xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision a7a0365565a41db2272e2742afb74d4c18160e3e)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 
12 /* Verbs header. */
13 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
14 #ifdef PEDANTIC
15 #pragma GCC diagnostic ignored "-Wpedantic"
16 #endif
17 #include <infiniband/verbs.h>
18 #ifdef PEDANTIC
19 #pragma GCC diagnostic error "-Wpedantic"
20 #endif
21 
22 #include <rte_common.h>
23 #include <rte_ether.h>
24 #include <rte_ethdev_driver.h>
25 #include <rte_flow.h>
26 #include <rte_flow_driver.h>
27 #include <rte_malloc.h>
28 #include <rte_ip.h>
29 
30 #include "mlx5.h"
31 #include "mlx5_defs.h"
32 #include "mlx5_flow.h"
33 #include "mlx5_glue.h"
34 #include "mlx5_prm.h"
35 #include "mlx5_rxtx.h"
36 
37 /* Dev ops structure defined in mlx5.c */
38 extern const struct eth_dev_ops mlx5_dev_ops;
39 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
40 
41 /** Device flow drivers. */
42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
43 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
44 #endif
45 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
46 
47 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
48 
49 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
50 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
51 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
52 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
53 #endif
54 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
55 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
56 };
57 
58 enum mlx5_expansion {
59 	MLX5_EXPANSION_ROOT,
60 	MLX5_EXPANSION_ROOT_OUTER,
61 	MLX5_EXPANSION_ROOT_ETH_VLAN,
62 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
63 	MLX5_EXPANSION_OUTER_ETH,
64 	MLX5_EXPANSION_OUTER_ETH_VLAN,
65 	MLX5_EXPANSION_OUTER_VLAN,
66 	MLX5_EXPANSION_OUTER_IPV4,
67 	MLX5_EXPANSION_OUTER_IPV4_UDP,
68 	MLX5_EXPANSION_OUTER_IPV4_TCP,
69 	MLX5_EXPANSION_OUTER_IPV6,
70 	MLX5_EXPANSION_OUTER_IPV6_UDP,
71 	MLX5_EXPANSION_OUTER_IPV6_TCP,
72 	MLX5_EXPANSION_VXLAN,
73 	MLX5_EXPANSION_VXLAN_GPE,
74 	MLX5_EXPANSION_GRE,
75 	MLX5_EXPANSION_MPLS,
76 	MLX5_EXPANSION_ETH,
77 	MLX5_EXPANSION_ETH_VLAN,
78 	MLX5_EXPANSION_VLAN,
79 	MLX5_EXPANSION_IPV4,
80 	MLX5_EXPANSION_IPV4_UDP,
81 	MLX5_EXPANSION_IPV4_TCP,
82 	MLX5_EXPANSION_IPV6,
83 	MLX5_EXPANSION_IPV6_UDP,
84 	MLX5_EXPANSION_IPV6_TCP,
85 };
86 
87 /** Supported expansion of items. */
88 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
89 	[MLX5_EXPANSION_ROOT] = {
90 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
91 						 MLX5_EXPANSION_IPV4,
92 						 MLX5_EXPANSION_IPV6),
93 		.type = RTE_FLOW_ITEM_TYPE_END,
94 	},
95 	[MLX5_EXPANSION_ROOT_OUTER] = {
96 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
97 						 MLX5_EXPANSION_OUTER_IPV4,
98 						 MLX5_EXPANSION_OUTER_IPV6),
99 		.type = RTE_FLOW_ITEM_TYPE_END,
100 	},
101 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
102 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
103 		.type = RTE_FLOW_ITEM_TYPE_END,
104 	},
105 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
106 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
107 		.type = RTE_FLOW_ITEM_TYPE_END,
108 	},
109 	[MLX5_EXPANSION_OUTER_ETH] = {
110 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
111 						 MLX5_EXPANSION_OUTER_IPV6,
112 						 MLX5_EXPANSION_MPLS),
113 		.type = RTE_FLOW_ITEM_TYPE_ETH,
114 		.rss_types = 0,
115 	},
116 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
117 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
118 		.type = RTE_FLOW_ITEM_TYPE_ETH,
119 		.rss_types = 0,
120 	},
121 	[MLX5_EXPANSION_OUTER_VLAN] = {
122 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
123 						 MLX5_EXPANSION_OUTER_IPV6),
124 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
125 	},
126 	[MLX5_EXPANSION_OUTER_IPV4] = {
127 		.next = RTE_FLOW_EXPAND_RSS_NEXT
128 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
129 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
130 			 MLX5_EXPANSION_GRE),
131 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
132 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
133 			ETH_RSS_NONFRAG_IPV4_OTHER,
134 	},
135 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
136 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
137 						 MLX5_EXPANSION_VXLAN_GPE),
138 		.type = RTE_FLOW_ITEM_TYPE_UDP,
139 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
140 	},
141 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
142 		.type = RTE_FLOW_ITEM_TYPE_TCP,
143 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
144 	},
145 	[MLX5_EXPANSION_OUTER_IPV6] = {
146 		.next = RTE_FLOW_EXPAND_RSS_NEXT
147 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
148 			 MLX5_EXPANSION_OUTER_IPV6_TCP),
149 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
150 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
151 			ETH_RSS_NONFRAG_IPV6_OTHER,
152 	},
153 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
154 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
155 						 MLX5_EXPANSION_VXLAN_GPE),
156 		.type = RTE_FLOW_ITEM_TYPE_UDP,
157 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
158 	},
159 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
160 		.type = RTE_FLOW_ITEM_TYPE_TCP,
161 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
162 	},
163 	[MLX5_EXPANSION_VXLAN] = {
164 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
165 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
166 	},
167 	[MLX5_EXPANSION_VXLAN_GPE] = {
168 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
169 						 MLX5_EXPANSION_IPV4,
170 						 MLX5_EXPANSION_IPV6),
171 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
172 	},
173 	[MLX5_EXPANSION_GRE] = {
174 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
175 		.type = RTE_FLOW_ITEM_TYPE_GRE,
176 	},
177 	[MLX5_EXPANSION_MPLS] = {
178 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
179 						 MLX5_EXPANSION_IPV6),
180 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
181 	},
182 	[MLX5_EXPANSION_ETH] = {
183 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
184 						 MLX5_EXPANSION_IPV6),
185 		.type = RTE_FLOW_ITEM_TYPE_ETH,
186 	},
187 	[MLX5_EXPANSION_ETH_VLAN] = {
188 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
189 		.type = RTE_FLOW_ITEM_TYPE_ETH,
190 	},
191 	[MLX5_EXPANSION_VLAN] = {
192 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
193 						 MLX5_EXPANSION_IPV6),
194 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
195 	},
196 	[MLX5_EXPANSION_IPV4] = {
197 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
198 						 MLX5_EXPANSION_IPV4_TCP),
199 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
200 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
201 			ETH_RSS_NONFRAG_IPV4_OTHER,
202 	},
203 	[MLX5_EXPANSION_IPV4_UDP] = {
204 		.type = RTE_FLOW_ITEM_TYPE_UDP,
205 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
206 	},
207 	[MLX5_EXPANSION_IPV4_TCP] = {
208 		.type = RTE_FLOW_ITEM_TYPE_TCP,
209 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
210 	},
211 	[MLX5_EXPANSION_IPV6] = {
212 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
213 						 MLX5_EXPANSION_IPV6_TCP),
214 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
215 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
216 			ETH_RSS_NONFRAG_IPV6_OTHER,
217 	},
218 	[MLX5_EXPANSION_IPV6_UDP] = {
219 		.type = RTE_FLOW_ITEM_TYPE_UDP,
220 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
221 	},
222 	[MLX5_EXPANSION_IPV6_TCP] = {
223 		.type = RTE_FLOW_ITEM_TYPE_TCP,
224 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
225 	},
226 };
227 
228 static const struct rte_flow_ops mlx5_flow_ops = {
229 	.validate = mlx5_flow_validate,
230 	.create = mlx5_flow_create,
231 	.destroy = mlx5_flow_destroy,
232 	.flush = mlx5_flow_flush,
233 	.isolate = mlx5_flow_isolate,
234 	.query = mlx5_flow_query,
235 };
236 
237 /* Convert FDIR request to Generic flow. */
238 struct mlx5_fdir {
239 	struct rte_flow_attr attr;
240 	struct rte_flow_item items[4];
241 	struct rte_flow_item_eth l2;
242 	struct rte_flow_item_eth l2_mask;
243 	union {
244 		struct rte_flow_item_ipv4 ipv4;
245 		struct rte_flow_item_ipv6 ipv6;
246 	} l3;
247 	union {
248 		struct rte_flow_item_ipv4 ipv4;
249 		struct rte_flow_item_ipv6 ipv6;
250 	} l3_mask;
251 	union {
252 		struct rte_flow_item_udp udp;
253 		struct rte_flow_item_tcp tcp;
254 	} l4;
255 	union {
256 		struct rte_flow_item_udp udp;
257 		struct rte_flow_item_tcp tcp;
258 	} l4_mask;
259 	struct rte_flow_action actions[2];
260 	struct rte_flow_action_queue queue;
261 };
262 
263 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
264 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
265 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
266 };
267 
268 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
269 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
270 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
271 	{ 9, 10, 11 }, { 12, 13, 14 },
272 };
273 
274 /* Tunnel information. */
275 struct mlx5_flow_tunnel_info {
276 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
277 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
278 };
279 
280 static struct mlx5_flow_tunnel_info tunnels_info[] = {
281 	{
282 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
283 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
284 	},
285 	{
286 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
287 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
288 	},
289 	{
290 		.tunnel = MLX5_FLOW_LAYER_GRE,
291 		.ptype = RTE_PTYPE_TUNNEL_GRE,
292 	},
293 	{
294 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
295 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
296 	},
297 	{
298 		.tunnel = MLX5_FLOW_LAYER_MPLS,
299 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
300 	},
301 };
302 
303 /**
304  * Discover the maximum number of priority available.
305  *
306  * @param[in] dev
307  *   Pointer to the Ethernet device structure.
308  *
309  * @return
310  *   number of supported flow priority on success, a negative errno
311  *   value otherwise and rte_errno is set.
