xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 5b4557ec8d16859215fafe9e24d3fb7722f43b8a)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 
12 /* Verbs header. */
13 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
14 #ifdef PEDANTIC
15 #pragma GCC diagnostic ignored "-Wpedantic"
16 #endif
17 #include <infiniband/verbs.h>
18 #ifdef PEDANTIC
19 #pragma GCC diagnostic error "-Wpedantic"
20 #endif
21 
22 #include <rte_common.h>
23 #include <rte_ether.h>
24 #include <rte_eth_ctrl.h>
25 #include <rte_ethdev_driver.h>
26 #include <rte_flow.h>
27 #include <rte_flow_driver.h>
28 #include <rte_malloc.h>
29 #include <rte_ip.h>
30 
31 #include "mlx5.h"
32 #include "mlx5_defs.h"
33 #include "mlx5_prm.h"
34 #include "mlx5_glue.h"
35 #include "mlx5_flow.h"
36 
37 /* Dev ops structure defined in mlx5.c */
38 extern const struct eth_dev_ops mlx5_dev_ops;
39 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
40 
41 /** Device flow drivers. */
42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
43 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
44 #endif
45 extern const struct mlx5_flow_driver_ops mlx5_flow_tcf_drv_ops;
46 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
47 
48 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
49 
50 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
51 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
52 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
53 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
54 #endif
55 	[MLX5_FLOW_TYPE_TCF] = &mlx5_flow_tcf_drv_ops,
56 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
57 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
58 };
59 
60 enum mlx5_expansion {
61 	MLX5_EXPANSION_ROOT,
62 	MLX5_EXPANSION_ROOT_OUTER,
63 	MLX5_EXPANSION_ROOT_ETH_VLAN,
64 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
65 	MLX5_EXPANSION_OUTER_ETH,
66 	MLX5_EXPANSION_OUTER_ETH_VLAN,
67 	MLX5_EXPANSION_OUTER_VLAN,
68 	MLX5_EXPANSION_OUTER_IPV4,
69 	MLX5_EXPANSION_OUTER_IPV4_UDP,
70 	MLX5_EXPANSION_OUTER_IPV4_TCP,
71 	MLX5_EXPANSION_OUTER_IPV6,
72 	MLX5_EXPANSION_OUTER_IPV6_UDP,
73 	MLX5_EXPANSION_OUTER_IPV6_TCP,
74 	MLX5_EXPANSION_VXLAN,
75 	MLX5_EXPANSION_VXLAN_GPE,
76 	MLX5_EXPANSION_GRE,
77 	MLX5_EXPANSION_MPLS,
78 	MLX5_EXPANSION_ETH,
79 	MLX5_EXPANSION_ETH_VLAN,
80 	MLX5_EXPANSION_VLAN,
81 	MLX5_EXPANSION_IPV4,
82 	MLX5_EXPANSION_IPV4_UDP,
83 	MLX5_EXPANSION_IPV4_TCP,
84 	MLX5_EXPANSION_IPV6,
85 	MLX5_EXPANSION_IPV6_UDP,
86 	MLX5_EXPANSION_IPV6_TCP,
87 };
88 
89 /** Supported expansion of items. */
90 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
91 	[MLX5_EXPANSION_ROOT] = {
92 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
93 						 MLX5_EXPANSION_IPV4,
94 						 MLX5_EXPANSION_IPV6),
95 		.type = RTE_FLOW_ITEM_TYPE_END,
96 	},
97 	[MLX5_EXPANSION_ROOT_OUTER] = {
98 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
99 						 MLX5_EXPANSION_OUTER_IPV4,
100 						 MLX5_EXPANSION_OUTER_IPV6),
101 		.type = RTE_FLOW_ITEM_TYPE_END,
102 	},
103 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
104 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
105 		.type = RTE_FLOW_ITEM_TYPE_END,
106 	},
107 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
108 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
109 		.type = RTE_FLOW_ITEM_TYPE_END,
110 	},
111 	[MLX5_EXPANSION_OUTER_ETH] = {
112 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
113 						 MLX5_EXPANSION_OUTER_IPV6,
114 						 MLX5_EXPANSION_MPLS),
115 		.type = RTE_FLOW_ITEM_TYPE_ETH,
116 		.rss_types = 0,
117 	},
118 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
119 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
120 		.type = RTE_FLOW_ITEM_TYPE_ETH,
121 		.rss_types = 0,
122 	},
123 	[MLX5_EXPANSION_OUTER_VLAN] = {
124 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
125 						 MLX5_EXPANSION_OUTER_IPV6),
126 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
127 	},
128 	[MLX5_EXPANSION_OUTER_IPV4] = {
129 		.next = RTE_FLOW_EXPAND_RSS_NEXT
130 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
131 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
132 			 MLX5_EXPANSION_GRE),
133 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
134 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
135 			ETH_RSS_NONFRAG_IPV4_OTHER,
136 	},
137 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
138 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
139 						 MLX5_EXPANSION_VXLAN_GPE),
140 		.type = RTE_FLOW_ITEM_TYPE_UDP,
141 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
142 	},
143 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
144 		.type = RTE_FLOW_ITEM_TYPE_TCP,
145 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
146 	},
147 	[MLX5_EXPANSION_OUTER_IPV6] = {
148 		.next = RTE_FLOW_EXPAND_RSS_NEXT
149 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
150 			 MLX5_EXPANSION_OUTER_IPV6_TCP),
151 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
152 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
153 			ETH_RSS_NONFRAG_IPV6_OTHER,
154 	},
155 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
156 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
157 						 MLX5_EXPANSION_VXLAN_GPE),
158 		.type = RTE_FLOW_ITEM_TYPE_UDP,
159 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
160 	},
161 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
162 		.type = RTE_FLOW_ITEM_TYPE_TCP,
163 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
164 	},
165 	[MLX5_EXPANSION_VXLAN] = {
166 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
167 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
168 	},
169 	[MLX5_EXPANSION_VXLAN_GPE] = {
170 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
171 						 MLX5_EXPANSION_IPV4,
172 						 MLX5_EXPANSION_IPV6),
173 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
174 	},
175 	[MLX5_EXPANSION_GRE] = {
176 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
177 		.type = RTE_FLOW_ITEM_TYPE_GRE,
178 	},
179 	[MLX5_EXPANSION_MPLS] = {
180 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
181 						 MLX5_EXPANSION_IPV6),
182 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
183 	},
184 	[MLX5_EXPANSION_ETH] = {
185 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
186 						 MLX5_EXPANSION_IPV6),
187 		.type = RTE_FLOW_ITEM_TYPE_ETH,
188 	},
189 	[MLX5_EXPANSION_ETH_VLAN] = {
190 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
191 		.type = RTE_FLOW_ITEM_TYPE_ETH,
192 	},
193 	[MLX5_EXPANSION_VLAN] = {
194 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
195 						 MLX5_EXPANSION_IPV6),
196 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
197 	},
198 	[MLX5_EXPANSION_IPV4] = {
199 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
200 						 MLX5_EXPANSION_IPV4_TCP),
201 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
202 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
203 			ETH_RSS_NONFRAG_IPV4_OTHER,
204 	},
205 	[MLX5_EXPANSION_IPV4_UDP] = {
206 		.type = RTE_FLOW_ITEM_TYPE_UDP,
207 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
208 	},
209 	[MLX5_EXPANSION_IPV4_TCP] = {
210 		.type = RTE_FLOW_ITEM_TYPE_TCP,
211 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
212 	},
213 	[MLX5_EXPANSION_IPV6] = {
214 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
215 						 MLX5_EXPANSION_IPV6_TCP),
216 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
217 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
218 			ETH_RSS_NONFRAG_IPV6_OTHER,
219 	},
220 	[MLX5_EXPANSION_IPV6_UDP] = {
221 		.type = RTE_FLOW_ITEM_TYPE_UDP,
222 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
223 	},
224 	[MLX5_EXPANSION_IPV6_TCP] = {
225 		.type = RTE_FLOW_ITEM_TYPE_TCP,
226 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
227 	},
228 };
229 
230 static const struct rte_flow_ops mlx5_flow_ops = {
231 	.validate = mlx5_flow_validate,
232 	.create = mlx5_flow_create,
233 	.destroy = mlx5_flow_destroy,
234 	.flush = mlx5_flow_flush,
235 	.isolate = mlx5_flow_isolate,
236 	.query = mlx5_flow_query,
237 };
238 
239 /* Convert FDIR request to Generic flow. */
240 struct mlx5_fdir {
241 	struct rte_flow_attr attr;
242 	struct rte_flow_action actions[2];
243 	struct rte_flow_item items[4];
244 	struct rte_flow_item_eth l2;
245 	struct rte_flow_item_eth l2_mask;
246 	union {
247 		struct rte_flow_item_ipv4 ipv4;
248 		struct rte_flow_item_ipv6 ipv6;
249 	} l3;
250 	union {
251 		struct rte_flow_item_ipv4 ipv4;
252 		struct rte_flow_item_ipv6 ipv6;
253 	} l3_mask;
254 	union {
255 		struct rte_flow_item_udp udp;
256 		struct rte_flow_item_tcp tcp;
257 	} l4;
258 	union {
259 		struct rte_flow_item_udp udp;
260 		struct rte_flow_item_tcp tcp;
261 	} l4_mask;
262 	struct rte_flow_action_queue queue;
263 };
264 
265 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
266 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
267 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
268 };
269 
270 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
271 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
272 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
273 	{ 9, 10, 11 }, { 12, 13, 14 },
274 };
275 
276 /* Tunnel information. */
277 struct mlx5_flow_tunnel_info {
278 	uint32_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
279 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
280 };
281 
282 static struct mlx5_flow_tunnel_info tunnels_info[] = {
283 	{
284 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
285 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
286 	},
287 	{
288 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
289 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
290 	},
291 	{
292 		.tunnel = MLX5_FLOW_LAYER_GRE,
293 		.ptype = RTE_PTYPE_TUNNEL_GRE,
294 	},
295 	{
296 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
297 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE | RTE_PTYPE_L4_UDP,
298 	},
299 	{
300 		.tunnel = MLX5_FLOW_LAYER_MPLS,
301 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
302 	},
303 };
304 
305 /**
306  * Discover the maximum number of priority available.
307  *
308  * @param[in] dev
309  *   Pointer to the Ethernet device structure.
310  *
311  * @return
312  *   number of supported flow priority on success, a negative errno
313  *   value otherwise and rte_errno is set.
