xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 4f1ed78ebd26f2393fd3cf29a9a9fa95ce14eb44)
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, uint64_t layer_types,
481 			    uint64_t hash_fields)
482 {
483 	struct rte_flow *flow = dev_flow->flow;
484 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
485 	int rss_request_inner = flow->rss.level >= 2;
486 
487 	/* Check RSS hash level for tunnel. */
488 	if (tunnel && rss_request_inner)
489 		hash_fields |= IBV_RX_HASH_INNER;
490 	else if (tunnel || rss_request_inner)
491 		return 0;
492 #endif
493 	/* Check if requested layer matches RSS hash fields. */
494 	if (!(flow->rss.types & layer_types))
495 		return 0;
496 	return hash_fields;
497 }
498 
499 /**
500  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
501  * if several tunnel rules are used on this queue, the tunnel ptype will be
502  * cleared.
503  *
504  * @param rxq_ctrl
505  *   Rx queue to update.
506  */
507 static void
508 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
509 {
510 	unsigned int i;
511 	uint32_t tunnel_ptype = 0;
512 
513 	/* Look up for the ptype to use. */
514 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
515 		if (!rxq_ctrl->flow_tunnels_n[i])
516 			continue;
517 		if (!tunnel_ptype) {
518 			tunnel_ptype = tunnels_info[i].ptype;
519 		} else {
520 			tunnel_ptype = 0;
521 			break;
522 		}
523 	}
524 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
525 }
526 
527 /**
528  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
529  * flow.
530  *
531  * @param[in] dev
532  *   Pointer to the Ethernet device structure.
533  * @param[in] dev_flow
534  *   Pointer to device flow structure.
535  */
536 static void
537 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
538 {
539 	struct priv *priv = dev->data->dev_private;
540 	struct rte_flow *flow = dev_flow->flow;
541 	const int mark = !!(flow->actions &
542 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
543 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
544 	unsigned int i;
545 
546 	for (i = 0; i != flow->rss.queue_num; ++i) {
547 		int idx = (*flow->queue)[i];
548 		struct mlx5_rxq_ctrl *rxq_ctrl =
549 			container_of((*priv->rxqs)[idx],
550 				     struct mlx5_rxq_ctrl, rxq);
551 
552 		if (mark) {
553 			rxq_ctrl->rxq.mark = 1;
554 			rxq_ctrl->flow_mark_n++;
555 		}
556 		if (tunnel) {
557 			unsigned int j;
558 
559 			/* Increase the counter matching the flow. */
560 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
561 				if ((tunnels_info[j].tunnel &
562 				     dev_flow->layers) ==
563 				    tunnels_info[j].tunnel) {
564 					rxq_ctrl->flow_tunnels_n[j]++;
565 					break;
566 				}
567 			}
568 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
569 		}
570 	}
571 }
572 
573 /**
574  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
575  *
576  * @param[in] dev
577  *   Pointer to the Ethernet device structure.
578  * @param[in] flow
579  *   Pointer to flow structure.
580  */
581 static void
582 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
583 {
584 	struct mlx5_flow *dev_flow;
585 
586 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
587 		flow_drv_rxq_flags_set(dev, dev_flow);
588 }
589 
590 /**
591  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
592  * device flow if no other flow uses it with the same kind of request.
593  *
594  * @param dev
595  *   Pointer to Ethernet device.
596  * @param[in] dev_flow
597  *   Pointer to the device flow.
598  */
599 static void
600 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
601 {
602 	struct priv *priv = dev->data->dev_private;
603 	struct rte_flow *flow = dev_flow->flow;
604 	const int mark = !!(flow->actions &
605 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
606 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
607 	unsigned int i;
608 
609 	assert(dev->data->dev_started);
610 	for (i = 0; i != flow->rss.queue_num; ++i) {
611 		int idx = (*flow->queue)[i];
612 		struct mlx5_rxq_ctrl *rxq_ctrl =
613 			container_of((*priv->rxqs)[idx],
614 				     struct mlx5_rxq_ctrl, rxq);
615 
616 		if (mark) {
617 			rxq_ctrl->flow_mark_n--;
618 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
619 		}
620 		if (tunnel) {
621 			unsigned int j;
622 
623 			/* Decrease the counter matching the flow. */
624 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
625 				if ((tunnels_info[j].tunnel &
626 				     dev_flow->layers) ==
627 				    tunnels_info[j].tunnel) {
628 					rxq_ctrl->flow_tunnels_n[j]--;
629 					break;
630 				}
631 			}
632 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
633 		}
634 	}
635 }
636 
637 /**
638  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
639  * @p flow if no other flow uses it with the same kind of request.
640  *
641  * @param dev
642  *   Pointer to Ethernet device.
643  * @param[in] flow
644  *   Pointer to the flow.
645  */
646 static void
647 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
648 {
649 	struct mlx5_flow *dev_flow;
650 
651 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
652 		flow_drv_rxq_flags_trim(dev, dev_flow);
653 }
654 
655 /**
656  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
657  *
658  * @param dev
659  *   Pointer to Ethernet device.
660  */
661 static void
662 flow_rxq_flags_clear(struct rte_eth_dev *dev)
663 {
664 	struct priv *priv = dev->data->dev_private;
665 	unsigned int i;
666 
667 	for (i = 0; i != priv->rxqs_n; ++i) {
668 		struct mlx5_rxq_ctrl *rxq_ctrl;
669 		unsigned int j;
670 
671 		if (!(*priv->rxqs)[i])
672 			continue;
673 		rxq_ctrl = container_of((*priv->rxqs)[i],
674 					struct mlx5_rxq_ctrl, rxq);
675 		rxq_ctrl->flow_mark_n = 0;
676 		rxq_ctrl->rxq.mark = 0;
677 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
678 			rxq_ctrl->flow_tunnels_n[j] = 0;
679 		rxq_ctrl->rxq.tunnel = 0;
680 	}
681 }
682 
683 /*
684  * Validate the flag action.
685  *
686  * @param[in] action_flags
687  *   Bit-fields that holds the actions detected until now.
688  * @param[in] attr
689  *   Attributes of flow that includes this action.
690  * @param[out] error
691  *   Pointer to error structure.
692  *
693  * @return
694  *   0 on success, a negative errno value otherwise and rte_errno is set.
695  */
696 int
697 mlx5_flow_validate_action_flag(uint64_t action_flags,
698 			       const struct rte_flow_attr *attr,
699 			       struct rte_flow_error *error)
700 {
701 
702 	if (action_flags & MLX5_FLOW_ACTION_DROP)
703 		return rte_flow_error_set(error, EINVAL,
704 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
705 					  "can't drop and flag in same flow");
706 	if (action_flags & MLX5_FLOW_ACTION_MARK)
707 		return rte_flow_error_set(error, EINVAL,
708 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
709 					  "can't mark and flag in same flow");
710 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
711 		return rte_flow_error_set(error, EINVAL,
712 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
713 					  "can't have 2 flag"
714 					  " actions in same flow");
715 	if (attr->egress)
716 		return rte_flow_error_set(error, ENOTSUP,
717 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
718 					  "flag action not supported for "
719 					  "egress");
720 	return 0;
721 }
722 
723 /*
724  * Validate the mark action.
725  *
726  * @param[in] action
727  *   Pointer to the queue action.
728  * @param[in] action_flags
729  *   Bit-fields that holds the actions detected until now.
730  * @param[in] attr
731  *   Attributes of flow that includes this action.
732  * @param[out] error
733  *   Pointer to error structure.
734  *
735  * @return
736  *   0 on success, a negative errno value otherwise and rte_errno is set.
737  */
738 int
739 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
740 			       uint64_t action_flags,
741 			       const struct rte_flow_attr *attr,
742 			       struct rte_flow_error *error)
743 {
744 	const struct rte_flow_action_mark *mark = action->conf;
745 
746 	if (!mark)
747 		return rte_flow_error_set(error, EINVAL,
748 					  RTE_FLOW_ERROR_TYPE_ACTION,
749 					  action,
750 					  "configuration cannot be null");
751 	if (mark->id >= MLX5_FLOW_MARK_MAX)
752 		return rte_flow_error_set(error, EINVAL,
753 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
754 					  &mark->id,
755 					  "mark id must in 0 <= id < "
756 					  RTE_STR(MLX5_FLOW_MARK_MAX));
757 	if (action_flags & MLX5_FLOW_ACTION_DROP)
758 		return rte_flow_error_set(error, EINVAL,
759 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
760 					  "can't drop and mark in same flow");
761 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
762 		return rte_flow_error_set(error, EINVAL,
763 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
764 					  "can't flag and mark in same flow");
765 	if (action_flags & MLX5_FLOW_ACTION_MARK)
766 		return rte_flow_error_set(error, EINVAL,
767 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
768 					  "can't have 2 mark actions in same"
769 					  " flow");
770 	if (attr->egress)
771 		return rte_flow_error_set(error, ENOTSUP,
772 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
773 					  "mark action not supported for "
774 					  "egress");
775 	return 0;
776 }
777 
778 /*
779  * Validate the drop action.
780  *
781  * @param[in] action_flags
782  *   Bit-fields that holds the actions detected until now.
783  * @param[in] attr
784  *   Attributes of flow that includes this action.
785  * @param[out] error
786  *   Pointer to error structure.
787  *
788  * @return
789  *   0 on success, a negative errno value otherwise and rte_ernno is set.