312  */
313 int
314 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
315 {
316 	struct mlx5_priv *priv = dev->data->dev_private;
317 	struct {
318 		struct ibv_flow_attr attr;
319 		struct ibv_flow_spec_eth eth;
320 		struct ibv_flow_spec_action_drop drop;
321 	} flow_attr = {
322 		.attr = {
323 			.num_of_specs = 2,
324 			.port = (uint8_t)priv->ibv_port,
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 	mlx5_hrxq_drop_release(dev);
354 	switch (priority) {
355 	case 8:
356 		priority = RTE_DIM(priority_map_3);
357 		break;
358 	case 16:
359 		priority = RTE_DIM(priority_map_5);
360 		break;
361 	default:
362 		rte_errno = ENOTSUP;
363 		DRV_LOG(ERR,
364 			"port %u verbs maximum priority: %d expected 8/16",
365 			dev->data->port_id, priority);
366 		return -rte_errno;
367 	}
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 mlx5_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 mlx5_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 mlx5_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 mlx5_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_errno 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_errno 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 mlx5_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 (!priv->rxqs_n)
850 		return rte_flow_error_set(error, EINVAL,
851 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
852 					  NULL, "No Rx queues configured");
853 	if (queue->index >= priv->rxqs_n)
854 		return rte_flow_error_set(error, EINVAL,
855 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
856 					  &queue->index,
857 					  "queue index out of range");
858 	if (!(*priv->rxqs)[queue->index])
859 		return rte_flow_error_set(error, EINVAL,
860 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
861 					  &queue->index,
862 					  "queue is not configured");
863 	if (attr->egress)
864 		return rte_flow_error_set(error, ENOTSUP,
865 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
866 					  "queue action not supported for "
867 					  "egress");
868 	return 0;
869 }
870 
871 /*
872  * Validate the rss action.
873  *
874  * @param[in] action
875  *   Pointer to the queue action.
876  * @param[in] action_flags
877  *   Bit-fields that holds the actions detected until now.
878  * @param[in] dev
879  *   Pointer to the Ethernet device structure.
880  * @param[in] attr
881  *   Attributes of flow that includes this action.
882  * @param[in] item_flags
883  *   Items that were detected.
884  * @param[out] error
885  *   Pointer to error structure.
886  *
887  * @return
888  *   0 on success, a negative errno value otherwise and rte_errno is set.
889  */
890 int
891 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
892 			      uint64_t action_flags,
893 			      struct rte_eth_dev *dev,
894 			      const struct rte_flow_attr *attr,
895 			      uint64_t item_flags,
896 			      struct rte_flow_error *error)
897 {
898 	struct mlx5_priv *priv = dev->data->dev_private;
899 	const struct rte_flow_action_rss *rss = action->conf;
900 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
901 	unsigned int i;
902 
903 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
904 		return rte_flow_error_set(error, EINVAL,
905 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
906 					  "can't have 2 fate actions"
907 					  " in same flow");
908 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
909 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
910 		return rte_flow_error_set(error, ENOTSUP,
911 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
912 					  &rss->func,
913 					  "RSS hash function not supported");
914 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
915 	if (rss->level > 2)
916 #else
917 	if (rss->level > 1)
918 #endif
919 		return rte_flow_error_set(error, ENOTSUP,
920 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
921 					  &rss->level,
922 					  "tunnel RSS is not supported");
923 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
924 	if (rss->key_len == 0 && rss->key != NULL)
925 		return rte_flow_error_set(error, ENOTSUP,
926 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
927 					  &rss->key_len,
928 					  "RSS hash key length 0");
929 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
930 		return rte_flow_error_set(error, ENOTSUP,
931 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
932 					  &rss->key_len,
933 					  "RSS hash key too small");
934 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
935 		return rte_flow_error_set(error, ENOTSUP,
936 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
937 					  &rss->key_len,
938 					  "RSS hash key too large");
939 	if (rss->queue_num > priv->config.ind_table_max_size)
940 		return rte_flow_error_set(error, ENOTSUP,
941 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
942 					  &rss->queue_num,
943 					  "number of queues too large");
944 	if (rss->types & MLX5_RSS_HF_MASK)
945 		return rte_flow_error_set(error, ENOTSUP,
946 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
947 					  &rss->types,
948 					  "some RSS protocols are not"
949 					  " supported");
950 	if (!priv->rxqs_n)
951 		return rte_flow_error_set(error, EINVAL,
952 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
953 					  NULL, "No Rx queues configured");
954 	if (!rss->queue_num)
955 		return rte_flow_error_set(error, EINVAL,
956 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
957 					  NULL, "No queues configured");
958 	for (i = 0; i != rss->queue_num; ++i) {
959 		if (!(*priv->rxqs)[rss->queue[i]])
960 			return rte_flow_error_set
961 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
962 				 &rss->queue[i], "queue is not configured");
963 	}
964 	if (attr->egress)
965 		return rte_flow_error_set(error, ENOTSUP,
966 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
967 					  "rss action not supported for "
968 					  "egress");
969 	if (rss->level > 1 &&  !tunnel)
970 		return rte_flow_error_set(error, EINVAL,
971 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
972 					  "inner RSS is not supported for "
973 					  "non-tunnel flows");
974 	return 0;
975 }
976 
977 /*
978  * Validate the count action.
979  *
980  * @param[in] dev
981  *   Pointer to the Ethernet device structure.
982  * @param[in] attr
983  *   Attributes of flow that includes this action.
984  * @param[out] error
985  *   Pointer to error structure.
986  *
987  * @return
988  *   0 on success, a negative errno value otherwise and rte_errno is set.
989  */
990 int
991 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
992 				const struct rte_flow_attr *attr,
993 				struct rte_flow_error *error)
994 {
995 	if (attr->egress)
996 		return rte_flow_error_set(error, ENOTSUP,
997 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
998 					  "count action not supported for "
999 					  "egress");
1000 	return 0;
1001 }
1002 
1003 /**
1004  * Verify the @p attributes will be correctly understood by the NIC and store
1005  * them in the @p flow if everything is correct.
1006  *
1007  * @param[in] dev
1008  *   Pointer to the Ethernet device structure.
1009  * @param[in] attributes
1010  *   Pointer to flow attributes
1011  * @param[out] error
1012  *   Pointer to error structure.
1013  *
1014  * @return
1015  *   0 on success, a negative errno value otherwise and rte_errno is set.
1016  */
1017 int
1018 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1019 			      const struct rte_flow_attr *attributes,
1020 			      struct rte_flow_error *error)
1021 {
1022 	struct mlx5_priv *priv = dev->data->dev_private;
1023 	uint32_t priority_max = priv->config.flow_prio - 1;
1024 
1025 	if (attributes->group)
1026 		return rte_flow_error_set(error, ENOTSUP,
1027 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1028 					  NULL, "groups is not supported");
1029 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1030 	    attributes->priority >= priority_max)
1031 		return rte_flow_error_set(error, ENOTSUP,
1032 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1033 					  NULL, "priority out of range");
1034 	if (attributes->egress)
1035 		return rte_flow_error_set(error, ENOTSUP,
1036 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1037 					  "egress is not supported");
1038 	if (attributes->transfer && !priv->config.dv_esw_en)
1039 		return rte_flow_error_set(error, ENOTSUP,
1040 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1041 					  NULL, "transfer is not supported");
1042 	if (!attributes->ingress)
1043 		return rte_flow_error_set(error, EINVAL,
1044 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1045 					  NULL,
1046 					  "ingress attribute is mandatory");
1047 	return 0;
1048 }
1049 
1050 /**
1051  * Validate ICMP6 item.
1052  *
1053  * @param[in] item
1054  *   Item specification.
1055  * @param[in] item_flags
1056  *   Bit-fields that holds the items detected until now.
1057  * @param[out] error
1058  *   Pointer to error structure.
1059  *
1060  * @return
1061  *   0 on success, a negative errno value otherwise and rte_errno is set.
1062  */
1063 int
1064 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1065 			       uint64_t item_flags,
1066 			       uint8_t target_protocol,
1067 			       struct rte_flow_error *error)
1068 {
1069 	const struct rte_flow_item_icmp6 *mask = item->mask;
1070 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1071 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1072 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1073 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1074 				      MLX5_FLOW_LAYER_OUTER_L4;
1075 	int ret;
1076 
1077 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1078 		return rte_flow_error_set(error, EINVAL,
1079 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1080 					  "protocol filtering not compatible"
1081 					  " with ICMP6 layer");
1082 	if (!(item_flags & l3m))
1083 		return rte_flow_error_set(error, EINVAL,
1084 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1085 					  "IPv6 is mandatory to filter on"
1086 					  " ICMP6");
1087 	if (item_flags & l4m)
1088 		return rte_flow_error_set(error, EINVAL,
1089 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1090 					  "multiple L4 layers not supported");
1091 	if (!mask)
1092 		mask = &rte_flow_item_icmp6_mask;
1093 	ret = mlx5_flow_item_acceptable
1094 		(item, (const uint8_t *)mask,
1095 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1096 		 sizeof(struct rte_flow_item_icmp6), error);
1097 	if (ret < 0)
1098 		return ret;
1099 	return 0;
1100 }
1101 
1102 /**
1103  * Validate ICMP item.
1104  *
1105  * @param[in] item
1106  *   Item specification.
1107  * @param[in] item_flags
1108  *   Bit-fields that holds the items detected until now.
1109  * @param[out] error
1110  *   Pointer to error structure.
1111  *
1112  * @return
1113  *   0 on success, a negative errno value otherwise and rte_errno is set.
1114  */
1115 int
1116 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1117 			     uint64_t item_flags,
1118 			     uint8_t target_protocol,
1119 			     struct rte_flow_error *error)
1120 {
1121 	const struct rte_flow_item_icmp *mask = item->mask;
1122 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1123 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1124 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1125 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1126 				      MLX5_FLOW_LAYER_OUTER_L4;
1127 	int ret;
1128 
1129 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1130 		return rte_flow_error_set(error, EINVAL,
1131 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1132 					  "protocol filtering not compatible"
1133 					  " with ICMP layer");
1134 	if (!(item_flags & l3m))
1135 		return rte_flow_error_set(error, EINVAL,
1136 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1137 					  "IPv4 is mandatory to filter"
1138 					  " on ICMP");
1139 	if (item_flags & l4m)
1140 		return rte_flow_error_set(error, EINVAL,
1141 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1142 					  "multiple L4 layers not supported");
1143 	if (!mask)
1144 		mask = &rte_flow_item_icmp_mask;
1145 	ret = mlx5_flow_item_acceptable
1146 		(item, (const uint8_t *)mask,
1147 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1148 		 sizeof(struct rte_flow_item_icmp), error);
1149 	if (ret < 0)
1150 		return ret;
1151 	return 0;
1152 }
1153 
1154 /**
1155  * Validate Ethernet item.
1156  *
1157  * @param[in] item
1158  *   Item specification.
1159  * @param[in] item_flags
1160  *   Bit-fields that holds the items detected until now.
1161  * @param[out] error
1162  *   Pointer to error structure.
1163  *
1164  * @return
1165  *   0 on success, a negative errno value otherwise and rte_errno is set.
