xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 58b1312e9d193f91c20eace59555b3ae3c88272d)
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 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 __rte_unused,
925 				const struct rte_flow_attr *attr,
926 				struct rte_flow_error *error)
927 {
928 	if (attr->egress)
929 		return rte_flow_error_set(error, ENOTSUP,
930 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
931 					  "count action not supported for "
932 					  "egress");
933 	return 0;
934 }
935 
936 /**
937  * Verify the @p attributes will be correctly understood by the NIC and store
938  * them in the @p flow if everything is correct.
939  *
940  * @param[in] dev
941  *   Pointer to the Ethernet device structure.
942  * @param[in] attributes
943  *   Pointer to flow attributes
944  * @param[out] error
945  *   Pointer to error structure.
946  *
947  * @return
948  *   0 on success, a negative errno value otherwise and rte_errno is set.
949  */
950 int
951 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
952 			      const struct rte_flow_attr *attributes,
953 			      struct rte_flow_error *error)
954 {
955 	struct priv *priv = dev->data->dev_private;
956 	uint32_t priority_max = priv->config.flow_prio - 1;
957 
958 	if (attributes->group)
959 		return rte_flow_error_set(error, ENOTSUP,
960 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
961 					  NULL, "groups is not supported");
962 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
963 	    attributes->priority >= priority_max)
964 		return rte_flow_error_set(error, ENOTSUP,
965 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
966 					  NULL, "priority out of range");
967 	if (attributes->egress)
968 		return rte_flow_error_set(error, ENOTSUP,
969 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
970 					  "egress is not supported");
971 	if (attributes->transfer)
972 		return rte_flow_error_set(error, ENOTSUP,
973 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
974 					  NULL, "transfer is not supported");
975 	if (!attributes->ingress)
976 		return rte_flow_error_set(error, EINVAL,
977 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
978 					  NULL,
979 					  "ingress attribute is mandatory");
980 	return 0;
981 }
982 
983 /**
984  * Validate Ethernet item.
985  *
986  * @param[in] item
987  *   Item specification.
988  * @param[in] item_flags
989  *   Bit-fields that holds the items detected until now.
990  * @param[out] error
991  *   Pointer to error structure.
992  *
993  * @return
994  *   0 on success, a negative errno value otherwise and rte_errno is set.
995  */
996 int
997 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
998 			    uint64_t item_flags,
999 			    struct rte_flow_error *error)
1000 {
1001 	const struct rte_flow_item_eth *mask = item->mask;
1002 	const struct rte_flow_item_eth nic_mask = {
1003 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1004 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1005 		.type = RTE_BE16(0xffff),
1006 	};
1007 	int ret;
1008 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1009 
1010 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L2)
1011 		return rte_flow_error_set(error, ENOTSUP,
1012 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1013 					  "3 levels of l2 are not supported");
1014 	if ((item_flags & MLX5_FLOW_LAYER_INNER_L2) && !tunnel)
1015 		return rte_flow_error_set(error, ENOTSUP,
1016 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1017 					  "2 L2 without tunnel are not supported");
1018 	if (!mask)
1019 		mask = &rte_flow_item_eth_mask;
1020 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1021 					(const uint8_t *)&nic_mask,
1022 					sizeof(struct rte_flow_item_eth),
1023 					error);
1024 	return ret;
1025 }
1026 
1027 /**
1028  * Validate VLAN item.
1029  *
1030  * @param[in] item
1031  *   Item specification.
1032  * @param[in] item_flags
1033  *   Bit-fields that holds the items detected until now.
1034  * @param[out] error
1035  *   Pointer to error structure.
1036  *
1037  * @return
1038  *   0 on success, a negative errno value otherwise and rte_errno is set.
1039  */
1040 int
1041 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1042 			     int64_t item_flags,
1043 			     struct rte_flow_error *error)
1044 {
1045 	const struct rte_flow_item_vlan *spec = item->spec;
1046 	const struct rte_flow_item_vlan *mask = item->mask;
1047 	const struct rte_flow_item_vlan nic_mask = {
1048 		.tci = RTE_BE16(0x0fff),
1049 		.inner_type = RTE_BE16(0xffff),
1050 	};
1051 	uint16_t vlan_tag = 0;
1052 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1053 	int ret;
1054 	const uint32_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1055 					MLX5_FLOW_LAYER_INNER_L4) :
1056 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1057 					MLX5_FLOW_LAYER_OUTER_L4);
1058 	const uint32_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1059 					MLX5_FLOW_LAYER_OUTER_VLAN;
1060 
1061 	if (item_flags & vlanm)
1062 		return rte_flow_error_set(error, EINVAL,
1063 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1064 					  "VLAN layer already configured");
1065 	else if ((item_flags & l34m) != 0)
1066 		return rte_flow_error_set(error, EINVAL,
1067 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1068 					  "L2 layer cannot follow L3/L4 layer");
1069 	if (!mask)
1070 		mask = &rte_flow_item_vlan_mask;
1071 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1072 					(const uint8_t *)&nic_mask,
1073 					sizeof(struct rte_flow_item_vlan),
1074 					error);
1075 	if (ret)
1076 		return ret;
1077 	if (spec) {
1078 		vlan_tag = spec->tci;
1079 		vlan_tag &= mask->tci;
1080 	}
1081 	/*
1082 	 * From verbs perspective an empty VLAN is equivalent
1083 	 * to a packet without VLAN layer.