790  */
791 int
792 mlx5_flow_validate_action_drop(uint64_t action_flags,
793 			       const struct rte_flow_attr *attr,
794 			       struct rte_flow_error *error)
795 {
796 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
797 		return rte_flow_error_set(error, EINVAL,
798 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
799 					  "can't drop and flag in same flow");
800 	if (action_flags & MLX5_FLOW_ACTION_MARK)
801 		return rte_flow_error_set(error, EINVAL,
802 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
803 					  "can't drop and mark in same flow");
804 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
805 		return rte_flow_error_set(error, EINVAL,
806 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
807 					  "can't have 2 fate actions in"
808 					  " same flow");
809 	if (attr->egress)
810 		return rte_flow_error_set(error, ENOTSUP,
811 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
812 					  "drop action not supported for "
813 					  "egress");
814 	return 0;
815 }
816 
817 /*
818  * Validate the queue action.
819  *
820  * @param[in] action
821  *   Pointer to the queue action.
822  * @param[in] action_flags
823  *   Bit-fields that holds the actions detected until now.
824  * @param[in] dev
825  *   Pointer to the Ethernet device structure.
826  * @param[in] attr
827  *   Attributes of flow that includes this action.
828  * @param[out] error
829  *   Pointer to error structure.
830  *
831  * @return
832  *   0 on success, a negative errno value otherwise and rte_ernno is set.
833  */
834 int
835 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
836 				uint64_t action_flags,
837 				struct rte_eth_dev *dev,
838 				const struct rte_flow_attr *attr,
839 				struct rte_flow_error *error)
840 {
841 	struct priv *priv = dev->data->dev_private;
842 	const struct rte_flow_action_queue *queue = action->conf;
843 
844 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
845 		return rte_flow_error_set(error, EINVAL,
846 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
847 					  "can't have 2 fate actions in"
848 					  " same flow");
849 	if (queue->index >= priv->rxqs_n)
850 		return rte_flow_error_set(error, EINVAL,
851 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
852 					  &queue->index,
853 					  "queue index out of range");
854 	if (!(*priv->rxqs)[queue->index])
855 		return rte_flow_error_set(error, EINVAL,
856 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
857 					  &queue->index,
858 					  "queue is not configured");
859 	if (attr->egress)
860 		return rte_flow_error_set(error, ENOTSUP,
861 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
862 					  "queue action not supported for "
863 					  "egress");
864 	return 0;
865 }
866 
867 /*
868  * Validate the rss action.
869  *
870  * @param[in] action
871  *   Pointer to the queue action.
872  * @param[in] action_flags
873  *   Bit-fields that holds the actions detected until now.
874  * @param[in] dev
875  *   Pointer to the Ethernet device structure.
876  * @param[in] attr
877  *   Attributes of flow that includes this action.
878  * @param[out] error
879  *   Pointer to error structure.
880  *
881  * @return
882  *   0 on success, a negative errno value otherwise and rte_ernno is set.
883  */
884 int
885 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
886 			      uint64_t action_flags,
887 			      struct rte_eth_dev *dev,
888 			      const struct rte_flow_attr *attr,
889 			      struct rte_flow_error *error)
890 {
891 	struct priv *priv = dev->data->dev_private;
892 	const struct rte_flow_action_rss *rss = action->conf;
893 	unsigned int i;
894 
895 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
896 		return rte_flow_error_set(error, EINVAL,
897 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
898 					  "can't have 2 fate actions"
899 					  " in same flow");
900 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
901 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
902 		return rte_flow_error_set(error, ENOTSUP,
903 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
904 					  &rss->func,
905 					  "RSS hash function not supported");
906 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
907 	if (rss->level > 2)
908 #else
909 	if (rss->level > 1)
910 #endif
911 		return rte_flow_error_set(error, ENOTSUP,
912 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
913 					  &rss->level,
914 					  "tunnel RSS is not supported");
915 	if (rss->key_len < MLX5_RSS_HASH_KEY_LEN)
916 		return rte_flow_error_set(error, ENOTSUP,
917 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
918 					  &rss->key_len,
919 					  "RSS hash key too small");
920 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
921 		return rte_flow_error_set(error, ENOTSUP,
922 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
923 					  &rss->key_len,
924 					  "RSS hash key too large");
925 	if (rss->queue_num > priv->config.ind_table_max_size)
926 		return rte_flow_error_set(error, ENOTSUP,
927 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
928 					  &rss->queue_num,
929 					  "number of queues too large");
930 	if (rss->types & MLX5_RSS_HF_MASK)
931 		return rte_flow_error_set(error, ENOTSUP,
932 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
933 					  &rss->types,
934 					  "some RSS protocols are not"
935 					  " supported");
936 	for (i = 0; i != rss->queue_num; ++i) {
937 		if (!(*priv->rxqs)[rss->queue[i]])
938 			return rte_flow_error_set
939 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
940 				 &rss->queue[i], "queue is not configured");
941 	}
942 	if (attr->egress)
943 		return rte_flow_error_set(error, ENOTSUP,
944 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
945 					  "rss action not supported for "
946 					  "egress");
947 	return 0;
948 }
949 
950 /*
951  * Validate the count action.
952  *
953  * @param[in] dev
954  *   Pointer to the Ethernet device structure.
955  * @param[in] attr
956  *   Attributes of flow that includes this action.
957  * @param[out] error
958  *   Pointer to error structure.
959  *
960  * @return
961  *   0 on success, a negative errno value otherwise and rte_ernno is set.
962  */
963 int
964 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
965 				const struct rte_flow_attr *attr,
966 				struct rte_flow_error *error)
967 {
968 	if (attr->egress)
969 		return rte_flow_error_set(error, ENOTSUP,
970 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
971 					  "count action not supported for "
972 					  "egress");
973 	return 0;
974 }
975 
976 /**
977  * Verify the @p attributes will be correctly understood by the NIC and store
978  * them in the @p flow if everything is correct.
979  *
980  * @param[in] dev
981  *   Pointer to the Ethernet device structure.
982  * @param[in] attributes
983  *   Pointer to flow attributes
984  * @param[out] error
985  *   Pointer to error structure.
986  *
987  * @return
988  *   0 on success, a negative errno value otherwise and rte_errno is set.
989  */
990 int
991 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
992 			      const struct rte_flow_attr *attributes,
993 			      struct rte_flow_error *error)
994 {
995 	struct priv *priv = dev->data->dev_private;
996 	uint32_t priority_max = priv->config.flow_prio - 1;
997 
998 	if (attributes->group)
999 		return rte_flow_error_set(error, ENOTSUP,
1000 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1001 					  NULL, "groups is not supported");
1002 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1003 	    attributes->priority >= priority_max)
1004 		return rte_flow_error_set(error, ENOTSUP,
1005 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1006 					  NULL, "priority out of range");
1007 	if (attributes->egress)
1008 		return rte_flow_error_set(error, ENOTSUP,
1009 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1010 					  "egress is not supported");
1011 	if (attributes->transfer)
1012 		return rte_flow_error_set(error, ENOTSUP,
1013 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1014 					  NULL, "transfer is not supported");
1015 	if (!attributes->ingress)
1016 		return rte_flow_error_set(error, EINVAL,
1017 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1018 					  NULL,
1019 					  "ingress attribute is mandatory");
1020 	return 0;
1021 }
1022 
1023 /**
1024  * Validate Ethernet item.
1025  *
1026  * @param[in] item
1027  *   Item specification.
1028  * @param[in] item_flags
1029  *   Bit-fields that holds the items detected until now.
1030  * @param[out] error
1031  *   Pointer to error structure.
1032  *
1033  * @return
1034  *   0 on success, a negative errno value otherwise and rte_errno is set.
1035  */
1036 int
1037 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1038 			    uint64_t item_flags,
1039 			    struct rte_flow_error *error)
1040 {
1041 	const struct rte_flow_item_eth *mask = item->mask;
1042 	const struct rte_flow_item_eth nic_mask = {
1043 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1044 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1045 		.type = RTE_BE16(0xffff),
1046 	};
1047 	int ret;
1048 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1049 
1050 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L2)
1051 		return rte_flow_error_set(error, ENOTSUP,
1052 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1053 					  "3 levels of l2 are not supported");
1054 	if ((item_flags & MLX5_FLOW_LAYER_INNER_L2) && !tunnel)
1055 		return rte_flow_error_set(error, ENOTSUP,
1056 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1057 					  "2 L2 without tunnel are not supported");
1058 	if (!mask)
1059 		mask = &rte_flow_item_eth_mask;
1060 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1061 					(const uint8_t *)&nic_mask,
1062 					sizeof(struct rte_flow_item_eth),
1063 					error);
1064 	return ret;
1065 }
1066 
1067 /**
1068  * Validate VLAN item.
1069  *
1070  * @param[in] item
1071  *   Item specification.
1072  * @param[in] item_flags
1073  *   Bit-fields that holds the items detected until now.
1074  * @param[out] error
1075  *   Pointer to error structure.
1076  *
1077  * @return
1078  *   0 on success, a negative errno value otherwise and rte_errno is set.