314  */
315 int
316 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
317 {
318 	struct {
319 		struct ibv_flow_attr attr;
320 		struct ibv_flow_spec_eth eth;
321 		struct ibv_flow_spec_action_drop drop;
322 	} flow_attr = {
323 		.attr = {
324 			.num_of_specs = 2,
325 		},
326 		.eth = {
327 			.type = IBV_FLOW_SPEC_ETH,
328 			.size = sizeof(struct ibv_flow_spec_eth),
329 		},
330 		.drop = {
331 			.size = sizeof(struct ibv_flow_spec_action_drop),
332 			.type = IBV_FLOW_SPEC_ACTION_DROP,
333 		},
334 	};
335 	struct ibv_flow *flow;
336 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
337 	uint16_t vprio[] = { 8, 16 };
338 	int i;
339 	int priority = 0;
340 
341 	if (!drop) {
342 		rte_errno = ENOTSUP;
343 		return -rte_errno;
344 	}
345 	for (i = 0; i != RTE_DIM(vprio); i++) {
346 		flow_attr.attr.priority = vprio[i] - 1;
347 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
348 		if (!flow)
349 			break;
350 		claim_zero(mlx5_glue->destroy_flow(flow));
351 		priority = vprio[i];
352 	}
353 	switch (priority) {
354 	case 8:
355 		priority = RTE_DIM(priority_map_3);
356 		break;
357 	case 16:
358 		priority = RTE_DIM(priority_map_5);
359 		break;
360 	default:
361 		rte_errno = ENOTSUP;
362 		DRV_LOG(ERR,
363 			"port %u verbs maximum priority: %d expected 8/16",
364 			dev->data->port_id, vprio[i]);
365 		return -rte_errno;
366 	}
367 	mlx5_hrxq_drop_release(dev);
368 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
369 		dev->data->port_id, priority);
370 	return priority;
371 }
372 
373 /**
374  * Adjust flow priority based on the highest layer and the request priority.
375  *
376  * @param[in] dev
377  *   Pointer to the Ethernet device structure.
378  * @param[in] priority
379  *   The rule base priority.
380  * @param[in] subpriority
381  *   The priority based on the items.
382  *
383  * @return
384  *   The new priority.
385  */
386 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
387 				   uint32_t subpriority)
388 {
389 	uint32_t res = 0;
390 	struct priv *priv = dev->data->dev_private;
391 
392 	switch (priv->config.flow_prio) {
393 	case RTE_DIM(priority_map_3):
394 		res = priority_map_3[priority][subpriority];
395 		break;
396 	case RTE_DIM(priority_map_5):
397 		res = priority_map_5[priority][subpriority];
398 		break;
399 	}
400 	return  res;
401 }
402 
403 /**
404  * Verify the @p item specifications (spec, last, mask) are compatible with the
405  * NIC capabilities.
406  *
407  * @param[in] item
408  *   Item specification.
409  * @param[in] mask
410  *   @p item->mask or flow default bit-masks.
411  * @param[in] nic_mask
412  *   Bit-masks covering supported fields by the NIC to compare with user mask.
413  * @param[in] size
414  *   Bit-masks size in bytes.
415  * @param[out] error
416  *   Pointer to error structure.
417  *
418  * @return
419  *   0 on success, a negative errno value otherwise and rte_errno is set.
420  */
421 static 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, ENOTSUP,
456 						  RTE_FLOW_ERROR_TYPE_ITEM,
457 						  item,
458 						  "range is not supported");
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, uint32_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 mlx5_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 flow.
529  *
530  * @param[in] dev
531  *   Pointer to the Ethernet device structure.
532  * @param[in] flow
533  *   Pointer to flow structure.
534  */
535 static void
536 mlx5_flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
537 {
538 	struct priv *priv = dev->data->dev_private;
539 	const int mark = !!(flow->actions &
540 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
541 	const int tunnel = !!(flow->layers & MLX5_FLOW_LAYER_TUNNEL);
542 	unsigned int i;
543 
544 	for (i = 0; i != flow->rss.queue_num; ++i) {
545 		int idx = (*flow->queue)[i];
546 		struct mlx5_rxq_ctrl *rxq_ctrl =
547 			container_of((*priv->rxqs)[idx],
548 				     struct mlx5_rxq_ctrl, rxq);
549 
550 		if (mark) {
551 			rxq_ctrl->rxq.mark = 1;
552 			rxq_ctrl->flow_mark_n++;
553 		}
554 		if (tunnel) {
555 			unsigned int j;
556 
557 			/* Increase the counter matching the flow. */
558 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
559 				if ((tunnels_info[j].tunnel & flow->layers) ==
560 				    tunnels_info[j].tunnel) {
561 					rxq_ctrl->flow_tunnels_n[j]++;
562 					break;
563 				}
564 			}
565 			mlx5_flow_rxq_tunnel_ptype_update(rxq_ctrl);
566 		}
567 	}
568 }
569 
570 /**
571  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
572  * @p flow if no other flow uses it with the same kind of request.
573  *
574  * @param dev
575  *   Pointer to Ethernet device.
576  * @param[in] flow
577  *   Pointer to the flow.
578  */
579 static void
580 mlx5_flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
581 {
582 	struct priv *priv = dev->data->dev_private;
583 	const int mark = !!(flow->actions &
584 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
585 	const int tunnel = !!(flow->layers & MLX5_FLOW_LAYER_TUNNEL);
586 	unsigned int i;
587 
588 	assert(dev->data->dev_started);
589 	for (i = 0; i != flow->rss.queue_num; ++i) {
590 		int idx = (*flow->queue)[i];
591 		struct mlx5_rxq_ctrl *rxq_ctrl =
592 			container_of((*priv->rxqs)[idx],
593 				     struct mlx5_rxq_ctrl, rxq);
594 
595 		if (mark) {
596 			rxq_ctrl->flow_mark_n--;
597 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
598 		}
599 		if (tunnel) {
600 			unsigned int j;
601 
602 			/* Decrease the counter matching the flow. */
603 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
604 				if ((tunnels_info[j].tunnel & flow->layers) ==
605 				    tunnels_info[j].tunnel) {
606 					rxq_ctrl->flow_tunnels_n[j]--;
607 					break;
608 				}
609 			}
610 			mlx5_flow_rxq_tunnel_ptype_update(rxq_ctrl);
611 		}
612 	}
613 }
614 
615 /**
616  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
617  *
618  * @param dev
619  *   Pointer to Ethernet device.
620  */
621 static void
622 mlx5_flow_rxq_flags_clear(struct rte_eth_dev *dev)
623 {
624 	struct priv *priv = dev->data->dev_private;
625 	unsigned int i;
626 
627 	for (i = 0; i != priv->rxqs_n; ++i) {
628 		struct mlx5_rxq_ctrl *rxq_ctrl;
629 		unsigned int j;
630 
631 		if (!(*priv->rxqs)[i])
632 			continue;
633 		rxq_ctrl = container_of((*priv->rxqs)[i],
634 					struct mlx5_rxq_ctrl, rxq);
635 		rxq_ctrl->flow_mark_n = 0;
636 		rxq_ctrl->rxq.mark = 0;
637 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
638 			rxq_ctrl->flow_tunnels_n[j] = 0;
639 		rxq_ctrl->rxq.tunnel = 0;
640 	}
641 }
642 
643 /*
644  * Validate the flag action.
645  *
646  * @param[in] action_flags
647  *   Bit-fields that holds the actions detected until now.
648  * @param[out] error
649  *   Pointer to error structure.
650  *
651  * @return
652  *   0 on success, a negative errno value otherwise and rte_errno is set.
653  */
654 int
655 mlx5_flow_validate_action_flag(uint64_t action_flags,
656 			       struct rte_flow_error *error)
657 {
658 
659 	if (action_flags & MLX5_FLOW_ACTION_DROP)
660 		return rte_flow_error_set(error, EINVAL,
661 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
662 					  "can't drop and flag in same flow");
663 	if (action_flags & MLX5_FLOW_ACTION_MARK)
664 		return rte_flow_error_set(error, EINVAL,
665 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
666 					  "can't mark and flag in same flow");
667 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
668 		return rte_flow_error_set(error, EINVAL,
669 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
670 					  "can't have 2 flag"
671 					  " actions in same flow");
672 	return 0;
673 }
674 
675 /*
676  * Validate the mark action.
677  *
678  * @param[in] action
679  *   Pointer to the queue action.
680  * @param[in] action_flags
681  *   Bit-fields that holds the actions detected until now.
682  * @param[out] error
683  *   Pointer to error structure.
684  *
685  * @return
686  *   0 on success, a negative errno value otherwise and rte_errno is set.
687  */
688 int
689 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
690 			       uint64_t action_flags,
691 			       struct rte_flow_error *error)
692 {
693 	const struct rte_flow_action_mark *mark = action->conf;
694 
695 	if (!mark)
696 		return rte_flow_error_set(error, EINVAL,
697 					  RTE_FLOW_ERROR_TYPE_ACTION,
698 					  action,
699 					  "configuration cannot be null");
700 	if (mark->id >= MLX5_FLOW_MARK_MAX)
701 		return rte_flow_error_set(error, EINVAL,
702 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
703 					  &mark->id,
704 					  "mark id must in 0 <= id < "
705 					  RTE_STR(MLX5_FLOW_MARK_MAX));
706 	if (action_flags & MLX5_FLOW_ACTION_DROP)
707 		return rte_flow_error_set(error, EINVAL,
708 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
709 					  "can't drop and mark 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 flag and mark in same flow");
714 	if (action_flags & MLX5_FLOW_ACTION_MARK)
715 		return rte_flow_error_set(error, EINVAL,
716 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
717 					  "can't have 2 mark actions in same"
718 					  " flow");
719 	return 0;
720 }
721 
722 /*
723  * Validate the drop action.
724  *
725  * @param[in] action_flags
726  *   Bit-fields that holds the actions detected until now.
727  * @param[out] error
728  *   Pointer to error structure.
729  *
730  * @return
731  *   0 on success, a negative errno value otherwise and rte_ernno is set.
732  */
733 int
734 mlx5_flow_validate_action_drop(uint64_t action_flags,
735 			       struct rte_flow_error *error)
736 {
737 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
738 		return rte_flow_error_set(error, EINVAL,
739 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
740 					  "can't drop and flag in same flow");
741 	if (action_flags & MLX5_FLOW_ACTION_MARK)
742 		return rte_flow_error_set(error, EINVAL,
743 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
744 					  "can't drop and mark in same flow");
745 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
746 		return rte_flow_error_set(error, EINVAL,
747 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
748 					  "can't have 2 fate actions in"
749 					  " same flow");
750 	return 0;
751 }
752 
753 /*
754  * Validate the queue action.
755  *
756  * @param[in] action
757  *   Pointer to the queue action.
758  * @param[in] action_flags
759  *   Bit-fields that holds the actions detected until now.
760  * @param[in] dev
761  *   Pointer to the Ethernet device structure.
762  * @param[out] error
763  *   Pointer to error structure.
764  *
765  * @return
766  *   0 on success, a negative errno value otherwise and rte_ernno is set.
767  */
768 int
769 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
770 				uint64_t action_flags,
771 				struct rte_eth_dev *dev,
772 				struct rte_flow_error *error)
773 {
774 	struct priv *priv = dev->data->dev_private;
775 	const struct rte_flow_action_queue *queue = action->conf;
776 
777 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
778 		return rte_flow_error_set(error, EINVAL,
779 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
780 					  "can't have 2 fate actions in"
781 					  " same flow");
782 	if (queue->index >= priv->rxqs_n)
783 		return rte_flow_error_set(error, EINVAL,
784 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
785 					  &queue->index,
786 					  "queue index out of range");
787 	if (!(*priv->rxqs)[queue->index])
788 		return rte_flow_error_set(error, EINVAL,
789 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
790 					  &queue->index,
791 					  "queue is not configured");
792 	return 0;
793 }
794 
795 /*
796  * Validate the rss action.
