xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision d56ec3dcad056c47cef4e837d5191d04c936d87e)
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 	enum modify_reg start_reg;
348 
349 	switch (feature) {
350 	case MLX5_HAIRPIN_RX:
351 		return REG_B;
352 	case MLX5_HAIRPIN_TX:
353 		return REG_A;
354 	case MLX5_METADATA_RX:
355 		switch (config->dv_xmeta_en) {
356 		case MLX5_XMETA_MODE_LEGACY:
357 			return REG_B;
358 		case MLX5_XMETA_MODE_META16:
359 			return REG_C_0;
360 		case MLX5_XMETA_MODE_META32:
361 			return REG_C_1;
362 		}
363 		break;
364 	case MLX5_METADATA_TX:
365 		return REG_A;
366 	case MLX5_METADATA_FDB:
367 		return REG_C_0;
368 	case MLX5_FLOW_MARK:
369 		switch (config->dv_xmeta_en) {
370 		case MLX5_XMETA_MODE_LEGACY:
371 			return REG_NONE;
372 		case MLX5_XMETA_MODE_META16:
373 			return REG_C_1;
374 		case MLX5_XMETA_MODE_META32:
375 			return REG_C_0;
376 		}
377 		break;
378 	case MLX5_COPY_MARK:
379 	case MLX5_MTR_SFX:
380 		/*
381 		 * Metadata COPY_MARK register using is in meter suffix sub
382 		 * flow while with meter. It's safe to share the same register.
383 		 */
384 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
385 	case MLX5_MTR_COLOR:
386 		RTE_ASSERT(priv->mtr_color_reg != REG_NONE);
387 		return priv->mtr_color_reg;
388 	case MLX5_APP_TAG:
389 		/*
390 		 * If meter is enable, it will engage two registers for color
391 		 * match and flow match. If meter color match is not using the
392 		 * REG_C_2, need to skip the REG_C_x be used by meter color
393 		 * match.
394 		 * If meter is disable, free to use all available registers.
395 		 */
396 		if (priv->mtr_color_reg != REG_NONE)
397 			start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_3 :
398 				    REG_C_4;
399 		else
400 			start_reg = REG_C_2;
401 		if (id > (REG_C_7 - start_reg))
402 			return rte_flow_error_set(error, EINVAL,
403 						  RTE_FLOW_ERROR_TYPE_ITEM,
404 						  NULL, "invalid tag id");
405 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
406 			return rte_flow_error_set(error, ENOTSUP,
407 						  RTE_FLOW_ERROR_TYPE_ITEM,
408 						  NULL, "unsupported tag id");
409 		/*
410 		 * This case means meter is using the REG_C_x great than 2.
411 		 * Take care not to conflict with meter color REG_C_x.
412 		 * If the available index REG_C_y >= REG_C_x, skip the
413 		 * color register.
414 		 */
415 		if (start_reg == REG_C_3 && config->flow_mreg_c
416 		    [id + REG_C_3 - REG_C_0] >= priv->mtr_color_reg) {
417 			if (config->flow_mreg_c[id + 1 + REG_C_3 - REG_C_0] !=
418 			    REG_NONE)
419 				return config->flow_mreg_c
420 						[id + 1 + REG_C_3 - REG_C_0];
421 			return rte_flow_error_set(error, ENOTSUP,
422 						  RTE_FLOW_ERROR_TYPE_ITEM,
423 						  NULL, "unsupported tag id");
424 		}
425 		return config->flow_mreg_c[id + start_reg - REG_C_0];
426 	}
427 	assert(false);
428 	return rte_flow_error_set(error, EINVAL,
429 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
430 				  NULL, "invalid feature name");
431 }
432 
433 /**
434  * Check extensive flow metadata register support.
435  *
436  * @param dev
437  *   Pointer to rte_eth_dev structure.
438  *
439  * @return
440  *   True if device supports extensive flow metadata register, otherwise false.
441  */
442 bool
443 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
444 {
445 	struct mlx5_priv *priv = dev->data->dev_private;
446 	struct mlx5_dev_config *config = &priv->config;
447 
448 	/*
449 	 * Having available reg_c can be regarded inclusively as supporting
450 	 * extensive flow metadata register, which could mean,
451 	 * - metadata register copy action by modify header.
452 	 * - 16 modify header actions is supported.
453 	 * - reg_c's are preserved across different domain (FDB and NIC) on
454 	 *   packet loopback by flow lookup miss.
455 	 */
456 	return config->flow_mreg_c[2] != REG_NONE;
457 }
458 
459 /**
460  * Discover the maximum number of priority available.
461  *
462  * @param[in] dev
463  *   Pointer to the Ethernet device structure.
464  *
465  * @return
466  *   number of supported flow priority on success, a negative errno
467  *   value otherwise and rte_errno is set.
468  */
469 int
470 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
471 {
472 	struct mlx5_priv *priv = dev->data->dev_private;
473 	struct {
474 		struct ibv_flow_attr attr;
475 		struct ibv_flow_spec_eth eth;
476 		struct ibv_flow_spec_action_drop drop;
477 	} flow_attr = {
478 		.attr = {
479 			.num_of_specs = 2,
480 			.port = (uint8_t)priv->ibv_port,
481 		},
482 		.eth = {
483 			.type = IBV_FLOW_SPEC_ETH,
484 			.size = sizeof(struct ibv_flow_spec_eth),
485 		},
486 		.drop = {
487 			.size = sizeof(struct ibv_flow_spec_action_drop),
488 			.type = IBV_FLOW_SPEC_ACTION_DROP,
489 		},
490 	};
491 	struct ibv_flow *flow;
492 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
493 	uint16_t vprio[] = { 8, 16 };
494 	int i;
495 	int priority = 0;
496 
497 	if (!drop) {
498 		rte_errno = ENOTSUP;
499 		return -rte_errno;
500 	}
501 	for (i = 0; i != RTE_DIM(vprio); i++) {
502 		flow_attr.attr.priority = vprio[i] - 1;
503 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
504 		if (!flow)
505 			break;
506 		claim_zero(mlx5_glue->destroy_flow(flow));
507 		priority = vprio[i];
508 	}
509 	mlx5_hrxq_drop_release(dev);
510 	switch (priority) {
511 	case 8:
512 		priority = RTE_DIM(priority_map_3);
513 		break;
514 	case 16:
515 		priority = RTE_DIM(priority_map_5);
516 		break;
517 	default:
518 		rte_errno = ENOTSUP;
519 		DRV_LOG(ERR,
520 			"port %u verbs maximum priority: %d expected 8/16",
521 			dev->data->port_id, priority);
522 		return -rte_errno;
523 	}
524 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
525 		dev->data->port_id, priority);
526 	return priority;
527 }
528 
529 /**
530  * Adjust flow priority based on the highest layer and the request priority.
531  *
532  * @param[in] dev
533  *   Pointer to the Ethernet device structure.
534  * @param[in] priority
535  *   The rule base priority.
536  * @param[in] subpriority
537  *   The priority based on the items.
538  *
539  * @return
540  *   The new priority.
541  */
542 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
543 				   uint32_t subpriority)
544 {
545 	uint32_t res = 0;
546 	struct mlx5_priv *priv = dev->data->dev_private;
547 
548 	switch (priv->config.flow_prio) {
549 	case RTE_DIM(priority_map_3):
550 		res = priority_map_3[priority][subpriority];
551 		break;
552 	case RTE_DIM(priority_map_5):
553 		res = priority_map_5[priority][subpriority];
554 		break;
555 	}
556 	return  res;
557 }
558 
559 /**
560  * Verify the @p item specifications (spec, last, mask) are compatible with the
561  * NIC capabilities.
562  *
563  * @param[in] item
564  *   Item specification.
565  * @param[in] mask
566  *   @p item->mask or flow default bit-masks.
567  * @param[in] nic_mask
568  *   Bit-masks covering supported fields by the NIC to compare with user mask.
569  * @param[in] size
570  *   Bit-masks size in bytes.
571  * @param[out] error
572  *   Pointer to error structure.
573  *
574  * @return
575  *   0 on success, a negative errno value otherwise and rte_errno is set.
576  */
577 int
578 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
579 			  const uint8_t *mask,
580 			  const uint8_t *nic_mask,
581 			  unsigned int size,
582 			  struct rte_flow_error *error)
583 {
584 	unsigned int i;
585 
586 	assert(nic_mask);
587 	for (i = 0; i < size; ++i)
588 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
589 			return rte_flow_error_set(error, ENOTSUP,
590 						  RTE_FLOW_ERROR_TYPE_ITEM,
591 						  item,
592 						  "mask enables non supported"
593 						  " bits");
594 	if (!item->spec && (item->mask || item->last))
595 		return rte_flow_error_set(error, EINVAL,
596 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
597 					  "mask/last without a spec is not"
598 					  " supported");
599 	if (item->spec && item->last) {
600 		uint8_t spec[size];
601 		uint8_t last[size];
602 		unsigned int i;
603 		int ret;
604 
605 		for (i = 0; i < size; ++i) {
606 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
607 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
608 		}
609 		ret = memcmp(spec, last, size);
610 		if (ret != 0)
611 			return rte_flow_error_set(error, EINVAL,
612 						  RTE_FLOW_ERROR_TYPE_ITEM,
613 						  item,
614 						  "range is not valid");
615 	}
616 	return 0;
617 }
618 
619 /**
620  * Adjust the hash fields according to the @p flow information.
621  *
622  * @param[in] dev_flow.
623  *   Pointer to the mlx5_flow.
624  * @param[in] tunnel
625  *   1 when the hash field is for a tunnel item.
626  * @param[in] layer_types
627  *   ETH_RSS_* types.
628  * @param[in] hash_fields
629  *   Item hash fields.
630  *
631  * @return
632  *   The hash fields that should be used.
633  */
634 uint64_t
635 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
636 			    int tunnel __rte_unused, uint64_t layer_types,
637 			    uint64_t hash_fields)
638 {
639 	struct rte_flow *flow = dev_flow->flow;
640 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
641 	int rss_request_inner = flow->rss.level >= 2;
642 
643 	/* Check RSS hash level for tunnel. */
644 	if (tunnel && rss_request_inner)
645 		hash_fields |= IBV_RX_HASH_INNER;
646 	else if (tunnel || rss_request_inner)
647 		return 0;
648 #endif
649 	/* Check if requested layer matches RSS hash fields. */
650 	if (!(flow->rss.types & layer_types))
651 		return 0;
652 	return hash_fields;
653 }
654 
655 /**
656  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
657  * if several tunnel rules are used on this queue, the tunnel ptype will be
658  * cleared.
659  *
660  * @param rxq_ctrl
661  *   Rx queue to update.
662  */
663 static void
664 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
665 {
666 	unsigned int i;
667 	uint32_t tunnel_ptype = 0;
668 
669 	/* Look up for the ptype to use. */
670 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
671 		if (!rxq_ctrl->flow_tunnels_n[i])
672 			continue;
673 		if (!tunnel_ptype) {
674 			tunnel_ptype = tunnels_info[i].ptype;
675 		} else {
676 			tunnel_ptype = 0;
677 			break;
678 		}
679 	}
680 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
681 }
682 
683 /**
684  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
685  * flow.
686  *
687  * @param[in] dev
688  *   Pointer to the Ethernet device structure.
689  * @param[in] dev_flow
690  *   Pointer to device flow structure.
691  */
692 static void
693 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
694 {
695 	struct mlx5_priv *priv = dev->data->dev_private;
696 	struct rte_flow *flow = dev_flow->flow;
697 	const int mark = !!(dev_flow->actions &
698 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
699 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
700 	unsigned int i;
701 
702 	for (i = 0; i != flow->rss.queue_num; ++i) {
703 		int idx = (*flow->rss.queue)[i];
704 		struct mlx5_rxq_ctrl *rxq_ctrl =
705 			container_of((*priv->rxqs)[idx],
706 				     struct mlx5_rxq_ctrl, rxq);
707 
708 		/*
709 		 * To support metadata register copy on Tx loopback,
710 		 * this must be always enabled (metadata may arive
711 		 * from other port - not from local flows only.
712 		 */
713 		if (priv->config.dv_flow_en &&
714 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
715 		    mlx5_flow_ext_mreg_supported(dev)) {
716 			rxq_ctrl->rxq.mark = 1;
717 			rxq_ctrl->flow_mark_n = 1;
718 		} else if (mark) {
719 			rxq_ctrl->rxq.mark = 1;
720 			rxq_ctrl->flow_mark_n++;
721 		}
722 		if (tunnel) {
723 			unsigned int j;
724 
725 			/* Increase the counter matching the flow. */
726 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
727 				if ((tunnels_info[j].tunnel &
728 				     dev_flow->layers) ==
729 				    tunnels_info[j].tunnel) {
730 					rxq_ctrl->flow_tunnels_n[j]++;
731 					break;
732 				}
733 			}
734 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
735 		}
736 	}
737 }
738 
739 /**
740  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
741  *
742  * @param[in] dev
743  *   Pointer to the Ethernet device structure.
744  * @param[in] flow
745  *   Pointer to flow structure.
746  */
747 static void
748 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
749 {
750 	struct mlx5_flow *dev_flow;
751 
752 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
753 		flow_drv_rxq_flags_set(dev, dev_flow);
754 }
755 
756 /**
757  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
758  * device flow if no other flow uses it with the same kind of request.
759  *
760  * @param dev
761  *   Pointer to Ethernet device.
762  * @param[in] dev_flow
763  *   Pointer to the device flow.
764  */
765 static void
766 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
767 {
768 	struct mlx5_priv *priv = dev->data->dev_private;
769 	struct rte_flow *flow = dev_flow->flow;
770 	const int mark = !!(dev_flow->actions &
771 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
772 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
773 	unsigned int i;
774 
775 	assert(dev->data->dev_started);
776 	for (i = 0; i != flow->rss.queue_num; ++i) {
777 		int idx = (*flow->rss.queue)[i];
778 		struct mlx5_rxq_ctrl *rxq_ctrl =
779 			container_of((*priv->rxqs)[idx],
780 				     struct mlx5_rxq_ctrl, rxq);
781 
782 		if (priv->config.dv_flow_en &&
783 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
784 		    mlx5_flow_ext_mreg_supported(dev)) {
785 			rxq_ctrl->rxq.mark = 1;
786 			rxq_ctrl->flow_mark_n = 1;
787 		} else if (mark) {
788 			rxq_ctrl->flow_mark_n--;
789 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
790 		}
791 		if (tunnel) {
792 			unsigned int j;
793 
794 			/* Decrease the counter matching the flow. */
795 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
796 				if ((tunnels_info[j].tunnel &
797 				     dev_flow->layers) ==
798 				    tunnels_info[j].tunnel) {
799 					rxq_ctrl->flow_tunnels_n[j]--;
800 					break;
801 				}
802 			}
803 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
804 		}
805 	}
806 }
807 
808 /**
809  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
810  * @p flow if no other flow uses it with the same kind of request.
811  *
812  * @param dev
813  *   Pointer to Ethernet device.
814  * @param[in] flow
815  *   Pointer to the flow.
816  */
817 static void
818 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
819 {
820 	struct mlx5_flow *dev_flow;
821 
822 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
823 		flow_drv_rxq_flags_trim(dev, dev_flow);
824 }
825 
826 /**
827  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
828  *
829  * @param dev
830  *   Pointer to Ethernet device.
831  */
832 static void
833 flow_rxq_flags_clear(struct rte_eth_dev *dev)
834 {
835 	struct mlx5_priv *priv = dev->data->dev_private;
836 	unsigned int i;
837 
838 	for (i = 0; i != priv->rxqs_n; ++i) {
839 		struct mlx5_rxq_ctrl *rxq_ctrl;
840 		unsigned int j;
841 
842 		if (!(*priv->rxqs)[i])
843 			continue;
844 		rxq_ctrl = container_of((*priv->rxqs)[i],
845 					struct mlx5_rxq_ctrl, rxq);
846 		rxq_ctrl->flow_mark_n = 0;
847 		rxq_ctrl->rxq.mark = 0;
848 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
849 			rxq_ctrl->flow_tunnels_n[j] = 0;
850 		rxq_ctrl->rxq.tunnel = 0;
851 	}
852 }
853 
854 /*
855  * return a pointer to the desired action in the list of actions.
856  *
857  * @param[in] actions
858  *   The list of actions to search the action in.
859  * @param[in] action
860  *   The action to find.
861  *
862  * @return
863  *   Pointer to the action in the list, if found. NULL otherwise.
864  */
865 const struct rte_flow_action *
866 mlx5_flow_find_action(const struct rte_flow_action *actions,
867 		      enum rte_flow_action_type action)
868 {
869 	if (actions == NULL)
870 		return NULL;
871 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
872 		if (actions->type == action)
873 			return actions;
874 	return NULL;
875 }
876 
877 /*
878  * Validate the flag action.
879  *
880  * @param[in] action_flags
881  *   Bit-fields that holds the actions detected until now.
882  * @param[in] attr
883  *   Attributes of flow that includes this action.
884  * @param[out] error
885  *   Pointer to error structure.
886  *
887  * @return
888  *   0 on success, a negative errno value otherwise and rte_errno is set.
889  */
890 int
891 mlx5_flow_validate_action_flag(uint64_t action_flags,
892 			       const struct rte_flow_attr *attr,
893 			       struct rte_flow_error *error)
894 {
895 
896 	if (action_flags & MLX5_FLOW_ACTION_DROP)
897 		return rte_flow_error_set(error, EINVAL,
898 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
899 					  "can't drop and flag in same flow");
900 	if (action_flags & MLX5_FLOW_ACTION_MARK)
901 		return rte_flow_error_set(error, EINVAL,
902 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
903 					  "can't mark and flag in same flow");
904 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
905 		return rte_flow_error_set(error, EINVAL,
906 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
907 					  "can't have 2 flag"
908 					  " actions in same flow");
909 	if (attr->egress)
910 		return rte_flow_error_set(error, ENOTSUP,
911 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
912 					  "flag action not supported for "
913 					  "egress");
914 	return 0;
915 }
916 
917 /*
918  * Validate the mark action.
919  *
920  * @param[in] action
921  *   Pointer to the queue action.
922  * @param[in] action_flags
923  *   Bit-fields that holds the actions detected until now.
924  * @param[in] attr
925  *   Attributes of flow that includes this action.
926  * @param[out] error
927  *   Pointer to error structure.
928  *
929  * @return
930  *   0 on success, a negative errno value otherwise and rte_errno is set.
931  */
932 int
933 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
934 			       uint64_t action_flags,
935 			       const struct rte_flow_attr *attr,
936 			       struct rte_flow_error *error)
937 {
938 	const struct rte_flow_action_mark *mark = action->conf;
939 
940 	if (!mark)
941 		return rte_flow_error_set(error, EINVAL,
942 					  RTE_FLOW_ERROR_TYPE_ACTION,
943 					  action,
944 					  "configuration cannot be null");
945 	if (mark->id >= MLX5_FLOW_MARK_MAX)
946 		return rte_flow_error_set(error, EINVAL,
947 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
948 					  &mark->id,
949 					  "mark id must in 0 <= id < "
950 					  RTE_STR(MLX5_FLOW_MARK_MAX));
951 	if (action_flags & MLX5_FLOW_ACTION_DROP)
952 		return rte_flow_error_set(error, EINVAL,
953 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
954 					  "can't drop and mark in same flow");
955 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
956 		return rte_flow_error_set(error, EINVAL,
957 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
958 					  "can't flag and mark in same flow");
959 	if (action_flags & MLX5_FLOW_ACTION_MARK)
960 		return rte_flow_error_set(error, EINVAL,
961 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
962 					  "can't have 2 mark actions in same"
963 					  " flow");
964 	if (attr->egress)
965 		return rte_flow_error_set(error, ENOTSUP,
966 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
967 					  "mark action not supported for "
968 					  "egress");
969 	return 0;
970 }
971 
972 /*
973  * Validate the drop action.
974  *
975  * @param[in] action_flags
976  *   Bit-fields that holds the actions detected until now.
977  * @param[in] attr
978  *   Attributes of flow that includes this action.
979  * @param[out] error
980  *   Pointer to error structure.
981  *
982  * @return
983  *   0 on success, a negative errno value otherwise and rte_errno is set.