1079  */
1080 int
1081 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1082 			     int64_t item_flags,
1083 			     struct rte_flow_error *error)
1084 {
1085 	const struct rte_flow_item_vlan *spec = item->spec;
1086 	const struct rte_flow_item_vlan *mask = item->mask;
1087 	const struct rte_flow_item_vlan nic_mask = {
1088 		.tci = RTE_BE16(0x0fff),
1089 		.inner_type = RTE_BE16(0xffff),
1090 	};
1091 	uint16_t vlan_tag = 0;
1092 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1093 	int ret;
1094 	const uint32_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1095 					MLX5_FLOW_LAYER_INNER_L4) :
1096 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1097 					MLX5_FLOW_LAYER_OUTER_L4);
1098 	const uint32_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1099 					MLX5_FLOW_LAYER_OUTER_VLAN;
1100 
1101 	if (item_flags & vlanm)
1102 		return rte_flow_error_set(error, EINVAL,
1103 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1104 					  "VLAN layer already configured");
1105 	else if ((item_flags & l34m) != 0)
1106 		return rte_flow_error_set(error, EINVAL,
1107 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1108 					  "L2 layer cannot follow L3/L4 layer");
1109 	if (!mask)
1110 		mask = &rte_flow_item_vlan_mask;
1111 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1112 					(const uint8_t *)&nic_mask,
1113 					sizeof(struct rte_flow_item_vlan),
1114 					error);
1115 	if (ret)
1116 		return ret;
1117 	if (spec) {
1118 		vlan_tag = spec->tci;
1119 		vlan_tag &= mask->tci;
1120 	}
1121 	/*
1122 	 * From verbs perspective an empty VLAN is equivalent
1123 	 * to a packet without VLAN layer.
1124 	 */
1125 	if (!vlan_tag)
1126 		return rte_flow_error_set(error, EINVAL,
1127 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1128 					  item->spec,
1129 					  "VLAN cannot be empty");
1130 	return 0;
1131 }
1132 
1133 /**
1134  * Validate IPV4 item.
1135  *
1136  * @param[in] item
1137  *   Item specification.
1138  * @param[in] item_flags
1139  *   Bit-fields that holds the items detected until now.
1140  * @param[out] error
1141  *   Pointer to error structure.
1142  *
1143  * @return
1144  *   0 on success, a negative errno value otherwise and rte_errno is set.
1145  */
1146 int
1147 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1148 			     int64_t item_flags,
1149 			     struct rte_flow_error *error)
1150 {
1151 	const struct rte_flow_item_ipv4 *mask = item->mask;
1152 	const struct rte_flow_item_ipv4 nic_mask = {
1153 		.hdr = {
1154 			.src_addr = RTE_BE32(0xffffffff),
1155 			.dst_addr = RTE_BE32(0xffffffff),
1156 			.type_of_service = 0xff,
1157 			.next_proto_id = 0xff,
1158 		},
1159 	};
1160 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1161 	int ret;
1162 
1163 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1164 				   MLX5_FLOW_LAYER_OUTER_L3))
1165 		return rte_flow_error_set(error, ENOTSUP,
1166 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1167 					  "multiple L3 layers not supported");
1168 	else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1169 					MLX5_FLOW_LAYER_OUTER_L4))
1170 		return rte_flow_error_set(error, EINVAL,
1171 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1172 					  "L3 cannot follow an L4 layer.");
1173 	if (!mask)
1174 		mask = &rte_flow_item_ipv4_mask;
1175 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1176 					(const uint8_t *)&nic_mask,
1177 					sizeof(struct rte_flow_item_ipv4),
1178 					error);
1179 	if (ret < 0)
1180 		return ret;
1181 	return 0;
1182 }
1183 
1184 /**
1185  * Validate IPV6 item.
1186  *
1187  * @param[in] item
1188  *   Item specification.
1189  * @param[in] item_flags
1190  *   Bit-fields that holds the items detected until now.
1191  * @param[out] error
1192  *   Pointer to error structure.
1193  *
1194  * @return
1195  *   0 on success, a negative errno value otherwise and rte_errno is set.
1196  */
1197 int
1198 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1199 			     uint64_t item_flags,
1200 			     struct rte_flow_error *error)
1201 {
1202 	const struct rte_flow_item_ipv6 *mask = item->mask;
1203 	const struct rte_flow_item_ipv6 nic_mask = {
1204 		.hdr = {
1205 			.src_addr =
1206 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1207 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1208 			.dst_addr =
1209 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1210 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1211 			.vtc_flow = RTE_BE32(0xffffffff),
1212 			.proto = 0xff,
1213 			.hop_limits = 0xff,
1214 		},
1215 	};
1216 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1217 	int ret;
1218 
1219 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1220 				   MLX5_FLOW_LAYER_OUTER_L3))
1221 		return rte_flow_error_set(error, ENOTSUP,
1222 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1223 					  "multiple L3 layers not supported");
1224 	else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1225 					MLX5_FLOW_LAYER_OUTER_L4))
1226 		return rte_flow_error_set(error, EINVAL,
1227 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1228 					  "L3 cannot follow an L4 layer.");
1229 	/*
1230 	 * IPv6 is not recognised by the NIC inside a GRE tunnel.
1231 	 * Such support has to be disabled as the rule will be
1232 	 * accepted.  Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1233 	 * Mellanox OFED 4.4-1.0.0.0.
1234 	 */
1235 	if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1236 		return rte_flow_error_set(error, ENOTSUP,
1237 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1238 					  "IPv6 inside a GRE tunnel is"
1239 					  " not recognised.");
1240 	if (!mask)
1241 		mask = &rte_flow_item_ipv6_mask;
1242 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1243 					(const uint8_t *)&nic_mask,
1244 					sizeof(struct rte_flow_item_ipv6),
1245 					error);
1246 	if (ret < 0)
1247 		return ret;
1248 	return 0;
1249 }
1250 
1251 /**
1252  * Validate UDP item.
1253  *
1254  * @param[in] item
1255  *   Item specification.
1256  * @param[in] item_flags
1257  *   Bit-fields that holds the items detected until now.
1258  * @param[in] target_protocol
1259  *   The next protocol in the previous item.
1260  * @param[in] flow_mask
1261  *   mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1262  * @param[out] error
1263  *   Pointer to error structure.
1264  *
1265  * @return
1266  *   0 on success, a negative errno value otherwise and rte_errno is set.
1267  */
1268 int
1269 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1270 			    uint64_t item_flags,
1271 			    uint8_t target_protocol,
1272 			    struct rte_flow_error *error)
1273 {
1274 	const struct rte_flow_item_udp *mask = item->mask;
1275 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1276 	int ret;
1277 
1278 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1279 		return rte_flow_error_set(error, EINVAL,
1280 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1281 					  "protocol filtering not compatible"
1282 					  " with UDP layer");
1283 	if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1284 				     MLX5_FLOW_LAYER_OUTER_L3)))
1285 		return rte_flow_error_set(error, EINVAL,
1286 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1287 					  "L3 is mandatory to filter on L4");
1288 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1289 				   MLX5_FLOW_LAYER_OUTER_L4))
1290 		return rte_flow_error_set(error, EINVAL,
1291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1292 					  "L4 layer is already present");
1293 	if (!mask)
1294 		mask = &rte_flow_item_udp_mask;
1295 	ret = mlx5_flow_item_acceptable
1296 		(item, (const uint8_t *)mask,
1297 		 (const uint8_t *)&rte_flow_item_udp_mask,
1298 		 sizeof(struct rte_flow_item_udp), error);
1299 	if (ret < 0)
1300 		return ret;
1301 	return 0;
1302 }
1303 
1304 /**
1305  * Validate TCP item.
1306  *
1307  * @param[in] item
1308  *   Item specification.
1309  * @param[in] item_flags
1310  *   Bit-fields that holds the items detected until now.
1311  * @param[in] target_protocol
1312  *   The next protocol in the previous item.
1313  * @param[out] error
1314  *   Pointer to error structure.
1315  *
1316  * @return
1317  *   0 on success, a negative errno value otherwise and rte_errno is set.
1318  */
1319 int
1320 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1321 			    uint64_t item_flags,
1322 			    uint8_t target_protocol,
1323 			    const struct rte_flow_item_tcp *flow_mask,
1324 			    struct rte_flow_error *error)
1325 {
1326 	const struct rte_flow_item_tcp *mask = item->mask;
1327 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1328 	int ret;
1329 
1330 	assert(flow_mask);
1331 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1332 		return rte_flow_error_set(error, EINVAL,
1333 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1334 					  "protocol filtering not compatible"
1335 					  " with TCP layer");
1336 	if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1337 				     MLX5_FLOW_LAYER_OUTER_L3)))
1338 		return rte_flow_error_set(error, EINVAL,
1339 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1340 					  "L3 is mandatory to filter on L4");
1341 	if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1342 				   MLX5_FLOW_LAYER_OUTER_L4))
1343 		return rte_flow_error_set(error, EINVAL,
1344 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1345 					  "L4 layer is already present");
1346 	if (!mask)
1347 		mask = &rte_flow_item_tcp_mask;
1348 	ret = mlx5_flow_item_acceptable
1349 		(item, (const uint8_t *)mask,
1350 		 (const uint8_t *)flow_mask,
1351 		 sizeof(struct rte_flow_item_tcp), error);
1352 	if (ret < 0)
1353 		return ret;
1354 	return 0;
1355 }
1356 
1357 /**
1358  * Validate VXLAN item.
1359  *
1360  * @param[in] item
1361  *   Item specification.
1362  * @param[in] item_flags
1363  *   Bit-fields that holds the items detected until now.
1364  * @param[in] target_protocol
1365  *   The next protocol in the previous item.
1366  * @param[out] error
1367  *   Pointer to error structure.
1368  *
1369  * @return
1370  *   0 on success, a negative errno value otherwise and rte_errno is set.
