xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 59c5779dafc5b405eaeb5aadfad3d596aa4873c5)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 
12 /* Verbs header. */
13 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
14 #ifdef PEDANTIC
15 #pragma GCC diagnostic ignored "-Wpedantic"
16 #endif
17 #include <infiniband/verbs.h>
18 #ifdef PEDANTIC
19 #pragma GCC diagnostic error "-Wpedantic"
20 #endif
21 
22 #include <rte_common.h>
23 #include <rte_ether.h>
24 #include <rte_ethdev_driver.h>
25 #include <rte_flow.h>
26 #include <rte_flow_driver.h>
27 #include <rte_malloc.h>
28 #include <rte_ip.h>
29 
30 #include "mlx5.h"
31 #include "mlx5_defs.h"
32 #include "mlx5_flow.h"
33 #include "mlx5_glue.h"
34 #include "mlx5_prm.h"
35 #include "mlx5_rxtx.h"
36 
37 /* Dev ops structure defined in mlx5.c */
38 extern const struct eth_dev_ops mlx5_dev_ops;
39 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
40 
41 /** Device flow drivers. */
42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
43 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
44 #endif
45 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
46 
47 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
48 
49 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
50 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
51 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
52 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
53 #endif
54 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
55 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
56 };
57 
58 enum mlx5_expansion {
59 	MLX5_EXPANSION_ROOT,
60 	MLX5_EXPANSION_ROOT_OUTER,
61 	MLX5_EXPANSION_ROOT_ETH_VLAN,
62 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
63 	MLX5_EXPANSION_OUTER_ETH,
64 	MLX5_EXPANSION_OUTER_ETH_VLAN,
65 	MLX5_EXPANSION_OUTER_VLAN,
66 	MLX5_EXPANSION_OUTER_IPV4,
67 	MLX5_EXPANSION_OUTER_IPV4_UDP,
68 	MLX5_EXPANSION_OUTER_IPV4_TCP,
69 	MLX5_EXPANSION_OUTER_IPV6,
70 	MLX5_EXPANSION_OUTER_IPV6_UDP,
71 	MLX5_EXPANSION_OUTER_IPV6_TCP,
72 	MLX5_EXPANSION_VXLAN,
73 	MLX5_EXPANSION_VXLAN_GPE,
74 	MLX5_EXPANSION_GRE,
75 	MLX5_EXPANSION_MPLS,
76 	MLX5_EXPANSION_ETH,
77 	MLX5_EXPANSION_ETH_VLAN,
78 	MLX5_EXPANSION_VLAN,
79 	MLX5_EXPANSION_IPV4,
80 	MLX5_EXPANSION_IPV4_UDP,
81 	MLX5_EXPANSION_IPV4_TCP,
82 	MLX5_EXPANSION_IPV6,
83 	MLX5_EXPANSION_IPV6_UDP,
84 	MLX5_EXPANSION_IPV6_TCP,
85 };
86 
87 /** Supported expansion of items. */
88 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
89 	[MLX5_EXPANSION_ROOT] = {
90 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
91 						 MLX5_EXPANSION_IPV4,
92 						 MLX5_EXPANSION_IPV6),
93 		.type = RTE_FLOW_ITEM_TYPE_END,
94 	},
95 	[MLX5_EXPANSION_ROOT_OUTER] = {
96 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
97 						 MLX5_EXPANSION_OUTER_IPV4,
98 						 MLX5_EXPANSION_OUTER_IPV6),
99 		.type = RTE_FLOW_ITEM_TYPE_END,
100 	},
101 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
102 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
103 		.type = RTE_FLOW_ITEM_TYPE_END,
104 	},
105 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
106 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
107 		.type = RTE_FLOW_ITEM_TYPE_END,
108 	},
109 	[MLX5_EXPANSION_OUTER_ETH] = {
110 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
111 						 MLX5_EXPANSION_OUTER_IPV6,
112 						 MLX5_EXPANSION_MPLS),
113 		.type = RTE_FLOW_ITEM_TYPE_ETH,
114 		.rss_types = 0,
115 	},
116 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
117 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
118 		.type = RTE_FLOW_ITEM_TYPE_ETH,
119 		.rss_types = 0,
120 	},
121 	[MLX5_EXPANSION_OUTER_VLAN] = {
122 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
123 						 MLX5_EXPANSION_OUTER_IPV6),
124 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
125 	},
126 	[MLX5_EXPANSION_OUTER_IPV4] = {
127 		.next = RTE_FLOW_EXPAND_RSS_NEXT
128 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
129 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
130 			 MLX5_EXPANSION_GRE,
131 			 MLX5_EXPANSION_IPV4,
132 			 MLX5_EXPANSION_IPV6),
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 			 MLX5_EXPANSION_IPV4,
152 			 MLX5_EXPANSION_IPV6),
153 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
154 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
155 			ETH_RSS_NONFRAG_IPV6_OTHER,
156 	},
157 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
158 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
159 						 MLX5_EXPANSION_VXLAN_GPE),
160 		.type = RTE_FLOW_ITEM_TYPE_UDP,
161 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
162 	},
163 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
164 		.type = RTE_FLOW_ITEM_TYPE_TCP,
165 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
166 	},
167 	[MLX5_EXPANSION_VXLAN] = {
168 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
169 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
170 	},
171 	[MLX5_EXPANSION_VXLAN_GPE] = {
172 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
173 						 MLX5_EXPANSION_IPV4,
174 						 MLX5_EXPANSION_IPV6),
175 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
176 	},
177 	[MLX5_EXPANSION_GRE] = {
178 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
179 		.type = RTE_FLOW_ITEM_TYPE_GRE,
180 	},
181 	[MLX5_EXPANSION_MPLS] = {
182 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
183 						 MLX5_EXPANSION_IPV6),
184 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
185 	},
186 	[MLX5_EXPANSION_ETH] = {
187 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 						 MLX5_EXPANSION_IPV6),
189 		.type = RTE_FLOW_ITEM_TYPE_ETH,
190 	},
191 	[MLX5_EXPANSION_ETH_VLAN] = {
192 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
193 		.type = RTE_FLOW_ITEM_TYPE_ETH,
194 	},
195 	[MLX5_EXPANSION_VLAN] = {
196 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
197 						 MLX5_EXPANSION_IPV6),
198 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
199 	},
200 	[MLX5_EXPANSION_IPV4] = {
201 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
202 						 MLX5_EXPANSION_IPV4_TCP),
203 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
204 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
205 			ETH_RSS_NONFRAG_IPV4_OTHER,
206 	},
207 	[MLX5_EXPANSION_IPV4_UDP] = {
208 		.type = RTE_FLOW_ITEM_TYPE_UDP,
209 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
210 	},
211 	[MLX5_EXPANSION_IPV4_TCP] = {
212 		.type = RTE_FLOW_ITEM_TYPE_TCP,
213 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
214 	},
215 	[MLX5_EXPANSION_IPV6] = {
216 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
217 						 MLX5_EXPANSION_IPV6_TCP),
218 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
219 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
220 			ETH_RSS_NONFRAG_IPV6_OTHER,
221 	},
222 	[MLX5_EXPANSION_IPV6_UDP] = {
223 		.type = RTE_FLOW_ITEM_TYPE_UDP,
224 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
225 	},
226 	[MLX5_EXPANSION_IPV6_TCP] = {
227 		.type = RTE_FLOW_ITEM_TYPE_TCP,
228 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
229 	},
230 };
231 
232 static const struct rte_flow_ops mlx5_flow_ops = {
233 	.validate = mlx5_flow_validate,
234 	.create = mlx5_flow_create,
235 	.destroy = mlx5_flow_destroy,
236 	.flush = mlx5_flow_flush,
237 	.isolate = mlx5_flow_isolate,
238 	.query = mlx5_flow_query,
239 };
240 
241 /* Convert FDIR request to Generic flow. */
242 struct mlx5_fdir {
243 	struct rte_flow_attr attr;
244 	struct rte_flow_item items[4];
245 	struct rte_flow_item_eth l2;
246 	struct rte_flow_item_eth l2_mask;
247 	union {
248 		struct rte_flow_item_ipv4 ipv4;
249 		struct rte_flow_item_ipv6 ipv6;
250 	} l3;
251 	union {
252 		struct rte_flow_item_ipv4 ipv4;
253 		struct rte_flow_item_ipv6 ipv6;
254 	} l3_mask;
255 	union {
256 		struct rte_flow_item_udp udp;
257 		struct rte_flow_item_tcp tcp;
258 	} l4;
259 	union {
260 		struct rte_flow_item_udp udp;
261 		struct rte_flow_item_tcp tcp;
262 	} l4_mask;
263 	struct rte_flow_action actions[2];
264 	struct rte_flow_action_queue queue;
265 };
266 
267 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
268 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
269 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
270 };
271 
272 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
273 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
274 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
275 	{ 9, 10, 11 }, { 12, 13, 14 },
276 };
277 
278 /* Tunnel information. */
279 struct mlx5_flow_tunnel_info {
280 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
281 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
282 };
283 
284 static struct mlx5_flow_tunnel_info tunnels_info[] = {
285 	{
286 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
287 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
288 	},
289 	{
290 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
291 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
292 	},
293 	{
294 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
295 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
296 	},
297 	{
298 		.tunnel = MLX5_FLOW_LAYER_GRE,
299 		.ptype = RTE_PTYPE_TUNNEL_GRE,
300 	},
301 	{
302 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
303 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
304 	},
305 	{
306 		.tunnel = MLX5_FLOW_LAYER_MPLS,
307 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
308 	},
309 	{
310 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
311 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
312 	},
313 	{
314 		.tunnel = MLX5_FLOW_LAYER_IPIP,
315 		.ptype = RTE_PTYPE_TUNNEL_IP,
316 	},
317 	{
318 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
319 		.ptype = RTE_PTYPE_TUNNEL_IP,
320 	},
321 };
322 
323 /**
324  * Translate tag ID to register.
325  *
326  * @param[in] dev
327  *   Pointer to the Ethernet device structure.
328  * @param[in] feature
329  *   The feature that request the register.
330  * @param[in] id
331  *   The request register ID.
332  * @param[out] error
333  *   Error description in case of any.
334  *
335  * @return
336  *   The request register on success, a negative errno
337  *   value otherwise and rte_errno is set.
338  */
339 enum modify_reg
340 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
341 		     enum mlx5_feature_name feature,
342 		     uint32_t id,
343 		     struct rte_flow_error *error)
344 {
345 	struct mlx5_priv *priv = dev->data->dev_private;
346 	struct mlx5_dev_config *config = &priv->config;
347 
348 	switch (feature) {
349 	case MLX5_HAIRPIN_RX:
350 		return REG_B;
351 	case MLX5_HAIRPIN_TX:
352 		return REG_A;
353 	case MLX5_METADATA_RX:
354 		switch (config->dv_xmeta_en) {
355 		case MLX5_XMETA_MODE_LEGACY:
356 			return REG_B;
357 		case MLX5_XMETA_MODE_META16:
358 			return REG_C_0;
359 		case MLX5_XMETA_MODE_META32:
360 			return REG_C_1;
361 		}
362 		break;
363 	case MLX5_METADATA_TX:
364 		return REG_A;
365 	case MLX5_METADATA_FDB:
366 		return REG_C_0;
367 	case MLX5_FLOW_MARK:
368 		switch (config->dv_xmeta_en) {
369 		case MLX5_XMETA_MODE_LEGACY:
370 			return REG_NONE;
371 		case MLX5_XMETA_MODE_META16:
372 			return REG_C_1;
373 		case MLX5_XMETA_MODE_META32:
374 			return REG_C_0;
375 		}
376 		break;
377 	case MLX5_COPY_MARK:
378 		return REG_C_3;
379 	case MLX5_APP_TAG:
380 		/*
381 		 * Suppose engaging reg_c_2 .. reg_c_7 registers.
382 		 * reg_c_2 is reserved for coloring by meters.
383 		 * reg_c_3 is reserved for split flows TAG.
384 		 */
385 		if (id > (REG_C_7 - REG_C_4))
386 			return rte_flow_error_set
387 					(error, EINVAL,
388 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
389 					 NULL, "invalid tag id");
390 		if (config->flow_mreg_c[id + REG_C_4 - REG_C_0] == REG_NONE)
391 			return rte_flow_error_set
392 					(error, ENOTSUP,
393 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
394 					 NULL, "unsupported tag id");
395 		return config->flow_mreg_c[id + REG_C_4 - REG_C_0];
396 	}
397 	assert(false);
398 	return rte_flow_error_set(error, EINVAL,
399 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
400 				  NULL, "invalid feature name");
401 }
402 
403 /**
404  * Check extensive flow metadata register support.
405  *
406  * @param dev
407  *   Pointer to rte_eth_dev structure.
408  *
409  * @return
410  *   True if device supports extensive flow metadata register, otherwise false.
411  */
412 bool
413 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
414 {
415 	struct mlx5_priv *priv = dev->data->dev_private;
416 	struct mlx5_dev_config *config = &priv->config;
417 
418 	/*
419 	 * Having available reg_c can be regarded inclusively as supporting
420 	 * extensive flow metadata register, which could mean,
421 	 * - metadata register copy action by modify header.
422 	 * - 16 modify header actions is supported.
423 	 * - reg_c's are preserved across different domain (FDB and NIC) on
424 	 *   packet loopback by flow lookup miss.
425 	 */
426 	return config->flow_mreg_c[2] != REG_NONE;
427 }
428 
429 /**
430  * Discover the maximum number of priority available.
431  *
432  * @param[in] dev
433  *   Pointer to the Ethernet device structure.
434  *
435  * @return
436  *   number of supported flow priority on success, a negative errno
437  *   value otherwise and rte_errno is set.
438  */
439 int
440 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
441 {
442 	struct mlx5_priv *priv = dev->data->dev_private;
443 	struct {
444 		struct ibv_flow_attr attr;
445 		struct ibv_flow_spec_eth eth;
446 		struct ibv_flow_spec_action_drop drop;
447 	} flow_attr = {
448 		.attr = {
449 			.num_of_specs = 2,
450 			.port = (uint8_t)priv->ibv_port,
451 		},
452 		.eth = {
453 			.type = IBV_FLOW_SPEC_ETH,
454 			.size = sizeof(struct ibv_flow_spec_eth),
455 		},
456 		.drop = {
457 			.size = sizeof(struct ibv_flow_spec_action_drop),
458 			.type = IBV_FLOW_SPEC_ACTION_DROP,
459 		},
460 	};
461 	struct ibv_flow *flow;
462 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
463 	uint16_t vprio[] = { 8, 16 };
464 	int i;
465 	int priority = 0;
466 
467 	if (!drop) {
468 		rte_errno = ENOTSUP;
469 		return -rte_errno;
470 	}
471 	for (i = 0; i != RTE_DIM(vprio); i++) {
472 		flow_attr.attr.priority = vprio[i] - 1;
473 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
474 		if (!flow)
475 			break;
476 		claim_zero(mlx5_glue->destroy_flow(flow));
477 		priority = vprio[i];
478 	}
479 	mlx5_hrxq_drop_release(dev);
480 	switch (priority) {
481 	case 8:
482 		priority = RTE_DIM(priority_map_3);
483 		break;
484 	case 16:
485 		priority = RTE_DIM(priority_map_5);
486 		break;
487 	default:
488 		rte_errno = ENOTSUP;
489 		DRV_LOG(ERR,
490 			"port %u verbs maximum priority: %d expected 8/16",
491 			dev->data->port_id, priority);
492 		return -rte_errno;
493 	}
494 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
495 		dev->data->port_id, priority);
496 	return priority;
497 }
498 
499 /**
500  * Adjust flow priority based on the highest layer and the request priority.
501  *
502  * @param[in] dev
503  *   Pointer to the Ethernet device structure.
504  * @param[in] priority
505  *   The rule base priority.
506  * @param[in] subpriority
507  *   The priority based on the items.
508  *
509  * @return
510  *   The new priority.
511  */
512 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
513 				   uint32_t subpriority)
514 {
515 	uint32_t res = 0;
516 	struct mlx5_priv *priv = dev->data->dev_private;
517 
518 	switch (priv->config.flow_prio) {
519 	case RTE_DIM(priority_map_3):
520 		res = priority_map_3[priority][subpriority];
521 		break;
522 	case RTE_DIM(priority_map_5):
523 		res = priority_map_5[priority][subpriority];
524 		break;
525 	}
526 	return  res;
527 }
528 
529 /**
530  * Verify the @p item specifications (spec, last, mask) are compatible with the
531  * NIC capabilities.
532  *
533  * @param[in] item
534  *   Item specification.
535  * @param[in] mask
536  *   @p item->mask or flow default bit-masks.
537  * @param[in] nic_mask
538  *   Bit-masks covering supported fields by the NIC to compare with user mask.
539  * @param[in] size
540  *   Bit-masks size in bytes.
541  * @param[out] error
542  *   Pointer to error structure.
543  *
544  * @return
545  *   0 on success, a negative errno value otherwise and rte_errno is set.
