xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 1edccebcccdbe600dc0a3a418fae68336648a87e)
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, 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, 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[in] attr
649  *   Attributes of flow that includes this action.
650  * @param[out] error
651  *   Pointer to error structure.
652  *
653  * @return
654  *   0 on success, a negative errno value otherwise and rte_errno is set.
655  */
656 int
657 mlx5_flow_validate_action_flag(uint64_t action_flags,
658 			       const struct rte_flow_attr *attr,
659 			       struct rte_flow_error *error)
660 {
661 
662 	if (action_flags & MLX5_FLOW_ACTION_DROP)
663 		return rte_flow_error_set(error, EINVAL,
664 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
665 					  "can't drop and flag in same flow");
666 	if (action_flags & MLX5_FLOW_ACTION_MARK)
667 		return rte_flow_error_set(error, EINVAL,
668 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
669 					  "can't mark and flag in same flow");
670 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
671 		return rte_flow_error_set(error, EINVAL,
672 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
673 					  "can't have 2 flag"
674 					  " actions in same flow");
675 	if (attr->egress)
676 		return rte_flow_error_set(error, ENOTSUP,
677 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
678 					  "flag action not supported for "
679 					  "egress");
680 	return 0;
681 }
682 
683 /*
684  * Validate the mark action.
685  *
686  * @param[in] action
687  *   Pointer to the queue action.
688  * @param[in] action_flags
689  *   Bit-fields that holds the actions detected until now.
690  * @param[in] attr
691  *   Attributes of flow that includes this action.
692  * @param[out] error
693  *   Pointer to error structure.
694  *
695  * @return
696  *   0 on success, a negative errno value otherwise and rte_errno is set.
697  */
698 int
699 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
700 			       uint64_t action_flags,
701 			       const struct rte_flow_attr *attr,
702 			       struct rte_flow_error *error)
703 {
704 	const struct rte_flow_action_mark *mark = action->conf;
705 
706 	if (!mark)
707 		return rte_flow_error_set(error, EINVAL,
708 					  RTE_FLOW_ERROR_TYPE_ACTION,
709 					  action,
710 					  "configuration cannot be null");
711 	if (mark->id >= MLX5_FLOW_MARK_MAX)
712 		return rte_flow_error_set(error, EINVAL,
713 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
714 					  &mark->id,
715 					  "mark id must in 0 <= id < "
716 					  RTE_STR(MLX5_FLOW_MARK_MAX));
717 	if (action_flags & MLX5_FLOW_ACTION_DROP)
718 		return rte_flow_error_set(error, EINVAL,
719 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
720 					  "can't drop and mark in same flow");
721 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
722 		return rte_flow_error_set(error, EINVAL,
723 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
724 					  "can't flag and mark in same flow");
725 	if (action_flags & MLX5_FLOW_ACTION_MARK)
726 		return rte_flow_error_set(error, EINVAL,
727 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
728 					  "can't have 2 mark actions in same"
729 					  " flow");
730 	if (attr->egress)
731 		return rte_flow_error_set(error, ENOTSUP,
732 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
733 					  "mark action not supported for "
734 					  "egress");
735 	return 0;
736 }
737 
738 /*
739  * Validate the drop action.
740  *
741  * @param[in] action_flags
742  *   Bit-fields that holds the actions detected until now.
743  * @param[in] attr
744  *   Attributes of flow that includes this action.
745  * @param[out] error
746  *   Pointer to error structure.
747  *
748  * @return
749  *   0 on success, a negative errno value otherwise and rte_ernno is set.
750  */
751 int
752 mlx5_flow_validate_action_drop(uint64_t action_flags,
753 			       const struct rte_flow_attr *attr,
754 			       struct rte_flow_error *error)
755 {
756 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
757 		return rte_flow_error_set(error, EINVAL,
758 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
759 					  "can't drop and flag in same flow");
760 	if (action_flags & MLX5_FLOW_ACTION_MARK)
761 		return rte_flow_error_set(error, EINVAL,
762 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
763 					  "can't drop and mark in same flow");
764 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
765 		return rte_flow_error_set(error, EINVAL,
766 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
767 					  "can't have 2 fate actions in"
768 					  " same flow");
769 	if (attr->egress)
770 		return rte_flow_error_set(error, ENOTSUP,
771 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
772 					  "drop action not supported for "
773 					  "egress");
774 	return 0;
775 }
776 
777 /*
778  * Validate the queue action.
779  *
780  * @param[in] action
781  *   Pointer to the queue action.
782  * @param[in] action_flags
783  *   Bit-fields that holds the actions detected until now.
784  * @param[in] dev
785  *   Pointer to the Ethernet device structure.
786  * @param[in] attr
787  *   Attributes of flow that includes this action.
788  * @param[out] error
789  *   Pointer to error structure.
790  *
791  * @return
792  *   0 on success, a negative errno value otherwise and rte_ernno is set.
793  */
794 int
795 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
796 				uint64_t action_flags,
797 				struct rte_eth_dev *dev,
798 				const struct rte_flow_attr *attr,
799 				struct rte_flow_error *error)
800 {
801 	struct priv *priv = dev->data->dev_private;
802 	const struct rte_flow_action_queue *queue = action->conf;
803 
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 (queue->index >= priv->rxqs_n)
810 		return rte_flow_error_set(error, EINVAL,
811 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
812 					  &queue->index,
813 					  "queue index out of range");
814 	if (!(*priv->rxqs)[queue->index])
815 		return rte_flow_error_set(error, EINVAL,
816 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
817 					  &queue->index,
818 					  "queue is not configured");
819 	if (attr->egress)
820 		return rte_flow_error_set(error, ENOTSUP,
821 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
822 					  "queue action not supported for "
823 					  "egress");
824 	return 0;
825 }
826 
827 /*
828  * Validate the rss action.
829  *
830  * @param[in] action
831  *   Pointer to the queue action.
832  * @param[in] action_flags
833  *   Bit-fields that holds the actions detected until now.
834  * @param[in] dev
835  *   Pointer to the Ethernet device structure.
836  * @param[in] attr
837  *   Attributes of flow that includes this action.
838  * @param[out] error
839  *   Pointer to error structure.
840  *
841  * @return
842  *   0 on success, a negative errno value otherwise and rte_ernno is set.
843  */
844 int
845 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
846 			      uint64_t action_flags,
847 			      struct rte_eth_dev *dev,
848 			      const struct rte_flow_attr *attr,
849 			      struct rte_flow_error *error)
850 {
851 	struct priv *priv = dev->data->dev_private;
852 	const struct rte_flow_action_rss *rss = action->conf;
853 	unsigned int i;
854 
855 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
856 		return rte_flow_error_set(error, EINVAL,
857 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
858 					  "can't have 2 fate actions"
859 					  " in same flow");
860 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
861 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
862 		return rte_flow_error_set(error, ENOTSUP,
863 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
864 					  &rss->func,
865 					  "RSS hash function not supported");
866 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
867 	if (rss->level > 2)
868 #else
869 	if (rss->level > 1)
870 #endif
871 		return rte_flow_error_set(error, ENOTSUP,
872 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
873 					  &rss->level,
874 					  "tunnel RSS is not supported");
875 	if (rss->key_len < MLX5_RSS_HASH_KEY_LEN)
876 		return rte_flow_error_set(error, ENOTSUP,
877 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
878 					  &rss->key_len,
879 					  "RSS hash key too small");
880 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
881 		return rte_flow_error_set(error, ENOTSUP,
882 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
883 					  &rss->key_len,
884 					  "RSS hash key too large");
885 	if (rss->queue_num > priv->config.ind_table_max_size)
886 		return rte_flow_error_set(error, ENOTSUP,
887 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
888 					  &rss->queue_num,
889 					  "number of queues too large");
890 	if (rss->types & MLX5_RSS_HF_MASK)
891 		return rte_flow_error_set(error, ENOTSUP,
892 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
893 					  &rss->types,
894 					  "some RSS protocols are not"
895 					  " supported");
896 	for (i = 0; i != rss->queue_num; ++i) {
897 		if (!(*priv->rxqs)[rss->queue[i]])
898 			return rte_flow_error_set
899 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
900 				 &rss->queue[i], "queue is not configured");
901 	}
902 	if (attr->egress)
903 		return rte_flow_error_set(error, ENOTSUP,
904 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
905 					  "rss action not supported for "
906 					  "egress");
907 	return 0;
908 }
909 
910 /*
911  * Validate the count action.
912  *
913  * @param[in] dev
914  *   Pointer to the Ethernet device structure.
915  * @param[in] attr
916  *   Attributes of flow that includes this action.
917  * @param[out] error
918  *   Pointer to error structure.
919  *
920  * @return
921  *   0 on success, a negative errno value otherwise and rte_ernno is set.
922  */
923 int
924 mlx5_flow_validate_action_count(struct rte_eth_dev *dev,
925 				const struct rte_flow_attr *attr,
926 				struct rte_flow_error *error)
927 {
928 	struct priv *priv = dev->data->dev_private;
929 
930 	if (!priv->config.flow_counter_en)
931 		return rte_flow_error_set(error, ENOTSUP,
932 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
933 					  "flow counters are not supported.");
934 	if (attr->egress)
935 		return rte_flow_error_set(error, ENOTSUP,
936 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
937 					  "count action not supported for "
938 					  "egress");
939 	return 0;
940 }
941 
942 /**
943  * Verify the @p attributes will be correctly understood by the NIC and store
944  * them in the @p flow if everything is correct.