1371  */
1372 int
1373 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1374 			      uint64_t item_flags,
1375 			      struct rte_flow_error *error)
1376 {
1377 	const struct rte_flow_item_vxlan *spec = item->spec;
1378 	const struct rte_flow_item_vxlan *mask = item->mask;
1379 	int ret;
1380 	union vni {
1381 		uint32_t vlan_id;
1382 		uint8_t vni[4];
1383 	} id = { .vlan_id = 0, };
1384 	uint32_t vlan_id = 0;
1385 
1386 
1387 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1388 		return rte_flow_error_set(error, ENOTSUP,
1389 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1390 					  "a tunnel is already present");
1391 	/*
1392 	 * Verify only UDPv4 is present as defined in
1393 	 * https://tools.ietf.org/html/rfc7348
1394 	 */
1395 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1396 		return rte_flow_error_set(error, EINVAL,
1397 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1398 					  "no outer UDP layer found");
1399 	if (!mask)
1400 		mask = &rte_flow_item_vxlan_mask;
1401 	ret = mlx5_flow_item_acceptable
1402 		(item, (const uint8_t *)mask,
1403 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1404 		 sizeof(struct rte_flow_item_vxlan),
1405 		 error);
1406 	if (ret < 0)
1407 		return ret;
1408 	if (spec) {
1409 		memcpy(&id.vni[1], spec->vni, 3);
1410 		vlan_id = id.vlan_id;
1411 		memcpy(&id.vni[1], mask->vni, 3);
1412 		vlan_id &= id.vlan_id;
1413 	}
1414 	/*
1415 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1416 	 * only this layer is defined in the Verbs specification it is
1417 	 * interpreted as wildcard and all packets will match this
1418 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1419 	 * udp), all packets matching the layers before will also
1420 	 * match this rule.  To avoid such situation, VNI 0 is
1421 	 * currently refused.
1422 	 */
1423 	if (!vlan_id)
1424 		return rte_flow_error_set(error, ENOTSUP,
1425 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1426 					  "VXLAN vni cannot be 0");
1427 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1428 		return rte_flow_error_set(error, ENOTSUP,
1429 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1430 					  "VXLAN tunnel must be fully defined");
1431 	return 0;
1432 }
1433 
1434 /**
1435  * Validate VXLAN_GPE item.
1436  *
1437  * @param[in] item
1438  *   Item specification.
1439  * @param[in] item_flags
1440  *   Bit-fields that holds the items detected until now.
1441  * @param[in] priv
1442  *   Pointer to the private data structure.
1443  * @param[in] target_protocol
1444  *   The next protocol in the previous item.
1445  * @param[out] error
1446  *   Pointer to error structure.
1447  *
1448  * @return
1449  *   0 on success, a negative errno value otherwise and rte_errno is set.
1450  */
1451 int
1452 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1453 				  uint64_t item_flags,
1454 				  struct rte_eth_dev *dev,
1455 				  struct rte_flow_error *error)
1456 {
1457 	struct priv *priv = dev->data->dev_private;
1458 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1459 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1460 	int ret;
1461 	union vni {
1462 		uint32_t vlan_id;
1463 		uint8_t vni[4];
1464 	} id = { .vlan_id = 0, };
1465 	uint32_t vlan_id = 0;
1466 
1467 	if (!priv->config.l3_vxlan_en)
1468 		return rte_flow_error_set(error, ENOTSUP,
1469 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1470 					  "L3 VXLAN is not enabled by device"
1471 					  " parameter and/or not configured in"
1472 					  " firmware");
1473 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1474 		return rte_flow_error_set(error, ENOTSUP,
1475 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1476 					  "a tunnel is already present");
1477 	/*
1478 	 * Verify only UDPv4 is present as defined in
1479 	 * https://tools.ietf.org/html/rfc7348
1480 	 */
1481 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1482 		return rte_flow_error_set(error, EINVAL,
1483 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1484 					  "no outer UDP layer found");
1485 	if (!mask)
1486 		mask = &rte_flow_item_vxlan_gpe_mask;
1487 	ret = mlx5_flow_item_acceptable
1488 		(item, (const uint8_t *)mask,
1489 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1490 		 sizeof(struct rte_flow_item_vxlan_gpe),
1491 		 error);
1492 	if (ret < 0)
1493 		return ret;
1494 	if (spec) {
1495 		if (spec->protocol)
1496 			return rte_flow_error_set(error, ENOTSUP,
1497 						  RTE_FLOW_ERROR_TYPE_ITEM,
1498 						  item,
1499 						  "VxLAN-GPE protocol"
1500 						  " not supported");
1501 		memcpy(&id.vni[1], spec->vni, 3);
1502 		vlan_id = id.vlan_id;
1503 		memcpy(&id.vni[1], mask->vni, 3);
1504 		vlan_id &= id.vlan_id;
1505 	}
1506 	/*
1507 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1508 	 * layer is defined in the Verbs specification it is interpreted as
1509 	 * wildcard and all packets will match this rule, if it follows a full
1510 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1511 	 * before will also match this rule.  To avoid such situation, VNI 0
1512 	 * is currently refused.
1513 	 */
1514 	if (!vlan_id)
1515 		return rte_flow_error_set(error, ENOTSUP,
1516 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1517 					  "VXLAN-GPE vni cannot be 0");
1518 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1519 		return rte_flow_error_set(error, ENOTSUP,
1520 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1521 					  "VXLAN-GPE tunnel must be fully"
1522 					  " defined");
1523 	return 0;
1524 }
1525 
1526 /**
1527  * Validate GRE item.
1528  *
1529  * @param[in] item
1530  *   Item specification.
1531  * @param[in] item_flags
1532  *   Bit flags to mark detected items.
1533  * @param[in] target_protocol
1534  *   The next protocol in the previous item.
1535  * @param[out] error
1536  *   Pointer to error structure.
1537  *
1538  * @return
1539  *   0 on success, a negative errno value otherwise and rte_errno is set.
1540  */
1541 int
1542 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1543 			    uint64_t item_flags,
1544 			    uint8_t target_protocol,
1545 			    struct rte_flow_error *error)
1546 {
1547 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1548 	const struct rte_flow_item_gre *mask = item->mask;
1549 	int ret;
1550 
1551 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1552 		return rte_flow_error_set(error, EINVAL,
1553 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1554 					  "protocol filtering not compatible"
1555 					  " with this GRE layer");
1556 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1557 		return rte_flow_error_set(error, ENOTSUP,
1558 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1559 					  "a tunnel is already present");
1560 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1561 		return rte_flow_error_set(error, ENOTSUP,
1562 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1563 					  "L3 Layer is missing");
1564 	if (!mask)
1565 		mask = &rte_flow_item_gre_mask;
1566 	ret = mlx5_flow_item_acceptable
1567 		(item, (const uint8_t *)mask,
1568 		 (const uint8_t *)&rte_flow_item_gre_mask,
1569 		 sizeof(struct rte_flow_item_gre), error);
1570 	if (ret < 0)
1571 		return ret;
1572 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1573 	if (spec && (spec->protocol & mask->protocol))
1574 		return rte_flow_error_set(error, ENOTSUP,
1575 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1576 					  "without MPLS support the"
1577 					  " specification cannot be used for"
1578 					  " filtering");
1579 #endif
1580 	return 0;
1581 }
1582 
1583 /**
1584  * Validate MPLS item.
1585  *
1586  * @param[in] item
1587  *   Item specification.
1588  * @param[in] item_flags
1589  *   Bit-fields that holds the items detected until now.
1590  * @param[in] target_protocol
1591  *   The next protocol in the previous item.
1592  * @param[out] error
1593  *   Pointer to error structure.
1594  *
1595  * @return
1596  *   0 on success, a negative errno value otherwise and rte_errno is set.
1597  */
1598 int
1599 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1600 			     uint64_t item_flags __rte_unused,
1601 			     uint8_t target_protocol __rte_unused,
1602 			     struct rte_flow_error *error)
1603 {
1604 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1605 	const struct rte_flow_item_mpls *mask = item->mask;
1606 	int ret;
1607 
1608 	if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1609 		return rte_flow_error_set(error, EINVAL,
1610 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1611 					  "protocol filtering not compatible"
1612 					  " with MPLS layer");
1613 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1614 		return rte_flow_error_set(error, ENOTSUP,
1615 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1616 					  "a tunnel is already"
1617 					  " present");
1618 	if (!mask)
1619 		mask = &rte_flow_item_mpls_mask;
1620 	ret = mlx5_flow_item_acceptable
1621 		(item, (const uint8_t *)mask,
1622 		 (const uint8_t *)&rte_flow_item_mpls_mask,
1623 		 sizeof(struct rte_flow_item_mpls), error);
1624 	if (ret < 0)
1625 		return ret;
1626 	return 0;
1627 #endif
1628 	return rte_flow_error_set(error, ENOTSUP,
1629 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
1630 				  "MPLS is not supported by Verbs, please"
1631 				  " update.");
1632 }
1633 
1634 static int
1635 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1636 		   const struct rte_flow_attr *attr __rte_unused,
1637 		   const struct rte_flow_item items[] __rte_unused,
1638 		   const struct rte_flow_action actions[] __rte_unused,
1639 		   struct rte_flow_error *error __rte_unused)
1640 {
1641 	rte_errno = ENOTSUP;
1642 	return -rte_errno;
1643 }
1644 
1645 static struct mlx5_flow *
1646 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1647 		  const struct rte_flow_item items[] __rte_unused,
1648 		  const struct rte_flow_action actions[] __rte_unused,
1649 		  uint64_t *item_flags __rte_unused,
1650 		  uint64_t *action_flags __rte_unused,
1651 		  struct rte_flow_error *error __rte_unused)
1652 {
1653 	rte_errno = ENOTSUP;
1654 	return NULL;
1655 }
1656 
1657 static int
1658 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1659 		    struct mlx5_flow *dev_flow __rte_unused,
1660 		    const struct rte_flow_attr *attr __rte_unused,
1661 		    const struct rte_flow_item items[] __rte_unused,
1662 		    const struct rte_flow_action actions[] __rte_unused,
1663 		    struct rte_flow_error *error __rte_unused)
1664 {
1665 	rte_errno = ENOTSUP;
1666 	return -rte_errno;
1667 }
1668 
1669 static int
1670 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1671 		struct rte_flow *flow __rte_unused,
1672 		struct rte_flow_error *error __rte_unused)
1673 {
1674 	rte_errno = ENOTSUP;
1675 	return -rte_errno;
1676 }
1677 
1678 static void
1679 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1680 		 struct rte_flow *flow __rte_unused)
1681 {
1682 }
1683 
1684 static void
1685 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1686 		  struct rte_flow *flow __rte_unused)
1687 {
1688 }
1689 
1690 static int
1691 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1692 		struct rte_flow *flow __rte_unused,
1693 		const struct rte_flow_action *actions __rte_unused,
1694 		void *data __rte_unused,
1695 		struct rte_flow_error *error __rte_unused)
1696 {
1697 	rte_errno = ENOTSUP;
1698 	return -rte_errno;
1699 }
1700 
1701 /* Void driver to protect from null pointer reference. */
1702 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1703 	.validate = flow_null_validate,
1704 	.prepare = flow_null_prepare,
1705 	.translate = flow_null_translate,
1706 	.apply = flow_null_apply,
1707 	.remove = flow_null_remove,
1708 	.destroy = flow_null_destroy,
1709 	.query = flow_null_query,
1710 };
1711 
1712 /**
1713  * Select flow driver type according to flow attributes and device
1714  * configuration.