546  */
547 int
548 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
549 			  const uint8_t *mask,
550 			  const uint8_t *nic_mask,
551 			  unsigned int size,
552 			  struct rte_flow_error *error)
553 {
554 	unsigned int i;
555 
556 	assert(nic_mask);
557 	for (i = 0; i < size; ++i)
558 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
559 			return rte_flow_error_set(error, ENOTSUP,
560 						  RTE_FLOW_ERROR_TYPE_ITEM,
561 						  item,
562 						  "mask enables non supported"
563 						  " bits");
564 	if (!item->spec && (item->mask || item->last))
565 		return rte_flow_error_set(error, EINVAL,
566 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
567 					  "mask/last without a spec is not"
568 					  " supported");
569 	if (item->spec && item->last) {
570 		uint8_t spec[size];
571 		uint8_t last[size];
572 		unsigned int i;
573 		int ret;
574 
575 		for (i = 0; i < size; ++i) {
576 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
577 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
578 		}
579 		ret = memcmp(spec, last, size);
580 		if (ret != 0)
581 			return rte_flow_error_set(error, EINVAL,
582 						  RTE_FLOW_ERROR_TYPE_ITEM,
583 						  item,
584 						  "range is not valid");
585 	}
586 	return 0;
587 }
588 
589 /**
590  * Adjust the hash fields according to the @p flow information.
591  *
592  * @param[in] dev_flow.
593  *   Pointer to the mlx5_flow.
594  * @param[in] tunnel
595  *   1 when the hash field is for a tunnel item.
596  * @param[in] layer_types
597  *   ETH_RSS_* types.
598  * @param[in] hash_fields
599  *   Item hash fields.
600  *
601  * @return
602  *   The hash fields that should be used.
603  */
604 uint64_t
605 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
606 			    int tunnel __rte_unused, uint64_t layer_types,
607 			    uint64_t hash_fields)
608 {
609 	struct rte_flow *flow = dev_flow->flow;
610 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
611 	int rss_request_inner = flow->rss.level >= 2;
612 
613 	/* Check RSS hash level for tunnel. */
614 	if (tunnel && rss_request_inner)
615 		hash_fields |= IBV_RX_HASH_INNER;
616 	else if (tunnel || rss_request_inner)
617 		return 0;
618 #endif
619 	/* Check if requested layer matches RSS hash fields. */
620 	if (!(flow->rss.types & layer_types))
621 		return 0;
622 	return hash_fields;
623 }
624 
625 /**
626  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
627  * if several tunnel rules are used on this queue, the tunnel ptype will be
628  * cleared.
629  *
630  * @param rxq_ctrl
631  *   Rx queue to update.
632  */
633 static void
634 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
635 {
636 	unsigned int i;
637 	uint32_t tunnel_ptype = 0;
638 
639 	/* Look up for the ptype to use. */
640 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
641 		if (!rxq_ctrl->flow_tunnels_n[i])
642 			continue;
643 		if (!tunnel_ptype) {
644 			tunnel_ptype = tunnels_info[i].ptype;
645 		} else {
646 			tunnel_ptype = 0;
647 			break;
648 		}
649 	}
650 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
651 }
652 
653 /**
654  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
655  * flow.
656  *
657  * @param[in] dev
658  *   Pointer to the Ethernet device structure.
659  * @param[in] dev_flow
660  *   Pointer to device flow structure.
661  */
662 static void
663 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
664 {
665 	struct mlx5_priv *priv = dev->data->dev_private;
666 	struct rte_flow *flow = dev_flow->flow;
667 	const int mark = !!(dev_flow->actions &
668 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
669 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
670 	unsigned int i;
671 
672 	for (i = 0; i != flow->rss.queue_num; ++i) {
673 		int idx = (*flow->rss.queue)[i];
674 		struct mlx5_rxq_ctrl *rxq_ctrl =
675 			container_of((*priv->rxqs)[idx],
676 				     struct mlx5_rxq_ctrl, rxq);
677 
678 		if (mark) {
679 			rxq_ctrl->rxq.mark = 1;
680 			rxq_ctrl->flow_mark_n++;
681 		}
682 		if (tunnel) {
683 			unsigned int j;
684 
685 			/* Increase the counter matching the flow. */
686 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
687 				if ((tunnels_info[j].tunnel &
688 				     dev_flow->layers) ==
689 				    tunnels_info[j].tunnel) {
690 					rxq_ctrl->flow_tunnels_n[j]++;
691 					break;
692 				}
693 			}
694 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
695 		}
696 	}
697 }
698 
699 /**
700  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
701  *
702  * @param[in] dev
703  *   Pointer to the Ethernet device structure.
704  * @param[in] flow
705  *   Pointer to flow structure.
706  */
707 static void
708 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
709 {
710 	struct mlx5_flow *dev_flow;
711 
712 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
713 		flow_drv_rxq_flags_set(dev, dev_flow);
714 }
715 
716 /**
717  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
718  * device flow if no other flow uses it with the same kind of request.
719  *
720  * @param dev
721  *   Pointer to Ethernet device.
722  * @param[in] dev_flow
723  *   Pointer to the device flow.
724  */
725 static void
726 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
727 {
728 	struct mlx5_priv *priv = dev->data->dev_private;
729 	struct rte_flow *flow = dev_flow->flow;
730 	const int mark = !!(dev_flow->actions &
731 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
732 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
733 	unsigned int i;
734 
735 	assert(dev->data->dev_started);
736 	for (i = 0; i != flow->rss.queue_num; ++i) {
737 		int idx = (*flow->rss.queue)[i];
738 		struct mlx5_rxq_ctrl *rxq_ctrl =
739 			container_of((*priv->rxqs)[idx],
740 				     struct mlx5_rxq_ctrl, rxq);
741 
742 		if (mark) {
743 			rxq_ctrl->flow_mark_n--;
744 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
745 		}
746 		if (tunnel) {
747 			unsigned int j;
748 
749 			/* Decrease the counter matching the flow. */
750 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
751 				if ((tunnels_info[j].tunnel &
752 				     dev_flow->layers) ==
753 				    tunnels_info[j].tunnel) {
754 					rxq_ctrl->flow_tunnels_n[j]--;
755 					break;
756 				}
757 			}
758 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
759 		}
760 	}
761 }
762 
763 /**
764  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
765  * @p flow if no other flow uses it with the same kind of request.
766  *
767  * @param dev
768  *   Pointer to Ethernet device.
769  * @param[in] flow
770  *   Pointer to the flow.
771  */
772 static void
773 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
774 {
775 	struct mlx5_flow *dev_flow;
776 
777 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
778 		flow_drv_rxq_flags_trim(dev, dev_flow);
779 }
780 
781 /**
782  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
783  *
784  * @param dev
785  *   Pointer to Ethernet device.
786  */
787 static void
788 flow_rxq_flags_clear(struct rte_eth_dev *dev)
789 {
790 	struct mlx5_priv *priv = dev->data->dev_private;
791 	unsigned int i;
792 
793 	for (i = 0; i != priv->rxqs_n; ++i) {
794 		struct mlx5_rxq_ctrl *rxq_ctrl;
795 		unsigned int j;
796 
797 		if (!(*priv->rxqs)[i])
798 			continue;
799 		rxq_ctrl = container_of((*priv->rxqs)[i],
800 					struct mlx5_rxq_ctrl, rxq);
801 		rxq_ctrl->flow_mark_n = 0;
802 		rxq_ctrl->rxq.mark = 0;
803 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
804 			rxq_ctrl->flow_tunnels_n[j] = 0;
805 		rxq_ctrl->rxq.tunnel = 0;
806 	}
807 }
808 
809 /*
810  * return a pointer to the desired action in the list of actions.
811  *
812  * @param[in] actions
813  *   The list of actions to search the action in.
814  * @param[in] action
815  *   The action to find.
816  *
817  * @return
818  *   Pointer to the action in the list, if found. NULL otherwise.
819  */
820 const struct rte_flow_action *
821 mlx5_flow_find_action(const struct rte_flow_action *actions,
822 		      enum rte_flow_action_type action)
823 {
824 	if (actions == NULL)
825 		return NULL;
826 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
827 		if (actions->type == action)
828 			return actions;
829 	return NULL;
830 }
831 
832 /*
833  * Validate the flag action.
834  *
835  * @param[in] action_flags
836  *   Bit-fields that holds the actions detected until now.
837  * @param[in] attr
838  *   Attributes of flow that includes this action.
839  * @param[out] error
840  *   Pointer to error structure.
841  *
842  * @return
843  *   0 on success, a negative errno value otherwise and rte_errno is set.
844  */
845 int
846 mlx5_flow_validate_action_flag(uint64_t action_flags,
847 			       const struct rte_flow_attr *attr,
848 			       struct rte_flow_error *error)
849 {
850 
851 	if (action_flags & MLX5_FLOW_ACTION_DROP)
852 		return rte_flow_error_set(error, EINVAL,
853 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
854 					  "can't drop and flag in same flow");
855 	if (action_flags & MLX5_FLOW_ACTION_MARK)
856 		return rte_flow_error_set(error, EINVAL,
857 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
858 					  "can't mark and flag in same flow");
859 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
860 		return rte_flow_error_set(error, EINVAL,
861 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
862 					  "can't have 2 flag"
863 					  " actions in same flow");
864 	if (attr->egress)
865 		return rte_flow_error_set(error, ENOTSUP,
866 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
867 					  "flag action not supported for "
868 					  "egress");
869 	return 0;
870 }
871 
872 /*
873  * Validate the mark action.
874  *
875  * @param[in] action
876  *   Pointer to the queue action.
877  * @param[in] action_flags
878  *   Bit-fields that holds the actions detected until now.
879  * @param[in] attr
880  *   Attributes of flow that includes this action.
881  * @param[out] error
882  *   Pointer to error structure.
883  *
884  * @return
885  *   0 on success, a negative errno value otherwise and rte_errno is set.
886  */
887 int
888 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
889 			       uint64_t action_flags,
890 			       const struct rte_flow_attr *attr,
891 			       struct rte_flow_error *error)
892 {
893 	const struct rte_flow_action_mark *mark = action->conf;
894 
895 	if (!mark)
896 		return rte_flow_error_set(error, EINVAL,
897 					  RTE_FLOW_ERROR_TYPE_ACTION,
898 					  action,
899 					  "configuration cannot be null");
900 	if (mark->id >= MLX5_FLOW_MARK_MAX)
901 		return rte_flow_error_set(error, EINVAL,
902 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
903 					  &mark->id,
904 					  "mark id must in 0 <= id < "
905 					  RTE_STR(MLX5_FLOW_MARK_MAX));
906 	if (action_flags & MLX5_FLOW_ACTION_DROP)
907 		return rte_flow_error_set(error, EINVAL,
908 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
909 					  "can't drop and mark in same flow");
910 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
911 		return rte_flow_error_set(error, EINVAL,
912 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
913 					  "can't flag and mark in same flow");
914 	if (action_flags & MLX5_FLOW_ACTION_MARK)
915 		return rte_flow_error_set(error, EINVAL,
916 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
917 					  "can't have 2 mark actions in same"
918 					  " flow");
919 	if (attr->egress)
920 		return rte_flow_error_set(error, ENOTSUP,
921 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
922 					  "mark action not supported for "
923 					  "egress");
924 	return 0;
925 }
926 
927 /*
928  * Validate the drop action.
929  *
930  * @param[in] action_flags
931  *   Bit-fields that holds the actions detected until now.
932  * @param[in] attr
933  *   Attributes of flow that includes this action.
934  * @param[out] error
935  *   Pointer to error structure.
936  *
937  * @return
938  *   0 on success, a negative errno value otherwise and rte_errno is set.
939  */
940 int
941 mlx5_flow_validate_action_drop(uint64_t action_flags,
942 			       const struct rte_flow_attr *attr,
943 			       struct rte_flow_error *error)
944 {
945 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
946 		return rte_flow_error_set(error, EINVAL,
947 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
948 					  "can't drop and flag in same flow");
949 	if (action_flags & MLX5_FLOW_ACTION_MARK)
950 		return rte_flow_error_set(error, EINVAL,
951 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
952 					  "can't drop and mark in same flow");
953 	if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
954 			    MLX5_FLOW_FATE_ESWITCH_ACTIONS))
955 		return rte_flow_error_set(error, EINVAL,
956 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
957 					  "can't have 2 fate actions in"
958 					  " same flow");
959 	if (attr->egress)
960 		return rte_flow_error_set(error, ENOTSUP,
961 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
962 					  "drop action not supported for "
963 					  "egress");
964 	return 0;
965 }
966 
967 /*
968  * Validate the queue action.
969  *
970  * @param[in] action
971  *   Pointer to the queue action.
972  * @param[in] action_flags
973  *   Bit-fields that holds the actions detected until now.
974  * @param[in] dev
975  *   Pointer to the Ethernet device structure.
976  * @param[in] attr
977  *   Attributes of flow that includes this action.
978  * @param[out] error
979  *   Pointer to error structure.
980  *
981  * @return
982  *   0 on success, a negative errno value otherwise and rte_errno is set.
983  */
984 int
985 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
986 				uint64_t action_flags,
987 				struct rte_eth_dev *dev,
988 				const struct rte_flow_attr *attr,
989 				struct rte_flow_error *error)
990 {
991 	struct mlx5_priv *priv = dev->data->dev_private;
992 	const struct rte_flow_action_queue *queue = action->conf;
993 
994 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
995 		return rte_flow_error_set(error, EINVAL,
996 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
997 					  "can't have 2 fate actions in"
998 					  " same flow");
999 	if (!priv->rxqs_n)
1000 		return rte_flow_error_set(error, EINVAL,
1001 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1002 					  NULL, "No Rx queues configured");
1003 	if (queue->index >= priv->rxqs_n)
1004 		return rte_flow_error_set(error, EINVAL,
1005 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1006 					  &queue->index,
1007 					  "queue index out of range");
1008 	if (!(*priv->rxqs)[queue->index])
1009 		return rte_flow_error_set(error, EINVAL,
1010 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1011 					  &queue->index,
1012 					  "queue is not configured");
1013 	if (attr->egress)
1014 		return rte_flow_error_set(error, ENOTSUP,
1015 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1016 					  "queue action not supported for "
1017 					  "egress");
1018 	return 0;
1019 }
1020 
1021 /*
1022  * Validate the rss action.
1023  *
1024  * @param[in] action
1025  *   Pointer to the queue action.
1026  * @param[in] action_flags
1027  *   Bit-fields that holds the actions detected until now.
1028  * @param[in] dev
1029  *   Pointer to the Ethernet device structure.
1030  * @param[in] attr
1031  *   Attributes of flow that includes this action.
1032  * @param[in] item_flags
1033  *   Items that were detected.
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_action_rss(const struct rte_flow_action *action,
1042 			      uint64_t action_flags,
1043 			      struct rte_eth_dev *dev,
1044 			      const struct rte_flow_attr *attr,
1045 			      uint64_t item_flags,
1046 			      struct rte_flow_error *error)
1047 {
1048 	struct mlx5_priv *priv = dev->data->dev_private;
1049 	const struct rte_flow_action_rss *rss = action->conf;
1050 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1051 	unsigned int i;
1052 
1053 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1054 		return rte_flow_error_set(error, EINVAL,
1055 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1056 					  "can't have 2 fate actions"
1057 					  " in same flow");
1058 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1059 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1060 		return rte_flow_error_set(error, ENOTSUP,
1061 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1062 					  &rss->func,
1063 					  "RSS hash function not supported");
1064 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1065 	if (rss->level > 2)
1066 #else
1067 	if (rss->level > 1)
1068 #endif
1069 		return rte_flow_error_set(error, ENOTSUP,
1070 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1071 					  &rss->level,
1072 					  "tunnel RSS is not supported");
1073 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1074 	if (rss->key_len == 0 && rss->key != NULL)
1075 		return rte_flow_error_set(error, ENOTSUP,
1076 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1077 					  &rss->key_len,
1078 					  "RSS hash key length 0");
1079 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1080 		return rte_flow_error_set(error, ENOTSUP,
1081 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1082 					  &rss->key_len,
1083 					  "RSS hash key too small");
1084 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1085 		return rte_flow_error_set(error, ENOTSUP,
1086 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1087 					  &rss->key_len,
1088 					  "RSS hash key too large");
1089 	if (rss->queue_num > priv->config.ind_table_max_size)
1090 		return rte_flow_error_set(error, ENOTSUP,
1091 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1092 					  &rss->queue_num,
1093 					  "number of queues too large");
1094 	if (rss->types & MLX5_RSS_HF_MASK)
1095 		return rte_flow_error_set(error, ENOTSUP,
1096 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1097 					  &rss->types,
1098 					  "some RSS protocols are not"
1099 					  " supported");
1100 	if (!priv->rxqs_n)
1101 		return rte_flow_error_set(error, EINVAL,
1102 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1103 					  NULL, "No Rx queues configured");
1104 	if (!rss->queue_num)
1105 		return rte_flow_error_set(error, EINVAL,
1106 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1107 					  NULL, "No queues configured");
1108 	for (i = 0; i != rss->queue_num; ++i) {
1109 		if (!(*priv->rxqs)[rss->queue[i]])
1110 			return rte_flow_error_set
1111 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1112 				 &rss->queue[i], "queue is not configured");
1113 	}
1114 	if (attr->egress)
1115 		return rte_flow_error_set(error, ENOTSUP,
1116 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1117 					  "rss action not supported for "
1118 					  "egress");
1119 	if (rss->level > 1 &&  !tunnel)
1120 		return rte_flow_error_set(error, EINVAL,
1121 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1122 					  "inner RSS is not supported for "
1123 					  "non-tunnel flows");
1124 	return 0;
1125 }
1126 
1127 /*
1128  * Validate the count action.