945  *
946  * @param[in] dev
947  *   Pointer to the Ethernet device structure.
948  * @param[in] attributes
949  *   Pointer to flow attributes
950  * @param[out] error
951  *   Pointer to error structure.
952  *
953  * @return
954  *   0 on success, a negative errno value otherwise and rte_errno is set.
955  */
956 int
957 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
958 			      const struct rte_flow_attr *attributes,
959 			      struct rte_flow_error *error)
960 {
961 	struct priv *priv = dev->data->dev_private;
962 	uint32_t priority_max = priv->config.flow_prio - 1;
963 
964 	if (attributes->group)
965 		return rte_flow_error_set(error, ENOTSUP,
966 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
967 					  NULL, "groups is not supported");
968 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
969 	    attributes->priority >= priority_max)
970 		return rte_flow_error_set(error, ENOTSUP,
971 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
972 					  NULL, "priority out of range");
973 	if (attributes->egress)
974 		return rte_flow_error_set(error, ENOTSUP,
975 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
976 					  "egress is not supported");
977 	if (attributes->transfer)
978 		return rte_flow_error_set(error, ENOTSUP,
979 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
980 					  NULL, "transfer is not supported");
981 	if (!attributes->ingress)
982 		return rte_flow_error_set(error, EINVAL,
983 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
984 					  NULL,
985 					  "ingress attribute is mandatory");
986 	return 0;
987 }
988 
989 /**
990  * Validate Ethernet item.
991  *
992  * @param[in] item
993  *   Item specification.
994  * @param[in] item_flags
995  *   Bit-fields that holds the items detected until now.
996  * @param[out] error
997  *   Pointer to error structure.
998  *
999  * @return
1000  *   0 on success, a negative errno value otherwise and rte_errno is set.
1001  */
1002 int
1003 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1004 			    uint64_t item_flags,
1005 			    struct rte_flow_error *error)
1006 {
1007 	const struct rte_flow_item_eth *mask = item->mask;
1008 	const struct rte_flow_item_eth nic_mask = {
1009 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1010 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1011 		.type = RTE_BE16(0xffff),
1012 	};
1013 	int ret;
1014 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1015 
1016 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L2)
1017 		return rte_flow_error_set(error, ENOTSUP,
1018 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1019 					  "3 levels of l2 are not supported");
1020 	if ((item_flags & MLX5_FLOW_LAYER_INNER_L2) && !tunnel)
1021 		return rte_flow_error_set(error, ENOTSUP,
1022 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1023 					  "2 L2 without tunnel are not supported");
1024 	if (!mask)
1025 		mask = &rte_flow_item_eth_mask;
1026 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1027 					(const uint8_t *)&nic_mask,
1028 					sizeof(struct rte_flow_item_eth),
1029 					error);
1030 	return ret;
1031 }
1032 
1033 /**
1034  * Validate VLAN item.
1035  *
1036  * @param[in] item
1037  *   Item specification.
1038  * @param[in] item_flags
1039  *   Bit-fields that holds the items detected until now.
1040  * @param[out] error
1041  *   Pointer to error structure.
1042  *
1043  * @return
1044  *   0 on success, a negative errno value otherwise and rte_errno is set.
1045  */
1046 int
1047 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1048 			     int64_t item_flags,
1049 			     struct rte_flow_error *error)
1050 {
1051 	const struct rte_flow_item_vlan *spec = item->spec;
1052 	const struct rte_flow_item_vlan *mask = item->mask;
1053 	const struct rte_flow_item_vlan nic_mask = {
1054 		.tci = RTE_BE16(0x0fff),
1055 		.inner_type = RTE_BE16(0xffff),
1056 	};
1057 	uint16_t vlan_tag = 0;
1058 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1059 	int ret;
1060 	const uint32_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1061 					MLX5_FLOW_LAYER_INNER_L4) :
1062 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1063 					MLX5_FLOW_LAYER_OUTER_L4);
1064 	const uint32_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1065 					MLX5_FLOW_LAYER_OUTER_VLAN;
1066 
1067 	if (item_flags & vlanm)
1068 		return rte_flow_error_set(error, EINVAL,
1069 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1070 					  "VLAN layer already configured");
1071 	else if ((item_flags & l34m) != 0)
1072 		return rte_flow_error_set(error, EINVAL,
1073 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1074 					  "L2 layer cannot follow L3/L4 layer");
1075 	if (!mask)
1076 		mask = &rte_flow_item_vlan_mask;
1077 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1078 					(const uint8_t *)&nic_mask,
1079 					sizeof(struct rte_flow_item_vlan),
1080 					error);
1081 	if (ret)
1082 		return ret;
1083 	if (spec) {
1084 		vlan_tag = spec->tci;
1085 		vlan_tag &= mask->tci;
1086 	}
1087 	/*
1088 	 * From verbs perspective an empty VLAN is equivalent
1089 	 * to a packet without VLAN layer.
1090 	 */
1091 	if (!vlan_tag)
1092 		return rte_flow_error_set(error, EINVAL,
1093 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1094 					  item->spec,
1095 					  "VLAN cannot be empty");
1096 	return 0;
1097 }
1098 
1099 /**
1100  * Validate IPV4 item.
1101  *
1102  * @param[in] item
1103  *   Item specification.
1104  * @param[in] item_flags
1105  *   Bit-fields that holds the items detected until now.
1106  * @param[out] error
1107  *   Pointer to error structure.
1108  *
1109  * @return
1110  *   0 on success, a negative errno value otherwise and rte_errno is set.
1111  */
1112 int
1113 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1114 			     int64_t item_flags,
1115 			     struct rte_flow_error *error)
1116 {
1117 	const struct rte_flow_item_ipv4 *mask = item->mask;
1118 	const struct rte_flow_item_ipv4 nic_mask = {
1119 		.hdr = {
1120 			.src_addr = RTE_BE32(0xffffffff),
1121 			.dst_addr = RTE_BE32(0xffffffff),
1122 			.type_of_service = 0xff,
1123 			.next_proto_id = 0xff,
1124 		},
1125 	};
1126 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1127 	int ret;
1128 
1129 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1130 				   MLX5_FLOW_LAYER_OUTER_L3))
1131 		return rte_flow_error_set(error, ENOTSUP,
1132 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1133 					  "multiple L3 layers not supported");
1134 	else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1135 					MLX5_FLOW_LAYER_OUTER_L4))
1136 		return rte_flow_error_set(error, EINVAL,
1137 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1138 					  "L3 cannot follow an L4 layer.");
1139 	if (!mask)
1140 		mask = &rte_flow_item_ipv4_mask;
1141 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1142 					(const uint8_t *)&nic_mask,
1143 					sizeof(struct rte_flow_item_ipv4),
1144 					error);
1145 	if (ret < 0)
1146 		return ret;
1147 	return 0;
1148 }
1149 
1150 /**
1151  * Validate IPV6 item.
1152  *
1153  * @param[in] item
1154  *   Item specification.
1155  * @param[in] item_flags
1156  *   Bit-fields that holds the items detected until now.
1157  * @param[out] error
1158  *   Pointer to error structure.
1159  *
1160  * @return
1161  *   0 on success, a negative errno value otherwise and rte_errno is set.
1162  */
1163 int
1164 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1165 			     uint64_t item_flags,
1166 			     struct rte_flow_error *error)
1167 {
1168 	const struct rte_flow_item_ipv6 *mask = item->mask;
1169 	const struct rte_flow_item_ipv6 nic_mask = {
1170 		.hdr = {
1171 			.src_addr =
1172 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1173 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1174 			.dst_addr =
1175 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1176 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1177 			.vtc_flow = RTE_BE32(0xffffffff),
1178 			.proto = 0xff,
1179 			.hop_limits = 0xff,
1180 		},
1181 	};
1182 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1183 	int ret;
1184 
1185 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1186 				   MLX5_FLOW_LAYER_OUTER_L3))
1187 		return rte_flow_error_set(error, ENOTSUP,
1188 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1189 					  "multiple L3 layers not supported");
1190 	else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1191 					MLX5_FLOW_LAYER_OUTER_L4))
1192 		return rte_flow_error_set(error, EINVAL,
1193 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1194 					  "L3 cannot follow an L4 layer.");
1195 	/*
1196 	 * IPv6 is not recognised by the NIC inside a GRE tunnel.
1197 	 * Such support has to be disabled as the rule will be
1198 	 * accepted.  Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1199 	 * Mellanox OFED 4.4-1.0.0.0.
1200 	 */
1201 	if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1202 		return rte_flow_error_set(error, ENOTSUP,
1203 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1204 					  "IPv6 inside a GRE tunnel is"
1205 					  " not recognised.");
1206 	if (!mask)
1207 		mask = &rte_flow_item_ipv6_mask;
1208 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1209 					(const uint8_t *)&nic_mask,
1210 					sizeof(struct rte_flow_item_ipv6),
1211 					error);
1212 	if (ret < 0)
1213 		return ret;
1214 	return 0;
1215 }
1216 
1217 /**
1218  * Validate UDP item.