1715  *
1716  * @param[in] dev
1717  *   Pointer to the dev structure.
1718  * @param[in] attr
1719  *   Pointer to the flow attributes.
1720  *
1721  * @return
1722  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1723  */
1724 static enum mlx5_flow_drv_type
1725 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1726 {
1727 	struct priv *priv = dev->data->dev_private;
1728 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1729 
1730 	if (attr->transfer)
1731 		type = MLX5_FLOW_TYPE_TCF;
1732 	else
1733 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1734 						 MLX5_FLOW_TYPE_VERBS;
1735 	return type;
1736 }
1737 
1738 #define flow_get_drv_ops(type) flow_drv_ops[type]
1739 
1740 /**
1741  * Flow driver validation API. This abstracts calling driver specific functions.
1742  * The type of flow driver is determined according to flow attributes.
1743  *
1744  * @param[in] dev
1745  *   Pointer to the dev structure.
1746  * @param[in] attr
1747  *   Pointer to the flow attributes.
1748  * @param[in] items
1749  *   Pointer to the list of items.
1750  * @param[in] actions
1751  *   Pointer to the list of actions.
1752  * @param[out] error
1753  *   Pointer to the error structure.
1754  *
1755  * @return
1756  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1757  */
1758 static inline int
1759 flow_drv_validate(struct rte_eth_dev *dev,
1760 		  const struct rte_flow_attr *attr,
1761 		  const struct rte_flow_item items[],
1762 		  const struct rte_flow_action actions[],
1763 		  struct rte_flow_error *error)
1764 {
1765 	const struct mlx5_flow_driver_ops *fops;
1766 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1767 
1768 	fops = flow_get_drv_ops(type);
1769 	return fops->validate(dev, attr, items, actions, error);
1770 }
1771 
1772 /**
1773  * Flow driver preparation API. This abstracts calling driver specific
1774  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1775  * calculates the size of memory required for device flow, allocates the memory,
1776  * initializes the device flow and returns the pointer.
1777  *
1778  * @param[in] attr
1779  *   Pointer to the flow attributes.
1780  * @param[in] items
1781  *   Pointer to the list of items.
1782  * @param[in] actions
1783  *   Pointer to the list of actions.
1784  * @param[out] item_flags
1785  *   Pointer to bit mask of all items detected.
1786  * @param[out] action_flags
1787  *   Pointer to bit mask of all actions detected.
1788  * @param[out] error
1789  *   Pointer to the error structure.
1790  *
1791  * @return
1792  *   Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1793  */
1794 static inline struct mlx5_flow *
1795 flow_drv_prepare(struct rte_flow *flow,
1796 		 const struct rte_flow_attr *attr,
1797 		 const struct rte_flow_item items[],
1798 		 const struct rte_flow_action actions[],
1799 		 uint64_t *item_flags,
1800 		 uint64_t *action_flags,
1801 		 struct rte_flow_error *error)
1802 {
1803 	const struct mlx5_flow_driver_ops *fops;
1804 	enum mlx5_flow_drv_type type = flow->drv_type;
1805 
1806 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1807 	fops = flow_get_drv_ops(type);
1808 	return fops->prepare(attr, items, actions, item_flags, action_flags,
1809 			     error);
1810 }
1811 
1812 /**
1813  * Flow driver translation API. This abstracts calling driver specific
1814  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1815  * translates a generic flow into a driver flow. flow_drv_prepare() must
1816  * precede.
1817  *
1818  *
1819  * @param[in] dev
1820  *   Pointer to the rte dev structure.
1821  * @param[in, out] dev_flow
1822  *   Pointer to the mlx5 flow.
1823  * @param[in] attr
1824  *   Pointer to the flow attributes.
1825  * @param[in] items
1826  *   Pointer to the list of items.
1827  * @param[in] actions
1828  *   Pointer to the list of actions.
1829  * @param[out] error
1830  *   Pointer to the error structure.
1831  *
1832  * @return
1833  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1834  */
1835 static inline int
1836 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1837 		   const struct rte_flow_attr *attr,
1838 		   const struct rte_flow_item items[],
1839 		   const struct rte_flow_action actions[],
1840 		   struct rte_flow_error *error)
1841 {
1842 	const struct mlx5_flow_driver_ops *fops;
1843 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1844 
1845 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1846 	fops = flow_get_drv_ops(type);
1847 	return fops->translate(dev, dev_flow, attr, items, actions, error);
1848 }
1849 
1850 /**
1851  * Flow driver apply API. This abstracts calling driver specific functions.
1852  * Parent flow (rte_flow) should have driver type (drv_type). It applies
1853  * translated driver flows on to device. flow_drv_translate() must precede.
1854  *
1855  * @param[in] dev
1856  *   Pointer to Ethernet device structure.
1857  * @param[in, out] flow
1858  *   Pointer to flow structure.
1859  * @param[out] error
1860  *   Pointer to error structure.
1861  *
1862  * @return
1863  *   0 on success, a negative errno value otherwise and rte_errno is set.
1864  */
1865 static inline int
1866 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1867 	       struct rte_flow_error *error)
1868 {
1869 	const struct mlx5_flow_driver_ops *fops;
1870 	enum mlx5_flow_drv_type type = flow->drv_type;
1871 
1872 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1873 	fops = flow_get_drv_ops(type);
1874 	return fops->apply(dev, flow, error);
1875 }
1876 
1877 /**
1878  * Flow driver remove API. This abstracts calling driver specific functions.
1879  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1880  * on device. All the resources of the flow should be freed by calling
1881  * flow_dv_destroy().
1882  *
1883  * @param[in] dev
1884  *   Pointer to Ethernet device.
1885  * @param[in, out] flow
1886  *   Pointer to flow structure.
1887  */
1888 static inline void
1889 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1890 {
1891 	const struct mlx5_flow_driver_ops *fops;
1892 	enum mlx5_flow_drv_type type = flow->drv_type;
1893 
1894 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1895 	fops = flow_get_drv_ops(type);
1896 	fops->remove(dev, flow);
1897 }
1898 
1899 /**
1900  * Flow driver destroy API. This abstracts calling driver specific functions.
1901  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1902  * on device and releases resources of the flow.
1903  *
1904  * @param[in] dev
1905  *   Pointer to Ethernet device.
1906  * @param[in, out] flow
1907  *   Pointer to flow structure.
1908  */
1909 static inline void
1910 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1911 {
1912 	const struct mlx5_flow_driver_ops *fops;
1913 	enum mlx5_flow_drv_type type = flow->drv_type;
1914 
1915 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1916 	fops = flow_get_drv_ops(type);
1917 	fops->destroy(dev, flow);
1918 }
1919 
1920 /**
1921  * Validate a flow supported by the NIC.
1922  *
1923  * @see rte_flow_validate()
1924  * @see rte_flow_ops
1925  */
1926 int
1927 mlx5_flow_validate(struct rte_eth_dev *dev,
1928 		   const struct rte_flow_attr *attr,
1929 		   const struct rte_flow_item items[],
1930 		   const struct rte_flow_action actions[],
1931 		   struct rte_flow_error *error)
1932 {
1933 	int ret;
1934 
1935 	ret = flow_drv_validate(dev, attr, items, actions, error);
1936 	if (ret < 0)
1937 		return ret;
1938 	return 0;
1939 }
1940 
1941 /**
1942  * Get RSS action from the action list.
1943  *
1944  * @param[in] actions
1945  *   Pointer to the list of actions.
1946  *
1947  * @return
1948  *   Pointer to the RSS action if exist, else return NULL.
1949  */
1950 static const struct rte_flow_action_rss*
1951 flow_get_rss_action(const struct rte_flow_action actions[])
1952 {
1953 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1954 		switch (actions->type) {
1955 		case RTE_FLOW_ACTION_TYPE_RSS:
1956 			return (const struct rte_flow_action_rss *)
1957 			       actions->conf;
1958 		default:
1959 			break;
1960 		}
1961 	}
1962 	return NULL;
1963 }
1964 
1965 static unsigned int
1966 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
1967 {
1968 	const struct rte_flow_item *item;
1969 	unsigned int has_vlan = 0;
1970 
1971 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
1972 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
1973 			has_vlan = 1;
1974 			break;
1975 		}
1976 	}
1977 	if (has_vlan)
1978 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
1979 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
1980 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
1981 			       MLX5_EXPANSION_ROOT_OUTER;
1982 }
1983 
1984 /**
1985  * Create a flow and add it to @p list.