984  */
985 int
986 mlx5_flow_validate_action_drop(uint64_t action_flags,
987 			       const struct rte_flow_attr *attr,
988 			       struct rte_flow_error *error)
989 {
990 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
991 		return rte_flow_error_set(error, EINVAL,
992 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
993 					  "can't drop and flag in same flow");
994 	if (action_flags & MLX5_FLOW_ACTION_MARK)
995 		return rte_flow_error_set(error, EINVAL,
996 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
997 					  "can't drop and mark in same flow");
998 	if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
999 			    MLX5_FLOW_FATE_ESWITCH_ACTIONS))
1000 		return rte_flow_error_set(error, EINVAL,
1001 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1002 					  "can't have 2 fate actions in"
1003 					  " same flow");
1004 	if (attr->egress)
1005 		return rte_flow_error_set(error, ENOTSUP,
1006 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1007 					  "drop action not supported for "
1008 					  "egress");
1009 	return 0;
1010 }
1011 
1012 /*
1013  * Validate the queue action.
1014  *
1015  * @param[in] action
1016  *   Pointer to the queue action.
1017  * @param[in] action_flags
1018  *   Bit-fields that holds the actions detected until now.
1019  * @param[in] dev
1020  *   Pointer to the Ethernet device structure.
1021  * @param[in] attr
1022  *   Attributes of flow that includes this action.
1023  * @param[out] error
1024  *   Pointer to error structure.
1025  *
1026  * @return
1027  *   0 on success, a negative errno value otherwise and rte_errno is set.
1028  */
1029 int
1030 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1031 				uint64_t action_flags,
1032 				struct rte_eth_dev *dev,
1033 				const struct rte_flow_attr *attr,
1034 				struct rte_flow_error *error)
1035 {
1036 	struct mlx5_priv *priv = dev->data->dev_private;
1037 	const struct rte_flow_action_queue *queue = action->conf;
1038 
1039 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1040 		return rte_flow_error_set(error, EINVAL,
1041 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1042 					  "can't have 2 fate actions in"
1043 					  " same flow");
1044 	if (!priv->rxqs_n)
1045 		return rte_flow_error_set(error, EINVAL,
1046 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1047 					  NULL, "No Rx queues configured");
1048 	if (queue->index >= priv->rxqs_n)
1049 		return rte_flow_error_set(error, EINVAL,
1050 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1051 					  &queue->index,
1052 					  "queue index out of range");
1053 	if (!(*priv->rxqs)[queue->index])
1054 		return rte_flow_error_set(error, EINVAL,
1055 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1056 					  &queue->index,
1057 					  "queue is not configured");
1058 	if (attr->egress)
1059 		return rte_flow_error_set(error, ENOTSUP,
1060 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1061 					  "queue action not supported for "
1062 					  "egress");
1063 	return 0;
1064 }
1065 
1066 /*
1067  * Validate the rss action.
1068  *
1069  * @param[in] action
1070  *   Pointer to the queue action.
1071  * @param[in] action_flags
1072  *   Bit-fields that holds the actions detected until now.
1073  * @param[in] dev
1074  *   Pointer to the Ethernet device structure.
1075  * @param[in] attr
1076  *   Attributes of flow that includes this action.
1077  * @param[in] item_flags
1078  *   Items that were detected.
1079  * @param[out] error
1080  *   Pointer to error structure.
1081  *
1082  * @return
1083  *   0 on success, a negative errno value otherwise and rte_errno is set.
1084  */
1085 int
1086 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1087 			      uint64_t action_flags,
1088 			      struct rte_eth_dev *dev,
1089 			      const struct rte_flow_attr *attr,
1090 			      uint64_t item_flags,
1091 			      struct rte_flow_error *error)
1092 {
1093 	struct mlx5_priv *priv = dev->data->dev_private;
1094 	const struct rte_flow_action_rss *rss = action->conf;
1095 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1096 	unsigned int i;
1097 
1098 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1099 		return rte_flow_error_set(error, EINVAL,
1100 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1101 					  "can't have 2 fate actions"
1102 					  " in same flow");
1103 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1104 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1105 		return rte_flow_error_set(error, ENOTSUP,
1106 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1107 					  &rss->func,
1108 					  "RSS hash function not supported");
1109 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1110 	if (rss->level > 2)
1111 #else
1112 	if (rss->level > 1)
1113 #endif
1114 		return rte_flow_error_set(error, ENOTSUP,
1115 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1116 					  &rss->level,
1117 					  "tunnel RSS is not supported");
1118 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1119 	if (rss->key_len == 0 && rss->key != NULL)
1120 		return rte_flow_error_set(error, ENOTSUP,
1121 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1122 					  &rss->key_len,
1123 					  "RSS hash key length 0");
1124 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1125 		return rte_flow_error_set(error, ENOTSUP,
1126 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1127 					  &rss->key_len,
1128 					  "RSS hash key too small");
1129 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1130 		return rte_flow_error_set(error, ENOTSUP,
1131 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1132 					  &rss->key_len,
1133 					  "RSS hash key too large");
1134 	if (rss->queue_num > priv->config.ind_table_max_size)
1135 		return rte_flow_error_set(error, ENOTSUP,
1136 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1137 					  &rss->queue_num,
1138 					  "number of queues too large");
1139 	if (rss->types & MLX5_RSS_HF_MASK)
1140 		return rte_flow_error_set(error, ENOTSUP,
1141 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1142 					  &rss->types,
1143 					  "some RSS protocols are not"
1144 					  " supported");
1145 	if (!priv->rxqs_n)
1146 		return rte_flow_error_set(error, EINVAL,
1147 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1148 					  NULL, "No Rx queues configured");
1149 	if (!rss->queue_num)
1150 		return rte_flow_error_set(error, EINVAL,
1151 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1152 					  NULL, "No queues configured");
1153 	for (i = 0; i != rss->queue_num; ++i) {
1154 		if (rss->queue[i] >= priv->rxqs_n)
1155 			return rte_flow_error_set
1156 				(error, EINVAL,
1157 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1158 				 &rss->queue[i], "queue index out of range");
1159 		if (!(*priv->rxqs)[rss->queue[i]])
1160 			return rte_flow_error_set
1161 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1162 				 &rss->queue[i], "queue is not configured");
1163 	}
1164 	if (attr->egress)
1165 		return rte_flow_error_set(error, ENOTSUP,
1166 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1167 					  "rss action not supported for "
1168 					  "egress");
1169 	if (rss->level > 1 &&  !tunnel)
1170 		return rte_flow_error_set(error, EINVAL,
1171 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1172 					  "inner RSS is not supported for "
1173 					  "non-tunnel flows");
1174 	return 0;
1175 }
1176 
1177 /*
1178  * Validate the count action.
1179  *
1180  * @param[in] dev
1181  *   Pointer to the Ethernet device structure.
1182  * @param[in] attr
1183  *   Attributes of flow that includes this action.
1184  * @param[out] error
1185  *   Pointer to error structure.
1186  *
1187  * @return
1188  *   0 on success, a negative errno value otherwise and rte_errno is set.
1189  */
1190 int
1191 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1192 				const struct rte_flow_attr *attr,
1193 				struct rte_flow_error *error)
1194 {
1195 	if (attr->egress)
1196 		return rte_flow_error_set(error, ENOTSUP,
1197 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1198 					  "count action not supported for "
1199 					  "egress");
1200 	return 0;
1201 }
1202 
1203 /**
1204  * Verify the @p attributes will be correctly understood by the NIC and store
1205  * them in the @p flow if everything is correct.
1206  *
1207  * @param[in] dev
1208  *   Pointer to the Ethernet device structure.
1209  * @param[in] attributes
1210  *   Pointer to flow attributes
1211  * @param[out] error
1212  *   Pointer to error structure.
1213  *
1214  * @return
1215  *   0 on success, a negative errno value otherwise and rte_errno is set.
1216  */
1217 int
1218 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1219 			      const struct rte_flow_attr *attributes,
1220 			      struct rte_flow_error *error)
1221 {
1222 	struct mlx5_priv *priv = dev->data->dev_private;
1223 	uint32_t priority_max = priv->config.flow_prio - 1;
1224 
1225 	if (attributes->group)
1226 		return rte_flow_error_set(error, ENOTSUP,
1227 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1228 					  NULL, "groups is not supported");
1229 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1230 	    attributes->priority >= priority_max)
1231 		return rte_flow_error_set(error, ENOTSUP,
1232 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1233 					  NULL, "priority out of range");
1234 	if (attributes->egress)
1235 		return rte_flow_error_set(error, ENOTSUP,
1236 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1237 					  "egress is not supported");
1238 	if (attributes->transfer && !priv->config.dv_esw_en)
1239 		return rte_flow_error_set(error, ENOTSUP,
1240 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1241 					  NULL, "transfer is not supported");
1242 	if (!attributes->ingress)
1243 		return rte_flow_error_set(error, EINVAL,
1244 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1245 					  NULL,
1246 					  "ingress attribute is mandatory");
1247 	return 0;
1248 }
1249 
1250 /**
1251  * Validate ICMP6 item.
1252  *
1253  * @param[in] item
1254  *   Item specification.
1255  * @param[in] item_flags
1256  *   Bit-fields that holds the items detected until now.
1257  * @param[out] error
1258  *   Pointer to error structure.
1259  *
1260  * @return
1261  *   0 on success, a negative errno value otherwise and rte_errno is set.
1262  */
1263 int
1264 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1265 			       uint64_t item_flags,
1266 			       uint8_t target_protocol,
1267 			       struct rte_flow_error *error)
1268 {
1269 	const struct rte_flow_item_icmp6 *mask = item->mask;
1270 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1271 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1272 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1273 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1274 				      MLX5_FLOW_LAYER_OUTER_L4;
1275 	int ret;
1276 
1277 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1278 		return rte_flow_error_set(error, EINVAL,
1279 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1280 					  "protocol filtering not compatible"
1281 					  " with ICMP6 layer");
1282 	if (!(item_flags & l3m))
1283 		return rte_flow_error_set(error, EINVAL,
1284 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1285 					  "IPv6 is mandatory to filter on"
1286 					  " ICMP6");
1287 	if (item_flags & l4m)
1288 		return rte_flow_error_set(error, EINVAL,
1289 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1290 					  "multiple L4 layers not supported");
1291 	if (!mask)
1292 		mask = &rte_flow_item_icmp6_mask;
1293 	ret = mlx5_flow_item_acceptable
1294 		(item, (const uint8_t *)mask,
1295 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1296 		 sizeof(struct rte_flow_item_icmp6), error);
1297 	if (ret < 0)
1298 		return ret;
1299 	return 0;
1300 }
1301 
1302 /**
1303  * Validate ICMP item.
1304  *
1305  * @param[in] item
1306  *   Item specification.
1307  * @param[in] item_flags
1308  *   Bit-fields that holds the items detected until now.
1309  * @param[out] error
1310  *   Pointer to error structure.
1311  *
1312  * @return
1313  *   0 on success, a negative errno value otherwise and rte_errno is set.
1314  */
1315 int
1316 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1317 			     uint64_t item_flags,
1318 			     uint8_t target_protocol,
1319 			     struct rte_flow_error *error)
1320 {
1321 	const struct rte_flow_item_icmp *mask = item->mask;
1322 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1323 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1324 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1325 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1326 				      MLX5_FLOW_LAYER_OUTER_L4;
1327 	int ret;
1328 
1329 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1330 		return rte_flow_error_set(error, EINVAL,
1331 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1332 					  "protocol filtering not compatible"
1333 					  " with ICMP layer");
1334 	if (!(item_flags & l3m))
1335 		return rte_flow_error_set(error, EINVAL,
1336 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1337 					  "IPv4 is mandatory to filter"
1338 					  " on ICMP");
1339 	if (item_flags & l4m)
1340 		return rte_flow_error_set(error, EINVAL,
1341 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1342 					  "multiple L4 layers not supported");
1343 	if (!mask)
1344 		mask = &rte_flow_item_icmp_mask;
1345 	ret = mlx5_flow_item_acceptable
1346 		(item, (const uint8_t *)mask,
1347 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1348 		 sizeof(struct rte_flow_item_icmp), error);
1349 	if (ret < 0)
1350 		return ret;
1351 	return 0;
1352 }
1353 
1354 /**
1355  * Validate Ethernet item.
1356  *
1357  * @param[in] item
1358  *   Item specification.
1359  * @param[in] item_flags
1360  *   Bit-fields that holds the items detected until now.
1361  * @param[out] error
1362  *   Pointer to error structure.
1363  *
1364  * @return
1365  *   0 on success, a negative errno value otherwise and rte_errno is set.
1366  */
1367 int
1368 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1369 			    uint64_t item_flags,
1370 			    struct rte_flow_error *error)
1371 {
1372 	const struct rte_flow_item_eth *mask = item->mask;
1373 	const struct rte_flow_item_eth nic_mask = {
1374 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1375 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1376 		.type = RTE_BE16(0xffff),
1377 	};
1378 	int ret;
1379 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1380 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1381 				       MLX5_FLOW_LAYER_OUTER_L2;
1382 
1383 	if (item_flags & ethm)
1384 		return rte_flow_error_set(error, ENOTSUP,
1385 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1386 					  "multiple L2 layers not supported");
1387 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1388 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1389 		return rte_flow_error_set(error, EINVAL,
1390 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1391 					  "L2 layer should not follow "
1392 					  "L3 layers");
1393 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1394 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1395 		return rte_flow_error_set(error, EINVAL,
1396 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1397 					  "L2 layer should not follow VLAN");
1398 	if (!mask)
1399 		mask = &rte_flow_item_eth_mask;
1400 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1401 					(const uint8_t *)&nic_mask,
1402 					sizeof(struct rte_flow_item_eth),
1403 					error);
1404 	return ret;
1405 }
1406 
1407 /**
1408  * Validate VLAN item.
1409  *
1410  * @param[in] item
1411  *   Item specification.
1412  * @param[in] item_flags
1413  *   Bit-fields that holds the items detected until now.
1414  * @param[in] dev
1415  *   Ethernet device flow is being created on.
1416  * @param[out] error
1417  *   Pointer to error structure.
1418  *
1419  * @return
1420  *   0 on success, a negative errno value otherwise and rte_errno is set.
1421  */
1422 int
1423 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1424 			     uint64_t item_flags,
1425 			     struct rte_eth_dev *dev,
1426 			     struct rte_flow_error *error)
1427 {
1428 	const struct rte_flow_item_vlan *spec = item->spec;
1429 	const struct rte_flow_item_vlan *mask = item->mask;
1430 	const struct rte_flow_item_vlan nic_mask = {
1431 		.tci = RTE_BE16(UINT16_MAX),
1432 		.inner_type = RTE_BE16(UINT16_MAX),
1433 	};
1434 	uint16_t vlan_tag = 0;
1435 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1436 	int ret;
1437 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1438 					MLX5_FLOW_LAYER_INNER_L4) :
1439 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1440 					MLX5_FLOW_LAYER_OUTER_L4);
1441 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1442 					MLX5_FLOW_LAYER_OUTER_VLAN;
1443 
1444 	if (item_flags & vlanm)
1445 		return rte_flow_error_set(error, EINVAL,
1446 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1447 					  "multiple VLAN layers not supported");
1448 	else if ((item_flags & l34m) != 0)
1449 		return rte_flow_error_set(error, EINVAL,
1450 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1451 					  "VLAN cannot follow L3/L4 layer");
1452 	if (!mask)
1453 		mask = &rte_flow_item_vlan_mask;
1454 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1455 					(const uint8_t *)&nic_mask,
1456 					sizeof(struct rte_flow_item_vlan),
1457 					error);
1458 	if (ret)
1459 		return ret;
1460 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1461 		struct mlx5_priv *priv = dev->data->dev_private;
1462 
1463 		if (priv->vmwa_context) {
1464 			/*
1465 			 * Non-NULL context means we have a virtual machine
1466 			 * and SR-IOV enabled, we have to create VLAN interface
1467 			 * to make hypervisor to setup E-Switch vport
1468 			 * context correctly. We avoid creating the multiple
1469 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1470 			 */
1471 			return rte_flow_error_set(error, EINVAL,
1472 						  RTE_FLOW_ERROR_TYPE_ITEM,
1473 						  item,
1474 						  "VLAN tag mask is not"
1475 						  " supported in virtual"
1476 						  " environment");
1477 		}
1478 	}
1479 	if (spec) {
1480 		vlan_tag = spec->tci;
1481 		vlan_tag &= mask->tci;
1482 	}
1483 	/*
1484 	 * From verbs perspective an empty VLAN is equivalent
1485 	 * to a packet without VLAN layer.
1486 	 */
1487 	if (!vlan_tag)
1488 		return rte_flow_error_set(error, EINVAL,
1489 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1490 					  item->spec,
1491 					  "VLAN cannot be empty");
1492 	return 0;
1493 }
1494 
1495 /**
1496  * Validate IPV4 item.
1497  *
1498  * @param[in] item
1499  *   Item specification.
1500  * @param[in] item_flags
1501  *   Bit-fields that holds the items detected until now.
1502  * @param[in] acc_mask
1503  *   Acceptable mask, if NULL default internal default mask
1504  *   will be used to check whether item fields are supported.
1505  * @param[out] error
1506  *   Pointer to error structure.
1507  *
1508  * @return
1509  *   0 on success, a negative errno value otherwise and rte_errno is set.
1510  */
1511 int
1512 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1513 			     uint64_t item_flags,
1514 			     uint64_t last_item,
1515 			     uint16_t ether_type,
1516 			     const struct rte_flow_item_ipv4 *acc_mask,
1517 			     struct rte_flow_error *error)
1518 {
1519 	const struct rte_flow_item_ipv4 *mask = item->mask;
1520 	const struct rte_flow_item_ipv4 *spec = item->spec;
1521 	const struct rte_flow_item_ipv4 nic_mask = {
1522 		.hdr = {
1523 			.src_addr = RTE_BE32(0xffffffff),
1524 			.dst_addr = RTE_BE32(0xffffffff),
1525 			.type_of_service = 0xff,
1526 			.next_proto_id = 0xff,
1527 		},
1528 	};
1529 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1530 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1531 				      MLX5_FLOW_LAYER_OUTER_L3;
1532 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1533 				      MLX5_FLOW_LAYER_OUTER_L4;
1534 	int ret;
1535 	uint8_t next_proto = 0xFF;
1536 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1537 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1538 				  MLX5_FLOW_LAYER_INNER_VLAN);
1539 
1540 	if ((last_item & l2_vlan) && ether_type &&
1541 	    ether_type != RTE_ETHER_TYPE_IPV4)
1542 		return rte_flow_error_set(error, EINVAL,
1543 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1544 					  "IPv4 cannot follow L2/VLAN layer "
1545 					  "which ether type is not IPv4");
1546 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1547 		if (mask && spec)
1548 			next_proto = mask->hdr.next_proto_id &
1549 				     spec->hdr.next_proto_id;
1550 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1551 			return rte_flow_error_set(error, EINVAL,
1552 						  RTE_FLOW_ERROR_TYPE_ITEM,
1553 						  item,
1554 						  "multiple tunnel "
1555 						  "not supported");
1556 	}
1557 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1558 		return rte_flow_error_set(error, EINVAL,
1559 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1560 					  "wrong tunnel type - IPv6 specified "
1561 					  "but IPv4 item provided");
1562 	if (item_flags & l3m)
1563 		return rte_flow_error_set(error, ENOTSUP,
1564 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1565 					  "multiple L3 layers not supported");
1566 	else if (item_flags & l4m)
1567 		return rte_flow_error_set(error, EINVAL,
1568 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1569 					  "L3 cannot follow an L4 layer.");
1570 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1571 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1572 		return rte_flow_error_set(error, EINVAL,
1573 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1574 					  "L3 cannot follow an NVGRE layer.");
1575 	if (!mask)
1576 		mask = &rte_flow_item_ipv4_mask;
1577 	else if (mask->hdr.next_proto_id != 0 &&
1578 		 mask->hdr.next_proto_id != 0xff)
1579 		return rte_flow_error_set(error, EINVAL,
1580 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1581 					  "partial mask is not supported"
1582 					  " for protocol");
1583 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1584 					acc_mask ? (const uint8_t *)acc_mask
1585 						 : (const uint8_t *)&nic_mask,
1586 					sizeof(struct rte_flow_item_ipv4),
1587 					error);
1588 	if (ret < 0)
1589 		return ret;
1590 	return 0;
1591 }
1592 
1593 /**
1594  * Validate IPV6 item.
1595  *
1596  * @param[in] item
1597  *   Item specification.
1598  * @param[in] item_flags
1599  *   Bit-fields that holds the items detected until now.
1600  * @param[in] acc_mask
1601  *   Acceptable mask, if NULL default internal default mask
1602  *   will be used to check whether item fields are supported.