1129  *
1130  * @param[in] dev
1131  *   Pointer to the Ethernet device structure.
1132  * @param[in] attr
1133  *   Attributes of flow that includes this action.
1134  * @param[out] error
1135  *   Pointer to error structure.
1136  *
1137  * @return
1138  *   0 on success, a negative errno value otherwise and rte_errno is set.
1139  */
1140 int
1141 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1142 				const struct rte_flow_attr *attr,
1143 				struct rte_flow_error *error)
1144 {
1145 	if (attr->egress)
1146 		return rte_flow_error_set(error, ENOTSUP,
1147 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1148 					  "count action not supported for "
1149 					  "egress");
1150 	return 0;
1151 }
1152 
1153 /**
1154  * Verify the @p attributes will be correctly understood by the NIC and store
1155  * them in the @p flow if everything is correct.
1156  *
1157  * @param[in] dev
1158  *   Pointer to the Ethernet device structure.
1159  * @param[in] attributes
1160  *   Pointer to flow attributes
1161  * @param[out] error
1162  *   Pointer to error structure.
1163  *
1164  * @return
1165  *   0 on success, a negative errno value otherwise and rte_errno is set.
1166  */
1167 int
1168 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1169 			      const struct rte_flow_attr *attributes,
1170 			      struct rte_flow_error *error)
1171 {
1172 	struct mlx5_priv *priv = dev->data->dev_private;
1173 	uint32_t priority_max = priv->config.flow_prio - 1;
1174 
1175 	if (attributes->group)
1176 		return rte_flow_error_set(error, ENOTSUP,
1177 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1178 					  NULL, "groups is not supported");
1179 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1180 	    attributes->priority >= priority_max)
1181 		return rte_flow_error_set(error, ENOTSUP,
1182 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1183 					  NULL, "priority out of range");
1184 	if (attributes->egress)
1185 		return rte_flow_error_set(error, ENOTSUP,
1186 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1187 					  "egress is not supported");
1188 	if (attributes->transfer && !priv->config.dv_esw_en)
1189 		return rte_flow_error_set(error, ENOTSUP,
1190 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1191 					  NULL, "transfer is not supported");
1192 	if (!attributes->ingress)
1193 		return rte_flow_error_set(error, EINVAL,
1194 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1195 					  NULL,
1196 					  "ingress attribute is mandatory");
1197 	return 0;
1198 }
1199 
1200 /**
1201  * Validate ICMP6 item.
1202  *
1203  * @param[in] item
1204  *   Item specification.
1205  * @param[in] item_flags
1206  *   Bit-fields that holds the items detected until now.
1207  * @param[out] error
1208  *   Pointer to error structure.
1209  *
1210  * @return
1211  *   0 on success, a negative errno value otherwise and rte_errno is set.
1212  */
1213 int
1214 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1215 			       uint64_t item_flags,
1216 			       uint8_t target_protocol,
1217 			       struct rte_flow_error *error)
1218 {
1219 	const struct rte_flow_item_icmp6 *mask = item->mask;
1220 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1221 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1222 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1223 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1224 				      MLX5_FLOW_LAYER_OUTER_L4;
1225 	int ret;
1226 
1227 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1228 		return rte_flow_error_set(error, EINVAL,
1229 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1230 					  "protocol filtering not compatible"
1231 					  " with ICMP6 layer");
1232 	if (!(item_flags & l3m))
1233 		return rte_flow_error_set(error, EINVAL,
1234 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1235 					  "IPv6 is mandatory to filter on"
1236 					  " ICMP6");
1237 	if (item_flags & l4m)
1238 		return rte_flow_error_set(error, EINVAL,
1239 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1240 					  "multiple L4 layers not supported");
1241 	if (!mask)
1242 		mask = &rte_flow_item_icmp6_mask;
1243 	ret = mlx5_flow_item_acceptable
1244 		(item, (const uint8_t *)mask,
1245 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1246 		 sizeof(struct rte_flow_item_icmp6), error);
1247 	if (ret < 0)
1248 		return ret;
1249 	return 0;
1250 }
1251 
1252 /**
1253  * Validate ICMP item.
1254  *
1255  * @param[in] item
1256  *   Item specification.
1257  * @param[in] item_flags
1258  *   Bit-fields that holds the items detected until now.
1259  * @param[out] error
1260  *   Pointer to error structure.
1261  *
1262  * @return
1263  *   0 on success, a negative errno value otherwise and rte_errno is set.
1264  */
1265 int
1266 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1267 			     uint64_t item_flags,
1268 			     uint8_t target_protocol,
1269 			     struct rte_flow_error *error)
1270 {
1271 	const struct rte_flow_item_icmp *mask = item->mask;
1272 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1273 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1274 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1275 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1276 				      MLX5_FLOW_LAYER_OUTER_L4;
1277 	int ret;
1278 
1279 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1280 		return rte_flow_error_set(error, EINVAL,
1281 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1282 					  "protocol filtering not compatible"
1283 					  " with ICMP layer");
1284 	if (!(item_flags & l3m))
1285 		return rte_flow_error_set(error, EINVAL,
1286 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1287 					  "IPv4 is mandatory to filter"
1288 					  " on ICMP");
1289 	if (item_flags & l4m)
1290 		return rte_flow_error_set(error, EINVAL,
1291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1292 					  "multiple L4 layers not supported");
1293 	if (!mask)
1294 		mask = &rte_flow_item_icmp_mask;
1295 	ret = mlx5_flow_item_acceptable
1296 		(item, (const uint8_t *)mask,
1297 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1298 		 sizeof(struct rte_flow_item_icmp), error);
1299 	if (ret < 0)
1300 		return ret;
1301 	return 0;
1302 }
1303 
1304 /**
1305  * Validate Ethernet 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[out] error
1312  *   Pointer to error structure.
1313  *
1314  * @return
1315  *   0 on success, a negative errno value otherwise and rte_errno is set.
1316  */
1317 int
1318 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1319 			    uint64_t item_flags,
1320 			    struct rte_flow_error *error)
1321 {
1322 	const struct rte_flow_item_eth *mask = item->mask;
1323 	const struct rte_flow_item_eth nic_mask = {
1324 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1325 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1326 		.type = RTE_BE16(0xffff),
1327 	};
1328 	int ret;
1329 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1330 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1331 				       MLX5_FLOW_LAYER_OUTER_L2;
1332 
1333 	if (item_flags & ethm)
1334 		return rte_flow_error_set(error, ENOTSUP,
1335 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1336 					  "multiple L2 layers not supported");
1337 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1338 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1339 		return rte_flow_error_set(error, EINVAL,
1340 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1341 					  "L2 layer should not follow "
1342 					  "L3 layers");
1343 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1344 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1345 		return rte_flow_error_set(error, EINVAL,
1346 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1347 					  "L2 layer should not follow VLAN");
1348 	if (!mask)
1349 		mask = &rte_flow_item_eth_mask;
1350 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1351 					(const uint8_t *)&nic_mask,
1352 					sizeof(struct rte_flow_item_eth),
1353 					error);
1354 	return ret;
1355 }
1356 
1357 /**
1358  * Validate VLAN 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] dev
1365  *   Ethernet device flow is being created on.
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_vlan(const struct rte_flow_item *item,
1374 			     uint64_t item_flags,
1375 			     struct rte_eth_dev *dev,
1376 			     struct rte_flow_error *error)
1377 {
1378 	const struct rte_flow_item_vlan *spec = item->spec;
1379 	const struct rte_flow_item_vlan *mask = item->mask;
1380 	const struct rte_flow_item_vlan nic_mask = {
1381 		.tci = RTE_BE16(UINT16_MAX),
1382 		.inner_type = RTE_BE16(UINT16_MAX),
1383 	};
1384 	uint16_t vlan_tag = 0;
1385 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1386 	int ret;
1387 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1388 					MLX5_FLOW_LAYER_INNER_L4) :
1389 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1390 					MLX5_FLOW_LAYER_OUTER_L4);
1391 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1392 					MLX5_FLOW_LAYER_OUTER_VLAN;
1393 
1394 	if (item_flags & vlanm)
1395 		return rte_flow_error_set(error, EINVAL,
1396 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1397 					  "multiple VLAN layers not supported");
1398 	else if ((item_flags & l34m) != 0)
1399 		return rte_flow_error_set(error, EINVAL,
1400 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1401 					  "VLAN cannot follow L3/L4 layer");
1402 	if (!mask)
1403 		mask = &rte_flow_item_vlan_mask;
1404 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1405 					(const uint8_t *)&nic_mask,
1406 					sizeof(struct rte_flow_item_vlan),
1407 					error);
1408 	if (ret)
1409 		return ret;
1410 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1411 		struct mlx5_priv *priv = dev->data->dev_private;
1412 
1413 		if (priv->vmwa_context) {
1414 			/*
1415 			 * Non-NULL context means we have a virtual machine
1416 			 * and SR-IOV enabled, we have to create VLAN interface
1417 			 * to make hypervisor to setup E-Switch vport
1418 			 * context correctly. We avoid creating the multiple
1419 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1420 			 */
1421 			return rte_flow_error_set(error, EINVAL,
1422 						  RTE_FLOW_ERROR_TYPE_ITEM,
1423 						  item,
1424 						  "VLAN tag mask is not"
1425 						  " supported in virtual"
1426 						  " environment");
1427 		}
1428 	}
1429 	if (spec) {
1430 		vlan_tag = spec->tci;
1431 		vlan_tag &= mask->tci;
1432 	}
1433 	/*
1434 	 * From verbs perspective an empty VLAN is equivalent
1435 	 * to a packet without VLAN layer.
1436 	 */
1437 	if (!vlan_tag)
1438 		return rte_flow_error_set(error, EINVAL,
1439 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1440 					  item->spec,
1441 					  "VLAN cannot be empty");
1442 	return 0;
1443 }
1444 
1445 /**
1446  * Validate IPV4 item.
1447  *
1448  * @param[in] item
1449  *   Item specification.
1450  * @param[in] item_flags
1451  *   Bit-fields that holds the items detected until now.
1452  * @param[in] acc_mask
1453  *   Acceptable mask, if NULL default internal default mask
1454  *   will be used to check whether item fields are supported.
1455  * @param[out] error
1456  *   Pointer to error structure.
1457  *
1458  * @return
1459  *   0 on success, a negative errno value otherwise and rte_errno is set.
1460  */
1461 int
1462 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1463 			     uint64_t item_flags,
1464 			     uint64_t last_item,
1465 			     uint16_t ether_type,
1466 			     const struct rte_flow_item_ipv4 *acc_mask,
1467 			     struct rte_flow_error *error)
1468 {
1469 	const struct rte_flow_item_ipv4 *mask = item->mask;
1470 	const struct rte_flow_item_ipv4 *spec = item->spec;
1471 	const struct rte_flow_item_ipv4 nic_mask = {
1472 		.hdr = {
1473 			.src_addr = RTE_BE32(0xffffffff),
1474 			.dst_addr = RTE_BE32(0xffffffff),
1475 			.type_of_service = 0xff,
1476 			.next_proto_id = 0xff,
1477 		},
1478 	};
1479 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1480 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1481 				      MLX5_FLOW_LAYER_OUTER_L3;
1482 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1483 				      MLX5_FLOW_LAYER_OUTER_L4;
1484 	int ret;
1485 	uint8_t next_proto = 0xFF;
1486 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1487 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1488 				  MLX5_FLOW_LAYER_INNER_VLAN);
1489 
1490 	if ((last_item & l2_vlan) && ether_type &&
1491 	    ether_type != RTE_ETHER_TYPE_IPV4)
1492 		return rte_flow_error_set(error, EINVAL,
1493 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1494 					  "IPv4 cannot follow L2/VLAN layer "
1495 					  "which ether type is not IPv4");
1496 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1497 		if (mask && spec)
1498 			next_proto = mask->hdr.next_proto_id &
1499 				     spec->hdr.next_proto_id;
1500 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1501 			return rte_flow_error_set(error, EINVAL,
1502 						  RTE_FLOW_ERROR_TYPE_ITEM,
1503 						  item,
1504 						  "multiple tunnel "
1505 						  "not supported");
1506 	}
1507 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1508 		return rte_flow_error_set(error, EINVAL,
1509 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1510 					  "wrong tunnel type - IPv6 specified "
1511 					  "but IPv4 item provided");
1512 	if (item_flags & l3m)
1513 		return rte_flow_error_set(error, ENOTSUP,
1514 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1515 					  "multiple L3 layers not supported");
1516 	else if (item_flags & l4m)
1517 		return rte_flow_error_set(error, EINVAL,
1518 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1519 					  "L3 cannot follow an L4 layer.");
1520 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1521 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1522 		return rte_flow_error_set(error, EINVAL,
1523 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1524 					  "L3 cannot follow an NVGRE layer.");
1525 	if (!mask)
1526 		mask = &rte_flow_item_ipv4_mask;
1527 	else if (mask->hdr.next_proto_id != 0 &&
1528 		 mask->hdr.next_proto_id != 0xff)
1529 		return rte_flow_error_set(error, EINVAL,
1530 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1531 					  "partial mask is not supported"
1532 					  " for protocol");
1533 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1534 					acc_mask ? (const uint8_t *)acc_mask
1535 						 : (const uint8_t *)&nic_mask,
1536 					sizeof(struct rte_flow_item_ipv4),
1537 					error);
1538 	if (ret < 0)
1539 		return ret;
1540 	return 0;
1541 }
1542 
1543 /**
1544  * Validate IPV6 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] acc_mask
1551  *   Acceptable mask, if NULL default internal default mask
1552  *   will be used to check whether item fields are supported.
1553  * @param[out] error
1554  *   Pointer to error structure.
1555  *
1556  * @return
1557  *   0 on success, a negative errno value otherwise and rte_errno is set.
1558  */
1559 int
1560 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1561 			     uint64_t item_flags,
1562 			     uint64_t last_item,
1563 			     uint16_t ether_type,
1564 			     const struct rte_flow_item_ipv6 *acc_mask,
1565 			     struct rte_flow_error *error)
1566 {
1567 	const struct rte_flow_item_ipv6 *mask = item->mask;
1568 	const struct rte_flow_item_ipv6 *spec = item->spec;
1569 	const struct rte_flow_item_ipv6 nic_mask = {
1570 		.hdr = {
1571 			.src_addr =
1572 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1573 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1574 			.dst_addr =
1575 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1576 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1577 			.vtc_flow = RTE_BE32(0xffffffff),
1578 			.proto = 0xff,
1579 			.hop_limits = 0xff,
1580 		},
1581 	};
1582 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1583 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1584 				      MLX5_FLOW_LAYER_OUTER_L3;
1585 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1586 				      MLX5_FLOW_LAYER_OUTER_L4;
1587 	int ret;
1588 	uint8_t next_proto = 0xFF;
1589 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1590 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1591 				  MLX5_FLOW_LAYER_INNER_VLAN);
1592 
1593 	if ((last_item & l2_vlan) && ether_type &&
1594 	    ether_type != RTE_ETHER_TYPE_IPV6)
1595 		return rte_flow_error_set(error, EINVAL,
1596 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1597 					  "IPv6 cannot follow L2/VLAN layer "
1598 					  "which ether type is not IPv6");
1599 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1600 		if (mask && spec)
1601 			next_proto = mask->hdr.proto & spec->hdr.proto;
1602 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1603 			return rte_flow_error_set(error, EINVAL,
1604 						  RTE_FLOW_ERROR_TYPE_ITEM,
1605 						  item,
1606 						  "multiple tunnel "
1607 						  "not supported");
1608 	}
1609 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1610 		return rte_flow_error_set(error, EINVAL,
1611 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1612 					  "wrong tunnel type - IPv4 specified "
1613 					  "but IPv6 item provided");
1614 	if (item_flags & l3m)
1615 		return rte_flow_error_set(error, ENOTSUP,
1616 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1617 					  "multiple L3 layers not supported");
1618 	else if (item_flags & l4m)
1619 		return rte_flow_error_set(error, EINVAL,
1620 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1621 					  "L3 cannot follow an L4 layer.");
1622 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1623 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1624 		return rte_flow_error_set(error, EINVAL,
1625 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1626 					  "L3 cannot follow an NVGRE layer.");
1627 	if (!mask)
1628 		mask = &rte_flow_item_ipv6_mask;
1629 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1630 					acc_mask ? (const uint8_t *)acc_mask
1631 						 : (const uint8_t *)&nic_mask,
1632 					sizeof(struct rte_flow_item_ipv6),
1633 					error);
1634 	if (ret < 0)
1635 		return ret;
1636 	return 0;
1637 }
1638 
1639 /**
1640  * Validate UDP item.
1641  *
1642  * @param[in] item
1643  *   Item specification.
1644  * @param[in] item_flags
1645  *   Bit-fields that holds the items detected until now.
1646  * @param[in] target_protocol
1647  *   The next protocol in the previous item.
1648  * @param[in] flow_mask
1649  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1650  * @param[out] error
1651  *   Pointer to error structure.
1652  *
1653  * @return
1654  *   0 on success, a negative errno value otherwise and rte_errno is set.