1986  *
1987  * @param dev
1988  *   Pointer to Ethernet device.
1989  * @param list
1990  *   Pointer to a TAILQ flow list.
1991  * @param[in] attr
1992  *   Flow rule attributes.
1993  * @param[in] items
1994  *   Pattern specification (list terminated by the END pattern item).
1995  * @param[in] actions
1996  *   Associated actions (list terminated by the END action).
1997  * @param[out] error
1998  *   Perform verbose error reporting if not NULL.
1999  *
2000  * @return
2001  *   A flow on success, NULL otherwise and rte_errno is set.
2002  */
2003 static struct rte_flow *
2004 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2005 		 const struct rte_flow_attr *attr,
2006 		 const struct rte_flow_item items[],
2007 		 const struct rte_flow_action actions[],
2008 		 struct rte_flow_error *error)
2009 {
2010 	struct rte_flow *flow = NULL;
2011 	struct mlx5_flow *dev_flow;
2012 	uint64_t action_flags = 0;
2013 	uint64_t item_flags = 0;
2014 	const struct rte_flow_action_rss *rss;
2015 	union {
2016 		struct rte_flow_expand_rss buf;
2017 		uint8_t buffer[2048];
2018 	} expand_buffer;
2019 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2020 	int ret;
2021 	uint32_t i;
2022 	uint32_t flow_size;
2023 
2024 	ret = flow_drv_validate(dev, attr, items, actions, error);
2025 	if (ret < 0)
2026 		return NULL;
2027 	flow_size = sizeof(struct rte_flow);
2028 	rss = flow_get_rss_action(actions);
2029 	if (rss)
2030 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2031 					    sizeof(void *));
2032 	else
2033 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2034 	flow = rte_calloc(__func__, 1, flow_size, 0);
2035 	flow->drv_type = flow_get_drv_type(dev, attr);
2036 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2037 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
2038 	flow->queue = (void *)(flow + 1);
2039 	LIST_INIT(&flow->dev_flows);
2040 	if (rss && rss->types) {
2041 		unsigned int graph_root;
2042 
2043 		graph_root = find_graph_root(items, rss->level);
2044 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2045 					  items, rss->types,
2046 					  mlx5_support_expansion,
2047 					  graph_root);
2048 		assert(ret > 0 &&
2049 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
2050 	} else {
2051 		buf->entries = 1;
2052 		buf->entry[0].pattern = (void *)(uintptr_t)items;
2053 	}
2054 	for (i = 0; i < buf->entries; ++i) {
2055 		dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2056 					    actions, &item_flags, &action_flags,
2057 					    error);
2058 		if (!dev_flow)
2059 			goto error;
2060 		dev_flow->flow = flow;
2061 		dev_flow->layers = item_flags;
2062 		/* Store actions once as expanded flows have same actions. */
2063 		if (i == 0)
2064 			flow->actions = action_flags;
2065 		assert(flow->actions == action_flags);
2066 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2067 		ret = flow_drv_translate(dev, dev_flow, attr,
2068 					 buf->entry[i].pattern,
2069 					 actions, error);
2070 		if (ret < 0)
2071 			goto error;
2072 	}
2073 	if (dev->data->dev_started) {
2074 		ret = flow_drv_apply(dev, flow, error);
2075 		if (ret < 0)
2076 			goto error;
2077 	}
2078 	TAILQ_INSERT_TAIL(list, flow, next);
2079 	flow_rxq_flags_set(dev, flow);
2080 	return flow;
2081 error:
2082 	ret = rte_errno; /* Save rte_errno before cleanup. */
2083 	assert(flow);
2084 	flow_drv_destroy(dev, flow);
2085 	rte_free(flow);
2086 	rte_errno = ret; /* Restore rte_errno. */
2087 	return NULL;
2088 }
2089 
2090 /**
2091  * Create a flow.
2092  *
2093  * @see rte_flow_create()
2094  * @see rte_flow_ops
2095  */
2096 struct rte_flow *
2097 mlx5_flow_create(struct rte_eth_dev *dev,
2098 		 const struct rte_flow_attr *attr,
2099 		 const struct rte_flow_item items[],
2100 		 const struct rte_flow_action actions[],
2101 		 struct rte_flow_error *error)
2102 {
2103 	return flow_list_create(dev,
2104 				&((struct priv *)dev->data->dev_private)->flows,
2105 				attr, items, actions, error);
2106 }
2107 
2108 /**
2109  * Destroy a flow in a list.
2110  *
2111  * @param dev
2112  *   Pointer to Ethernet device.
2113  * @param list
2114  *   Pointer to a TAILQ flow list.
2115  * @param[in] flow
2116  *   Flow to destroy.
2117  */
2118 static void
2119 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2120 		  struct rte_flow *flow)
2121 {
2122 	flow_drv_destroy(dev, flow);
2123 	TAILQ_REMOVE(list, flow, next);
2124 	/*
2125 	 * Update RX queue flags only if port is started, otherwise it is
2126 	 * already clean.
2127 	 */
2128 	if (dev->data->dev_started)
2129 		flow_rxq_flags_trim(dev, flow);
2130 	rte_free(flow);
2131 }
2132 
2133 /**
2134  * Destroy all flows.
2135  *
2136  * @param dev
2137  *   Pointer to Ethernet device.
2138  * @param list
2139  *   Pointer to a TAILQ flow list.
2140  */
2141 void
2142 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2143 {
2144 	while (!TAILQ_EMPTY(list)) {
2145 		struct rte_flow *flow;
2146 
2147 		flow = TAILQ_FIRST(list);
2148 		flow_list_destroy(dev, list, flow);
2149 	}
2150 }
2151 
2152 /**
2153  * Remove all flows.
2154  *
2155  * @param dev
2156  *   Pointer to Ethernet device.
2157  * @param list
2158  *   Pointer to a TAILQ flow list.
2159  */
2160 void
2161 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2162 {
2163 	struct rte_flow *flow;
2164 
2165 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2166 		flow_drv_remove(dev, flow);
2167 	flow_rxq_flags_clear(dev);
2168 }
2169 
2170 /**
2171  * Add all flows.
2172  *
2173  * @param dev
2174  *   Pointer to Ethernet device.
2175  * @param list
2176  *   Pointer to a TAILQ flow list.
2177  *
2178  * @return
2179  *   0 on success, a negative errno value otherwise and rte_errno is set.
2180  */
2181 int
2182 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2183 {
2184 	struct rte_flow *flow;
2185 	struct rte_flow_error error;
2186 	int ret = 0;
2187 
2188 	TAILQ_FOREACH(flow, list, next) {
2189 		ret = flow_drv_apply(dev, flow, &error);
2190 		if (ret < 0)
2191 			goto error;
2192 		flow_rxq_flags_set(dev, flow);
2193 	}
2194 	return 0;
2195 error:
2196 	ret = rte_errno; /* Save rte_errno before cleanup. */
2197 	mlx5_flow_stop(dev, list);
2198 	rte_errno = ret; /* Restore rte_errno. */
2199 	return -rte_errno;
2200 }
2201 
2202 /**
2203  * Verify the flow list is empty
2204  *
2205  * @param dev
2206  *  Pointer to Ethernet device.
2207  *
2208  * @return the number of flows not released.
2209  */
2210 int
2211 mlx5_flow_verify(struct rte_eth_dev *dev)
2212 {
2213 	struct priv *priv = dev->data->dev_private;
2214 	struct rte_flow *flow;
2215 	int ret = 0;
2216 
2217 	TAILQ_FOREACH(flow, &priv->flows, next) {
2218 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
2219 			dev->data->port_id, (void *)flow);
2220 		++ret;
2221 	}
2222 	return ret;
2223 }
2224 
2225 /**
2226  * Enable a control flow configured from the control plane.
2227  *
2228  * @param dev
2229  *   Pointer to Ethernet device.
2230  * @param eth_spec
2231  *   An Ethernet flow spec to apply.
2232  * @param eth_mask
2233  *   An Ethernet flow mask to apply.
2234  * @param vlan_spec
2235  *   A VLAN flow spec to apply.
2236  * @param vlan_mask
2237  *   A VLAN flow mask to apply.
2238  *
2239  * @return
2240  *   0 on success, a negative errno value otherwise and rte_errno is set.