1166  */
1167 int
1168 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1169 			    uint64_t item_flags,
1170 			    struct rte_flow_error *error)
1171 {
1172 	const struct rte_flow_item_eth *mask = item->mask;
1173 	const struct rte_flow_item_eth nic_mask = {
1174 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1175 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1176 		.type = RTE_BE16(0xffff),
1177 	};
1178 	int ret;
1179 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1180 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1181 				       MLX5_FLOW_LAYER_OUTER_L2;
1182 
1183 	if (item_flags & ethm)
1184 		return rte_flow_error_set(error, ENOTSUP,
1185 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1186 					  "multiple L2 layers not supported");
1187 	if (!mask)
1188 		mask = &rte_flow_item_eth_mask;
1189 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1190 					(const uint8_t *)&nic_mask,
1191 					sizeof(struct rte_flow_item_eth),
1192 					error);
1193 	return ret;
1194 }
1195 
1196 /**
1197  * Validate VLAN item.
1198  *
1199  * @param[in] item
1200  *   Item specification.
1201  * @param[in] item_flags
1202  *   Bit-fields that holds the items detected until now.
1203  * @param[out] error
1204  *   Pointer to error structure.
1205  *
1206  * @return
1207  *   0 on success, a negative errno value otherwise and rte_errno is set.
1208  */
1209 int
1210 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1211 			     uint64_t item_flags,
1212 			     struct rte_flow_error *error)
1213 {
1214 	const struct rte_flow_item_vlan *spec = item->spec;
1215 	const struct rte_flow_item_vlan *mask = item->mask;
1216 	const struct rte_flow_item_vlan nic_mask = {
1217 		.tci = RTE_BE16(0x0fff),
1218 		.inner_type = RTE_BE16(0xffff),
1219 	};
1220 	uint16_t vlan_tag = 0;
1221 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1222 	int ret;
1223 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1224 					MLX5_FLOW_LAYER_INNER_L4) :
1225 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1226 					MLX5_FLOW_LAYER_OUTER_L4);
1227 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1228 					MLX5_FLOW_LAYER_OUTER_VLAN;
1229 
1230 	if (item_flags & vlanm)
1231 		return rte_flow_error_set(error, EINVAL,
1232 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1233 					  "multiple VLAN layers not supported");
1234 	else if ((item_flags & l34m) != 0)
1235 		return rte_flow_error_set(error, EINVAL,
1236 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1237 					  "L2 layer cannot follow L3/L4 layer");
1238 	if (!mask)
1239 		mask = &rte_flow_item_vlan_mask;
1240 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1241 					(const uint8_t *)&nic_mask,
1242 					sizeof(struct rte_flow_item_vlan),
1243 					error);
1244 	if (ret)
1245 		return ret;
1246 	if (spec) {
1247 		vlan_tag = spec->tci;
1248 		vlan_tag &= mask->tci;
1249 	}
1250 	/*
1251 	 * From verbs perspective an empty VLAN is equivalent
1252 	 * to a packet without VLAN layer.
1253 	 */
1254 	if (!vlan_tag)
1255 		return rte_flow_error_set(error, EINVAL,
1256 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1257 					  item->spec,
1258 					  "VLAN cannot be empty");
1259 	return 0;
1260 }
1261 
1262 /**
1263  * Validate IPV4 item.
1264  *
1265  * @param[in] item
1266  *   Item specification.
1267  * @param[in] item_flags
1268  *   Bit-fields that holds the items detected until now.
1269  * @param[in] acc_mask
1270  *   Acceptable mask, if NULL default internal default mask
1271  *   will be used to check whether item fields are supported.
1272  * @param[out] error
1273  *   Pointer to error structure.
1274  *
1275  * @return
1276  *   0 on success, a negative errno value otherwise and rte_errno is set.
1277  */
1278 int
1279 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1280 			     uint64_t item_flags,
1281 			     const struct rte_flow_item_ipv4 *acc_mask,
1282 			     struct rte_flow_error *error)
1283 {
1284 	const struct rte_flow_item_ipv4 *mask = item->mask;
1285 	const struct rte_flow_item_ipv4 nic_mask = {
1286 		.hdr = {
1287 			.src_addr = RTE_BE32(0xffffffff),
1288 			.dst_addr = RTE_BE32(0xffffffff),
1289 			.type_of_service = 0xff,
1290 			.next_proto_id = 0xff,
1291 		},
1292 	};
1293 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1294 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1295 				      MLX5_FLOW_LAYER_OUTER_L3;
1296 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1297 				      MLX5_FLOW_LAYER_OUTER_L4;
1298 	int ret;
1299 
1300 	if (item_flags & l3m)
1301 		return rte_flow_error_set(error, ENOTSUP,
1302 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1303 					  "multiple L3 layers not supported");
1304 	else if (item_flags & l4m)
1305 		return rte_flow_error_set(error, EINVAL,
1306 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1307 					  "L3 cannot follow an L4 layer.");
1308 	if (!mask)
1309 		mask = &rte_flow_item_ipv4_mask;
1310 	else if (mask->hdr.next_proto_id != 0 &&
1311 		 mask->hdr.next_proto_id != 0xff)
1312 		return rte_flow_error_set(error, EINVAL,
1313 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1314 					  "partial mask is not supported"
1315 					  " for protocol");
1316 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1317 					acc_mask ? (const uint8_t *)acc_mask
1318 						 : (const uint8_t *)&nic_mask,
1319 					sizeof(struct rte_flow_item_ipv4),
1320 					error);
1321 	if (ret < 0)
1322 		return ret;
1323 	return 0;
1324 }
1325 
1326 /**
1327  * Validate IPV6 item.
1328  *
1329  * @param[in] item
1330  *   Item specification.
1331  * @param[in] item_flags
1332  *   Bit-fields that holds the items detected until now.
1333  * @param[in] acc_mask
1334  *   Acceptable mask, if NULL default internal default mask
1335  *   will be used to check whether item fields are supported.
1336  * @param[out] error
1337  *   Pointer to error structure.
1338  *
1339  * @return
1340  *   0 on success, a negative errno value otherwise and rte_errno is set.
1341  */
1342 int
1343 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1344 			     uint64_t item_flags,
1345 			     const struct rte_flow_item_ipv6 *acc_mask,
1346 			     struct rte_flow_error *error)
1347 {
1348 	const struct rte_flow_item_ipv6 *mask = item->mask;
1349 	const struct rte_flow_item_ipv6 nic_mask = {
1350 		.hdr = {
1351 			.src_addr =
1352 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1353 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1354 			.dst_addr =
1355 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1356 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1357 			.vtc_flow = RTE_BE32(0xffffffff),
1358 			.proto = 0xff,
1359 			.hop_limits = 0xff,
1360 		},
1361 	};
1362 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1363 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1364 				      MLX5_FLOW_LAYER_OUTER_L3;
1365 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1366 				      MLX5_FLOW_LAYER_OUTER_L4;
1367 	int ret;
1368 
1369 	if (item_flags & l3m)
1370 		return rte_flow_error_set(error, ENOTSUP,
1371 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1372 					  "multiple L3 layers not supported");
1373 	else if (item_flags & l4m)
1374 		return rte_flow_error_set(error, EINVAL,
1375 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1376 					  "L3 cannot follow an L4 layer.");
1377 	if (!mask)
1378 		mask = &rte_flow_item_ipv6_mask;
1379 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1380 					acc_mask ? (const uint8_t *)acc_mask
1381 						 : (const uint8_t *)&nic_mask,
1382 					sizeof(struct rte_flow_item_ipv6),
1383 					error);
1384 	if (ret < 0)
1385 		return ret;
1386 	return 0;
1387 }
1388 
1389 /**
1390  * Validate UDP item.
1391  *
1392  * @param[in] item
1393  *   Item specification.
1394  * @param[in] item_flags
1395  *   Bit-fields that holds the items detected until now.
1396  * @param[in] target_protocol
1397  *   The next protocol in the previous item.
1398  * @param[in] flow_mask
1399  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1400  * @param[out] error
1401  *   Pointer to error structure.
1402  *
1403  * @return
1404  *   0 on success, a negative errno value otherwise and rte_errno is set.
1405  */
1406 int
1407 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1408 			    uint64_t item_flags,
1409 			    uint8_t target_protocol,
1410 			    struct rte_flow_error *error)
1411 {
1412 	const struct rte_flow_item_udp *mask = item->mask;
1413 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1414 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1415 				      MLX5_FLOW_LAYER_OUTER_L3;
1416 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1417 				      MLX5_FLOW_LAYER_OUTER_L4;
1418 	int ret;
1419 
1420 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1421 		return rte_flow_error_set(error, EINVAL,
1422 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1423 					  "protocol filtering not compatible"
1424 					  " with UDP layer");
1425 	if (!(item_flags & l3m))
1426 		return rte_flow_error_set(error, EINVAL,
1427 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1428 					  "L3 is mandatory to filter on L4");
1429 	if (item_flags & l4m)
1430 		return rte_flow_error_set(error, EINVAL,
1431 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1432 					  "multiple L4 layers not supported");
1433 	if (!mask)
1434 		mask = &rte_flow_item_udp_mask;
1435 	ret = mlx5_flow_item_acceptable
1436 		(item, (const uint8_t *)mask,
1437 		 (const uint8_t *)&rte_flow_item_udp_mask,
1438 		 sizeof(struct rte_flow_item_udp), error);
1439 	if (ret < 0)
1440 		return ret;
1441 	return 0;
1442 }
1443 
1444 /**
1445  * Validate TCP item.
1446  *
1447  * @param[in] item
1448  *   Item specification.
1449  * @param[in] item_flags
1450  *   Bit-fields that holds the items detected until now.
1451  * @param[in] target_protocol
1452  *   The next protocol in the previous item.
1453  * @param[out] error
1454  *   Pointer to error structure.
1455  *
1456  * @return
1457  *   0 on success, a negative errno value otherwise and rte_errno is set.
1458  */
1459 int
1460 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1461 			    uint64_t item_flags,
1462 			    uint8_t target_protocol,
1463 			    const struct rte_flow_item_tcp *flow_mask,
1464 			    struct rte_flow_error *error)
1465 {
1466 	const struct rte_flow_item_tcp *mask = item->mask;
1467 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1468 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1469 				      MLX5_FLOW_LAYER_OUTER_L3;
1470 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1471 				      MLX5_FLOW_LAYER_OUTER_L4;
1472 	int ret;
1473 
1474 	assert(flow_mask);
1475 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1476 		return rte_flow_error_set(error, EINVAL,
1477 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1478 					  "protocol filtering not compatible"
1479 					  " with TCP layer");
1480 	if (!(item_flags & l3m))
1481 		return rte_flow_error_set(error, EINVAL,
1482 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1483 					  "L3 is mandatory to filter on L4");
1484 	if (item_flags & l4m)
1485 		return rte_flow_error_set(error, EINVAL,
1486 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1487 					  "multiple L4 layers not supported");
1488 	if (!mask)
1489 		mask = &rte_flow_item_tcp_mask;
1490 	ret = mlx5_flow_item_acceptable
1491 		(item, (const uint8_t *)mask,
1492 		 (const uint8_t *)flow_mask,
1493 		 sizeof(struct rte_flow_item_tcp), error);
1494 	if (ret < 0)
1495 		return ret;
1496 	return 0;
1497 }
1498 
1499 /**
1500  * Validate VXLAN item.