1655  */
1656 int
1657 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1658 			    uint64_t item_flags,
1659 			    uint8_t target_protocol,
1660 			    struct rte_flow_error *error)
1661 {
1662 	const struct rte_flow_item_udp *mask = item->mask;
1663 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1664 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1665 				      MLX5_FLOW_LAYER_OUTER_L3;
1666 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1667 				      MLX5_FLOW_LAYER_OUTER_L4;
1668 	int ret;
1669 
1670 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1671 		return rte_flow_error_set(error, EINVAL,
1672 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1673 					  "protocol filtering not compatible"
1674 					  " with UDP layer");
1675 	if (!(item_flags & l3m))
1676 		return rte_flow_error_set(error, EINVAL,
1677 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1678 					  "L3 is mandatory to filter on L4");
1679 	if (item_flags & l4m)
1680 		return rte_flow_error_set(error, EINVAL,
1681 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1682 					  "multiple L4 layers not supported");
1683 	if (!mask)
1684 		mask = &rte_flow_item_udp_mask;
1685 	ret = mlx5_flow_item_acceptable
1686 		(item, (const uint8_t *)mask,
1687 		 (const uint8_t *)&rte_flow_item_udp_mask,
1688 		 sizeof(struct rte_flow_item_udp), error);
1689 	if (ret < 0)
1690 		return ret;
1691 	return 0;
1692 }
1693 
1694 /**
1695  * Validate TCP item.
1696  *
1697  * @param[in] item
1698  *   Item specification.
1699  * @param[in] item_flags
1700  *   Bit-fields that holds the items detected until now.
1701  * @param[in] target_protocol
1702  *   The next protocol in the previous item.
1703  * @param[out] error
1704  *   Pointer to error structure.
1705  *
1706  * @return
1707  *   0 on success, a negative errno value otherwise and rte_errno is set.
1708  */
1709 int
1710 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1711 			    uint64_t item_flags,
1712 			    uint8_t target_protocol,
1713 			    const struct rte_flow_item_tcp *flow_mask,
1714 			    struct rte_flow_error *error)
1715 {
1716 	const struct rte_flow_item_tcp *mask = item->mask;
1717 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1718 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1719 				      MLX5_FLOW_LAYER_OUTER_L3;
1720 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1721 				      MLX5_FLOW_LAYER_OUTER_L4;
1722 	int ret;
1723 
1724 	assert(flow_mask);
1725 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1726 		return rte_flow_error_set(error, EINVAL,
1727 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1728 					  "protocol filtering not compatible"
1729 					  " with TCP layer");
1730 	if (!(item_flags & l3m))
1731 		return rte_flow_error_set(error, EINVAL,
1732 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1733 					  "L3 is mandatory to filter on L4");
1734 	if (item_flags & l4m)
1735 		return rte_flow_error_set(error, EINVAL,
1736 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1737 					  "multiple L4 layers not supported");
1738 	if (!mask)
1739 		mask = &rte_flow_item_tcp_mask;
1740 	ret = mlx5_flow_item_acceptable
1741 		(item, (const uint8_t *)mask,
1742 		 (const uint8_t *)flow_mask,
1743 		 sizeof(struct rte_flow_item_tcp), error);
1744 	if (ret < 0)
1745 		return ret;
1746 	return 0;
1747 }
1748 
1749 /**
1750  * Validate VXLAN item.
1751  *
1752  * @param[in] item
1753  *   Item specification.
1754  * @param[in] item_flags
1755  *   Bit-fields that holds the items detected until now.
1756  * @param[in] target_protocol
1757  *   The next protocol in the previous item.
1758  * @param[out] error
1759  *   Pointer to error structure.
1760  *
1761  * @return
1762  *   0 on success, a negative errno value otherwise and rte_errno is set.
1763  */
1764 int
1765 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1766 			      uint64_t item_flags,
1767 			      struct rte_flow_error *error)
1768 {
1769 	const struct rte_flow_item_vxlan *spec = item->spec;
1770 	const struct rte_flow_item_vxlan *mask = item->mask;
1771 	int ret;
1772 	union vni {
1773 		uint32_t vlan_id;
1774 		uint8_t vni[4];
1775 	} id = { .vlan_id = 0, };
1776 	uint32_t vlan_id = 0;
1777 
1778 
1779 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1780 		return rte_flow_error_set(error, ENOTSUP,
1781 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1782 					  "multiple tunnel layers not"
1783 					  " supported");
1784 	/*
1785 	 * Verify only UDPv4 is present as defined in
1786 	 * https://tools.ietf.org/html/rfc7348
1787 	 */
1788 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1789 		return rte_flow_error_set(error, EINVAL,
1790 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1791 					  "no outer UDP layer found");
1792 	if (!mask)
1793 		mask = &rte_flow_item_vxlan_mask;
1794 	ret = mlx5_flow_item_acceptable
1795 		(item, (const uint8_t *)mask,
1796 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1797 		 sizeof(struct rte_flow_item_vxlan),
1798 		 error);
1799 	if (ret < 0)
1800 		return ret;
1801 	if (spec) {
1802 		memcpy(&id.vni[1], spec->vni, 3);
1803 		vlan_id = id.vlan_id;
1804 		memcpy(&id.vni[1], mask->vni, 3);
1805 		vlan_id &= id.vlan_id;
1806 	}
1807 	/*
1808 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1809 	 * only this layer is defined in the Verbs specification it is
1810 	 * interpreted as wildcard and all packets will match this
1811 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1812 	 * udp), all packets matching the layers before will also
1813 	 * match this rule.  To avoid such situation, VNI 0 is
1814 	 * currently refused.
1815 	 */
1816 	if (!vlan_id)
1817 		return rte_flow_error_set(error, ENOTSUP,
1818 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1819 					  "VXLAN vni cannot be 0");
1820 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1821 		return rte_flow_error_set(error, ENOTSUP,
1822 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1823 					  "VXLAN tunnel must be fully defined");
1824 	return 0;
1825 }
1826 
1827 /**
1828  * Validate VXLAN_GPE item.
1829  *
1830  * @param[in] item
1831  *   Item specification.
1832  * @param[in] item_flags
1833  *   Bit-fields that holds the items detected until now.
1834  * @param[in] priv
1835  *   Pointer to the private data structure.
1836  * @param[in] target_protocol
1837  *   The next protocol in the previous item.
1838  * @param[out] error
1839  *   Pointer to error structure.
1840  *
1841  * @return
1842  *   0 on success, a negative errno value otherwise and rte_errno is set.
1843  */
1844 int
1845 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1846 				  uint64_t item_flags,
1847 				  struct rte_eth_dev *dev,
1848 				  struct rte_flow_error *error)
1849 {
1850 	struct mlx5_priv *priv = dev->data->dev_private;
1851 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1852 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1853 	int ret;
1854 	union vni {
1855 		uint32_t vlan_id;
1856 		uint8_t vni[4];
1857 	} id = { .vlan_id = 0, };
1858 	uint32_t vlan_id = 0;
1859 
1860 	if (!priv->config.l3_vxlan_en)
1861 		return rte_flow_error_set(error, ENOTSUP,
1862 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1863 					  "L3 VXLAN is not enabled by device"
1864 					  " parameter and/or not configured in"
1865 					  " firmware");
1866 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1867 		return rte_flow_error_set(error, ENOTSUP,
1868 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1869 					  "multiple tunnel layers not"
1870 					  " supported");
1871 	/*
1872 	 * Verify only UDPv4 is present as defined in
1873 	 * https://tools.ietf.org/html/rfc7348
1874 	 */
1875 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1876 		return rte_flow_error_set(error, EINVAL,
1877 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1878 					  "no outer UDP layer found");
1879 	if (!mask)
1880 		mask = &rte_flow_item_vxlan_gpe_mask;
1881 	ret = mlx5_flow_item_acceptable
1882 		(item, (const uint8_t *)mask,
1883 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1884 		 sizeof(struct rte_flow_item_vxlan_gpe),
1885 		 error);
1886 	if (ret < 0)
1887 		return ret;
1888 	if (spec) {
1889 		if (spec->protocol)
1890 			return rte_flow_error_set(error, ENOTSUP,
1891 						  RTE_FLOW_ERROR_TYPE_ITEM,
1892 						  item,
1893 						  "VxLAN-GPE protocol"
1894 						  " not supported");
1895 		memcpy(&id.vni[1], spec->vni, 3);
1896 		vlan_id = id.vlan_id;
1897 		memcpy(&id.vni[1], mask->vni, 3);
1898 		vlan_id &= id.vlan_id;
1899 	}
1900 	/*
1901 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1902 	 * layer is defined in the Verbs specification it is interpreted as
1903 	 * wildcard and all packets will match this rule, if it follows a full
1904 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1905 	 * before will also match this rule.  To avoid such situation, VNI 0
1906 	 * is currently refused.
1907 	 */
1908 	if (!vlan_id)
1909 		return rte_flow_error_set(error, ENOTSUP,
1910 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1911 					  "VXLAN-GPE vni cannot be 0");
1912 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1913 		return rte_flow_error_set(error, ENOTSUP,
1914 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1915 					  "VXLAN-GPE tunnel must be fully"
1916 					  " defined");
1917 	return 0;
1918 }
1919 /**
1920  * Validate GRE Key item.
1921  *
1922  * @param[in] item
1923  *   Item specification.
1924  * @param[in] item_flags
1925  *   Bit flags to mark detected items.
1926  * @param[in] gre_item
1927  *   Pointer to gre_item
1928  * @param[out] error
1929  *   Pointer to error structure.
1930  *
1931  * @return
1932  *   0 on success, a negative errno value otherwise and rte_errno is set.
1933  */
1934 int
1935 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1936 				uint64_t item_flags,
1937 				const struct rte_flow_item *gre_item,
1938 				struct rte_flow_error *error)
1939 {
1940 	const rte_be32_t *mask = item->mask;
1941 	int ret = 0;
1942 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1943 	const struct rte_flow_item_gre *gre_spec = gre_item->spec;
1944 	const struct rte_flow_item_gre *gre_mask = gre_item->mask;
1945 
1946 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1947 		return rte_flow_error_set(error, ENOTSUP,
1948 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1949 					  "Multiple GRE key not support");
1950 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1951 		return rte_flow_error_set(error, ENOTSUP,
1952 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1953 					  "No preceding GRE header");
1954 	if (item_flags & MLX5_FLOW_LAYER_INNER)
1955 		return rte_flow_error_set(error, ENOTSUP,
1956 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1957 					  "GRE key following a wrong item");
1958 	if (!gre_mask)
1959 		gre_mask = &rte_flow_item_gre_mask;
1960 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
1961 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
1962 		return rte_flow_error_set(error, EINVAL,
1963 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1964 					  "Key bit must be on");
1965 
1966 	if (!mask)
1967 		mask = &gre_key_default_mask;
1968 	ret = mlx5_flow_item_acceptable
1969 		(item, (const uint8_t *)mask,
1970 		 (const uint8_t *)&gre_key_default_mask,
1971 		 sizeof(rte_be32_t), error);
1972 	return ret;
1973 }
1974 
1975 /**
1976  * Validate GRE item.
1977  *
1978  * @param[in] item
1979  *   Item specification.
1980  * @param[in] item_flags
1981  *   Bit flags to mark detected items.
1982  * @param[in] target_protocol
1983  *   The next protocol in the previous item.
1984  * @param[out] error
1985  *   Pointer to error structure.
1986  *
1987  * @return
1988  *   0 on success, a negative errno value otherwise and rte_errno is set.
1989  */
1990 int
1991 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1992 			    uint64_t item_flags,
1993 			    uint8_t target_protocol,
1994 			    struct rte_flow_error *error)
1995 {
1996 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1997 	const struct rte_flow_item_gre *mask = item->mask;
1998 	int ret;
1999 	const struct rte_flow_item_gre nic_mask = {
2000 		.c_rsvd0_ver = RTE_BE16(0xB000),
2001 		.protocol = RTE_BE16(UINT16_MAX),
2002 	};
2003 
2004 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2005 		return rte_flow_error_set(error, EINVAL,
2006 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2007 					  "protocol filtering not compatible"
2008 					  " with this GRE layer");
2009 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2010 		return rte_flow_error_set(error, ENOTSUP,
2011 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2012 					  "multiple tunnel layers not"
2013 					  " supported");
2014 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2015 		return rte_flow_error_set(error, ENOTSUP,
2016 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2017 					  "L3 Layer is missing");
2018 	if (!mask)
2019 		mask = &rte_flow_item_gre_mask;
2020 	ret = mlx5_flow_item_acceptable
2021 		(item, (const uint8_t *)mask,
2022 		 (const uint8_t *)&nic_mask,
2023 		 sizeof(struct rte_flow_item_gre), error);
2024 	if (ret < 0)
2025 		return ret;
2026 #ifndef HAVE_MLX5DV_DR
2027 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2028 	if (spec && (spec->protocol & mask->protocol))
2029 		return rte_flow_error_set(error, ENOTSUP,
2030 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2031 					  "without MPLS support the"
2032 					  " specification cannot be used for"
2033 					  " filtering");
2034 #endif
2035 #endif
2036 	return 0;
2037 }
2038 
2039 /**
2040  * Validate Geneve item.
2041  *
2042  * @param[in] item
2043  *   Item specification.
2044  * @param[in] itemFlags
2045  *   Bit-fields that holds the items detected until now.
2046  * @param[in] enPriv
2047  *   Pointer to the private data structure.
2048  * @param[out] error
2049  *   Pointer to error structure.
2050  *
2051  * @return
2052  *   0 on success, a negative errno value otherwise and rte_errno is set.
2053  */
2054 
2055 int
2056 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2057 			       uint64_t item_flags,
2058 			       struct rte_eth_dev *dev,
2059 			       struct rte_flow_error *error)
2060 {
2061 	struct mlx5_priv *priv = dev->data->dev_private;
2062 	const struct rte_flow_item_geneve *spec = item->spec;
2063 	const struct rte_flow_item_geneve *mask = item->mask;
2064 	int ret;
2065 	uint16_t gbhdr;
2066 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2067 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2068 	const struct rte_flow_item_geneve nic_mask = {
2069 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2070 		.vni = "\xff\xff\xff",
2071 		.protocol = RTE_BE16(UINT16_MAX),
2072 	};
2073 
2074 	if (!(priv->config.hca_attr.flex_parser_protocols &
2075 	      MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2076 	    !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2077 		return rte_flow_error_set(error, ENOTSUP,
2078 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2079 					  "L3 Geneve is not enabled by device"
2080 					  " parameter and/or not configured in"
2081 					  " firmware");
2082 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2083 		return rte_flow_error_set(error, ENOTSUP,
2084 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2085 					  "multiple tunnel layers not"
2086 					  " supported");
2087 	/*
2088 	 * Verify only UDPv4 is present as defined in
2089 	 * https://tools.ietf.org/html/rfc7348
2090 	 */
2091 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2092 		return rte_flow_error_set(error, EINVAL,
2093 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2094 					  "no outer UDP layer found");
2095 	if (!mask)
2096 		mask = &rte_flow_item_geneve_mask;
2097 	ret = mlx5_flow_item_acceptable
2098 				  (item, (const uint8_t *)mask,
2099 				   (const uint8_t *)&nic_mask,
2100 				   sizeof(struct rte_flow_item_geneve), error);
2101 	if (ret)
2102 		return ret;
2103 	if (spec) {
2104 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2105 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2106 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2107 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2108 			return rte_flow_error_set(error, ENOTSUP,
2109 						  RTE_FLOW_ERROR_TYPE_ITEM,
2110 						  item,
2111 						  "Geneve protocol unsupported"
2112 						  " fields are being used");
2113 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2114 			return rte_flow_error_set
2115 					(error, ENOTSUP,
2116 					 RTE_FLOW_ERROR_TYPE_ITEM,
2117 					 item,
2118 					 "Unsupported Geneve options length");
2119 	}
2120 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2121 		return rte_flow_error_set
2122 				    (error, ENOTSUP,
2123 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2124 				     "Geneve tunnel must be fully defined");
2125 	return 0;
2126 }
2127 
2128 /**
2129  * Validate MPLS item.
2130  *
2131  * @param[in] dev
2132  *   Pointer to the rte_eth_dev structure.
2133  * @param[in] item
2134  *   Item specification.
2135  * @param[in] item_flags
2136  *   Bit-fields that holds the items detected until now.
2137  * @param[in] prev_layer
2138  *   The protocol layer indicated in previous item.
2139  * @param[out] error
2140  *   Pointer to error structure.
2141  *
2142  * @return
2143  *   0 on success, a negative errno value otherwise and rte_errno is set.