1084 	 */
1085 	if (!vlan_tag)
1086 		return rte_flow_error_set(error, EINVAL,
1087 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1088 					  item->spec,
1089 					  "VLAN cannot be empty");
1090 	return 0;
1091 }
1092 
1093 /**
1094  * Validate IPV4 item.
1095  *
1096  * @param[in] item
1097  *   Item specification.
1098  * @param[in] item_flags
1099  *   Bit-fields that holds the items detected until now.
1100  * @param[out] error
1101  *   Pointer to error structure.
1102  *
1103  * @return
1104  *   0 on success, a negative errno value otherwise and rte_errno is set.
1105  */
1106 int
1107 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1108 			     int64_t item_flags,
1109 			     struct rte_flow_error *error)
1110 {
1111 	const struct rte_flow_item_ipv4 *mask = item->mask;
1112 	const struct rte_flow_item_ipv4 nic_mask = {
1113 		.hdr = {
1114 			.src_addr = RTE_BE32(0xffffffff),
1115 			.dst_addr = RTE_BE32(0xffffffff),
1116 			.type_of_service = 0xff,
1117 			.next_proto_id = 0xff,
1118 		},
1119 	};
1120 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1121 	int ret;
1122 
1123 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1124 				   MLX5_FLOW_LAYER_OUTER_L3))
1125 		return rte_flow_error_set(error, ENOTSUP,
1126 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1127 					  "multiple L3 layers not supported");
1128 	else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1129 					MLX5_FLOW_LAYER_OUTER_L4))
1130 		return rte_flow_error_set(error, EINVAL,
1131 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1132 					  "L3 cannot follow an L4 layer.");
1133 	if (!mask)
1134 		mask = &rte_flow_item_ipv4_mask;
1135 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1136 					(const uint8_t *)&nic_mask,
1137 					sizeof(struct rte_flow_item_ipv4),
1138 					error);
1139 	if (ret < 0)
1140 		return ret;
1141 	return 0;
1142 }
1143 
1144 /**
1145  * Validate IPV6 item.
1146  *
1147  * @param[in] item
1148  *   Item specification.
1149  * @param[in] item_flags
1150  *   Bit-fields that holds the items detected until now.
1151  * @param[out] error
1152  *   Pointer to error structure.
1153  *
1154  * @return
1155  *   0 on success, a negative errno value otherwise and rte_errno is set.
1156  */
1157 int
1158 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1159 			     uint64_t item_flags,
1160 			     struct rte_flow_error *error)
1161 {
1162 	const struct rte_flow_item_ipv6 *mask = item->mask;
1163 	const struct rte_flow_item_ipv6 nic_mask = {
1164 		.hdr = {
1165 			.src_addr =
1166 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1167 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1168 			.dst_addr =
1169 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1170 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1171 			.vtc_flow = RTE_BE32(0xffffffff),
1172 			.proto = 0xff,
1173 			.hop_limits = 0xff,
1174 		},
1175 	};
1176 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1177 	int ret;
1178 
1179 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1180 				   MLX5_FLOW_LAYER_OUTER_L3))
1181 		return rte_flow_error_set(error, ENOTSUP,
1182 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1183 					  "multiple L3 layers not supported");
1184 	else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1185 					MLX5_FLOW_LAYER_OUTER_L4))
1186 		return rte_flow_error_set(error, EINVAL,
1187 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1188 					  "L3 cannot follow an L4 layer.");
1189 	/*
1190 	 * IPv6 is not recognised by the NIC inside a GRE tunnel.
1191 	 * Such support has to be disabled as the rule will be
1192 	 * accepted.  Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1193 	 * Mellanox OFED 4.4-1.0.0.0.
1194 	 */
1195 	if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1196 		return rte_flow_error_set(error, ENOTSUP,
1197 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1198 					  "IPv6 inside a GRE tunnel is"
1199 					  " not recognised.");
1200 	if (!mask)
1201 		mask = &rte_flow_item_ipv6_mask;
1202 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1203 					(const uint8_t *)&nic_mask,
1204 					sizeof(struct rte_flow_item_ipv6),
1205 					error);
1206 	if (ret < 0)
1207 		return ret;
1208 	return 0;
1209 }
1210 
1211 /**
1212  * Validate UDP item.
1213  *
1214  * @param[in] item
1215  *   Item specification.
1216  * @param[in] item_flags
1217  *   Bit-fields that holds the items detected until now.
1218  * @param[in] target_protocol
1219  *   The next protocol in the previous item.
1220  * @param[in] flow_mask
1221  *   mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1222  * @param[out] error
1223  *   Pointer to error structure.
1224  *
1225  * @return
1226  *   0 on success, a negative errno value otherwise and rte_errno is set.