1501  *
1502  * @param[in] item
1503  *   Item specification.
1504  * @param[in] item_flags
1505  *   Bit-fields that holds the items detected until now.
1506  * @param[in] target_protocol
1507  *   The next protocol in the previous item.
1508  * @param[out] error
1509  *   Pointer to error structure.
1510  *
1511  * @return
1512  *   0 on success, a negative errno value otherwise and rte_errno is set.
1513  */
1514 int
1515 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1516 			      uint64_t item_flags,
1517 			      struct rte_flow_error *error)
1518 {
1519 	const struct rte_flow_item_vxlan *spec = item->spec;
1520 	const struct rte_flow_item_vxlan *mask = item->mask;
1521 	int ret;
1522 	union vni {
1523 		uint32_t vlan_id;
1524 		uint8_t vni[4];
1525 	} id = { .vlan_id = 0, };
1526 	uint32_t vlan_id = 0;
1527 
1528 
1529 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1530 		return rte_flow_error_set(error, ENOTSUP,
1531 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1532 					  "multiple tunnel layers not"
1533 					  " supported");
1534 	/*
1535 	 * Verify only UDPv4 is present as defined in
1536 	 * https://tools.ietf.org/html/rfc7348
1537 	 */
1538 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1539 		return rte_flow_error_set(error, EINVAL,
1540 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1541 					  "no outer UDP layer found");
1542 	if (!mask)
1543 		mask = &rte_flow_item_vxlan_mask;
1544 	ret = mlx5_flow_item_acceptable
1545 		(item, (const uint8_t *)mask,
1546 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1547 		 sizeof(struct rte_flow_item_vxlan),
1548 		 error);
1549 	if (ret < 0)
1550 		return ret;
1551 	if (spec) {
1552 		memcpy(&id.vni[1], spec->vni, 3);
1553 		vlan_id = id.vlan_id;
1554 		memcpy(&id.vni[1], mask->vni, 3);
1555 		vlan_id &= id.vlan_id;
1556 	}
1557 	/*
1558 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1559 	 * only this layer is defined in the Verbs specification it is
1560 	 * interpreted as wildcard and all packets will match this
1561 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1562 	 * udp), all packets matching the layers before will also
1563 	 * match this rule.  To avoid such situation, VNI 0 is
1564 	 * currently refused.
1565 	 */
1566 	if (!vlan_id)
1567 		return rte_flow_error_set(error, ENOTSUP,
1568 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1569 					  "VXLAN vni cannot be 0");
1570 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1571 		return rte_flow_error_set(error, ENOTSUP,
1572 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1573 					  "VXLAN tunnel must be fully defined");
1574 	return 0;
1575 }
1576 
1577 /**
1578  * Validate VXLAN_GPE item.
1579  *
1580  * @param[in] item
1581  *   Item specification.
1582  * @param[in] item_flags
1583  *   Bit-fields that holds the items detected until now.
1584  * @param[in] priv
1585  *   Pointer to the private data structure.
1586  * @param[in] target_protocol
1587  *   The next protocol in the previous item.
1588  * @param[out] error
1589  *   Pointer to error structure.
1590  *
1591  * @return
1592  *   0 on success, a negative errno value otherwise and rte_errno is set.
1593  */
1594 int
1595 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1596 				  uint64_t item_flags,
1597 				  struct rte_eth_dev *dev,
1598 				  struct rte_flow_error *error)
1599 {
1600 	struct mlx5_priv *priv = dev->data->dev_private;
1601 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1602 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1603 	int ret;
1604 	union vni {
1605 		uint32_t vlan_id;
1606 		uint8_t vni[4];
1607 	} id = { .vlan_id = 0, };
1608 	uint32_t vlan_id = 0;
1609 
1610 	if (!priv->config.l3_vxlan_en)
1611 		return rte_flow_error_set(error, ENOTSUP,
1612 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1613 					  "L3 VXLAN is not enabled by device"
1614 					  " parameter and/or not configured in"
1615 					  " firmware");
1616 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1617 		return rte_flow_error_set(error, ENOTSUP,
1618 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1619 					  "multiple tunnel layers not"
1620 					  " supported");
1621 	/*
1622 	 * Verify only UDPv4 is present as defined in
1623 	 * https://tools.ietf.org/html/rfc7348
1624 	 */
1625 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1626 		return rte_flow_error_set(error, EINVAL,
1627 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1628 					  "no outer UDP layer found");
1629 	if (!mask)
1630 		mask = &rte_flow_item_vxlan_gpe_mask;
1631 	ret = mlx5_flow_item_acceptable
1632 		(item, (const uint8_t *)mask,
1633 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1634 		 sizeof(struct rte_flow_item_vxlan_gpe),
1635 		 error);
1636 	if (ret < 0)
1637 		return ret;
1638 	if (spec) {
1639 		if (spec->protocol)
1640 			return rte_flow_error_set(error, ENOTSUP,
1641 						  RTE_FLOW_ERROR_TYPE_ITEM,
1642 						  item,
1643 						  "VxLAN-GPE protocol"
1644 						  " not supported");
1645 		memcpy(&id.vni[1], spec->vni, 3);
1646 		vlan_id = id.vlan_id;
1647 		memcpy(&id.vni[1], mask->vni, 3);
1648 		vlan_id &= id.vlan_id;
1649 	}
1650 	/*
1651 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1652 	 * layer is defined in the Verbs specification it is interpreted as
1653 	 * wildcard and all packets will match this rule, if it follows a full
1654 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1655 	 * before will also match this rule.  To avoid such situation, VNI 0
1656 	 * is currently refused.
1657 	 */
1658 	if (!vlan_id)
1659 		return rte_flow_error_set(error, ENOTSUP,
1660 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1661 					  "VXLAN-GPE vni cannot be 0");
1662 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1663 		return rte_flow_error_set(error, ENOTSUP,
1664 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1665 					  "VXLAN-GPE tunnel must be fully"
1666 					  " defined");
1667 	return 0;
1668 }
1669 /**
1670  * Validate GRE Key item.
1671  *
1672  * @param[in] item
1673  *   Item specification.
1674  * @param[in] item_flags
1675  *   Bit flags to mark detected items.
1676  * @param[in] gre_item
1677  *   Pointer to gre_item
1678  * @param[out] error
1679  *   Pointer to error structure.
1680  *
1681  * @return
1682  *   0 on success, a negative errno value otherwise and rte_errno is set.
1683  */
1684 int
1685 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1686 				uint64_t item_flags,
1687 				const struct rte_flow_item *gre_item,
1688 				struct rte_flow_error *error)
1689 {
1690 	const rte_be32_t *mask = item->mask;
1691 	int ret = 0;
1692 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1693 	const struct rte_flow_item_gre *gre_spec = gre_item->spec;
1694 	const struct rte_flow_item_gre *gre_mask = gre_item->mask;
1695 
1696 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1697 		return rte_flow_error_set(error, ENOTSUP,
1698 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1699 					  "Multiple GRE key not support");
1700 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1701 		return rte_flow_error_set(error, ENOTSUP,
1702 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1703 					  "No preceding GRE header");
1704 	if (item_flags & MLX5_FLOW_LAYER_INNER)
1705 		return rte_flow_error_set(error, ENOTSUP,
1706 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1707 					  "GRE key following a wrong item");
1708 	if (!gre_mask)
1709 		gre_mask = &rte_flow_item_gre_mask;
1710 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
1711 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
1712 		return rte_flow_error_set(error, EINVAL,
1713 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1714 					  "Key bit must be on");
1715 
1716 	if (!mask)
1717 		mask = &gre_key_default_mask;
1718 	ret = mlx5_flow_item_acceptable
1719 		(item, (const uint8_t *)mask,
1720 		 (const uint8_t *)&gre_key_default_mask,
1721 		 sizeof(rte_be32_t), error);
1722 	return ret;
1723 }
1724 
1725 /**
1726  * Validate GRE item.
1727  *
1728  * @param[in] item
1729  *   Item specification.
1730  * @param[in] item_flags
1731  *   Bit flags to mark detected items.
1732  * @param[in] target_protocol
1733  *   The next protocol in the previous item.
1734  * @param[out] error
1735  *   Pointer to error structure.
1736  *
1737  * @return
1738  *   0 on success, a negative errno value otherwise and rte_errno is set.
1739  */
1740 int
1741 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1742 			    uint64_t item_flags,
1743 			    uint8_t target_protocol,
1744 			    struct rte_flow_error *error)
1745 {
1746 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1747 	const struct rte_flow_item_gre *mask = item->mask;
1748 	int ret;
1749 	const struct rte_flow_item_gre nic_mask = {
1750 		.c_rsvd0_ver = RTE_BE16(0xB000),
1751 		.protocol = RTE_BE16(UINT16_MAX),
1752 	};
1753 
1754 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1755 		return rte_flow_error_set(error, EINVAL,
1756 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1757 					  "protocol filtering not compatible"
1758 					  " with this GRE layer");
1759 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1760 		return rte_flow_error_set(error, ENOTSUP,
1761 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1762 					  "multiple tunnel layers not"
1763 					  " supported");
1764 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1765 		return rte_flow_error_set(error, ENOTSUP,
1766 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1767 					  "L3 Layer is missing");
1768 	if (!mask)
1769 		mask = &rte_flow_item_gre_mask;
1770 	ret = mlx5_flow_item_acceptable
1771 		(item, (const uint8_t *)mask,
1772 		 (const uint8_t *)&nic_mask,
1773 		 sizeof(struct rte_flow_item_gre), error);
1774 	if (ret < 0)
1775 		return ret;
1776 #ifndef HAVE_MLX5DV_DR
1777 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1778 	if (spec && (spec->protocol & mask->protocol))
1779 		return rte_flow_error_set(error, ENOTSUP,
1780 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1781 					  "without MPLS support the"
1782 					  " specification cannot be used for"
1783 					  " filtering");
1784 #endif
1785 #endif
1786 	return 0;
1787 }
1788 
1789 /**
1790  * Validate MPLS item.
1791  *
1792  * @param[in] dev
1793  *   Pointer to the rte_eth_dev structure.
1794  * @param[in] item
1795  *   Item specification.
1796  * @param[in] item_flags
1797  *   Bit-fields that holds the items detected until now.
1798  * @param[in] prev_layer
1799  *   The protocol layer indicated in previous item.
1800  * @param[out] error
1801  *   Pointer to error structure.
1802  *
1803  * @return
1804  *   0 on success, a negative errno value otherwise and rte_errno is set.