1219  *
1220  * @param[in] item
1221  *   Item specification.
1222  * @param[in] item_flags
1223  *   Bit-fields that holds the items detected until now.
1224  * @param[in] target_protocol
1225  *   The next protocol in the previous item.
1226  * @param[in] flow_mask
1227  *   mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1228  * @param[out] error
1229  *   Pointer to error structure.
1230  *
1231  * @return
1232  *   0 on success, a negative errno value otherwise and rte_errno is set.
1233  */
1234 int
1235 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1236 			    uint64_t item_flags,
1237 			    uint8_t target_protocol,
1238 			    struct rte_flow_error *error)
1239 {
1240 	const struct rte_flow_item_udp *mask = item->mask;
1241 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1242 	int ret;
1243 
1244 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1245 		return rte_flow_error_set(error, EINVAL,
1246 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1247 					  "protocol filtering not compatible"
1248 					  " with UDP layer");
1249 	if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1250 				     MLX5_FLOW_LAYER_OUTER_L3)))
1251 		return rte_flow_error_set(error, EINVAL,
1252 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1253 					  "L3 is mandatory to filter on L4");
1254 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1255 				   MLX5_FLOW_LAYER_OUTER_L4))
1256 		return rte_flow_error_set(error, EINVAL,
1257 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1258 					  "L4 layer is already present");
1259 	if (!mask)
1260 		mask = &rte_flow_item_udp_mask;
1261 	ret = mlx5_flow_item_acceptable
1262 		(item, (const uint8_t *)mask,
1263 		 (const uint8_t *)&rte_flow_item_udp_mask,
1264 		 sizeof(struct rte_flow_item_udp), error);
1265 	if (ret < 0)
1266 		return ret;
1267 	return 0;
1268 }
1269 
1270 /**
1271  * Validate TCP item.
1272  *
1273  * @param[in] item
1274  *   Item specification.
1275  * @param[in] item_flags
1276  *   Bit-fields that holds the items detected until now.
1277  * @param[in] target_protocol
1278  *   The next protocol in the previous item.
1279  * @param[out] error
1280  *   Pointer to error structure.
1281  *
1282  * @return
1283  *   0 on success, a negative errno value otherwise and rte_errno is set.
1284  */
1285 int
1286 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1287 			    uint64_t item_flags,
1288 			    uint8_t target_protocol,
1289 			    const struct rte_flow_item_tcp *flow_mask,
1290 			    struct rte_flow_error *error)
1291 {
1292 	const struct rte_flow_item_tcp *mask = item->mask;
1293 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1294 	int ret;
1295 
1296 	assert(flow_mask);
1297 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1298 		return rte_flow_error_set(error, EINVAL,
1299 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1300 					  "protocol filtering not compatible"
1301 					  " with TCP layer");
1302 	if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1303 				     MLX5_FLOW_LAYER_OUTER_L3)))
1304 		return rte_flow_error_set(error, EINVAL,
1305 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1306 					  "L3 is mandatory to filter on L4");
1307 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1308 				   MLX5_FLOW_LAYER_OUTER_L4))
1309 		return rte_flow_error_set(error, EINVAL,
1310 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1311 					  "L4 layer is already present");
1312 	if (!mask)
1313 		mask = &rte_flow_item_tcp_mask;
1314 	ret = mlx5_flow_item_acceptable
1315 		(item, (const uint8_t *)mask,
1316 		 (const uint8_t *)flow_mask,
1317 		 sizeof(struct rte_flow_item_tcp), error);
1318 	if (ret < 0)
1319 		return ret;
1320 	return 0;
1321 }
1322 
1323 /**
1324  * Validate VXLAN item.
1325  *
1326  * @param[in] item
1327  *   Item specification.
1328  * @param[in] item_flags
1329  *   Bit-fields that holds the items detected until now.
1330  * @param[in] target_protocol
1331  *   The next protocol in the previous item.
1332  * @param[out] error
1333  *   Pointer to error structure.
1334  *
1335  * @return
1336  *   0 on success, a negative errno value otherwise and rte_errno is set.
1337  */
1338 int
1339 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1340 			      uint64_t item_flags,
1341 			      struct rte_flow_error *error)
1342 {
1343 	const struct rte_flow_item_vxlan *spec = item->spec;
1344 	const struct rte_flow_item_vxlan *mask = item->mask;
1345 	int ret;
1346 	union vni {
1347 		uint32_t vlan_id;
1348 		uint8_t vni[4];
1349 	} id = { .vlan_id = 0, };
1350 	uint32_t vlan_id = 0;
1351 
1352 
1353 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1354 		return rte_flow_error_set(error, ENOTSUP,
1355 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1356 					  "a tunnel is already present");
1357 	/*
1358 	 * Verify only UDPv4 is present as defined in
1359 	 * https://tools.ietf.org/html/rfc7348
1360 	 */
1361 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1362 		return rte_flow_error_set(error, EINVAL,
1363 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1364 					  "no outer UDP layer found");
1365 	if (!mask)
1366 		mask = &rte_flow_item_vxlan_mask;
1367 	ret = mlx5_flow_item_acceptable
1368 		(item, (const uint8_t *)mask,
1369 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1370 		 sizeof(struct rte_flow_item_vxlan),
1371 		 error);
1372 	if (ret < 0)
1373 		return ret;
1374 	if (spec) {
1375 		memcpy(&id.vni[1], spec->vni, 3);
1376 		vlan_id = id.vlan_id;
1377 		memcpy(&id.vni[1], mask->vni, 3);
1378 		vlan_id &= id.vlan_id;
1379 	}
1380 	/*
1381 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1382 	 * only this layer is defined in the Verbs specification it is
1383 	 * interpreted as wildcard and all packets will match this
1384 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1385 	 * udp), all packets matching the layers before will also
1386 	 * match this rule.  To avoid such situation, VNI 0 is
1387 	 * currently refused.
1388 	 */
1389 	if (!vlan_id)
1390 		return rte_flow_error_set(error, ENOTSUP,
1391 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1392 					  "VXLAN vni cannot be 0");
1393 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1394 		return rte_flow_error_set(error, ENOTSUP,
1395 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1396 					  "VXLAN tunnel must be fully defined");
1397 	return 0;
1398 }
1399 
1400 /**
1401  * Validate VXLAN_GPE item.
1402  *
1403  * @param[in] item
1404  *   Item specification.
1405  * @param[in] item_flags
1406  *   Bit-fields that holds the items detected until now.
1407  * @param[in] priv
1408  *   Pointer to the private data structure.
1409  * @param[in] target_protocol
1410  *   The next protocol in the previous item.
1411  * @param[out] error
1412  *   Pointer to error structure.
1413  *
1414  * @return
1415  *   0 on success, a negative errno value otherwise and rte_errno is set.
1416  */
1417 int
1418 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1419 				  uint64_t item_flags,
1420 				  struct rte_eth_dev *dev,
1421 				  struct rte_flow_error *error)
1422 {
1423 	struct priv *priv = dev->data->dev_private;
1424 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1425 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1426 	int ret;
1427 	union vni {
1428 		uint32_t vlan_id;
1429 		uint8_t vni[4];
1430 	} id = { .vlan_id = 0, };
1431 	uint32_t vlan_id = 0;
1432 
1433 	if (!priv->config.l3_vxlan_en)
1434 		return rte_flow_error_set(error, ENOTSUP,
1435 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1436 					  "L3 VXLAN is not enabled by device"
1437 					  " parameter and/or not configured in"
1438 					  " firmware");
1439 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1440 		return rte_flow_error_set(error, ENOTSUP,
1441 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1442 					  "a tunnel is already present");
1443 	/*
1444 	 * Verify only UDPv4 is present as defined in
1445 	 * https://tools.ietf.org/html/rfc7348
1446 	 */
1447 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1448 		return rte_flow_error_set(error, EINVAL,
1449 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1450 					  "no outer UDP layer found");
1451 	if (!mask)
1452 		mask = &rte_flow_item_vxlan_gpe_mask;
1453 	ret = mlx5_flow_item_acceptable
1454 		(item, (const uint8_t *)mask,
1455 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1456 		 sizeof(struct rte_flow_item_vxlan_gpe),
1457 		 error);
1458 	if (ret < 0)
1459 		return ret;
1460 	if (spec) {
1461 		if (spec->protocol)
1462 			return rte_flow_error_set(error, ENOTSUP,
1463 						  RTE_FLOW_ERROR_TYPE_ITEM,
1464 						  item,
1465 						  "VxLAN-GPE protocol"
1466 						  " not supported");
1467 		memcpy(&id.vni[1], spec->vni, 3);
1468 		vlan_id = id.vlan_id;
1469 		memcpy(&id.vni[1], mask->vni, 3);
1470 		vlan_id &= id.vlan_id;
1471 	}
1472 	/*
1473 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1474 	 * layer is defined in the Verbs specification it is interpreted as
1475 	 * wildcard and all packets will match this rule, if it follows a full
1476 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1477 	 * before will also match this rule.  To avoid such situation, VNI 0
1478 	 * is currently refused.