797  *
798  * @param[in] action
799  *   Pointer to the queue action.
800  * @param[in] action_flags
801  *   Bit-fields that holds the actions detected until now.
802  * @param[in] dev
803  *   Pointer to the Ethernet device structure.
804  * @param[out] error
805  *   Pointer to error structure.
806  *
807  * @return
808  *   0 on success, a negative errno value otherwise and rte_ernno is set.
809  */
810 int
811 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
812 			      uint64_t action_flags,
813 			      struct rte_eth_dev *dev,
814 			      struct rte_flow_error *error)
815 {
816 	struct priv *priv = dev->data->dev_private;
817 	const struct rte_flow_action_rss *rss = action->conf;
818 	unsigned int i;
819 
820 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
821 		return rte_flow_error_set(error, EINVAL,
822 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
823 					  "can't have 2 fate actions"
824 					  " in same flow");
825 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
826 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
827 		return rte_flow_error_set(error, ENOTSUP,
828 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
829 					  &rss->func,
830 					  "RSS hash function not supported");
831 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
832 	if (rss->level > 2)
833 #else
834 	if (rss->level > 1)
835 #endif
836 		return rte_flow_error_set(error, ENOTSUP,
837 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
838 					  &rss->level,
839 					  "tunnel RSS is not supported");
840 	if (rss->key_len < MLX5_RSS_HASH_KEY_LEN)
841 		return rte_flow_error_set(error, ENOTSUP,
842 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
843 					  &rss->key_len,
844 					  "RSS hash key too small");
845 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
846 		return rte_flow_error_set(error, ENOTSUP,
847 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
848 					  &rss->key_len,
849 					  "RSS hash key too large");
850 	if (rss->queue_num > priv->config.ind_table_max_size)
851 		return rte_flow_error_set(error, ENOTSUP,
852 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
853 					  &rss->queue_num,
854 					  "number of queues too large");
855 	if (rss->types & MLX5_RSS_HF_MASK)
856 		return rte_flow_error_set(error, ENOTSUP,
857 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
858 					  &rss->types,
859 					  "some RSS protocols are not"
860 					  " supported");
861 	for (i = 0; i != rss->queue_num; ++i) {
862 		if (!(*priv->rxqs)[rss->queue[i]])
863 			return rte_flow_error_set
864 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
865 				 &rss->queue[i], "queue is not configured");
866 	}
867 	return 0;
868 }
869 
870 /*
871  * Validate the count action.
872  *
873  * @param[in] dev
874  *   Pointer to the Ethernet device structure.
875  * @param[out] error
876  *   Pointer to error structure.
877  *
878  * @return
879  *   0 on success, a negative errno value otherwise and rte_ernno is set.
880  */
881 int
882 mlx5_flow_validate_action_count(struct rte_eth_dev *dev,
883 				struct rte_flow_error *error)
884 {
885 	struct priv *priv = dev->data->dev_private;
886 
887 	if (!priv->config.flow_counter_en)
888 		return rte_flow_error_set(error, ENOTSUP,
889 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
890 					  "flow counters are not supported.");
891 	return 0;
892 }
893 
894 /**
895  * Verify the @p attributes will be correctly understood by the NIC and store
896  * them in the @p flow if everything is correct.
897  *
898  * @param[in] dev
899  *   Pointer to the Ethernet device structure.
900  * @param[in] attributes
901  *   Pointer to flow attributes
902  * @param[out] error
903  *   Pointer to error structure.
904  *
905  * @return
906  *   0 on success, a negative errno value otherwise and rte_errno is set.
907  */
908 int
909 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
910 			      const struct rte_flow_attr *attributes,
911 			      struct rte_flow_error *error)
912 {
913 	struct priv *priv = dev->data->dev_private;
914 	uint32_t priority_max = priv->config.flow_prio - 1;
915 
916 	if (attributes->group)
917 		return rte_flow_error_set(error, ENOTSUP,
918 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
919 					  NULL, "groups is not supported");
920 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
921 	    attributes->priority >= priority_max)
922 		return rte_flow_error_set(error, ENOTSUP,
923 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
924 					  NULL, "priority out of range");
925 	if (attributes->egress)
926 		return rte_flow_error_set(error, ENOTSUP,
927 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
928 					  "egress is not supported");
929 	if (attributes->transfer)
930 		return rte_flow_error_set(error, ENOTSUP,
931 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
932 					  NULL, "transfer is not supported");
933 	if (!attributes->ingress)
934 		return rte_flow_error_set(error, EINVAL,
935 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
936 					  NULL,
937 					  "ingress attribute is mandatory");
938 	return 0;
939 }
940 
941 /**
942  * Validate Ethernet item.
943  *
944  * @param[in] item
945  *   Item specification.
946  * @param[in] item_flags
947  *   Bit-fields that holds the items detected until now.
948  * @param[out] error
949  *   Pointer to error structure.
950  *
951  * @return
952  *   0 on success, a negative errno value otherwise and rte_errno is set.
953  */
954 int
955 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
956 			    uint64_t item_flags,
957 			    struct rte_flow_error *error)
958 {
959 	const struct rte_flow_item_eth *mask = item->mask;
960 	const struct rte_flow_item_eth nic_mask = {
961 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
962 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
963 		.type = RTE_BE16(0xffff),
964 	};
965 	int ret;
966 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
967 
968 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L2)
969 		return rte_flow_error_set(error, ENOTSUP,
970 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
971 					  "3 levels of l2 are not supported");
972 	if ((item_flags & MLX5_FLOW_LAYER_INNER_L2) && !tunnel)
973 		return rte_flow_error_set(error, ENOTSUP,
974 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
975 					  "2 L2 without tunnel are not supported");
976 	if (!mask)
977 		mask = &rte_flow_item_eth_mask;
978 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
979 					(const uint8_t *)&nic_mask,
980 					sizeof(struct rte_flow_item_eth),
981 					error);
982 	return ret;
983 }
984 
985 /**
986  * Validate VLAN item.
987  *
988  * @param[in] item
989  *   Item specification.
990  * @param[in] item_flags
991  *   Bit-fields that holds the items detected until now.
992  * @param[out] error
993  *   Pointer to error structure.
994  *
995  * @return
996  *   0 on success, a negative errno value otherwise and rte_errno is set.
997  */
998 int
999 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1000 			     int64_t item_flags,
1001 			     struct rte_flow_error *error)
1002 {
1003 	const struct rte_flow_item_vlan *spec = item->spec;
1004 	const struct rte_flow_item_vlan *mask = item->mask;
1005 	const struct rte_flow_item_vlan nic_mask = {
1006 		.tci = RTE_BE16(0x0fff),
1007 		.inner_type = RTE_BE16(0xffff),
1008 	};
1009 	uint16_t vlan_tag = 0;
1010 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1011 	int ret;
1012 	const uint32_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1013 					MLX5_FLOW_LAYER_INNER_L4) :
1014 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1015 					MLX5_FLOW_LAYER_OUTER_L4);
1016 	const uint32_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1017 					MLX5_FLOW_LAYER_OUTER_VLAN;
1018 
1019 	if (item_flags & vlanm)
1020 		return rte_flow_error_set(error, EINVAL,
1021 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1022 					  "VLAN layer already configured");
1023 	else if ((item_flags & l34m) != 0)
1024 		return rte_flow_error_set(error, EINVAL,
1025 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1026 					  "L2 layer cannot follow L3/L4 layer");
1027 	if (!mask)
1028 		mask = &rte_flow_item_vlan_mask;
1029 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1030 					(const uint8_t *)&nic_mask,
1031 					sizeof(struct rte_flow_item_vlan),
1032 					error);
1033 	if (ret)
1034 		return ret;
1035 	if (spec) {
1036 		vlan_tag = spec->tci;
1037 		vlan_tag &= mask->tci;
1038 	}
1039 	/*
1040 	 * From verbs perspective an empty VLAN is equivalent
1041 	 * to a packet without VLAN layer.
1042 	 */
1043 	if (!vlan_tag)
1044 		return rte_flow_error_set(error, EINVAL,
1045 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1046 					  item->spec,
1047 					  "VLAN cannot be empty");
1048 	return 0;
1049 }
1050 
1051 /**
1052  * Validate IPV4 item.
1053  *
1054  * @param[in] item
1055  *   Item specification.
1056  * @param[in] item_flags
1057  *   Bit-fields that holds the items detected until now.
1058  * @param[out] error
1059  *   Pointer to error structure.
1060  *
1061  * @return
1062  *   0 on success, a negative errno value otherwise and rte_errno is set.
1063  */
1064 int
1065 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1066 			     int64_t item_flags,
1067 			     struct rte_flow_error *error)
1068 {
1069 	const struct rte_flow_item_ipv4 *mask = item->mask;
1070 	const struct rte_flow_item_ipv4 nic_mask = {
1071 		.hdr = {
1072 			.src_addr = RTE_BE32(0xffffffff),
1073 			.dst_addr = RTE_BE32(0xffffffff),
1074 			.type_of_service = 0xff,
1075 			.next_proto_id = 0xff,
1076 		},
1077 	};
1078 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1079 	int ret;
1080 
1081 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1082 				   MLX5_FLOW_LAYER_OUTER_L3))
1083 		return rte_flow_error_set(error, ENOTSUP,
1084 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1085 					  "multiple L3 layers not supported");
1086 	else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1087 					MLX5_FLOW_LAYER_OUTER_L4))
1088 		return rte_flow_error_set(error, EINVAL,
1089 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1090 					  "L3 cannot follow an L4 layer.");
1091 	if (!mask)
1092 		mask = &rte_flow_item_ipv4_mask;
1093 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1094 					(const uint8_t *)&nic_mask,
1095 					sizeof(struct rte_flow_item_ipv4),
1096 					error);
1097 	if (ret < 0)
1098 		return ret;
1099 	return 0;
1100 }
1101 
1102 /**
1103  * Validate IPV6 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_ipv6(const struct rte_flow_item *item,
1117 			     uint64_t item_flags,
1118 			     struct rte_flow_error *error)
1119 {
1120 	const struct rte_flow_item_ipv6 *mask = item->mask;
1121 	const struct rte_flow_item_ipv6 nic_mask = {
1122 		.hdr = {
1123 			.src_addr =
1124 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1125 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1126 			.dst_addr =
1127 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1128 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1129 			.vtc_flow = RTE_BE32(0xffffffff),
1130 			.proto = 0xff,
1131 			.hop_limits = 0xff,
1132 		},
1133 	};
1134 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1135 	int ret;
1136 
1137 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1138 				   MLX5_FLOW_LAYER_OUTER_L3))
1139 		return rte_flow_error_set(error, ENOTSUP,
1140 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1141 					  "multiple L3 layers not supported");
1142 	else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1143 					MLX5_FLOW_LAYER_OUTER_L4))
1144 		return rte_flow_error_set(error, EINVAL,
1145 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1146 					  "L3 cannot follow an L4 layer.");
1147 	/*
1148 	 * IPv6 is not recognised by the NIC inside a GRE tunnel.