1805  */
1806 int
1807 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
1808 			     const struct rte_flow_item *item __rte_unused,
1809 			     uint64_t item_flags __rte_unused,
1810 			     uint64_t prev_layer __rte_unused,
1811 			     struct rte_flow_error *error)
1812 {
1813 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1814 	const struct rte_flow_item_mpls *mask = item->mask;
1815 	struct mlx5_priv *priv = dev->data->dev_private;
1816 	int ret;
1817 
1818 	if (!priv->config.mpls_en)
1819 		return rte_flow_error_set(error, ENOTSUP,
1820 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1821 					  "MPLS not supported or"
1822 					  " disabled in firmware"
1823 					  " configuration.");
1824 	/* MPLS over IP, UDP, GRE is allowed */
1825 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
1826 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
1827 			    MLX5_FLOW_LAYER_GRE)))
1828 		return rte_flow_error_set(error, EINVAL,
1829 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1830 					  "protocol filtering not compatible"
1831 					  " with MPLS layer");
1832 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
1833 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1834 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
1835 		return rte_flow_error_set(error, ENOTSUP,
1836 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1837 					  "multiple tunnel layers not"
1838 					  " supported");
1839 	if (!mask)
1840 		mask = &rte_flow_item_mpls_mask;
1841 	ret = mlx5_flow_item_acceptable
1842 		(item, (const uint8_t *)mask,
1843 		 (const uint8_t *)&rte_flow_item_mpls_mask,
1844 		 sizeof(struct rte_flow_item_mpls), error);
1845 	if (ret < 0)
1846 		return ret;
1847 	return 0;
1848 #endif
1849 	return rte_flow_error_set(error, ENOTSUP,
1850 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
1851 				  "MPLS is not supported by Verbs, please"
1852 				  " update.");
1853 }
1854 
1855 static int
1856 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1857 		   const struct rte_flow_attr *attr __rte_unused,
1858 		   const struct rte_flow_item items[] __rte_unused,
1859 		   const struct rte_flow_action actions[] __rte_unused,
1860 		   struct rte_flow_error *error)
1861 {
1862 	return rte_flow_error_set(error, ENOTSUP,
1863 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
1864 }
1865 
1866 static struct mlx5_flow *
1867 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1868 		  const struct rte_flow_item items[] __rte_unused,
1869 		  const struct rte_flow_action actions[] __rte_unused,
1870 		  struct rte_flow_error *error)
1871 {
1872 	rte_flow_error_set(error, ENOTSUP,
1873 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
1874 	return NULL;
1875 }
1876 
1877 static int
1878 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1879 		    struct mlx5_flow *dev_flow __rte_unused,
1880 		    const struct rte_flow_attr *attr __rte_unused,
1881 		    const struct rte_flow_item items[] __rte_unused,
1882 		    const struct rte_flow_action actions[] __rte_unused,
1883 		    struct rte_flow_error *error)
1884 {
1885 	return rte_flow_error_set(error, ENOTSUP,
1886 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
1887 }
1888 
1889 static int
1890 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1891 		struct rte_flow *flow __rte_unused,
1892 		struct rte_flow_error *error)
1893 {
1894 	return rte_flow_error_set(error, ENOTSUP,
1895 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
1896 }
1897 
1898 static void
1899 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1900 		 struct rte_flow *flow __rte_unused)
1901 {
1902 }
1903 
1904 static void
1905 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1906 		  struct rte_flow *flow __rte_unused)
1907 {
1908 }
1909 
1910 static int
1911 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1912 		struct rte_flow *flow __rte_unused,
1913 		const struct rte_flow_action *actions __rte_unused,
1914 		void *data __rte_unused,
1915 		struct rte_flow_error *error)
1916 {
1917 	return rte_flow_error_set(error, ENOTSUP,
1918 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
1919 }
1920 
1921 /* Void driver to protect from null pointer reference. */
1922 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1923 	.validate = flow_null_validate,
1924 	.prepare = flow_null_prepare,
1925 	.translate = flow_null_translate,
1926 	.apply = flow_null_apply,
1927 	.remove = flow_null_remove,
1928 	.destroy = flow_null_destroy,
1929 	.query = flow_null_query,
1930 };
1931 
1932 /**
1933  * Select flow driver type according to flow attributes and device
1934  * configuration.
1935  *
1936  * @param[in] dev
1937  *   Pointer to the dev structure.
1938  * @param[in] attr
1939  *   Pointer to the flow attributes.
1940  *
1941  * @return
1942  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1943  */
1944 static enum mlx5_flow_drv_type
1945 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1946 {
1947 	struct mlx5_priv *priv = dev->data->dev_private;
1948 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1949 
1950 	if (attr->transfer && priv->config.dv_esw_en)
1951 		type = MLX5_FLOW_TYPE_DV;
1952 	if (!attr->transfer)
1953 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1954 						 MLX5_FLOW_TYPE_VERBS;
1955 	return type;
1956 }
1957 
1958 #define flow_get_drv_ops(type) flow_drv_ops[type]
1959 
1960 /**
1961  * Flow driver validation API. This abstracts calling driver specific functions.
1962  * The type of flow driver is determined according to flow attributes.
1963  *
1964  * @param[in] dev
1965  *   Pointer to the dev structure.
1966  * @param[in] attr
1967  *   Pointer to the flow attributes.
1968  * @param[in] items
1969  *   Pointer to the list of items.
1970  * @param[in] actions
1971  *   Pointer to the list of actions.
1972  * @param[out] error
1973  *   Pointer to the error structure.
1974  *
1975  * @return
1976  *   0 on success, a negative errno value otherwise and rte_errno is set.
1977  */
1978 static inline int
1979 flow_drv_validate(struct rte_eth_dev *dev,
1980 		  const struct rte_flow_attr *attr,
1981 		  const struct rte_flow_item items[],
1982 		  const struct rte_flow_action actions[],
1983 		  struct rte_flow_error *error)
1984 {
1985 	const struct mlx5_flow_driver_ops *fops;
1986 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1987 
1988 	fops = flow_get_drv_ops(type);
1989 	return fops->validate(dev, attr, items, actions, error);
1990 }
1991 
1992 /**
1993  * Flow driver preparation API. This abstracts calling driver specific
1994  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1995  * calculates the size of memory required for device flow, allocates the memory,
1996  * initializes the device flow and returns the pointer.
1997  *
1998  * @note
1999  *   This function initializes device flow structure such as dv or verbs in
2000  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2001  *   rest. For example, adding returning device flow to flow->dev_flow list and
2002  *   setting backward reference to the flow should be done out of this function.
2003  *   layers field is not filled either.
2004  *
2005  * @param[in] attr
2006  *   Pointer to the flow attributes.
2007  * @param[in] items
2008  *   Pointer to the list of items.
2009  * @param[in] actions
2010  *   Pointer to the list of actions.
2011  * @param[out] error
2012  *   Pointer to the error structure.
2013  *
2014  * @return
2015  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2016  */
2017 static inline struct mlx5_flow *
2018 flow_drv_prepare(const struct rte_flow *flow,
2019 		 const struct rte_flow_attr *attr,
2020 		 const struct rte_flow_item items[],
2021 		 const struct rte_flow_action actions[],
2022 		 struct rte_flow_error *error)
2023 {
2024 	const struct mlx5_flow_driver_ops *fops;
2025 	enum mlx5_flow_drv_type type = flow->drv_type;
2026 
2027 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2028 	fops = flow_get_drv_ops(type);
2029 	return fops->prepare(attr, items, actions, error);
2030 }
2031 
2032 /**
2033  * Flow driver translation API. This abstracts calling driver specific
2034  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2035  * translates a generic flow into a driver flow. flow_drv_prepare() must
2036  * precede.
2037  *
2038  * @note
2039  *   dev_flow->layers could be filled as a result of parsing during translation
2040  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2041  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2042  *   flow->actions could be overwritten even though all the expanded dev_flows
2043  *   have the same actions.
2044  *
2045  * @param[in] dev
2046  *   Pointer to the rte dev structure.
2047  * @param[in, out] dev_flow
2048  *   Pointer to the mlx5 flow.
2049  * @param[in] attr
2050  *   Pointer to the flow attributes.
2051  * @param[in] items
2052  *   Pointer to the list of items.
2053  * @param[in] actions
2054  *   Pointer to the list of actions.
2055  * @param[out] error
2056  *   Pointer to the error structure.
2057  *
2058  * @return
2059  *   0 on success, a negative errno value otherwise and rte_errno is set.
2060  */
2061 static inline int
2062 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2063 		   const struct rte_flow_attr *attr,
2064 		   const struct rte_flow_item items[],
2065 		   const struct rte_flow_action actions[],
2066 		   struct rte_flow_error *error)
2067 {
2068 	const struct mlx5_flow_driver_ops *fops;
2069 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2070 
2071 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2072 	fops = flow_get_drv_ops(type);
2073 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2074 }
2075 
2076 /**
2077  * Flow driver apply API. This abstracts calling driver specific functions.
2078  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2079  * translated driver flows on to device. flow_drv_translate() must precede.
2080  *
2081  * @param[in] dev
2082  *   Pointer to Ethernet device structure.
2083  * @param[in, out] flow
2084  *   Pointer to flow structure.
2085  * @param[out] error
2086  *   Pointer to error structure.
2087  *
2088  * @return
2089  *   0 on success, a negative errno value otherwise and rte_errno is set.
2090  */
2091 static inline int
2092 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2093 	       struct rte_flow_error *error)
2094 {
2095 	const struct mlx5_flow_driver_ops *fops;
2096 	enum mlx5_flow_drv_type type = flow->drv_type;
2097 
2098 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2099 	fops = flow_get_drv_ops(type);
2100 	return fops->apply(dev, flow, error);
2101 }
2102 
2103 /**
2104  * Flow driver remove API. This abstracts calling driver specific functions.
2105  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2106  * on device. All the resources of the flow should be freed by calling
2107  * flow_drv_destroy().
2108  *
2109  * @param[in] dev
2110  *   Pointer to Ethernet device.
2111  * @param[in, out] flow
2112  *   Pointer to flow structure.
2113  */
2114 static inline void
2115 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2116 {
2117 	const struct mlx5_flow_driver_ops *fops;
2118 	enum mlx5_flow_drv_type type = flow->drv_type;
2119 
2120 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2121 	fops = flow_get_drv_ops(type);
2122 	fops->remove(dev, flow);
2123 }
2124 
2125 /**
2126  * Flow driver destroy API. This abstracts calling driver specific functions.
2127  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2128  * on device and releases resources of the flow.
2129  *
2130  * @param[in] dev
2131  *   Pointer to Ethernet device.
2132  * @param[in, out] flow
2133  *   Pointer to flow structure.