2144  */
2145 int
2146 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2147 			     const struct rte_flow_item *item __rte_unused,
2148 			     uint64_t item_flags __rte_unused,
2149 			     uint64_t prev_layer __rte_unused,
2150 			     struct rte_flow_error *error)
2151 {
2152 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2153 	const struct rte_flow_item_mpls *mask = item->mask;
2154 	struct mlx5_priv *priv = dev->data->dev_private;
2155 	int ret;
2156 
2157 	if (!priv->config.mpls_en)
2158 		return rte_flow_error_set(error, ENOTSUP,
2159 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2160 					  "MPLS not supported or"
2161 					  " disabled in firmware"
2162 					  " configuration.");
2163 	/* MPLS over IP, UDP, GRE is allowed */
2164 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2165 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2166 			    MLX5_FLOW_LAYER_GRE)))
2167 		return rte_flow_error_set(error, EINVAL,
2168 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2169 					  "protocol filtering not compatible"
2170 					  " with MPLS layer");
2171 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2172 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2173 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2174 		return rte_flow_error_set(error, ENOTSUP,
2175 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2176 					  "multiple tunnel layers not"
2177 					  " supported");
2178 	if (!mask)
2179 		mask = &rte_flow_item_mpls_mask;
2180 	ret = mlx5_flow_item_acceptable
2181 		(item, (const uint8_t *)mask,
2182 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2183 		 sizeof(struct rte_flow_item_mpls), error);
2184 	if (ret < 0)
2185 		return ret;
2186 	return 0;
2187 #endif
2188 	return rte_flow_error_set(error, ENOTSUP,
2189 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2190 				  "MPLS is not supported by Verbs, please"
2191 				  " update.");
2192 }
2193 
2194 /**
2195  * Validate NVGRE item.
2196  *
2197  * @param[in] item
2198  *   Item specification.
2199  * @param[in] item_flags
2200  *   Bit flags to mark detected items.
2201  * @param[in] target_protocol
2202  *   The next protocol in the previous item.
2203  * @param[out] error
2204  *   Pointer to error structure.
2205  *
2206  * @return
2207  *   0 on success, a negative errno value otherwise and rte_errno is set.
2208  */
2209 int
2210 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2211 			      uint64_t item_flags,
2212 			      uint8_t target_protocol,
2213 			      struct rte_flow_error *error)
2214 {
2215 	const struct rte_flow_item_nvgre *mask = item->mask;
2216 	int ret;
2217 
2218 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2219 		return rte_flow_error_set(error, EINVAL,
2220 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2221 					  "protocol filtering not compatible"
2222 					  " with this GRE layer");
2223 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2224 		return rte_flow_error_set(error, ENOTSUP,
2225 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2226 					  "multiple tunnel layers not"
2227 					  " supported");
2228 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2229 		return rte_flow_error_set(error, ENOTSUP,
2230 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2231 					  "L3 Layer is missing");
2232 	if (!mask)
2233 		mask = &rte_flow_item_nvgre_mask;
2234 	ret = mlx5_flow_item_acceptable
2235 		(item, (const uint8_t *)mask,
2236 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2237 		 sizeof(struct rte_flow_item_nvgre), error);
2238 	if (ret < 0)
2239 		return ret;
2240 	return 0;
2241 }
2242 
2243 static int
2244 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2245 		   const struct rte_flow_attr *attr __rte_unused,
2246 		   const struct rte_flow_item items[] __rte_unused,
2247 		   const struct rte_flow_action actions[] __rte_unused,
2248 		   bool external __rte_unused,
2249 		   struct rte_flow_error *error)
2250 {
2251 	return rte_flow_error_set(error, ENOTSUP,
2252 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2253 }
2254 
2255 static struct mlx5_flow *
2256 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2257 		  const struct rte_flow_item items[] __rte_unused,
2258 		  const struct rte_flow_action actions[] __rte_unused,
2259 		  struct rte_flow_error *error)
2260 {
2261 	rte_flow_error_set(error, ENOTSUP,
2262 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2263 	return NULL;
2264 }
2265 
2266 static int
2267 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2268 		    struct mlx5_flow *dev_flow __rte_unused,
2269 		    const struct rte_flow_attr *attr __rte_unused,
2270 		    const struct rte_flow_item items[] __rte_unused,
2271 		    const struct rte_flow_action actions[] __rte_unused,
2272 		    struct rte_flow_error *error)
2273 {
2274 	return rte_flow_error_set(error, ENOTSUP,
2275 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2276 }
2277 
2278 static int
2279 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2280 		struct rte_flow *flow __rte_unused,
2281 		struct rte_flow_error *error)
2282 {
2283 	return rte_flow_error_set(error, ENOTSUP,
2284 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2285 }
2286 
2287 static void
2288 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2289 		 struct rte_flow *flow __rte_unused)
2290 {
2291 }
2292 
2293 static void
2294 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2295 		  struct rte_flow *flow __rte_unused)
2296 {
2297 }
2298 
2299 static int
2300 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2301 		struct rte_flow *flow __rte_unused,
2302 		const struct rte_flow_action *actions __rte_unused,
2303 		void *data __rte_unused,
2304 		struct rte_flow_error *error)
2305 {
2306 	return rte_flow_error_set(error, ENOTSUP,
2307 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2308 }
2309 
2310 /* Void driver to protect from null pointer reference. */
2311 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2312 	.validate = flow_null_validate,
2313 	.prepare = flow_null_prepare,
2314 	.translate = flow_null_translate,
2315 	.apply = flow_null_apply,
2316 	.remove = flow_null_remove,
2317 	.destroy = flow_null_destroy,
2318 	.query = flow_null_query,
2319 };
2320 
2321 /**
2322  * Select flow driver type according to flow attributes and device
2323  * configuration.
2324  *
2325  * @param[in] dev
2326  *   Pointer to the dev structure.
2327  * @param[in] attr
2328  *   Pointer to the flow attributes.
2329  *
2330  * @return
2331  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2332  */
2333 static enum mlx5_flow_drv_type
2334 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2335 {
2336 	struct mlx5_priv *priv = dev->data->dev_private;
2337 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2338 
2339 	if (attr->transfer && priv->config.dv_esw_en)
2340 		type = MLX5_FLOW_TYPE_DV;
2341 	if (!attr->transfer)
2342 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2343 						 MLX5_FLOW_TYPE_VERBS;
2344 	return type;
2345 }
2346 
2347 #define flow_get_drv_ops(type) flow_drv_ops[type]
2348 
2349 /**
2350  * Flow driver validation API. This abstracts calling driver specific functions.
2351  * The type of flow driver is determined according to flow attributes.
2352  *
2353  * @param[in] dev
2354  *   Pointer to the dev structure.
2355  * @param[in] attr
2356  *   Pointer to the flow attributes.
2357  * @param[in] items
2358  *   Pointer to the list of items.
2359  * @param[in] actions
2360  *   Pointer to the list of actions.
2361  * @param[in] external
2362  *   This flow rule is created by request external to PMD.
2363  * @param[out] error
2364  *   Pointer to the error structure.
2365  *
2366  * @return
2367  *   0 on success, a negative errno value otherwise and rte_errno is set.
2368  */
2369 static inline int
2370 flow_drv_validate(struct rte_eth_dev *dev,
2371 		  const struct rte_flow_attr *attr,
2372 		  const struct rte_flow_item items[],
2373 		  const struct rte_flow_action actions[],
2374 		  bool external, struct rte_flow_error *error)
2375 {
2376 	const struct mlx5_flow_driver_ops *fops;
2377 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2378 
2379 	fops = flow_get_drv_ops(type);
2380 	return fops->validate(dev, attr, items, actions, external, error);
2381 }
2382 
2383 /**
2384  * Flow driver preparation API. This abstracts calling driver specific
2385  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2386  * calculates the size of memory required for device flow, allocates the memory,
2387  * initializes the device flow and returns the pointer.
2388  *
2389  * @note
2390  *   This function initializes device flow structure such as dv or verbs in
2391  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2392  *   rest. For example, adding returning device flow to flow->dev_flow list and
2393  *   setting backward reference to the flow should be done out of this function.
2394  *   layers field is not filled either.
2395  *
2396  * @param[in] attr
2397  *   Pointer to the flow attributes.
2398  * @param[in] items
2399  *   Pointer to the list of items.
2400  * @param[in] actions
2401  *   Pointer to the list of actions.
2402  * @param[out] error
2403  *   Pointer to the error structure.
2404  *
2405  * @return
2406  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2407  */
2408 static inline struct mlx5_flow *
2409 flow_drv_prepare(const struct rte_flow *flow,
2410 		 const struct rte_flow_attr *attr,
2411 		 const struct rte_flow_item items[],
2412 		 const struct rte_flow_action actions[],
2413 		 struct rte_flow_error *error)
2414 {
2415 	const struct mlx5_flow_driver_ops *fops;
2416 	enum mlx5_flow_drv_type type = flow->drv_type;
2417 
2418 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2419 	fops = flow_get_drv_ops(type);
2420 	return fops->prepare(attr, items, actions, error);
2421 }
2422 
2423 /**
2424  * Flow driver translation API. This abstracts calling driver specific
2425  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2426  * translates a generic flow into a driver flow. flow_drv_prepare() must
2427  * precede.
2428  *
2429  * @note
2430  *   dev_flow->layers could be filled as a result of parsing during translation
2431  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2432  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2433  *   flow->actions could be overwritten even though all the expanded dev_flows
2434  *   have the same actions.
2435  *
2436  * @param[in] dev
2437  *   Pointer to the rte dev structure.
2438  * @param[in, out] dev_flow
2439  *   Pointer to the mlx5 flow.
2440  * @param[in] attr
2441  *   Pointer to the flow attributes.
2442  * @param[in] items
2443  *   Pointer to the list of items.
2444  * @param[in] actions
2445  *   Pointer to the list of actions.
2446  * @param[out] error
2447  *   Pointer to the error structure.
2448  *
2449  * @return
2450  *   0 on success, a negative errno value otherwise and rte_errno is set.
2451  */
2452 static inline int
2453 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2454 		   const struct rte_flow_attr *attr,
2455 		   const struct rte_flow_item items[],
2456 		   const struct rte_flow_action actions[],
2457 		   struct rte_flow_error *error)
2458 {
2459 	const struct mlx5_flow_driver_ops *fops;
2460 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2461 
2462 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2463 	fops = flow_get_drv_ops(type);
2464 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2465 }
2466 
2467 /**
2468  * Flow driver apply API. This abstracts calling driver specific functions.
2469  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2470  * translated driver flows on to device. flow_drv_translate() must precede.
2471  *
2472  * @param[in] dev
2473  *   Pointer to Ethernet device structure.
2474  * @param[in, out] flow
2475  *   Pointer to flow structure.
2476  * @param[out] error
2477  *   Pointer to error structure.
2478  *
2479  * @return
2480  *   0 on success, a negative errno value otherwise and rte_errno is set.
2481  */
2482 static inline int
2483 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2484 	       struct rte_flow_error *error)
2485 {
2486 	const struct mlx5_flow_driver_ops *fops;
2487 	enum mlx5_flow_drv_type type = flow->drv_type;
2488 
2489 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2490 	fops = flow_get_drv_ops(type);
2491 	return fops->apply(dev, flow, error);
2492 }
2493 
2494 /**
2495  * Flow driver remove API. This abstracts calling driver specific functions.
2496  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2497  * on device. All the resources of the flow should be freed by calling
2498  * flow_drv_destroy().
2499  *
2500  * @param[in] dev
2501  *   Pointer to Ethernet device.
2502  * @param[in, out] flow
2503  *   Pointer to flow structure.
2504  */
2505 static inline void
2506 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2507 {
2508 	const struct mlx5_flow_driver_ops *fops;
2509 	enum mlx5_flow_drv_type type = flow->drv_type;
2510 
2511 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2512 	fops = flow_get_drv_ops(type);
2513 	fops->remove(dev, flow);
2514 }
2515 
2516 /**
2517  * Flow driver destroy API. This abstracts calling driver specific functions.
2518  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2519  * on device and releases resources of the flow.
2520  *
2521  * @param[in] dev
2522  *   Pointer to Ethernet device.
2523  * @param[in, out] flow
2524  *   Pointer to flow structure.
2525  */
2526 static inline void
2527 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2528 {
2529 	const struct mlx5_flow_driver_ops *fops;
2530 	enum mlx5_flow_drv_type type = flow->drv_type;
2531 
2532 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2533 	fops = flow_get_drv_ops(type);
2534 	fops->destroy(dev, flow);
2535 }
2536 
2537 /**
2538  * Validate a flow supported by the NIC.
2539  *
2540  * @see rte_flow_validate()
2541  * @see rte_flow_ops
2542  */
2543 int
2544 mlx5_flow_validate(struct rte_eth_dev *dev,
2545 		   const struct rte_flow_attr *attr,
2546 		   const struct rte_flow_item items[],
2547 		   const struct rte_flow_action actions[],
2548 		   struct rte_flow_error *error)
2549 {
2550 	int ret;
2551 
2552 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2553 	if (ret < 0)
2554 		return ret;
2555 	return 0;
2556 }
2557 
2558 /**
2559  * Get RSS action from the action list.
2560  *
2561  * @param[in] actions
2562  *   Pointer to the list of actions.
2563  *
2564  * @return
2565  *   Pointer to the RSS action if exist, else return NULL.
2566  */
2567 static const struct rte_flow_action_rss*
2568 flow_get_rss_action(const struct rte_flow_action actions[])
2569 {
2570 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2571 		switch (actions->type) {
2572 		case RTE_FLOW_ACTION_TYPE_RSS:
2573 			return (const struct rte_flow_action_rss *)
2574 			       actions->conf;
2575 		default:
2576 			break;
2577 		}
2578 	}
2579 	return NULL;
2580 }
2581 
2582 static unsigned int
2583 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2584 {
2585 	const struct rte_flow_item *item;
2586 	unsigned int has_vlan = 0;
2587 
2588 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2589 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2590 			has_vlan = 1;
2591 			break;
2592 		}
2593 	}
2594 	if (has_vlan)
2595 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2596 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2597 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2598 			       MLX5_EXPANSION_ROOT_OUTER;
2599 }
2600 
2601 /**
2602  * Check if the flow should be splited due to hairpin.
2603  * The reason for the split is that in current HW we can't
2604  * support encap on Rx, so if a flow have encap we move it
2605  * to Tx.
2606  *
2607  * @param dev
2608  *   Pointer to Ethernet device.
2609  * @param[in] attr
2610  *   Flow rule attributes.
2611  * @param[in] actions
2612  *   Associated actions (list terminated by the END action).
2613  *
2614  * @return
2615  *   > 0 the number of actions and the flow should be split,
2616  *   0 when no split required.
2617  */
2618 static int
2619 flow_check_hairpin_split(struct rte_eth_dev *dev,
2620 			 const struct rte_flow_attr *attr,
2621 			 const struct rte_flow_action actions[])
2622 {
2623 	int queue_action = 0;
2624 	int action_n = 0;
2625 	int encap = 0;
2626 	const struct rte_flow_action_queue *queue;
2627 	const struct rte_flow_action_rss *rss;
2628 	const struct rte_flow_action_raw_encap *raw_encap;
2629 
2630 	if (!attr->ingress)
2631 		return 0;
2632 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2633 		switch (actions->type) {
2634 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2635 			queue = actions->conf;
2636 			if (mlx5_rxq_get_type(dev, queue->index) !=
2637 			    MLX5_RXQ_TYPE_HAIRPIN)
2638 				return 0;
2639 			queue_action = 1;
2640 			action_n++;
2641 			break;
2642 		case RTE_FLOW_ACTION_TYPE_RSS:
2643 			rss = actions->conf;
2644 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2645 			    MLX5_RXQ_TYPE_HAIRPIN)
2646 				return 0;
2647 			queue_action = 1;
2648 			action_n++;
2649 			break;
2650 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2651 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2652 			encap = 1;
2653 			action_n++;
2654 			break;
2655 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2656 			raw_encap = actions->conf;
2657 			if (raw_encap->size >
2658 			    (sizeof(struct rte_flow_item_eth) +
2659 			     sizeof(struct rte_flow_item_ipv4)))
2660 				encap = 1;
2661 			action_n++;
2662 			break;
2663 		default:
2664 			action_n++;
2665 			break;
2666 		}
2667 	}
2668 	if (encap == 1 && queue_action)
2669 		return action_n;
2670 	return 0;
2671 }
2672 
2673 #define MLX5_MAX_SPLIT_ACTIONS 24
2674 #define MLX5_MAX_SPLIT_ITEMS 24
2675 
2676 /**
2677  * Split the hairpin flow.
2678  * Since HW can't support encap on Rx we move the encap to Tx.
2679  * If the count action is after the encap then we also
2680  * move the count action. in this case the count will also measure
2681  * the outer bytes.
2682  *
2683  * @param dev
2684  *   Pointer to Ethernet device.
2685  * @param[in] actions
2686  *   Associated actions (list terminated by the END action).
2687  * @param[out] actions_rx
2688  *   Rx flow actions.
2689  * @param[out] actions_tx
2690  *   Tx flow actions..
2691  * @param[out] pattern_tx
2692  *   The pattern items for the Tx flow.
2693  * @param[out] flow_id
2694  *   The flow ID connected to this flow.
2695  *
2696  * @return
2697  *   0 on success.