1479 	 */
1480 	if (!vlan_id)
1481 		return rte_flow_error_set(error, ENOTSUP,
1482 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1483 					  "VXLAN-GPE vni cannot be 0");
1484 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1485 		return rte_flow_error_set(error, ENOTSUP,
1486 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1487 					  "VXLAN-GPE tunnel must be fully"
1488 					  " defined");
1489 	return 0;
1490 }
1491 
1492 /**
1493  * Validate GRE item.
1494  *
1495  * @param[in] item
1496  *   Item specification.
1497  * @param[in] item_flags
1498  *   Bit flags to mark detected items.
1499  * @param[in] target_protocol
1500  *   The next protocol in the previous item.
1501  * @param[out] error
1502  *   Pointer to error structure.
1503  *
1504  * @return
1505  *   0 on success, a negative errno value otherwise and rte_errno is set.
1506  */
1507 int
1508 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1509 			    uint64_t item_flags,
1510 			    uint8_t target_protocol,
1511 			    struct rte_flow_error *error)
1512 {
1513 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1514 	const struct rte_flow_item_gre *mask = item->mask;
1515 	int ret;
1516 
1517 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1518 		return rte_flow_error_set(error, EINVAL,
1519 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1520 					  "protocol filtering not compatible"
1521 					  " with this GRE layer");
1522 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1523 		return rte_flow_error_set(error, ENOTSUP,
1524 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1525 					  "a tunnel is already present");
1526 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1527 		return rte_flow_error_set(error, ENOTSUP,
1528 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1529 					  "L3 Layer is missing");
1530 	if (!mask)
1531 		mask = &rte_flow_item_gre_mask;
1532 	ret = mlx5_flow_item_acceptable
1533 		(item, (const uint8_t *)mask,
1534 		 (const uint8_t *)&rte_flow_item_gre_mask,
1535 		 sizeof(struct rte_flow_item_gre), error);
1536 	if (ret < 0)
1537 		return ret;
1538 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1539 	if (spec && (spec->protocol & mask->protocol))
1540 		return rte_flow_error_set(error, ENOTSUP,
1541 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1542 					  "without MPLS support the"
1543 					  " specification cannot be used for"
1544 					  " filtering");
1545 #endif
1546 	return 0;
1547 }
1548 
1549 /**
1550  * Validate MPLS item.
1551  *
1552  * @param[in] item
1553  *   Item specification.
1554  * @param[in] item_flags
1555  *   Bit-fields that holds the items detected until now.
1556  * @param[in] target_protocol
1557  *   The next protocol in the previous item.
1558  * @param[out] error
1559  *   Pointer to error structure.
1560  *
1561  * @return
1562  *   0 on success, a negative errno value otherwise and rte_errno is set.
1563  */
1564 int
1565 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1566 			     uint64_t item_flags __rte_unused,
1567 			     uint8_t target_protocol __rte_unused,
1568 			     struct rte_flow_error *error)
1569 {
1570 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1571 	const struct rte_flow_item_mpls *mask = item->mask;
1572 	int ret;
1573 
1574 	if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1575 		return rte_flow_error_set(error, EINVAL,
1576 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1577 					  "protocol filtering not compatible"
1578 					  " with MPLS layer");
1579 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1580 		return rte_flow_error_set(error, ENOTSUP,
1581 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1582 					  "a tunnel is already"
1583 					  " present");
1584 	if (!mask)
1585 		mask = &rte_flow_item_mpls_mask;
1586 	ret = mlx5_flow_item_acceptable
1587 		(item, (const uint8_t *)mask,
1588 		 (const uint8_t *)&rte_flow_item_mpls_mask,
1589 		 sizeof(struct rte_flow_item_mpls), error);
1590 	if (ret < 0)
1591 		return ret;
1592 	return 0;
1593 #endif
1594 	return rte_flow_error_set(error, ENOTSUP,
1595 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
1596 				  "MPLS is not supported by Verbs, please"
1597 				  " update.");
1598 }
1599 
1600 static int
1601 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1602 		   const struct rte_flow_attr *attr __rte_unused,
1603 		   const struct rte_flow_item items[] __rte_unused,
1604 		   const struct rte_flow_action actions[] __rte_unused,
1605 		   struct rte_flow_error *error __rte_unused)
1606 {
1607 	rte_errno = ENOTSUP;
1608 	return -rte_errno;
1609 }
1610 
1611 static struct mlx5_flow *
1612 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1613 		  const struct rte_flow_item items[] __rte_unused,
1614 		  const struct rte_flow_action actions[] __rte_unused,
1615 		  uint64_t *item_flags __rte_unused,
1616 		  uint64_t *action_flags __rte_unused,
1617 		  struct rte_flow_error *error __rte_unused)
1618 {
1619 	rte_errno = ENOTSUP;
1620 	return NULL;
1621 }
1622 
1623 static int
1624 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1625 		    struct mlx5_flow *dev_flow __rte_unused,
1626 		    const struct rte_flow_attr *attr __rte_unused,
1627 		    const struct rte_flow_item items[] __rte_unused,
1628 		    const struct rte_flow_action actions[] __rte_unused,
1629 		    struct rte_flow_error *error __rte_unused)
1630 {
1631 	rte_errno = ENOTSUP;
1632 	return -rte_errno;
1633 }
1634 
1635 static int
1636 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1637 		struct rte_flow *flow __rte_unused,
1638 		struct rte_flow_error *error __rte_unused)
1639 {
1640 	rte_errno = ENOTSUP;
1641 	return -rte_errno;
1642 }
1643 
1644 static void
1645 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1646 		 struct rte_flow *flow __rte_unused)
1647 {
1648 }
1649 
1650 static void
1651 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1652 		  struct rte_flow *flow __rte_unused)
1653 {
1654 }
1655 
1656 /* Void driver to protect from null pointer reference. */
1657 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1658 	.validate = flow_null_validate,
1659 	.prepare = flow_null_prepare,
1660 	.translate = flow_null_translate,
1661 	.apply = flow_null_apply,
1662 	.remove = flow_null_remove,
1663 	.destroy = flow_null_destroy,
1664 };
1665 
1666 /**
1667  * Select flow driver type according to flow attributes and device
1668  * configuration.
1669  *
1670  * @param[in] dev
1671  *   Pointer to the dev structure.
1672  * @param[in] attr
1673  *   Pointer to the flow attributes.
1674  *
1675  * @return
1676  *   flow driver type if supported, MLX5_FLOW_TYPE_MAX otherwise.
1677  */
1678 static enum mlx5_flow_drv_type
1679 flow_get_drv_type(struct rte_eth_dev *dev __rte_unused,
1680 		  const struct rte_flow_attr *attr)
1681 {
1682 	struct priv *priv __rte_unused = dev->data->dev_private;
1683 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1684 
1685 	if (attr->transfer) {
1686 		type = MLX5_FLOW_TYPE_TCF;
1687 	} else {
1688 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
1689 		type = priv->config.dv_flow_en ?  MLX5_FLOW_TYPE_DV :
1690 						  MLX5_FLOW_TYPE_VERBS;
1691 #else
1692 		type = MLX5_FLOW_TYPE_VERBS;
1693 #endif
1694 	}
1695 	return type;
1696 }
1697 
1698 #define flow_get_drv_ops(type) flow_drv_ops[type]
1699 
1700 /**
1701  * Flow driver validation API. This abstracts calling driver specific functions.
1702  * The type of flow driver is determined according to flow attributes.
1703  *
1704  * @param[in] dev
1705  *   Pointer to the dev structure.
1706  * @param[in] attr
1707  *   Pointer to the flow attributes.
1708  * @param[in] items
1709  *   Pointer to the list of items.
1710  * @param[in] actions
1711  *   Pointer to the list of actions.
1712  * @param[out] error
1713  *   Pointer to the error structure.
1714  *
1715  * @return
1716  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1717  */
1718 static inline int
1719 flow_drv_validate(struct rte_eth_dev *dev,
1720 		  const struct rte_flow_attr *attr,
1721 		  const struct rte_flow_item items[],
1722 		  const struct rte_flow_action actions[],
1723 		  struct rte_flow_error *error)
1724 {
1725 	const struct mlx5_flow_driver_ops *fops;
1726 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1727 
1728 	fops = flow_get_drv_ops(type);
1729 	return fops->validate(dev, attr, items, actions, error);
1730 }
1731 
1732 /**
1733  * Flow driver preparation API. This abstracts calling driver specific
1734  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1735  * calculates the size of memory required for device flow, allocates the memory,
1736  * initializes the device flow and returns the pointer.
1737  *
1738  * @param[in] attr
1739  *   Pointer to the flow attributes.
1740  * @param[in] items
1741  *   Pointer to the list of items.
1742  * @param[in] actions
1743  *   Pointer to the list of actions.
1744  * @param[out] item_flags
1745  *   Pointer to bit mask of all items detected.
1746  * @param[out] action_flags
1747  *   Pointer to bit mask of all actions detected.