1149 	 * Such support has to be disabled as the rule will be
1150 	 * accepted.  Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1151 	 * Mellanox OFED 4.4-1.0.0.0.
1152 	 */
1153 	if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1154 		return rte_flow_error_set(error, ENOTSUP,
1155 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1156 					  "IPv6 inside a GRE tunnel is"
1157 					  " not recognised.");
1158 	if (!mask)
1159 		mask = &rte_flow_item_ipv6_mask;
1160 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1161 					(const uint8_t *)&nic_mask,
1162 					sizeof(struct rte_flow_item_ipv6),
1163 					error);
1164 	if (ret < 0)
1165 		return ret;
1166 	return 0;
1167 }
1168 
1169 /**
1170  * Validate UDP item.
1171  *
1172  * @param[in] item
1173  *   Item specification.
1174  * @param[in] item_flags
1175  *   Bit-fields that holds the items detected until now.
1176  * @param[in] target_protocol
1177  *   The next protocol in the previous item.
1178  * @param[out] error
1179  *   Pointer to error structure.
1180  *
1181  * @return
1182  *   0 on success, a negative errno value otherwise and rte_errno is set.
1183  */
1184 int
1185 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1186 			    uint64_t item_flags,
1187 			    uint8_t target_protocol,
1188 			    struct rte_flow_error *error)
1189 {
1190 	const struct rte_flow_item_udp *mask = item->mask;
1191 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1192 	int ret;
1193 
1194 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1195 		return rte_flow_error_set(error, EINVAL,
1196 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1197 					  "protocol filtering not compatible"
1198 					  " with UDP layer");
1199 	if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1200 				     MLX5_FLOW_LAYER_OUTER_L3)))
1201 		return rte_flow_error_set(error, EINVAL,
1202 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1203 					  "L3 is mandatory to filter on L4");
1204 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1205 				   MLX5_FLOW_LAYER_OUTER_L4))
1206 		return rte_flow_error_set(error, EINVAL,
1207 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1208 					  "L4 layer is already present");
1209 	if (!mask)
1210 		mask = &rte_flow_item_udp_mask;
1211 	ret = mlx5_flow_item_acceptable
1212 		(item, (const uint8_t *)mask,
1213 		 (const uint8_t *)&rte_flow_item_udp_mask,
1214 		 sizeof(struct rte_flow_item_udp), error);
1215 	if (ret < 0)
1216 		return ret;
1217 	return 0;
1218 }
1219 
1220 /**
1221  * Validate TCP item.
1222  *
1223  * @param[in] item
1224  *   Item specification.
1225  * @param[in] item_flags
1226  *   Bit-fields that holds the items detected until now.
1227  * @param[in] target_protocol
1228  *   The next protocol in the previous item.
1229  * @param[out] error
1230  *   Pointer to error structure.
1231  *
1232  * @return
1233  *   0 on success, a negative errno value otherwise and rte_errno is set.
1234  */
1235 int
1236 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1237 			    uint64_t item_flags,
1238 			    uint8_t target_protocol,
1239 			    struct rte_flow_error *error)
1240 {
1241 	const struct rte_flow_item_tcp *mask = item->mask;
1242 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1243 	int ret;
1244 
1245 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1246 		return rte_flow_error_set(error, EINVAL,
1247 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1248 					  "protocol filtering not compatible"
1249 					  " with TCP layer");
1250 	if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1251 				     MLX5_FLOW_LAYER_OUTER_L3)))
1252 		return rte_flow_error_set(error, EINVAL,
1253 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1254 					  "L3 is mandatory to filter on L4");
1255 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1256 				   MLX5_FLOW_LAYER_OUTER_L4))
1257 		return rte_flow_error_set(error, EINVAL,
1258 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1259 					  "L4 layer is already present");
1260 	if (!mask)
1261 		mask = &rte_flow_item_tcp_mask;
1262 	ret = mlx5_flow_item_acceptable
1263 		(item, (const uint8_t *)mask,
1264 		 (const uint8_t *)&rte_flow_item_tcp_mask,
1265 		 sizeof(struct rte_flow_item_tcp), error);
1266 	if (ret < 0)
1267 		return ret;
1268 	return 0;
1269 }
1270 
1271 /**
1272  * Validate VXLAN item.
1273  *
1274  * @param[in] item
1275  *   Item specification.
1276  * @param[in] item_flags
1277  *   Bit-fields that holds the items detected until now.
1278  * @param[in] target_protocol
1279  *   The next protocol in the previous item.
1280  * @param[out] error
1281  *   Pointer to error structure.
1282  *
1283  * @return
1284  *   0 on success, a negative errno value otherwise and rte_errno is set.
1285  */
1286 int
1287 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1288 			      uint64_t item_flags,
1289 			      struct rte_flow_error *error)
1290 {
1291 	const struct rte_flow_item_vxlan *spec = item->spec;
1292 	const struct rte_flow_item_vxlan *mask = item->mask;
1293 	int ret;
1294 	union vni {
1295 		uint32_t vlan_id;
1296 		uint8_t vni[4];
1297 	} id = { .vlan_id = 0, };
1298 	uint32_t vlan_id = 0;
1299 
1300 
1301 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1302 		return rte_flow_error_set(error, ENOTSUP,
1303 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1304 					  "a tunnel is already present");
1305 	/*
1306 	 * Verify only UDPv4 is present as defined in
1307 	 * https://tools.ietf.org/html/rfc7348
1308 	 */
1309 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1310 		return rte_flow_error_set(error, EINVAL,
1311 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1312 					  "no outer UDP layer found");
1313 	if (!mask)
1314 		mask = &rte_flow_item_vxlan_mask;
1315 	ret = mlx5_flow_item_acceptable
1316 		(item, (const uint8_t *)mask,
1317 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1318 		 sizeof(struct rte_flow_item_vxlan),
1319 		 error);
1320 	if (ret < 0)
1321 		return ret;
1322 	if (spec) {
1323 		memcpy(&id.vni[1], spec->vni, 3);
1324 		vlan_id = id.vlan_id;
1325 		memcpy(&id.vni[1], mask->vni, 3);
1326 		vlan_id &= id.vlan_id;
1327 	}
1328 	/*
1329 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1330 	 * only this layer is defined in the Verbs specification it is
1331 	 * interpreted as wildcard and all packets will match this
1332 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1333 	 * udp), all packets matching the layers before will also
1334 	 * match this rule.  To avoid such situation, VNI 0 is
1335 	 * currently refused.
1336 	 */
1337 	if (!vlan_id)
1338 		return rte_flow_error_set(error, ENOTSUP,
1339 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1340 					  "VXLAN vni cannot be 0");
1341 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1342 		return rte_flow_error_set(error, ENOTSUP,
1343 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1344 					  "VXLAN tunnel must be fully defined");
1345 	return 0;
1346 }
1347 
1348 /**
1349  * Validate VXLAN_GPE item.
1350  *
1351  * @param[in] item
1352  *   Item specification.
1353  * @param[in] item_flags
1354  *   Bit-fields that holds the items detected until now.
1355  * @param[in] priv
1356  *   Pointer to the private data structure.
1357  * @param[in] target_protocol
1358  *   The next protocol in the previous item.
1359  * @param[out] error
1360  *   Pointer to error structure.
1361  *
1362  * @return
1363  *   0 on success, a negative errno value otherwise and rte_errno is set.
1364  */
1365 int
1366 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1367 				  uint64_t item_flags,
1368 				  struct rte_eth_dev *dev,
1369 				  struct rte_flow_error *error)
1370 {
1371 	struct priv *priv = dev->data->dev_private;
1372 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1373 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1374 	int ret;
1375 	union vni {
1376 		uint32_t vlan_id;
1377 		uint8_t vni[4];
1378 	} id = { .vlan_id = 0, };
1379 	uint32_t vlan_id = 0;
1380 
1381 	if (!priv->config.l3_vxlan_en)
1382 		return rte_flow_error_set(error, ENOTSUP,
1383 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1384 					  "L3 VXLAN is not enabled by device"
1385 					  " parameter and/or not configured in"
1386 					  " firmware");
1387 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1388 		return rte_flow_error_set(error, ENOTSUP,
1389 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1390 					  "a tunnel is already present");
1391 	/*
1392 	 * Verify only UDPv4 is present as defined in
1393 	 * https://tools.ietf.org/html/rfc7348
1394 	 */
1395 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1396 		return rte_flow_error_set(error, EINVAL,
1397 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1398 					  "no outer UDP layer found");
1399 	if (!mask)
1400 		mask = &rte_flow_item_vxlan_gpe_mask;
1401 	ret = mlx5_flow_item_acceptable
1402 		(item, (const uint8_t *)mask,
1403 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1404 		 sizeof(struct rte_flow_item_vxlan_gpe),
1405 		 error);
1406 	if (ret < 0)
1407 		return ret;
1408 	if (spec) {
1409 		if (spec->protocol)
1410 			return rte_flow_error_set(error, ENOTSUP,
1411 						  RTE_FLOW_ERROR_TYPE_ITEM,
1412 						  item,
1413 						  "VxLAN-GPE protocol"
1414 						  " not supported");
1415 		memcpy(&id.vni[1], spec->vni, 3);
1416 		vlan_id = id.vlan_id;
1417 		memcpy(&id.vni[1], mask->vni, 3);
1418 		vlan_id &= id.vlan_id;
1419 	}
1420 	/*
1421 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1422 	 * layer is defined in the Verbs specification it is interpreted as
1423 	 * wildcard and all packets will match this rule, if it follows a full
1424 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1425 	 * before will also match this rule.  To avoid such situation, VNI 0
1426 	 * is currently refused.
1427 	 */
1428 	if (!vlan_id)
1429 		return rte_flow_error_set(error, ENOTSUP,
1430 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1431 					  "VXLAN-GPE vni cannot be 0");
1432 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1433 		return rte_flow_error_set(error, ENOTSUP,
1434 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1435 					  "VXLAN-GPE tunnel must be fully"
1436 					  " defined");
1437 	return 0;
1438 }
1439 
1440 /**
1441  * Validate GRE item.
1442  *
1443  * @param[in] item
1444  *   Item specification.
1445  * @param[in] item_flags
1446  *   Bit flags to mark detected items.
1447  * @param[in] target_protocol
1448  *   The next protocol in the previous item.
1449  * @param[out] error
1450  *   Pointer to error structure.
1451  *
1452  * @return
1453  *   0 on success, a negative errno value otherwise and rte_errno is set.