1603  * @param[out] error
1604  *   Pointer to error structure.
1605  *
1606  * @return
1607  *   0 on success, a negative errno value otherwise and rte_errno is set.
1608  */
1609 int
1610 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1611 			     uint64_t item_flags,
1612 			     uint64_t last_item,
1613 			     uint16_t ether_type,
1614 			     const struct rte_flow_item_ipv6 *acc_mask,
1615 			     struct rte_flow_error *error)
1616 {
1617 	const struct rte_flow_item_ipv6 *mask = item->mask;
1618 	const struct rte_flow_item_ipv6 *spec = item->spec;
1619 	const struct rte_flow_item_ipv6 nic_mask = {
1620 		.hdr = {
1621 			.src_addr =
1622 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1623 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1624 			.dst_addr =
1625 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1626 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1627 			.vtc_flow = RTE_BE32(0xffffffff),
1628 			.proto = 0xff,
1629 			.hop_limits = 0xff,
1630 		},
1631 	};
1632 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1633 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1634 				      MLX5_FLOW_LAYER_OUTER_L3;
1635 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1636 				      MLX5_FLOW_LAYER_OUTER_L4;
1637 	int ret;
1638 	uint8_t next_proto = 0xFF;
1639 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1640 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1641 				  MLX5_FLOW_LAYER_INNER_VLAN);
1642 
1643 	if ((last_item & l2_vlan) && ether_type &&
1644 	    ether_type != RTE_ETHER_TYPE_IPV6)
1645 		return rte_flow_error_set(error, EINVAL,
1646 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1647 					  "IPv6 cannot follow L2/VLAN layer "
1648 					  "which ether type is not IPv6");
1649 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1650 		if (mask && spec)
1651 			next_proto = mask->hdr.proto & spec->hdr.proto;
1652 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1653 			return rte_flow_error_set(error, EINVAL,
1654 						  RTE_FLOW_ERROR_TYPE_ITEM,
1655 						  item,
1656 						  "multiple tunnel "
1657 						  "not supported");
1658 	}
1659 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1660 		return rte_flow_error_set(error, EINVAL,
1661 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1662 					  "wrong tunnel type - IPv4 specified "
1663 					  "but IPv6 item provided");
1664 	if (item_flags & l3m)
1665 		return rte_flow_error_set(error, ENOTSUP,
1666 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1667 					  "multiple L3 layers not supported");
1668 	else if (item_flags & l4m)
1669 		return rte_flow_error_set(error, EINVAL,
1670 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1671 					  "L3 cannot follow an L4 layer.");
1672 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1673 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1674 		return rte_flow_error_set(error, EINVAL,
1675 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1676 					  "L3 cannot follow an NVGRE layer.");
1677 	if (!mask)
1678 		mask = &rte_flow_item_ipv6_mask;
1679 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1680 					acc_mask ? (const uint8_t *)acc_mask
1681 						 : (const uint8_t *)&nic_mask,
1682 					sizeof(struct rte_flow_item_ipv6),
1683 					error);
1684 	if (ret < 0)
1685 		return ret;
1686 	return 0;
1687 }
1688 
1689 /**
1690  * Validate UDP item.
1691  *
1692  * @param[in] item
1693  *   Item specification.
1694  * @param[in] item_flags
1695  *   Bit-fields that holds the items detected until now.
1696  * @param[in] target_protocol
1697  *   The next protocol in the previous item.
1698  * @param[in] flow_mask
1699  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1700  * @param[out] error
1701  *   Pointer to error structure.
1702  *
1703  * @return
1704  *   0 on success, a negative errno value otherwise and rte_errno is set.
1705  */
1706 int
1707 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1708 			    uint64_t item_flags,
1709 			    uint8_t target_protocol,
1710 			    struct rte_flow_error *error)
1711 {
1712 	const struct rte_flow_item_udp *mask = item->mask;
1713 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1714 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1715 				      MLX5_FLOW_LAYER_OUTER_L3;
1716 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1717 				      MLX5_FLOW_LAYER_OUTER_L4;
1718 	int ret;
1719 
1720 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1721 		return rte_flow_error_set(error, EINVAL,
1722 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1723 					  "protocol filtering not compatible"
1724 					  " with UDP layer");
1725 	if (!(item_flags & l3m))
1726 		return rte_flow_error_set(error, EINVAL,
1727 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1728 					  "L3 is mandatory to filter on L4");
1729 	if (item_flags & l4m)
1730 		return rte_flow_error_set(error, EINVAL,
1731 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1732 					  "multiple L4 layers not supported");
1733 	if (!mask)
1734 		mask = &rte_flow_item_udp_mask;
1735 	ret = mlx5_flow_item_acceptable
1736 		(item, (const uint8_t *)mask,
1737 		 (const uint8_t *)&rte_flow_item_udp_mask,
1738 		 sizeof(struct rte_flow_item_udp), error);
1739 	if (ret < 0)
1740 		return ret;
1741 	return 0;
1742 }
1743 
1744 /**
1745  * Validate TCP item.
1746  *
1747  * @param[in] item
1748  *   Item specification.
1749  * @param[in] item_flags
1750  *   Bit-fields that holds the items detected until now.
1751  * @param[in] target_protocol
1752  *   The next protocol in the previous item.
1753  * @param[out] error
1754  *   Pointer to error structure.
1755  *
1756  * @return
1757  *   0 on success, a negative errno value otherwise and rte_errno is set.
1758  */
1759 int
1760 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1761 			    uint64_t item_flags,
1762 			    uint8_t target_protocol,
1763 			    const struct rte_flow_item_tcp *flow_mask,
1764 			    struct rte_flow_error *error)
1765 {
1766 	const struct rte_flow_item_tcp *mask = item->mask;
1767 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1768 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1769 				      MLX5_FLOW_LAYER_OUTER_L3;
1770 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1771 				      MLX5_FLOW_LAYER_OUTER_L4;
1772 	int ret;
1773 
1774 	assert(flow_mask);
1775 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1776 		return rte_flow_error_set(error, EINVAL,
1777 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1778 					  "protocol filtering not compatible"
1779 					  " with TCP layer");
1780 	if (!(item_flags & l3m))
1781 		return rte_flow_error_set(error, EINVAL,
1782 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1783 					  "L3 is mandatory to filter on L4");
1784 	if (item_flags & l4m)
1785 		return rte_flow_error_set(error, EINVAL,
1786 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1787 					  "multiple L4 layers not supported");
1788 	if (!mask)
1789 		mask = &rte_flow_item_tcp_mask;
1790 	ret = mlx5_flow_item_acceptable
1791 		(item, (const uint8_t *)mask,
1792 		 (const uint8_t *)flow_mask,
1793 		 sizeof(struct rte_flow_item_tcp), error);
1794 	if (ret < 0)
1795 		return ret;
1796 	return 0;
1797 }
1798 
1799 /**
1800  * Validate VXLAN item.
1801  *
1802  * @param[in] item
1803  *   Item specification.
1804  * @param[in] item_flags
1805  *   Bit-fields that holds the items detected until now.
1806  * @param[in] target_protocol
1807  *   The next protocol in the previous item.
1808  * @param[out] error
1809  *   Pointer to error structure.
1810  *
1811  * @return
1812  *   0 on success, a negative errno value otherwise and rte_errno is set.
1813  */
1814 int
1815 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1816 			      uint64_t item_flags,
1817 			      struct rte_flow_error *error)
1818 {
1819 	const struct rte_flow_item_vxlan *spec = item->spec;
1820 	const struct rte_flow_item_vxlan *mask = item->mask;
1821 	int ret;
1822 	union vni {
1823 		uint32_t vlan_id;
1824 		uint8_t vni[4];
1825 	} id = { .vlan_id = 0, };
1826 	uint32_t vlan_id = 0;
1827 
1828 
1829 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1830 		return rte_flow_error_set(error, ENOTSUP,
1831 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1832 					  "multiple tunnel layers not"
1833 					  " supported");
1834 	/*
1835 	 * Verify only UDPv4 is present as defined in
1836 	 * https://tools.ietf.org/html/rfc7348
1837 	 */
1838 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1839 		return rte_flow_error_set(error, EINVAL,
1840 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1841 					  "no outer UDP layer found");
1842 	if (!mask)
1843 		mask = &rte_flow_item_vxlan_mask;
1844 	ret = mlx5_flow_item_acceptable
1845 		(item, (const uint8_t *)mask,
1846 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1847 		 sizeof(struct rte_flow_item_vxlan),
1848 		 error);
1849 	if (ret < 0)
1850 		return ret;
1851 	if (spec) {
1852 		memcpy(&id.vni[1], spec->vni, 3);
1853 		vlan_id = id.vlan_id;
1854 		memcpy(&id.vni[1], mask->vni, 3);
1855 		vlan_id &= id.vlan_id;
1856 	}
1857 	/*
1858 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1859 	 * only this layer is defined in the Verbs specification it is
1860 	 * interpreted as wildcard and all packets will match this
1861 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1862 	 * udp), all packets matching the layers before will also
1863 	 * match this rule.  To avoid such situation, VNI 0 is
1864 	 * currently refused.
1865 	 */
1866 	if (!vlan_id)
1867 		return rte_flow_error_set(error, ENOTSUP,
1868 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1869 					  "VXLAN vni cannot be 0");
1870 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1871 		return rte_flow_error_set(error, ENOTSUP,
1872 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1873 					  "VXLAN tunnel must be fully defined");
1874 	return 0;
1875 }
1876 
1877 /**
1878  * Validate VXLAN_GPE item.
1879  *
1880  * @param[in] item
1881  *   Item specification.
1882  * @param[in] item_flags
1883  *   Bit-fields that holds the items detected until now.
1884  * @param[in] priv
1885  *   Pointer to the private data structure.
1886  * @param[in] target_protocol
1887  *   The next protocol in the previous item.
1888  * @param[out] error
1889  *   Pointer to error structure.
1890  *
1891  * @return
1892  *   0 on success, a negative errno value otherwise and rte_errno is set.
1893  */
1894 int
1895 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1896 				  uint64_t item_flags,
1897 				  struct rte_eth_dev *dev,
1898 				  struct rte_flow_error *error)
1899 {
1900 	struct mlx5_priv *priv = dev->data->dev_private;
1901 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1902 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1903 	int ret;
1904 	union vni {
1905 		uint32_t vlan_id;
1906 		uint8_t vni[4];
1907 	} id = { .vlan_id = 0, };
1908 	uint32_t vlan_id = 0;
1909 
1910 	if (!priv->config.l3_vxlan_en)
1911 		return rte_flow_error_set(error, ENOTSUP,
1912 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1913 					  "L3 VXLAN is not enabled by device"
1914 					  " parameter and/or not configured in"
1915 					  " firmware");
1916 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1917 		return rte_flow_error_set(error, ENOTSUP,
1918 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1919 					  "multiple tunnel layers not"
1920 					  " supported");
1921 	/*
1922 	 * Verify only UDPv4 is present as defined in
1923 	 * https://tools.ietf.org/html/rfc7348
1924 	 */
1925 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1926 		return rte_flow_error_set(error, EINVAL,
1927 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1928 					  "no outer UDP layer found");
1929 	if (!mask)
1930 		mask = &rte_flow_item_vxlan_gpe_mask;
1931 	ret = mlx5_flow_item_acceptable
1932 		(item, (const uint8_t *)mask,
1933 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1934 		 sizeof(struct rte_flow_item_vxlan_gpe),
1935 		 error);
1936 	if (ret < 0)
1937 		return ret;
1938 	if (spec) {
1939 		if (spec->protocol)
1940 			return rte_flow_error_set(error, ENOTSUP,
1941 						  RTE_FLOW_ERROR_TYPE_ITEM,
1942 						  item,
1943 						  "VxLAN-GPE protocol"
1944 						  " not supported");
1945 		memcpy(&id.vni[1], spec->vni, 3);
1946 		vlan_id = id.vlan_id;
1947 		memcpy(&id.vni[1], mask->vni, 3);
1948 		vlan_id &= id.vlan_id;
1949 	}
1950 	/*
1951 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1952 	 * layer is defined in the Verbs specification it is interpreted as
1953 	 * wildcard and all packets will match this rule, if it follows a full
1954 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1955 	 * before will also match this rule.  To avoid such situation, VNI 0
1956 	 * is currently refused.
1957 	 */
1958 	if (!vlan_id)
1959 		return rte_flow_error_set(error, ENOTSUP,
1960 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1961 					  "VXLAN-GPE vni cannot be 0");
1962 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1963 		return rte_flow_error_set(error, ENOTSUP,
1964 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1965 					  "VXLAN-GPE tunnel must be fully"
1966 					  " defined");
1967 	return 0;
1968 }
1969 /**
1970  * Validate GRE Key item.
1971  *
1972  * @param[in] item
1973  *   Item specification.
1974  * @param[in] item_flags
1975  *   Bit flags to mark detected items.
1976  * @param[in] gre_item
1977  *   Pointer to gre_item
1978  * @param[out] error
1979  *   Pointer to error structure.
1980  *
1981  * @return
1982  *   0 on success, a negative errno value otherwise and rte_errno is set.
1983  */
1984 int
1985 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1986 				uint64_t item_flags,
1987 				const struct rte_flow_item *gre_item,
1988 				struct rte_flow_error *error)
1989 {
1990 	const rte_be32_t *mask = item->mask;
1991 	int ret = 0;
1992 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1993 	const struct rte_flow_item_gre *gre_spec = gre_item->spec;
1994 	const struct rte_flow_item_gre *gre_mask = gre_item->mask;
1995 
1996 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1997 		return rte_flow_error_set(error, ENOTSUP,
1998 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1999 					  "Multiple GRE key not support");
2000 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2001 		return rte_flow_error_set(error, ENOTSUP,
2002 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2003 					  "No preceding GRE header");
2004 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2005 		return rte_flow_error_set(error, ENOTSUP,
2006 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2007 					  "GRE key following a wrong item");
2008 	if (!gre_mask)
2009 		gre_mask = &rte_flow_item_gre_mask;
2010 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2011 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2012 		return rte_flow_error_set(error, EINVAL,
2013 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2014 					  "Key bit must be on");
2015 
2016 	if (!mask)
2017 		mask = &gre_key_default_mask;
2018 	ret = mlx5_flow_item_acceptable
2019 		(item, (const uint8_t *)mask,
2020 		 (const uint8_t *)&gre_key_default_mask,
2021 		 sizeof(rte_be32_t), error);
2022 	return ret;
2023 }
2024 
2025 /**
2026  * Validate GRE item.
2027  *
2028  * @param[in] item
2029  *   Item specification.
2030  * @param[in] item_flags
2031  *   Bit flags to mark detected items.
2032  * @param[in] target_protocol
2033  *   The next protocol in the previous item.
2034  * @param[out] error
2035  *   Pointer to error structure.
2036  *
2037  * @return
2038  *   0 on success, a negative errno value otherwise and rte_errno is set.
2039  */
2040 int
2041 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2042 			    uint64_t item_flags,
2043 			    uint8_t target_protocol,
2044 			    struct rte_flow_error *error)
2045 {
2046 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2047 	const struct rte_flow_item_gre *mask = item->mask;
2048 	int ret;
2049 	const struct rte_flow_item_gre nic_mask = {
2050 		.c_rsvd0_ver = RTE_BE16(0xB000),
2051 		.protocol = RTE_BE16(UINT16_MAX),
2052 	};
2053 
2054 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2055 		return rte_flow_error_set(error, EINVAL,
2056 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2057 					  "protocol filtering not compatible"
2058 					  " with this GRE layer");
2059 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2060 		return rte_flow_error_set(error, ENOTSUP,
2061 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2062 					  "multiple tunnel layers not"
2063 					  " supported");
2064 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2065 		return rte_flow_error_set(error, ENOTSUP,
2066 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2067 					  "L3 Layer is missing");
2068 	if (!mask)
2069 		mask = &rte_flow_item_gre_mask;
2070 	ret = mlx5_flow_item_acceptable
2071 		(item, (const uint8_t *)mask,
2072 		 (const uint8_t *)&nic_mask,
2073 		 sizeof(struct rte_flow_item_gre), error);
2074 	if (ret < 0)
2075 		return ret;
2076 #ifndef HAVE_MLX5DV_DR
2077 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2078 	if (spec && (spec->protocol & mask->protocol))
2079 		return rte_flow_error_set(error, ENOTSUP,
2080 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2081 					  "without MPLS support the"
2082 					  " specification cannot be used for"
2083 					  " filtering");
2084 #endif
2085 #endif
2086 	return 0;
2087 }
2088 
2089 /**
2090  * Validate Geneve item.
2091  *
2092  * @param[in] item
2093  *   Item specification.
2094  * @param[in] itemFlags
2095  *   Bit-fields that holds the items detected until now.
2096  * @param[in] enPriv
2097  *   Pointer to the private data structure.
2098  * @param[out] error
2099  *   Pointer to error structure.
2100  *
2101  * @return
2102  *   0 on success, a negative errno value otherwise and rte_errno is set.
2103  */
2104 
2105 int
2106 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2107 			       uint64_t item_flags,
2108 			       struct rte_eth_dev *dev,
2109 			       struct rte_flow_error *error)
2110 {
2111 	struct mlx5_priv *priv = dev->data->dev_private;
2112 	const struct rte_flow_item_geneve *spec = item->spec;
2113 	const struct rte_flow_item_geneve *mask = item->mask;
2114 	int ret;
2115 	uint16_t gbhdr;
2116 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2117 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2118 	const struct rte_flow_item_geneve nic_mask = {
2119 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2120 		.vni = "\xff\xff\xff",
2121 		.protocol = RTE_BE16(UINT16_MAX),
2122 	};
2123 
2124 	if (!(priv->config.hca_attr.flex_parser_protocols &
2125 	      MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2126 	    !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2127 		return rte_flow_error_set(error, ENOTSUP,
2128 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2129 					  "L3 Geneve is not enabled by device"
2130 					  " parameter and/or not configured in"
2131 					  " firmware");
2132 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2133 		return rte_flow_error_set(error, ENOTSUP,
2134 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2135 					  "multiple tunnel layers not"
2136 					  " supported");
2137 	/*
2138 	 * Verify only UDPv4 is present as defined in
2139 	 * https://tools.ietf.org/html/rfc7348
2140 	 */
2141 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2142 		return rte_flow_error_set(error, EINVAL,
2143 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2144 					  "no outer UDP layer found");
2145 	if (!mask)
2146 		mask = &rte_flow_item_geneve_mask;
2147 	ret = mlx5_flow_item_acceptable
2148 				  (item, (const uint8_t *)mask,
2149 				   (const uint8_t *)&nic_mask,
2150 				   sizeof(struct rte_flow_item_geneve), error);
2151 	if (ret)
2152 		return ret;
2153 	if (spec) {
2154 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2155 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2156 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2157 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2158 			return rte_flow_error_set(error, ENOTSUP,
2159 						  RTE_FLOW_ERROR_TYPE_ITEM,
2160 						  item,
2161 						  "Geneve protocol unsupported"
2162 						  " fields are being used");
2163 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2164 			return rte_flow_error_set
2165 					(error, ENOTSUP,
2166 					 RTE_FLOW_ERROR_TYPE_ITEM,
2167 					 item,
2168 					 "Unsupported Geneve options length");
2169 	}
2170 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2171 		return rte_flow_error_set
2172 				    (error, ENOTSUP,
2173 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2174 				     "Geneve tunnel must be fully defined");
2175 	return 0;
2176 }
2177 
2178 /**
2179  * Validate MPLS item.
2180  *
2181  * @param[in] dev
2182  *   Pointer to the rte_eth_dev structure.
2183  * @param[in] item
2184  *   Item specification.
2185  * @param[in] item_flags
2186  *   Bit-fields that holds the items detected until now.
2187  * @param[in] prev_layer
2188  *   The protocol layer indicated in previous item.
2189  * @param[out] error
2190  *   Pointer to error structure.
2191  *
2192  * @return
2193  *   0 on success, a negative errno value otherwise and rte_errno is set.