1748  * @param[out] error
1749  *   Pointer to the error structure.
1750  *
1751  * @return
1752  *   Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1753  */
1754 static inline struct mlx5_flow *
1755 flow_drv_prepare(struct rte_flow *flow,
1756 		 const struct rte_flow_attr *attr,
1757 		 const struct rte_flow_item items[],
1758 		 const struct rte_flow_action actions[],
1759 		 uint64_t *item_flags,
1760 		 uint64_t *action_flags,
1761 		 struct rte_flow_error *error)
1762 {
1763 	const struct mlx5_flow_driver_ops *fops;
1764 	enum mlx5_flow_drv_type type = flow->drv_type;
1765 
1766 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1767 	fops = flow_get_drv_ops(type);
1768 	return fops->prepare(attr, items, actions, item_flags, action_flags,
1769 			     error);
1770 }
1771 
1772 /**
1773  * Flow driver translation API. This abstracts calling driver specific
1774  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1775  * translates a generic flow into a driver flow. flow_drv_prepare() must
1776  * precede.
1777  *
1778  *
1779  * @param[in] dev
1780  *   Pointer to the rte dev structure.
1781  * @param[in, out] dev_flow
1782  *   Pointer to the mlx5 flow.
1783  * @param[in] attr
1784  *   Pointer to the flow attributes.
1785  * @param[in] items
1786  *   Pointer to the list of items.
1787  * @param[in] actions
1788  *   Pointer to the list of actions.
1789  * @param[out] error
1790  *   Pointer to the error structure.
1791  *
1792  * @return
1793  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1794  */
1795 static inline int
1796 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1797 		   const struct rte_flow_attr *attr,
1798 		   const struct rte_flow_item items[],
1799 		   const struct rte_flow_action actions[],
1800 		   struct rte_flow_error *error)
1801 {
1802 	const struct mlx5_flow_driver_ops *fops;
1803 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1804 
1805 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1806 	fops = flow_get_drv_ops(type);
1807 	return fops->translate(dev, dev_flow, attr, items, actions, error);
1808 }
1809 
1810 /**
1811  * Flow driver apply API. This abstracts calling driver specific functions.
1812  * Parent flow (rte_flow) should have driver type (drv_type). It applies
1813  * translated driver flows on to device. flow_drv_translate() must precede.
1814  *
1815  * @param[in] dev
1816  *   Pointer to Ethernet device structure.
1817  * @param[in, out] flow
1818  *   Pointer to flow structure.
1819  * @param[out] error
1820  *   Pointer to error structure.
1821  *
1822  * @return
1823  *   0 on success, a negative errno value otherwise and rte_errno is set.
1824  */
1825 static inline int
1826 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1827 	       struct rte_flow_error *error)
1828 {
1829 	const struct mlx5_flow_driver_ops *fops;
1830 	enum mlx5_flow_drv_type type = flow->drv_type;
1831 
1832 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1833 	fops = flow_get_drv_ops(type);
1834 	return fops->apply(dev, flow, error);
1835 }
1836 
1837 /**
1838  * Flow driver remove API. This abstracts calling driver specific functions.
1839  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1840  * on device. All the resources of the flow should be freed by calling
1841  * flow_dv_destroy().
1842  *
1843  * @param[in] dev
1844  *   Pointer to Ethernet device.
1845  * @param[in, out] flow
1846  *   Pointer to flow structure.
1847  */
1848 static inline void
1849 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1850 {
1851 	const struct mlx5_flow_driver_ops *fops;
1852 	enum mlx5_flow_drv_type type = flow->drv_type;
1853 
1854 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1855 	fops = flow_get_drv_ops(type);
1856 	fops->remove(dev, flow);
1857 }
1858 
1859 /**
1860  * Flow driver destroy API. This abstracts calling driver specific functions.
1861  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1862  * on device and releases resources of the flow.
1863  *
1864  * @param[in] dev
1865  *   Pointer to Ethernet device.
1866  * @param[in, out] flow
1867  *   Pointer to flow structure.
1868  */
1869 static inline void
1870 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1871 {
1872 	const struct mlx5_flow_driver_ops *fops;
1873 	enum mlx5_flow_drv_type type = flow->drv_type;
1874 
1875 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1876 	fops = flow_get_drv_ops(type);
1877 	fops->destroy(dev, flow);
1878 }
1879 
1880 /**
1881  * Validate a flow supported by the NIC.
1882  *
1883  * @see rte_flow_validate()
1884  * @see rte_flow_ops
1885  */
1886 int
1887 mlx5_flow_validate(struct rte_eth_dev *dev,
1888 		   const struct rte_flow_attr *attr,
1889 		   const struct rte_flow_item items[],
1890 		   const struct rte_flow_action actions[],
1891 		   struct rte_flow_error *error)
1892 {
1893 	int ret;
1894 
1895 	ret = flow_drv_validate(dev, attr, items, actions, error);
1896 	if (ret < 0)
1897 		return ret;
1898 	return 0;
1899 }
1900 
1901 /**
1902  * Get RSS action from the action list.
1903  *
1904  * @param[in] actions
1905  *   Pointer to the list of actions.
1906  *
1907  * @return
1908  *   Pointer to the RSS action if exist, else return NULL.
1909  */
1910 static const struct rte_flow_action_rss*
1911 mlx5_flow_get_rss_action(const struct rte_flow_action actions[])
1912 {
1913 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1914 		switch (actions->type) {
1915 		case RTE_FLOW_ACTION_TYPE_RSS:
1916 			return (const struct rte_flow_action_rss *)
1917 			       actions->conf;
1918 		default:
1919 			break;
1920 		}
1921 	}
1922 	return NULL;
1923 }
1924 
1925 static unsigned int
1926 mlx5_find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
1927 {
1928 	const struct rte_flow_item *item;
1929 	unsigned int has_vlan = 0;
1930 
1931 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
1932 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
1933 			has_vlan = 1;
1934 			break;
1935 		}
1936 	}
1937 	if (has_vlan)
1938 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
1939 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
1940 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
1941 			       MLX5_EXPANSION_ROOT_OUTER;
1942 }
1943 
1944 /**
1945  * Create a flow and add it to @p list.
1946  *
1947  * @param dev
1948  *   Pointer to Ethernet device.
1949  * @param list
1950  *   Pointer to a TAILQ flow list.
1951  * @param[in] attr
1952  *   Flow rule attributes.
1953  * @param[in] items
1954  *   Pattern specification (list terminated by the END pattern item).
1955  * @param[in] actions
1956  *   Associated actions (list terminated by the END action).
1957  * @param[out] error
1958  *   Perform verbose error reporting if not NULL.
1959  *
1960  * @return
1961  *   A flow on success, NULL otherwise and rte_errno is set.
1962  */
1963 static struct rte_flow *
1964 mlx5_flow_list_create(struct rte_eth_dev *dev,
1965 		      struct mlx5_flows *list,
1966 		      const struct rte_flow_attr *attr,
1967 		      const struct rte_flow_item items[],
1968 		      const struct rte_flow_action actions[],
1969 		      struct rte_flow_error *error)
1970 {
1971 	struct rte_flow *flow = NULL;
1972 	struct mlx5_flow *dev_flow;
1973 	uint64_t action_flags = 0;
1974 	uint64_t item_flags = 0;
1975 	const struct rte_flow_action_rss *rss;
1976 	union {
1977 		struct rte_flow_expand_rss buf;
1978 		uint8_t buffer[2048];
1979 	} expand_buffer;
1980 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
1981 	int ret;
1982 	uint32_t i;
1983 	uint32_t flow_size;
1984 
1985 	ret = flow_drv_validate(dev, attr, items, actions, error);
1986 	if (ret < 0)
1987 		return NULL;
1988 	flow_size = sizeof(struct rte_flow);
1989 	rss = mlx5_flow_get_rss_action(actions);
1990 	if (rss)
1991 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
1992 					    sizeof(void *));
1993 	else
1994 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
1995 	flow = rte_calloc(__func__, 1, flow_size, 0);
1996 	flow->drv_type = flow_get_drv_type(dev, attr);
1997 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
1998 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
1999 	flow->queue = (void *)(flow + 1);
2000 	LIST_INIT(&flow->dev_flows);
2001 	if (rss && rss->types) {
2002 		unsigned int graph_root;
2003 
2004 		graph_root = mlx5_find_graph_root(items, rss->level);
2005 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2006 					  items, rss->types,
2007 					  mlx5_support_expansion,
2008 					  graph_root);
2009 		assert(ret > 0 &&
2010 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
2011 	} else {
2012 		buf->entries = 1;
2013 		buf->entry[0].pattern = (void *)(uintptr_t)items;
2014 	}
2015 	for (i = 0; i < buf->entries; ++i) {
2016 		dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2017 					    actions, &item_flags, &action_flags,
2018 					    error);
2019 		if (!dev_flow)
2020 			goto error;
2021 		dev_flow->flow = flow;
2022 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2023 		ret = flow_drv_translate(dev, dev_flow, attr,
2024 					 buf->entry[i].pattern,
2025 					 actions, error);
2026 		if (ret < 0)
2027 			goto error;
2028 	}
2029 	if (dev->data->dev_started) {
2030 		ret = flow_drv_apply(dev, flow, error);
2031 		if (ret < 0)
2032 			goto error;
2033 	}
2034 	TAILQ_INSERT_TAIL(list, flow, next);
2035 	mlx5_flow_rxq_flags_set(dev, flow);
2036 	return flow;
2037 error:
2038 	ret = rte_errno; /* Save rte_errno before cleanup. */
2039 	assert(flow);
2040 	flow_drv_destroy(dev, flow);
2041 	rte_free(flow);
2042 	rte_errno = ret; /* Restore rte_errno. */
2043 	return NULL;
2044 }
2045 
2046 /**
2047  * Create a flow.