2698  */
2699 static int
2700 flow_hairpin_split(struct rte_eth_dev *dev,
2701 		   const struct rte_flow_action actions[],
2702 		   struct rte_flow_action actions_rx[],
2703 		   struct rte_flow_action actions_tx[],
2704 		   struct rte_flow_item pattern_tx[],
2705 		   uint32_t *flow_id)
2706 {
2707 	struct mlx5_priv *priv = dev->data->dev_private;
2708 	const struct rte_flow_action_raw_encap *raw_encap;
2709 	const struct rte_flow_action_raw_decap *raw_decap;
2710 	struct mlx5_rte_flow_action_set_tag *set_tag;
2711 	struct rte_flow_action *tag_action;
2712 	struct mlx5_rte_flow_item_tag *tag_item;
2713 	struct rte_flow_item *item;
2714 	char *addr;
2715 	int encap = 0;
2716 
2717 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
2718 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2719 		switch (actions->type) {
2720 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2721 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2722 			rte_memcpy(actions_tx, actions,
2723 			       sizeof(struct rte_flow_action));
2724 			actions_tx++;
2725 			break;
2726 		case RTE_FLOW_ACTION_TYPE_COUNT:
2727 			if (encap) {
2728 				rte_memcpy(actions_tx, actions,
2729 					   sizeof(struct rte_flow_action));
2730 				actions_tx++;
2731 			} else {
2732 				rte_memcpy(actions_rx, actions,
2733 					   sizeof(struct rte_flow_action));
2734 				actions_rx++;
2735 			}
2736 			break;
2737 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2738 			raw_encap = actions->conf;
2739 			if (raw_encap->size >
2740 			    (sizeof(struct rte_flow_item_eth) +
2741 			     sizeof(struct rte_flow_item_ipv4))) {
2742 				memcpy(actions_tx, actions,
2743 				       sizeof(struct rte_flow_action));
2744 				actions_tx++;
2745 				encap = 1;
2746 			} else {
2747 				rte_memcpy(actions_rx, actions,
2748 					   sizeof(struct rte_flow_action));
2749 				actions_rx++;
2750 			}
2751 			break;
2752 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2753 			raw_decap = actions->conf;
2754 			if (raw_decap->size <
2755 			    (sizeof(struct rte_flow_item_eth) +
2756 			     sizeof(struct rte_flow_item_ipv4))) {
2757 				memcpy(actions_tx, actions,
2758 				       sizeof(struct rte_flow_action));
2759 				actions_tx++;
2760 			} else {
2761 				rte_memcpy(actions_rx, actions,
2762 					   sizeof(struct rte_flow_action));
2763 				actions_rx++;
2764 			}
2765 			break;
2766 		default:
2767 			rte_memcpy(actions_rx, actions,
2768 				   sizeof(struct rte_flow_action));
2769 			actions_rx++;
2770 			break;
2771 		}
2772 	}
2773 	/* Add set meta action and end action for the Rx flow. */
2774 	tag_action = actions_rx;
2775 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
2776 	actions_rx++;
2777 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
2778 	actions_rx++;
2779 	set_tag = (void *)actions_rx;
2780 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
2781 	assert(set_tag->id > REG_NONE);
2782 	set_tag->data = *flow_id;
2783 	tag_action->conf = set_tag;
2784 	/* Create Tx item list. */
2785 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
2786 	addr = (void *)&pattern_tx[2];
2787 	item = pattern_tx;
2788 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
2789 	tag_item = (void *)addr;
2790 	tag_item->data = *flow_id;
2791 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
2792 	assert(set_tag->id > REG_NONE);
2793 	item->spec = tag_item;
2794 	addr += sizeof(struct mlx5_rte_flow_item_tag);
2795 	tag_item = (void *)addr;
2796 	tag_item->data = UINT32_MAX;
2797 	tag_item->id = UINT16_MAX;
2798 	item->mask = tag_item;
2799 	addr += sizeof(struct mlx5_rte_flow_item_tag);
2800 	item->last = NULL;
2801 	item++;
2802 	item->type = RTE_FLOW_ITEM_TYPE_END;
2803 	return 0;
2804 }
2805 
2806 /**
2807  * The last stage of splitting chain, just creates the subflow
2808  * without any modification.
2809  *
2810  * @param dev
2811  *   Pointer to Ethernet device.
2812  * @param[in] flow
2813  *   Parent flow structure pointer.
2814  * @param[in, out] sub_flow
2815  *   Pointer to return the created subflow, may be NULL.
2816  * @param[in] attr
2817  *   Flow rule attributes.
2818  * @param[in] items
2819  *   Pattern specification (list terminated by the END pattern item).
2820  * @param[in] actions
2821  *   Associated actions (list terminated by the END action).
2822  * @param[in] external
2823  *   This flow rule is created by request external to PMD.
2824  * @param[out] error
2825  *   Perform verbose error reporting if not NULL.
2826  * @return
2827  *   0 on success, negative value otherwise
2828  */
2829 static int
2830 flow_create_split_inner(struct rte_eth_dev *dev,
2831 			struct rte_flow *flow,
2832 			struct mlx5_flow **sub_flow,
2833 			const struct rte_flow_attr *attr,
2834 			const struct rte_flow_item items[],
2835 			const struct rte_flow_action actions[],
2836 			bool external, struct rte_flow_error *error)
2837 {
2838 	struct mlx5_flow *dev_flow;
2839 
2840 	dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
2841 	if (!dev_flow)
2842 		return -rte_errno;
2843 	dev_flow->flow = flow;
2844 	dev_flow->external = external;
2845 	/* Subflow object was created, we must include one in the list. */
2846 	LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2847 	if (sub_flow)
2848 		*sub_flow = dev_flow;
2849 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
2850 }
2851 
2852 /**
2853  * Split the flow to subflow set. The splitters might be linked
2854  * in the chain, like this:
2855  * flow_create_split_outer() calls:
2856  *   flow_create_split_meter() calls:
2857  *     flow_create_split_metadata(meter_subflow_0) calls:
2858  *       flow_create_split_inner(metadata_subflow_0)
2859  *       flow_create_split_inner(metadata_subflow_1)
2860  *       flow_create_split_inner(metadata_subflow_2)
2861  *     flow_create_split_metadata(meter_subflow_1) calls:
2862  *       flow_create_split_inner(metadata_subflow_0)
2863  *       flow_create_split_inner(metadata_subflow_1)
2864  *       flow_create_split_inner(metadata_subflow_2)
2865  *
2866  * This provide flexible way to add new levels of flow splitting.
2867  * The all of successfully created subflows are included to the
2868  * parent flow dev_flow list.
2869  *
2870  * @param dev
2871  *   Pointer to Ethernet device.
2872  * @param[in] flow
2873  *   Parent flow structure pointer.
2874  * @param[in] attr
2875  *   Flow rule attributes.
2876  * @param[in] items
2877  *   Pattern specification (list terminated by the END pattern item).
2878  * @param[in] actions
2879  *   Associated actions (list terminated by the END action).
2880  * @param[in] external
2881  *   This flow rule is created by request external to PMD.
2882  * @param[out] error
2883  *   Perform verbose error reporting if not NULL.
2884  * @return
2885  *   0 on success, negative value otherwise
2886  */
2887 static int
2888 flow_create_split_outer(struct rte_eth_dev *dev,
2889 			struct rte_flow *flow,
2890 			const struct rte_flow_attr *attr,
2891 			const struct rte_flow_item items[],
2892 			const struct rte_flow_action actions[],
2893 			bool external, struct rte_flow_error *error)
2894 {
2895 	int ret;
2896 
2897 	ret = flow_create_split_inner(dev, flow, NULL, attr, items,
2898 				      actions, external, error);
2899 	assert(ret <= 0);
2900 	return ret;
2901 }
2902 
2903 /**
2904  * Create a flow and add it to @p list.
2905  *
2906  * @param dev
2907  *   Pointer to Ethernet device.
2908  * @param list
2909  *   Pointer to a TAILQ flow list. If this parameter NULL,
2910  *   no list insertion occurred, flow is just created,
2911  *   this is caller's responsibility to track the
2912  *   created flow.
2913  * @param[in] attr
2914  *   Flow rule attributes.
2915  * @param[in] items
2916  *   Pattern specification (list terminated by the END pattern item).
2917  * @param[in] actions
2918  *   Associated actions (list terminated by the END action).
2919  * @param[in] external
2920  *   This flow rule is created by request external to PMD.
2921  * @param[out] error
2922  *   Perform verbose error reporting if not NULL.
2923  *
2924  * @return
2925  *   A flow on success, NULL otherwise and rte_errno is set.
2926  */
2927 static struct rte_flow *
2928 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2929 		 const struct rte_flow_attr *attr,
2930 		 const struct rte_flow_item items[],
2931 		 const struct rte_flow_action actions[],
2932 		 bool external, struct rte_flow_error *error)
2933 {
2934 	struct mlx5_priv *priv = dev->data->dev_private;
2935 	struct rte_flow *flow = NULL;
2936 	struct mlx5_flow *dev_flow;
2937 	const struct rte_flow_action_rss *rss;
2938 	union {
2939 		struct rte_flow_expand_rss buf;
2940 		uint8_t buffer[2048];
2941 	} expand_buffer;
2942 	union {
2943 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
2944 		uint8_t buffer[2048];
2945 	} actions_rx;
2946 	union {
2947 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
2948 		uint8_t buffer[2048];
2949 	} actions_hairpin_tx;
2950 	union {
2951 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
2952 		uint8_t buffer[2048];
2953 	} items_tx;
2954 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2955 	const struct rte_flow_action *p_actions_rx = actions;
2956 	int ret;
2957 	uint32_t i;
2958 	uint32_t flow_size;
2959 	int hairpin_flow = 0;
2960 	uint32_t hairpin_id = 0;
2961 	struct rte_flow_attr attr_tx = { .priority = 0 };
2962 
2963 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
2964 	if (hairpin_flow > 0) {
2965 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
2966 			rte_errno = EINVAL;
2967 			return NULL;
2968 		}
2969 		flow_hairpin_split(dev, actions, actions_rx.actions,
2970 				   actions_hairpin_tx.actions, items_tx.items,
2971 				   &hairpin_id);
2972 		p_actions_rx = actions_rx.actions;
2973 	}
2974 	ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
2975 				error);
2976 	if (ret < 0)
2977 		goto error_before_flow;
2978 	flow_size = sizeof(struct rte_flow);
2979 	rss = flow_get_rss_action(p_actions_rx);
2980 	if (rss)
2981 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2982 					    sizeof(void *));
2983 	else
2984 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2985 	flow = rte_calloc(__func__, 1, flow_size, 0);
2986 	if (!flow) {
2987 		rte_errno = ENOMEM;
2988 		goto error_before_flow;
2989 	}
2990 	flow->drv_type = flow_get_drv_type(dev, attr);
2991 	if (hairpin_id != 0)
2992 		flow->hairpin_flow_id = hairpin_id;
2993 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2994 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
2995 	flow->rss.queue = (void *)(flow + 1);
2996 	if (rss) {
2997 		/*
2998 		 * The following information is required by
2999 		 * mlx5_flow_hashfields_adjust() in advance.
3000 		 */
3001 		flow->rss.level = rss->level;
3002 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
3003 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
3004 	}
3005 	LIST_INIT(&flow->dev_flows);
3006 	if (rss && rss->types) {
3007 		unsigned int graph_root;
3008 
3009 		graph_root = find_graph_root(items, rss->level);
3010 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
3011 					  items, rss->types,
3012 					  mlx5_support_expansion,
3013 					  graph_root);
3014 		assert(ret > 0 &&
3015 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
3016 	} else {
3017 		buf->entries = 1;
3018 		buf->entry[0].pattern = (void *)(uintptr_t)items;
3019 	}
3020 	for (i = 0; i < buf->entries; ++i) {
3021 		/*
3022 		 * The splitter may create multiple dev_flows,
3023 		 * depending on configuration. In the simplest
3024 		 * case it just creates unmodified original flow.
3025 		 */
3026 		ret = flow_create_split_outer(dev, flow, attr,
3027 					      buf->entry[i].pattern,
3028 					      p_actions_rx, external,
3029 					      error);
3030 		if (ret < 0)
3031 			goto error;
3032 	}
3033 	/* Create the tx flow. */
3034 	if (hairpin_flow) {
3035 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
3036 		attr_tx.ingress = 0;
3037 		attr_tx.egress = 1;
3038 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
3039 					    actions_hairpin_tx.actions, error);
3040 		if (!dev_flow)
3041 			goto error;
3042 		dev_flow->flow = flow;
3043 		dev_flow->external = 0;
3044 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3045 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
3046 					 items_tx.items,
3047 					 actions_hairpin_tx.actions, error);
3048 		if (ret < 0)
3049 			goto error;
3050 	}
3051 	if (dev->data->dev_started) {
3052 		ret = flow_drv_apply(dev, flow, error);
3053 		if (ret < 0)
3054 			goto error;
3055 	}
3056 	if (list)
3057 		TAILQ_INSERT_TAIL(list, flow, next);
3058 	flow_rxq_flags_set(dev, flow);
3059 	return flow;
3060 error_before_flow:
3061 	if (hairpin_id)
3062 		mlx5_flow_id_release(priv->sh->flow_id_pool,
3063 				     hairpin_id);
3064 	return NULL;
3065 error:
3066 	ret = rte_errno; /* Save rte_errno before cleanup. */
3067 	if (flow->hairpin_flow_id)
3068 		mlx5_flow_id_release(priv->sh->flow_id_pool,
3069 				     flow->hairpin_flow_id);
3070 	assert(flow);
3071 	flow_drv_destroy(dev, flow);
3072 	rte_free(flow);
3073 	rte_errno = ret; /* Restore rte_errno. */
3074 	return NULL;
3075 }
3076 
3077 /**
3078  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
3079  * incoming packets to table 1.
3080  *
3081  * Other flow rules, requested for group n, will be created in
3082  * e-switch table n+1.
3083  * Jump action to e-switch group n will be created to group n+1.
3084  *
3085  * Used when working in switchdev mode, to utilise advantages of table 1
3086  * and above.
3087  *
3088  * @param dev
3089  *   Pointer to Ethernet device.
3090  *
3091  * @return
3092  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
3093  */
3094 struct rte_flow *
3095 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
3096 {
3097 	const struct rte_flow_attr attr = {
3098 		.group = 0,
3099 		.priority = 0,
3100 		.ingress = 1,
3101 		.egress = 0,
3102 		.transfer = 1,
3103 	};
3104 	const struct rte_flow_item pattern = {
3105 		.type = RTE_FLOW_ITEM_TYPE_END,
3106 	};
3107 	struct rte_flow_action_jump jump = {
3108 		.group = 1,
3109 	};
3110 	const struct rte_flow_action actions[] = {
3111 		{
3112 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3113 			.conf = &jump,
3114 		},
3115 		{
3116 			.type = RTE_FLOW_ACTION_TYPE_END,
3117 		},
3118 	};
3119 	struct mlx5_priv *priv = dev->data->dev_private;
3120 	struct rte_flow_error error;
3121 
3122 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
3123 				actions, false, &error);
3124 }
3125 
3126 /**
3127  * Create a flow.
3128  *
3129  * @see rte_flow_create()
3130  * @see rte_flow_ops
3131  */
3132 struct rte_flow *
3133 mlx5_flow_create(struct rte_eth_dev *dev,
3134 		 const struct rte_flow_attr *attr,
3135 		 const struct rte_flow_item items[],
3136 		 const struct rte_flow_action actions[],
3137 		 struct rte_flow_error *error)
3138 {
3139 	struct mlx5_priv *priv = dev->data->dev_private;
3140 
3141 	return flow_list_create(dev, &priv->flows,
3142 				attr, items, actions, true, error);
3143 }
3144 
3145 /**
3146  * Destroy a flow in a list.
3147  *
3148  * @param dev
3149  *   Pointer to Ethernet device.
3150  * @param list
3151  *   Pointer to a TAILQ flow list. If this parameter NULL,
3152  *   there is no flow removal from the list.
3153  * @param[in] flow
3154  *   Flow to destroy.
3155  */
3156 static void
3157 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
3158 		  struct rte_flow *flow)
3159 {
3160 	struct mlx5_priv *priv = dev->data->dev_private;
3161 
3162 	/*
3163 	 * Update RX queue flags only if port is started, otherwise it is
3164 	 * already clean.
3165 	 */
3166 	if (dev->data->dev_started)
3167 		flow_rxq_flags_trim(dev, flow);
3168 	if (flow->hairpin_flow_id)
3169 		mlx5_flow_id_release(priv->sh->flow_id_pool,
3170 				     flow->hairpin_flow_id);
3171 	flow_drv_destroy(dev, flow);
3172 	if (list)
3173 		TAILQ_REMOVE(list, flow, next);
3174 	rte_free(flow->fdir);
3175 	rte_free(flow);
3176 }
3177 
3178 /**
3179  * Destroy all flows.
3180  *
3181  * @param dev
3182  *   Pointer to Ethernet device.
3183  * @param list
3184  *   Pointer to a TAILQ flow list.
3185  */
3186 void
3187 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
3188 {
3189 	while (!TAILQ_EMPTY(list)) {
3190 		struct rte_flow *flow;
3191 
3192 		flow = TAILQ_FIRST(list);
3193 		flow_list_destroy(dev, list, flow);
3194 	}
3195 }
3196 
3197 /**
3198  * Remove all flows.
3199  *
3200  * @param dev
3201  *   Pointer to Ethernet device.
3202  * @param list
3203  *   Pointer to a TAILQ flow list.
3204  */
3205 void
3206 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
3207 {
3208 	struct rte_flow *flow;
3209 
3210 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
3211 		flow_drv_remove(dev, flow);
3212 	flow_rxq_flags_clear(dev);
3213 }
3214 
3215 /**
3216  * Add all flows.
3217  *
3218  * @param dev
3219  *   Pointer to Ethernet device.
3220  * @param list
3221  *   Pointer to a TAILQ flow list.
3222  *
3223  * @return
3224  *   0 on success, a negative errno value otherwise and rte_errno is set.