1454  */
1455 int
1456 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1457 			    uint64_t item_flags,
1458 			    uint8_t target_protocol,
1459 			    struct rte_flow_error *error)
1460 {
1461 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1462 	const struct rte_flow_item_gre *mask = item->mask;
1463 	int ret;
1464 
1465 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1466 		return rte_flow_error_set(error, EINVAL,
1467 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1468 					  "protocol filtering not compatible"
1469 					  " with this GRE layer");
1470 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1471 		return rte_flow_error_set(error, ENOTSUP,
1472 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1473 					  "a tunnel is already present");
1474 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1475 		return rte_flow_error_set(error, ENOTSUP,
1476 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1477 					  "L3 Layer is missing");
1478 	if (!mask)
1479 		mask = &rte_flow_item_gre_mask;
1480 	ret = mlx5_flow_item_acceptable
1481 		(item, (const uint8_t *)mask,
1482 		 (const uint8_t *)&rte_flow_item_gre_mask,
1483 		 sizeof(struct rte_flow_item_gre), error);
1484 	if (ret < 0)
1485 		return ret;
1486 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1487 	if (spec && (spec->protocol & mask->protocol))
1488 		return rte_flow_error_set(error, ENOTSUP,
1489 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1490 					  "without MPLS support the"
1491 					  " specification cannot be used for"
1492 					  " filtering");
1493 #endif
1494 	return 0;
1495 }
1496 
1497 /**
1498  * Validate MPLS item.
1499  *
1500  * @param[in] item
1501  *   Item specification.
1502  * @param[in] item_flags
1503  *   Bit-fields that holds the items detected until now.
1504  * @param[in] target_protocol
1505  *   The next protocol in the previous item.
1506  * @param[out] error
1507  *   Pointer to error structure.
1508  *
1509  * @return
1510  *   0 on success, a negative errno value otherwise and rte_errno is set.
1511  */
1512 int
1513 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1514 			     uint64_t item_flags __rte_unused,
1515 			     uint8_t target_protocol __rte_unused,
1516 			     struct rte_flow_error *error)
1517 {
1518 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1519 	const struct rte_flow_item_mpls *mask = item->mask;
1520 	int ret;
1521 
1522 	if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1523 		return rte_flow_error_set(error, EINVAL,
1524 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1525 					  "protocol filtering not compatible"
1526 					  " with MPLS layer");
1527 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1528 		return rte_flow_error_set(error, ENOTSUP,
1529 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1530 					  "a tunnel is already"
1531 					  " present");
1532 	if (!mask)
1533 		mask = &rte_flow_item_mpls_mask;
1534 	ret = mlx5_flow_item_acceptable
1535 		(item, (const uint8_t *)mask,
1536 		 (const uint8_t *)&rte_flow_item_mpls_mask,
1537 		 sizeof(struct rte_flow_item_mpls), error);
1538 	if (ret < 0)
1539 		return ret;
1540 	return 0;
1541 #endif
1542 	return rte_flow_error_set(error, ENOTSUP,
1543 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
1544 				  "MPLS is not supported by Verbs, please"
1545 				  " update.");
1546 }
1547 
1548 static int
1549 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1550 		   const struct rte_flow_attr *attr __rte_unused,
1551 		   const struct rte_flow_item items[] __rte_unused,
1552 		   const struct rte_flow_action actions[] __rte_unused,
1553 		   struct rte_flow_error *error __rte_unused)
1554 {
1555 	rte_errno = ENOTSUP;
1556 	return -rte_errno;
1557 }
1558 
1559 static struct mlx5_flow *
1560 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1561 		  const struct rte_flow_item items[] __rte_unused,
1562 		  const struct rte_flow_action actions[] __rte_unused,
1563 		  uint64_t *item_flags __rte_unused,
1564 		  uint64_t *action_flags __rte_unused,
1565 		  struct rte_flow_error *error __rte_unused)
1566 {
1567 	rte_errno = ENOTSUP;
1568 	return NULL;
1569 }
1570 
1571 static int
1572 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1573 		    struct mlx5_flow *dev_flow __rte_unused,
1574 		    const struct rte_flow_attr *attr __rte_unused,
1575 		    const struct rte_flow_item items[] __rte_unused,
1576 		    const struct rte_flow_action actions[] __rte_unused,
1577 		    struct rte_flow_error *error __rte_unused)
1578 {
1579 	rte_errno = ENOTSUP;
1580 	return -rte_errno;
1581 }
1582 
1583 static int
1584 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1585 		struct rte_flow *flow __rte_unused,
1586 		struct rte_flow_error *error __rte_unused)
1587 {
1588 	rte_errno = ENOTSUP;
1589 	return -rte_errno;
1590 }
1591 
1592 static void
1593 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1594 		 struct rte_flow *flow __rte_unused)
1595 {
1596 }
1597 
1598 static void
1599 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1600 		  struct rte_flow *flow __rte_unused)
1601 {
1602 }
1603 
1604 /* Void driver to protect from null pointer reference. */
1605 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1606 	.validate = flow_null_validate,
1607 	.prepare = flow_null_prepare,
1608 	.translate = flow_null_translate,
1609 	.apply = flow_null_apply,
1610 	.remove = flow_null_remove,
1611 	.destroy = flow_null_destroy,
1612 };
1613 
1614 /**
1615  * Select flow driver type according to flow attributes and device
1616  * configuration.
1617  *
1618  * @param[in] dev
1619  *   Pointer to the dev structure.
1620  * @param[in] attr
1621  *   Pointer to the flow attributes.
1622  *
1623  * @return
1624  *   flow driver type if supported, MLX5_FLOW_TYPE_MAX otherwise.
1625  */
1626 static enum mlx5_flow_drv_type
1627 flow_get_drv_type(struct rte_eth_dev *dev __rte_unused,
1628 		  const struct rte_flow_attr *attr)
1629 {
1630 	struct priv *priv __rte_unused = dev->data->dev_private;
1631 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1632 
1633 	if (attr->transfer) {
1634 		type = MLX5_FLOW_TYPE_TCF;
1635 	} else {
1636 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
1637 		type = priv->config.dv_flow_en ?  MLX5_FLOW_TYPE_DV :
1638 						  MLX5_FLOW_TYPE_VERBS;
1639 #else
1640 		type = MLX5_FLOW_TYPE_VERBS;
1641 #endif
1642 	}
1643 	return type;
1644 }
1645 
1646 #define flow_get_drv_ops(type) flow_drv_ops[type]
1647 
1648 /**
1649  * Flow driver validation API. This abstracts calling driver specific functions.
1650  * The type of flow driver is determined according to flow attributes.
1651  *
1652  * @param[in] dev
1653  *   Pointer to the dev structure.
1654  * @param[in] attr
1655  *   Pointer to the flow attributes.
1656  * @param[in] items
1657  *   Pointer to the list of items.
1658  * @param[in] actions
1659  *   Pointer to the list of actions.
1660  * @param[out] error
1661  *   Pointer to the error structure.
1662  *
1663  * @return
1664  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1665  */
1666 static inline int
1667 flow_drv_validate(struct rte_eth_dev *dev,
1668 		  const struct rte_flow_attr *attr,
1669 		  const struct rte_flow_item items[],
1670 		  const struct rte_flow_action actions[],
1671 		  struct rte_flow_error *error)
1672 {
1673 	const struct mlx5_flow_driver_ops *fops;
1674 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1675 
1676 	fops = flow_get_drv_ops(type);
1677 	return fops->validate(dev, attr, items, actions, error);
1678 }
1679 
1680 /**
1681  * Flow driver preparation API. This abstracts calling driver specific
1682  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1683  * calculates the size of memory required for device flow, allocates the memory,
1684  * initializes the device flow and returns the pointer.
1685  *
1686  * @param[in] attr
1687  *   Pointer to the flow attributes.
1688  * @param[in] items
1689  *   Pointer to the list of items.
1690  * @param[in] actions
1691  *   Pointer to the list of actions.
1692  * @param[out] item_flags
1693  *   Pointer to bit mask of all items detected.
1694  * @param[out] action_flags
1695  *   Pointer to bit mask of all actions detected.
1696  * @param[out] error
1697  *   Pointer to the error structure.
1698  *
1699  * @return
1700  *   Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1701  */
1702 static inline struct mlx5_flow *
1703 flow_drv_prepare(struct rte_flow *flow,
1704 		 const struct rte_flow_attr *attr,
1705 		 const struct rte_flow_item items[],
1706 		 const struct rte_flow_action actions[],
1707 		 uint64_t *item_flags,
1708 		 uint64_t *action_flags,
1709 		 struct rte_flow_error *error)
1710 {
1711 	const struct mlx5_flow_driver_ops *fops;
1712 	enum mlx5_flow_drv_type type = flow->drv_type;
1713 
1714 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1715 	fops = flow_get_drv_ops(type);
1716 	return fops->prepare(attr, items, actions, item_flags, action_flags,
1717 			     error);
1718 }
1719 
1720 /**
1721  * Flow driver translation API. This abstracts calling driver specific
1722  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1723  * translates a generic flow into a driver flow. flow_drv_prepare() must
1724  * precede.
1725  *
1726  *
1727  * @param[in] dev
1728  *   Pointer to the rte dev structure.
1729  * @param[in, out] dev_flow
1730  *   Pointer to the mlx5 flow.
1731  * @param[in] attr
1732  *   Pointer to the flow attributes.
1733  * @param[in] items
1734  *   Pointer to the list of items.
1735  * @param[in] actions
1736  *   Pointer to the list of actions.
1737  * @param[out] error
1738  *   Pointer to the error structure.
1739  *
1740  * @return
1741  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1742  */
1743 static inline int
1744 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1745 		   const struct rte_flow_attr *attr,
1746 		   const struct rte_flow_item items[],
1747 		   const struct rte_flow_action actions[],
1748 		   struct rte_flow_error *error)
1749 {
1750 	const struct mlx5_flow_driver_ops *fops;
1751 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1752 
1753 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1754 	fops = flow_get_drv_ops(type);
1755 	return fops->translate(dev, dev_flow, attr, items, actions, error);
1756 }
1757 
1758 /**
1759  * Flow driver apply API. This abstracts calling driver specific functions.
1760  * Parent flow (rte_flow) should have driver type (drv_type). It applies
1761  * translated driver flows on to device. flow_drv_translate() must precede.
1762  *
1763  * @param[in] dev
1764  *   Pointer to Ethernet device structure.
1765  * @param[in, out] flow
1766  *   Pointer to flow structure.
1767  * @param[out] error
1768  *   Pointer to error structure.
1769  *
1770  * @return
1771  *   0 on success, a negative errno value otherwise and rte_errno is set.
1772  */
1773 static inline int
1774 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1775 	       struct rte_flow_error *error)
1776 {
1777 	const struct mlx5_flow_driver_ops *fops;
1778 	enum mlx5_flow_drv_type type = flow->drv_type;
1779 
1780 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1781 	fops = flow_get_drv_ops(type);
1782 	return fops->apply(dev, flow, error);
1783 }
1784 
1785 /**
1786  * Flow driver remove API. This abstracts calling driver specific functions.
1787  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1788  * on device. All the resources of the flow should be freed by calling
1789  * flow_dv_destroy().
1790  *
1791  * @param[in] dev
1792  *   Pointer to Ethernet device.