2194  */
2195 int
2196 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2197 			     const struct rte_flow_item *item __rte_unused,
2198 			     uint64_t item_flags __rte_unused,
2199 			     uint64_t prev_layer __rte_unused,
2200 			     struct rte_flow_error *error)
2201 {
2202 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2203 	const struct rte_flow_item_mpls *mask = item->mask;
2204 	struct mlx5_priv *priv = dev->data->dev_private;
2205 	int ret;
2206 
2207 	if (!priv->config.mpls_en)
2208 		return rte_flow_error_set(error, ENOTSUP,
2209 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2210 					  "MPLS not supported or"
2211 					  " disabled in firmware"
2212 					  " configuration.");
2213 	/* MPLS over IP, UDP, GRE is allowed */
2214 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2215 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2216 			    MLX5_FLOW_LAYER_GRE)))
2217 		return rte_flow_error_set(error, EINVAL,
2218 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2219 					  "protocol filtering not compatible"
2220 					  " with MPLS layer");
2221 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2222 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2223 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2224 		return rte_flow_error_set(error, ENOTSUP,
2225 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2226 					  "multiple tunnel layers not"
2227 					  " supported");
2228 	if (!mask)
2229 		mask = &rte_flow_item_mpls_mask;
2230 	ret = mlx5_flow_item_acceptable
2231 		(item, (const uint8_t *)mask,
2232 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2233 		 sizeof(struct rte_flow_item_mpls), error);
2234 	if (ret < 0)
2235 		return ret;
2236 	return 0;
2237 #endif
2238 	return rte_flow_error_set(error, ENOTSUP,
2239 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2240 				  "MPLS is not supported by Verbs, please"
2241 				  " update.");
2242 }
2243 
2244 /**
2245  * Validate NVGRE item.
2246  *
2247  * @param[in] item
2248  *   Item specification.
2249  * @param[in] item_flags
2250  *   Bit flags to mark detected items.
2251  * @param[in] target_protocol
2252  *   The next protocol in the previous item.
2253  * @param[out] error
2254  *   Pointer to error structure.
2255  *
2256  * @return
2257  *   0 on success, a negative errno value otherwise and rte_errno is set.
2258  */
2259 int
2260 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2261 			      uint64_t item_flags,
2262 			      uint8_t target_protocol,
2263 			      struct rte_flow_error *error)
2264 {
2265 	const struct rte_flow_item_nvgre *mask = item->mask;
2266 	int ret;
2267 
2268 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2269 		return rte_flow_error_set(error, EINVAL,
2270 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2271 					  "protocol filtering not compatible"
2272 					  " with this GRE layer");
2273 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2274 		return rte_flow_error_set(error, ENOTSUP,
2275 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2276 					  "multiple tunnel layers not"
2277 					  " supported");
2278 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2279 		return rte_flow_error_set(error, ENOTSUP,
2280 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2281 					  "L3 Layer is missing");
2282 	if (!mask)
2283 		mask = &rte_flow_item_nvgre_mask;
2284 	ret = mlx5_flow_item_acceptable
2285 		(item, (const uint8_t *)mask,
2286 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2287 		 sizeof(struct rte_flow_item_nvgre), error);
2288 	if (ret < 0)
2289 		return ret;
2290 	return 0;
2291 }
2292 
2293 /* Allocate unique ID for the split Q/RSS subflows. */
2294 static uint32_t
2295 flow_qrss_get_id(struct rte_eth_dev *dev)
2296 {
2297 	struct mlx5_priv *priv = dev->data->dev_private;
2298 	uint32_t qrss_id, ret;
2299 
2300 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2301 	if (ret)
2302 		return 0;
2303 	assert(qrss_id);
2304 	return qrss_id;
2305 }
2306 
2307 /* Free unique ID for the split Q/RSS subflows. */
2308 static void
2309 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2310 {
2311 	struct mlx5_priv *priv = dev->data->dev_private;
2312 
2313 	if (qrss_id)
2314 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2315 }
2316 
2317 /**
2318  * Release resource related QUEUE/RSS action split.
2319  *
2320  * @param dev
2321  *   Pointer to Ethernet device.
2322  * @param flow
2323  *   Flow to release id's from.
2324  */
2325 static void
2326 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2327 			     struct rte_flow *flow)
2328 {
2329 	struct mlx5_flow *dev_flow;
2330 
2331 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2332 		if (dev_flow->qrss_id)
2333 			flow_qrss_free_id(dev, dev_flow->qrss_id);
2334 }
2335 
2336 static int
2337 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2338 		   const struct rte_flow_attr *attr __rte_unused,
2339 		   const struct rte_flow_item items[] __rte_unused,
2340 		   const struct rte_flow_action actions[] __rte_unused,
2341 		   bool external __rte_unused,
2342 		   struct rte_flow_error *error)
2343 {
2344 	return rte_flow_error_set(error, ENOTSUP,
2345 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2346 }
2347 
2348 static struct mlx5_flow *
2349 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2350 		  const struct rte_flow_item items[] __rte_unused,
2351 		  const struct rte_flow_action actions[] __rte_unused,
2352 		  struct rte_flow_error *error)
2353 {
2354 	rte_flow_error_set(error, ENOTSUP,
2355 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2356 	return NULL;
2357 }
2358 
2359 static int
2360 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2361 		    struct mlx5_flow *dev_flow __rte_unused,
2362 		    const struct rte_flow_attr *attr __rte_unused,
2363 		    const struct rte_flow_item items[] __rte_unused,
2364 		    const struct rte_flow_action actions[] __rte_unused,
2365 		    struct rte_flow_error *error)
2366 {
2367 	return rte_flow_error_set(error, ENOTSUP,
2368 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2369 }
2370 
2371 static int
2372 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2373 		struct rte_flow *flow __rte_unused,
2374 		struct rte_flow_error *error)
2375 {
2376 	return rte_flow_error_set(error, ENOTSUP,
2377 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2378 }
2379 
2380 static void
2381 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2382 		 struct rte_flow *flow __rte_unused)
2383 {
2384 }
2385 
2386 static void
2387 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2388 		  struct rte_flow *flow __rte_unused)
2389 {
2390 }
2391 
2392 static int
2393 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2394 		struct rte_flow *flow __rte_unused,
2395 		const struct rte_flow_action *actions __rte_unused,
2396 		void *data __rte_unused,
2397 		struct rte_flow_error *error)
2398 {
2399 	return rte_flow_error_set(error, ENOTSUP,
2400 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2401 }
2402 
2403 /* Void driver to protect from null pointer reference. */
2404 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2405 	.validate = flow_null_validate,
2406 	.prepare = flow_null_prepare,
2407 	.translate = flow_null_translate,
2408 	.apply = flow_null_apply,
2409 	.remove = flow_null_remove,
2410 	.destroy = flow_null_destroy,
2411 	.query = flow_null_query,
2412 };
2413 
2414 /**
2415  * Select flow driver type according to flow attributes and device
2416  * configuration.
2417  *
2418  * @param[in] dev
2419  *   Pointer to the dev structure.
2420  * @param[in] attr
2421  *   Pointer to the flow attributes.
2422  *
2423  * @return
2424  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2425  */
2426 static enum mlx5_flow_drv_type
2427 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2428 {
2429 	struct mlx5_priv *priv = dev->data->dev_private;
2430 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2431 
2432 	if (attr->transfer && priv->config.dv_esw_en)
2433 		type = MLX5_FLOW_TYPE_DV;
2434 	if (!attr->transfer)
2435 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2436 						 MLX5_FLOW_TYPE_VERBS;
2437 	return type;
2438 }
2439 
2440 #define flow_get_drv_ops(type) flow_drv_ops[type]
2441 
2442 /**
2443  * Flow driver validation API. This abstracts calling driver specific functions.
2444  * The type of flow driver is determined according to flow attributes.
2445  *
2446  * @param[in] dev
2447  *   Pointer to the dev structure.
2448  * @param[in] attr
2449  *   Pointer to the flow attributes.
2450  * @param[in] items
2451  *   Pointer to the list of items.
2452  * @param[in] actions
2453  *   Pointer to the list of actions.
2454  * @param[in] external
2455  *   This flow rule is created by request external to PMD.
2456  * @param[out] error
2457  *   Pointer to the error structure.
2458  *
2459  * @return
2460  *   0 on success, a negative errno value otherwise and rte_errno is set.
2461  */
2462 static inline int
2463 flow_drv_validate(struct rte_eth_dev *dev,
2464 		  const struct rte_flow_attr *attr,
2465 		  const struct rte_flow_item items[],
2466 		  const struct rte_flow_action actions[],
2467 		  bool external, struct rte_flow_error *error)
2468 {
2469 	const struct mlx5_flow_driver_ops *fops;
2470 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2471 
2472 	fops = flow_get_drv_ops(type);
2473 	return fops->validate(dev, attr, items, actions, external, error);
2474 }
2475 
2476 /**
2477  * Flow driver preparation API. This abstracts calling driver specific
2478  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2479  * calculates the size of memory required for device flow, allocates the memory,
2480  * initializes the device flow and returns the pointer.
2481  *
2482  * @note
2483  *   This function initializes device flow structure such as dv or verbs in
2484  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2485  *   rest. For example, adding returning device flow to flow->dev_flow list and
2486  *   setting backward reference to the flow should be done out of this function.
2487  *   layers field is not filled either.
2488  *
2489  * @param[in] attr
2490  *   Pointer to the flow attributes.
2491  * @param[in] items
2492  *   Pointer to the list of items.
2493  * @param[in] actions
2494  *   Pointer to the list of actions.
2495  * @param[out] error
2496  *   Pointer to the error structure.
2497  *
2498  * @return
2499  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2500  */
2501 static inline struct mlx5_flow *
2502 flow_drv_prepare(const struct rte_flow *flow,
2503 		 const struct rte_flow_attr *attr,
2504 		 const struct rte_flow_item items[],
2505 		 const struct rte_flow_action actions[],
2506 		 struct rte_flow_error *error)
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 	return fops->prepare(attr, items, actions, error);
2514 }
2515 
2516 /**
2517  * Flow driver translation API. This abstracts calling driver specific
2518  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2519  * translates a generic flow into a driver flow. flow_drv_prepare() must
2520  * precede.
2521  *
2522  * @note
2523  *   dev_flow->layers could be filled as a result of parsing during translation
2524  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2525  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2526  *   flow->actions could be overwritten even though all the expanded dev_flows
2527  *   have the same actions.
2528  *
2529  * @param[in] dev
2530  *   Pointer to the rte dev structure.
2531  * @param[in, out] dev_flow
2532  *   Pointer to the mlx5 flow.
2533  * @param[in] attr
2534  *   Pointer to the flow attributes.
2535  * @param[in] items
2536  *   Pointer to the list of items.
2537  * @param[in] actions
2538  *   Pointer to the list of actions.
2539  * @param[out] error
2540  *   Pointer to the error structure.
2541  *
2542  * @return
2543  *   0 on success, a negative errno value otherwise and rte_errno is set.
2544  */
2545 static inline int
2546 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2547 		   const struct rte_flow_attr *attr,
2548 		   const struct rte_flow_item items[],
2549 		   const struct rte_flow_action actions[],
2550 		   struct rte_flow_error *error)
2551 {
2552 	const struct mlx5_flow_driver_ops *fops;
2553 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2554 
2555 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2556 	fops = flow_get_drv_ops(type);
2557 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2558 }
2559 
2560 /**
2561  * Flow driver apply API. This abstracts calling driver specific functions.
2562  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2563  * translated driver flows on to device. flow_drv_translate() must precede.
2564  *
2565  * @param[in] dev
2566  *   Pointer to Ethernet device structure.
2567  * @param[in, out] flow
2568  *   Pointer to flow structure.
2569  * @param[out] error
2570  *   Pointer to error structure.
2571  *
2572  * @return
2573  *   0 on success, a negative errno value otherwise and rte_errno is set.
2574  */
2575 static inline int
2576 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2577 	       struct rte_flow_error *error)
2578 {
2579 	const struct mlx5_flow_driver_ops *fops;
2580 	enum mlx5_flow_drv_type type = flow->drv_type;
2581 
2582 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2583 	fops = flow_get_drv_ops(type);
2584 	return fops->apply(dev, flow, error);
2585 }
2586 
2587 /**
2588  * Flow driver remove API. This abstracts calling driver specific functions.
2589  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2590  * on device. All the resources of the flow should be freed by calling
2591  * flow_drv_destroy().
2592  *
2593  * @param[in] dev
2594  *   Pointer to Ethernet device.
2595  * @param[in, out] flow
2596  *   Pointer to flow structure.
2597  */
2598 static inline void
2599 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2600 {
2601 	const struct mlx5_flow_driver_ops *fops;
2602 	enum mlx5_flow_drv_type type = flow->drv_type;
2603 
2604 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2605 	fops = flow_get_drv_ops(type);
2606 	fops->remove(dev, flow);
2607 }
2608 
2609 /**
2610  * Flow driver destroy API. This abstracts calling driver specific functions.
2611  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2612  * on device and releases resources of the flow.
2613  *
2614  * @param[in] dev
2615  *   Pointer to Ethernet device.
2616  * @param[in, out] flow
2617  *   Pointer to flow structure.
2618  */
2619 static inline void
2620 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2621 {
2622 	const struct mlx5_flow_driver_ops *fops;
2623 	enum mlx5_flow_drv_type type = flow->drv_type;
2624 
2625 	flow_mreg_split_qrss_release(dev, flow);
2626 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2627 	fops = flow_get_drv_ops(type);
2628 	fops->destroy(dev, flow);
2629 }
2630 
2631 /**
2632  * Validate a flow supported by the NIC.
2633  *
2634  * @see rte_flow_validate()
2635  * @see rte_flow_ops
2636  */
2637 int
2638 mlx5_flow_validate(struct rte_eth_dev *dev,
2639 		   const struct rte_flow_attr *attr,
2640 		   const struct rte_flow_item items[],
2641 		   const struct rte_flow_action actions[],
2642 		   struct rte_flow_error *error)
2643 {
2644 	int ret;
2645 
2646 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2647 	if (ret < 0)
2648 		return ret;
2649 	return 0;
2650 }
2651 
2652 /**
2653  * Get port id item from the item list.
2654  *
2655  * @param[in] item
2656  *   Pointer to the list of items.
2657  *
2658  * @return
2659  *   Pointer to the port id item if exist, else return NULL.
2660  */
2661 static const struct rte_flow_item *
2662 find_port_id_item(const struct rte_flow_item *item)
2663 {
2664 	assert(item);
2665 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2666 		if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2667 			return item;
2668 	}
2669 	return NULL;
2670 }
2671 
2672 /**
2673  * Get RSS action from the action list.
2674  *
2675  * @param[in] actions
2676  *   Pointer to the list of actions.
2677  *
2678  * @return
2679  *   Pointer to the RSS action if exist, else return NULL.
2680  */
2681 static const struct rte_flow_action_rss*
2682 flow_get_rss_action(const struct rte_flow_action actions[])
2683 {
2684 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2685 		switch (actions->type) {
2686 		case RTE_FLOW_ACTION_TYPE_RSS:
2687 			return (const struct rte_flow_action_rss *)
2688 			       actions->conf;
2689 		default:
2690 			break;
2691 		}
2692 	}
2693 	return NULL;
2694 }
2695 
2696 static unsigned int
2697 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2698 {
2699 	const struct rte_flow_item *item;
2700 	unsigned int has_vlan = 0;
2701 
2702 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2703 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2704 			has_vlan = 1;
2705 			break;
2706 		}
2707 	}
2708 	if (has_vlan)
2709 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2710 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2711 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2712 			       MLX5_EXPANSION_ROOT_OUTER;
2713 }
2714 
2715 /**
2716  * Get QUEUE/RSS action from the action list.
2717  *
2718  * @param[in] actions
2719  *   Pointer to the list of actions.
2720  * @param[out] qrss
2721  *   Pointer to the return pointer.
2722  * @param[out] qrss_type
2723  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2724  *   if no QUEUE/RSS is found.
2725  *
2726  * @return
2727  *   Total number of actions.
2728  */
2729 static int
2730 flow_parse_qrss_action(const struct rte_flow_action actions[],
2731 		       const struct rte_flow_action **qrss)
2732 {
2733 	int actions_n = 0;
2734 
2735 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2736 		switch (actions->type) {
2737 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2738 		case RTE_FLOW_ACTION_TYPE_RSS:
2739 			*qrss = actions;
2740 			break;
2741 		default:
2742 			break;
2743 		}
2744 		actions_n++;
2745 	}
2746 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2747 	return actions_n + 1;
2748 }
2749 
2750 /**
2751  * Check meter action from the action list.
2752  *
2753  * @param[in] actions
2754  *   Pointer to the list of actions.
2755  * @param[out] mtr
2756  *   Pointer to the meter exist flag.
2757  *
2758  * @return
2759  *   Total number of actions.
2760  */
2761 static int
2762 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2763 {
2764 	int actions_n = 0;
2765 
2766 	assert(mtr);
2767 	*mtr = 0;
2768 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2769 		switch (actions->type) {
2770 		case RTE_FLOW_ACTION_TYPE_METER:
2771 			*mtr = 1;
2772 			break;
2773 		default:
2774 			break;
2775 		}
2776 		actions_n++;
2777 	}
2778 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2779 	return actions_n + 1;
2780 }
2781 
2782 /**
2783  * Check if the flow should be splited due to hairpin.
2784  * The reason for the split is that in current HW we can't
2785  * support encap on Rx, so if a flow have encap we move it
2786  * to Tx.
2787  *
2788  * @param dev
2789  *   Pointer to Ethernet device.
2790  * @param[in] attr
2791  *   Flow rule attributes.
2792  * @param[in] actions
2793  *   Associated actions (list terminated by the END action).
2794  *
2795  * @return
2796  *   > 0 the number of actions and the flow should be split,
2797  *   0 when no split required.
2798  */
2799 static int
2800 flow_check_hairpin_split(struct rte_eth_dev *dev,
2801 			 const struct rte_flow_attr *attr,
2802 			 const struct rte_flow_action actions[])
2803 {
2804 	int queue_action = 0;
2805 	int action_n = 0;
2806 	int encap = 0;
2807 	const struct rte_flow_action_queue *queue;
2808 	const struct rte_flow_action_rss *rss;
2809 	const struct rte_flow_action_raw_encap *raw_encap;
2810 
2811 	if (!attr->ingress)
2812 		return 0;
2813 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2814 		switch (actions->type) {
2815 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2816 			queue = actions->conf;
2817 			if (mlx5_rxq_get_type(dev, queue->index) !=
2818 			    MLX5_RXQ_TYPE_HAIRPIN)
2819 				return 0;
2820 			queue_action = 1;
2821 			action_n++;
2822 			break;
2823 		case RTE_FLOW_ACTION_TYPE_RSS:
2824 			rss = actions->conf;
2825 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2826 			    MLX5_RXQ_TYPE_HAIRPIN)
2827 				return 0;
2828 			queue_action = 1;
2829 			action_n++;
2830 			break;
2831 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2832 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2833 			encap = 1;
2834 			action_n++;
2835 			break;
2836 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2837 			raw_encap = actions->conf;
2838 			if (raw_encap->size >
2839 			    (sizeof(struct rte_flow_item_eth) +
2840 			     sizeof(struct rte_flow_item_ipv4)))
2841 				encap = 1;
2842 			action_n++;
2843 			break;
2844 		default:
2845 			action_n++;
2846 			break;
2847 		}
2848 	}
2849 	if (encap == 1 && queue_action)
2850 		return action_n;
2851 	return 0;
2852 }
2853 
2854 /* Declare flow create/destroy prototype in advance. */
2855 static struct rte_flow *
2856 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2857 		 const struct rte_flow_attr *attr,
2858 		 const struct rte_flow_item items[],
2859 		 const struct rte_flow_action actions[],
2860 		 bool external, struct rte_flow_error *error);
2861 
2862 static void
2863 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2864 		  struct rte_flow *flow);
2865 
2866 /**
2867  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2868  *
2869  * As mark_id is unique, if there's already a registered flow for the mark_id,
2870  * return by increasing the reference counter of the resource. Otherwise, create
2871  * the resource (mcp_res) and flow.
2872  *
2873  * Flow looks like,
2874  *   - If ingress port is ANY and reg_c[1] is mark_id,
2875  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2876  *
2877  * For default flow (zero mark_id), flow is like,
2878  *   - If ingress port is ANY,
2879  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2880  *
2881  * @param dev
2882  *   Pointer to Ethernet device.
2883  * @param mark_id
2884  *   ID of MARK action, zero means default flow for META.
2885  * @param[out] error
2886  *   Perform verbose error reporting if not NULL.
2887  *
2888  * @return
2889  *   Associated resource on success, NULL otherwise and rte_errno is set.
