 			.type = IBV_FLOW_SPEC_ETH,
328 			.size = sizeof(struct ibv_flow_spec_eth),
329 		},
330 		.drop = {
331 			.size = sizeof(struct ibv_flow_spec_action_drop),
332 			.type = IBV_FLOW_SPEC_ACTION_DROP,
333 		},
334 	};
335 	struct ibv_flow *flow;
336 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
337 	uint16_t vprio[] = { 8, 16 };
338 	int i;
339 	int priority = 0;
340 
341 	if (!drop) {
342 		rte_errno = ENOTSUP;
343 		return -rte_errno;
344 	}
345 	for (i = 0; i != RTE_DIM(vprio); i++) {
346 		flow_attr.attr.priority = vprio[i] - 1;
347 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
348 		if (!flow)
349 			break;
350 		claim_zero(mlx5_glue->destroy_flow(flow));
351 		priority = vprio[i];
352 	}
353 	switch (priority) {
354 	case 8:
355 		priority = RTE_DIM(priority_map_3);
356 		break;
357 	case 16:
358 		priority = RTE_DIM(priority_map_5);
359 		break;
360 	default:
361 		rte_errno = ENOTSUP;
362 		DRV_LOG(ERR,
363 			"port %u verbs maximum priority: %d expected 8/16",
364 			dev->data->port_id, vprio[i]);
365 		return -rte_errno;
366 	}
367 	mlx5_hrxq_drop_release(dev);
368 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
369 		dev->data->port_id, priority);
370 	return priority;
371 }
372 
373 /**
374  * Adjust flow priority based on the highest layer and the request priority.
375  *
376  * @param[in] dev
377  *   Pointer to the Ethernet device structure.
378  * @param[in] priority
379  *   The rule base priority.
380  * @param[in] subpriority
381  *   The priority based on the items.
382  *
383  * @return
384  *   The new priority.
385  */
386 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
387 				   uint32_t subpriority)
388 {
389 	uint32_t res = 0;
390 	struct priv *priv = dev->data->dev_private;
391 
392 	switch (priv->config.flow_prio) {
393 	case RTE_DIM(priority_map_3):
394 		res = priority_map_3[priority][subpriority];
395 		break;
396 	case RTE_DIM(priority_map_5):
397 		res = priority_map_5[priority][subpriority];
398 		break;
399 	}
400 	return  res;
401 }
402 
403 /**
404  * Verify the @p item specifications (spec, last, mask) are compatible with the
405  * NIC capabilities.
406  *
407  * @param[in] item
408  *   Item specification.
409  * @param[in] mask
410  *   @p item->mask or flow default bit-masks.
411  * @param[in] nic_mask
412  *   Bit-masks covering supported fields by the NIC to compare with user mask.
413  * @param[in] size
414  *   Bit-masks size in bytes.
415  * @param[out] error
416  *   Pointer to error structure.
417  *
418  * @return
419  *   0 on success, a negative errno value otherwise and rte_errno is set.
420  */
421 int
422 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
423 			  const uint8_t *mask,
424 			  const uint8_t *nic_mask,
425 			  unsigned int size,
426 			  struct rte_flow_error *error)
427 {
428 	unsigned int i;
429 
430 	assert(nic_mask);
431 	for (i = 0; i < size; ++i)
432 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
433 			return rte_flow_error_set(error, ENOTSUP,
434 						  RTE_FLOW_ERROR_TYPE_ITEM,
435 						  item,
436 						  "mask enables non supported"
437 						  " bits");
438 	if (!item->spec && (item->mask || item->last))
439 		return rte_flow_error_set(error, EINVAL,
440 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
441 					  "mask/last without a spec is not"
442 					  " supported");
443 	if (item->spec && item->last) {
444 		uint8_t spec[size];
445 		uint8_t last[size];
446 		unsigned int i;
447 		int ret;
448 
449 		for (i = 0; i < size; ++i) {
450 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
451 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
452 		}
453 		ret = memcmp(spec, last, size);
454 		if (ret != 0)
455 			return rte_flow_error_set(error, EINVAL,
456 						  RTE_FLOW_ERROR_TYPE_ITEM,
457 						  item,
458 						  "range is not valid");
459 	}
460 	return 0;
461 }
462 
463 /**
464  * Adjust the hash fields according to the @p flow information.
465  *
466  * @param[in] dev_flow.
467  *   Pointer to the mlx5_flow.
468  * @param[in] tunnel
469  *   1 when the hash field is for a tunnel item.
470  * @param[in] layer_types
471  *   ETH_RSS_* types.
472  * @param[in] hash_fields
473  *   Item hash fields.
474  *
475  * @return
476  *   The hash fileds that should be used.
477  */
478 uint64_t
479 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
480 			    int tunnel __rte_unused, uint64_t layer_types,
481 			    uint64_t hash_fields)
482 {
483 	struct rte_flow *flow = dev_flow->flow;
484 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
485 	int rss_request_inner = flow->rss.level >= 2;
486 
487 	/* Check RSS hash level for tunnel. */
488 	if (tunnel && rss_request_inner)
489 		hash_fields |= IBV_RX_HASH_INNER;
490 	else if (tunnel || rss_request_inner)
491 		return 0;
492 #endif
493 	/* Check if requested layer matches RSS hash fields. */
494 	if (!(flow->rss.types & layer_types))
495 		return 0;
496 	return hash_fields;
497 }
498 
499 /**
500  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
501  * if several tunnel rules are used on this queue, the tunnel ptype will be
502  * cleared.
503  *
504  * @param rxq_ctrl
505  *   Rx queue to update.
506  */
507 static void
508 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
509 {
510 	unsigned int i;
511 	uint32_t tunnel_ptype = 0;
512 
513 	/* Look up for the ptype to use. */
514 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
515 		if (!rxq_ctrl->flow_tunnels_n[i])
516 			continue;
517 		if (!tunnel_ptype) {
518 			tunnel_ptype = tunnels_info[i].ptype;
519 		} else {
520 			tunnel_ptype = 0;
521 			break;
522 		}
523 	}
524 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
525 }
526 
527 /**
528  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
529  * flow.
530  *
531  * @param[in] dev
532  *   Pointer to the Ethernet device structure.
533  * @param[in] dev_flow
534  *   Pointer to device flow structure.
535  */
536 static void
537 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
538 {
539 	struct priv *priv = dev->data->dev_private;
540 	struct rte_flow *flow = dev_flow->flow;
541 	const int mark = !!(flow->actions &
542 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
543 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
544 	unsigned int i;
545 
546 	for (i = 0; i != flow->rss.queue_num; ++i) {
547 		int idx = (*flow->queue)[i];
548 		struct mlx5_rxq_ctrl *rxq_ctrl =
549 			container_of((*priv->rxqs)[idx],
550 				     struct mlx5_rxq_ctrl, rxq);
551 
552 		if (mark) {
553 			rxq_ctrl->rxq.mark = 1;
554 			rxq_ctrl->flow_mark_n++;
555 		}
556 		if (tunnel) {
557 			unsigned int j;
558 
559 			/* Increase the counter matching the flow. */
560 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
561 				if ((tunnels_info[j].tunnel &
562 				     dev_flow->layers) ==
563 				    tunnels_info[j].tunnel) {
564 					rxq_ctrl->flow_tunnels_n[j]++;
565 					break;
566 				}
567 			}
568 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
569 		}
570 	}
571 }
572 
573 /**
574  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
575  *
576  * @param[in] dev
577  *   Pointer to the Ethernet device structure.
578  * @param[in] flow
579  *   Pointer to flow structure.
580  */
581 static void
582 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
583 {
584 	struct mlx5_flow *dev_flow;
585 
586 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
587 		flow_drv_rxq_flags_set(dev, dev_flow);
588 }
589 
590 /**
591  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
592  * device flow if no other flow uses it with the same kind of request.
593  *
594  * @param dev
595  *   Pointer to Ethernet device.
596  * @param[in] dev_flow
597  *   Pointer to the device flow.
598  */
599 static void
600 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
601 {
602 	struct priv *priv = dev->data->dev_private;
603 	struct rte_flow *flow = dev_flow->flow;
604 	const int mark = !!(flow->actions &
605 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
606 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
607 	unsigned int i;
608 
609 	assert(dev->data->dev_started);
610 	for (i = 0; i != flow->rss.queue_num; ++i) {
611 		int idx = (*flow->queue)[i];
612 		struct mlx5_rxq_ctrl *rxq_ctrl =
613 			container_of((*priv->rxqs)[idx],
614 				     struct mlx5_rxq_ctrl, rxq);
615 
616 		if (mark) {
617 			rxq_ctrl->flow_mark_n--;
618 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
619 		}
620 		if (tunnel) {
621 			unsigned int j;
622 
623 			/* Decrease the counter matching the flow. */
624 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
625 				if ((tunnels_info[j].tunnel &
626 				     dev_flow->layers) ==
627 				    tunnels_info[j].tunnel) {
628 					rxq_ctrl->flow_tunnels_n[j]--;
629 					break;
630 				}
631 			}
632 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
633 		}
634 	}
635 }
636 
637 /**
638  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
639  * @p flow if no other flow uses it with the same kind of request.
640  *
641  * @param dev
642  *   Pointer to Ethernet device.
643  * @param[in] flow
644  *   Pointer to the flow.
645  */
646 static void
647 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
648 {
649 	struct mlx5_flow *dev_flow;
650 
651 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
652 		flow_drv_rxq_flags_trim(dev, dev_flow);
653 }
654 
655 /**
656  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
657  *
658  * @param dev
659  *   Pointer to Ethernet device.
660  */
661 static void
662 flow_rxq_flags_clear(struct rte_eth_dev *dev)
663 {
664 	struct priv *priv = dev->data->dev_private;
665 	unsigned int i;
666 
667 	for (i = 0; i != priv->rxqs_n; ++i) {
668 		struct mlx5_rxq_ctrl *rxq_ctrl;
669 		unsigned int j;
670 
671 		if (!(*priv->rxqs)[i])
672 			continue;
673 		rxq_ctrl = container_of((*priv->rxqs)[i],
674 					struct mlx5_rxq_ctrl, rxq);
675 		rxq_ctrl->flow_mark_n = 0;
676 		rxq_ctrl->rxq.mark = 0;
677 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
678 			rxq_ctrl->flow_tunnels_n[j] = 0;
679 		rxq_ctrl->rxq.tunnel = 0;
680 	}
681 }
682 
683 /*
684  * Validate the flag action.