2048  *
2049  * @see rte_flow_create()
2050  * @see rte_flow_ops
2051  */
2052 struct rte_flow *
2053 mlx5_flow_create(struct rte_eth_dev *dev,
2054 		 const struct rte_flow_attr *attr,
2055 		 const struct rte_flow_item items[],
2056 		 const struct rte_flow_action actions[],
2057 		 struct rte_flow_error *error)
2058 {
2059 	return mlx5_flow_list_create
2060 		(dev, &((struct priv *)dev->data->dev_private)->flows,
2061 		 attr, items, actions, error);
2062 }
2063 
2064 /**
2065  * Destroy a flow in a list.
2066  *
2067  * @param dev
2068  *   Pointer to Ethernet device.
2069  * @param list
2070  *   Pointer to a TAILQ flow list.
2071  * @param[in] flow
2072  *   Flow to destroy.
2073  */
2074 static void
2075 mlx5_flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2076 		       struct rte_flow *flow)
2077 {
2078 	flow_drv_destroy(dev, flow);
2079 	TAILQ_REMOVE(list, flow, next);
2080 	/*
2081 	 * Update RX queue flags only if port is started, otherwise it is
2082 	 * already clean.
2083 	 */
2084 	if (dev->data->dev_started)
2085 		mlx5_flow_rxq_flags_trim(dev, flow);
2086 	rte_free(flow);
2087 }
2088 
2089 /**
2090  * Destroy all flows.
2091  *
2092  * @param dev
2093  *   Pointer to Ethernet device.
2094  * @param list
2095  *   Pointer to a TAILQ flow list.
2096  */
2097 void
2098 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2099 {
2100 	while (!TAILQ_EMPTY(list)) {
2101 		struct rte_flow *flow;
2102 
2103 		flow = TAILQ_FIRST(list);
2104 		mlx5_flow_list_destroy(dev, list, flow);
2105 	}
2106 }
2107 
2108 /**
2109  * Remove all flows.
2110  *
2111  * @param dev
2112  *   Pointer to Ethernet device.
2113  * @param list
2114  *   Pointer to a TAILQ flow list.
2115  */
2116 void
2117 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2118 {
2119 	struct rte_flow *flow;
2120 
2121 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2122 		flow_drv_remove(dev, flow);
2123 	mlx5_flow_rxq_flags_clear(dev);
2124 }
2125 
2126 /**
2127  * Add all flows.
2128  *
2129  * @param dev
2130  *   Pointer to Ethernet device.
2131  * @param list
2132  *   Pointer to a TAILQ flow list.
2133  *
2134  * @return
2135  *   0 on success, a negative errno value otherwise and rte_errno is set.
2136  */
2137 int
2138 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2139 {
2140 	struct rte_flow *flow;
2141 	struct rte_flow_error error;
2142 	int ret = 0;
2143 
2144 	TAILQ_FOREACH(flow, list, next) {
2145 		ret = flow_drv_apply(dev, flow, &error);
2146 		if (ret < 0)
2147 			goto error;
2148 		mlx5_flow_rxq_flags_set(dev, flow);
2149 	}
2150 	return 0;
2151 error:
2152 	ret = rte_errno; /* Save rte_errno before cleanup. */
2153 	mlx5_flow_stop(dev, list);
2154 	rte_errno = ret; /* Restore rte_errno. */
2155 	return -rte_errno;
2156 }
2157 
2158 /**
2159  * Verify the flow list is empty
2160  *
2161  * @param dev
2162  *  Pointer to Ethernet device.
2163  *
2164  * @return the number of flows not released.
2165  */
2166 int
2167 mlx5_flow_verify(struct rte_eth_dev *dev)
2168 {
2169 	struct priv *priv = dev->data->dev_private;
2170 	struct rte_flow *flow;
2171 	int ret = 0;
2172 
2173 	TAILQ_FOREACH(flow, &priv->flows, next) {
2174 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
2175 			dev->data->port_id, (void *)flow);
2176 		++ret;
2177 	}
2178 	return ret;
2179 }
2180 
2181 /**
2182  * Enable a control flow configured from the control plane.
2183  *
2184  * @param dev
2185  *   Pointer to Ethernet device.
2186  * @param eth_spec
2187  *   An Ethernet flow spec to apply.
2188  * @param eth_mask
2189  *   An Ethernet flow mask to apply.
2190  * @param vlan_spec
2191  *   A VLAN flow spec to apply.
2192  * @param vlan_mask
2193  *   A VLAN flow mask to apply.
2194  *
2195  * @return
2196  *   0 on success, a negative errno value otherwise and rte_errno is set.
2197  */
2198 int
2199 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2200 		    struct rte_flow_item_eth *eth_spec,
2201 		    struct rte_flow_item_eth *eth_mask,
2202 		    struct rte_flow_item_vlan *vlan_spec,
2203 		    struct rte_flow_item_vlan *vlan_mask)
2204 {
2205 	struct priv *priv = dev->data->dev_private;
2206 	const struct rte_flow_attr attr = {
2207 		.ingress = 1,
2208 		.priority = MLX5_FLOW_PRIO_RSVD,
2209 	};
2210 	struct rte_flow_item items[] = {
2211 		{
2212 			.type = RTE_FLOW_ITEM_TYPE_ETH,
2213 			.spec = eth_spec,
2214 			.last = NULL,
2215 			.mask = eth_mask,
2216 		},
2217 		{
2218 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2219 					      RTE_FLOW_ITEM_TYPE_END,
2220 			.spec = vlan_spec,
2221 			.last = NULL,
2222 			.mask = vlan_mask,
2223 		},
2224 		{
2225 			.type = RTE_FLOW_ITEM_TYPE_END,
2226 		},
2227 	};
2228 	uint16_t queue[priv->reta_idx_n];
2229 	struct rte_flow_action_rss action_rss = {
2230 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2231 		.level = 0,
2232 		.types = priv->rss_conf.rss_hf,
2233 		.key_len = priv->rss_conf.rss_key_len,
2234 		.queue_num = priv->reta_idx_n,
2235 		.key = priv->rss_conf.rss_key,
2236 		.queue = queue,
2237 	};
2238 	struct rte_flow_action actions[] = {
2239 		{
2240 			.type = RTE_FLOW_ACTION_TYPE_RSS,
2241 			.conf = &action_rss,
2242 		},
2243 		{
2244 			.type = RTE_FLOW_ACTION_TYPE_END,
2245 		},
2246 	};
2247 	struct rte_flow *flow;
2248 	struct rte_flow_error error;
2249 	unsigned int i;
2250 
2251 	if (!priv->reta_idx_n) {
2252 		rte_errno = EINVAL;
2253 		return -rte_errno;
2254 	}
2255 	for (i = 0; i != priv->reta_idx_n; ++i)
2256 		queue[i] = (*priv->reta_idx)[i];
2257 	flow = mlx5_flow_list_create(dev, &priv->ctrl_flows, &attr, items,
2258 				     actions, &error);
2259 	if (!flow)
2260 		return -rte_errno;
2261 	return 0;
2262 }
2263 
2264 /**
2265  * Enable a flow control configured from the control plane.
2266  *
2267  * @param dev
2268  *   Pointer to Ethernet device.
2269  * @param eth_spec
2270  *   An Ethernet flow spec to apply.
2271  * @param eth_mask
2272  *   An Ethernet flow mask to apply.
2273  *
2274  * @return
2275  *   0 on success, a negative errno value otherwise and rte_errno is set.
2276  */
2277 int
2278 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2279 	       struct rte_flow_item_eth *eth_spec,
2280 	       struct rte_flow_item_eth *eth_mask)
2281 {
2282 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2283 }
2284 
2285 /**
2286  * Destroy a flow.
2287  *
2288  * @see rte_flow_destroy()
2289  * @see rte_flow_ops
2290  */
2291 int
2292 mlx5_flow_destroy(struct rte_eth_dev *dev,
2293 		  struct rte_flow *flow,
2294 		  struct rte_flow_error *error __rte_unused)
2295 {
2296 	struct priv *priv = dev->data->dev_private;
2297 
2298 	mlx5_flow_list_destroy(dev, &priv->flows, flow);
2299 	return 0;
2300 }
2301 
2302 /**
2303  * Destroy all flows.