1793  * @param[in, out] flow
1794  *   Pointer to flow structure.
1795  */
1796 static inline void
1797 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1798 {
1799 	const struct mlx5_flow_driver_ops *fops;
1800 	enum mlx5_flow_drv_type type = flow->drv_type;
1801 
1802 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1803 	fops = flow_get_drv_ops(type);
1804 	fops->remove(dev, flow);
1805 }
1806 
1807 /**
1808  * Flow driver destroy API. This abstracts calling driver specific functions.
1809  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1810  * on device and releases resources of the flow.
1811  *
1812  * @param[in] dev
1813  *   Pointer to Ethernet device.
1814  * @param[in, out] flow
1815  *   Pointer to flow structure.
1816  */
1817 static inline void
1818 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1819 {
1820 	const struct mlx5_flow_driver_ops *fops;
1821 	enum mlx5_flow_drv_type type = flow->drv_type;
1822 
1823 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1824 	fops = flow_get_drv_ops(type);
1825 	fops->destroy(dev, flow);
1826 }
1827 
1828 /**
1829  * Validate a flow supported by the NIC.
1830  *
1831  * @see rte_flow_validate()
1832  * @see rte_flow_ops
1833  */
1834 int
1835 mlx5_flow_validate(struct rte_eth_dev *dev,
1836 		   const struct rte_flow_attr *attr,
1837 		   const struct rte_flow_item items[],
1838 		   const struct rte_flow_action actions[],
1839 		   struct rte_flow_error *error)
1840 {
1841 	int ret;
1842 
1843 	ret = flow_drv_validate(dev, attr, items, actions, error);
1844 	if (ret < 0)
1845 		return ret;
1846 	return 0;
1847 }
1848 
1849 /**
1850  * Get RSS action from the action list.
1851  *
1852  * @param[in] actions
1853  *   Pointer to the list of actions.
1854  *
1855  * @return
1856  *   Pointer to the RSS action if exist, else return NULL.
1857  */
1858 static const struct rte_flow_action_rss*
1859 mlx5_flow_get_rss_action(const struct rte_flow_action actions[])
1860 {
1861 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1862 		switch (actions->type) {
1863 		case RTE_FLOW_ACTION_TYPE_RSS:
1864 			return (const struct rte_flow_action_rss *)
1865 			       actions->conf;
1866 		default:
1867 			break;
1868 		}
1869 	}
1870 	return NULL;
1871 }
1872 
1873 static unsigned int
1874 mlx5_find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
1875 {
1876 	const struct rte_flow_item *item;
1877 	unsigned int has_vlan = 0;
1878 
1879 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
1880 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
1881 			has_vlan = 1;
1882 			break;
1883 		}
1884 	}
1885 	if (has_vlan)
1886 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
1887 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
1888 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
1889 			       MLX5_EXPANSION_ROOT_OUTER;
1890 }
1891 
1892 /**
1893  * Create a flow and add it to @p list.
1894  *
1895  * @param dev
1896  *   Pointer to Ethernet device.
1897  * @param list
1898  *   Pointer to a TAILQ flow list.
1899  * @param[in] attr
1900  *   Flow rule attributes.
1901  * @param[in] items
1902  *   Pattern specification (list terminated by the END pattern item).
1903  * @param[in] actions
1904  *   Associated actions (list terminated by the END action).
1905  * @param[out] error
1906  *   Perform verbose error reporting if not NULL.
1907  *
1908  * @return
1909  *   A flow on success, NULL otherwise and rte_errno is set.
1910  */
1911 static struct rte_flow *
1912 mlx5_flow_list_create(struct rte_eth_dev *dev,
1913 		      struct mlx5_flows *list,
1914 		      const struct rte_flow_attr *attr,
1915 		      const struct rte_flow_item items[],
1916 		      const struct rte_flow_action actions[],
1917 		      struct rte_flow_error *error)
1918 {
1919 	struct rte_flow *flow = NULL;
1920 	struct mlx5_flow *dev_flow;
1921 	uint64_t action_flags = 0;
1922 	uint64_t item_flags = 0;
1923 	const struct rte_flow_action_rss *rss;
1924 	union {
1925 		struct rte_flow_expand_rss buf;
1926 		uint8_t buffer[2048];
1927 	} expand_buffer;
1928 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
1929 	int ret;
1930 	uint32_t i;
1931 	uint32_t flow_size;
1932 
1933 	ret = flow_drv_validate(dev, attr, items, actions, error);
1934 	if (ret < 0)
1935 		return NULL;
1936 	flow_size = sizeof(struct rte_flow);
1937 	rss = mlx5_flow_get_rss_action(actions);
1938 	if (rss)
1939 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
1940 					    sizeof(void *));
1941 	else
1942 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
1943 	flow = rte_calloc(__func__, 1, flow_size, 0);
1944 	flow->drv_type = flow_get_drv_type(dev, attr);
1945 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
1946 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
1947 	flow->queue = (void *)(flow + 1);
1948 	LIST_INIT(&flow->dev_flows);
1949 	if (rss && rss->types) {
1950 		unsigned int graph_root;
1951 
1952 		graph_root = mlx5_find_graph_root(items, rss->level);
1953 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
1954 					  items, rss->types,
1955 					  mlx5_support_expansion,
1956 					  graph_root);
1957 		assert(ret > 0 &&
1958 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
1959 	} else {
1960 		buf->entries = 1;
1961 		buf->entry[0].pattern = (void *)(uintptr_t)items;
1962 	}
1963 	for (i = 0; i < buf->entries; ++i) {
1964 		dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
1965 					    actions, &item_flags, &action_flags,
1966 					    error);
1967 		if (!dev_flow)
1968 			goto error;
1969 		dev_flow->flow = flow;
1970 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
1971 		ret = flow_drv_translate(dev, dev_flow, attr,
1972 					 buf->entry[i].pattern,
1973 					 actions, error);
1974 		if (ret < 0)
1975 			goto error;
1976 	}
1977 	if (dev->data->dev_started) {
1978 		ret = flow_drv_apply(dev, flow, error);
1979 		if (ret < 0)
1980 			goto error;
1981 	}
1982 	TAILQ_INSERT_TAIL(list, flow, next);
1983 	mlx5_flow_rxq_flags_set(dev, flow);
1984 	return flow;
1985 error:
1986 	ret = rte_errno; /* Save rte_errno before cleanup. */
1987 	assert(flow);
1988 	flow_drv_destroy(dev, flow);
1989 	rte_free(flow);
1990 	rte_errno = ret; /* Restore rte_errno. */
1991 	return NULL;
1992 }
1993 
1994 /**
1995  * Create a flow.
1996  *
1997  * @see rte_flow_create()
1998  * @see rte_flow_ops
1999  */
2000 struct rte_flow *
2001 mlx5_flow_create(struct rte_eth_dev *dev,
2002 		 const struct rte_flow_attr *attr,
2003 		 const struct rte_flow_item items[],
2004 		 const struct rte_flow_action actions[],
2005 		 struct rte_flow_error *error)
2006 {
2007 	return mlx5_flow_list_create
2008 		(dev, &((struct priv *)dev->data->dev_private)->flows,
2009 		 attr, items, actions, error);
2010 }
2011 
2012 /**
2013  * Destroy a flow in a list.
2014  *
2015  * @param dev
2016  *   Pointer to Ethernet device.
2017  * @param list
2018  *   Pointer to a TAILQ flow list.
2019  * @param[in] flow
2020  *   Flow to destroy.
2021  */
2022 static void
2023 mlx5_flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2024 		       struct rte_flow *flow)
2025 {
2026 	flow_drv_destroy(dev, flow);
2027 	TAILQ_REMOVE(list, flow, next);
2028 	/*
2029 	 * Update RX queue flags only if port is started, otherwise it is
2030 	 * already clean.
2031 	 */
2032 	if (dev->data->dev_started)
2033 		mlx5_flow_rxq_flags_trim(dev, flow);
2034 	rte_free(flow);
2035 }
2036 
2037 /**
2038  * Destroy all flows.
2039  *
2040  * @param dev
2041  *   Pointer to Ethernet device.
2042  * @param list
2043  *   Pointer to a TAILQ flow list.
2044  */
2045 void
2046 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2047 {
2048 	while (!TAILQ_EMPTY(list)) {
2049 		struct rte_flow *flow;
2050 
2051 		flow = TAILQ_FIRST(list);
2052 		mlx5_flow_list_destroy(dev, list, flow);
2053 	}
2054 }
2055 
2056 /**
2057  * Remove all flows.
2058  *
2059  * @param dev
2060  *   Pointer to Ethernet device.
2061  * @param list
2062  *   Pointer to a TAILQ flow list.
2063  */
2064 void
2065 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2066 {
2067 	struct rte_flow *flow;
2068 
2069 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2070 		flow_drv_remove(dev, flow);
2071 	mlx5_flow_rxq_flags_clear(dev);
2072 }
2073 
2074 /**
2075  * Add all flows.
2076  *
2077  * @param dev
2078  *   Pointer to Ethernet device.
2079  * @param list
2080  *   Pointer to a TAILQ flow list.
2081  *
2082  * @return
2083  *   0 on success, a negative errno value otherwise and rte_errno is set.
2084  */
2085 int
2086 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2087 {
2088 	struct rte_flow *flow;
2089 	struct rte_flow_error error;
2090 	int ret = 0;
2091 
2092 	TAILQ_FOREACH(flow, list, next) {
2093 		ret = flow_drv_apply(dev, flow, &error);
2094 		if (ret < 0)
2095 			goto error;
2096 		mlx5_flow_rxq_flags_set(dev, flow);
2097 	}
2098 	return 0;
2099 error:
2100 	ret = rte_errno; /* Save rte_errno before cleanup. */
2101 	mlx5_flow_stop(dev, list);
2102 	rte_errno = ret; /* Restore rte_errno. */
2103 	return -rte_errno;
2104 }
2105 
2106 /**
2107  * Verify the flow list is empty
2108  *
2109  * @param dev
2110  *  Pointer to Ethernet device.
2111  *
2112  * @return the number of flows not released.
2113  */
2114 int
2115 mlx5_flow_verify(struct rte_eth_dev *dev)
2116 {
2117 	struct priv *priv = dev->data->dev_private;
2118 	struct rte_flow *flow;
2119 	int ret = 0;
2120 
2121 	TAILQ_FOREACH(flow, &priv->flows, next) {
2122 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
2123 			dev->data->port_id, (void *)flow);
2124 		++ret;
2125 	}
2126 	return ret;
2127 }
2128 
2129 /**
2130  * Enable a control flow configured from the control plane.
2131  *
2132  * @param dev
2133  *   Pointer to Ethernet device.
2134  * @param eth_spec
2135  *   An Ethernet flow spec to apply.
2136  * @param eth_mask
2137  *   An Ethernet flow mask to apply.
2138  * @param vlan_spec
2139  *   A VLAN flow spec to apply.
2140  * @param vlan_mask
2141  *   A VLAN flow mask to apply.