2890  */
2891 static struct mlx5_flow_mreg_copy_resource *
2892 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2893 			  struct rte_flow_error *error)
2894 {
2895 	struct mlx5_priv *priv = dev->data->dev_private;
2896 	struct rte_flow_attr attr = {
2897 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2898 		.ingress = 1,
2899 	};
2900 	struct mlx5_rte_flow_item_tag tag_spec = {
2901 		.data = mark_id,
2902 	};
2903 	struct rte_flow_item items[] = {
2904 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2905 	};
2906 	struct rte_flow_action_mark ftag = {
2907 		.id = mark_id,
2908 	};
2909 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2910 		.dst = REG_B,
2911 		.src = 0,
2912 	};
2913 	struct rte_flow_action_jump jump = {
2914 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2915 	};
2916 	struct rte_flow_action actions[] = {
2917 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2918 	};
2919 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2920 	int ret;
2921 
2922 	/* Fill the register fileds in the flow. */
2923 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2924 	if (ret < 0)
2925 		return NULL;
2926 	tag_spec.id = ret;
2927 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2928 	if (ret < 0)
2929 		return NULL;
2930 	cp_mreg.src = ret;
2931 	/* Check if already registered. */
2932 	assert(priv->mreg_cp_tbl);
2933 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2934 	if (mcp_res) {
2935 		/* For non-default rule. */
2936 		if (mark_id)
2937 			mcp_res->refcnt++;
2938 		assert(mark_id || mcp_res->refcnt == 1);
2939 		return mcp_res;
2940 	}
2941 	/* Provide the full width of FLAG specific value. */
2942 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2943 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2944 	/* Build a new flow. */
2945 	if (mark_id) {
2946 		items[0] = (struct rte_flow_item){
2947 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2948 			.spec = &tag_spec,
2949 		};
2950 		items[1] = (struct rte_flow_item){
2951 			.type = RTE_FLOW_ITEM_TYPE_END,
2952 		};
2953 		actions[0] = (struct rte_flow_action){
2954 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2955 			.conf = &ftag,
2956 		};
2957 		actions[1] = (struct rte_flow_action){
2958 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2959 			.conf = &cp_mreg,
2960 		};
2961 		actions[2] = (struct rte_flow_action){
2962 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2963 			.conf = &jump,
2964 		};
2965 		actions[3] = (struct rte_flow_action){
2966 			.type = RTE_FLOW_ACTION_TYPE_END,
2967 		};
2968 	} else {
2969 		/* Default rule, wildcard match. */
2970 		attr.priority = MLX5_FLOW_PRIO_RSVD;
2971 		items[0] = (struct rte_flow_item){
2972 			.type = RTE_FLOW_ITEM_TYPE_END,
2973 		};
2974 		actions[0] = (struct rte_flow_action){
2975 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2976 			.conf = &cp_mreg,
2977 		};
2978 		actions[1] = (struct rte_flow_action){
2979 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2980 			.conf = &jump,
2981 		};
2982 		actions[2] = (struct rte_flow_action){
2983 			.type = RTE_FLOW_ACTION_TYPE_END,
2984 		};
2985 	}
2986 	/* Build a new entry. */
2987 	mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
2988 	if (!mcp_res) {
2989 		rte_errno = ENOMEM;
2990 		return NULL;
2991 	}
2992 	/*
2993 	 * The copy Flows are not included in any list. There
2994 	 * ones are referenced from other Flows and can not
2995 	 * be applied, removed, deleted in ardbitrary order
2996 	 * by list traversing.
2997 	 */
2998 	mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
2999 					 actions, false, error);
3000 	if (!mcp_res->flow)
3001 		goto error;
3002 	mcp_res->refcnt++;
3003 	mcp_res->hlist_ent.key = mark_id;
3004 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3005 				&mcp_res->hlist_ent);
3006 	assert(!ret);
3007 	if (ret)
3008 		goto error;
3009 	return mcp_res;
3010 error:
3011 	if (mcp_res->flow)
3012 		flow_list_destroy(dev, NULL, mcp_res->flow);
3013 	rte_free(mcp_res);
3014 	return NULL;
3015 }
3016 
3017 /**
3018  * Release flow in RX_CP_TBL.
3019  *
3020  * @param dev
3021  *   Pointer to Ethernet device.
3022  * @flow
3023  *   Parent flow for wich copying is provided.
3024  */
3025 static void
3026 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3027 			  struct rte_flow *flow)
3028 {
3029 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3030 	struct mlx5_priv *priv = dev->data->dev_private;
3031 
3032 	if (!mcp_res || !priv->mreg_cp_tbl)
3033 		return;
3034 	if (flow->copy_applied) {
3035 		assert(mcp_res->appcnt);
3036 		flow->copy_applied = 0;
3037 		--mcp_res->appcnt;
3038 		if (!mcp_res->appcnt)
3039 			flow_drv_remove(dev, mcp_res->flow);
3040 	}
3041 	/*
3042 	 * We do not check availability of metadata registers here,
3043 	 * because copy resources are allocated in this case.
3044 	 */
3045 	if (--mcp_res->refcnt)
3046 		return;
3047 	assert(mcp_res->flow);
3048 	flow_list_destroy(dev, NULL, mcp_res->flow);
3049 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3050 	rte_free(mcp_res);
3051 	flow->mreg_copy = NULL;
3052 }
3053 
3054 /**
3055  * Start flow in RX_CP_TBL.
3056  *
3057  * @param dev
3058  *   Pointer to Ethernet device.
3059  * @flow
3060  *   Parent flow for wich copying is provided.
3061  *
3062  * @return
3063  *   0 on success, a negative errno value otherwise and rte_errno is set.
3064  */
3065 static int
3066 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3067 			    struct rte_flow *flow)
3068 {
3069 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3070 	int ret;
3071 
3072 	if (!mcp_res || flow->copy_applied)
3073 		return 0;
3074 	if (!mcp_res->appcnt) {
3075 		ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3076 		if (ret)
3077 			return ret;
3078 	}
3079 	++mcp_res->appcnt;
3080 	flow->copy_applied = 1;
3081 	return 0;
3082 }
3083 
3084 /**
3085  * Stop flow in RX_CP_TBL.
3086  *
3087  * @param dev
3088  *   Pointer to Ethernet device.
3089  * @flow
3090  *   Parent flow for wich copying is provided.
3091  */
3092 static void
3093 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3094 			   struct rte_flow *flow)
3095 {
3096 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3097 
3098 	if (!mcp_res || !flow->copy_applied)
3099 		return;
3100 	assert(mcp_res->appcnt);
3101 	--mcp_res->appcnt;
3102 	flow->copy_applied = 0;
3103 	if (!mcp_res->appcnt)
3104 		flow_drv_remove(dev, mcp_res->flow);
3105 }
3106 
3107 /**
3108  * Remove the default copy action from RX_CP_TBL.
3109  *
3110  * @param dev
3111  *   Pointer to Ethernet device.
3112  */
3113 static void
3114 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3115 {
3116 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3117 	struct mlx5_priv *priv = dev->data->dev_private;
3118 
3119 	/* Check if default flow is registered. */
3120 	if (!priv->mreg_cp_tbl)
3121 		return;
3122 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, 0ULL);
3123 	if (!mcp_res)
3124 		return;
3125 	assert(mcp_res->flow);
3126 	flow_list_destroy(dev, NULL, mcp_res->flow);
3127 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3128 	rte_free(mcp_res);
3129 }
3130 
3131 /**
3132  * Add the default copy action in in RX_CP_TBL.
3133  *
3134  * @param dev
3135  *   Pointer to Ethernet device.
3136  * @param[out] error
3137  *   Perform verbose error reporting if not NULL.
3138  *
3139  * @return
3140  *   0 for success, negative value otherwise and rte_errno is set.
3141  */
3142 static int
3143 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3144 				  struct rte_flow_error *error)
3145 {
3146 	struct mlx5_priv *priv = dev->data->dev_private;
3147 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3148 
3149 	/* Check whether extensive metadata feature is engaged. */
3150 	if (!priv->config.dv_flow_en ||
3151 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3152 	    !mlx5_flow_ext_mreg_supported(dev) ||
3153 	    !priv->sh->dv_regc0_mask)
3154 		return 0;
3155 	mcp_res = flow_mreg_add_copy_action(dev, 0, error);
3156 	if (!mcp_res)
3157 		return -rte_errno;
3158 	return 0;
3159 }
3160 
3161 /**
3162  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3163  *
3164  * All the flow having Q/RSS action should be split by
3165  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3166  * performs the following,
3167  *   - CQE->flow_tag := reg_c[1] (MARK)
3168  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3169  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3170  * but there should be a flow per each MARK ID set by MARK action.
3171  *
3172  * For the aforementioned reason, if there's a MARK action in flow's action
3173  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3174  * the MARK ID to CQE's flow_tag like,
3175  *   - If reg_c[1] is mark_id,
3176  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3177  *
3178  * For SET_META action which stores value in reg_c[0], as the destination is
3179  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3180  * MARK ID means the default flow. The default flow looks like,
3181  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3182  *
3183  * @param dev
3184  *   Pointer to Ethernet device.
3185  * @param flow
3186  *   Pointer to flow structure.
3187  * @param[in] actions
3188  *   Pointer to the list of actions.
3189  * @param[out] error
3190  *   Perform verbose error reporting if not NULL.
3191  *
3192  * @return
3193  *   0 on success, negative value otherwise and rte_errno is set.
3194  */
3195 static int
3196 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3197 			    struct rte_flow *flow,
3198 			    const struct rte_flow_action *actions,
3199 			    struct rte_flow_error *error)
3200 {
3201 	struct mlx5_priv *priv = dev->data->dev_private;
3202 	struct mlx5_dev_config *config = &priv->config;
3203 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3204 	const struct rte_flow_action_mark *mark;
3205 
3206 	/* Check whether extensive metadata feature is engaged. */
3207 	if (!config->dv_flow_en ||
3208 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3209 	    !mlx5_flow_ext_mreg_supported(dev) ||
3210 	    !priv->sh->dv_regc0_mask)
3211 		return 0;
3212 	/* Find MARK action. */
3213 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3214 		switch (actions->type) {
3215 		case RTE_FLOW_ACTION_TYPE_FLAG:
3216 			mcp_res = flow_mreg_add_copy_action
3217 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3218 			if (!mcp_res)
3219 				return -rte_errno;
3220 			flow->mreg_copy = mcp_res;
3221 			if (dev->data->dev_started) {
3222 				mcp_res->appcnt++;
3223 				flow->copy_applied = 1;
3224 			}
3225 			return 0;
3226 		case RTE_FLOW_ACTION_TYPE_MARK:
3227 			mark = (const struct rte_flow_action_mark *)
3228 				actions->conf;
3229 			mcp_res =
3230 				flow_mreg_add_copy_action(dev, mark->id, error);
3231 			if (!mcp_res)
3232 				return -rte_errno;
3233 			flow->mreg_copy = mcp_res;
3234 			if (dev->data->dev_started) {
3235 				mcp_res->appcnt++;
3236 				flow->copy_applied = 1;
3237 			}
3238 			return 0;
3239 		default:
3240 			break;
3241 		}
3242 	}
3243 	return 0;
3244 }
3245 
3246 #define MLX5_MAX_SPLIT_ACTIONS 24
3247 #define MLX5_MAX_SPLIT_ITEMS 24
3248 
3249 /**
3250  * Split the hairpin flow.
3251  * Since HW can't support encap on Rx we move the encap to Tx.
3252  * If the count action is after the encap then we also
3253  * move the count action. in this case the count will also measure
3254  * the outer bytes.
3255  *
3256  * @param dev
3257  *   Pointer to Ethernet device.
3258  * @param[in] actions
3259  *   Associated actions (list terminated by the END action).
3260  * @param[out] actions_rx
3261  *   Rx flow actions.
3262  * @param[out] actions_tx
3263  *   Tx flow actions..
3264  * @param[out] pattern_tx
3265  *   The pattern items for the Tx flow.
3266  * @param[out] flow_id
3267  *   The flow ID connected to this flow.
3268  *
3269  * @return
3270  *   0 on success.
3271  */
3272 static int
3273 flow_hairpin_split(struct rte_eth_dev *dev,
3274 		   const struct rte_flow_action actions[],
3275 		   struct rte_flow_action actions_rx[],
3276 		   struct rte_flow_action actions_tx[],
3277 		   struct rte_flow_item pattern_tx[],
3278 		   uint32_t *flow_id)
3279 {
3280 	struct mlx5_priv *priv = dev->data->dev_private;
3281 	const struct rte_flow_action_raw_encap *raw_encap;
3282 	const struct rte_flow_action_raw_decap *raw_decap;
3283 	struct mlx5_rte_flow_action_set_tag *set_tag;
3284 	struct rte_flow_action *tag_action;
3285 	struct mlx5_rte_flow_item_tag *tag_item;
3286 	struct rte_flow_item *item;
3287 	char *addr;
3288 	int encap = 0;
3289 
3290 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3291 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3292 		switch (actions->type) {
3293 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3294 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3295 			rte_memcpy(actions_tx, actions,
3296 			       sizeof(struct rte_flow_action));
3297 			actions_tx++;
3298 			break;
3299 		case RTE_FLOW_ACTION_TYPE_COUNT:
3300 			if (encap) {
3301 				rte_memcpy(actions_tx, actions,
3302 					   sizeof(struct rte_flow_action));
3303 				actions_tx++;
3304 			} else {
3305 				rte_memcpy(actions_rx, actions,
3306 					   sizeof(struct rte_flow_action));
3307 				actions_rx++;
3308 			}
3309 			break;
3310 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3311 			raw_encap = actions->conf;
3312 			if (raw_encap->size >
3313 			    (sizeof(struct rte_flow_item_eth) +
3314 			     sizeof(struct rte_flow_item_ipv4))) {
3315 				memcpy(actions_tx, actions,
3316 				       sizeof(struct rte_flow_action));
3317 				actions_tx++;
3318 				encap = 1;
3319 			} else {
3320 				rte_memcpy(actions_rx, actions,
3321 					   sizeof(struct rte_flow_action));
3322 				actions_rx++;
3323 			}
3324 			break;
3325 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3326 			raw_decap = actions->conf;
3327 			if (raw_decap->size <
3328 			    (sizeof(struct rte_flow_item_eth) +
3329 			     sizeof(struct rte_flow_item_ipv4))) {
3330 				memcpy(actions_tx, actions,
3331 				       sizeof(struct rte_flow_action));
3332 				actions_tx++;
3333 			} else {
3334 				rte_memcpy(actions_rx, actions,
3335 					   sizeof(struct rte_flow_action));
3336 				actions_rx++;
3337 			}
3338 			break;
3339 		default:
3340 			rte_memcpy(actions_rx, actions,
3341 				   sizeof(struct rte_flow_action));
3342 			actions_rx++;
3343 			break;
3344 		}
3345 	}
3346 	/* Add set meta action and end action for the Rx flow. */
3347 	tag_action = actions_rx;
3348 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3349 	actions_rx++;
3350 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3351 	actions_rx++;
3352 	set_tag = (void *)actions_rx;
3353 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3354 	assert(set_tag->id > REG_NONE);
3355 	set_tag->data = *flow_id;
3356 	tag_action->conf = set_tag;
3357 	/* Create Tx item list. */
3358 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3359 	addr = (void *)&pattern_tx[2];
3360 	item = pattern_tx;
3361 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3362 	tag_item = (void *)addr;
3363 	tag_item->data = *flow_id;
3364 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3365 	assert(set_tag->id > REG_NONE);
3366 	item->spec = tag_item;
3367 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3368 	tag_item = (void *)addr;
3369 	tag_item->data = UINT32_MAX;
3370 	tag_item->id = UINT16_MAX;
3371 	item->mask = tag_item;
3372 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3373 	item->last = NULL;
3374 	item++;
3375 	item->type = RTE_FLOW_ITEM_TYPE_END;
3376 	return 0;
3377 }
3378 
3379 /**
3380  * The last stage of splitting chain, just creates the subflow
3381  * without any modification.
3382  *
3383  * @param dev
3384  *   Pointer to Ethernet device.
3385  * @param[in] flow
3386  *   Parent flow structure pointer.
3387  * @param[in, out] sub_flow
3388  *   Pointer to return the created subflow, may be NULL.
3389  * @param[in] attr
3390  *   Flow rule attributes.
3391  * @param[in] items
3392  *   Pattern specification (list terminated by the END pattern item).
3393  * @param[in] actions
3394  *   Associated actions (list terminated by the END action).
3395  * @param[in] external
3396  *   This flow rule is created by request external to PMD.
3397  * @param[out] error
3398  *   Perform verbose error reporting if not NULL.
3399  * @return
3400  *   0 on success, negative value otherwise
3401  */
3402 static int
3403 flow_create_split_inner(struct rte_eth_dev *dev,
3404 			struct rte_flow *flow,
3405 			struct mlx5_flow **sub_flow,
3406 			const struct rte_flow_attr *attr,
3407 			const struct rte_flow_item items[],
3408 			const struct rte_flow_action actions[],
3409 			bool external, struct rte_flow_error *error)
3410 {
3411 	struct mlx5_flow *dev_flow;
3412 
3413 	dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3414 	if (!dev_flow)
3415 		return -rte_errno;
3416 	dev_flow->flow = flow;
3417 	dev_flow->external = external;
3418 	/* Subflow object was created, we must include one in the list. */
3419 	LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3420 	if (sub_flow)
3421 		*sub_flow = dev_flow;
3422 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3423 }
3424 
3425 /**
3426  * Split the meter flow.
3427  *
3428  * As meter flow will split to three sub flow, other than meter
3429  * action, the other actions make sense to only meter accepts
3430  * the packet. If it need to be dropped, no other additional
3431  * actions should be take.
3432  *
3433  * One kind of special action which decapsulates the L3 tunnel
3434  * header will be in the prefix sub flow, as not to take the
3435  * L3 tunnel header into account.
3436  *
3437  * @param dev
3438  *   Pointer to Ethernet device.
3439  * @param[in] actions
3440  *   Associated actions (list terminated by the END action).
3441  * @param[out] actions_sfx
3442  *   Suffix flow actions.
3443  * @param[out] actions_pre
3444  *   Prefix flow actions.
3445  * @param[out] pattern_sfx
3446  *   The pattern items for the suffix flow.
3447  * @param[out] tag_sfx
3448  *   Pointer to suffix flow tag.
3449  *
3450  * @return
3451  *   0 on success.
3452  */
3453 static int
3454 flow_meter_split_prep(struct rte_eth_dev *dev,
3455 		 const struct rte_flow_action actions[],
3456 		 struct rte_flow_action actions_sfx[],
3457 		 struct rte_flow_action actions_pre[])
3458 {
3459 	struct rte_flow_action *tag_action;
3460 	struct mlx5_rte_flow_action_set_tag *set_tag;
3461 	struct rte_flow_error error;
3462 	const struct rte_flow_action_raw_encap *raw_encap;
3463 	const struct rte_flow_action_raw_decap *raw_decap;
3464 	uint32_t tag_id;
3465 
3466 	/* Add the extra tag action first. */
3467 	tag_action = actions_pre;
3468 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3469 	actions_pre++;
3470 	/* Prepare the actions for prefix and suffix flow. */
3471 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3472 		switch (actions->type) {
3473 		case RTE_FLOW_ACTION_TYPE_METER:
3474 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3475 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3476 			memcpy(actions_pre, actions,
3477 			       sizeof(struct rte_flow_action));
3478 			actions_pre++;
3479 			break;
3480 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3481 			raw_encap = actions->conf;
3482 			if (raw_encap->size >
3483 			    (sizeof(struct rte_flow_item_eth) +
3484 			     sizeof(struct rte_flow_item_ipv4))) {
3485 				memcpy(actions_sfx, actions,
3486 				       sizeof(struct rte_flow_action));
3487 				actions_sfx++;
3488 			} else {
3489 				rte_memcpy(actions_pre, actions,
3490 					   sizeof(struct rte_flow_action));
3491 				actions_pre++;
3492 			}
3493 			break;
3494 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3495 			raw_decap = actions->conf;
3496 			/* Size 0 decap means 50 bytes as vxlan decap. */
3497 			if (raw_decap->size && (raw_decap->size <
3498 			    (sizeof(struct rte_flow_item_eth) +
3499 			     sizeof(struct rte_flow_item_ipv4)))) {
3500 				memcpy(actions_sfx, actions,
3501 				       sizeof(struct rte_flow_action));
3502 				actions_sfx++;
3503 			} else {
3504 				rte_memcpy(actions_pre, actions,
3505 					   sizeof(struct rte_flow_action));
3506 				actions_pre++;
3507 			}
3508 			break;
3509 		default:
3510 			memcpy(actions_sfx, actions,
3511 				sizeof(struct rte_flow_action));
3512 			actions_sfx++;
3513 			break;
3514 		}
3515 	}
3516 	/* Add end action to the actions. */
3517 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3518 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3519 	actions_pre++;
3520 	/* Set the tag. */
3521 	set_tag = (void *)actions_pre;
3522 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3523 	/*
3524 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3525 	 */
3526 	tag_id = flow_qrss_get_id(dev);
3527 	set_tag->data = rte_cpu_to_be_32(tag_id);
3528 	tag_action->conf = set_tag;
3529 	return tag_id;
3530 }
3531 
3532 /**
3533  * Split action list having QUEUE/RSS for metadata register copy.