2134  */
2135 static inline void
2136 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2137 {
2138 	const struct mlx5_flow_driver_ops *fops;
2139 	enum mlx5_flow_drv_type type = flow->drv_type;
2140 
2141 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2142 	fops = flow_get_drv_ops(type);
2143 	fops->destroy(dev, flow);
2144 }
2145 
2146 /**
2147  * Validate a flow supported by the NIC.
2148  *
2149  * @see rte_flow_validate()
2150  * @see rte_flow_ops
2151  */
2152 int
2153 mlx5_flow_validate(struct rte_eth_dev *dev,
2154 		   const struct rte_flow_attr *attr,
2155 		   const struct rte_flow_item items[],
2156 		   const struct rte_flow_action actions[],
2157 		   struct rte_flow_error *error)
2158 {
2159 	int ret;
2160 
2161 	ret = flow_drv_validate(dev, attr, items, actions, error);
2162 	if (ret < 0)
2163 		return ret;
2164 	return 0;
2165 }
2166 
2167 /**
2168  * Get RSS action from the action list.
2169  *
2170  * @param[in] actions
2171  *   Pointer to the list of actions.
2172  *
2173  * @return
2174  *   Pointer to the RSS action if exist, else return NULL.
2175  */
2176 static const struct rte_flow_action_rss*
2177 flow_get_rss_action(const struct rte_flow_action actions[])
2178 {
2179 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2180 		switch (actions->type) {
2181 		case RTE_FLOW_ACTION_TYPE_RSS:
2182 			return (const struct rte_flow_action_rss *)
2183 			       actions->conf;
2184 		default:
2185 			break;
2186 		}
2187 	}
2188 	return NULL;
2189 }
2190 
2191 static unsigned int
2192 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2193 {
2194 	const struct rte_flow_item *item;
2195 	unsigned int has_vlan = 0;
2196 
2197 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2198 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2199 			has_vlan = 1;
2200 			break;
2201 		}
2202 	}
2203 	if (has_vlan)
2204 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2205 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2206 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2207 			       MLX5_EXPANSION_ROOT_OUTER;
2208 }
2209 
2210 /**
2211  * Create a flow and add it to @p list.
2212  *
2213  * @param dev
2214  *   Pointer to Ethernet device.
2215  * @param list
2216  *   Pointer to a TAILQ flow list.
2217  * @param[in] attr
2218  *   Flow rule attributes.
2219  * @param[in] items
2220  *   Pattern specification (list terminated by the END pattern item).
2221  * @param[in] actions
2222  *   Associated actions (list terminated by the END action).
2223  * @param[out] error
2224  *   Perform verbose error reporting if not NULL.
2225  *
2226  * @return
2227  *   A flow on success, NULL otherwise and rte_errno is set.
2228  */
2229 static struct rte_flow *
2230 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2231 		 const struct rte_flow_attr *attr,
2232 		 const struct rte_flow_item items[],
2233 		 const struct rte_flow_action actions[],
2234 		 struct rte_flow_error *error)
2235 {
2236 	struct rte_flow *flow = NULL;
2237 	struct mlx5_flow *dev_flow;
2238 	const struct rte_flow_action_rss *rss;
2239 	union {
2240 		struct rte_flow_expand_rss buf;
2241 		uint8_t buffer[2048];
2242 	} expand_buffer;
2243 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2244 	int ret;
2245 	uint32_t i;
2246 	uint32_t flow_size;
2247 
2248 	ret = flow_drv_validate(dev, attr, items, actions, error);
2249 	if (ret < 0)
2250 		return NULL;
2251 	flow_size = sizeof(struct rte_flow);
2252 	rss = flow_get_rss_action(actions);
2253 	if (rss)
2254 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2255 					    sizeof(void *));
2256 	else
2257 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2258 	flow = rte_calloc(__func__, 1, flow_size, 0);
2259 	if (!flow) {
2260 		rte_errno = ENOMEM;
2261 		return NULL;
2262 	}
2263 	flow->drv_type = flow_get_drv_type(dev, attr);
2264 	flow->ingress = attr->ingress;
2265 	flow->transfer = attr->transfer;
2266 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2267 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
2268 	flow->queue = (void *)(flow + 1);
2269 	LIST_INIT(&flow->dev_flows);
2270 	if (rss && rss->types) {
2271 		unsigned int graph_root;
2272 
2273 		graph_root = find_graph_root(items, rss->level);
2274 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2275 					  items, rss->types,
2276 					  mlx5_support_expansion,
2277 					  graph_root);
2278 		assert(ret > 0 &&
2279 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
2280 	} else {
2281 		buf->entries = 1;
2282 		buf->entry[0].pattern = (void *)(uintptr_t)items;
2283 	}
2284 	for (i = 0; i < buf->entries; ++i) {
2285 		dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2286 					    actions, error);
2287 		if (!dev_flow)
2288 			goto error;
2289 		dev_flow->flow = flow;
2290 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2291 		ret = flow_drv_translate(dev, dev_flow, attr,
2292 					 buf->entry[i].pattern,
2293 					 actions, error);
2294 		if (ret < 0)
2295 			goto error;
2296 	}
2297 	if (dev->data->dev_started) {
2298 		ret = flow_drv_apply(dev, flow, error);
2299 		if (ret < 0)
2300 			goto error;
2301 	}
2302 	TAILQ_INSERT_TAIL(list, flow, next);
2303 	flow_rxq_flags_set(dev, flow);
2304 	return flow;
2305 error:
2306 	ret = rte_errno; /* Save rte_errno before cleanup. */
2307 	assert(flow);
2308 	flow_drv_destroy(dev, flow);
2309 	rte_free(flow);
2310 	rte_errno = ret; /* Restore rte_errno. */
2311 	return NULL;
2312 }
2313 
2314 /**
2315  * Create a flow.
2316  *
2317  * @see rte_flow_create()
2318  * @see rte_flow_ops
2319  */
2320 struct rte_flow *
2321 mlx5_flow_create(struct rte_eth_dev *dev,
2322 		 const struct rte_flow_attr *attr,
2323 		 const struct rte_flow_item items[],
2324 		 const struct rte_flow_action actions[],
2325 		 struct rte_flow_error *error)
2326 {
2327 	struct mlx5_priv *priv = dev->data->dev_private;
2328 
2329 	return flow_list_create(dev, &priv->flows,
2330 				attr, items, actions, error);
2331 }
2332 
2333 /**
2334  * Destroy a flow in a list.
2335  *
2336  * @param dev
2337  *   Pointer to Ethernet device.
2338  * @param list
2339  *   Pointer to a TAILQ flow list.
2340  * @param[in] flow
2341  *   Flow to destroy.
2342  */
2343 static void
2344 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2345 		  struct rte_flow *flow)
2346 {
2347 	/*
2348 	 * Update RX queue flags only if port is started, otherwise it is
2349 	 * already clean.
2350 	 */
2351 	if (dev->data->dev_started)
2352 		flow_rxq_flags_trim(dev, flow);
2353 	flow_drv_destroy(dev, flow);
2354 	TAILQ_REMOVE(list, flow, next);
2355 	rte_free(flow->fdir);
2356 	rte_free(flow);
2357 }
2358 
2359 /**
2360  * Destroy all flows.
2361  *
2362  * @param dev
2363  *   Pointer to Ethernet device.
2364  * @param list
2365  *   Pointer to a TAILQ flow list.
2366  */
2367 void
2368 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2369 {
2370 	while (!TAILQ_EMPTY(list)) {
2371 		struct rte_flow *flow;
2372 
2373 		flow = TAILQ_FIRST(list);
2374 		flow_list_destroy(dev, list, flow);
2375 	}
2376 }
2377 
2378 /**
2379  * Remove all flows.
2380  *
2381  * @param dev
2382  *   Pointer to Ethernet device.
2383  * @param list
2384  *   Pointer to a TAILQ flow list.
2385  */
2386 void
2387 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2388 {
2389 	struct rte_flow *flow;
2390 
2391 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2392 		flow_drv_remove(dev, flow);
2393 	flow_rxq_flags_clear(dev);
2394 }
2395 
2396 /**
2397  * Add all flows.
2398  *
2399  * @param dev
2400  *   Pointer to Ethernet device.
2401  * @param list
2402  *   Pointer to a TAILQ flow list.
2403  *
2404  * @return
2405  *   0 on success, a negative errno value otherwise and rte_errno is set.
2406  */
2407 int
2408 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2409 {
2410 	struct rte_flow *flow;
2411 	struct rte_flow_error error;
2412 	int ret = 0;
2413 
2414 	TAILQ_FOREACH(flow, list, next) {
2415 		ret = flow_drv_apply(dev, flow, &error);
2416 		if (ret < 0)
2417 			goto error;
2418 		flow_rxq_flags_set(dev, flow);
2419 	}
2420 	return 0;
2421 error:
2422 	ret = rte_errno; /* Save rte_errno before cleanup. */
2423 	mlx5_flow_stop(dev, list);
2424 	rte_errno = ret; /* Restore rte_errno. */
2425 	return -rte_errno;
2426 }
2427 
2428 /**
2429  * Verify the flow list is empty
2430  *
2431  * @param dev
2432  *  Pointer to Ethernet device.
2433  *
2434  * @return the number of flows not released.
2435  */
2436 int
2437 mlx5_flow_verify(struct rte_eth_dev *dev)
2438 {
2439 	struct mlx5_priv *priv = dev->data->dev_private;
2440 	struct rte_flow *flow;
2441 	int ret = 0;
2442 
2443 	TAILQ_FOREACH(flow, &priv->flows, next) {
2444 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
2445 			dev->data->port_id, (void *)flow);
2446 		++ret;
2447 	}
2448 	return ret;
2449 }
2450 
2451 /**
2452  * Enable a control flow configured from the control plane.
2453  *
2454  * @param dev
2455  *   Pointer to Ethernet device.
2456  * @param eth_spec
2457  *   An Ethernet flow spec to apply.
2458  * @param eth_mask
2459  *   An Ethernet flow mask to apply.
2460  * @param vlan_spec
2461  *   A VLAN flow spec to apply.
2462  * @param vlan_mask
2463  *   A VLAN flow mask to apply.
2464  *
2465  * @return
2466  *   0 on success, a negative errno value otherwise and rte_errno is set.