2142  *
2143  * @return
2144  *   0 on success, a negative errno value otherwise and rte_errno is set.
2145  */
2146 int
2147 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2148 		    struct rte_flow_item_eth *eth_spec,
2149 		    struct rte_flow_item_eth *eth_mask,
2150 		    struct rte_flow_item_vlan *vlan_spec,
2151 		    struct rte_flow_item_vlan *vlan_mask)
2152 {
2153 	struct priv *priv = dev->data->dev_private;
2154 	const struct rte_flow_attr attr = {
2155 		.ingress = 1,
2156 		.priority = MLX5_FLOW_PRIO_RSVD,
2157 	};
2158 	struct rte_flow_item items[] = {
2159 		{
2160 			.type = RTE_FLOW_ITEM_TYPE_ETH,
2161 			.spec = eth_spec,
2162 			.last = NULL,
2163 			.mask = eth_mask,
2164 		},
2165 		{
2166 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2167 					      RTE_FLOW_ITEM_TYPE_END,
2168 			.spec = vlan_spec,
2169 			.last = NULL,
2170 			.mask = vlan_mask,
2171 		},
2172 		{
2173 			.type = RTE_FLOW_ITEM_TYPE_END,
2174 		},
2175 	};
2176 	uint16_t queue[priv->reta_idx_n];
2177 	struct rte_flow_action_rss action_rss = {
2178 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2179 		.level = 0,
2180 		.types = priv->rss_conf.rss_hf,
2181 		.key_len = priv->rss_conf.rss_key_len,
2182 		.queue_num = priv->reta_idx_n,
2183 		.key = priv->rss_conf.rss_key,
2184 		.queue = queue,
2185 	};
2186 	struct rte_flow_action actions[] = {
2187 		{
2188 			.type = RTE_FLOW_ACTION_TYPE_RSS,
2189 			.conf = &action_rss,
2190 		},
2191 		{
2192 			.type = RTE_FLOW_ACTION_TYPE_END,
2193 		},
2194 	};
2195 	struct rte_flow *flow;
2196 	struct rte_flow_error error;
2197 	unsigned int i;
2198 
2199 	if (!priv->reta_idx_n) {
2200 		rte_errno = EINVAL;
2201 		return -rte_errno;
2202 	}
2203 	for (i = 0; i != priv->reta_idx_n; ++i)
2204 		queue[i] = (*priv->reta_idx)[i];
2205 	flow = mlx5_flow_list_create(dev, &priv->ctrl_flows, &attr, items,
2206 				     actions, &error);
2207 	if (!flow)
2208 		return -rte_errno;
2209 	return 0;
2210 }
2211 
2212 /**
2213  * Enable a flow control configured from the control plane.
2214  *
2215  * @param dev
2216  *   Pointer to Ethernet device.
2217  * @param eth_spec
2218  *   An Ethernet flow spec to apply.
2219  * @param eth_mask
2220  *   An Ethernet flow mask to apply.
2221  *
2222  * @return
2223  *   0 on success, a negative errno value otherwise and rte_errno is set.
2224  */
2225 int
2226 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2227 	       struct rte_flow_item_eth *eth_spec,
2228 	       struct rte_flow_item_eth *eth_mask)
2229 {
2230 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2231 }
2232 
2233 /**
2234  * Destroy a flow.
2235  *
2236  * @see rte_flow_destroy()
2237  * @see rte_flow_ops
2238  */
2239 int
2240 mlx5_flow_destroy(struct rte_eth_dev *dev,
2241 		  struct rte_flow *flow,
2242 		  struct rte_flow_error *error __rte_unused)
2243 {
2244 	struct priv *priv = dev->data->dev_private;
2245 
2246 	mlx5_flow_list_destroy(dev, &priv->flows, flow);
2247 	return 0;
2248 }
2249 
2250 /**
2251  * Destroy all flows.
2252  *
2253  * @see rte_flow_flush()
2254  * @see rte_flow_ops
2255  */
2256 int
2257 mlx5_flow_flush(struct rte_eth_dev *dev,
2258 		struct rte_flow_error *error __rte_unused)
2259 {
2260 	struct priv *priv = dev->data->dev_private;
2261 
2262 	mlx5_flow_list_flush(dev, &priv->flows);
2263 	return 0;
2264 }
2265 
2266 /**
2267  * Isolated mode.
2268  *
2269  * @see rte_flow_isolate()
2270  * @see rte_flow_ops
2271  */
2272 int
2273 mlx5_flow_isolate(struct rte_eth_dev *dev,
2274 		  int enable,
2275 		  struct rte_flow_error *error)
2276 {
2277 	struct priv *priv = dev->data->dev_private;
2278 
2279 	if (dev->data->dev_started) {
2280 		rte_flow_error_set(error, EBUSY,
2281 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2282 				   NULL,
2283 				   "port must be stopped first");
2284 		return -rte_errno;
2285 	}
2286 	priv->isolated = !!enable;
2287 	if (enable)
2288 		dev->dev_ops = &mlx5_dev_ops_isolate;
2289 	else
2290 		dev->dev_ops = &mlx5_dev_ops;
2291 	return 0;
2292 }
2293 
2294 /**
2295  * Query flow counter.
2296  *
2297  * @param flow
2298  *   Pointer to the flow.
2299  *
2300  * @return
2301  *   0 on success, a negative errno value otherwise and rte_errno is set.
2302  */
2303 static int
2304 mlx5_flow_query_count(struct rte_flow *flow __rte_unused,
2305 		      void *data __rte_unused,
2306 		      struct rte_flow_error *error)
2307 {
2308 #ifdef HAVE_IBV_DEVICE_COUNTERS_SET_SUPPORT
2309 	if (flow->actions & MLX5_FLOW_ACTION_COUNT) {
2310 		struct rte_flow_query_count *qc = data;
2311 		uint64_t counters[2] = {0, 0};
2312 		struct ibv_query_counter_set_attr query_cs_attr = {
2313 			.cs = flow->counter->cs,
2314 			.query_flags = IBV_COUNTER_SET_FORCE_UPDATE,
2315 		};
2316 		struct ibv_counter_set_data query_out = {
2317 			.out = counters,
2318 			.outlen = 2 * sizeof(uint64_t),
2319 		};
2320 		int err = mlx5_glue->query_counter_set(&query_cs_attr,
2321 						       &query_out);
2322 
2323 		if (err)
2324 			return rte_flow_error_set
2325 				(error, err,
2326 				 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2327 				 NULL,
2328 				 "cannot read counter");
2329 		qc->hits_set = 1;
2330 		qc->bytes_set = 1;
2331 		qc->hits = counters[0] - flow->counter->hits;
2332 		qc->bytes = counters[1] - flow->counter->bytes;
2333 		if (qc->reset) {
2334 			flow->counter->hits = counters[0];
2335 			flow->counter->bytes = counters[1];
2336 		}
2337 		return 0;
2338 	}
2339 	return rte_flow_error_set(error, ENOTSUP,
2340 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2341 				  NULL,
2342 				  "flow does not have counter");
2343 #endif
2344 	return rte_flow_error_set(error, ENOTSUP,
2345 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2346 				  NULL,
2347 				  "counters are not available");
2348 }
2349 
2350 /**
2351  * Query a flows.
2352  *
2353  * @see rte_flow_query()
2354  * @see rte_flow_ops
2355  */
2356 int
2357 mlx5_flow_query(struct rte_eth_dev *dev __rte_unused,
2358 		struct rte_flow *flow,
2359 		const struct rte_flow_action *actions,
2360 		void *data,
2361 		struct rte_flow_error *error)
2362 {
2363 	int ret = 0;
2364 
2365 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2366 		switch (actions->type) {
2367 		case RTE_FLOW_ACTION_TYPE_VOID:
2368 			break;
2369 		case RTE_FLOW_ACTION_TYPE_COUNT:
2370 			ret = mlx5_flow_query_count(flow, data, error);
2371 			break;
2372 		default:
2373 			return rte_flow_error_set(error, ENOTSUP,
2374 						  RTE_FLOW_ERROR_TYPE_ACTION,
2375 						  actions,
2376 						  "action not supported");
2377 		}
2378 		if (ret < 0)
2379 			return ret;
2380 	}
2381 	return 0;
2382 }
2383 
2384 /**
2385  * Convert a flow director filter to a generic flow.
2386  *
2387  * @param dev
2388  *   Pointer to Ethernet device.
2389  * @param fdir_filter
2390  *   Flow director filter to add.
2391  * @param attributes
2392  *   Generic flow parameters structure.
2393  *
2394  * @return
2395  *   0 on success, a negative errno value otherwise and rte_errno is set.