685  *
686  * @param[in] action_flags
687  *   Bit-fields that holds the actions detected until now.
688  * @param[in] attr
689  *   Attributes of flow that includes this action.
690  * @param[out] error
691  *   Pointer to error structure.
692  *
693  * @return
694  *   0 on success, a negative errno value otherwise and rte_errno is set.
695  */
696 int
697 mlx5_flow_validate_action_flag(uint64_t action_flags,
698 			       const struct rte_flow_attr *attr,
699 			       struct rte_flow_error *error)
700 {
701 
702 	if (action_flags & MLX5_FLOW_ACTION_DROP)
703 		return rte_flow_error_set(error, EINVAL,
704 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
705 					  "can't drop and flag in same flow");
706 	if (action_flags & MLX5_FLOW_ACTION_MARK)
707 		return rte_flow_error_set(error, EINVAL,
708 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
709 					  "can't mark and flag in same flow");
710 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
711 		return rte_flow_error_set(error, EINVAL,
712 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
713 					  "can't have 2 flag"
714 					  " actions in same flow");
715 	if (attr->egress)
716 		return rte_flow_error_set(error, ENOTSUP,
717 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
718 					  "flag action not supported for "
719 					  "egress");
720 	return 0;
721 }
722 
723 /*
724  * Validate the mark action.
725  *
726  * @param[in] action
727  *   Pointer to the queue action.
728  * @param[in] action_flags
729  *   Bit-fields that holds the actions detected until now.
730  * @param[in] attr
731  *   Attributes of flow that includes this action.
732  * @param[out] error
733  *   Pointer to error structure.
734  *
735  * @return
736  *   0 on success, a negative errno value otherwise and rte_errno is set.
737  */
738 int
739 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
740 			       uint64_t action_flags,
741 			       const struct rte_flow_attr *attr,
742 			       struct rte_flow_error *error)
743 {
744 	const struct rte_flow_action_mark *mark = action->conf;
745 
746 	if (!mark)
747 		return rte_flow_error_set(error, EINVAL,
748 					  RTE_FLOW_ERROR_TYPE_ACTION,
749 					  action,
750 					  "configuration cannot be null");
751 	if (mark->id >= MLX5_FLOW_MARK_MAX)
752 		return rte_flow_error_set(error, EINVAL,
753 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
754 					  &mark->id,
755 					  "mark id must in 0 <= id < "
756 					  RTE_STR(MLX5_FLOW_MARK_MAX));
757 	if (action_flags & MLX5_FLOW_ACTION_DROP)
758 		return rte_flow_error_set(error, EINVAL,
759 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
760 					  "can't drop and mark in same flow");
761 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
762 		return rte_flow_error_set(error, EINVAL,
763 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
764 					  "can't flag and mark in same flow");
765 	if (action_flags & MLX5_FLOW_ACTION_MARK)
766 		return rte_flow_error_set(error, EINVAL,
767 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
768 					  "can't have 2 mark actions in same"
769 					  " flow");
770 	if (attr->egress)
771 		return rte_flow_error_set(error, ENOTSUP,
772 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
773 					  "mark action not supported for "
774 					  "egress");
775 	return 0;
776 }
777 
778 /*
779  * Validate the drop action.
780  *
781  * @param[in] action_flags
782  *   Bit-fields that holds the actions detected until now.
783  * @param[in] attr
784  *   Attributes of flow that includes this action.
785  * @param[out] error
786  *   Pointer to error structure.
787  *
788  * @return
789  *   0 on success, a negative errno value otherwise and rte_ernno is set.
790  */
791 int
792 mlx5_flow_validate_action_drop(uint64_t action_flags,
793 			       const struct rte_flow_attr *attr,
794 			       struct rte_flow_error *error)
795 {
796 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
797 		return rte_flow_error_set(error, EINVAL,
798 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
799 					  "can't drop and flag in same flow");
800 	if (action_flags & MLX5_FLOW_ACTION_MARK)
801 		return rte_flow_error_set(error, EINVAL,
802 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
803 					  "can't drop and mark in same flow");
804 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
805 		return rte_flow_error_set(error, EINVAL,
806 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
807 					  "can't have 2 fate actions in"
808 					  " same flow");
809 	if (attr->egress)
810 		return rte_flow_error_set(error, ENOTSUP,
811 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
812 					  "drop action not supported for "
813 					  "egress");
814 	return 0;
815 }
816 
817 /*
818  * Validate the queue action.
819  *
820  * @param[in] action
821  *   Pointer to the queue action.
822  * @param[in] action_flags
823  *   Bit-fields that holds the actions detected until now.
824  * @param[in] dev
825  *   Pointer to the Ethernet device structure.
826  * @param[in] attr
827  *   Attributes of flow that includes this action.
828  * @param[out] error
829  *   Pointer to error structure.
830  *
831  * @return
832  *   0 on success, a negative errno value otherwise and rte_ernno is set.
833  */
834 int
835 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
836 				uint64_t action_flags,
837 				struct rte_eth_dev *dev,
838 				const struct rte_flow_attr *attr,
839 				struct rte_flow_error *error)
840 {
841 	struct priv *priv = dev->data->dev_private;
842 	const struct rte_flow_action_queue *queue = action->conf;
843 
844 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
845 		return rte_flow_error_set(error, EINVAL,
846 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
847 					  "can't have 2 fate actions in"
848 					  " same flow");
849 	if (queue->index >= priv->rxqs_n)
850 		return rte_flow_error_set(error, EINVAL,
851 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
852 					  &queue->index,
853 					  "queue index out of range");
854 	if (!(*priv->rxqs)[queue->index])
855 		return rte_flow_error_set(error, EINVAL,
856 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
857 					  &queue->index,
858 					  "queue is not configured");
859 	if (attr->egress)
860 		return rte_flow_error_set(error, ENOTSUP,
861 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
862 					  "queue action not supported for "
863 					  "egress");
864 	return 0;
865 }
866 
867 /*
868  * Validate the rss action.
869  *
870  * @param[in] action
871  *   Pointer to the queue action.
872  * @param[in] action_flags
873  *   Bit-fields that holds the actions detected until now.
874  * @param[in] dev
875  *   Pointer to the Ethernet device structure.
876  * @param[in] attr
877  *   Attributes of flow that includes this action.
878  * @param[out] error
879  *   Pointer to error structure.
880  *
881  * @return
882  *   0 on success, a negative errno value otherwise and rte_ernno is set.
883  */
884 int
885 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
886 			      uint64_t action_flags,
887 			      struct rte_eth_dev *dev,
888 			      const struct rte_flow_attr *attr,
889 			      struct rte_flow_error *error)
890 {
891 	struct priv *priv = dev->data->dev_private;
892 	const struct rte_flow_action_rss *rss = action->conf;
893 	unsigned int i;
894 
895 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
896 		return rte_flow_error_set(error, EINVAL,
897 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
898 					  "can't have 2 fate actions"
899 					  " in same flow");
900 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
901 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
902 		return rte_flow_error_set(error, ENOTSUP,
903 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
904 					  &rss->func,
905 					  "RSS hash function not supported");
906 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
907 	if (rss->level > 2)
908 #else
909 	if (rss->level > 1)
910 #endif
911 		return rte_flow_error_set(error, ENOTSUP,
912 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
913 					  &rss->level,
914 					  "tunnel RSS is not supported");
915 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
916 	if (rss->key_len == 0 && rss->key != NULL)
917 		return rte_flow_error_set(error, ENOTSUP,
918 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
919 					  &rss->key_len,
920 					  "RSS hash key length 0");
921 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
922 		return rte_flow_error_set(error, ENOTSUP,
923 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
924 					  &rss->key_len,
925 					  "RSS hash key too small");
926 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
927 		return rte_flow_error_set(error, ENOTSUP,
928 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
929 					  &rss->key_len,
930 					  "RSS hash key too large");
931 	if (rss->queue_num > priv->config.ind_table_max_size)
932 		return rte_flow_error_set(error, ENOTSUP,
933 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
934 					  &rss->queue_num,
935 					  "number of queues too large");
936 	if (rss->types & MLX5_RSS_HF_MASK)
937 		return rte_flow_error_set(error, ENOTSUP,
938 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
939 					  &rss->types,
940 					  "some RSS protocols are not"
941 					  " supported");
942 	for (i = 0; i != rss->queue_num; ++i) {
943 		if (!(*priv->rxqs)[rss->queue[i]])
944 			return rte_flow_error_set
945 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
946 				 &rss->queue[i], "queue is not configured");
947 	}
948 	if (attr->egress)
949 		return rte_flow_error_set(error, ENOTSUP,
950 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
951 					  "rss action not supported for "
952 					  "egress");
953 	return 0;
954 }
955 
956 /*
957  * Validate the count action.
958  *
959  * @param[in] dev
960  *   Pointer to the Ethernet device structure.
961  * @param[in] attr
962  *   Attributes of flow that includes this action.
963  * @param[out] error
964  *   Pointer to error structure.
965  *
966  * @return
967  *   0 on success, a negative errno value otherwise and rte_ernno is set.
968  */
969 int
970 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
971 				const struct rte_flow_attr *attr,
972 				struct rte_flow_error *error)
973 {
974 	if (attr->egress)
975 		return rte_flow_error_set(error, ENOTSUP,
976 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
977 					  "count action not supported for "
978 					  "egress");
979 	return 0;
980 }
981 
982 /**
983  * Verify the @p attributes will be correctly understood by the NIC and store
984  * them in the @p flow if everything is correct.
985  *
986  * @param[in] dev
987  *   Pointer to the Ethernet device structure.
988  * @param[in] attributes
989  *   Pointer to flow attributes
990  * @param[out] error
991  *   Pointer to error structure.
992  *
993  * @return
994  *   0 on success, a negative errno value otherwise and rte_errno is set.
995  */
996 int
997 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
998 			      const struct rte_flow_attr *attributes,
999 			      struct rte_flow_error *error)
1000 {
1001 	struct priv *priv = dev->data->dev_private;
1002 	uint32_t priority_max = priv->config.flow_prio - 1;
1003 
1004 	if (attributes->group)
1005 		return rte_flow_error_set(error, ENOTSUP,
1006 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1007 					  NULL, "groups is not supported");
1008 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1009 	    attributes->priority >= priority_max)
1010 		return rte_flow_error_set(error, ENOTSUP,
1011 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1012 					  NULL, "priority out of range");
1013 	if (attributes->egress)
1014 		return rte_flow_error_set(error, ENOTSUP,
1015 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1016 					  "egress is not supported");
1017 	if (attributes->transfer)
1018 		return rte_flow_error_set(error, ENOTSUP,
1019 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1020 					  NULL, "transfer is not supported");
1021 	if (!attributes->ingress)
1022 		return rte_flow_error_set(error, EINVAL,
1023 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1024 					  NULL,
1025 					  "ingress attribute is mandatory");
1026 	return 0;
1027 }
1028 
1029 /**
1030  * Validate Ethernet item.