2241  */
2242 int
2243 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2244 		    struct rte_flow_item_eth *eth_spec,
2245 		    struct rte_flow_item_eth *eth_mask,
2246 		    struct rte_flow_item_vlan *vlan_spec,
2247 		    struct rte_flow_item_vlan *vlan_mask)
2248 {
2249 	struct priv *priv = dev->data->dev_private;
2250 	const struct rte_flow_attr attr = {
2251 		.ingress = 1,
2252 		.priority = MLX5_FLOW_PRIO_RSVD,
2253 	};
2254 	struct rte_flow_item items[] = {
2255 		{
2256 			.type = RTE_FLOW_ITEM_TYPE_ETH,
2257 			.spec = eth_spec,
2258 			.last = NULL,
2259 			.mask = eth_mask,
2260 		},
2261 		{
2262 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2263 					      RTE_FLOW_ITEM_TYPE_END,
2264 			.spec = vlan_spec,
2265 			.last = NULL,
2266 			.mask = vlan_mask,
2267 		},
2268 		{
2269 			.type = RTE_FLOW_ITEM_TYPE_END,
2270 		},
2271 	};
2272 	uint16_t queue[priv->reta_idx_n];
2273 	struct rte_flow_action_rss action_rss = {
2274 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2275 		.level = 0,
2276 		.types = priv->rss_conf.rss_hf,
2277 		.key_len = priv->rss_conf.rss_key_len,
2278 		.queue_num = priv->reta_idx_n,
2279 		.key = priv->rss_conf.rss_key,
2280 		.queue = queue,
2281 	};
2282 	struct rte_flow_action actions[] = {
2283 		{
2284 			.type = RTE_FLOW_ACTION_TYPE_RSS,
2285 			.conf = &action_rss,
2286 		},
2287 		{
2288 			.type = RTE_FLOW_ACTION_TYPE_END,
2289 		},
2290 	};
2291 	struct rte_flow *flow;
2292 	struct rte_flow_error error;
2293 	unsigned int i;
2294 
2295 	if (!priv->reta_idx_n) {
2296 		rte_errno = EINVAL;
2297 		return -rte_errno;
2298 	}
2299 	for (i = 0; i != priv->reta_idx_n; ++i)
2300 		queue[i] = (*priv->reta_idx)[i];
2301 	flow = flow_list_create(dev, &priv->ctrl_flows,
2302 				&attr, items, actions, &error);
2303 	if (!flow)
2304 		return -rte_errno;
2305 	return 0;
2306 }
2307 
2308 /**
2309  * Enable a flow control configured from the control plane.
2310  *
2311  * @param dev
2312  *   Pointer to Ethernet device.
2313  * @param eth_spec
2314  *   An Ethernet flow spec to apply.
2315  * @param eth_mask
2316  *   An Ethernet flow mask to apply.
2317  *
2318  * @return
2319  *   0 on success, a negative errno value otherwise and rte_errno is set.
2320  */
2321 int
2322 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2323 	       struct rte_flow_item_eth *eth_spec,
2324 	       struct rte_flow_item_eth *eth_mask)
2325 {
2326 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2327 }
2328 
2329 /**
2330  * Destroy a flow.
2331  *
2332  * @see rte_flow_destroy()
2333  * @see rte_flow_ops
2334  */
2335 int
2336 mlx5_flow_destroy(struct rte_eth_dev *dev,
2337 		  struct rte_flow *flow,
2338 		  struct rte_flow_error *error __rte_unused)
2339 {
2340 	struct priv *priv = dev->data->dev_private;
2341 
2342 	flow_list_destroy(dev, &priv->flows, flow);
2343 	return 0;
2344 }
2345 
2346 /**
2347  * Destroy all flows.
2348  *
2349  * @see rte_flow_flush()
2350  * @see rte_flow_ops
2351  */
2352 int
2353 mlx5_flow_flush(struct rte_eth_dev *dev,
2354 		struct rte_flow_error *error __rte_unused)
2355 {
2356 	struct priv *priv = dev->data->dev_private;
2357 
2358 	mlx5_flow_list_flush(dev, &priv->flows);
2359 	return 0;
2360 }
2361 
2362 /**
2363  * Isolated mode.
2364  *
2365  * @see rte_flow_isolate()
2366  * @see rte_flow_ops
2367  */
2368 int
2369 mlx5_flow_isolate(struct rte_eth_dev *dev,
2370 		  int enable,
2371 		  struct rte_flow_error *error)
2372 {
2373 	struct priv *priv = dev->data->dev_private;
2374 
2375 	if (dev->data->dev_started) {
2376 		rte_flow_error_set(error, EBUSY,
2377 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2378 				   NULL,
2379 				   "port must be stopped first");
2380 		return -rte_errno;
2381 	}
2382 	priv->isolated = !!enable;
2383 	if (enable)
2384 		dev->dev_ops = &mlx5_dev_ops_isolate;
2385 	else
2386 		dev->dev_ops = &mlx5_dev_ops;
2387 	return 0;
2388 }
2389 
2390 /**
2391  * Query a flow.
2392  *
2393  * @see rte_flow_query()
2394  * @see rte_flow_ops
2395  */
2396 static int
2397 flow_drv_query(struct rte_eth_dev *dev,
2398 	       struct rte_flow *flow,
2399 	       const struct rte_flow_action *actions,
2400 	       void *data,
2401 	       struct rte_flow_error *error)
2402 {
2403 	const struct mlx5_flow_driver_ops *fops;
2404 	enum mlx5_flow_drv_type ftype = flow->drv_type;
2405 
2406 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2407 	fops = flow_get_drv_ops(ftype);
2408 
2409 	return fops->query(dev, flow, actions, data, error);
2410 }
2411 
2412 /**
2413  * Query a flow.
2414  *
2415  * @see rte_flow_query()
2416  * @see rte_flow_ops
2417  */
2418 int
2419 mlx5_flow_query(struct rte_eth_dev *dev,
2420 		struct rte_flow *flow,
2421 		const struct rte_flow_action *actions,
2422 		void *data,
2423 		struct rte_flow_error *error)
2424 {
2425 	int ret;
2426 
2427 	ret = flow_drv_query(dev, flow, actions, data, error);
2428 	if (ret < 0)
2429 		return ret;
2430 	return 0;
2431 }
2432 
2433 /**
2434  * Convert a flow director filter to a generic flow.
2435  *
2436  * @param dev
2437  *   Pointer to Ethernet device.
2438  * @param fdir_filter
2439  *   Flow director filter to add.
2440  * @param attributes
2441  *   Generic flow parameters structure.
2442  *
2443  * @return
2444  *   0 on success, a negative errno value otherwise and rte_errno is set.
2445  */
2446 static int
2447 mlx5_fdir_filter_convert(struct rte_eth_dev *dev,
2448 			 const struct rte_eth_fdir_filter *fdir_filter,
2449 			 struct mlx5_fdir *attributes)
2450 {
2451 	struct priv *priv = dev->data->dev_private;
2452 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
2453 	const struct rte_eth_fdir_masks *mask =
2454 		&dev->data->dev_conf.fdir_conf.mask;
2455 
2456 	/* Validate queue number. */
2457 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2458 		DRV_LOG(ERR, "port %u invalid queue number %d",
2459 			dev->data->port_id, fdir_filter->action.rx_queue);
2460 		rte_errno = EINVAL;
2461 		return -rte_errno;
2462 	}
2463 	attributes->attr.ingress = 1;
2464 	attributes->items[0] = (struct rte_flow_item) {
2465 		.type = RTE_FLOW_ITEM_TYPE_ETH,
2466 		.spec = &attributes->l2,
2467 		.mask = &attributes->l2_mask,
2468 	};
2469 	switch (fdir_filter->action.behavior) {
2470 	case RTE_ETH_FDIR_ACCEPT:
2471 		attributes->actions[0] = (struct rte_flow_action){
2472 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
2473 			.conf = &attributes->queue,
2474 		};
2475 		break;
2476 	case RTE_ETH_FDIR_REJECT:
2477 		attributes->actions[0] = (struct rte_flow_action){
2478 			.type = RTE_FLOW_ACTION_TYPE_DROP,
2479 		};
2480 		break;
2481 	default:
2482 		DRV_LOG(ERR, "port %u invalid behavior %d",
2483 			dev->data->port_id,
2484 			fdir_filter->action.behavior);
2485 		rte_errno = ENOTSUP;
2486 		return -rte_errno;
2487 	}
2488 	attributes->queue.index = fdir_filter->action.rx_queue;
2489 	/* Handle L3. */
2490 	switch (fdir_filter->input.flow_type) {
2491 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2492 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2493 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2494 		attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2495 			.src_addr = input->flow.ip4_flow.src_ip,
2496 			.dst_addr = input->flow.ip4_flow.dst_ip,
2497 			.time_to_live = input->flow.ip4_flow.ttl,
2498 			.type_of_service = input->flow.ip4_flow.tos,
2499 		};
2500 		attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2501 			.src_addr = mask->ipv4_mask.src_ip,
2502 			.dst_addr = mask->ipv4_mask.dst_ip,
2503 			.time_to_live = mask->ipv4_mask.ttl,
2504 			.type_of_service = mask->ipv4_mask.tos,
2505 			.next_proto_id = mask->ipv4_mask.proto,
2506 		};
2507 		attributes->items[1] = (struct rte_flow_item){
2508 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
2509 			.spec = &attributes->l3,
2510 			.mask = &attributes->l3_mask,
2511 		};
2512 		break;
2513 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2514 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2515 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2516 		attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2517 			.hop_limits = input->flow.ipv6_flow.hop_limits,
2518 			.proto = input->flow.ipv6_flow.proto,
2519 		};
2520 
2521 		memcpy(attributes->l3.ipv6.hdr.src_addr,
2522 		       input->flow.ipv6_flow.src_ip,
2523 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2524 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
2525 		       input->flow.ipv6_flow.dst_ip,
2526 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2527 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2528 		       mask->ipv6_mask.src_ip,
2529 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2530 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2531 		       mask->ipv6_mask.dst_ip,
2532 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2533 		attributes->items[1] = (struct rte_flow_item){
2534 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
2535 			.spec = &attributes->l3,
2536 			.mask = &attributes->l3_mask,
2537 		};
2538 		break;
2539 	default:
2540 		DRV_LOG(ERR, "port %u invalid flow type%d",
2541 			dev->data->port_id, fdir_filter->input.flow_type);
2542 		rte_errno = ENOTSUP;
2543 		return -rte_errno;
2544 	}
2545 	/* Handle L4. */
2546 	switch (fdir_filter->input.flow_type) {
2547 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2548 		attributes->l4.udp.hdr = (struct udp_hdr){
2549 			.src_port = input->flow.udp4_flow.src_port,
2550 			.dst_port = input->flow.udp4_flow.dst_port,
2551 		};
2552 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2553 			.src_port = mask->src_port_mask,
2554 			.dst_port = mask->dst_port_mask,
2555 		};
2556 		attributes->items[2] = (struct rte_flow_item){
2557 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2558 			.spec = &attributes->l4,
2559 			.mask = &attributes->l4_mask,
2560 		};
2561 		break;
2562 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2563 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2564 			.src_port = input->flow.tcp4_flow.src_port,
2565 			.dst_port = input->flow.tcp4_flow.dst_port,
2566 		};
2567 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2568 			.src_port = mask->src_port_mask,
2569 			.dst_port = mask->dst_port_mask,
2570 		};
2571 		attributes->items[2] = (struct rte_flow_item){
2572 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2573 			.spec = &attributes->l4,
2574 			.mask = &attributes->l4_mask,
2575 		};
2576 		break;
2577 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2578 		attributes->l4.udp.hdr = (struct udp_hdr){
2579 			.src_port = input->flow.udp6_flow.src_port,
2580 			.dst_port = input->flow.udp6_flow.dst_port,
2581 		};
2582 		attributes->l4_mask.udp.hdr = (struct udp_hdr){
2583 			.src_port = mask->src_port_mask,
2584 			.dst_port = mask->dst_port_mask,
2585 		};
2586 		attributes->items[2] = (struct rte_flow_item){
2587 			.type = RTE_FLOW_ITEM_TYPE_UDP,
2588 			.spec = &attributes->l4,
2589 			.mask = &attributes->l4_mask,
2590 		};
2591 		break;
2592 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2593 		attributes->l4.tcp.hdr = (struct tcp_hdr){
2594 			.src_port = input->flow.tcp6_flow.src_port,
2595 			.dst_port = input->flow.tcp6_flow.dst_port,
2596 		};
2597 		attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2598 			.src_port = mask->src_port_mask,
2599 			.dst_port = mask->dst_port_mask,
2600 		};
2601 		attributes->items[2] = (struct rte_flow_item){
2602 			.type = RTE_FLOW_ITEM_TYPE_TCP,
2603 			.spec = &attributes->l4,
2604 			.mask = &attributes->l4_mask,
2605 		};
2606 		break;
2607 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2608 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2609 		break;
2610 	default:
2611 		DRV_LOG(ERR, "port %u invalid flow type%d",
2612 			dev->data->port_id, fdir_filter->input.flow_type);
2613 		rte_errno = ENOTSUP;
2614 		return -rte_errno;
2615 	}
2616 	return 0;
2617 }
2618 
2619 /**
2620  * Add new flow director filter and store it in list.