1227  */
1228 int
1229 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1230 			    uint64_t item_flags,
1231 			    uint8_t target_protocol,
1232 			    struct rte_flow_error *error)
1233 {
1234 	const struct rte_flow_item_udp *mask = item->mask;
1235 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1236 	int ret;
1237 
1238 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1239 		return rte_flow_error_set(error, EINVAL,
1240 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1241 					  "protocol filtering not compatible"
1242 					  " with UDP layer");
1243 	if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1244 				     MLX5_FLOW_LAYER_OUTER_L3)))
1245 		return rte_flow_error_set(error, EINVAL,
1246 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1247 					  "L3 is mandatory to filter on L4");
1248 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1249 				   MLX5_FLOW_LAYER_OUTER_L4))
1250 		return rte_flow_error_set(error, EINVAL,
1251 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1252 					  "L4 layer is already present");
1253 	if (!mask)
1254 		mask = &rte_flow_item_udp_mask;
1255 	ret = mlx5_flow_item_acceptable
1256 		(item, (const uint8_t *)mask,
1257 		 (const uint8_t *)&rte_flow_item_udp_mask,
1258 		 sizeof(struct rte_flow_item_udp), error);
1259 	if (ret < 0)
1260 		return ret;
1261 	return 0;
1262 }
1263 
1264 /**
1265  * Validate TCP item.
1266  *
1267  * @param[in] item
1268  *   Item specification.
1269  * @param[in] item_flags
1270  *   Bit-fields that holds the items detected until now.
1271  * @param[in] target_protocol
1272  *   The next protocol in the previous item.
1273  * @param[out] error
1274  *   Pointer to error structure.
1275  *
1276  * @return
1277  *   0 on success, a negative errno value otherwise and rte_errno is set.
1278  */
1279 int
1280 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1281 			    uint64_t item_flags,
1282 			    uint8_t target_protocol,
1283 			    const struct rte_flow_item_tcp *flow_mask,
1284 			    struct rte_flow_error *error)
1285 {
1286 	const struct rte_flow_item_tcp *mask = item->mask;
1287 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1288 	int ret;
1289 
1290 	assert(flow_mask);
1291 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1292 		return rte_flow_error_set(error, EINVAL,
1293 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1294 					  "protocol filtering not compatible"
1295 					  " with TCP layer");
1296 	if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1297 				     MLX5_FLOW_LAYER_OUTER_L3)))
1298 		return rte_flow_error_set(error, EINVAL,
1299 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1300 					  "L3 is mandatory to filter on L4");
1301 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1302 				   MLX5_FLOW_LAYER_OUTER_L4))
1303 		return rte_flow_error_set(error, EINVAL,
1304 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1305 					  "L4 layer is already present");
1306 	if (!mask)
1307 		mask = &rte_flow_item_tcp_mask;
1308 	ret = mlx5_flow_item_acceptable
1309 		(item, (const uint8_t *)mask,
1310 		 (const uint8_t *)flow_mask,
1311 		 sizeof(struct rte_flow_item_tcp), error);
1312 	if (ret < 0)
1313 		return ret;
1314 	return 0;
1315 }
1316 
1317 /**
1318  * Validate VXLAN item.
1319  *
1320  * @param[in] item
1321  *   Item specification.
1322  * @param[in] item_flags
1323  *   Bit-fields that holds the items detected until now.
1324  * @param[in] target_protocol
1325  *   The next protocol in the previous item.
1326  * @param[out] error
1327  *   Pointer to error structure.
1328  *
1329  * @return
1330  *   0 on success, a negative errno value otherwise and rte_errno is set.
1331  */
1332 int
1333 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1334 			      uint64_t item_flags,
1335 			      struct rte_flow_error *error)
1336 {
1337 	const struct rte_flow_item_vxlan *spec = item->spec;
1338 	const struct rte_flow_item_vxlan *mask = item->mask;
1339 	int ret;
1340 	union vni {
1341 		uint32_t vlan_id;
1342 		uint8_t vni[4];
1343 	} id = { .vlan_id = 0, };
1344 	uint32_t vlan_id = 0;
1345 
1346 
1347 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1348 		return rte_flow_error_set(error, ENOTSUP,
1349 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1350 					  "a tunnel is already present");
1351 	/*
1352 	 * Verify only UDPv4 is present as defined in
1353 	 * https://tools.ietf.org/html/rfc7348
1354 	 */
1355 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1356 		return rte_flow_error_set(error, EINVAL,
1357 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1358 					  "no outer UDP layer found");
1359 	if (!mask)
1360 		mask = &rte_flow_item_vxlan_mask;
1361 	ret = mlx5_flow_item_acceptable
1362 		(item, (const uint8_t *)mask,
1363 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1364 		 sizeof(struct rte_flow_item_vxlan),
1365 		 error);
1366 	if (ret < 0)
1367 		return ret;
1368 	if (spec) {
1369 		memcpy(&id.vni[1], spec->vni, 3);
1370 		vlan_id = id.vlan_id;
1371 		memcpy(&id.vni[1], mask->vni, 3);
1372 		vlan_id &= id.vlan_id;
1373 	}
1374 	/*
1375 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1376 	 * only this layer is defined in the Verbs specification it is
1377 	 * interpreted as wildcard and all packets will match this
1378 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1379 	 * udp), all packets matching the layers before will also
1380 	 * match this rule.  To avoid such situation, VNI 0 is
1381 	 * currently refused.