1031  *
1032  * @param[in] item
1033  *   Item specification.
1034  * @param[in] item_flags
1035  *   Bit-fields that holds the items detected until now.
1036  * @param[out] error
1037  *   Pointer to error structure.
1038  *
1039  * @return
1040  *   0 on success, a negative errno value otherwise and rte_errno is set.
1041  */
1042 int
1043 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1044 			    uint64_t item_flags,
1045 			    struct rte_flow_error *error)
1046 {
1047 	const struct rte_flow_item_eth *mask = item->mask;
1048 	const struct rte_flow_item_eth nic_mask = {
1049 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1050 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1051 		.type = RTE_BE16(0xffff),
1052 	};
1053 	int ret;
1054 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1055 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1056 				       MLX5_FLOW_LAYER_OUTER_L2;
1057 
1058 	if (item_flags & ethm)
1059 		return rte_flow_error_set(error, ENOTSUP,
1060 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1061 					  "multiple L2 layers not supported");
1062 	if (!mask)
1063 		mask = &rte_flow_item_eth_mask;
1064 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1065 					(const uint8_t *)&nic_mask,
1066 					sizeof(struct rte_flow_item_eth),
1067 					error);
1068 	return ret;
1069 }
1070 
1071 /**
1072  * Validate VLAN item.
1073  *
1074  * @param[in] item
1075  *   Item specification.
1076  * @param[in] item_flags
1077  *   Bit-fields that holds the items detected until now.
1078  * @param[out] error
1079  *   Pointer to error structure.
1080  *
1081  * @return
1082  *   0 on success, a negative errno value otherwise and rte_errno is set.
1083  */
1084 int
1085 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1086 			     uint64_t item_flags,
1087 			     struct rte_flow_error *error)
1088 {
1089 	const struct rte_flow_item_vlan *spec = item->spec;
1090 	const struct rte_flow_item_vlan *mask = item->mask;
1091 	const struct rte_flow_item_vlan nic_mask = {
1092 		.tci = RTE_BE16(0x0fff),
1093 		.inner_type = RTE_BE16(0xffff),
1094 	};
1095 	uint16_t vlan_tag = 0;
1096 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1097 	int ret;
1098 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1099 					MLX5_FLOW_LAYER_INNER_L4) :
1100 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1101 					MLX5_FLOW_LAYER_OUTER_L4);
1102 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1103 					MLX5_FLOW_LAYER_OUTER_VLAN;
1104 
1105 	if (item_flags & vlanm)
1106 		return rte_flow_error_set(error, EINVAL,
1107 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1108 					  "multiple VLAN layers not supported");
1109 	else if ((item_flags & l34m) != 0)
1110 		return rte_flow_error_set(error, EINVAL,
1111 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1112 					  "L2 layer cannot follow L3/L4 layer");
1113 	if (!mask)
1114 		mask = &rte_flow_item_vlan_mask;
1115 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1116 					(const uint8_t *)&nic_mask,
1117 					sizeof(struct rte_flow_item_vlan),
1118 					error);
1119 	if (ret)
1120 		return ret;
1121 	if (spec) {
1122 		vlan_tag = spec->tci;
1123 		vlan_tag &= mask->tci;
1124 	}
1125 	/*
1126 	 * From verbs perspective an empty VLAN is equivalent
1127 	 * to a packet without VLAN layer.
1128 	 */
1129 	if (!vlan_tag)
1130 		return rte_flow_error_set(error, EINVAL,
1131 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1132 					  item->spec,
1133 					  "VLAN cannot be empty");
1134 	return 0;
1135 }
1136 
1137 /**
1138  * Validate IPV4 item.
1139  *
1140  * @param[in] item
1141  *   Item specification.
1142  * @param[in] item_flags
1143  *   Bit-fields that holds the items detected until now.
1144  * @param[out] error
1145  *   Pointer to error structure.
1146  *
1147  * @return
1148  *   0 on success, a negative errno value otherwise and rte_errno is set.
1149  */
1150 int
1151 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1152 			     uint64_t item_flags,
1153 			     struct rte_flow_error *error)
1154 {
1155 	const struct rte_flow_item_ipv4 *mask = item->mask;
1156 	const struct rte_flow_item_ipv4 nic_mask = {
1157 		.hdr = {
1158 			.src_addr = RTE_BE32(0xffffffff),
1159 			.dst_addr = RTE_BE32(0xffffffff),
1160 			.type_of_service = 0xff,
1161 			.next_proto_id = 0xff,
1162 		},
1163 	};
1164 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1165 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1166 				      MLX5_FLOW_LAYER_OUTER_L3;
1167 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1168 				      MLX5_FLOW_LAYER_OUTER_L4;
1169 	int ret;
1170 
1171 	if (item_flags & l3m)
1172 		return rte_flow_error_set(error, ENOTSUP,
1173 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1174 					  "multiple L3 layers not supported");
1175 	else if (item_flags & l4m)
1176 		return rte_flow_error_set(error, EINVAL,
1177 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1178 					  "L3 cannot follow an L4 layer.");
1179 	if (!mask)
1180 		mask = &rte_flow_item_ipv4_mask;
1181 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1182 					(const uint8_t *)&nic_mask,
1183 					sizeof(struct rte_flow_item_ipv4),
1184 					error);
1185 	if (ret < 0)
1186 		return ret;
1187 	return 0;
1188 }
1189 
1190 /**
1191  * Validate IPV6 item.
1192  *
1193  * @param[in] item
1194  *   Item specification.
1195  * @param[in] item_flags
1196  *   Bit-fields that holds the items detected until now.
1197  * @param[out] error
1198  *   Pointer to error structure.
1199  *
1200  * @return
1201  *   0 on success, a negative errno value otherwise and rte_errno is set.
1202  */
1203 int
1204 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1205 			     uint64_t item_flags,
1206 			     struct rte_flow_error *error)
1207 {
1208 	const struct rte_flow_item_ipv6 *mask = item->mask;
1209 	const struct rte_flow_item_ipv6 nic_mask = {
1210 		.hdr = {
1211 			.src_addr =
1212 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1213 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1214 			.dst_addr =
1215 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1216 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1217 			.vtc_flow = RTE_BE32(0xffffffff),
1218 			.proto = 0xff,
1219 			.hop_limits = 0xff,
1220 		},
1221 	};
1222 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1223 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1224 				      MLX5_FLOW_LAYER_OUTER_L3;
1225 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1226 				      MLX5_FLOW_LAYER_OUTER_L4;
1227 	int ret;
1228 
1229 	if (item_flags & l3m)
1230 		return rte_flow_error_set(error, ENOTSUP,
1231 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1232 					  "multiple L3 layers not supported");
1233 	else if (item_flags & l4m)
1234 		return rte_flow_error_set(error, EINVAL,
1235 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1236 					  "L3 cannot follow an L4 layer.");
1237 	/*
1238 	 * IPv6 is not recognised by the NIC inside a GRE tunnel.
1239 	 * Such support has to be disabled as the rule will be
1240 	 * accepted.  Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1241 	 * Mellanox OFED 4.4-1.0.0.0.
1242 	 */
1243 	if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1244 		return rte_flow_error_set(error, ENOTSUP,
1245 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1246 					  "IPv6 inside a GRE tunnel is"
1247 					  " not recognised.");
1248 	if (!mask)
1249 		mask = &rte_flow_item_ipv6_mask;
1250 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1251 					(const uint8_t *)&nic_mask,
1252 					sizeof(struct rte_flow_item_ipv6),
1253 					error);
1254 	if (ret < 0)
1255 		return ret;
1256 	return 0;
1257 }
1258 
1259 /**
1260  * Validate UDP item.
1261  *
1262  * @param[in] item
1263  *   Item specification.
1264  * @param[in] item_flags
1265  *   Bit-fields that holds the items detected until now.
1266  * @param[in] target_protocol
1267  *   The next protocol in the previous item.
1268  * @param[in] flow_mask
1269  *   mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1270  * @param[out] error
1271  *   Pointer to error structure.
1272  *
1273  * @return
1274  *   0 on success, a negative errno value otherwise and rte_errno is set.
1275  */
1276 int
1277 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1278 			    uint64_t item_flags,
1279 			    uint8_t target_protocol,
1280 			    struct rte_flow_error *error)
1281 {
1282 	const struct rte_flow_item_udp *mask = item->mask;
1283 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1284 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1285 				      MLX5_FLOW_LAYER_OUTER_L3;
1286 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1287 				      MLX5_FLOW_LAYER_OUTER_L4;
1288 	int ret;
1289 
1290 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1291 		return rte_flow_error_set(error, EINVAL,
1292 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1293 					  "protocol filtering not compatible"
1294 					  " with UDP layer");
1295 	if (!(item_flags & l3m))
1296 		return rte_flow_error_set(error, EINVAL,
1297 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1298 					  "L3 is mandatory to filter on L4");
1299 	if (item_flags & l4m)
1300 		return rte_flow_error_set(error, EINVAL,
1301 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1302 					  "multiple L4 layers not supported");
1303 	if (!mask)
1304 		mask = &rte_flow_item_udp_mask;
1305 	ret = mlx5_flow_item_acceptable
1306 		(item, (const uint8_t *)mask,
1307 		 (const uint8_t *)&rte_flow_item_udp_mask,
1308 		 sizeof(struct rte_flow_item_udp), error);
1309 	if (ret < 0)
1310 		return ret;
1311 	return 0;
1312 }
1313 
1314 /**
1315  * Validate TCP item.
1316  *
1317  * @param[in] item
1318  *   Item specification.
1319  * @param[in] item_flags
1320  *   Bit-fields that holds the items detected until now.
1321  * @param[in] target_protocol
1322  *   The next protocol in the previous item.
1323  * @param[out] error
1324  *   Pointer to error structure.
1325  *
1326  * @return
1327  *   0 on success, a negative errno value otherwise and rte_errno is set.