2304  *
2305  * @see rte_flow_flush()
2306  * @see rte_flow_ops
2307  */
2308 int
2309 mlx5_flow_flush(struct rte_eth_dev *dev,
2310 		struct rte_flow_error *error __rte_unused)
2311 {
2312 	struct priv *priv = dev->data->dev_private;
2313 
2314 	mlx5_flow_list_flush(dev, &priv->flows);
2315 	return 0;
2316 }
2317 
2318 /**
2319  * Isolated mode.
2320  *
2321  * @see rte_flow_isolate()
2322  * @see rte_flow_ops
2323  */
2324 int
2325 mlx5_flow_isolate(struct rte_eth_dev *dev,
2326 		  int enable,
2327 		  struct rte_flow_error *error)
2328 {
2329 	struct priv *priv = dev->data->dev_private;
2330 
2331 	if (dev->data->dev_started) {
2332 		rte_flow_error_set(error, EBUSY,
2333 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2334 				   NULL,
2335 				   "port must be stopped first");
2336 		return -rte_errno;
2337 	}
2338 	priv->isolated = !!enable;
2339 	if (enable)
2340 		dev->dev_ops = &mlx5_dev_ops_isolate;
2341 	else
2342 		dev->dev_ops = &mlx5_dev_ops;
2343 	return 0;
2344 }
2345 
2346 /**
2347  * Query flow counter.
2348  *
2349  * @param flow
2350  *   Pointer to the flow.
2351  *
2352  * @return
2353  *   0 on success, a negative errno value otherwise and rte_errno is set.
2354  */
2355 static int
2356 mlx5_flow_query_count(struct rte_flow *flow __rte_unused,
2357 		      void *data __rte_unused,
2358 		      struct rte_flow_error *error)
2359 {
2360 #ifdef HAVE_IBV_DEVICE_COUNTERS_SET_SUPPORT
2361 	if (flow->actions & MLX5_FLOW_ACTION_COUNT) {
2362 		struct rte_flow_query_count *qc = data;
2363 		uint64_t counters[2] = {0, 0};
2364 		struct ibv_query_counter_set_attr query_cs_attr = {
2365 			.cs = flow->counter->cs,
2366 			.query_flags = IBV_COUNTER_SET_FORCE_UPDATE,
2367 		};
2368 		struct ibv_counter_set_data query_out = {
2369 			.out = counters,
2370 			.outlen = 2 * sizeof(uint64_t),
2371 		};
2372 		int err = mlx5_glue->query_counter_set(&query_cs_attr,
2373 						       &query_out);
2374 
2375 		if (err)
2376 			return rte_flow_error_set
2377 				(error, err,
2378 				 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2379 				 NULL,
2380 				 "cannot read counter");
2381 		qc->hits_set = 1;
2382 		qc->bytes_set = 1;
2383 		qc->hits = counters[0] - flow->counter->hits;
2384 		qc->bytes = counters[1] - flow->counter->bytes;
2385 		if (qc->reset) {
2386 			flow->counter->hits = counters[0];
2387 			flow->counter->bytes = counters[1];
2388 		}
2389 		return 0;
2390 	}
2391 	return rte_flow_error_set(error, EINVAL,
2392 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2393 				  NULL,
2394 				  "flow does not have counter");
2395 #endif
2396 	return rte_flow_error_set(error, ENOTSUP,
2397 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2398 				  NULL,
2399 				  "counters are not available");
2400 }
2401 
2402 /**
2403  * Query a flows.
2404  *
2405  * @see rte_flow_query()
2406  * @see rte_flow_ops
2407  */
2408 int
2409 mlx5_flow_query(struct rte_eth_dev *dev __rte_unused,
2410 		struct rte_flow *flow,
2411 		const struct rte_flow_action *actions,
2412 		void *data,
2413 		struct rte_flow_error *error)
2414 {
2415 	int ret = 0;
2416 
2417 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2418 		switch (actions->type) {
2419 		case RTE_FLOW_ACTION_TYPE_VOID:
2420 			break;
2421 		case RTE_FLOW_ACTION_TYPE_COUNT:
2422 			ret = mlx5_flow_query_count(flow, data, error);
2423 			break;
2424 		default:
2425 			return rte_flow_error_set(error, ENOTSUP,
2426 						  RTE_FLOW_ERROR_TYPE_ACTION,
2427 						  actions,
2428 						  "action not supported");
2429 		}
2430 		if (ret < 0)
2431 			return ret;
2432 	}
2433 	return 0;
2434 }
2435 
2436 /**
2437  * Convert a flow director filter to a generic flow.
2438  *
2439  * @param dev
2440  *   Pointer to Ethernet device.
2441  * @param fdir_filter
2442  *   Flow director filter to add.
2443  * @param attributes
2444  *   Generic flow parameters structure.
2445  *
2446  * @return
2447  *   0 on success, a negative errno value otherwise and rte_errno is set.
2448  */
2449 static int
2450 mlx5_fdir_filter_convert(struct rte_eth_dev *dev,
2451 			 const struct rte_eth_fdir_filter *fdir_filter,
2452 			 struct mlx5_fdir *attributes)
2453 {
2454 	struct priv *priv = dev->data->dev_private;
2455 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
2456 	const struct rte_eth_fdir_masks *mask =
2457 		&dev->data->dev_conf.fdir_conf.mask;
2458 
2459 	/* Validate queue number. */
2460 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2461 		DRV_LOG(ERR, "port %u invalid queue number %d",
2462 			dev->data->port_id, fdir_filter->action.rx_queue);
2463 		rte_errno = EINVAL;
2464 		return -rte_errno;
2465 	}
2466 	attributes->attr.ingress = 1;
2467 	attributes->items[0] = (struct rte_flow_item) {
2468 		.type = RTE_FLOW_ITEM_TYPE_ETH,
2469 		.spec = &attributes->l2,
2470 		.mask = &attributes->l2_mask,
2471 	};
2472 	switch (fdir_filter->action.behavior) {
2473 	case RTE_ETH_FDIR_ACCEPT:
2474 		attributes->actions[0] = (struct rte_flow_action){
2475 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
2476 			.conf = &attributes->queue,
2477 		};
2478 		break;
2479 	case RTE_ETH_FDIR_REJECT:
2480 		attributes->actions[0] = (struct rte_flow_action){
2481 			.type = RTE_FLOW_ACTION_TYPE_DROP,
2482 		};
2483 		break;
2484 	default:
2485 		DRV_LOG(ERR, "port %u invalid behavior %d",
2486 			dev->data->port_id,
2487 			fdir_filter->action.behavior);
2488 		rte_errno = ENOTSUP;
2489 		return -rte_errno;
2490 	}
2491 	attributes->queue.index = fdir_filter->action.rx_queue;
2492 	/* Handle L3. */
2493 	switch (fdir_filter->input.flow_type) {
2494 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2495 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2496 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2497 		attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2498 			.src_addr = input->flow.ip4_flow.src_ip,
2499 			.dst_addr = input->flow.ip4_flow.dst_ip,
2500 			.time_to_live = input->flow.ip4_flow.ttl,
2501 			.type_of_service = input->flow.ip4_flow.tos,
2502 		};
2503 		attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2504 			.src_addr = mask->ipv4_mask.src_ip,
2505 			.dst_addr = mask->ipv4_mask.dst_ip,
2506 			.time_to_live = mask->ipv4_mask.ttl,
2507 			.type_of_service = mask->ipv4_mask.tos,
2508 			.next_proto_id = mask->ipv4_mask.proto,
2509 		};
2510 		attributes->items[1] = (struct rte_flow_item){
2511 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
2512 			.spec = &attributes->l3,
2513 			.mask = &attributes->l3_mask,
2514 		};
2515 		break;
2516 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2517 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2518 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2519 		attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2520 			.hop_limits = input->flow.ipv6_flow.hop_limits,
2521 			.proto = input->flow.ipv6_flow.proto,
2522 		};
2523 
2524 		memcpy(attributes->l3.ipv6.hdr.src_addr,
2525 		       input->flow.ipv6_flow.src_ip,
2526 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2527 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
2528 		       input->flow.ipv6_flow.dst_ip,
2529 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2530 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2531 		       mask->ipv6_mask.src_ip,
2532 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2533 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2534 		       mask->ipv6_mask.dst_ip,
2535 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2536 		attributes->items[1] = (struct rte_flow_item){
2537 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
2538 			.spec = &attributes->l3,
2539 			.mask = &attributes->l3_mask,
2540 		};
2541 		break;
2542 	default:
2543 		DRV_LOG(ERR, "port %u invalid flow type%d",
2544 			dev->data->port_id, fdir_filter->input.flow_type);
2545 		rte_errno = ENOTSUP;
2546 		return -rte_errno;
2547 	}
2548 	/* Handle L4. */
2549 	switch (fdir_filter->input.flow_type) {
2550 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2551 		attributes->l4.udp.hdr = (struct udp_hdr){
2552 			.src_port = input->flow.udp4_flow.src_port,
2553 			.dst_port = input->flow.udp4_flow.dst_port,
2554 		};
2555 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2556 			.src_port = mask->src_port_mask,
2557 			.dst_port = mask->dst_port_mask,
2558 		};
2559 		attributes->items[2] = (struct rte_flow_item){
2560 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2561 			.spec = &attributes->l4,
2562 			.mask = &attributes->l4_mask,
2563 		};
2564 		break;
2565 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2566 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2567 			.src_port = input->flow.tcp4_flow.src_port,
2568 			.dst_port = input->flow.tcp4_flow.dst_port,
2569 		};
2570 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2571 			.src_port = mask->src_port_mask,
2572 			.dst_port = mask->dst_port_mask,
2573 		};
2574 		attributes->items[2] = (struct rte_flow_item){
2575 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2576 			.spec = &attributes->l4,
2577 			.mask = &attributes->l4_mask,
2578 		};
2579 		break;
2580 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2581 		attributes->l4.udp.hdr = (struct udp_hdr){
2582 			.src_port = input->flow.udp6_flow.src_port,
2583 			.dst_port = input->flow.udp6_flow.dst_port,
2584 		};
2585 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2586 			.src_port = mask->src_port_mask,
2587 			.dst_port = mask->dst_port_mask,
2588 		};
2589 		attributes->items[2] = (struct rte_flow_item){
2590 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2591 			.spec = &attributes->l4,
2592 			.mask = &attributes->l4_mask,
2593 		};
2594 		break;
2595 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2596 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2597 			.src_port = input->flow.tcp6_flow.src_port,
2598 			.dst_port = input->flow.tcp6_flow.dst_port,
2599 		};
2600 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2601 			.src_port = mask->src_port_mask,
2602 			.dst_port = mask->dst_port_mask,
2603 		};
2604 		attributes->items[2] = (struct rte_flow_item){
2605 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2606 			.spec = &attributes->l4,
2607 			.mask = &attributes->l4_mask,
2608 		};
2609 		break;
2610 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2611 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2612 		break;
2613 	default:
2614 		DRV_LOG(ERR, "port %u invalid flow type%d",
2615 			dev->data->port_id, fdir_filter->input.flow_type);
2616 		rte_errno = ENOTSUP;
2617 		return -rte_errno;
2618 	}
2619 	return 0;
2620 }
2621 
2622 /**
2623  * Add new flow director filter and store it in list.