1382 	 */
1383 	if (!vlan_id)
1384 		return rte_flow_error_set(error, ENOTSUP,
1385 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1386 					  "VXLAN vni cannot be 0");
1387 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1388 		return rte_flow_error_set(error, ENOTSUP,
1389 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1390 					  "VXLAN tunnel must be fully defined");
1391 	return 0;
1392 }
1393 
1394 /**
1395  * Validate VXLAN_GPE item.
1396  *
1397  * @param[in] item
1398  *   Item specification.
1399  * @param[in] item_flags
1400  *   Bit-fields that holds the items detected until now.
1401  * @param[in] priv
1402  *   Pointer to the private data structure.
1403  * @param[in] target_protocol
1404  *   The next protocol in the previous item.
1405  * @param[out] error
1406  *   Pointer to error structure.
1407  *
1408  * @return
1409  *   0 on success, a negative errno value otherwise and rte_errno is set.
1410  */
1411 int
1412 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1413 				  uint64_t item_flags,
1414 				  struct rte_eth_dev *dev,
1415 				  struct rte_flow_error *error)
1416 {
1417 	struct priv *priv = dev->data->dev_private;
1418 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1419 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1420 	int ret;
1421 	union vni {
1422 		uint32_t vlan_id;
1423 		uint8_t vni[4];
1424 	} id = { .vlan_id = 0, };
1425 	uint32_t vlan_id = 0;
1426 
1427 	if (!priv->config.l3_vxlan_en)
1428 		return rte_flow_error_set(error, ENOTSUP,
1429 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1430 					  "L3 VXLAN is not enabled by device"
1431 					  " parameter and/or not configured in"
1432 					  " firmware");
1433 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1434 		return rte_flow_error_set(error, ENOTSUP,
1435 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1436 					  "a tunnel is already present");
1437 	/*
1438 	 * Verify only UDPv4 is present as defined in
1439 	 * https://tools.ietf.org/html/rfc7348
1440 	 */
1441 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1442 		return rte_flow_error_set(error, EINVAL,
1443 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1444 					  "no outer UDP layer found");
1445 	if (!mask)
1446 		mask = &rte_flow_item_vxlan_gpe_mask;
1447 	ret = mlx5_flow_item_acceptable
1448 		(item, (const uint8_t *)mask,
1449 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1450 		 sizeof(struct rte_flow_item_vxlan_gpe),
1451 		 error);
1452 	if (ret < 0)
1453 		return ret;
1454 	if (spec) {
1455 		if (spec->protocol)
1456 			return rte_flow_error_set(error, ENOTSUP,
1457 						  RTE_FLOW_ERROR_TYPE_ITEM,
1458 						  item,
1459 						  "VxLAN-GPE protocol"
1460 						  " not supported");
1461 		memcpy(&id.vni[1], spec->vni, 3);
1462 		vlan_id = id.vlan_id;
1463 		memcpy(&id.vni[1], mask->vni, 3);
1464 		vlan_id &= id.vlan_id;
1465 	}
1466 	/*
1467 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1468 	 * layer is defined in the Verbs specification it is interpreted as
1469 	 * wildcard and all packets will match this rule, if it follows a full
1470 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1471 	 * before will also match this rule.  To avoid such situation, VNI 0
1472 	 * is currently refused.
1473 	 */
1474 	if (!vlan_id)
1475 		return rte_flow_error_set(error, ENOTSUP,
1476 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1477 					  "VXLAN-GPE vni cannot be 0");
1478 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1479 		return rte_flow_error_set(error, ENOTSUP,
1480 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1481 					  "VXLAN-GPE tunnel must be fully"
1482 					  " defined");
1483 	return 0;
1484 }
1485 
1486 /**
1487  * Validate GRE item.
1488  *
1489  * @param[in] item
1490  *   Item specification.
1491  * @param[in] item_flags
1492  *   Bit flags to mark detected items.
1493  * @param[in] target_protocol
1494  *   The next protocol in the previous item.
1495  * @param[out] error
1496  *   Pointer to error structure.
1497  *
1498  * @return
1499  *   0 on success, a negative errno value otherwise and rte_errno is set.
1500  */
1501 int
1502 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1503 			    uint64_t item_flags,
1504 			    uint8_t target_protocol,
1505 			    struct rte_flow_error *error)
1506 {
1507 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1508 	const struct rte_flow_item_gre *mask = item->mask;
1509 	int ret;
1510 
1511 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1512 		return rte_flow_error_set(error, EINVAL,
1513 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1514 					  "protocol filtering not compatible"
1515 					  " with this GRE layer");
1516 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1517 		return rte_flow_error_set(error, ENOTSUP,
1518 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1519 					  "a tunnel is already present");
1520 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1521 		return rte_flow_error_set(error, ENOTSUP,
1522 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1523 					  "L3 Layer is missing");
1524 	if (!mask)
1525 		mask = &rte_flow_item_gre_mask;
1526 	ret = mlx5_flow_item_acceptable
1527 		(item, (const uint8_t *)mask,
1528 		 (const uint8_t *)&rte_flow_item_gre_mask,
1529 		 sizeof(struct rte_flow_item_gre), error);
1530 	if (ret < 0)
1531 		return ret;
1532 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1533 	if (spec && (spec->protocol & mask->protocol))
1534 		return rte_flow_error_set(error, ENOTSUP,
1535 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1536 					  "without MPLS support the"
1537 					  " specification cannot be used for"
1538 					  " filtering");
1539 #endif
1540 	return 0;
1541 }
1542 
1543 /**
1544  * Validate MPLS item.