1328  */
1329 int
1330 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1331 			    uint64_t item_flags,
1332 			    uint8_t target_protocol,
1333 			    const struct rte_flow_item_tcp *flow_mask,
1334 			    struct rte_flow_error *error)
1335 {
1336 	const struct rte_flow_item_tcp *mask = item->mask;
1337 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1338 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1339 				      MLX5_FLOW_LAYER_OUTER_L3;
1340 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1341 				      MLX5_FLOW_LAYER_OUTER_L4;
1342 	int ret;
1343 
1344 	assert(flow_mask);
1345 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1346 		return rte_flow_error_set(error, EINVAL,
1347 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1348 					  "protocol filtering not compatible"
1349 					  " with TCP layer");
1350 	if (!(item_flags & l3m))
1351 		return rte_flow_error_set(error, EINVAL,
1352 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1353 					  "L3 is mandatory to filter on L4");
1354 	if (item_flags & l4m)
1355 		return rte_flow_error_set(error, EINVAL,
1356 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1357 					  "multiple L4 layers not supported");
1358 	if (!mask)
1359 		mask = &rte_flow_item_tcp_mask;
1360 	ret = mlx5_flow_item_acceptable
1361 		(item, (const uint8_t *)mask,
1362 		 (const uint8_t *)flow_mask,
1363 		 sizeof(struct rte_flow_item_tcp), error);
1364 	if (ret < 0)
1365 		return ret;
1366 	return 0;
1367 }
1368 
1369 /**
1370  * Validate VXLAN item.
1371  *
1372  * @param[in] item
1373  *   Item specification.
1374  * @param[in] item_flags
1375  *   Bit-fields that holds the items detected until now.
1376  * @param[in] target_protocol
1377  *   The next protocol in the previous item.
1378  * @param[out] error
1379  *   Pointer to error structure.
1380  *
1381  * @return
1382  *   0 on success, a negative errno value otherwise and rte_errno is set.
1383  */
1384 int
1385 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1386 			      uint64_t item_flags,
1387 			      struct rte_flow_error *error)
1388 {
1389 	const struct rte_flow_item_vxlan *spec = item->spec;
1390 	const struct rte_flow_item_vxlan *mask = item->mask;
1391 	int ret;
1392 	union vni {
1393 		uint32_t vlan_id;
1394 		uint8_t vni[4];
1395 	} id = { .vlan_id = 0, };
1396 	uint32_t vlan_id = 0;
1397 
1398 
1399 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1400 		return rte_flow_error_set(error, ENOTSUP,
1401 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1402 					  "multiple tunnel layers not"
1403 					  " supported");
1404 	/*
1405 	 * Verify only UDPv4 is present as defined in
1406 	 * https://tools.ietf.org/html/rfc7348
1407 	 */
1408 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1409 		return rte_flow_error_set(error, EINVAL,
1410 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1411 					  "no outer UDP layer found");
1412 	if (!mask)
1413 		mask = &rte_flow_item_vxlan_mask;
1414 	ret = mlx5_flow_item_acceptable
1415 		(item, (const uint8_t *)mask,
1416 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1417 		 sizeof(struct rte_flow_item_vxlan),
1418 		 error);
1419 	if (ret < 0)
1420 		return ret;
1421 	if (spec) {
1422 		memcpy(&id.vni[1], spec->vni, 3);
1423 		vlan_id = id.vlan_id;
1424 		memcpy(&id.vni[1], mask->vni, 3);
1425 		vlan_id &= id.vlan_id;
1426 	}
1427 	/*
1428 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1429 	 * only this layer is defined in the Verbs specification it is
1430 	 * interpreted as wildcard and all packets will match this
1431 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1432 	 * udp), all packets matching the layers before will also
1433 	 * match this rule.  To avoid such situation, VNI 0 is
1434 	 * currently refused.
1435 	 */
1436 	if (!vlan_id)
1437 		return rte_flow_error_set(error, ENOTSUP,
1438 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1439 					  "VXLAN vni cannot be 0");
1440 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1441 		return rte_flow_error_set(error, ENOTSUP,
1442 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1443 					  "VXLAN tunnel must be fully defined");
1444 	return 0;
1445 }
1446 
1447 /**
1448  * Validate VXLAN_GPE item.
1449  *
1450  * @param[in] item
1451  *   Item specification.
1452  * @param[in] item_flags
1453  *   Bit-fields that holds the items detected until now.
1454  * @param[in] priv
1455  *   Pointer to the private data structure.
1456  * @param[in] target_protocol
1457  *   The next protocol in the previous item.
1458  * @param[out] error
1459  *   Pointer to error structure.
1460  *
1461  * @return
1462  *   0 on success, a negative errno value otherwise and rte_errno is set.
1463  */
1464 int
1465 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1466 				  uint64_t item_flags,
1467 				  struct rte_eth_dev *dev,
1468 				  struct rte_flow_error *error)
1469 {
1470 	struct priv *priv = dev->data->dev_private;
1471 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1472 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1473 	int ret;
1474 	union vni {
1475 		uint32_t vlan_id;
1476 		uint8_t vni[4];
1477 	} id = { .vlan_id = 0, };
1478 	uint32_t vlan_id = 0;
1479 
1480 	if (!priv->config.l3_vxlan_en)
1481 		return rte_flow_error_set(error, ENOTSUP,
1482 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1483 					  "L3 VXLAN is not enabled by device"
1484 					  " parameter and/or not configured in"
1485 					  " firmware");
1486 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1487 		return rte_flow_error_set(error, ENOTSUP,
1488 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1489 					  "multiple tunnel layers not"
1490 					  " supported");
1491 	/*
1492 	 * Verify only UDPv4 is present as defined in
1493 	 * https://tools.ietf.org/html/rfc7348
1494 	 */
1495 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1496 		return rte_flow_error_set(error, EINVAL,
1497 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1498 					  "no outer UDP layer found");
1499 	if (!mask)
1500 		mask = &rte_flow_item_vxlan_gpe_mask;
1501 	ret = mlx5_flow_item_acceptable
1502 		(item, (const uint8_t *)mask,
1503 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1504 		 sizeof(struct rte_flow_item_vxlan_gpe),
1505 		 error);
1506 	if (ret < 0)
1507 		return ret;
1508 	if (spec) {
1509 		if (spec->protocol)
1510 			return rte_flow_error_set(error, ENOTSUP,
1511 						  RTE_FLOW_ERROR_TYPE_ITEM,
1512 						  item,
1513 						  "VxLAN-GPE protocol"
1514 						  " not supported");
1515 		memcpy(&id.vni[1], spec->vni, 3);
1516 		vlan_id = id.vlan_id;
1517 		memcpy(&id.vni[1], mask->vni, 3);
1518 		vlan_id &= id.vlan_id;
1519 	}
1520 	/*
1521 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1522 	 * layer is defined in the Verbs specification it is interpreted as
1523 	 * wildcard and all packets will match this rule, if it follows a full
1524 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1525 	 * before will also match this rule.  To avoid such situation, VNI 0
1526 	 * is currently refused.
1527 	 */
1528 	if (!vlan_id)
1529 		return rte_flow_error_set(error, ENOTSUP,
1530 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1531 					  "VXLAN-GPE vni cannot be 0");
1532 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1533 		return rte_flow_error_set(error, ENOTSUP,
1534 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1535 					  "VXLAN-GPE tunnel must be fully"
1536 					  " defined");
1537 	return 0;
1538 }
1539 
1540 /**
1541  * Validate GRE item.
1542  *
1543  * @param[in] item
1544  *   Item specification.
1545  * @param[in] item_flags
1546  *   Bit flags to mark detected items.
1547  * @param[in] target_protocol
1548  *   The next protocol in the previous item.
1549  * @param[out] error
1550  *   Pointer to error structure.
1551  *
1552  * @return
1553  *   0 on success, a negative errno value otherwise and rte_errno is set.
1554  */
1555 int
1556 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1557 			    uint64_t item_flags,
1558 			    uint8_t target_protocol,
1559 			    struct rte_flow_error *error)
1560 {
1561 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1562 	const struct rte_flow_item_gre *mask = item->mask;
1563 	int ret;
1564 
1565 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1566 		return rte_flow_error_set(error, EINVAL,
1567 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1568 					  "protocol filtering not compatible"
1569 					  " with this GRE layer");
1570 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1571 		return rte_flow_error_set(error, ENOTSUP,
1572 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1573 					  "multiple tunnel layers not"
1574 					  " supported");
1575 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1576 		return rte_flow_error_set(error, ENOTSUP,
1577 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1578 					  "L3 Layer is missing");
1579 	if (!mask)
1580 		mask = &rte_flow_item_gre_mask;
1581 	ret = mlx5_flow_item_acceptable
1582 		(item, (const uint8_t *)mask,
1583 		 (const uint8_t *)&rte_flow_item_gre_mask,
1584 		 sizeof(struct rte_flow_item_gre), error);
1585 	if (ret < 0)
1586 		return ret;
1587 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1588 	if (spec && (spec->protocol & mask->protocol))
1589 		return rte_flow_error_set(error, ENOTSUP,
1590 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1591 					  "without MPLS support the"
1592 					  " specification cannot be used for"
1593 					  " filtering");
1594 #endif
1595 	return 0;
1596 }
1597 
1598 /**
1599  * Validate MPLS item.
1600  *
1601  * @param[in] item
1602  *   Item specification.
1603  * @param[in] item_flags
1604  *   Bit-fields that holds the items detected until now.
1605  * @param[in] target_protocol
1606  *   The next protocol in the previous item.
1607  * @param[out] error
1608  *   Pointer to error structure.
1609  *
1610  * @return
1611  *   0 on success, a negative errno value otherwise and rte_errno is set.