3225  */
3226 int
3227 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
3228 {
3229 	struct rte_flow *flow;
3230 	struct rte_flow_error error;
3231 	int ret = 0;
3232 
3233 	TAILQ_FOREACH(flow, list, next) {
3234 		ret = flow_drv_apply(dev, flow, &error);
3235 		if (ret < 0)
3236 			goto error;
3237 		flow_rxq_flags_set(dev, flow);
3238 	}
3239 	return 0;
3240 error:
3241 	ret = rte_errno; /* Save rte_errno before cleanup. */
3242 	mlx5_flow_stop(dev, list);
3243 	rte_errno = ret; /* Restore rte_errno. */
3244 	return -rte_errno;
3245 }
3246 
3247 /**
3248  * Verify the flow list is empty
3249  *
3250  * @param dev
3251  *  Pointer to Ethernet device.
3252  *
3253  * @return the number of flows not released.
3254  */
3255 int
3256 mlx5_flow_verify(struct rte_eth_dev *dev)
3257 {
3258 	struct mlx5_priv *priv = dev->data->dev_private;
3259 	struct rte_flow *flow;
3260 	int ret = 0;
3261 
3262 	TAILQ_FOREACH(flow, &priv->flows, next) {
3263 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
3264 			dev->data->port_id, (void *)flow);
3265 		++ret;
3266 	}
3267 	return ret;
3268 }
3269 
3270 /**
3271  * Enable default hairpin egress flow.
3272  *
3273  * @param dev
3274  *   Pointer to Ethernet device.
3275  * @param queue
3276  *   The queue index.
3277  *
3278  * @return
3279  *   0 on success, a negative errno value otherwise and rte_errno is set.
3280  */
3281 int
3282 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
3283 			    uint32_t queue)
3284 {
3285 	struct mlx5_priv *priv = dev->data->dev_private;
3286 	const struct rte_flow_attr attr = {
3287 		.egress = 1,
3288 		.priority = 0,
3289 	};
3290 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
3291 		.queue = queue,
3292 	};
3293 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
3294 		.queue = UINT32_MAX,
3295 	};
3296 	struct rte_flow_item items[] = {
3297 		{
3298 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
3299 			.spec = &queue_spec,
3300 			.last = NULL,
3301 			.mask = &queue_mask,
3302 		},
3303 		{
3304 			.type = RTE_FLOW_ITEM_TYPE_END,
3305 		},
3306 	};
3307 	struct rte_flow_action_jump jump = {
3308 		.group = MLX5_HAIRPIN_TX_TABLE,
3309 	};
3310 	struct rte_flow_action actions[2];
3311 	struct rte_flow *flow;
3312 	struct rte_flow_error error;
3313 
3314 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
3315 	actions[0].conf = &jump;
3316 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
3317 	flow = flow_list_create(dev, &priv->ctrl_flows,
3318 				&attr, items, actions, false, &error);
3319 	if (!flow) {
3320 		DRV_LOG(DEBUG,
3321 			"Failed to create ctrl flow: rte_errno(%d),"
3322 			" type(%d), message(%s)",
3323 			rte_errno, error.type,
3324 			error.message ? error.message : " (no stated reason)");
3325 		return -rte_errno;
3326 	}
3327 	return 0;
3328 }
3329 
3330 /**
3331  * Enable a control flow configured from the control plane.
3332  *
3333  * @param dev
3334  *   Pointer to Ethernet device.
3335  * @param eth_spec
3336  *   An Ethernet flow spec to apply.
3337  * @param eth_mask
3338  *   An Ethernet flow mask to apply.
3339  * @param vlan_spec
3340  *   A VLAN flow spec to apply.
3341  * @param vlan_mask
3342  *   A VLAN flow mask to apply.
3343  *
3344  * @return
3345  *   0 on success, a negative errno value otherwise and rte_errno is set.
3346  */
3347 int
3348 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
3349 		    struct rte_flow_item_eth *eth_spec,
3350 		    struct rte_flow_item_eth *eth_mask,
3351 		    struct rte_flow_item_vlan *vlan_spec,
3352 		    struct rte_flow_item_vlan *vlan_mask)
3353 {
3354 	struct mlx5_priv *priv = dev->data->dev_private;
3355 	const struct rte_flow_attr attr = {
3356 		.ingress = 1,
3357 		.priority = MLX5_FLOW_PRIO_RSVD,
3358 	};
3359 	struct rte_flow_item items[] = {
3360 		{
3361 			.type = RTE_FLOW_ITEM_TYPE_ETH,
3362 			.spec = eth_spec,
3363 			.last = NULL,
3364 			.mask = eth_mask,
3365 		},
3366 		{
3367 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
3368 					      RTE_FLOW_ITEM_TYPE_END,
3369 			.spec = vlan_spec,
3370 			.last = NULL,
3371 			.mask = vlan_mask,
3372 		},
3373 		{
3374 			.type = RTE_FLOW_ITEM_TYPE_END,
3375 		},
3376 	};
3377 	uint16_t queue[priv->reta_idx_n];
3378 	struct rte_flow_action_rss action_rss = {
3379 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
3380 		.level = 0,
3381 		.types = priv->rss_conf.rss_hf,
3382 		.key_len = priv->rss_conf.rss_key_len,
3383 		.queue_num = priv->reta_idx_n,
3384 		.key = priv->rss_conf.rss_key,
3385 		.queue = queue,
3386 	};
3387 	struct rte_flow_action actions[] = {
3388 		{
3389 			.type = RTE_FLOW_ACTION_TYPE_RSS,
3390 			.conf = &action_rss,
3391 		},
3392 		{
3393 			.type = RTE_FLOW_ACTION_TYPE_END,
3394 		},
3395 	};
3396 	struct rte_flow *flow;
3397 	struct rte_flow_error error;
3398 	unsigned int i;
3399 
3400 	if (!priv->reta_idx_n || !priv->rxqs_n) {
3401 		return 0;
3402 	}
3403 	for (i = 0; i != priv->reta_idx_n; ++i)
3404 		queue[i] = (*priv->reta_idx)[i];
3405 	flow = flow_list_create(dev, &priv->ctrl_flows,
3406 				&attr, items, actions, false, &error);
3407 	if (!flow)
3408 		return -rte_errno;
3409 	return 0;
3410 }
3411 
3412 /**
3413  * Enable a flow control configured from the control plane.
3414  *
3415  * @param dev
3416  *   Pointer to Ethernet device.
3417  * @param eth_spec
3418  *   An Ethernet flow spec to apply.
3419  * @param eth_mask
3420  *   An Ethernet flow mask to apply.
3421  *
3422  * @return
3423  *   0 on success, a negative errno value otherwise and rte_errno is set.
3424  */
3425 int
3426 mlx5_ctrl_flow(struct rte_eth_dev *dev,
3427 	       struct rte_flow_item_eth *eth_spec,
3428 	       struct rte_flow_item_eth *eth_mask)
3429 {
3430 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
3431 }
3432 
3433 /**
3434  * Destroy a flow.
3435  *
3436  * @see rte_flow_destroy()
3437  * @see rte_flow_ops
3438  */
3439 int
3440 mlx5_flow_destroy(struct rte_eth_dev *dev,
3441 		  struct rte_flow *flow,
3442 		  struct rte_flow_error *error __rte_unused)
3443 {
3444 	struct mlx5_priv *priv = dev->data->dev_private;
3445 
3446 	flow_list_destroy(dev, &priv->flows, flow);
3447 	return 0;
3448 }
3449 
3450 /**
3451  * Destroy all flows.
3452  *
3453  * @see rte_flow_flush()
3454  * @see rte_flow_ops
3455  */
3456 int
3457 mlx5_flow_flush(struct rte_eth_dev *dev,
3458 		struct rte_flow_error *error __rte_unused)
3459 {
3460 	struct mlx5_priv *priv = dev->data->dev_private;
3461 
3462 	mlx5_flow_list_flush(dev, &priv->flows);
3463 	return 0;
3464 }
3465 
3466 /**
3467  * Isolated mode.
3468  *
3469  * @see rte_flow_isolate()
3470  * @see rte_flow_ops
3471  */
3472 int
3473 mlx5_flow_isolate(struct rte_eth_dev *dev,
3474 		  int enable,
3475 		  struct rte_flow_error *error)
3476 {
3477 	struct mlx5_priv *priv = dev->data->dev_private;
3478 
3479 	if (dev->data->dev_started) {
3480 		rte_flow_error_set(error, EBUSY,
3481 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3482 				   NULL,
3483 				   "port must be stopped first");
3484 		return -rte_errno;
3485 	}
3486 	priv->isolated = !!enable;
3487 	if (enable)
3488 		dev->dev_ops = &mlx5_dev_ops_isolate;
3489 	else
3490 		dev->dev_ops = &mlx5_dev_ops;
3491 	return 0;
3492 }
3493 
3494 /**
3495  * Query a flow.
3496  *
3497  * @see rte_flow_query()
3498  * @see rte_flow_ops
3499  */
3500 static int
3501 flow_drv_query(struct rte_eth_dev *dev,
3502 	       struct rte_flow *flow,
3503 	       const struct rte_flow_action *actions,
3504 	       void *data,
3505 	       struct rte_flow_error *error)
3506 {
3507 	const struct mlx5_flow_driver_ops *fops;
3508 	enum mlx5_flow_drv_type ftype = flow->drv_type;
3509 
3510 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
3511 	fops = flow_get_drv_ops(ftype);
3512 
3513 	return fops->query(dev, flow, actions, data, error);
3514 }
3515 
3516 /**
3517  * Query a flow.
3518  *
3519  * @see rte_flow_query()
3520  * @see rte_flow_ops
3521  */
3522 int
3523 mlx5_flow_query(struct rte_eth_dev *dev,
3524 		struct rte_flow *flow,
3525 		const struct rte_flow_action *actions,
3526 		void *data,
3527 		struct rte_flow_error *error)
3528 {
3529 	int ret;
3530 
3531 	ret = flow_drv_query(dev, flow, actions, data, error);
3532 	if (ret < 0)
3533 		return ret;
3534 	return 0;
3535 }
3536 
3537 /**
3538  * Convert a flow director filter to a generic flow.
3539  *
3540  * @param dev
3541  *   Pointer to Ethernet device.
3542  * @param fdir_filter
3543  *   Flow director filter to add.
3544  * @param attributes
3545  *   Generic flow parameters structure.
3546  *
3547  * @return
3548  *   0 on success, a negative errno value otherwise and rte_errno is set.
3549  */
3550 static int
3551 flow_fdir_filter_convert(struct rte_eth_dev *dev,
3552 			 const struct rte_eth_fdir_filter *fdir_filter,
3553 			 struct mlx5_fdir *attributes)
3554 {
3555 	struct mlx5_priv *priv = dev->data->dev_private;
3556 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
3557 	const struct rte_eth_fdir_masks *mask =
3558 		&dev->data->dev_conf.fdir_conf.mask;
3559 
3560 	/* Validate queue number. */
3561 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
3562 		DRV_LOG(ERR, "port %u invalid queue number %d",
3563 			dev->data->port_id, fdir_filter->action.rx_queue);
3564 		rte_errno = EINVAL;
3565 		return -rte_errno;
3566 	}
3567 	attributes->attr.ingress = 1;
3568 	attributes->items[0] = (struct rte_flow_item) {
3569 		.type = RTE_FLOW_ITEM_TYPE_ETH,
3570 		.spec = &attributes->l2,
3571 		.mask = &attributes->l2_mask,
3572 	};
3573 	switch (fdir_filter->action.behavior) {
3574 	case RTE_ETH_FDIR_ACCEPT:
3575 		attributes->actions[0] = (struct rte_flow_action){
3576 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
3577 			.conf = &attributes->queue,
3578 		};
3579 		break;
3580 	case RTE_ETH_FDIR_REJECT:
3581 		attributes->actions[0] = (struct rte_flow_action){
3582 			.type = RTE_FLOW_ACTION_TYPE_DROP,
3583 		};
3584 		break;
3585 	default:
3586 		DRV_LOG(ERR, "port %u invalid behavior %d",
3587 			dev->data->port_id,
3588 			fdir_filter->action.behavior);
3589 		rte_errno = ENOTSUP;
3590 		return -rte_errno;
3591 	}
3592 	attributes->queue.index = fdir_filter->action.rx_queue;
3593 	/* Handle L3. */
3594 	switch (fdir_filter->input.flow_type) {
3595 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
3596 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
3597 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
3598 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
3599 			.src_addr = input->flow.ip4_flow.src_ip,
3600 			.dst_addr = input->flow.ip4_flow.dst_ip,
3601 			.time_to_live = input->flow.ip4_flow.ttl,
3602 			.type_of_service = input->flow.ip4_flow.tos,
3603 		};
3604 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
3605 			.src_addr = mask->ipv4_mask.src_ip,
3606 			.dst_addr = mask->ipv4_mask.dst_ip,
3607 			.time_to_live = mask->ipv4_mask.ttl,
3608 			.type_of_service = mask->ipv4_mask.tos,
3609 			.next_proto_id = mask->ipv4_mask.proto,
3610 		};
3611 		attributes->items[1] = (struct rte_flow_item){
3612 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
3613 			.spec = &attributes->l3,
3614 			.mask = &attributes->l3_mask,
3615 		};
3616 		break;
3617 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
3618 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
3619 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
3620 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
3621 			.hop_limits = input->flow.ipv6_flow.hop_limits,
3622 			.proto = input->flow.ipv6_flow.proto,
3623 		};
3624 
3625 		memcpy(attributes->l3.ipv6.hdr.src_addr,
3626 		       input->flow.ipv6_flow.src_ip,
3627 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
3628 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
3629 		       input->flow.ipv6_flow.dst_ip,
3630 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
3631 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
3632 		       mask->ipv6_mask.src_ip,
3633 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
3634 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
3635 		       mask->ipv6_mask.dst_ip,
3636 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
3637 		attributes->items[1] = (struct rte_flow_item){
3638 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
3639 			.spec = &attributes->l3,
3640 			.mask = &attributes->l3_mask,
3641 		};
3642 		break;
3643 	default:
3644 		DRV_LOG(ERR, "port %u invalid flow type%d",
3645 			dev->data->port_id, fdir_filter->input.flow_type);
3646 		rte_errno = ENOTSUP;
3647 		return -rte_errno;
3648 	}
3649 	/* Handle L4. */
3650 	switch (fdir_filter->input.flow_type) {
3651 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
3652 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
3653 			.src_port = input->flow.udp4_flow.src_port,
3654 			.dst_port = input->flow.udp4_flow.dst_port,
3655 		};
3656 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
3657 			.src_port = mask->src_port_mask,
3658 			.dst_port = mask->dst_port_mask,
3659 		};
3660 		attributes->items[2] = (struct rte_flow_item){
3661 			.type = RTE_FLOW_ITEM_TYPE_UDP,
3662 			.spec = &attributes->l4,
3663 			.mask = &attributes->l4_mask,
3664 		};
3665 		break;
3666 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
3667 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
3668 			.src_port = input->flow.tcp4_flow.src_port,
3669 			.dst_port = input->flow.tcp4_flow.dst_port,
3670 		};
3671 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
3672 			.src_port = mask->src_port_mask,
3673 			.dst_port = mask->dst_port_mask,
3674 		};
3675 		attributes->items[2] = (struct rte_flow_item){
3676 			.type = RTE_FLOW_ITEM_TYPE_TCP,
3677 			.spec = &attributes->l4,
3678 			.mask = &attributes->l4_mask,
3679 		};
3680 		break;
3681 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
3682 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
3683 			.src_port = input->flow.udp6_flow.src_port,
3684 			.dst_port = input->flow.udp6_flow.dst_port,
3685 		};
3686 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
3687 			.src_port = mask->src_port_mask,
3688 			.dst_port = mask->dst_port_mask,
3689 		};
3690 		attributes->items[2] = (struct rte_flow_item){
3691 			.type = RTE_FLOW_ITEM_TYPE_UDP,
3692 			.spec = &attributes->l4,
3693 			.mask = &attributes->l4_mask,
3694 		};
3695 		break;
3696 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
3697 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
3698 			.src_port = input->flow.tcp6_flow.src_port,
3699 			.dst_port = input->flow.tcp6_flow.dst_port,
3700 		};
3701 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
3702 			.src_port = mask->src_port_mask,
3703 			.dst_port = mask->dst_port_mask,
3704 		};
3705 		attributes->items[2] = (struct rte_flow_item){
3706 			.type = RTE_FLOW_ITEM_TYPE_TCP,
3707 			.spec = &attributes->l4,
3708 			.mask = &attributes->l4_mask,
3709 		};
3710 		break;
3711 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
3712 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
3713 		break;
3714 	default:
3715 		DRV_LOG(ERR, "port %u invalid flow type%d",
3716 			dev->data->port_id, fdir_filter->input.flow_type);
3717 		rte_errno = ENOTSUP;
3718 		return -rte_errno;
3719 	}
3720 	return 0;
3721 }
3722 
3723 #define FLOW_FDIR_CMP(f1, f2, fld) \
3724 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
3725 
3726 /**
3727  * Compare two FDIR flows. If items and actions are identical, the two flows are
3728  * regarded as same.
3729  *
3730  * @param dev
3731  *   Pointer to Ethernet device.