3534  *
3535  * Once Q/RSS action is detected in user's action list, the flow action
3536  * should be split in order to copy metadata registers, which will happen in
3537  * RX_CP_TBL like,
3538  *   - CQE->flow_tag := reg_c[1] (MARK)
3539  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3540  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3541  * This is because the last action of each flow must be a terminal action
3542  * (QUEUE, RSS or DROP).
3543  *
3544  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3545  * stored and kept in the mlx5_flow structure per each sub_flow.
3546  *
3547  * The Q/RSS action is replaced with,
3548  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3549  * And the following JUMP action is added at the end,
3550  *   - JUMP, to RX_CP_TBL.
3551  *
3552  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3553  * flow_create_split_metadata() routine. The flow will look like,
3554  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3555  *
3556  * @param dev
3557  *   Pointer to Ethernet device.
3558  * @param[out] split_actions
3559  *   Pointer to store split actions to jump to CP_TBL.
3560  * @param[in] actions
3561  *   Pointer to the list of original flow actions.
3562  * @param[in] qrss
3563  *   Pointer to the Q/RSS action.
3564  * @param[in] actions_n
3565  *   Number of original actions.
3566  * @param[out] error
3567  *   Perform verbose error reporting if not NULL.
3568  *
3569  * @return
3570  *   non-zero unique flow_id on success, otherwise 0 and
3571  *   error/rte_error are set.
3572  */
3573 static uint32_t
3574 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3575 			  struct rte_flow_action *split_actions,
3576 			  const struct rte_flow_action *actions,
3577 			  const struct rte_flow_action *qrss,
3578 			  int actions_n, struct rte_flow_error *error)
3579 {
3580 	struct mlx5_rte_flow_action_set_tag *set_tag;
3581 	struct rte_flow_action_jump *jump;
3582 	const int qrss_idx = qrss - actions;
3583 	uint32_t flow_id = 0;
3584 	int ret = 0;
3585 
3586 	/*
3587 	 * Given actions will be split
3588 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3589 	 * - Add jump to mreg CP_TBL.
3590 	 * As a result, there will be one more action.
3591 	 */
3592 	++actions_n;
3593 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3594 	set_tag = (void *)(split_actions + actions_n);
3595 	/*
3596 	 * If tag action is not set to void(it means we are not the meter
3597 	 * suffix flow), add the tag action. Since meter suffix flow already
3598 	 * has the tag added.
3599 	 */
3600 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3601 		/*
3602 		 * Allocate the new subflow ID. This one is unique within
3603 		 * device and not shared with representors. Otherwise,
3604 		 * we would have to resolve multi-thread access synch
3605 		 * issue. Each flow on the shared device is appended
3606 		 * with source vport identifier, so the resulting
3607 		 * flows will be unique in the shared (by master and
3608 		 * representors) domain even if they have coinciding
3609 		 * IDs.
3610 		 */
3611 		flow_id = flow_qrss_get_id(dev);
3612 		if (!flow_id)
3613 			return rte_flow_error_set(error, ENOMEM,
3614 						  RTE_FLOW_ERROR_TYPE_ACTION,
3615 						  NULL, "can't allocate id "
3616 						  "for split Q/RSS subflow");
3617 		/* Internal SET_TAG action to set flow ID. */
3618 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3619 			.data = flow_id,
3620 		};
3621 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3622 		if (ret < 0)
3623 			return ret;
3624 		set_tag->id = ret;
3625 		/* Construct new actions array. */
3626 		/* Replace QUEUE/RSS action. */
3627 		split_actions[qrss_idx] = (struct rte_flow_action){
3628 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3629 			.conf = set_tag,
3630 		};
3631 	}
3632 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3633 	jump = (void *)(set_tag + 1);
3634 	*jump = (struct rte_flow_action_jump){
3635 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3636 	};
3637 	split_actions[actions_n - 2] = (struct rte_flow_action){
3638 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3639 		.conf = jump,
3640 	};
3641 	split_actions[actions_n - 1] = (struct rte_flow_action){
3642 		.type = RTE_FLOW_ACTION_TYPE_END,
3643 	};
3644 	return flow_id;
3645 }
3646 
3647 /**
3648  * Extend the given action list for Tx metadata copy.
3649  *
3650  * Copy the given action list to the ext_actions and add flow metadata register
3651  * copy action in order to copy reg_a set by WQE to reg_c[0].
3652  *
3653  * @param[out] ext_actions
3654  *   Pointer to the extended action list.
3655  * @param[in] actions
3656  *   Pointer to the list of actions.
3657  * @param[in] actions_n
3658  *   Number of actions in the list.
3659  * @param[out] error
3660  *   Perform verbose error reporting if not NULL.
3661  *
3662  * @return
3663  *   0 on success, negative value otherwise
3664  */
3665 static int
3666 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3667 		       struct rte_flow_action *ext_actions,
3668 		       const struct rte_flow_action *actions,
3669 		       int actions_n, struct rte_flow_error *error)
3670 {
3671 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3672 		(struct mlx5_flow_action_copy_mreg *)
3673 			(ext_actions + actions_n + 1);
3674 	int ret;
3675 
3676 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3677 	if (ret < 0)
3678 		return ret;
3679 	cp_mreg->dst = ret;
3680 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3681 	if (ret < 0)
3682 		return ret;
3683 	cp_mreg->src = ret;
3684 	memcpy(ext_actions, actions,
3685 			sizeof(*ext_actions) * actions_n);
3686 	ext_actions[actions_n - 1] = (struct rte_flow_action){
3687 		.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3688 		.conf = cp_mreg,
3689 	};
3690 	ext_actions[actions_n] = (struct rte_flow_action){
3691 		.type = RTE_FLOW_ACTION_TYPE_END,
3692 	};
3693 	return 0;
3694 }
3695 
3696 /**
3697  * The splitting for metadata feature.
3698  *
3699  * - Q/RSS action on NIC Rx should be split in order to pass by
3700  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3701  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3702  *
3703  * - All the actions on NIC Tx should have a mreg copy action to
3704  *   copy reg_a from WQE to reg_c[0].
3705  *
3706  * @param dev
3707  *   Pointer to Ethernet device.
3708  * @param[in] flow
3709  *   Parent flow structure pointer.
3710  * @param[in] attr
3711  *   Flow rule attributes.
3712  * @param[in] items
3713  *   Pattern specification (list terminated by the END pattern item).
3714  * @param[in] actions
3715  *   Associated actions (list terminated by the END action).
3716  * @param[in] external
3717  *   This flow rule is created by request external to PMD.
3718  * @param[out] error
3719  *   Perform verbose error reporting if not NULL.
3720  * @return
3721  *   0 on success, negative value otherwise
3722  */
3723 static int
3724 flow_create_split_metadata(struct rte_eth_dev *dev,
3725 			   struct rte_flow *flow,
3726 			   const struct rte_flow_attr *attr,
3727 			   const struct rte_flow_item items[],
3728 			   const struct rte_flow_action actions[],
3729 			   bool external, struct rte_flow_error *error)
3730 {
3731 	struct mlx5_priv *priv = dev->data->dev_private;
3732 	struct mlx5_dev_config *config = &priv->config;
3733 	const struct rte_flow_action *qrss = NULL;
3734 	struct rte_flow_action *ext_actions = NULL;
3735 	struct mlx5_flow *dev_flow = NULL;
3736 	uint32_t qrss_id = 0;
3737 	int mtr_sfx = 0;
3738 	size_t act_size;
3739 	int actions_n;
3740 	int ret;
3741 
3742 	/* Check whether extensive metadata feature is engaged. */
3743 	if (!config->dv_flow_en ||
3744 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3745 	    !mlx5_flow_ext_mreg_supported(dev))
3746 		return flow_create_split_inner(dev, flow, NULL, attr, items,
3747 					       actions, external, error);
3748 	actions_n = flow_parse_qrss_action(actions, &qrss);
3749 	if (qrss) {
3750 		/* Exclude hairpin flows from splitting. */
3751 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3752 			const struct rte_flow_action_queue *queue;
3753 
3754 			queue = qrss->conf;
3755 			if (mlx5_rxq_get_type(dev, queue->index) ==
3756 			    MLX5_RXQ_TYPE_HAIRPIN)
3757 				qrss = NULL;
3758 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3759 			const struct rte_flow_action_rss *rss;
3760 
3761 			rss = qrss->conf;
3762 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3763 			    MLX5_RXQ_TYPE_HAIRPIN)
3764 				qrss = NULL;
3765 		}
3766 	}
3767 	if (qrss) {
3768 		/* Check if it is in meter suffix table. */
3769 		mtr_sfx = attr->group == (attr->transfer ?
3770 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3771 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3772 		/*
3773 		 * Q/RSS action on NIC Rx should be split in order to pass by
3774 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3775 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3776 		 */
3777 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3778 			   sizeof(struct rte_flow_action_set_tag) +
3779 			   sizeof(struct rte_flow_action_jump);
3780 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3781 		if (!ext_actions)
3782 			return rte_flow_error_set(error, ENOMEM,
3783 						  RTE_FLOW_ERROR_TYPE_ACTION,
3784 						  NULL, "no memory to split "
3785 						  "metadata flow");
3786 		/*
3787 		 * If we are the suffix flow of meter, tag already exist.
3788 		 * Set the tag action to void.
3789 		 */
3790 		if (mtr_sfx)
3791 			ext_actions[qrss - actions].type =
3792 						RTE_FLOW_ACTION_TYPE_VOID;
3793 		else
3794 			ext_actions[qrss - actions].type =
3795 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3796 		/*
3797 		 * Create the new actions list with removed Q/RSS action
3798 		 * and appended set tag and jump to register copy table
3799 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3800 		 * in advance, because it is needed for set tag action.
3801 		 */
3802 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3803 						    qrss, actions_n, error);
3804 		if (!mtr_sfx && !qrss_id) {
3805 			ret = -rte_errno;
3806 			goto exit;
3807 		}
3808 	} else if (attr->egress && !attr->transfer) {
3809 		/*
3810 		 * All the actions on NIC Tx should have a metadata register
3811 		 * copy action to copy reg_a from WQE to reg_c[meta]
3812 		 */
3813 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3814 			   sizeof(struct mlx5_flow_action_copy_mreg);
3815 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3816 		if (!ext_actions)
3817 			return rte_flow_error_set(error, ENOMEM,
3818 						  RTE_FLOW_ERROR_TYPE_ACTION,
3819 						  NULL, "no memory to split "
3820 						  "metadata flow");
3821 		/* Create the action list appended with copy register. */
3822 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3823 					     actions_n, error);
3824 		if (ret < 0)
3825 			goto exit;
3826 	}
3827 	/* Add the unmodified original or prefix subflow. */
3828 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3829 				      ext_actions ? ext_actions : actions,
3830 				      external, error);
3831 	if (ret < 0)
3832 		goto exit;
3833 	assert(dev_flow);
3834 	if (qrss) {
3835 		const struct rte_flow_attr q_attr = {
3836 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3837 			.ingress = 1,
3838 		};
3839 		/* Internal PMD action to set register. */
3840 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3841 			.data = qrss_id,
3842 			.id = 0,
3843 		};
3844 		struct rte_flow_item q_items[] = {
3845 			{
3846 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3847 				.spec = &q_tag_spec,
3848 				.last = NULL,
3849 				.mask = NULL,
3850 			},
3851 			{
3852 				.type = RTE_FLOW_ITEM_TYPE_END,
3853 			},
3854 		};
3855 		struct rte_flow_action q_actions[] = {
3856 			{
3857 				.type = qrss->type,
3858 				.conf = qrss->conf,
3859 			},
3860 			{
3861 				.type = RTE_FLOW_ACTION_TYPE_END,
3862 			},
3863 		};
3864 		uint64_t hash_fields = dev_flow->hash_fields;
3865 
3866 		/*
3867 		 * Configure the tag item only if there is no meter subflow.
3868 		 * Since tag is already marked in the meter suffix subflow
3869 		 * we can just use the meter suffix items as is.
3870 		 */
3871 		if (qrss_id) {
3872 			/* Not meter subflow. */
3873 			assert(!mtr_sfx);
3874 			/*
3875 			 * Put unique id in prefix flow due to it is destroyed
3876 			 * after suffix flow and id will be freed after there
3877 			 * is no actual flows with this id and identifier
3878 			 * reallocation becomes possible (for example, for
3879 			 * other flows in other threads).
3880 			 */
3881 			dev_flow->qrss_id = qrss_id;
3882 			qrss_id = 0;
3883 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3884 						   error);
3885 			if (ret < 0)
3886 				goto exit;
3887 			q_tag_spec.id = ret;
3888 		}
3889 		dev_flow = NULL;
3890 		/* Add suffix subflow to execute Q/RSS. */
3891 		ret = flow_create_split_inner(dev, flow, &dev_flow,
3892 					      &q_attr, mtr_sfx ? items :
3893 					      q_items, q_actions,
3894 					      external, error);
3895 		if (ret < 0)
3896 			goto exit;
3897 		assert(dev_flow);
3898 		dev_flow->hash_fields = hash_fields;
3899 	}
3900 
3901 exit:
3902 	/*
3903 	 * We do not destroy the partially created sub_flows in case of error.
3904 	 * These ones are included into parent flow list and will be destroyed
3905 	 * by flow_drv_destroy.
3906 	 */
3907 	flow_qrss_free_id(dev, qrss_id);
3908 	rte_free(ext_actions);
3909 	return ret;
3910 }
3911 
3912 /**
3913  * The splitting for meter feature.
3914  *
3915  * - The meter flow will be split to two flows as prefix and
3916  *   suffix flow. The packets make sense only it pass the prefix
3917  *   meter action.
3918  *
3919  * - Reg_C_5 is used for the packet to match betweend prefix and
3920  *   suffix flow.
3921  *
3922  * @param dev
3923  *   Pointer to Ethernet device.
3924  * @param[in] flow
3925  *   Parent flow structure pointer.
3926  * @param[in] attr
3927  *   Flow rule attributes.
3928  * @param[in] items
3929  *   Pattern specification (list terminated by the END pattern item).
3930  * @param[in] actions
3931  *   Associated actions (list terminated by the END action).
3932  * @param[in] external
3933  *   This flow rule is created by request external to PMD.
3934  * @param[out] error
3935  *   Perform verbose error reporting if not NULL.
3936  * @return
3937  *   0 on success, negative value otherwise
3938  */
3939 static int
3940 flow_create_split_meter(struct rte_eth_dev *dev,
3941 			   struct rte_flow *flow,
3942 			   const struct rte_flow_attr *attr,
3943 			   const struct rte_flow_item items[],
3944 			   const struct rte_flow_action actions[],
3945 			   bool external, struct rte_flow_error *error)
3946 {
3947 	struct mlx5_priv *priv = dev->data->dev_private;
3948 	struct rte_flow_action *sfx_actions = NULL;
3949 	struct rte_flow_action *pre_actions = NULL;
3950 	struct rte_flow_item *sfx_items = NULL;
3951 	const  struct rte_flow_item *sfx_port_id_item;
3952 	struct mlx5_flow *dev_flow = NULL;
3953 	struct rte_flow_attr sfx_attr = *attr;
3954 	uint32_t mtr = 0;
3955 	uint32_t mtr_tag_id = 0;
3956 	size_t act_size;
3957 	size_t item_size;
3958 	int actions_n = 0;
3959 	int ret;
3960 
3961 	if (priv->mtr_en)
3962 		actions_n = flow_check_meter_action(actions, &mtr);
3963 	if (mtr) {
3964 		struct mlx5_rte_flow_item_tag *tag_spec;
3965 		/* The five prefix actions: meter, decap, encap, tag, end. */
3966 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3967 			   sizeof(struct rte_flow_action_set_tag);
3968 		/* tag, end. */
3969 #define METER_SUFFIX_ITEM 3
3970 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
3971 			    sizeof(struct mlx5_rte_flow_item_tag);
3972 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
3973 		if (!sfx_actions)
3974 			return rte_flow_error_set(error, ENOMEM,
3975 						  RTE_FLOW_ERROR_TYPE_ACTION,
3976 						  NULL, "no memory to split "
3977 						  "meter flow");
3978 		pre_actions = sfx_actions + actions_n;
3979 		mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
3980 						     pre_actions);
3981 		if (!mtr_tag_id) {
3982 			ret = -rte_errno;
3983 			goto exit;
3984 		}
3985 		/* Add the prefix subflow. */
3986 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3987 						  pre_actions, external, error);
3988 		if (ret) {
3989 			ret = -rte_errno;
3990 			goto exit;
3991 		}
3992 		dev_flow->mtr_flow_id = mtr_tag_id;
3993 		/* Prepare the suffix flow match pattern. */
3994 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
3995 			     act_size);
3996 		tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
3997 			    METER_SUFFIX_ITEM);
3998 		tag_spec->data = rte_cpu_to_be_32(dev_flow->mtr_flow_id);
3999 		tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
4000 						    error);
4001 		sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4002 		sfx_items->spec = tag_spec;
4003 		sfx_items->last = NULL;
4004 		sfx_items->mask = NULL;
4005 		sfx_items++;
4006 		sfx_port_id_item = find_port_id_item(items);
4007 		if (sfx_port_id_item) {
4008 			memcpy(sfx_items, sfx_port_id_item,
4009 			       sizeof(*sfx_items));
4010 			sfx_items++;
4011 		}
4012 		sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4013 		sfx_items -= METER_SUFFIX_ITEM;
4014 		/* Setting the sfx group atrr. */
4015 		sfx_attr.group = sfx_attr.transfer ?
4016 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4017 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4018 	}
4019 	/* Add the prefix subflow. */
4020 	ret = flow_create_split_metadata(dev, flow, &sfx_attr,
4021 					 sfx_items ? sfx_items : items,
4022 					 sfx_actions ? sfx_actions : actions,
4023 					 external, error);
4024 exit:
4025 	if (sfx_actions)
4026 		rte_free(sfx_actions);
4027 	return ret;
4028 }
4029 
4030 /**
4031  * Split the flow to subflow set. The splitters might be linked
4032  * in the chain, like this:
4033  * flow_create_split_outer() calls:
4034  *   flow_create_split_meter() calls:
4035  *     flow_create_split_metadata(meter_subflow_0) calls:
4036  *       flow_create_split_inner(metadata_subflow_0)
4037  *       flow_create_split_inner(metadata_subflow_1)
4038  *       flow_create_split_inner(metadata_subflow_2)
4039  *     flow_create_split_metadata(meter_subflow_1) calls:
4040  *       flow_create_split_inner(metadata_subflow_0)
4041  *       flow_create_split_inner(metadata_subflow_1)
4042  *       flow_create_split_inner(metadata_subflow_2)
4043  *
4044  * This provide flexible way to add new levels of flow splitting.
4045  * The all of successfully created subflows are included to the
4046  * parent flow dev_flow list.
4047  *
4048  * @param dev
4049  *   Pointer to Ethernet device.
4050  * @param[in] flow
4051  *   Parent flow structure pointer.
4052  * @param[in] attr
4053  *   Flow rule attributes.
4054  * @param[in] items
4055  *   Pattern specification (list terminated by the END pattern item).
4056  * @param[in] actions
4057  *   Associated actions (list terminated by the END action).
4058  * @param[in] external
4059  *   This flow rule is created by request external to PMD.
4060  * @param[out] error
4061  *   Perform verbose error reporting if not NULL.
4062  * @return
4063  *   0 on success, negative value otherwise
4064  */
4065 static int
4066 flow_create_split_outer(struct rte_eth_dev *dev,
4067 			struct rte_flow *flow,
4068 			const struct rte_flow_attr *attr,
4069 			const struct rte_flow_item items[],
4070 			const struct rte_flow_action actions[],
4071 			bool external, struct rte_flow_error *error)
4072 {
4073 	int ret;
4074 
4075 	ret = flow_create_split_meter(dev, flow, attr, items,
4076 					 actions, external, error);
4077 	assert(ret <= 0);
4078 	return ret;
4079 }
4080 
4081 /**
4082  * Create a flow and add it to @p list.