1612  */
1613 int
1614 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1615 			     uint64_t item_flags __rte_unused,
1616 			     uint8_t target_protocol __rte_unused,
1617 			     struct rte_flow_error *error)
1618 {
1619 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1620 	const struct rte_flow_item_mpls *mask = item->mask;
1621 	int ret;
1622 
1623 	if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1624 		return rte_flow_error_set(error, EINVAL,
1625 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1626 					  "protocol filtering not compatible"
1627 					  " with MPLS layer");
1628 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
1629 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1630 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
1631 		return rte_flow_error_set(error, ENOTSUP,
1632 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1633 					  "multiple tunnel layers not"
1634 					  " supported");
1635 	if (!mask)
1636 		mask = &rte_flow_item_mpls_mask;
1637 	ret = mlx5_flow_item_acceptable
1638 		(item, (const uint8_t *)mask,
1639 		 (const uint8_t *)&rte_flow_item_mpls_mask,
1640 		 sizeof(struct rte_flow_item_mpls), error);
1641 	if (ret < 0)
1642 		return ret;
1643 	return 0;
1644 #endif
1645 	return rte_flow_error_set(error, ENOTSUP,
1646 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
1647 				  "MPLS is not supported by Verbs, please"
1648 				  " update.");
1649 }
1650 
1651 static int
1652 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1653 		   const struct rte_flow_attr *attr __rte_unused,
1654 		   const struct rte_flow_item items[] __rte_unused,
1655 		   const struct rte_flow_action actions[] __rte_unused,
1656 		   struct rte_flow_error *error __rte_unused)
1657 {
1658 	rte_errno = ENOTSUP;
1659 	return -rte_errno;
1660 }
1661 
1662 static struct mlx5_flow *
1663 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1664 		  const struct rte_flow_item items[] __rte_unused,
1665 		  const struct rte_flow_action actions[] __rte_unused,
1666 		  struct rte_flow_error *error __rte_unused)
1667 {
1668 	rte_errno = ENOTSUP;
1669 	return NULL;
1670 }
1671 
1672 static int
1673 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1674 		    struct mlx5_flow *dev_flow __rte_unused,
1675 		    const struct rte_flow_attr *attr __rte_unused,
1676 		    const struct rte_flow_item items[] __rte_unused,
1677 		    const struct rte_flow_action actions[] __rte_unused,
1678 		    struct rte_flow_error *error __rte_unused)
1679 {
1680 	rte_errno = ENOTSUP;
1681 	return -rte_errno;
1682 }
1683 
1684 static int
1685 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1686 		struct rte_flow *flow __rte_unused,
1687 		struct rte_flow_error *error __rte_unused)
1688 {
1689 	rte_errno = ENOTSUP;
1690 	return -rte_errno;
1691 }
1692 
1693 static void
1694 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1695 		 struct rte_flow *flow __rte_unused)
1696 {
1697 }
1698 
1699 static void
1700 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1701 		  struct rte_flow *flow __rte_unused)
1702 {
1703 }
1704 
1705 static int
1706 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1707 		struct rte_flow *flow __rte_unused,
1708 		const struct rte_flow_action *actions __rte_unused,
1709 		void *data __rte_unused,
1710 		struct rte_flow_error *error __rte_unused)
1711 {
1712 	rte_errno = ENOTSUP;
1713 	return -rte_errno;
1714 }
1715 
1716 /* Void driver to protect from null pointer reference. */
1717 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1718 	.validate = flow_null_validate,
1719 	.prepare = flow_null_prepare,
1720 	.translate = flow_null_translate,
1721 	.apply = flow_null_apply,
1722 	.remove = flow_null_remove,
1723 	.destroy = flow_null_destroy,
1724 	.query = flow_null_query,
1725 };
1726 
1727 /**
1728  * Select flow driver type according to flow attributes and device
1729  * configuration.
1730  *
1731  * @param[in] dev
1732  *   Pointer to the dev structure.
1733  * @param[in] attr
1734  *   Pointer to the flow attributes.
1735  *
1736  * @return
1737  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1738  */
1739 static enum mlx5_flow_drv_type
1740 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1741 {
1742 	struct priv *priv = dev->data->dev_private;
1743 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1744 
1745 	if (attr->transfer)
1746 		type = MLX5_FLOW_TYPE_TCF;
1747 	else
1748 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1749 						 MLX5_FLOW_TYPE_VERBS;
1750 	return type;
1751 }
1752 
1753 #define flow_get_drv_ops(type) flow_drv_ops[type]
1754 
1755 /**
1756  * Flow driver validation API. This abstracts calling driver specific functions.
1757  * The type of flow driver is determined according to flow attributes.
1758  *
1759  * @param[in] dev
1760  *   Pointer to the dev structure.
1761  * @param[in] attr
1762  *   Pointer to the flow attributes.
1763  * @param[in] items
1764  *   Pointer to the list of items.
1765  * @param[in] actions
1766  *   Pointer to the list of actions.
1767  * @param[out] error
1768  *   Pointer to the error structure.
1769  *
1770  * @return
1771  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1772  */
1773 static inline int
1774 flow_drv_validate(struct rte_eth_dev *dev,
1775 		  const struct rte_flow_attr *attr,
1776 		  const struct rte_flow_item items[],
1777 		  const struct rte_flow_action actions[],
1778 		  struct rte_flow_error *error)
1779 {
1780 	const struct mlx5_flow_driver_ops *fops;
1781 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1782 
1783 	fops = flow_get_drv_ops(type);
1784 	return fops->validate(dev, attr, items, actions, error);
1785 }
1786 
1787 /**
1788  * Flow driver preparation API. This abstracts calling driver specific
1789  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1790  * calculates the size of memory required for device flow, allocates the memory,
1791  * initializes the device flow and returns the pointer.
1792  *
1793  * @note
1794  *   This function initializes device flow structure such as dv, tcf or verbs in
1795  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
1796  *   rest. For example, adding returning device flow to flow->dev_flow list and
1797  *   setting backward reference to the flow should be done out of this function.
1798  *   layers field is not filled either.
1799  *
1800  * @param[in] attr
1801  *   Pointer to the flow attributes.
1802  * @param[in] items
1803  *   Pointer to the list of items.
1804  * @param[in] actions
1805  *   Pointer to the list of actions.
1806  * @param[out] error
1807  *   Pointer to the error structure.
1808  *
1809  * @return
1810  *   Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1811  */
1812 static inline struct mlx5_flow *
1813 flow_drv_prepare(const struct rte_flow *flow,
1814 		 const struct rte_flow_attr *attr,
1815 		 const struct rte_flow_item items[],
1816 		 const struct rte_flow_action actions[],
1817 		 struct rte_flow_error *error)
1818 {
1819 	const struct mlx5_flow_driver_ops *fops;
1820 	enum mlx5_flow_drv_type type = flow->drv_type;
1821 
1822 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1823 	fops = flow_get_drv_ops(type);
1824 	return fops->prepare(attr, items, actions, error);
1825 }
1826 
1827 /**
1828  * Flow driver translation API. This abstracts calling driver specific
1829  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1830  * translates a generic flow into a driver flow. flow_drv_prepare() must
1831  * precede.
1832  *
1833  * @note
1834  *   dev_flow->layers could be filled as a result of parsing during translation
1835  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
1836  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
1837  *   flow->actions could be overwritten even though all the expanded dev_flows
1838  *   have the same actions.
1839  *
1840  * @param[in] dev
1841  *   Pointer to the rte dev structure.
1842  * @param[in, out] dev_flow
1843  *   Pointer to the mlx5 flow.
1844  * @param[in] attr
1845  *   Pointer to the flow attributes.
1846  * @param[in] items
1847  *   Pointer to the list of items.
1848  * @param[in] actions
1849  *   Pointer to the list of actions.
1850  * @param[out] error
1851  *   Pointer to the error structure.
1852  *
1853  * @return
1854  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1855  */
1856 static inline int
1857 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1858 		   const struct rte_flow_attr *attr,
1859 		   const struct rte_flow_item items[],
1860 		   const struct rte_flow_action actions[],
1861 		   struct rte_flow_error *error)
1862 {
1863 	const struct mlx5_flow_driver_ops *fops;
1864 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1865 
1866 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1867 	fops = flow_get_drv_ops(type);
1868 	return fops->translate(dev, dev_flow, attr, items, actions, error);
1869 }
1870 
1871 /**
1872  * Flow driver apply API. This abstracts calling driver specific functions.
1873  * Parent flow (rte_flow) should have driver type (drv_type). It applies
1874  * translated driver flows on to device. flow_drv_translate() must precede.
1875  *
1876  * @param[in] dev
1877  *   Pointer to Ethernet device structure.
1878  * @param[in, out] flow
1879  *   Pointer to flow structure.
1880  * @param[out] error
1881  *   Pointer to error structure.
1882  *
1883  * @return
1884  *   0 on success, a negative errno value otherwise and rte_errno is set.
1885  */
1886 static inline int
1887 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1888 	       struct rte_flow_error *error)
1889 {
1890 	const struct mlx5_flow_driver_ops *fops;
1891 	enum mlx5_flow_drv_type type = flow->drv_type;
1892 
1893 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1894 	fops = flow_get_drv_ops(type);
1895 	return fops->apply(dev, flow, error);
1896 }
1897 
1898 /**
1899  * Flow driver remove API. This abstracts calling driver specific functions.
1900  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1901  * on device. All the resources of the flow should be freed by calling
1902  * flow_drv_destroy().
1903  *
1904  * @param[in] dev
1905  *   Pointer to Ethernet device.
1906  * @param[in, out] flow
1907  *   Pointer to flow structure.
1908  */
1909 static inline void
1910 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1911 {
1912 	const struct mlx5_flow_driver_ops *fops;
1913 	enum mlx5_flow_drv_type type = flow->drv_type;
1914 
1915 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1916 	fops = flow_get_drv_ops(type);
1917 	fops->remove(dev, flow);
1918 }
1919 
1920 /**
1921  * Flow driver destroy API. This abstracts calling driver specific functions.
1922  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1923  * on device and releases resources of the flow.
1924  *
1925  * @param[in] dev
1926  *   Pointer to Ethernet device.
1927  * @param[in, out] flow
1928  *   Pointer to flow structure.
1929  */
1930 static inline void
1931 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1932 {
1933 	const struct mlx5_flow_driver_ops *fops;
1934 	enum mlx5_flow_drv_type type = flow->drv_type;
1935 
1936 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1937 	fops = flow_get_drv_ops(type);
1938 	fops->destroy(dev, flow);
1939 }
1940 
1941 /**
1942  * Validate a flow supported by the NIC.
1943  *
1944  * @see rte_flow_validate()
1945  * @see rte_flow_ops
1946  */
1947 int
1948 mlx5_flow_validate(struct rte_eth_dev *dev,
1949 		   const struct rte_flow_attr *attr,
1950 		   const struct rte_flow_item items[],
1951 		   const struct rte_flow_action actions[],
1952 		   struct rte_flow_error *error)
1953 {
1954 	int ret;
1955 
1956 	ret = flow_drv_validate(dev, attr, items, actions, error);
1957 	if (ret < 0)
1958 		return ret;
1959 	return 0;
1960 }
1961 
1962 /**
1963  * Get RSS action from the action list.