1545  *
1546  * @param[in] item
1547  *   Item specification.
1548  * @param[in] item_flags
1549  *   Bit-fields that holds the items detected until now.
1550  * @param[in] target_protocol
1551  *   The next protocol in the previous item.
1552  * @param[out] error
1553  *   Pointer to error structure.
1554  *
1555  * @return
1556  *   0 on success, a negative errno value otherwise and rte_errno is set.
1557  */
1558 int
1559 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1560 			     uint64_t item_flags __rte_unused,
1561 			     uint8_t target_protocol __rte_unused,
1562 			     struct rte_flow_error *error)
1563 {
1564 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1565 	const struct rte_flow_item_mpls *mask = item->mask;
1566 	int ret;
1567 
1568 	if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1569 		return rte_flow_error_set(error, EINVAL,
1570 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1571 					  "protocol filtering not compatible"
1572 					  " with MPLS layer");
1573 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1574 		return rte_flow_error_set(error, ENOTSUP,
1575 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1576 					  "a tunnel is already"
1577 					  " present");
1578 	if (!mask)
1579 		mask = &rte_flow_item_mpls_mask;
1580 	ret = mlx5_flow_item_acceptable
1581 		(item, (const uint8_t *)mask,
1582 		 (const uint8_t *)&rte_flow_item_mpls_mask,
1583 		 sizeof(struct rte_flow_item_mpls), error);
1584 	if (ret < 0)
1585 		return ret;
1586 	return 0;
1587 #endif
1588 	return rte_flow_error_set(error, ENOTSUP,
1589 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
1590 				  "MPLS is not supported by Verbs, please"
1591 				  " update.");
1592 }
1593 
1594 static int
1595 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1596 		   const struct rte_flow_attr *attr __rte_unused,
1597 		   const struct rte_flow_item items[] __rte_unused,
1598 		   const struct rte_flow_action actions[] __rte_unused,
1599 		   struct rte_flow_error *error __rte_unused)
1600 {
1601 	rte_errno = ENOTSUP;
1602 	return -rte_errno;
1603 }
1604 
1605 static struct mlx5_flow *
1606 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1607 		  const struct rte_flow_item items[] __rte_unused,
1608 		  const struct rte_flow_action actions[] __rte_unused,
1609 		  uint64_t *item_flags __rte_unused,
1610 		  uint64_t *action_flags __rte_unused,
1611 		  struct rte_flow_error *error __rte_unused)
1612 {
1613 	rte_errno = ENOTSUP;
1614 	return NULL;
1615 }
1616 
1617 static int
1618 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1619 		    struct mlx5_flow *dev_flow __rte_unused,
1620 		    const struct rte_flow_attr *attr __rte_unused,
1621 		    const struct rte_flow_item items[] __rte_unused,
1622 		    const struct rte_flow_action actions[] __rte_unused,
1623 		    struct rte_flow_error *error __rte_unused)
1624 {
1625 	rte_errno = ENOTSUP;
1626 	return -rte_errno;
1627 }
1628 
1629 static int
1630 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1631 		struct rte_flow *flow __rte_unused,
1632 		struct rte_flow_error *error __rte_unused)
1633 {
1634 	rte_errno = ENOTSUP;
1635 	return -rte_errno;
1636 }
1637 
1638 static void
1639 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1640 		 struct rte_flow *flow __rte_unused)
1641 {
1642 }
1643 
1644 static void
1645 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1646 		  struct rte_flow *flow __rte_unused)
1647 {
1648 }
1649 
1650 static int
1651 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1652 		struct rte_flow *flow __rte_unused,
1653 		const struct rte_flow_action *actions __rte_unused,
1654 		void *data __rte_unused,
1655 		struct rte_flow_error *error __rte_unused)
1656 {
1657 	rte_errno = ENOTSUP;
1658 	return -rte_errno;
1659 }
1660 
1661 /* Void driver to protect from null pointer reference. */
1662 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1663 	.validate = flow_null_validate,
1664 	.prepare = flow_null_prepare,
1665 	.translate = flow_null_translate,
1666 	.apply = flow_null_apply,
1667 	.remove = flow_null_remove,
1668 	.destroy = flow_null_destroy,
1669 	.query = flow_null_query,
1670 };
1671 
1672 /**
1673  * Select flow driver type according to flow attributes and device
1674  * configuration.
1675  *
1676  * @param[in] dev
1677  *   Pointer to the dev structure.
1678  * @param[in] attr
1679  *   Pointer to the flow attributes.
1680  *
1681  * @return
1682  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1683  */
1684 static enum mlx5_flow_drv_type
1685 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1686 {
1687 	struct priv *priv = dev->data->dev_private;
1688 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1689 
1690 	if (attr->transfer)
1691 		type = MLX5_FLOW_TYPE_TCF;
1692 	else
1693 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1694 						 MLX5_FLOW_TYPE_VERBS;
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 a flow.