3732  * @param f1
3733  *   FDIR flow to compare.
3734  * @param f2
3735  *   FDIR flow to compare.
3736  *
3737  * @return
3738  *   Zero on match, 1 otherwise.
3739  */
3740 static int
3741 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
3742 {
3743 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
3744 	    FLOW_FDIR_CMP(f1, f2, l2) ||
3745 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
3746 	    FLOW_FDIR_CMP(f1, f2, l3) ||
3747 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
3748 	    FLOW_FDIR_CMP(f1, f2, l4) ||
3749 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
3750 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
3751 		return 1;
3752 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
3753 	    FLOW_FDIR_CMP(f1, f2, queue))
3754 		return 1;
3755 	return 0;
3756 }
3757 
3758 /**
3759  * Search device flow list to find out a matched FDIR flow.
3760  *
3761  * @param dev
3762  *   Pointer to Ethernet device.
3763  * @param fdir_flow
3764  *   FDIR flow to lookup.
3765  *
3766  * @return
3767  *   Pointer of flow if found, NULL otherwise.
3768  */
3769 static struct rte_flow *
3770 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
3771 {
3772 	struct mlx5_priv *priv = dev->data->dev_private;
3773 	struct rte_flow *flow = NULL;
3774 
3775 	assert(fdir_flow);
3776 	TAILQ_FOREACH(flow, &priv->flows, next) {
3777 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
3778 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
3779 				dev->data->port_id, (void *)flow);
3780 			break;
3781 		}
3782 	}
3783 	return flow;
3784 }
3785 
3786 /**
3787  * Add new flow director filter and store it in list.
3788  *
3789  * @param dev
3790  *   Pointer to Ethernet device.
3791  * @param fdir_filter
3792  *   Flow director filter to add.
3793  *
3794  * @return
3795  *   0 on success, a negative errno value otherwise and rte_errno is set.
3796  */
3797 static int
3798 flow_fdir_filter_add(struct rte_eth_dev *dev,
3799 		     const struct rte_eth_fdir_filter *fdir_filter)
3800 {
3801 	struct mlx5_priv *priv = dev->data->dev_private;
3802 	struct mlx5_fdir *fdir_flow;
3803 	struct rte_flow *flow;
3804 	int ret;
3805 
3806 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
3807 	if (!fdir_flow) {
3808 		rte_errno = ENOMEM;
3809 		return -rte_errno;
3810 	}
3811 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
3812 	if (ret)
3813 		goto error;
3814 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
3815 	if (flow) {
3816 		rte_errno = EEXIST;
3817 		goto error;
3818 	}
3819 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
3820 				fdir_flow->items, fdir_flow->actions, true,
3821 				NULL);
3822 	if (!flow)
3823 		goto error;
3824 	assert(!flow->fdir);
3825 	flow->fdir = fdir_flow;
3826 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
3827 		dev->data->port_id, (void *)flow);
3828 	return 0;
3829 error:
3830 	rte_free(fdir_flow);
3831 	return -rte_errno;
3832 }
3833 
3834 /**
3835  * Delete specific filter.
3836  *
3837  * @param dev
3838  *   Pointer to Ethernet device.
3839  * @param fdir_filter
3840  *   Filter to be deleted.
3841  *
3842  * @return
3843  *   0 on success, a negative errno value otherwise and rte_errno is set.
3844  */
3845 static int
3846 flow_fdir_filter_delete(struct rte_eth_dev *dev,
3847 			const struct rte_eth_fdir_filter *fdir_filter)
3848 {
3849 	struct mlx5_priv *priv = dev->data->dev_private;
3850 	struct rte_flow *flow;
3851 	struct mlx5_fdir fdir_flow = {
3852 		.attr.group = 0,
3853 	};
3854 	int ret;
3855 
3856 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
3857 	if (ret)
3858 		return -rte_errno;
3859 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
3860 	if (!flow) {
3861 		rte_errno = ENOENT;
3862 		return -rte_errno;
3863 	}
3864 	flow_list_destroy(dev, &priv->flows, flow);
3865 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
3866 		dev->data->port_id, (void *)flow);
3867 	return 0;
3868 }
3869 
3870 /**
3871  * Update queue for specific filter.
3872  *
3873  * @param dev
3874  *   Pointer to Ethernet device.
3875  * @param fdir_filter
3876  *   Filter to be updated.
3877  *
3878  * @return
3879  *   0 on success, a negative errno value otherwise and rte_errno is set.
3880  */
3881 static int
3882 flow_fdir_filter_update(struct rte_eth_dev *dev,
3883 			const struct rte_eth_fdir_filter *fdir_filter)
3884 {
3885 	int ret;
3886 
3887 	ret = flow_fdir_filter_delete(dev, fdir_filter);
3888 	if (ret)
3889 		return ret;
3890 	return flow_fdir_filter_add(dev, fdir_filter);
3891 }
3892 
3893 /**
3894  * Flush all filters.
3895  *
3896  * @param dev
3897  *   Pointer to Ethernet device.
3898  */
3899 static void
3900 flow_fdir_filter_flush(struct rte_eth_dev *dev)
3901 {
3902 	struct mlx5_priv *priv = dev->data->dev_private;
3903 
3904 	mlx5_flow_list_flush(dev, &priv->flows);
3905 }
3906 
3907 /**
3908  * Get flow director information.
3909  *
3910  * @param dev
3911  *   Pointer to Ethernet device.
3912  * @param[out] fdir_info
3913  *   Resulting flow director information.
3914  */
3915 static void
3916 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
3917 {
3918 	struct rte_eth_fdir_masks *mask =
3919 		&dev->data->dev_conf.fdir_conf.mask;
3920 
3921 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
3922 	fdir_info->guarant_spc = 0;
3923 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
3924 	fdir_info->max_flexpayload = 0;
3925 	fdir_info->flow_types_mask[0] = 0;
3926 	fdir_info->flex_payload_unit = 0;
3927 	fdir_info->max_flex_payload_segment_num = 0;
3928 	fdir_info->flex_payload_limit = 0;
3929 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
3930 }
3931 
3932 /**
3933  * Deal with flow director operations.
3934  *
3935  * @param dev
3936  *   Pointer to Ethernet device.
3937  * @param filter_op
3938  *   Operation to perform.
3939  * @param arg
3940  *   Pointer to operation-specific structure.
3941  *
3942  * @return
3943  *   0 on success, a negative errno value otherwise and rte_errno is set.
3944  */
3945 static int
3946 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
3947 		    void *arg)
3948 {
3949 	enum rte_fdir_mode fdir_mode =
3950 		dev->data->dev_conf.fdir_conf.mode;
3951 
3952 	if (filter_op == RTE_ETH_FILTER_NOP)
3953 		return 0;
3954 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
3955 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
3956 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
3957 			dev->data->port_id, fdir_mode);
3958 		rte_errno = EINVAL;
3959 		return -rte_errno;
3960 	}
3961 	switch (filter_op) {
3962 	case RTE_ETH_FILTER_ADD:
3963 		return flow_fdir_filter_add(dev, arg);
3964 	case RTE_ETH_FILTER_UPDATE:
3965 		return flow_fdir_filter_update(dev, arg);
3966 	case RTE_ETH_FILTER_DELETE:
3967 		return flow_fdir_filter_delete(dev, arg);
3968 	case RTE_ETH_FILTER_FLUSH:
3969 		flow_fdir_filter_flush(dev);
3970 		break;
3971 	case RTE_ETH_FILTER_INFO:
3972 		flow_fdir_info_get(dev, arg);
3973 		break;
3974 	default:
3975 		DRV_LOG(DEBUG, "port %u unknown operation %u",
3976 			dev->data->port_id, filter_op);
3977 		rte_errno = EINVAL;
3978 		return -rte_errno;
3979 	}
3980 	return 0;
3981 }
3982 
3983 /**
3984  * Manage filter operations.
3985  *
3986  * @param dev
3987  *   Pointer to Ethernet device structure.
3988  * @param filter_type
3989  *   Filter type.
3990  * @param filter_op
3991  *   Operation to perform.
3992  * @param arg
3993  *   Pointer to operation-specific structure.
3994  *
3995  * @return
3996  *   0 on success, a negative errno value otherwise and rte_errno is set.
3997  */
3998 int
3999 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
4000 		     enum rte_filter_type filter_type,
4001 		     enum rte_filter_op filter_op,
4002 		     void *arg)
4003 {
4004 	switch (filter_type) {
4005 	case RTE_ETH_FILTER_GENERIC:
4006 		if (filter_op != RTE_ETH_FILTER_GET) {
4007 			rte_errno = EINVAL;
4008 			return -rte_errno;
4009 		}
4010 		*(const void **)arg = &mlx5_flow_ops;
4011 		return 0;
4012 	case RTE_ETH_FILTER_FDIR:
4013 		return flow_fdir_ctrl_func(dev, filter_op, arg);
4014 	default:
4015 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
4016 			dev->data->port_id, filter_type);
4017 		rte_errno = ENOTSUP;
4018 		return -rte_errno;
4019 	}
4020 	return 0;
4021 }
4022 
4023 #define MLX5_POOL_QUERY_FREQ_US 1000000
4024 
4025 /**
4026  * Set the periodic procedure for triggering asynchronous batch queries for all
4027  * the counter pools.
4028  *
4029  * @param[in] sh
4030  *   Pointer to mlx5_ibv_shared object.
4031  */
4032 void
4033 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
4034 {
4035 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
4036 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
4037 	uint32_t us;
4038 
4039 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
4040 	pools_n += rte_atomic16_read(&cont->n_valid);
4041 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
4042 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
4043 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
4044 		sh->cmng.query_thread_on = 0;
4045 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
4046 	} else {
4047 		sh->cmng.query_thread_on = 1;
4048 	}
4049 }
4050 
4051 /**
4052  * The periodic procedure for triggering asynchronous batch queries for all the
4053  * counter pools. This function is probably called by the host thread.
4054  *
4055  * @param[in] arg
4056  *   The parameter for the alarm process.
4057  */
4058 void
4059 mlx5_flow_query_alarm(void *arg)
4060 {
4061 	struct mlx5_ibv_shared *sh = arg;
4062 	struct mlx5_devx_obj *dcs;
4063 	uint16_t offset;
4064 	int ret;
4065 	uint8_t batch = sh->cmng.batch;
4066 	uint16_t pool_index = sh->cmng.pool_index;
4067 	struct mlx5_pools_container *cont;
4068 	struct mlx5_pools_container *mcont;
4069 	struct mlx5_flow_counter_pool *pool;
4070 
4071 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
4072 		goto set_alarm;
4073 next_container:
4074 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
4075 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
4076 	/* Check if resize was done and need to flip a container. */
4077 	if (cont != mcont) {
4078 		if (cont->pools) {
4079 			/* Clean the old container. */
4080 			rte_free(cont->pools);
4081 			memset(cont, 0, sizeof(*cont));
4082 		}
4083 		rte_cio_wmb();
4084 		 /* Flip the host container. */
4085 		sh->cmng.mhi[batch] ^= (uint8_t)2;
4086 		cont = mcont;
4087 	}
4088 	if (!cont->pools) {
4089 		/* 2 empty containers case is unexpected. */
4090 		if (unlikely(batch != sh->cmng.batch))
4091 			goto set_alarm;
4092 		batch ^= 0x1;
4093 		pool_index = 0;
4094 		goto next_container;
4095 	}
4096 	pool = cont->pools[pool_index];
4097 	if (pool->raw_hw)
4098 		/* There is a pool query in progress. */
4099 		goto set_alarm;
4100 	pool->raw_hw =
4101 		LIST_FIRST(&sh->cmng.free_stat_raws);
4102 	if (!pool->raw_hw)
4103 		/* No free counter statistics raw memory. */
4104 		goto set_alarm;
4105 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
4106 							      (&pool->a64_dcs);
4107 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
4108 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
4109 					       offset, NULL, NULL,
4110 					       pool->raw_hw->mem_mng->dm->id,
4111 					       (void *)(uintptr_t)
4112 					       (pool->raw_hw->data + offset),
4113 					       sh->devx_comp,
4114 					       (uint64_t)(uintptr_t)pool);
4115 	if (ret) {
4116 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
4117 			" %d", pool->min_dcs->id);
4118 		pool->raw_hw = NULL;
4119 		goto set_alarm;
4120 	}
4121 	pool->raw_hw->min_dcs_id = dcs->id;
4122 	LIST_REMOVE(pool->raw_hw, next);
4123 	sh->cmng.pending_queries++;
4124 	pool_index++;
4125 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
4126 		batch ^= 0x1;
4127 		pool_index = 0;
4128 	}
4129 set_alarm:
4130 	sh->cmng.batch = batch;
4131 	sh->cmng.pool_index = pool_index;
4132 	mlx5_set_query_alarm(sh);
4133 }
4134 
4135 /**
4136  * Handler for the HW respond about ready values from an asynchronous batch
4137  * query. This function is probably called by the host thread.
4138  *
4139  * @param[in] sh
4140  *   The pointer to the shared IB device context.
4141  * @param[in] async_id
4142  *   The Devx async ID.
4143  * @param[in] status
4144  *   The status of the completion.
4145  */
4146 void
4147 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
4148 				  uint64_t async_id, int status)
4149 {
4150 	struct mlx5_flow_counter_pool *pool =
4151 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
4152 	struct mlx5_counter_stats_raw *raw_to_free;
4153 
4154 	if (unlikely(status)) {
4155 		raw_to_free = pool->raw_hw;
4156 	} else {
4157 		raw_to_free = pool->raw;
4158 		rte_spinlock_lock(&pool->sl);
4159 		pool->raw = pool->raw_hw;
4160 		rte_spinlock_unlock(&pool->sl);
4161 		rte_atomic64_add(&pool->query_gen, 1);
4162 		/* Be sure the new raw counters data is updated in memory. */
4163 		rte_cio_wmb();
4164 	}
4165 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
4166 	pool->raw_hw = NULL;
4167 	sh->cmng.pending_queries--;
4168 }
4169 
4170 /**
4171  * Translate the rte_flow group index to HW table value.
4172  *
4173  * @param[in] attributes
4174  *   Pointer to flow attributes
4175  * @param[in] external
4176  *   Value is part of flow rule created by request external to PMD.
4177  * @param[in] group
4178  *   rte_flow group index value.
4179  * @param[out] table
4180  *   HW table value.
4181  * @param[out] error
4182  *   Pointer to error structure.
4183  *
4184  * @return
4185  *   0 on success, a negative errno value otherwise and rte_errno is set.
4186  */
4187 int
4188 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
4189 			 uint32_t group, uint32_t *table,
4190 			 struct rte_flow_error *error)
4191 {
4192 	if (attributes->transfer && external) {
4193 		if (group == UINT32_MAX)
4194 			return rte_flow_error_set
4195 						(error, EINVAL,
4196 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
4197 						 NULL,
4198 						 "group index not supported");
4199 		*table = group + 1;
4200 	} else {
4201 		*table = group;
4202 	}
4203 	return 0;
4204 }
4205 
4206 /**
4207  * Discover availability of metadata reg_c's.
4208  *
4209  * Iteratively use test flows to check availability.
4210  *
4211  * @param[in] dev
4212  *   Pointer to the Ethernet device structure.
4213  *
4214  * @return
4215  *   0 on success, a negative errno value otherwise and rte_errno is set.
4216  */
4217 int
4218 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
4219 {
4220 	struct mlx5_priv *priv = dev->data->dev_private;
4221 	struct mlx5_dev_config *config = &priv->config;
4222 	enum modify_reg idx;
4223 	int n = 0;
4224 
4225 	/* reg_c[0] and reg_c[1] are reserved. */
4226 	config->flow_mreg_c[n++] = REG_C_0;
4227 	config->flow_mreg_c[n++] = REG_C_1;
4228 	/* Discover availability of other reg_c's. */
4229 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
4230 		struct rte_flow_attr attr = {
4231 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4232 			.priority = MLX5_FLOW_PRIO_RSVD,
4233 			.ingress = 1,
4234 		};
4235 		struct rte_flow_item items[] = {
4236 			[0] = {
4237 				.type = RTE_FLOW_ITEM_TYPE_END,
4238 			},
4239 		};
4240 		struct rte_flow_action actions[] = {
4241 			[0] = {
4242 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4243 				.conf = &(struct mlx5_flow_action_copy_mreg){
4244 					.src = REG_C_1,
4245 					.dst = idx,
4246 				},
4247 			},
4248 			[1] = {
4249 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
4250 				.conf = &(struct rte_flow_action_jump){
4251 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4252 				},
4253 			},
4254 			[2] = {
4255 				.type = RTE_FLOW_ACTION_TYPE_END,
4256 			},
4257 		};
4258 		struct rte_flow *flow;
4259 		struct rte_flow_error error;
4260 
4261 		if (!config->dv_flow_en)
4262 			break;
4263 		/* Create internal flow, validation skips copy action. */
4264 		flow = flow_list_create(dev, NULL, &attr, items,
4265 					actions, false, &error);
4266 		if (!flow)
4267 			continue;
4268 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
4269 			config->flow_mreg_c[n++] = idx;
4270 		flow_list_destroy(dev, NULL, flow);
4271 	}
4272 	for (; n < MLX5_MREG_C_NUM; ++n)
4273 		config->flow_mreg_c[n] = REG_NONE;
4274 	return 0;
4275 }
4276