1964  *
1965  * @param[in] actions
1966  *   Pointer to the list of actions.
1967  *
1968  * @return
1969  *   Pointer to the RSS action if exist, else return NULL.
1970  */
1971 static const struct rte_flow_action_rss*
1972 flow_get_rss_action(const struct rte_flow_action actions[])
1973 {
1974 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1975 		switch (actions->type) {
1976 		case RTE_FLOW_ACTION_TYPE_RSS:
1977 			return (const struct rte_flow_action_rss *)
1978 			       actions->conf;
1979 		default:
1980 			break;
1981 		}
1982 	}
1983 	return NULL;
1984 }
1985 
1986 static unsigned int
1987 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
1988 {
1989 	const struct rte_flow_item *item;
1990 	unsigned int has_vlan = 0;
1991 
1992 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
1993 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
1994 			has_vlan = 1;
1995 			break;
1996 		}
1997 	}
1998 	if (has_vlan)
1999 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2000 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2001 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2002 			       MLX5_EXPANSION_ROOT_OUTER;
2003 }
2004 
2005 /**
2006  * Create a flow and add it to @p list.
2007  *
2008  * @param dev
2009  *   Pointer to Ethernet device.
2010  * @param list
2011  *   Pointer to a TAILQ flow list.
2012  * @param[in] attr
2013  *   Flow rule attributes.
2014  * @param[in] items
2015  *   Pattern specification (list terminated by the END pattern item).
2016  * @param[in] actions
2017  *   Associated actions (list terminated by the END action).
2018  * @param[out] error
2019  *   Perform verbose error reporting if not NULL.
2020  *
2021  * @return
2022  *   A flow on success, NULL otherwise and rte_errno is set.
2023  */
2024 static struct rte_flow *
2025 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2026 		 const struct rte_flow_attr *attr,
2027 		 const struct rte_flow_item items[],
2028 		 const struct rte_flow_action actions[],
2029 		 struct rte_flow_error *error)
2030 {
2031 	struct rte_flow *flow = NULL;
2032 	struct mlx5_flow *dev_flow;
2033 	const struct rte_flow_action_rss *rss;
2034 	union {
2035 		struct rte_flow_expand_rss buf;
2036 		uint8_t buffer[2048];
2037 	} expand_buffer;
2038 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2039 	int ret;
2040 	uint32_t i;
2041 	uint32_t flow_size;
2042 
2043 	ret = flow_drv_validate(dev, attr, items, actions, error);
2044 	if (ret < 0)
2045 		return NULL;
2046 	flow_size = sizeof(struct rte_flow);
2047 	rss = flow_get_rss_action(actions);
2048 	if (rss)
2049 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2050 					    sizeof(void *));
2051 	else
2052 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2053 	flow = rte_calloc(__func__, 1, flow_size, 0);
2054 	flow->drv_type = flow_get_drv_type(dev, attr);
2055 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2056 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
2057 	flow->queue = (void *)(flow + 1);
2058 	LIST_INIT(&flow->dev_flows);
2059 	if (rss && rss->types) {
2060 		unsigned int graph_root;
2061 
2062 		graph_root = find_graph_root(items, rss->level);
2063 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2064 					  items, rss->types,
2065 					  mlx5_support_expansion,
2066 					  graph_root);
2067 		assert(ret > 0 &&
2068 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
2069 	} else {
2070 		buf->entries = 1;
2071 		buf->entry[0].pattern = (void *)(uintptr_t)items;
2072 	}
2073 	for (i = 0; i < buf->entries; ++i) {
2074 		dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2075 					    actions, error);
2076 		if (!dev_flow)
2077 			goto error;
2078 		dev_flow->flow = flow;
2079 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2080 		ret = flow_drv_translate(dev, dev_flow, attr,
2081 					 buf->entry[i].pattern,
2082 					 actions, error);
2083 		if (ret < 0)
2084 			goto error;
2085 	}
2086 	if (dev->data->dev_started) {
2087 		ret = flow_drv_apply(dev, flow, error);
2088 		if (ret < 0)
2089 			goto error;
2090 	}
2091 	TAILQ_INSERT_TAIL(list, flow, next);
2092 	flow_rxq_flags_set(dev, flow);
2093 	return flow;
2094 error:
2095 	ret = rte_errno; /* Save rte_errno before cleanup. */
2096 	assert(flow);
2097 	flow_drv_destroy(dev, flow);
2098 	rte_free(flow);
2099 	rte_errno = ret; /* Restore rte_errno. */
2100 	return NULL;
2101 }
2102 
2103 /**
2104  * Create a flow.
2105  *
2106  * @see rte_flow_create()
2107  * @see rte_flow_ops
2108  */
2109 struct rte_flow *
2110 mlx5_flow_create(struct rte_eth_dev *dev,
2111 		 const struct rte_flow_attr *attr,
2112 		 const struct rte_flow_item items[],
2113 		 const struct rte_flow_action actions[],
2114 		 struct rte_flow_error *error)
2115 {
2116 	return flow_list_create(dev,
2117 				&((struct priv *)dev->data->dev_private)->flows,
2118 				attr, items, actions, error);
2119 }
2120 
2121 /**
2122  * Destroy a flow in a list.
2123  *
2124  * @param dev
2125  *   Pointer to Ethernet device.
2126  * @param list
2127  *   Pointer to a TAILQ flow list.
2128  * @param[in] flow
2129  *   Flow to destroy.
2130  */
2131 static void
2132 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2133 		  struct rte_flow *flow)
2134 {
2135 	flow_drv_destroy(dev, flow);
2136 	TAILQ_REMOVE(list, flow, next);
2137 	/*
2138 	 * Update RX queue flags only if port is started, otherwise it is
2139 	 * already clean.
2140 	 */
2141 	if (dev->data->dev_started)
2142 		flow_rxq_flags_trim(dev, flow);
2143 	rte_free(flow->fdir);
2144 	rte_free(flow);
2145 }
2146 
2147 /**
2148  * Destroy all flows.
2149  *
2150  * @param dev
2151  *   Pointer to Ethernet device.
2152  * @param list
2153  *   Pointer to a TAILQ flow list.
2154  */
2155 void
2156 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2157 {
2158 	while (!TAILQ_EMPTY(list)) {
2159 		struct rte_flow *flow;
2160 
2161 		flow = TAILQ_FIRST(list);
2162 		flow_list_destroy(dev, list, flow);
2163 	}
2164 }
2165 
2166 /**
2167  * Remove all flows.
2168  *
2169  * @param dev
2170  *   Pointer to Ethernet device.
2171  * @param list
2172  *   Pointer to a TAILQ flow list.
2173  */
2174 void
2175 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2176 {
2177 	struct rte_flow *flow;
2178 
2179 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2180 		flow_drv_remove(dev, flow);
2181 	flow_rxq_flags_clear(dev);
2182 }
2183 
2184 /**
2185  * Add all flows.
2186  *
2187  * @param dev
2188  *   Pointer to Ethernet device.
2189  * @param list
2190  *   Pointer to a TAILQ flow list.
2191  *
2192  * @return
2193  *   0 on success, a negative errno value otherwise and rte_errno is set.
2194  */
2195 int
2196 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2197 {
2198 	struct rte_flow *flow;
2199 	struct rte_flow_error error;
2200 	int ret = 0;
2201 
2202 	TAILQ_FOREACH(flow, list, next) {
2203 		ret = flow_drv_apply(dev, flow, &error);
2204 		if (ret < 0)
2205 			goto error;
2206 		flow_rxq_flags_set(dev, flow);
2207 	}
2208 	return 0;
2209 error:
2210 	ret = rte_errno; /* Save rte_errno before cleanup. */
2211 	mlx5_flow_stop(dev, list);
2212 	rte_errno = ret; /* Restore rte_errno. */
2213 	return -rte_errno;
2214 }
2215 
2216 /**
2217  * Verify the flow list is empty
2218  *
2219  * @param dev
2220  *  Pointer to Ethernet device.
2221  *
2222  * @return the number of flows not released.
2223  */
2224 int
2225 mlx5_flow_verify(struct rte_eth_dev *dev)
2226 {
2227 	struct priv *priv = dev->data->dev_private;
2228 	struct rte_flow *flow;
2229 	int ret = 0;
2230 
2231 	TAILQ_FOREACH(flow, &priv->flows, next) {
2232 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
2233 			dev->data->port_id, (void *)flow);
2234 		++ret;
2235 	}
2236 	return ret;
2237 }
2238 
2239 /**
2240  * Enable a control flow configured from the control plane.
2241  *
2242  * @param dev
2243  *   Pointer to Ethernet device.
2244  * @param eth_spec
2245  *   An Ethernet flow spec to apply.
2246  * @param eth_mask
2247  *   An Ethernet flow mask to apply.
2248  * @param vlan_spec
2249  *   A VLAN flow spec to apply.
2250  * @param vlan_mask
2251  *   A VLAN flow mask to apply.
2252  *
2253  * @return
2254  *   0 on success, a negative errno value otherwise and rte_errno is set.
2255  */
2256 int
2257 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2258 		    struct rte_flow_item_eth *eth_spec,
2259 		    struct rte_flow_item_eth *eth_mask,
2260 		    struct rte_flow_item_vlan *vlan_spec,
2261 		    struct rte_flow_item_vlan *vlan_mask)
2262 {
2263 	struct priv *priv = dev->data->dev_private;
2264 	const struct rte_flow_attr attr = {
2265 		.ingress = 1,
2266 		.priority = MLX5_FLOW_PRIO_RSVD,
2267 	};
2268 	struct rte_flow_item items[] = {
2269 		{
2270 			.type = RTE_FLOW_ITEM_TYPE_ETH,
2271 			.spec = eth_spec,
2272 			.last = NULL,
2273 			.mask = eth_mask,
2274 		},
2275 		{
2276 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2277 					      RTE_FLOW_ITEM_TYPE_END,
2278 			.spec = vlan_spec,
2279 			.last = NULL,
2280 			.mask = vlan_mask,
2281 		},
2282 		{
2283 			.type = RTE_FLOW_ITEM_TYPE_END,
2284 		},
2285 	};
2286 	uint16_t queue[priv->reta_idx_n];
2287 	struct rte_flow_action_rss action_rss = {
2288 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2289 		.level = 0,
2290 		.types = priv->rss_conf.rss_hf,
2291 		.key_len = priv->rss_conf.rss_key_len,
2292 		.queue_num = priv->reta_idx_n,
2293 		.key = priv->rss_conf.rss_key,
2294 		.queue = queue,
2295 	};
2296 	struct rte_flow_action actions[] = {
2297 		{
2298 			.type = RTE_FLOW_ACTION_TYPE_RSS,
2299 			.conf = &action_rss,
2300 		},
2301 		{
2302 			.type = RTE_FLOW_ACTION_TYPE_END,
2303 		},
2304 	};
2305 	struct rte_flow *flow;
2306 	struct rte_flow_error error;
2307 	unsigned int i;
2308 
2309 	if (!priv->reta_idx_n) {
2310 		rte_errno = EINVAL;
2311 		return -rte_errno;
2312 	}
2313 	for (i = 0; i != priv->reta_idx_n; ++i)
2314 		queue[i] = (*priv->reta_idx)[i];
2315 	flow = flow_list_create(dev, &priv->ctrl_flows,
2316 				&attr, items, actions, &error);
2317 	if (!flow)
2318 		return -rte_errno;
2319 	return 0;
2320 }
2321 
2322 /**
2323  * Enable a flow control configured from the control plane.