2621  *
2622  * @param dev
2623  *   Pointer to Ethernet device.
2624  * @param fdir_filter
2625  *   Flow director filter to add.
2626  *
2627  * @return
2628  *   0 on success, a negative errno value otherwise and rte_errno is set.
2629  */
2630 static int
2631 mlx5_fdir_filter_add(struct rte_eth_dev *dev,
2632 		     const struct rte_eth_fdir_filter *fdir_filter)
2633 {
2634 	struct priv *priv = dev->data->dev_private;
2635 	struct mlx5_fdir attributes = {
2636 		.attr.group = 0,
2637 		.l2_mask = {
2638 			.dst.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2639 			.src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2640 			.type = 0,
2641 		},
2642 	};
2643 	struct rte_flow_error error;
2644 	struct rte_flow *flow;
2645 	int ret;
2646 
2647 	ret = mlx5_fdir_filter_convert(dev, fdir_filter, &attributes);
2648 	if (ret)
2649 		return ret;
2650 	flow = flow_list_create(dev, &priv->flows, &attributes.attr,
2651 				attributes.items, attributes.actions, &error);
2652 	if (flow) {
2653 		DRV_LOG(DEBUG, "port %u FDIR created %p", dev->data->port_id,
2654 			(void *)flow);
2655 		return 0;
2656 	}
2657 	return -rte_errno;
2658 }
2659 
2660 /**
2661  * Delete specific filter.
2662  *
2663  * @param dev
2664  *   Pointer to Ethernet device.
2665  * @param fdir_filter
2666  *   Filter to be deleted.
2667  *
2668  * @return
2669  *   0 on success, a negative errno value otherwise and rte_errno is set.
2670  */
2671 static int
2672 mlx5_fdir_filter_delete(struct rte_eth_dev *dev __rte_unused,
2673 			const struct rte_eth_fdir_filter *fdir_filter
2674 			__rte_unused)
2675 {
2676 	rte_errno = ENOTSUP;
2677 	return -rte_errno;
2678 }
2679 
2680 /**
2681  * Update queue for specific filter.
2682  *
2683  * @param dev
2684  *   Pointer to Ethernet device.
2685  * @param fdir_filter
2686  *   Filter to be updated.
2687  *
2688  * @return
2689  *   0 on success, a negative errno value otherwise and rte_errno is set.
2690  */
2691 static int
2692 mlx5_fdir_filter_update(struct rte_eth_dev *dev,
2693 			const struct rte_eth_fdir_filter *fdir_filter)
2694 {
2695 	int ret;
2696 
2697 	ret = mlx5_fdir_filter_delete(dev, fdir_filter);
2698 	if (ret)
2699 		return ret;
2700 	return mlx5_fdir_filter_add(dev, fdir_filter);
2701 }
2702 
2703 /**
2704  * Flush all filters.
2705  *
2706  * @param dev
2707  *   Pointer to Ethernet device.
2708  */
2709 static void
2710 mlx5_fdir_filter_flush(struct rte_eth_dev *dev)
2711 {
2712 	struct priv *priv = dev->data->dev_private;
2713 
2714 	mlx5_flow_list_flush(dev, &priv->flows);
2715 }
2716 
2717 /**
2718  * Get flow director information.
2719  *
2720  * @param dev
2721  *   Pointer to Ethernet device.
2722  * @param[out] fdir_info
2723  *   Resulting flow director information.
2724  */
2725 static void
2726 mlx5_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2727 {
2728 	struct rte_eth_fdir_masks *mask =
2729 		&dev->data->dev_conf.fdir_conf.mask;
2730 
2731 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2732 	fdir_info->guarant_spc = 0;
2733 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2734 	fdir_info->max_flexpayload = 0;
2735 	fdir_info->flow_types_mask[0] = 0;
2736 	fdir_info->flex_payload_unit = 0;
2737 	fdir_info->max_flex_payload_segment_num = 0;
2738 	fdir_info->flex_payload_limit = 0;
2739 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2740 }
2741 
2742 /**
2743  * Deal with flow director operations.
2744  *
2745  * @param dev
2746  *   Pointer to Ethernet device.
2747  * @param filter_op
2748  *   Operation to perform.
2749  * @param arg
2750  *   Pointer to operation-specific structure.
2751  *
2752  * @return
2753  *   0 on success, a negative errno value otherwise and rte_errno is set.
2754  */
2755 static int
2756 mlx5_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2757 		    void *arg)
2758 {
2759 	enum rte_fdir_mode fdir_mode =
2760 		dev->data->dev_conf.fdir_conf.mode;
2761 
2762 	if (filter_op == RTE_ETH_FILTER_NOP)
2763 		return 0;
2764 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2765 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2766 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
2767 			dev->data->port_id, fdir_mode);
2768 		rte_errno = EINVAL;
2769 		return -rte_errno;
2770 	}
2771 	switch (filter_op) {
2772 	case RTE_ETH_FILTER_ADD:
2773 		return mlx5_fdir_filter_add(dev, arg);
2774 	case RTE_ETH_FILTER_UPDATE:
2775 		return mlx5_fdir_filter_update(dev, arg);
2776 	case RTE_ETH_FILTER_DELETE:
2777 		return mlx5_fdir_filter_delete(dev, arg);
2778 	case RTE_ETH_FILTER_FLUSH:
2779 		mlx5_fdir_filter_flush(dev);
2780 		break;
2781 	case RTE_ETH_FILTER_INFO:
2782 		mlx5_fdir_info_get(dev, arg);
2783 		break;
2784 	default:
2785 		DRV_LOG(DEBUG, "port %u unknown operation %u",
2786 			dev->data->port_id, filter_op);
2787 		rte_errno = EINVAL;
2788 		return -rte_errno;
2789 	}
2790 	return 0;
2791 }
2792 
2793 /**
2794  * Manage filter operations.
2795  *
2796  * @param dev
2797  *   Pointer to Ethernet device structure.
2798  * @param filter_type
2799  *   Filter type.
2800  * @param filter_op
2801  *   Operation to perform.
2802  * @param arg
2803  *   Pointer to operation-specific structure.
2804  *
2805  * @return
2806  *   0 on success, a negative errno value otherwise and rte_errno is set.
2807  */
2808 int
2809 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2810 		     enum rte_filter_type filter_type,
2811 		     enum rte_filter_op filter_op,
2812 		     void *arg)
2813 {
2814 	switch (filter_type) {
2815 	case RTE_ETH_FILTER_GENERIC:
2816 		if (filter_op != RTE_ETH_FILTER_GET) {
2817 			rte_errno = EINVAL;
2818 			return -rte_errno;
2819 		}
2820 		*(const void **)arg = &mlx5_flow_ops;
2821 		return 0;
2822 	case RTE_ETH_FILTER_FDIR:
2823 		return mlx5_fdir_ctrl_func(dev, filter_op, arg);
2824 	default:
2825 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
2826 			dev->data->port_id, filter_type);
2827 		rte_errno = ENOTSUP;
2828 		return -rte_errno;
2829 	}
2830 	return 0;
2831 }
2832