4083  *
4084  * @param dev
4085  *   Pointer to Ethernet device.
4086  * @param list
4087  *   Pointer to a TAILQ flow list. If this parameter NULL,
4088  *   no list insertion occurred, flow is just created,
4089  *   this is caller's responsibility to track the
4090  *   created flow.
4091  * @param[in] attr
4092  *   Flow rule attributes.
4093  * @param[in] items
4094  *   Pattern specification (list terminated by the END pattern item).
4095  * @param[in] actions
4096  *   Associated actions (list terminated by the END action).
4097  * @param[in] external
4098  *   This flow rule is created by request external to PMD.
4099  * @param[out] error
4100  *   Perform verbose error reporting if not NULL.
4101  *
4102  * @return
4103  *   A flow on success, NULL otherwise and rte_errno is set.
4104  */
4105 static struct rte_flow *
4106 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4107 		 const struct rte_flow_attr *attr,
4108 		 const struct rte_flow_item items[],
4109 		 const struct rte_flow_action actions[],
4110 		 bool external, struct rte_flow_error *error)
4111 {
4112 	struct mlx5_priv *priv = dev->data->dev_private;
4113 	struct rte_flow *flow = NULL;
4114 	struct mlx5_flow *dev_flow;
4115 	const struct rte_flow_action_rss *rss;
4116 	union {
4117 		struct rte_flow_expand_rss buf;
4118 		uint8_t buffer[2048];
4119 	} expand_buffer;
4120 	union {
4121 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4122 		uint8_t buffer[2048];
4123 	} actions_rx;
4124 	union {
4125 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4126 		uint8_t buffer[2048];
4127 	} actions_hairpin_tx;
4128 	union {
4129 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4130 		uint8_t buffer[2048];
4131 	} items_tx;
4132 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4133 	const struct rte_flow_action *p_actions_rx = actions;
4134 	int ret;
4135 	uint32_t i;
4136 	uint32_t flow_size;
4137 	int hairpin_flow = 0;
4138 	uint32_t hairpin_id = 0;
4139 	struct rte_flow_attr attr_tx = { .priority = 0 };
4140 
4141 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4142 	if (hairpin_flow > 0) {
4143 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4144 			rte_errno = EINVAL;
4145 			return NULL;
4146 		}
4147 		flow_hairpin_split(dev, actions, actions_rx.actions,
4148 				   actions_hairpin_tx.actions, items_tx.items,
4149 				   &hairpin_id);
4150 		p_actions_rx = actions_rx.actions;
4151 	}
4152 	ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4153 				error);
4154 	if (ret < 0)
4155 		goto error_before_flow;
4156 	flow_size = sizeof(struct rte_flow);
4157 	rss = flow_get_rss_action(p_actions_rx);
4158 	if (rss)
4159 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4160 					    sizeof(void *));
4161 	else
4162 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4163 	flow = rte_calloc(__func__, 1, flow_size, 0);
4164 	if (!flow) {
4165 		rte_errno = ENOMEM;
4166 		goto error_before_flow;
4167 	}
4168 	flow->drv_type = flow_get_drv_type(dev, attr);
4169 	if (hairpin_id != 0)
4170 		flow->hairpin_flow_id = hairpin_id;
4171 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4172 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
4173 	flow->rss.queue = (void *)(flow + 1);
4174 	if (rss) {
4175 		/*
4176 		 * The following information is required by
4177 		 * mlx5_flow_hashfields_adjust() in advance.
4178 		 */
4179 		flow->rss.level = rss->level;
4180 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4181 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4182 	}
4183 	LIST_INIT(&flow->dev_flows);
4184 	if (rss && rss->types) {
4185 		unsigned int graph_root;
4186 
4187 		graph_root = find_graph_root(items, rss->level);
4188 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4189 					  items, rss->types,
4190 					  mlx5_support_expansion,
4191 					  graph_root);
4192 		assert(ret > 0 &&
4193 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4194 	} else {
4195 		buf->entries = 1;
4196 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4197 	}
4198 	for (i = 0; i < buf->entries; ++i) {
4199 		/*
4200 		 * The splitter may create multiple dev_flows,
4201 		 * depending on configuration. In the simplest
4202 		 * case it just creates unmodified original flow.
4203 		 */
4204 		ret = flow_create_split_outer(dev, flow, attr,
4205 					      buf->entry[i].pattern,
4206 					      p_actions_rx, external,
4207 					      error);
4208 		if (ret < 0)
4209 			goto error;
4210 	}
4211 	/* Create the tx flow. */
4212 	if (hairpin_flow) {
4213 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4214 		attr_tx.ingress = 0;
4215 		attr_tx.egress = 1;
4216 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4217 					    actions_hairpin_tx.actions, error);
4218 		if (!dev_flow)
4219 			goto error;
4220 		dev_flow->flow = flow;
4221 		dev_flow->external = 0;
4222 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4223 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4224 					 items_tx.items,
4225 					 actions_hairpin_tx.actions, error);
4226 		if (ret < 0)
4227 			goto error;
4228 	}
4229 	/*
4230 	 * Update the metadata register copy table. If extensive
4231 	 * metadata feature is enabled and registers are supported
4232 	 * we might create the extra rte_flow for each unique
4233 	 * MARK/FLAG action ID.
4234 	 *
4235 	 * The table is updated for ingress Flows only, because
4236 	 * the egress Flows belong to the different device and
4237 	 * copy table should be updated in peer NIC Rx domain.
4238 	 */
4239 	if (attr->ingress &&
4240 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4241 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4242 		if (ret)
4243 			goto error;
4244 	}
4245 	if (dev->data->dev_started) {
4246 		ret = flow_drv_apply(dev, flow, error);
4247 		if (ret < 0)
4248 			goto error;
4249 	}
4250 	if (list)
4251 		TAILQ_INSERT_TAIL(list, flow, next);
4252 	flow_rxq_flags_set(dev, flow);
4253 	return flow;
4254 error_before_flow:
4255 	if (hairpin_id)
4256 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4257 				     hairpin_id);
4258 	return NULL;
4259 error:
4260 	assert(flow);
4261 	flow_mreg_del_copy_action(dev, flow);
4262 	ret = rte_errno; /* Save rte_errno before cleanup. */
4263 	if (flow->hairpin_flow_id)
4264 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4265 				     flow->hairpin_flow_id);
4266 	assert(flow);
4267 	flow_drv_destroy(dev, flow);
4268 	rte_free(flow);
4269 	rte_errno = ret; /* Restore rte_errno. */
4270 	return NULL;
4271 }
4272 
4273 /**
4274  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4275  * incoming packets to table 1.
4276  *
4277  * Other flow rules, requested for group n, will be created in
4278  * e-switch table n+1.
4279  * Jump action to e-switch group n will be created to group n+1.
4280  *
4281  * Used when working in switchdev mode, to utilise advantages of table 1
4282  * and above.
4283  *
4284  * @param dev
4285  *   Pointer to Ethernet device.
4286  *
4287  * @return
4288  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4289  */
4290 struct rte_flow *
4291 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4292 {
4293 	const struct rte_flow_attr attr = {
4294 		.group = 0,
4295 		.priority = 0,
4296 		.ingress = 1,
4297 		.egress = 0,
4298 		.transfer = 1,
4299 	};
4300 	const struct rte_flow_item pattern = {
4301 		.type = RTE_FLOW_ITEM_TYPE_END,
4302 	};
4303 	struct rte_flow_action_jump jump = {
4304 		.group = 1,
4305 	};
4306 	const struct rte_flow_action actions[] = {
4307 		{
4308 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4309 			.conf = &jump,
4310 		},
4311 		{
4312 			.type = RTE_FLOW_ACTION_TYPE_END,
4313 		},
4314 	};
4315 	struct mlx5_priv *priv = dev->data->dev_private;
4316 	struct rte_flow_error error;
4317 
4318 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4319 				actions, false, &error);
4320 }
4321 
4322 /**
4323  * Create a flow.
4324  *
4325  * @see rte_flow_create()
4326  * @see rte_flow_ops
4327  */
4328 struct rte_flow *
4329 mlx5_flow_create(struct rte_eth_dev *dev,
4330 		 const struct rte_flow_attr *attr,
4331 		 const struct rte_flow_item items[],
4332 		 const struct rte_flow_action actions[],
4333 		 struct rte_flow_error *error)
4334 {
4335 	struct mlx5_priv *priv = dev->data->dev_private;
4336 
4337 	return flow_list_create(dev, &priv->flows,
4338 				attr, items, actions, true, error);
4339 }
4340 
4341 /**
4342  * Destroy a flow in a list.
4343  *
4344  * @param dev
4345  *   Pointer to Ethernet device.
4346  * @param list
4347  *   Pointer to a TAILQ flow list. If this parameter NULL,
4348  *   there is no flow removal from the list.
4349  * @param[in] flow
4350  *   Flow to destroy.
4351  */
4352 static void
4353 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4354 		  struct rte_flow *flow)
4355 {
4356 	struct mlx5_priv *priv = dev->data->dev_private;
4357 
4358 	/*
4359 	 * Update RX queue flags only if port is started, otherwise it is
4360 	 * already clean.
4361 	 */
4362 	if (dev->data->dev_started)
4363 		flow_rxq_flags_trim(dev, flow);
4364 	if (flow->hairpin_flow_id)
4365 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4366 				     flow->hairpin_flow_id);
4367 	flow_drv_destroy(dev, flow);
4368 	if (list)
4369 		TAILQ_REMOVE(list, flow, next);
4370 	flow_mreg_del_copy_action(dev, flow);
4371 	rte_free(flow->fdir);
4372 	rte_free(flow);
4373 }
4374 
4375 /**
4376  * Destroy all flows.
4377  *
4378  * @param dev
4379  *   Pointer to Ethernet device.
4380  * @param list
4381  *   Pointer to a TAILQ flow list.
4382  */
4383 void
4384 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4385 {
4386 	while (!TAILQ_EMPTY(list)) {
4387 		struct rte_flow *flow;
4388 
4389 		flow = TAILQ_FIRST(list);
4390 		flow_list_destroy(dev, list, flow);
4391 	}
4392 }
4393 
4394 /**
4395  * Remove all flows.
4396  *
4397  * @param dev
4398  *   Pointer to Ethernet device.
4399  * @param list
4400  *   Pointer to a TAILQ flow list.
4401  */
4402 void
4403 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4404 {
4405 	struct rte_flow *flow;
4406 
4407 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4408 		flow_drv_remove(dev, flow);
4409 		flow_mreg_stop_copy_action(dev, flow);
4410 	}
4411 	flow_mreg_del_default_copy_action(dev);
4412 	flow_rxq_flags_clear(dev);
4413 }
4414 
4415 /**
4416  * Add all flows.
4417  *
4418  * @param dev
4419  *   Pointer to Ethernet device.
4420  * @param list
4421  *   Pointer to a TAILQ flow list.
4422  *
4423  * @return
4424  *   0 on success, a negative errno value otherwise and rte_errno is set.
4425  */
4426 int
4427 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4428 {
4429 	struct rte_flow *flow;
4430 	struct rte_flow_error error;
4431 	int ret = 0;
4432 
4433 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4434 	ret = flow_mreg_add_default_copy_action(dev, &error);
4435 	if (ret < 0)
4436 		return -rte_errno;
4437 	/* Apply Flows created by application. */
4438 	TAILQ_FOREACH(flow, list, next) {
4439 		ret = flow_mreg_start_copy_action(dev, flow);
4440 		if (ret < 0)
4441 			goto error;
4442 		ret = flow_drv_apply(dev, flow, &error);
4443 		if (ret < 0)
4444 			goto error;
4445 		flow_rxq_flags_set(dev, flow);
4446 	}
4447 	return 0;
4448 error:
4449 	ret = rte_errno; /* Save rte_errno before cleanup. */
4450 	mlx5_flow_stop(dev, list);
4451 	rte_errno = ret; /* Restore rte_errno. */
4452 	return -rte_errno;
4453 }
4454 
4455 /**
4456  * Verify the flow list is empty
4457  *
4458  * @param dev
4459  *  Pointer to Ethernet device.
4460  *
4461  * @return the number of flows not released.
4462  */
4463 int
4464 mlx5_flow_verify(struct rte_eth_dev *dev)
4465 {
4466 	struct mlx5_priv *priv = dev->data->dev_private;
4467 	struct rte_flow *flow;
4468 	int ret = 0;
4469 
4470 	TAILQ_FOREACH(flow, &priv->flows, next) {
4471 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4472 			dev->data->port_id, (void *)flow);
4473 		++ret;
4474 	}
4475 	return ret;
4476 }
4477 
4478 /**
4479  * Enable default hairpin egress flow.
4480  *
4481  * @param dev
4482  *   Pointer to Ethernet device.
4483  * @param queue
4484  *   The queue index.
4485  *
4486  * @return
4487  *   0 on success, a negative errno value otherwise and rte_errno is set.
4488  */
4489 int
4490 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4491 			    uint32_t queue)
4492 {
4493 	struct mlx5_priv *priv = dev->data->dev_private;
4494 	const struct rte_flow_attr attr = {
4495 		.egress = 1,
4496 		.priority = 0,
4497 	};
4498 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4499 		.queue = queue,
4500 	};
4501 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4502 		.queue = UINT32_MAX,
4503 	};
4504 	struct rte_flow_item items[] = {
4505 		{
4506 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4507 			.spec = &queue_spec,
4508 			.last = NULL,
4509 			.mask = &queue_mask,
4510 		},
4511 		{
4512 			.type = RTE_FLOW_ITEM_TYPE_END,
4513 		},
4514 	};
4515 	struct rte_flow_action_jump jump = {
4516 		.group = MLX5_HAIRPIN_TX_TABLE,
4517 	};
4518 	struct rte_flow_action actions[2];
4519 	struct rte_flow *flow;
4520 	struct rte_flow_error error;
4521 
4522 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4523 	actions[0].conf = &jump;
4524 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4525 	flow = flow_list_create(dev, &priv->ctrl_flows,
4526 				&attr, items, actions, false, &error);
4527 	if (!flow) {
4528 		DRV_LOG(DEBUG,
4529 			"Failed to create ctrl flow: rte_errno(%d),"
4530 			" type(%d), message(%s)",
4531 			rte_errno, error.type,
4532 			error.message ? error.message : " (no stated reason)");
4533 		return -rte_errno;
4534 	}
4535 	return 0;
4536 }
4537 
4538 /**
4539  * Enable a control flow configured from the control plane.
4540  *
4541  * @param dev
4542  *   Pointer to Ethernet device.
4543  * @param eth_spec
4544  *   An Ethernet flow spec to apply.
4545  * @param eth_mask
4546  *   An Ethernet flow mask to apply.
4547  * @param vlan_spec
4548  *   A VLAN flow spec to apply.
4549  * @param vlan_mask
4550  *   A VLAN flow mask to apply.
4551  *
4552  * @return
4553  *   0 on success, a negative errno value otherwise and rte_errno is set.
4554  */
4555 int
4556 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4557 		    struct rte_flow_item_eth *eth_spec,
4558 		    struct rte_flow_item_eth *eth_mask,
4559 		    struct rte_flow_item_vlan *vlan_spec,
4560 		    struct rte_flow_item_vlan *vlan_mask)
4561 {
4562 	struct mlx5_priv *priv = dev->data->dev_private;
4563 	const struct rte_flow_attr attr = {
4564 		.ingress = 1,
4565 		.priority = MLX5_FLOW_PRIO_RSVD,
4566 	};
4567 	struct rte_flow_item items[] = {
4568 		{
4569 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4570 			.spec = eth_spec,
4571 			.last = NULL,
4572 			.mask = eth_mask,
4573 		},
4574 		{
4575 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4576 					      RTE_FLOW_ITEM_TYPE_END,
4577 			.spec = vlan_spec,
4578 			.last = NULL,
4579 			.mask = vlan_mask,
4580 		},
4581 		{
4582 			.type = RTE_FLOW_ITEM_TYPE_END,
4583 		},
4584 	};
4585 	uint16_t queue[priv->reta_idx_n];
4586 	struct rte_flow_action_rss action_rss = {
4587 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4588 		.level = 0,
4589 		.types = priv->rss_conf.rss_hf,
4590 		.key_len = priv->rss_conf.rss_key_len,
4591 		.queue_num = priv->reta_idx_n,
4592 		.key = priv->rss_conf.rss_key,
4593 		.queue = queue,
4594 	};
4595 	struct rte_flow_action actions[] = {
4596 		{
4597 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4598 			.conf = &action_rss,
4599 		},
4600 		{
4601 			.type = RTE_FLOW_ACTION_TYPE_END,
4602 		},
4603 	};
4604 	struct rte_flow *flow;
4605 	struct rte_flow_error error;
4606 	unsigned int i;
4607 
4608 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4609 		return 0;
4610 	}
4611 	for (i = 0; i != priv->reta_idx_n; ++i)
4612 		queue[i] = (*priv->reta_idx)[i];
4613 	flow = flow_list_create(dev, &priv->ctrl_flows,
4614 				&attr, items, actions, false, &error);
4615 	if (!flow)
4616 		return -rte_errno;
4617 	return 0;
4618 }
4619 
4620 /**
4621  * Enable a flow control configured from the control plane.
4622  *
4623  * @param dev
4624  *   Pointer to Ethernet device.
4625  * @param eth_spec
4626  *   An Ethernet flow spec to apply.
4627  * @param eth_mask
4628  *   An Ethernet flow mask to apply.
4629  *
4630  * @return
4631  *   0 on success, a negative errno value otherwise and rte_errno is set.
4632  */
4633 int
4634 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4635 	       struct rte_flow_item_eth *eth_spec,
4636 	       struct rte_flow_item_eth *eth_mask)
4637 {
4638 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4639 }
4640 
4641 /**
4642  * Destroy a flow.
4643  *
4644  * @see rte_flow_destroy()
4645  * @see rte_flow_ops
4646  */
4647 int
4648 mlx5_flow_destroy(struct rte_eth_dev *dev,
4649 		  struct rte_flow *flow,
4650 		  struct rte_flow_error *error __rte_unused)
4651 {
4652 	struct mlx5_priv *priv = dev->data->dev_private;
4653 
4654 	flow_list_destroy(dev, &priv->flows, flow);
4655 	return 0;
4656 }
4657 
4658 /**
4659  * Destroy all flows.
4660  *
4661  * @see rte_flow_flush()
4662  * @see rte_flow_ops
4663  */
4664 int
4665 mlx5_flow_flush(struct rte_eth_dev *dev,
4666 		struct rte_flow_error *error __rte_unused)
4667 {
4668 	struct mlx5_priv *priv = dev->data->dev_private;
4669 
4670 	mlx5_flow_list_flush(dev, &priv->flows);
4671 	return 0;
4672 }
4673 
4674 /**
4675  * Isolated mode.
4676  *
4677  * @see rte_flow_isolate()
4678  * @see rte_flow_ops
4679  */
4680 int
4681 mlx5_flow_isolate(struct rte_eth_dev *dev,
4682 		  int enable,
4683 		  struct rte_flow_error *error)
4684 {
4685 	struct mlx5_priv *priv = dev->data->dev_private;
4686 
4687 	if (dev->data->dev_started) {
4688 		rte_flow_error_set(error, EBUSY,
4689 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4690 				   NULL,
4691 				   "port must be stopped first");
4692 		return -rte_errno;
4693 	}
4694 	priv->isolated = !!enable;
4695 	if (enable)
4696 		dev->dev_ops = &mlx5_dev_ops_isolate;
4697 	else
4698 		dev->dev_ops = &mlx5_dev_ops;
4699 	return 0;
4700 }
4701 
4702 /**
4703  * Query a flow.
4704  *
4705  * @see rte_flow_query()
4706  * @see rte_flow_ops
4707  */
4708 static int
4709 flow_drv_query(struct rte_eth_dev *dev,
4710 	       struct rte_flow *flow,
4711 	       const struct rte_flow_action *actions,
4712 	       void *data,
4713 	       struct rte_flow_error *error)
4714 {
4715 	const struct mlx5_flow_driver_ops *fops;
4716 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4717 
4718 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4719 	fops = flow_get_drv_ops(ftype);
4720 
4721 	return fops->query(dev, flow, actions, data, error);
4722 }
4723 
4724 /**
4725  * Query a flow.