2324  *
2325  * @param dev
2326  *   Pointer to Ethernet device.
2327  * @param eth_spec
2328  *   An Ethernet flow spec to apply.
2329  * @param eth_mask
2330  *   An Ethernet flow mask to apply.
2331  *
2332  * @return
2333  *   0 on success, a negative errno value otherwise and rte_errno is set.
2334  */
2335 int
2336 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2337 	       struct rte_flow_item_eth *eth_spec,
2338 	       struct rte_flow_item_eth *eth_mask)
2339 {
2340 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2341 }
2342 
2343 /**
2344  * Destroy a flow.
2345  *
2346  * @see rte_flow_destroy()
2347  * @see rte_flow_ops
2348  */
2349 int
2350 mlx5_flow_destroy(struct rte_eth_dev *dev,
2351 		  struct rte_flow *flow,
2352 		  struct rte_flow_error *error __rte_unused)
2353 {
2354 	struct priv *priv = dev->data->dev_private;
2355 
2356 	flow_list_destroy(dev, &priv->flows, flow);
2357 	return 0;
2358 }
2359 
2360 /**
2361  * Destroy all flows.
2362  *
2363  * @see rte_flow_flush()
2364  * @see rte_flow_ops
2365  */
2366 int
2367 mlx5_flow_flush(struct rte_eth_dev *dev,
2368 		struct rte_flow_error *error __rte_unused)
2369 {
2370 	struct priv *priv = dev->data->dev_private;
2371 
2372 	mlx5_flow_list_flush(dev, &priv->flows);
2373 	return 0;
2374 }
2375 
2376 /**
2377  * Isolated mode.
2378  *
2379  * @see rte_flow_isolate()
2380  * @see rte_flow_ops
2381  */
2382 int
2383 mlx5_flow_isolate(struct rte_eth_dev *dev,
2384 		  int enable,
2385 		  struct rte_flow_error *error)
2386 {
2387 	struct priv *priv = dev->data->dev_private;
2388 
2389 	if (dev->data->dev_started) {
2390 		rte_flow_error_set(error, EBUSY,
2391 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2392 				   NULL,
2393 				   "port must be stopped first");
2394 		return -rte_errno;
2395 	}
2396 	priv->isolated = !!enable;
2397 	if (enable)
2398 		dev->dev_ops = &mlx5_dev_ops_isolate;
2399 	else
2400 		dev->dev_ops = &mlx5_dev_ops;
2401 	return 0;
2402 }
2403 
2404 /**
2405  * Query a flow.
2406  *
2407  * @see rte_flow_query()
2408  * @see rte_flow_ops
2409  */
2410 static int
2411 flow_drv_query(struct rte_eth_dev *dev,
2412 	       struct rte_flow *flow,
2413 	       const struct rte_flow_action *actions,
2414 	       void *data,
2415 	       struct rte_flow_error *error)
2416 {
2417 	const struct mlx5_flow_driver_ops *fops;
2418 	enum mlx5_flow_drv_type ftype = flow->drv_type;
2419 
2420 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2421 	fops = flow_get_drv_ops(ftype);
2422 
2423 	return fops->query(dev, flow, actions, data, error);
2424 }
2425 
2426 /**
2427  * Query a flow.
2428  *
2429  * @see rte_flow_query()
2430  * @see rte_flow_ops
2431  */
2432 int
2433 mlx5_flow_query(struct rte_eth_dev *dev,
2434 		struct rte_flow *flow,
2435 		const struct rte_flow_action *actions,
2436 		void *data,
2437 		struct rte_flow_error *error)
2438 {
2439 	int ret;
2440 
2441 	ret = flow_drv_query(dev, flow, actions, data, error);
2442 	if (ret < 0)
2443 		return ret;
2444 	return 0;
2445 }
2446 
2447 /**
2448  * Convert a flow director filter to a generic flow.
2449  *
2450  * @param dev
2451  *   Pointer to Ethernet device.
2452  * @param fdir_filter
2453  *   Flow director filter to add.
2454  * @param attributes
2455  *   Generic flow parameters structure.
2456  *
2457  * @return
2458  *   0 on success, a negative errno value otherwise and rte_errno is set.
2459  */
2460 static int
2461 flow_fdir_filter_convert(struct rte_eth_dev *dev,
2462 			 const struct rte_eth_fdir_filter *fdir_filter,
2463 			 struct mlx5_fdir *attributes)
2464 {
2465 	struct priv *priv = dev->data->dev_private;
2466 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
2467 	const struct rte_eth_fdir_masks *mask =
2468 		&dev->data->dev_conf.fdir_conf.mask;
2469 
2470 	/* Validate queue number. */
2471 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2472 		DRV_LOG(ERR, "port %u invalid queue number %d",
2473 			dev->data->port_id, fdir_filter->action.rx_queue);
2474 		rte_errno = EINVAL;
2475 		return -rte_errno;
2476 	}
2477 	attributes->attr.ingress = 1;
2478 	attributes->items[0] = (struct rte_flow_item) {
2479 		.type = RTE_FLOW_ITEM_TYPE_ETH,
2480 		.spec = &attributes->l2,
2481 		.mask = &attributes->l2_mask,
2482 	};
2483 	switch (fdir_filter->action.behavior) {
2484 	case RTE_ETH_FDIR_ACCEPT:
2485 		attributes->actions[0] = (struct rte_flow_action){
2486 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
2487 			.conf = &attributes->queue,
2488 		};
2489 		break;
2490 	case RTE_ETH_FDIR_REJECT:
2491 		attributes->actions[0] = (struct rte_flow_action){
2492 			.type = RTE_FLOW_ACTION_TYPE_DROP,
2493 		};
2494 		break;
2495 	default:
2496 		DRV_LOG(ERR, "port %u invalid behavior %d",
2497 			dev->data->port_id,
2498 			fdir_filter->action.behavior);
2499 		rte_errno = ENOTSUP;
2500 		return -rte_errno;
2501 	}
2502 	attributes->queue.index = fdir_filter->action.rx_queue;
2503 	/* Handle L3. */
2504 	switch (fdir_filter->input.flow_type) {
2505 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2506 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2507 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2508 		attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2509 			.src_addr = input->flow.ip4_flow.src_ip,
2510 			.dst_addr = input->flow.ip4_flow.dst_ip,
2511 			.time_to_live = input->flow.ip4_flow.ttl,
2512 			.type_of_service = input->flow.ip4_flow.tos,
2513 		};
2514 		attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2515 			.src_addr = mask->ipv4_mask.src_ip,
2516 			.dst_addr = mask->ipv4_mask.dst_ip,
2517 			.time_to_live = mask->ipv4_mask.ttl,
2518 			.type_of_service = mask->ipv4_mask.tos,
2519 			.next_proto_id = mask->ipv4_mask.proto,
2520 		};
2521 		attributes->items[1] = (struct rte_flow_item){
2522 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
2523 			.spec = &attributes->l3,
2524 			.mask = &attributes->l3_mask,
2525 		};
2526 		break;
2527 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2528 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2529 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2530 		attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2531 			.hop_limits = input->flow.ipv6_flow.hop_limits,
2532 			.proto = input->flow.ipv6_flow.proto,
2533 		};
2534 
2535 		memcpy(attributes->l3.ipv6.hdr.src_addr,
2536 		       input->flow.ipv6_flow.src_ip,
2537 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2538 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
2539 		       input->flow.ipv6_flow.dst_ip,
2540 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2541 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2542 		       mask->ipv6_mask.src_ip,
2543 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2544 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2545 		       mask->ipv6_mask.dst_ip,
2546 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2547 		attributes->items[1] = (struct rte_flow_item){
2548 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
2549 			.spec = &attributes->l3,
2550 			.mask = &attributes->l3_mask,
2551 		};
2552 		break;
2553 	default:
2554 		DRV_LOG(ERR, "port %u invalid flow type%d",
2555 			dev->data->port_id, fdir_filter->input.flow_type);
2556 		rte_errno = ENOTSUP;
2557 		return -rte_errno;
2558 	}
2559 	/* Handle L4. */
2560 	switch (fdir_filter->input.flow_type) {
2561 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2562 		attributes->l4.udp.hdr = (struct udp_hdr){
2563 			.src_port = input->flow.udp4_flow.src_port,
2564 			.dst_port = input->flow.udp4_flow.dst_port,
2565 		};
2566 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2567 			.src_port = mask->src_port_mask,
2568 			.dst_port = mask->dst_port_mask,
2569 		};
2570 		attributes->items[2] = (struct rte_flow_item){
2571 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2572 			.spec = &attributes->l4,
2573 			.mask = &attributes->l4_mask,
2574 		};
2575 		break;
2576 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2577 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2578 			.src_port = input->flow.tcp4_flow.src_port,
2579 			.dst_port = input->flow.tcp4_flow.dst_port,
2580 		};
2581 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2582 			.src_port = mask->src_port_mask,
2583 			.dst_port = mask->dst_port_mask,
2584 		};
2585 		attributes->items[2] = (struct rte_flow_item){
2586 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2587 			.spec = &attributes->l4,
2588 			.mask = &attributes->l4_mask,
2589 		};
2590 		break;
2591 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2592 		attributes->l4.udp.hdr = (struct udp_hdr){
2593 			.src_port = input->flow.udp6_flow.src_port,
2594 			.dst_port = input->flow.udp6_flow.dst_port,
2595 		};
2596 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2597 			.src_port = mask->src_port_mask,
2598 			.dst_port = mask->dst_port_mask,
2599 		};
2600 		attributes->items[2] = (struct rte_flow_item){
2601 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2602 			.spec = &attributes->l4,
2603 			.mask = &attributes->l4_mask,
2604 		};
2605 		break;
2606 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2607 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2608 			.src_port = input->flow.tcp6_flow.src_port,
2609 			.dst_port = input->flow.tcp6_flow.dst_port,
2610 		};
2611 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2612 			.src_port = mask->src_port_mask,
2613 			.dst_port = mask->dst_port_mask,
2614 		};
2615 		attributes->items[2] = (struct rte_flow_item){
2616 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2617 			.spec = &attributes->l4,
2618 			.mask = &attributes->l4_mask,
2619 		};
2620 		break;
2621 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2622 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2623 		break;
2624 	default:
2625 		DRV_LOG(ERR, "port %u invalid flow type%d",
2626 			dev->data->port_id, fdir_filter->input.flow_type);
2627 		rte_errno = ENOTSUP;
2628 		return -rte_errno;
2629 	}
2630 	return 0;
2631 }
2632 
2633 #define FLOW_FDIR_CMP(f1, f2, fld) \
2634 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
2635 
2636 /**
2637  * Compare two FDIR flows. If items and actions are identical, the two flows are
2638  * regarded as same.