2348  *
2349  * @see rte_flow_query()
2350  * @see rte_flow_ops
2351  */
2352 static int
2353 flow_drv_query(struct rte_eth_dev *dev,
2354 	       struct rte_flow *flow,
2355 	       const struct rte_flow_action *actions,
2356 	       void *data,
2357 	       struct rte_flow_error *error)
2358 {
2359 	const struct mlx5_flow_driver_ops *fops;
2360 	enum mlx5_flow_drv_type ftype = flow->drv_type;
2361 
2362 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2363 	fops = flow_get_drv_ops(ftype);
2364 
2365 	return fops->query(dev, flow, actions, data, error);
2366 }
2367 
2368 /**
2369  * Query a flow.
2370  *
2371  * @see rte_flow_query()
2372  * @see rte_flow_ops
2373  */
2374 int
2375 mlx5_flow_query(struct rte_eth_dev *dev,
2376 		struct rte_flow *flow,
2377 		const struct rte_flow_action *actions,
2378 		void *data,
2379 		struct rte_flow_error *error)
2380 {
2381 	int ret;
2382 
2383 	ret = flow_drv_query(dev, flow, actions, data, error);
2384 	if (ret < 0)
2385 		return ret;
2386 	return 0;
2387 }
2388 
2389 /**
2390  * Convert a flow director filter to a generic flow.
2391  *
2392  * @param dev
2393  *   Pointer to Ethernet device.
2394  * @param fdir_filter
2395  *   Flow director filter to add.
2396  * @param attributes
2397  *   Generic flow parameters structure.
2398  *
2399  * @return
2400  *   0 on success, a negative errno value otherwise and rte_errno is set.
2401  */
2402 static int
2403 mlx5_fdir_filter_convert(struct rte_eth_dev *dev,
2404 			 const struct rte_eth_fdir_filter *fdir_filter,
2405 			 struct mlx5_fdir *attributes)
2406 {
2407 	struct priv *priv = dev->data->dev_private;
2408 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
2409 	const struct rte_eth_fdir_masks *mask =
2410 		&dev->data->dev_conf.fdir_conf.mask;
2411 
2412 	/* Validate queue number. */
2413 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2414 		DRV_LOG(ERR, "port %u invalid queue number %d",
2415 			dev->data->port_id, fdir_filter->action.rx_queue);
2416 		rte_errno = EINVAL;
2417 		return -rte_errno;
2418 	}
2419 	attributes->attr.ingress = 1;
2420 	attributes->items[0] = (struct rte_flow_item) {
2421 		.type = RTE_FLOW_ITEM_TYPE_ETH,
2422 		.spec = &attributes->l2,
2423 		.mask = &attributes->l2_mask,
2424 	};
2425 	switch (fdir_filter->action.behavior) {
2426 	case RTE_ETH_FDIR_ACCEPT:
2427 		attributes->actions[0] = (struct rte_flow_action){
2428 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
2429 			.conf = &attributes->queue,
2430 		};
2431 		break;
2432 	case RTE_ETH_FDIR_REJECT:
2433 		attributes->actions[0] = (struct rte_flow_action){
2434 			.type = RTE_FLOW_ACTION_TYPE_DROP,
2435 		};
2436 		break;
2437 	default:
2438 		DRV_LOG(ERR, "port %u invalid behavior %d",
2439 			dev->data->port_id,
2440 			fdir_filter->action.behavior);
2441 		rte_errno = ENOTSUP;
2442 		return -rte_errno;
2443 	}
2444 	attributes->queue.index = fdir_filter->action.rx_queue;
2445 	/* Handle L3. */
2446 	switch (fdir_filter->input.flow_type) {
2447 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2448 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2449 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2450 		attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2451 			.src_addr = input->flow.ip4_flow.src_ip,
2452 			.dst_addr = input->flow.ip4_flow.dst_ip,
2453 			.time_to_live = input->flow.ip4_flow.ttl,
2454 			.type_of_service = input->flow.ip4_flow.tos,
2455 		};
2456 		attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2457 			.src_addr = mask->ipv4_mask.src_ip,
2458 			.dst_addr = mask->ipv4_mask.dst_ip,
2459 			.time_to_live = mask->ipv4_mask.ttl,
2460 			.type_of_service = mask->ipv4_mask.tos,
2461 			.next_proto_id = mask->ipv4_mask.proto,
2462 		};
2463 		attributes->items[1] = (struct rte_flow_item){
2464 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
2465 			.spec = &attributes->l3,
2466 			.mask = &attributes->l3_mask,
2467 		};
2468 		break;
2469 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2470 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2471 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2472 		attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2473 			.hop_limits = input->flow.ipv6_flow.hop_limits,
2474 			.proto = input->flow.ipv6_flow.proto,
2475 		};
2476 
2477 		memcpy(attributes->l3.ipv6.hdr.src_addr,
2478 		       input->flow.ipv6_flow.src_ip,
2479 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2480 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
2481 		       input->flow.ipv6_flow.dst_ip,
2482 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2483 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2484 		       mask->ipv6_mask.src_ip,
2485 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2486 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2487 		       mask->ipv6_mask.dst_ip,
2488 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2489 		attributes->items[1] = (struct rte_flow_item){