2467  */
2468 int
2469 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2470 		    struct rte_flow_item_eth *eth_spec,
2471 		    struct rte_flow_item_eth *eth_mask,
2472 		    struct rte_flow_item_vlan *vlan_spec,
2473 		    struct rte_flow_item_vlan *vlan_mask)
2474 {
2475 	struct mlx5_priv *priv = dev->data->dev_private;
2476 	const struct rte_flow_attr attr = {
2477 		.ingress = 1,
2478 		.priority = MLX5_FLOW_PRIO_RSVD,
2479 	};
2480 	struct rte_flow_item items[] = {
2481 		{
2482 			.type = RTE_FLOW_ITEM_TYPE_ETH,
2483 			.spec = eth_spec,
2484 			.last = NULL,
2485 			.mask = eth_mask,
2486 		},
2487 		{
2488 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2489 					      RTE_FLOW_ITEM_TYPE_END,
2490 			.spec = vlan_spec,
2491 			.last = NULL,
2492 			.mask = vlan_mask,
2493 		},
2494 		{
2495 			.type = RTE_FLOW_ITEM_TYPE_END,
2496 		},
2497 	};
2498 	uint16_t queue[priv->reta_idx_n];
2499 	struct rte_flow_action_rss action_rss = {
2500 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2501 		.level = 0,
2502 		.types = priv->rss_conf.rss_hf,
2503 		.key_len = priv->rss_conf.rss_key_len,
2504 		.queue_num = priv->reta_idx_n,
2505 		.key = priv->rss_conf.rss_key,
2506 		.queue = queue,
2507 	};
2508 	struct rte_flow_action actions[] = {
2509 		{
2510 			.type = RTE_FLOW_ACTION_TYPE_RSS,
2511 			.conf = &action_rss,
2512 		},
2513 		{
2514 			.type = RTE_FLOW_ACTION_TYPE_END,
2515 		},
2516 	};
2517 	struct rte_flow *flow;
2518 	struct rte_flow_error error;
2519 	unsigned int i;
2520 
2521 	if (!priv->reta_idx_n || !priv->rxqs_n) {
2522 		return 0;
2523 	}
2524 	for (i = 0; i != priv->reta_idx_n; ++i)
2525 		queue[i] = (*priv->reta_idx)[i];
2526 	flow = flow_list_create(dev, &priv->ctrl_flows,
2527 				&attr, items, actions, &error);
2528 	if (!flow)
2529 		return -rte_errno;
2530 	return 0;
2531 }
2532 
2533 /**
2534  * Enable a flow control configured from the control plane.
2535  *
2536  * @param dev
2537  *   Pointer to Ethernet device.
2538  * @param eth_spec
2539  *   An Ethernet flow spec to apply.
2540  * @param eth_mask
2541  *   An Ethernet flow mask to apply.
2542  *
2543  * @return
2544  *   0 on success, a negative errno value otherwise and rte_errno is set.
2545  */
2546 int
2547 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2548 	       struct rte_flow_item_eth *eth_spec,
2549 	       struct rte_flow_item_eth *eth_mask)
2550 {
2551 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2552 }
2553 
2554 /**
2555  * Destroy a flow.
2556  *
2557  * @see rte_flow_destroy()
2558  * @see rte_flow_ops
2559  */
2560 int
2561 mlx5_flow_destroy(struct rte_eth_dev *dev,
2562 		  struct rte_flow *flow,
2563 		  struct rte_flow_error *error __rte_unused)
2564 {
2565 	struct mlx5_priv *priv = dev->data->dev_private;
2566 
2567 	flow_list_destroy(dev, &priv->flows, flow);
2568 	return 0;
2569 }
2570 
2571 /**
2572  * Destroy all flows.
2573  *
2574  * @see rte_flow_flush()
2575  * @see rte_flow_ops
2576  */
2577 int
2578 mlx5_flow_flush(struct rte_eth_dev *dev,
2579 		struct rte_flow_error *error __rte_unused)
2580 {
2581 	struct mlx5_priv *priv = dev->data->dev_private;
2582 
2583 	mlx5_flow_list_flush(dev, &priv->flows);
2584 	return 0;
2585 }
2586 
2587 /**
2588  * Isolated mode.
2589  *
2590  * @see rte_flow_isolate()
2591  * @see rte_flow_ops
2592  */
2593 int
2594 mlx5_flow_isolate(struct rte_eth_dev *dev,
2595 		  int enable,
2596 		  struct rte_flow_error *error)
2597 {
2598 	struct mlx5_priv *priv = dev->data->dev_private;
2599 
2600 	if (dev->data->dev_started) {
2601 		rte_flow_error_set(error, EBUSY,
2602 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2603 				   NULL,
2604 				   "port must be stopped first");
2605 		return -rte_errno;
2606 	}
2607 	priv->isolated = !!enable;
2608 	if (enable)
2609 		dev->dev_ops = &mlx5_dev_ops_isolate;
2610 	else
2611 		dev->dev_ops = &mlx5_dev_ops;
2612 	return 0;
2613 }
2614 
2615 /**
2616  * Query a flow.
2617  *
2618  * @see rte_flow_query()
2619  * @see rte_flow_ops
2620  */
2621 static int
2622 flow_drv_query(struct rte_eth_dev *dev,
2623 	       struct rte_flow *flow,
2624 	       const struct rte_flow_action *actions,
2625 	       void *data,
2626 	       struct rte_flow_error *error)
2627 {
2628 	const struct mlx5_flow_driver_ops *fops;
2629 	enum mlx5_flow_drv_type ftype = flow->drv_type;
2630 
2631 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2632 	fops = flow_get_drv_ops(ftype);
2633 
2634 	return fops->query(dev, flow, actions, data, error);
2635 }
2636 
2637 /**
2638  * Query a flow.
2639  *
2640  * @see rte_flow_query()
2641  * @see rte_flow_ops
2642  */
2643 int
2644 mlx5_flow_query(struct rte_eth_dev *dev,
2645 		struct rte_flow *flow,
2646 		const struct rte_flow_action *actions,
2647 		void *data,
2648 		struct rte_flow_error *error)
2649 {
2650 	int ret;
2651 
2652 	ret = flow_drv_query(dev, flow, actions, data, error);
2653 	if (ret < 0)
2654 		return ret;
2655 	return 0;
2656 }
2657 
2658 /**
2659  * Convert a flow director filter to a generic flow.
2660  *
2661  * @param dev
2662  *   Pointer to Ethernet device.
2663  * @param fdir_filter
2664  *   Flow director filter to add.
2665  * @param attributes
2666  *   Generic flow parameters structure.
2667  *
2668  * @return
2669  *   0 on success, a negative errno value otherwise and rte_errno is set.
2670  */
2671 static int
2672 flow_fdir_filter_convert(struct rte_eth_dev *dev,
2673 			 const struct rte_eth_fdir_filter *fdir_filter,
2674 			 struct mlx5_fdir *attributes)
2675 {
2676 	struct mlx5_priv *priv = dev->data->dev_private;
2677 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
2678 	const struct rte_eth_fdir_masks *mask =
2679 		&dev->data->dev_conf.fdir_conf.mask;
2680 
2681 	/* Validate queue number. */
2682 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2683 		DRV_LOG(ERR, "port %u invalid queue number %d",
2684 			dev->data->port_id, fdir_filter->action.rx_queue);
2685 		rte_errno = EINVAL;
2686 		return -rte_errno;
2687 	}
2688 	attributes->attr.ingress = 1;
2689 	attributes->items[0] = (struct rte_flow_item) {
2690 		.type = RTE_FLOW_ITEM_TYPE_ETH,
2691 		.spec = &attributes->l2,
2692 		.mask = &attributes->l2_mask,
2693 	};
2694 	switch (fdir_filter->action.behavior) {
2695 	case RTE_ETH_FDIR_ACCEPT:
2696 		attributes->actions[0] = (struct rte_flow_action){
2697 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
2698 			.conf = &attributes->queue,
2699 		};
2700 		break;
2701 	case RTE_ETH_FDIR_REJECT:
2702 		attributes->actions[0] = (struct rte_flow_action){
2703 			.type = RTE_FLOW_ACTION_TYPE_DROP,
2704 		};
2705 		break;
2706 	default:
2707 		DRV_LOG(ERR, "port %u invalid behavior %d",
2708 			dev->data->port_id,
2709 			fdir_filter->action.behavior);
2710 		rte_errno = ENOTSUP;
2711 		return -rte_errno;
2712 	}
2713 	attributes->queue.index = fdir_filter->action.rx_queue;
2714 	/* Handle L3. */
2715 	switch (fdir_filter->input.flow_type) {
2716 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2717 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2718 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2719 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
2720 			.src_addr = input->flow.ip4_flow.src_ip,
2721 			.dst_addr = input->flow.ip4_flow.dst_ip,
2722 			.time_to_live = input->flow.ip4_flow.ttl,
2723 			.type_of_service = input->flow.ip4_flow.tos,
2724 		};
2725 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
2726 			.src_addr = mask->ipv4_mask.src_ip,
2727 			.dst_addr = mask->ipv4_mask.dst_ip,
2728 			.time_to_live = mask->ipv4_mask.ttl,
2729 			.type_of_service = mask->ipv4_mask.tos,
2730 			.next_proto_id = mask->ipv4_mask.proto,
2731 		};
2732 		attributes->items[1] = (struct rte_flow_item){
2733 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
2734 			.spec = &attributes->l3,
2735 			.mask = &attributes->l3_mask,
2736 		};
2737 		break;
2738 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2739 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2740 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2741 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
2742 			.hop_limits = input->flow.ipv6_flow.hop_limits,
2743 			.proto = input->flow.ipv6_flow.proto,
2744 		};
2745 
2746 		memcpy(attributes->l3.ipv6.hdr.src_addr,
2747 		       input->flow.ipv6_flow.src_ip,
2748 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2749 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
2750 		       input->flow.ipv6_flow.dst_ip,
2751 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2752 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2753 		       mask->ipv6_mask.src_ip,
2754 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2755 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2756 		       mask->ipv6_mask.dst_ip,
2757 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2758 		attributes->items[1] = (struct rte_flow_item){
2759 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
2760 			.spec = &attributes->l3,
2761 			.mask = &attributes->l3_mask,
2762 		};
2763 		break;
2764 	default:
2765 		DRV_LOG(ERR, "port %u invalid flow type%d",
2766 			dev->data->port_id, fdir_filter->input.flow_type);
2767 		rte_errno = ENOTSUP;
2768 		return -rte_errno;
2769 	}
2770 	/* Handle L4. */
2771 	switch (fdir_filter->input.flow_type) {
2772 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2773 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
2774 			.src_port = input->flow.udp4_flow.src_port,
2775 			.dst_port = input->flow.udp4_flow.dst_port,
2776 		};
2777 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
2778 			.src_port = mask->src_port_mask,
2779 			.dst_port = mask->dst_port_mask,
2780 		};
2781 		attributes->items[2] = (struct rte_flow_item){
2782 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2783 			.spec = &attributes->l4,
2784 			.mask = &attributes->l4_mask,
2785 		};
2786 		break;
2787 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2788 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
2789 			.src_port = input->flow.tcp4_flow.src_port,
2790 			.dst_port = input->flow.tcp4_flow.dst_port,
2791 		};
2792 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
2793 			.src_port = mask->src_port_mask,
2794 			.dst_port = mask->dst_port_mask,
2795 		};
2796 		attributes->items[2] = (struct rte_flow_item){
2797 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2798 			.spec = &attributes->l4,
2799 			.mask = &attributes->l4_mask,
2800 		};
2801 		break;
2802 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2803 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
2804 			.src_port = input->flow.udp6_flow.src_port,
2805 			.dst_port = input->flow.udp6_flow.dst_port,
2806 		};
2807 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
2808 			.src_port = mask->src_port_mask,
2809 			.dst_port = mask->dst_port_mask,
2810 		};
2811 		attributes->items[2] = (struct rte_flow_item){
2812 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2813 			.spec = &attributes->l4,
2814 			.mask = &attributes->l4_mask,
2815 		};
2816 		break;
2817 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2818 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
2819 			.src_port = input->flow.tcp6_flow.src_port,
2820 			.dst_port = input->flow.tcp6_flow.dst_port,
2821 		};
2822 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
2823 			.src_port = mask->src_port_mask,
2824 			.dst_port = mask->dst_port_mask,
2825 		};
2826 		attributes->items[2] = (struct rte_flow_item){
2827 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2828 			.spec = &attributes->l4,
2829 			.mask = &attributes->l4_mask,
2830 		};
2831 		break;
2832 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2833 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2834 		break;
2835 	default:
2836 		DRV_LOG(ERR, "port %u invalid flow type%d",
2837 			dev->data->port_id, fdir_filter->input.flow_type);
2838 		rte_errno = ENOTSUP;
2839 		return -rte_errno;
2840 	}
2841 	return 0;
2842 }
2843 
2844 #define FLOW_FDIR_CMP(f1, f2, fld) \
2845 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
2846 
2847 /**
2848  * Compare two FDIR flows. If items and actions are identical, the two flows are
2849  * regarded as same.