2639  *
2640  * @param dev
2641  *   Pointer to Ethernet device.
2642  * @param f1
2643  *   FDIR flow to compare.
2644  * @param f2
2645  *   FDIR flow to compare.
2646  *
2647  * @return
2648  *   Zero on match, 1 otherwise.
2649  */
2650 static int
2651 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
2652 {
2653 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
2654 	    FLOW_FDIR_CMP(f1, f2, l2) ||
2655 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
2656 	    FLOW_FDIR_CMP(f1, f2, l3) ||
2657 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
2658 	    FLOW_FDIR_CMP(f1, f2, l4) ||
2659 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
2660 	    FLOW_FDIR_CMP(f1, f2, actions[0]))
2661 		return 1;
2662 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
2663 	    FLOW_FDIR_CMP(f1, f2, queue))
2664 		return 1;
2665 	return 0;
2666 }
2667 
2668 /**
2669  * Search device flow list to find out a matched FDIR flow.
2670  *
2671  * @param dev
2672  *   Pointer to Ethernet device.
2673  * @param fdir_flow
2674  *   FDIR flow to lookup.
2675  *
2676  * @return
2677  *   Pointer of flow if found, NULL otherwise.
2678  */
2679 static struct rte_flow *
2680 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
2681 {
2682 	struct priv *priv = dev->data->dev_private;
2683 	struct rte_flow *flow = NULL;
2684 
2685 	assert(fdir_flow);
2686 	TAILQ_FOREACH(flow, &priv->flows, next) {
2687 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
2688 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
2689 				dev->data->port_id, (void *)flow);
2690 			break;
2691 		}
2692 	}
2693 	return flow;
2694 }
2695 
2696 /**
2697  * Add new flow director filter and store it in list.
2698  *
2699  * @param dev
2700  *   Pointer to Ethernet device.
2701  * @param fdir_filter
2702  *   Flow director filter to add.
2703  *
2704  * @return
2705  *   0 on success, a negative errno value otherwise and rte_errno is set.
2706  */
2707 static int
2708 flow_fdir_filter_add(struct rte_eth_dev *dev,
2709 		     const struct rte_eth_fdir_filter *fdir_filter)
2710 {
2711 	struct priv *priv = dev->data->dev_private;
2712 	struct mlx5_fdir *fdir_flow;
2713 	struct rte_flow *flow;
2714 	int ret;
2715 
2716 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
2717 	if (!fdir_flow) {
2718 		rte_errno = ENOMEM;
2719 		return -rte_errno;
2720 	}
2721 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
2722 	if (ret)
2723 		goto error;
2724 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
2725 	if (flow) {
2726 		rte_errno = EEXIST;
2727 		goto error;
2728 	}
2729 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
2730 				fdir_flow->items, fdir_flow->actions, NULL);
2731 	if (!flow)
2732 		goto error;
2733 	assert(!flow->fdir);
2734 	flow->fdir = fdir_flow;
2735 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
2736 		dev->data->port_id, (void *)flow);
2737 	return 0;
2738 error:
2739 	rte_free(fdir_flow);
2740 	return -rte_errno;
2741 }
2742 
2743 /**
2744  * Delete specific filter.
2745  *
2746  * @param dev
2747  *   Pointer to Ethernet device.
2748  * @param fdir_filter
2749  *   Filter to be deleted.
2750  *
2751  * @return
2752  *   0 on success, a negative errno value otherwise and rte_errno is set.
2753  */
2754 static int
2755 flow_fdir_filter_delete(struct rte_eth_dev *dev,
2756 			const struct rte_eth_fdir_filter *fdir_filter)
2757 {
2758 	struct priv *priv = dev->data->dev_private;
2759 	struct rte_flow *flow;
2760 	struct mlx5_fdir fdir_flow = {
2761 		.attr.group = 0,
2762 	};
2763 	int ret;
2764 
2765 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
2766 	if (ret)
2767 		return -rte_errno;
2768 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
2769 	if (!flow) {
2770 		rte_errno = ENOENT;
2771 		return -rte_errno;
2772 	}
2773 	flow_list_destroy(dev, &priv->flows, flow);
2774 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
2775 		dev->data->port_id, (void *)flow);
2776 	return 0;
2777 }
2778 
2779 /**
2780  * Update queue for specific filter.
2781  *
2782  * @param dev
2783  *   Pointer to Ethernet device.
2784  * @param fdir_filter
2785  *   Filter to be updated.
2786  *
2787  * @return
2788  *   0 on success, a negative errno value otherwise and rte_errno is set.
2789  */
2790 static int
2791 flow_fdir_filter_update(struct rte_eth_dev *dev,
2792 			const struct rte_eth_fdir_filter *fdir_filter)
2793 {
2794 	int ret;
2795 
2796 	ret = flow_fdir_filter_delete(dev, fdir_filter);
2797 	if (ret)
2798 		return ret;
2799 	return flow_fdir_filter_add(dev, fdir_filter);
2800 }
2801 
2802 /**
2803  * Flush all filters.
2804  *
2805  * @param dev
2806  *   Pointer to Ethernet device.
2807  */
2808 static void
2809 flow_fdir_filter_flush(struct rte_eth_dev *dev)
2810 {
2811 	struct priv *priv = dev->data->dev_private;
2812 
2813 	mlx5_flow_list_flush(dev, &priv->flows);
2814 }
2815 
2816 /**
2817  * Get flow director information.
2818  *
2819  * @param dev
2820  *   Pointer to Ethernet device.
2821  * @param[out] fdir_info
2822  *   Resulting flow director information.
2823  */
2824 static void
2825 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2826 {
2827 	struct rte_eth_fdir_masks *mask =
2828 		&dev->data->dev_conf.fdir_conf.mask;
2829 
2830 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2831 	fdir_info->guarant_spc = 0;
2832 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2833 	fdir_info->max_flexpayload = 0;
2834 	fdir_info->flow_types_mask[0] = 0;
2835 	fdir_info->flex_payload_unit = 0;
2836 	fdir_info->max_flex_payload_segment_num = 0;
2837 	fdir_info->flex_payload_limit = 0;
2838 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2839 }
2840 
2841 /**
2842  * Deal with flow director operations.
2843  *
2844  * @param dev
2845  *   Pointer to Ethernet device.
2846  * @param filter_op
2847  *   Operation to perform.
2848  * @param arg
2849  *   Pointer to operation-specific structure.
2850  *
2851  * @return
2852  *   0 on success, a negative errno value otherwise and rte_errno is set.
2853  */
2854 static int
2855 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2856 		    void *arg)
2857 {
2858 	enum rte_fdir_mode fdir_mode =
2859 		dev->data->dev_conf.fdir_conf.mode;
2860 
2861 	if (filter_op == RTE_ETH_FILTER_NOP)
2862 		return 0;
2863 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2864 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2865 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
2866 			dev->data->port_id, fdir_mode);
2867 		rte_errno = EINVAL;
2868 		return -rte_errno;
2869 	}
2870 	switch (filter_op) {
2871 	case RTE_ETH_FILTER_ADD:
2872 		return flow_fdir_filter_add(dev, arg);
2873 	case RTE_ETH_FILTER_UPDATE:
2874 		return flow_fdir_filter_update(dev, arg);
2875 	case RTE_ETH_FILTER_DELETE:
2876 		return flow_fdir_filter_delete(dev, arg);
2877 	case RTE_ETH_FILTER_FLUSH:
2878 		flow_fdir_filter_flush(dev);
2879 		break;
2880 	case RTE_ETH_FILTER_INFO:
2881 		flow_fdir_info_get(dev, arg);
2882 		break;
2883 	default:
2884 		DRV_LOG(DEBUG, "port %u unknown operation %u",
2885 			dev->data->port_id, filter_op);
2886 		rte_errno = EINVAL;
2887 		return -rte_errno;
2888 	}
2889 	return 0;
2890 }
2891 
2892 /**
2893  * Manage filter operations.
2894  *
2895  * @param dev
2896  *   Pointer to Ethernet device structure.
2897  * @param filter_type
2898  *   Filter type.
2899  * @param filter_op
2900  *   Operation to perform.
2901  * @param arg
2902  *   Pointer to operation-specific structure.
2903  *
2904  * @return
2905  *   0 on success, a negative errno value otherwise and rte_errno is set.
2906  */
2907 int
2908 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2909 		     enum rte_filter_type filter_type,
2910 		     enum rte_filter_op filter_op,
2911 		     void *arg)
2912 {
2913 	switch (filter_type) {
2914 	case RTE_ETH_FILTER_GENERIC:
2915 		if (filter_op != RTE_ETH_FILTER_GET) {
2916 			rte_errno = EINVAL;
2917 			return -rte_errno;
2918 		}
2919 		*(const void **)arg = &mlx5_flow_ops;
2920 		return 0;
2921 	case RTE_ETH_FILTER_FDIR:
2922 		return flow_fdir_ctrl_func(dev, filter_op, arg);
2923 	default:
2924 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
2925 			dev->data->port_id, filter_type);
2926 		rte_errno = ENOTSUP;
2927 		return -rte_errno;
2928 	}
2929 	return 0;
2930 }
2931