2490 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
2491 			.spec = &attributes->l3,
2492 			.mask = &attributes->l3_mask,
2493 		};
2494 		break;
2495 	default:
2496 		DRV_LOG(ERR, "port %u invalid flow type%d",
2497 			dev->data->port_id, fdir_filter->input.flow_type);
2498 		rte_errno = ENOTSUP;
2499 		return -rte_errno;
2500 	}
2501 	/* Handle L4. */
2502 	switch (fdir_filter->input.flow_type) {
2503 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2504 		attributes->l4.udp.hdr = (struct udp_hdr){
2505 			.src_port = input->flow.udp4_flow.src_port,
2506 			.dst_port = input->flow.udp4_flow.dst_port,
2507 		};
2508 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2509 			.src_port = mask->src_port_mask,
2510 			.dst_port = mask->dst_port_mask,
2511 		};
2512 		attributes->items[2] = (struct rte_flow_item){
2513 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2514 			.spec = &attributes->l4,
2515 			.mask = &attributes->l4_mask,
2516 		};
2517 		break;
2518 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2519 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2520 			.src_port = input->flow.tcp4_flow.src_port,
2521 			.dst_port = input->flow.tcp4_flow.dst_port,
2522 		};
2523 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2524 			.src_port = mask->src_port_mask,
2525 			.dst_port = mask->dst_port_mask,
2526 		};
2527 		attributes->items[2] = (struct rte_flow_item){
2528 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2529 			.spec = &attributes->l4,
2530 			.mask = &attributes->l4_mask,
2531 		};
2532 		break;
2533 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2534 		attributes->l4.udp.hdr = (struct udp_hdr){
2535 			.src_port = input->flow.udp6_flow.src_port,
2536 			.dst_port = input->flow.udp6_flow.dst_port,
2537 		};
2538 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2539 			.src_port = mask->src_port_mask,
2540 			.dst_port = mask->dst_port_mask,
2541 		};
2542 		attributes->items[2] = (struct rte_flow_item){
2543 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2544 			.spec = &attributes->l4,
2545 			.mask = &attributes->l4_mask,
2546 		};
2547 		break;
2548 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2549 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2550 			.src_port = input->flow.tcp6_flow.src_port,
2551 			.dst_port = input->flow.tcp6_flow.dst_port,
2552 		};
2553 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2554 			.src_port = mask->src_port_mask,
2555 			.dst_port = mask->dst_port_mask,
2556 		};
2557 		attributes->items[2] = (struct rte_flow_item){
2558 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2559 			.spec = &attributes->l4,
2560 			.mask = &attributes->l4_mask,
2561 		};
2562 		break;
2563 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2564 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2565 		break;
2566 	default:
2567 		DRV_LOG(ERR, "port %u invalid flow type%d",
2568 			dev->data->port_id, fdir_filter->input.flow_type);
2569 		rte_errno = ENOTSUP;
2570 		return -rte_errno;
2571 	}
2572 	return 0;
2573 }
2574 
2575 /**
2576  * Add new flow director filter and store it in list.
2577  *
2578  * @param dev
2579  *   Pointer to Ethernet device.
2580  * @param fdir_filter
2581  *   Flow director filter to add.
2582  *
2583  * @return
2584  *   0 on success, a negative errno value otherwise and rte_errno is set.
2585  */
2586 static int
2587 mlx5_fdir_filter_add(struct rte_eth_dev *dev,
2588 		     const struct rte_eth_fdir_filter *fdir_filter)
2589 {
2590 	struct priv *priv = dev->data->dev_private;
2591 	struct mlx5_fdir attributes = {
2592 		.attr.group = 0,
2593 		.l2_mask = {
2594 			.dst.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2595 			.src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2596 			.type = 0,
2597 		},
2598 	};
2599 	struct rte_flow_error error;
2600 	struct rte_flow *flow;
2601 	int ret;
2602 
2603 	ret = mlx5_fdir_filter_convert(dev, fdir_filter, &attributes);
2604 	if (ret)
2605 		return ret;
2606 	flow = mlx5_flow_list_create(dev, &priv->flows, &attributes.attr,
2607 				     attributes.items, attributes.actions,
2608 				     &error);
2609 	if (flow) {
2610 		DRV_LOG(DEBUG, "port %u FDIR created %p", dev->data->port_id,
2611 			(void *)flow);
2612 		return 0;
2613 	}
2614 	return -rte_errno;
2615 }
2616 
2617 /**
2618  * Delete specific filter.
2619  *
2620  * @param dev
2621  *   Pointer to Ethernet device.
2622  * @param fdir_filter
2623  *   Filter to be deleted.
2624  *
2625  * @return
2626  *   0 on success, a negative errno value otherwise and rte_errno is set.
2627  */
2628 static int
2629 mlx5_fdir_filter_delete(struct rte_eth_dev *dev __rte_unused,
2630 			const struct rte_eth_fdir_filter *fdir_filter
2631 			__rte_unused)
2632 {
2633 	rte_errno = ENOTSUP;
2634 	return -rte_errno;
2635 }
2636 
2637 /**
2638  * Update queue for specific filter.