2850  *
2851  * @param dev
2852  *   Pointer to Ethernet device.
2853  * @param f1
2854  *   FDIR flow to compare.
2855  * @param f2
2856  *   FDIR flow to compare.
2857  *
2858  * @return
2859  *   Zero on match, 1 otherwise.
2860  */
2861 static int
2862 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
2863 {
2864 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
2865 	    FLOW_FDIR_CMP(f1, f2, l2) ||
2866 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
2867 	    FLOW_FDIR_CMP(f1, f2, l3) ||
2868 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
2869 	    FLOW_FDIR_CMP(f1, f2, l4) ||
2870 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
2871 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
2872 		return 1;
2873 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
2874 	    FLOW_FDIR_CMP(f1, f2, queue))
2875 		return 1;
2876 	return 0;
2877 }
2878 
2879 /**
2880  * Search device flow list to find out a matched FDIR flow.
2881  *
2882  * @param dev
2883  *   Pointer to Ethernet device.
2884  * @param fdir_flow
2885  *   FDIR flow to lookup.
2886  *
2887  * @return
2888  *   Pointer of flow if found, NULL otherwise.
2889  */
2890 static struct rte_flow *
2891 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
2892 {
2893 	struct mlx5_priv *priv = dev->data->dev_private;
2894 	struct rte_flow *flow = NULL;
2895 
2896 	assert(fdir_flow);
2897 	TAILQ_FOREACH(flow, &priv->flows, next) {
2898 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
2899 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
2900 				dev->data->port_id, (void *)flow);
2901 			break;
2902 		}
2903 	}
2904 	return flow;
2905 }
2906 
2907 /**
2908  * Add new flow director filter and store it in list.
2909  *
2910  * @param dev
2911  *   Pointer to Ethernet device.
2912  * @param fdir_filter
2913  *   Flow director filter to add.
2914  *
2915  * @return
2916  *   0 on success, a negative errno value otherwise and rte_errno is set.
2917  */
2918 static int
2919 flow_fdir_filter_add(struct rte_eth_dev *dev,
2920 		     const struct rte_eth_fdir_filter *fdir_filter)
2921 {
2922 	struct mlx5_priv *priv = dev->data->dev_private;
2923 	struct mlx5_fdir *fdir_flow;
2924 	struct rte_flow *flow;
2925 	int ret;
2926 
2927 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
2928 	if (!fdir_flow) {
2929 		rte_errno = ENOMEM;
2930 		return -rte_errno;
2931 	}
2932 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
2933 	if (ret)
2934 		goto error;
2935 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
2936 	if (flow) {
2937 		rte_errno = EEXIST;
2938 		goto error;
2939 	}
2940 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
2941 				fdir_flow->items, fdir_flow->actions, NULL);
2942 	if (!flow)
2943 		goto error;
2944 	assert(!flow->fdir);
2945 	flow->fdir = fdir_flow;
2946 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
2947 		dev->data->port_id, (void *)flow);
2948 	return 0;
2949 error:
2950 	rte_free(fdir_flow);
2951 	return -rte_errno;
2952 }
2953 
2954 /**
2955  * Delete specific filter.
2956  *
2957  * @param dev
2958  *   Pointer to Ethernet device.
2959  * @param fdir_filter
2960  *   Filter to be deleted.
2961  *
2962  * @return
2963  *   0 on success, a negative errno value otherwise and rte_errno is set.
2964  */
2965 static int
2966 flow_fdir_filter_delete(struct rte_eth_dev *dev,
2967 			const struct rte_eth_fdir_filter *fdir_filter)
2968 {
2969 	struct mlx5_priv *priv = dev->data->dev_private;
2970 	struct rte_flow *flow;
2971 	struct mlx5_fdir fdir_flow = {
2972 		.attr.group = 0,
2973 	};
2974 	int ret;
2975 
2976 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
2977 	if (ret)
2978 		return -rte_errno;
2979 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
2980 	if (!flow) {
2981 		rte_errno = ENOENT;
2982 		return -rte_errno;
2983 	}
2984 	flow_list_destroy(dev, &priv->flows, flow);
2985 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
2986 		dev->data->port_id, (void *)flow);
2987 	return 0;
2988 }
2989 
2990 /**
2991  * Update queue for specific filter.
2992  *
2993  * @param dev
2994  *   Pointer to Ethernet device.
2995  * @param fdir_filter
2996  *   Filter to be updated.
2997  *
2998  * @return
2999  *   0 on success, a negative errno value otherwise and rte_errno is set.
3000  */
3001 static int
3002 flow_fdir_filter_update(struct rte_eth_dev *dev,
3003 			const struct rte_eth_fdir_filter *fdir_filter)
3004 {
3005 	int ret;
3006 
3007 	ret = flow_fdir_filter_delete(dev, fdir_filter);
3008 	if (ret)
3009 		return ret;
3010 	return flow_fdir_filter_add(dev, fdir_filter);
3011 }
3012 
3013 /**
3014  * Flush all filters.
3015  *
3016  * @param dev
3017  *   Pointer to Ethernet device.
3018  */
3019 static void
3020 flow_fdir_filter_flush(struct rte_eth_dev *dev)
3021 {
3022 	struct mlx5_priv *priv = dev->data->dev_private;
3023 
3024 	mlx5_flow_list_flush(dev, &priv->flows);
3025 }
3026 
3027 /**
3028  * Get flow director information.
3029  *
3030  * @param dev
3031  *   Pointer to Ethernet device.
3032  * @param[out] fdir_info
3033  *   Resulting flow director information.
3034  */
3035 static void
3036 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
3037 {
3038 	struct rte_eth_fdir_masks *mask =
3039 		&dev->data->dev_conf.fdir_conf.mask;
3040 
3041 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
3042 	fdir_info->guarant_spc = 0;
3043 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
3044 	fdir_info->max_flexpayload = 0;
3045 	fdir_info->flow_types_mask[0] = 0;
3046 	fdir_info->flex_payload_unit = 0;
3047 	fdir_info->max_flex_payload_segment_num = 0;
3048 	fdir_info->flex_payload_limit = 0;
3049 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
3050 }
3051 
3052 /**
3053  * Deal with flow director operations.
3054  *
3055  * @param dev
3056  *   Pointer to Ethernet device.
3057  * @param filter_op
3058  *   Operation to perform.
3059  * @param arg
3060  *   Pointer to operation-specific structure.
3061  *
3062  * @return
3063  *   0 on success, a negative errno value otherwise and rte_errno is set.
3064  */
3065 static int
3066 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
3067 		    void *arg)
3068 {
3069 	enum rte_fdir_mode fdir_mode =
3070 		dev->data->dev_conf.fdir_conf.mode;
3071 
3072 	if (filter_op == RTE_ETH_FILTER_NOP)
3073 		return 0;
3074 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
3075 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
3076 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
3077 			dev->data->port_id, fdir_mode);
3078 		rte_errno = EINVAL;
3079 		return -rte_errno;
3080 	}
3081 	switch (filter_op) {
3082 	case RTE_ETH_FILTER_ADD:
3083 		return flow_fdir_filter_add(dev, arg);
3084 	case RTE_ETH_FILTER_UPDATE:
3085 		return flow_fdir_filter_update(dev, arg);
3086 	case RTE_ETH_FILTER_DELETE:
3087 		return flow_fdir_filter_delete(dev, arg);
3088 	case RTE_ETH_FILTER_FLUSH:
3089 		flow_fdir_filter_flush(dev);
3090 		break;
3091 	case RTE_ETH_FILTER_INFO:
3092 		flow_fdir_info_get(dev, arg);
3093 		break;
3094 	default:
3095 		DRV_LOG(DEBUG, "port %u unknown operation %u",
3096 			dev->data->port_id, filter_op);
3097 		rte_errno = EINVAL;
3098 		return -rte_errno;
3099 	}
3100 	return 0;
3101 }
3102 
3103 /**
3104  * Manage filter operations.
3105  *
3106  * @param dev
3107  *   Pointer to Ethernet device structure.
3108  * @param filter_type
3109  *   Filter type.
3110  * @param filter_op
3111  *   Operation to perform.
3112  * @param arg
3113  *   Pointer to operation-specific structure.
3114  *
3115  * @return
3116  *   0 on success, a negative errno value otherwise and rte_errno is set.
3117  */
3118 int
3119 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
3120 		     enum rte_filter_type filter_type,
3121 		     enum rte_filter_op filter_op,
3122 		     void *arg)
3123 {
3124 	switch (filter_type) {
3125 	case RTE_ETH_FILTER_GENERIC:
3126 		if (filter_op != RTE_ETH_FILTER_GET) {
3127 			rte_errno = EINVAL;
3128 			return -rte_errno;
3129 		}
3130 		*(const void **)arg = &mlx5_flow_ops;
3131 		return 0;
3132 	case RTE_ETH_FILTER_FDIR:
3133 		return flow_fdir_ctrl_func(dev, filter_op, arg);
3134 	default:
3135 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
3136 			dev->data->port_id, filter_type);
3137 		rte_errno = ENOTSUP;
3138 		return -rte_errno;
3139 	}
3140 	return 0;
3141 }
3142