2624  *
2625  * @param dev
2626  *   Pointer to Ethernet device.
2627  * @param fdir_filter
2628  *   Flow director filter to add.
2629  *
2630  * @return
2631  *   0 on success, a negative errno value otherwise and rte_errno is set.
2632  */
2633 static int
2634 mlx5_fdir_filter_add(struct rte_eth_dev *dev,
2635 		     const struct rte_eth_fdir_filter *fdir_filter)
2636 {
2637 	struct priv *priv = dev->data->dev_private;
2638 	struct mlx5_fdir attributes = {
2639 		.attr.group = 0,
2640 		.l2_mask = {
2641 			.dst.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2642 			.src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2643 			.type = 0,
2644 		},
2645 	};
2646 	struct rte_flow_error error;
2647 	struct rte_flow *flow;
2648 	int ret;
2649 
2650 	ret = mlx5_fdir_filter_convert(dev, fdir_filter, &attributes);
2651 	if (ret)
2652 		return ret;
2653 	flow = mlx5_flow_list_create(dev, &priv->flows, &attributes.attr,
2654 				     attributes.items, attributes.actions,
2655 				     &error);
2656 	if (flow) {
2657 		DRV_LOG(DEBUG, "port %u FDIR created %p", dev->data->port_id,
2658 			(void *)flow);
2659 		return 0;
2660 	}
2661 	return -rte_errno;
2662 }
2663 
2664 /**
2665  * Delete specific filter.
2666  *
2667  * @param dev
2668  *   Pointer to Ethernet device.
2669  * @param fdir_filter
2670  *   Filter to be deleted.
2671  *
2672  * @return
2673  *   0 on success, a negative errno value otherwise and rte_errno is set.
2674  */
2675 static int
2676 mlx5_fdir_filter_delete(struct rte_eth_dev *dev __rte_unused,
2677 			const struct rte_eth_fdir_filter *fdir_filter
2678 			__rte_unused)
2679 {
2680 	rte_errno = ENOTSUP;
2681 	return -rte_errno;
2682 }
2683 
2684 /**
2685  * Update queue for specific filter.
2686  *
2687  * @param dev
2688  *   Pointer to Ethernet device.
2689  * @param fdir_filter
2690  *   Filter to be updated.
2691  *
2692  * @return
2693  *   0 on success, a negative errno value otherwise and rte_errno is set.
2694  */
2695 static int
2696 mlx5_fdir_filter_update(struct rte_eth_dev *dev,
2697 			const struct rte_eth_fdir_filter *fdir_filter)
2698 {
2699 	int ret;
2700 
2701 	ret = mlx5_fdir_filter_delete(dev, fdir_filter);
2702 	if (ret)
2703 		return ret;
2704 	return mlx5_fdir_filter_add(dev, fdir_filter);
2705 }
2706 
2707 /**
2708  * Flush all filters.
2709  *
2710  * @param dev
2711  *   Pointer to Ethernet device.
2712  */
2713 static void
2714 mlx5_fdir_filter_flush(struct rte_eth_dev *dev)
2715 {
2716 	struct priv *priv = dev->data->dev_private;
2717 
2718 	mlx5_flow_list_flush(dev, &priv->flows);
2719 }
2720 
2721 /**
2722  * Get flow director information.
2723  *
2724  * @param dev
2725  *   Pointer to Ethernet device.
2726  * @param[out] fdir_info
2727  *   Resulting flow director information.
2728  */
2729 static void
2730 mlx5_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2731 {
2732 	struct rte_eth_fdir_masks *mask =
2733 		&dev->data->dev_conf.fdir_conf.mask;
2734 
2735 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2736 	fdir_info->guarant_spc = 0;
2737 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2738 	fdir_info->max_flexpayload = 0;
2739 	fdir_info->flow_types_mask[0] = 0;
2740 	fdir_info->flex_payload_unit = 0;
2741 	fdir_info->max_flex_payload_segment_num = 0;
2742 	fdir_info->flex_payload_limit = 0;
2743 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2744 }
2745 
2746 /**
2747  * Deal with flow director operations.
2748  *
2749  * @param dev
2750  *   Pointer to Ethernet device.
2751  * @param filter_op
2752  *   Operation to perform.
2753  * @param arg
2754  *   Pointer to operation-specific structure.
2755  *
2756  * @return
2757  *   0 on success, a negative errno value otherwise and rte_errno is set.
2758  */
2759 static int
2760 mlx5_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2761 		    void *arg)
2762 {
2763 	enum rte_fdir_mode fdir_mode =
2764 		dev->data->dev_conf.fdir_conf.mode;
2765 
2766 	if (filter_op == RTE_ETH_FILTER_NOP)
2767 		return 0;
2768 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2769 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2770 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
2771 			dev->data->port_id, fdir_mode);
2772 		rte_errno = EINVAL;
2773 		return -rte_errno;
2774 	}
2775 	switch (filter_op) {
2776 	case RTE_ETH_FILTER_ADD:
2777 		return mlx5_fdir_filter_add(dev, arg);
2778 	case RTE_ETH_FILTER_UPDATE:
2779 		return mlx5_fdir_filter_update(dev, arg);
2780 	case RTE_ETH_FILTER_DELETE:
2781 		return mlx5_fdir_filter_delete(dev, arg);
2782 	case RTE_ETH_FILTER_FLUSH:
2783 		mlx5_fdir_filter_flush(dev);
2784 		break;
2785 	case RTE_ETH_FILTER_INFO:
2786 		mlx5_fdir_info_get(dev, arg);
2787 		break;
2788 	default:
2789 		DRV_LOG(DEBUG, "port %u unknown operation %u",
2790 			dev->data->port_id, filter_op);
2791 		rte_errno = EINVAL;
2792 		return -rte_errno;
2793 	}
2794 	return 0;
2795 }
2796 
2797 /**
2798  * Manage filter operations.
2799  *
2800  * @param dev
2801  *   Pointer to Ethernet device structure.
2802  * @param filter_type
2803  *   Filter type.
2804  * @param filter_op
2805  *   Operation to perform.
2806  * @param arg
2807  *   Pointer to operation-specific structure.
2808  *
2809  * @return
2810  *   0 on success, a negative errno value otherwise and rte_errno is set.
2811  */
2812 int
2813 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2814 		     enum rte_filter_type filter_type,
2815 		     enum rte_filter_op filter_op,
2816 		     void *arg)
2817 {
2818 	switch (filter_type) {
2819 	case RTE_ETH_FILTER_GENERIC:
2820 		if (filter_op != RTE_ETH_FILTER_GET) {
2821 			rte_errno = EINVAL;
2822 			return -rte_errno;
2823 		}
2824 		*(const void **)arg = &mlx5_flow_ops;
2825 		return 0;
2826 	case RTE_ETH_FILTER_FDIR:
2827 		return mlx5_fdir_ctrl_func(dev, filter_op, arg);
2828 	default:
2829 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
2830 			dev->data->port_id, filter_type);
2831 		rte_errno = ENOTSUP;
2832 		return -rte_errno;
2833 	}
2834 	return 0;
2835 }
2836