2396  */
2397 static int
2398 mlx5_fdir_filter_convert(struct rte_eth_dev *dev,
2399 			 const struct rte_eth_fdir_filter *fdir_filter,
2400 			 struct mlx5_fdir *attributes)
2401 {
2402 	struct priv *priv = dev->data->dev_private;
2403 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
2404 	const struct rte_eth_fdir_masks *mask =
2405 		&dev->data->dev_conf.fdir_conf.mask;
2406 
2407 	/* Validate queue number. */
2408 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2409 		DRV_LOG(ERR, "port %u invalid queue number %d",
2410 			dev->data->port_id, fdir_filter->action.rx_queue);
2411 		rte_errno = EINVAL;
2412 		return -rte_errno;
2413 	}
2414 	attributes->attr.ingress = 1;
2415 	attributes->items[0] = (struct rte_flow_item) {
2416 		.type = RTE_FLOW_ITEM_TYPE_ETH,
2417 		.spec = &attributes->l2,
2418 		.mask = &attributes->l2_mask,
2419 	};
2420 	switch (fdir_filter->action.behavior) {
2421 	case RTE_ETH_FDIR_ACCEPT:
2422 		attributes->actions[0] = (struct rte_flow_action){
2423 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
2424 			.conf = &attributes->queue,
2425 		};
2426 		break;
2427 	case RTE_ETH_FDIR_REJECT:
2428 		attributes->actions[0] = (struct rte_flow_action){
2429 			.type = RTE_FLOW_ACTION_TYPE_DROP,
2430 		};
2431 		break;
2432 	default:
2433 		DRV_LOG(ERR, "port %u invalid behavior %d",
2434 			dev->data->port_id,
2435 			fdir_filter->action.behavior);
2436 		rte_errno = ENOTSUP;
2437 		return -rte_errno;
2438 	}
2439 	attributes->queue.index = fdir_filter->action.rx_queue;
2440 	/* Handle L3. */
2441 	switch (fdir_filter->input.flow_type) {
2442 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2443 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2444 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2445 		attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2446 			.src_addr = input->flow.ip4_flow.src_ip,
2447 			.dst_addr = input->flow.ip4_flow.dst_ip,
2448 			.time_to_live = input->flow.ip4_flow.ttl,
2449 			.type_of_service = input->flow.ip4_flow.tos,
2450 		};
2451 		attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2452 			.src_addr = mask->ipv4_mask.src_ip,
2453 			.dst_addr = mask->ipv4_mask.dst_ip,
2454 			.time_to_live = mask->ipv4_mask.ttl,
2455 			.type_of_service = mask->ipv4_mask.tos,
2456 			.next_proto_id = mask->ipv4_mask.proto,
2457 		};
2458 		attributes->items[1] = (struct rte_flow_item){
2459 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
2460 			.spec = &attributes->l3,
2461 			.mask = &attributes->l3_mask,
2462 		};
2463 		break;
2464 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2465 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2466 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2467 		attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2468 			.hop_limits = input->flow.ipv6_flow.hop_limits,
2469 			.proto = input->flow.ipv6_flow.proto,
2470 		};
2471 
2472 		memcpy(attributes->l3.ipv6.hdr.src_addr,
2473 		       input->flow.ipv6_flow.src_ip,
2474 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2475 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
2476 		       input->flow.ipv6_flow.dst_ip,
2477 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2478 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2479 		       mask->ipv6_mask.src_ip,
2480 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2481 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2482 		       mask->ipv6_mask.dst_ip,
2483 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2484 		attributes->items[1] = (struct rte_flow_item){
2485 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
2486 			.spec = &attributes->l3,
2487 			.mask = &attributes->l3_mask,
2488 		};
2489 		break;
2490 	default:
2491 		DRV_LOG(ERR, "port %u invalid flow type%d",
2492 			dev->data->port_id, fdir_filter->input.flow_type);
2493 		rte_errno = ENOTSUP;
2494 		return -rte_errno;
2495 	}
2496 	/* Handle L4. */
2497 	switch (fdir_filter->input.flow_type) {
2498 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2499 		attributes->l4.udp.hdr = (struct udp_hdr){
2500 			.src_port = input->flow.udp4_flow.src_port,
2501 			.dst_port = input->flow.udp4_flow.dst_port,
2502 		};
2503 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2504 			.src_port = mask->src_port_mask,
2505 			.dst_port = mask->dst_port_mask,
2506 		};
2507 		attributes->items[2] = (struct rte_flow_item){
2508 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2509 			.spec = &attributes->l4,
2510 			.mask = &attributes->l4_mask,
2511 		};
2512 		break;
2513 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2514 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2515 			.src_port = input->flow.tcp4_flow.src_port,
2516 			.dst_port = input->flow.tcp4_flow.dst_port,
2517 		};
2518 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2519 			.src_port = mask->src_port_mask,
2520 			.dst_port = mask->dst_port_mask,
2521 		};
2522 		attributes->items[2] = (struct rte_flow_item){
2523 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2524 			.spec = &attributes->l4,
2525 			.mask = &attributes->l4_mask,
2526 		};
2527 		break;
2528 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2529 		attributes->l4.udp.hdr = (struct udp_hdr){
2530 			.src_port = input->flow.udp6_flow.src_port,
2531 			.dst_port = input->flow.udp6_flow.dst_port,
2532 		};
2533 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2534 			.src_port = mask->src_port_mask,
2535 			.dst_port = mask->dst_port_mask,
2536 		};
2537 		attributes->items[2] = (struct rte_flow_item){
2538 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2539 			.spec = &attributes->l4,
2540 			.mask = &attributes->l4_mask,
2541 		};
2542 		break;
2543 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2544 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2545 			.src_port = input->flow.tcp6_flow.src_port,
2546 			.dst_port = input->flow.tcp6_flow.dst_port,
2547 		};
2548 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2549 			.src_port = mask->src_port_mask,
2550 			.dst_port = mask->dst_port_mask,
2551 		};
2552 		attributes->items[2] = (struct rte_flow_item){
2553 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2554 			.spec = &attributes->l4,
2555 			.mask = &attributes->l4_mask,
2556 		};
2557 		break;
2558 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2559 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2560 		break;
2561 	default:
2562 		DRV_LOG(ERR, "port %u invalid flow type%d",
2563 			dev->data->port_id, fdir_filter->input.flow_type);
2564 		rte_errno = ENOTSUP;
2565 		return -rte_errno;
2566 	}
2567 	return 0;
2568 }
2569 
2570 /**
2571  * Add new flow director filter and store it in list.
2572  *
2573  * @param dev
2574  *   Pointer to Ethernet device.
2575  * @param fdir_filter
2576  *   Flow director filter to add.
2577  *
2578  * @return
2579  *   0 on success, a negative errno value otherwise and rte_errno is set.
2580  */
2581 static int
2582 mlx5_fdir_filter_add(struct rte_eth_dev *dev,
2583 		     const struct rte_eth_fdir_filter *fdir_filter)
2584 {
2585 	struct priv *priv = dev->data->dev_private;
2586 	struct mlx5_fdir attributes = {
2587 		.attr.group = 0,
2588 		.l2_mask = {
2589 			.dst.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2590 			.src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2591 			.type = 0,
2592 		},
2593 	};
2594 	struct rte_flow_error error;
2595 	struct rte_flow *flow;
2596 	int ret;
2597 
2598 	ret = mlx5_fdir_filter_convert(dev, fdir_filter, &attributes);
2599 	if (ret)
2600 		return ret;
2601 	flow = mlx5_flow_list_create(dev, &priv->flows, &attributes.attr,
2602 				     attributes.items, attributes.actions,
2603 				     &error);
2604 	if (flow) {
2605 		DRV_LOG(DEBUG, "port %u FDIR created %p", dev->data->port_id,
2606 			(void *)flow);
2607 		return 0;
2608 	}
2609 	return -rte_errno;
2610 }
2611 
2612 /**
2613  * Delete specific filter.
2614  *
2615  * @param dev
2616  *   Pointer to Ethernet device.
2617  * @param fdir_filter
2618  *   Filter to be deleted.
2619  *
2620  * @return
2621  *   0 on success, a negative errno value otherwise and rte_errno is set.
2622  */
2623 static int
2624 mlx5_fdir_filter_delete(struct rte_eth_dev *dev __rte_unused,
2625 			const struct rte_eth_fdir_filter *fdir_filter
2626 			__rte_unused)
2627 {
2628 	rte_errno = ENOTSUP;
2629 	return -rte_errno;
2630 }
2631 
2632 /**
2633  * Update queue for specific filter.
2634  *
2635  * @param dev
2636  *   Pointer to Ethernet device.
2637  * @param fdir_filter
2638  *   Filter to be updated.
2639  *
2640  * @return
2641  *   0 on success, a negative errno value otherwise and rte_errno is set.
2642  */
2643 static int
2644 mlx5_fdir_filter_update(struct rte_eth_dev *dev,
2645 			const struct rte_eth_fdir_filter *fdir_filter)
2646 {
2647 	int ret;
2648 
2649 	ret = mlx5_fdir_filter_delete(dev, fdir_filter);
2650 	if (ret)
2651 		return ret;
2652 	return mlx5_fdir_filter_add(dev, fdir_filter);
2653 }
2654 
2655 /**
2656  * Flush all filters.
2657  *
2658  * @param dev
2659  *   Pointer to Ethernet device.
2660  */
2661 static void
2662 mlx5_fdir_filter_flush(struct rte_eth_dev *dev)
2663 {
2664 	struct priv *priv = dev->data->dev_private;
2665 
2666 	mlx5_flow_list_flush(dev, &priv->flows);
2667 }
2668 
2669 /**
2670  * Get flow director information.
2671  *
2672  * @param dev
2673  *   Pointer to Ethernet device.
2674  * @param[out] fdir_info
2675  *   Resulting flow director information.
2676  */
2677 static void
2678 mlx5_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2679 {
2680 	struct rte_eth_fdir_masks *mask =
2681 		&dev->data->dev_conf.fdir_conf.mask;
2682 
2683 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2684 	fdir_info->guarant_spc = 0;
2685 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2686 	fdir_info->max_flexpayload = 0;
2687 	fdir_info->flow_types_mask[0] = 0;
2688 	fdir_info->flex_payload_unit = 0;
2689 	fdir_info->max_flex_payload_segment_num = 0;
2690 	fdir_info->flex_payload_limit = 0;
2691 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2692 }
2693 
2694 /**
2695  * Deal with flow director operations.
2696  *
2697  * @param dev
2698  *   Pointer to Ethernet device.
2699  * @param filter_op
2700  *   Operation to perform.
2701  * @param arg
2702  *   Pointer to operation-specific structure.
2703  *
2704  * @return
2705  *   0 on success, a negative errno value otherwise and rte_errno is set.
2706  */
2707 static int
2708 mlx5_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2709 		    void *arg)
2710 {
2711 	enum rte_fdir_mode fdir_mode =
2712 		dev->data->dev_conf.fdir_conf.mode;
2713 
2714 	if (filter_op == RTE_ETH_FILTER_NOP)
2715 		return 0;
2716 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2717 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2718 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
2719 			dev->data->port_id, fdir_mode);
2720 		rte_errno = EINVAL;
2721 		return -rte_errno;
2722 	}
2723 	switch (filter_op) {
2724 	case RTE_ETH_FILTER_ADD:
2725 		return mlx5_fdir_filter_add(dev, arg);
2726 	case RTE_ETH_FILTER_UPDATE:
2727 		return mlx5_fdir_filter_update(dev, arg);
2728 	case RTE_ETH_FILTER_DELETE:
2729 		return mlx5_fdir_filter_delete(dev, arg);
2730 	case RTE_ETH_FILTER_FLUSH:
2731 		mlx5_fdir_filter_flush(dev);
2732 		break;
2733 	case RTE_ETH_FILTER_INFO:
2734 		mlx5_fdir_info_get(dev, arg);
2735 		break;
2736 	default:
2737 		DRV_LOG(DEBUG, "port %u unknown operation %u",
2738 			dev->data->port_id, filter_op);
2739 		rte_errno = EINVAL;
2740 		return -rte_errno;
2741 	}
2742 	return 0;
2743 }
2744 
2745 /**
2746  * Manage filter operations.
2747  *
2748  * @param dev
2749  *   Pointer to Ethernet device structure.
2750  * @param filter_type
2751  *   Filter type.
2752  * @param filter_op
2753  *   Operation to perform.
2754  * @param arg
2755  *   Pointer to operation-specific structure.
2756  *
2757  * @return
2758  *   0 on success, a negative errno value otherwise and rte_errno is set.
2759  */
2760 int
2761 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2762 		     enum rte_filter_type filter_type,
2763 		     enum rte_filter_op filter_op,
2764 		     void *arg)
2765 {
2766 	switch (filter_type) {
2767 	case RTE_ETH_FILTER_GENERIC:
2768 		if (filter_op != RTE_ETH_FILTER_GET) {
2769 			rte_errno = EINVAL;
2770 			return -rte_errno;
2771 		}
2772 		*(const void **)arg = &mlx5_flow_ops;
2773 		return 0;
2774 	case RTE_ETH_FILTER_FDIR:
2775 		return mlx5_fdir_ctrl_func(dev, filter_op, arg);
2776 	default:
2777 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
2778 			dev->data->port_id, filter_type);
2779 		rte_errno = ENOTSUP;
2780 		return -rte_errno;
2781 	}
2782 	return 0;
2783 }
2784