2639  *
2640  * @param dev
2641  *   Pointer to Ethernet device.
2642  * @param fdir_filter
2643  *   Filter to be updated.
2644  *
2645  * @return
2646  *   0 on success, a negative errno value otherwise and rte_errno is set.
2647  */
2648 static int
2649 mlx5_fdir_filter_update(struct rte_eth_dev *dev,
2650 			const struct rte_eth_fdir_filter *fdir_filter)
2651 {
2652 	int ret;
2653 
2654 	ret = mlx5_fdir_filter_delete(dev, fdir_filter);
2655 	if (ret)
2656 		return ret;
2657 	return mlx5_fdir_filter_add(dev, fdir_filter);
2658 }
2659 
2660 /**
2661  * Flush all filters.
2662  *
2663  * @param dev
2664  *   Pointer to Ethernet device.
2665  */
2666 static void
2667 mlx5_fdir_filter_flush(struct rte_eth_dev *dev)
2668 {
2669 	struct priv *priv = dev->data->dev_private;
2670 
2671 	mlx5_flow_list_flush(dev, &priv->flows);
2672 }
2673 
2674 /**
2675  * Get flow director information.
2676  *
2677  * @param dev
2678  *   Pointer to Ethernet device.
2679  * @param[out] fdir_info
2680  *   Resulting flow director information.
2681  */
2682 static void
2683 mlx5_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2684 {
2685 	struct rte_eth_fdir_masks *mask =
2686 		&dev->data->dev_conf.fdir_conf.mask;
2687 
2688 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2689 	fdir_info->guarant_spc = 0;
2690 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2691 	fdir_info->max_flexpayload = 0;
2692 	fdir_info->flow_types_mask[0] = 0;
2693 	fdir_info->flex_payload_unit = 0;
2694 	fdir_info->max_flex_payload_segment_num = 0;
2695 	fdir_info->flex_payload_limit = 0;
2696 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2697 }
2698 
2699 /**
2700  * Deal with flow director operations.
2701  *
2702  * @param dev
2703  *   Pointer to Ethernet device.
2704  * @param filter_op
2705  *   Operation to perform.
2706  * @param arg
2707  *   Pointer to operation-specific structure.
2708  *
2709  * @return
2710  *   0 on success, a negative errno value otherwise and rte_errno is set.
2711  */
2712 static int
2713 mlx5_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2714 		    void *arg)
2715 {
2716 	enum rte_fdir_mode fdir_mode =
2717 		dev->data->dev_conf.fdir_conf.mode;
2718 
2719 	if (filter_op == RTE_ETH_FILTER_NOP)
2720 		return 0;
2721 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2722 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2723 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
2724 			dev->data->port_id, fdir_mode);
2725 		rte_errno = EINVAL;
2726 		return -rte_errno;
2727 	}
2728 	switch (filter_op) {
2729 	case RTE_ETH_FILTER_ADD:
2730 		return mlx5_fdir_filter_add(dev, arg);
2731 	case RTE_ETH_FILTER_UPDATE:
2732 		return mlx5_fdir_filter_update(dev, arg);
2733 	case RTE_ETH_FILTER_DELETE:
2734 		return mlx5_fdir_filter_delete(dev, arg);
2735 	case RTE_ETH_FILTER_FLUSH:
2736 		mlx5_fdir_filter_flush(dev);
2737 		break;
2738 	case RTE_ETH_FILTER_INFO:
2739 		mlx5_fdir_info_get(dev, arg);
2740 		break;
2741 	default:
2742 		DRV_LOG(DEBUG, "port %u unknown operation %u",
2743 			dev->data->port_id, filter_op);
2744 		rte_errno = EINVAL;
2745 		return -rte_errno;
2746 	}
2747 	return 0;
2748 }
2749 
2750 /**
2751  * Manage filter operations.
2752  *
2753  * @param dev
2754  *   Pointer to Ethernet device structure.
2755  * @param filter_type
2756  *   Filter type.
2757  * @param filter_op
2758  *   Operation to perform.
2759  * @param arg
2760  *   Pointer to operation-specific structure.
2761  *
2762  * @return
2763  *   0 on success, a negative errno value otherwise and rte_errno is set.
2764  */
2765 int
2766 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2767 		     enum rte_filter_type filter_type,
2768 		     enum rte_filter_op filter_op,
2769 		     void *arg)
2770 {
2771 	switch (filter_type) {
2772 	case RTE_ETH_FILTER_GENERIC:
2773 		if (filter_op != RTE_ETH_FILTER_GET) {
2774 			rte_errno = EINVAL;
2775 			return -rte_errno;
2776 		}
2777 		*(const void **)arg = &mlx5_flow_ops;
2778 		return 0;
2779 	case RTE_ETH_FILTER_FDIR:
2780 		return mlx5_fdir_ctrl_func(dev, filter_op, arg);
2781 	default:
2782 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
2783 			dev->data->port_id, filter_type);
2784 		rte_errno = ENOTSUP;
2785 		return -rte_errno;
2786 	}
2787 	return 0;
2788 }
2789