4726  *
4727  * @see rte_flow_query()
4728  * @see rte_flow_ops
4729  */
4730 int
4731 mlx5_flow_query(struct rte_eth_dev *dev,
4732 		struct rte_flow *flow,
4733 		const struct rte_flow_action *actions,
4734 		void *data,
4735 		struct rte_flow_error *error)
4736 {
4737 	int ret;
4738 
4739 	ret = flow_drv_query(dev, flow, actions, data, error);
4740 	if (ret < 0)
4741 		return ret;
4742 	return 0;
4743 }
4744 
4745 /**
4746  * Convert a flow director filter to a generic flow.
4747  *
4748  * @param dev
4749  *   Pointer to Ethernet device.
4750  * @param fdir_filter
4751  *   Flow director filter to add.
4752  * @param attributes
4753  *   Generic flow parameters structure.
4754  *
4755  * @return
4756  *   0 on success, a negative errno value otherwise and rte_errno is set.
4757  */
4758 static int
4759 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4760 			 const struct rte_eth_fdir_filter *fdir_filter,
4761 			 struct mlx5_fdir *attributes)
4762 {
4763 	struct mlx5_priv *priv = dev->data->dev_private;
4764 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4765 	const struct rte_eth_fdir_masks *mask =
4766 		&dev->data->dev_conf.fdir_conf.mask;
4767 
4768 	/* Validate queue number. */
4769 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4770 		DRV_LOG(ERR, "port %u invalid queue number %d",
4771 			dev->data->port_id, fdir_filter->action.rx_queue);
4772 		rte_errno = EINVAL;
4773 		return -rte_errno;
4774 	}
4775 	attributes->attr.ingress = 1;
4776 	attributes->items[0] = (struct rte_flow_item) {
4777 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4778 		.spec = &attributes->l2,
4779 		.mask = &attributes->l2_mask,
4780 	};
4781 	switch (fdir_filter->action.behavior) {
4782 	case RTE_ETH_FDIR_ACCEPT:
4783 		attributes->actions[0] = (struct rte_flow_action){
4784 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4785 			.conf = &attributes->queue,
4786 		};
4787 		break;
4788 	case RTE_ETH_FDIR_REJECT:
4789 		attributes->actions[0] = (struct rte_flow_action){
4790 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4791 		};
4792 		break;
4793 	default:
4794 		DRV_LOG(ERR, "port %u invalid behavior %d",
4795 			dev->data->port_id,
4796 			fdir_filter->action.behavior);
4797 		rte_errno = ENOTSUP;
4798 		return -rte_errno;
4799 	}
4800 	attributes->queue.index = fdir_filter->action.rx_queue;
4801 	/* Handle L3. */
4802 	switch (fdir_filter->input.flow_type) {
4803 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4804 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4805 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4806 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4807 			.src_addr = input->flow.ip4_flow.src_ip,
4808 			.dst_addr = input->flow.ip4_flow.dst_ip,
4809 			.time_to_live = input->flow.ip4_flow.ttl,
4810 			.type_of_service = input->flow.ip4_flow.tos,
4811 		};
4812 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4813 			.src_addr = mask->ipv4_mask.src_ip,
4814 			.dst_addr = mask->ipv4_mask.dst_ip,
4815 			.time_to_live = mask->ipv4_mask.ttl,
4816 			.type_of_service = mask->ipv4_mask.tos,
4817 			.next_proto_id = mask->ipv4_mask.proto,
4818 		};
4819 		attributes->items[1] = (struct rte_flow_item){
4820 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
4821 			.spec = &attributes->l3,
4822 			.mask = &attributes->l3_mask,
4823 		};
4824 		break;
4825 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4826 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4827 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4828 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4829 			.hop_limits = input->flow.ipv6_flow.hop_limits,
4830 			.proto = input->flow.ipv6_flow.proto,
4831 		};
4832 
4833 		memcpy(attributes->l3.ipv6.hdr.src_addr,
4834 		       input->flow.ipv6_flow.src_ip,
4835 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4836 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
4837 		       input->flow.ipv6_flow.dst_ip,
4838 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4839 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4840 		       mask->ipv6_mask.src_ip,
4841 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4842 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4843 		       mask->ipv6_mask.dst_ip,
4844 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4845 		attributes->items[1] = (struct rte_flow_item){
4846 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
4847 			.spec = &attributes->l3,
4848 			.mask = &attributes->l3_mask,
4849 		};
4850 		break;
4851 	default:
4852 		DRV_LOG(ERR, "port %u invalid flow type%d",
4853 			dev->data->port_id, fdir_filter->input.flow_type);
4854 		rte_errno = ENOTSUP;
4855 		return -rte_errno;
4856 	}
4857 	/* Handle L4. */
4858 	switch (fdir_filter->input.flow_type) {
4859 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4860 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4861 			.src_port = input->flow.udp4_flow.src_port,
4862 			.dst_port = input->flow.udp4_flow.dst_port,
4863 		};
4864 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4865 			.src_port = mask->src_port_mask,
4866 			.dst_port = mask->dst_port_mask,
4867 		};
4868 		attributes->items[2] = (struct rte_flow_item){
4869 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4870 			.spec = &attributes->l4,
4871 			.mask = &attributes->l4_mask,
4872 		};
4873 		break;
4874 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4875 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4876 			.src_port = input->flow.tcp4_flow.src_port,
4877 			.dst_port = input->flow.tcp4_flow.dst_port,
4878 		};
4879 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4880 			.src_port = mask->src_port_mask,
4881 			.dst_port = mask->dst_port_mask,
4882 		};
4883 		attributes->items[2] = (struct rte_flow_item){
4884 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4885 			.spec = &attributes->l4,
4886 			.mask = &attributes->l4_mask,
4887 		};
4888 		break;
4889 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4890 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4891 			.src_port = input->flow.udp6_flow.src_port,
4892 			.dst_port = input->flow.udp6_flow.dst_port,
4893 		};
4894 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4895 			.src_port = mask->src_port_mask,
4896 			.dst_port = mask->dst_port_mask,
4897 		};
4898 		attributes->items[2] = (struct rte_flow_item){
4899 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4900 			.spec = &attributes->l4,
4901 			.mask = &attributes->l4_mask,
4902 		};
4903 		break;
4904 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4905 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4906 			.src_port = input->flow.tcp6_flow.src_port,
4907 			.dst_port = input->flow.tcp6_flow.dst_port,
4908 		};
4909 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4910 			.src_port = mask->src_port_mask,
4911 			.dst_port = mask->dst_port_mask,
4912 		};
4913 		attributes->items[2] = (struct rte_flow_item){
4914 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4915 			.spec = &attributes->l4,
4916 			.mask = &attributes->l4_mask,
4917 		};
4918 		break;
4919 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4920 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4921 		break;
4922 	default:
4923 		DRV_LOG(ERR, "port %u invalid flow type%d",
4924 			dev->data->port_id, fdir_filter->input.flow_type);
4925 		rte_errno = ENOTSUP;
4926 		return -rte_errno;
4927 	}
4928 	return 0;
4929 }
4930 
4931 #define FLOW_FDIR_CMP(f1, f2, fld) \
4932 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4933 
4934 /**
4935  * Compare two FDIR flows. If items and actions are identical, the two flows are
4936  * regarded as same.
4937  *
4938  * @param dev
4939  *   Pointer to Ethernet device.
4940  * @param f1
4941  *   FDIR flow to compare.
4942  * @param f2
4943  *   FDIR flow to compare.
4944  *
4945  * @return
4946  *   Zero on match, 1 otherwise.
4947  */
4948 static int
4949 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4950 {
4951 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
4952 	    FLOW_FDIR_CMP(f1, f2, l2) ||
4953 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4954 	    FLOW_FDIR_CMP(f1, f2, l3) ||
4955 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4956 	    FLOW_FDIR_CMP(f1, f2, l4) ||
4957 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4958 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
4959 		return 1;
4960 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4961 	    FLOW_FDIR_CMP(f1, f2, queue))
4962 		return 1;
4963 	return 0;
4964 }
4965 
4966 /**
4967  * Search device flow list to find out a matched FDIR flow.
4968  *
4969  * @param dev
4970  *   Pointer to Ethernet device.
4971  * @param fdir_flow
4972  *   FDIR flow to lookup.
4973  *
4974  * @return
4975  *   Pointer of flow if found, NULL otherwise.
4976  */
4977 static struct rte_flow *
4978 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
4979 {
4980 	struct mlx5_priv *priv = dev->data->dev_private;
4981 	struct rte_flow *flow = NULL;
4982 
4983 	assert(fdir_flow);
4984 	TAILQ_FOREACH(flow, &priv->flows, next) {
4985 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
4986 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
4987 				dev->data->port_id, (void *)flow);
4988 			break;
4989 		}
4990 	}
4991 	return flow;
4992 }
4993 
4994 /**
4995  * Add new flow director filter and store it in list.
4996  *
4997  * @param dev
4998  *   Pointer to Ethernet device.
4999  * @param fdir_filter
5000  *   Flow director filter to add.
5001  *
5002  * @return
5003  *   0 on success, a negative errno value otherwise and rte_errno is set.
5004  */
5005 static int
5006 flow_fdir_filter_add(struct rte_eth_dev *dev,
5007 		     const struct rte_eth_fdir_filter *fdir_filter)
5008 {
5009 	struct mlx5_priv *priv = dev->data->dev_private;
5010 	struct mlx5_fdir *fdir_flow;
5011 	struct rte_flow *flow;
5012 	int ret;
5013 
5014 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5015 	if (!fdir_flow) {
5016 		rte_errno = ENOMEM;
5017 		return -rte_errno;
5018 	}
5019 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5020 	if (ret)
5021 		goto error;
5022 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5023 	if (flow) {
5024 		rte_errno = EEXIST;
5025 		goto error;
5026 	}
5027 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5028 				fdir_flow->items, fdir_flow->actions, true,
5029 				NULL);
5030 	if (!flow)
5031 		goto error;
5032 	assert(!flow->fdir);
5033 	flow->fdir = fdir_flow;
5034 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5035 		dev->data->port_id, (void *)flow);
5036 	return 0;
5037 error:
5038 	rte_free(fdir_flow);
5039 	return -rte_errno;
5040 }
5041 
5042 /**
5043  * Delete specific filter.
5044  *
5045  * @param dev
5046  *   Pointer to Ethernet device.
5047  * @param fdir_filter
5048  *   Filter to be deleted.
5049  *
5050  * @return
5051  *   0 on success, a negative errno value otherwise and rte_errno is set.
5052  */
5053 static int
5054 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5055 			const struct rte_eth_fdir_filter *fdir_filter)
5056 {
5057 	struct mlx5_priv *priv = dev->data->dev_private;
5058 	struct rte_flow *flow;
5059 	struct mlx5_fdir fdir_flow = {
5060 		.attr.group = 0,
5061 	};
5062 	int ret;
5063 
5064 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5065 	if (ret)
5066 		return -rte_errno;
5067 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5068 	if (!flow) {
5069 		rte_errno = ENOENT;
5070 		return -rte_errno;
5071 	}
5072 	flow_list_destroy(dev, &priv->flows, flow);
5073 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5074 		dev->data->port_id, (void *)flow);
5075 	return 0;
5076 }
5077 
5078 /**
5079  * Update queue for specific filter.
5080  *
5081  * @param dev
5082  *   Pointer to Ethernet device.
5083  * @param fdir_filter
5084  *   Filter to be updated.
5085  *
5086  * @return
5087  *   0 on success, a negative errno value otherwise and rte_errno is set.
5088  */
5089 static int
5090 flow_fdir_filter_update(struct rte_eth_dev *dev,
5091 			const struct rte_eth_fdir_filter *fdir_filter)
5092 {
5093 	int ret;
5094 
5095 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5096 	if (ret)
5097 		return ret;
5098 	return flow_fdir_filter_add(dev, fdir_filter);
5099 }
5100 
5101 /**
5102  * Flush all filters.
5103  *
5104  * @param dev
5105  *   Pointer to Ethernet device.
5106  */
5107 static void
5108 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5109 {
5110 	struct mlx5_priv *priv = dev->data->dev_private;
5111 
5112 	mlx5_flow_list_flush(dev, &priv->flows);
5113 }
5114 
5115 /**
5116  * Get flow director information.
5117  *
5118  * @param dev
5119  *   Pointer to Ethernet device.
5120  * @param[out] fdir_info
5121  *   Resulting flow director information.
5122  */
5123 static void
5124 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5125 {
5126 	struct rte_eth_fdir_masks *mask =
5127 		&dev->data->dev_conf.fdir_conf.mask;
5128 
5129 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5130 	fdir_info->guarant_spc = 0;
5131 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5132 	fdir_info->max_flexpayload = 0;
5133 	fdir_info->flow_types_mask[0] = 0;
5134 	fdir_info->flex_payload_unit = 0;
5135 	fdir_info->max_flex_payload_segment_num = 0;
5136 	fdir_info->flex_payload_limit = 0;
5137 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5138 }
5139 
5140 /**
5141  * Deal with flow director operations.
5142  *
5143  * @param dev
5144  *   Pointer to Ethernet device.
5145  * @param filter_op
5146  *   Operation to perform.
5147  * @param arg
5148  *   Pointer to operation-specific structure.
5149  *
5150  * @return
5151  *   0 on success, a negative errno value otherwise and rte_errno is set.
5152  */
5153 static int
5154 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5155 		    void *arg)
5156 {
5157 	enum rte_fdir_mode fdir_mode =
5158 		dev->data->dev_conf.fdir_conf.mode;
5159 
5160 	if (filter_op == RTE_ETH_FILTER_NOP)
5161 		return 0;
5162 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5163 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5164 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5165 			dev->data->port_id, fdir_mode);
5166 		rte_errno = EINVAL;
5167 		return -rte_errno;
5168 	}
5169 	switch (filter_op) {
5170 	case RTE_ETH_FILTER_ADD:
5171 		return flow_fdir_filter_add(dev, arg);
5172 	case RTE_ETH_FILTER_UPDATE:
5173 		return flow_fdir_filter_update(dev, arg);
5174 	case RTE_ETH_FILTER_DELETE:
5175 		return flow_fdir_filter_delete(dev, arg);
5176 	case RTE_ETH_FILTER_FLUSH:
5177 		flow_fdir_filter_flush(dev);
5178 		break;
5179 	case RTE_ETH_FILTER_INFO:
5180 		flow_fdir_info_get(dev, arg);
5181 		break;
5182 	default:
5183 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5184 			dev->data->port_id, filter_op);
5185 		rte_errno = EINVAL;
5186 		return -rte_errno;
5187 	}
5188 	return 0;
5189 }
5190 
5191 /**
5192  * Manage filter operations.
5193  *
5194  * @param dev
5195  *   Pointer to Ethernet device structure.
5196  * @param filter_type
5197  *   Filter type.
5198  * @param filter_op
5199  *   Operation to perform.
5200  * @param arg
5201  *   Pointer to operation-specific structure.
5202  *
5203  * @return
5204  *   0 on success, a negative errno value otherwise and rte_errno is set.
5205  */
5206 int
5207 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5208 		     enum rte_filter_type filter_type,
5209 		     enum rte_filter_op filter_op,
5210 		     void *arg)
5211 {
5212 	switch (filter_type) {
5213 	case RTE_ETH_FILTER_GENERIC:
5214 		if (filter_op != RTE_ETH_FILTER_GET) {
5215 			rte_errno = EINVAL;
5216 			return -rte_errno;
5217 		}
5218 		*(const void **)arg = &mlx5_flow_ops;
5219 		return 0;
5220 	case RTE_ETH_FILTER_FDIR:
5221 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5222 	default:
5223 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5224 			dev->data->port_id, filter_type);
5225 		rte_errno = ENOTSUP;
5226 		return -rte_errno;
5227 	}
5228 	return 0;
5229 }
5230 
5231 /**
5232  * Create the needed meter and suffix tables.
5233  *
5234  * @param[in] dev
5235  *   Pointer to Ethernet device.
5236  * @param[in] fm
5237  *   Pointer to the flow meter.
5238  *
5239  * @return
5240  *   Pointer to table set on success, NULL otherwise.
5241  */
5242 struct mlx5_meter_domains_infos *
5243 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5244 			  const struct mlx5_flow_meter *fm)
5245 {
5246 	const struct mlx5_flow_driver_ops *fops;
5247 
5248 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5249 	return fops->create_mtr_tbls(dev, fm);
5250 }
5251 
5252 /**
5253  * Destroy the meter table set.
5254  *
5255  * @param[in] dev
5256  *   Pointer to Ethernet device.
5257  * @param[in] tbl
5258  *   Pointer to the meter table set.
5259  *
5260  * @return
5261  *   0 on success.
5262  */
5263 int
5264 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5265 			   struct mlx5_meter_domains_infos *tbls)
5266 {
5267 	const struct mlx5_flow_driver_ops *fops;
5268 
5269 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5270 	return fops->destroy_mtr_tbls(dev, tbls);
5271 }
5272 
5273 /**
5274  * Create policer rules.
5275  *
5276  * @param[in] dev
5277  *   Pointer to Ethernet device.
5278  * @param[in] fm
5279  *   Pointer to flow meter structure.
5280  * @param[in] attr
5281  *   Pointer to flow attributes.
5282  *
5283  * @return
5284  *   0 on success, -1 otherwise.
5285  */
5286 int
5287 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5288 			       struct mlx5_flow_meter *fm,
5289 			       const struct rte_flow_attr *attr)
5290 {
5291 	const struct mlx5_flow_driver_ops *fops;
5292 
5293 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5294 	return fops->create_policer_rules(dev, fm, attr);
5295 }
5296 
5297 /**
5298  * Destroy policer rules.
5299  *
5300  * @param[in] fm
5301  *   Pointer to flow meter structure.
5302  * @param[in] attr
5303  *   Pointer to flow attributes.
5304  *
5305  * @return
5306  *   0 on success, -1 otherwise.
5307  */
5308 int
5309 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5310 				struct mlx5_flow_meter *fm,
5311 				const struct rte_flow_attr *attr)
5312 {
5313 	const struct mlx5_flow_driver_ops *fops;
5314 
5315 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5316 	return fops->destroy_policer_rules(dev, fm, attr);
5317 }
5318 
5319 /**
5320  * Allocate a counter.
5321  *
5322  * @param[in] dev
5323  *   Pointer to Ethernet device structure.
5324  *
5325  * @return
5326  *   Pointer to allocated counter  on success, NULL otherwise.
5327  */
5328 struct mlx5_flow_counter *
5329 mlx5_counter_alloc(struct rte_eth_dev *dev)
5330 {
5331 	const struct mlx5_flow_driver_ops *fops;
5332 	struct rte_flow_attr attr = { .transfer = 0 };
5333 
5334 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5335 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5336 		return fops->counter_alloc(dev);
5337 	}
5338 	DRV_LOG(ERR,
5339 		"port %u counter allocate is not supported.",
5340 		 dev->data->port_id);
5341 	return NULL;
5342 }
5343 
5344 /**
5345  * Free a counter.
5346  *
5347  * @param[in] dev
5348  *   Pointer to Ethernet device structure.
5349  * @param[in] cnt
5350  *   Pointer to counter to be free.
5351  */
5352 void
5353 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5354 {
5355 	const struct mlx5_flow_driver_ops *fops;
5356 	struct rte_flow_attr attr = { .transfer = 0 };
5357 
5358 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5359 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5360 		fops->counter_free(dev, cnt);
5361 		return;
5362 	}
5363 	DRV_LOG(ERR,
5364 		"port %u counter free is not supported.",
5365 		 dev->data->port_id);
5366 }
5367 
5368 /**
5369  * Query counter statistics.
5370  *
5371  * @param[in] dev
5372  *   Pointer to Ethernet device structure.
5373  * @param[in] cnt
5374  *   Pointer to counter to query.
5375  * @param[in] clear
5376  *   Set to clear counter statistics.
5377  * @param[out] pkts
5378  *   The counter hits packets number to save.
5379  * @param[out] bytes
5380  *   The counter hits bytes number to save.
5381  *
5382  * @return
5383  *   0 on success, a negative errno value otherwise.
5384  */
5385 int
5386 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5387 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5388 {
5389 	const struct mlx5_flow_driver_ops *fops;
5390 	struct rte_flow_attr attr = { .transfer = 0 };
5391 
5392 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5393 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5394 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5395 	}
5396 	DRV_LOG(ERR,
5397 		"port %u counter query is not supported.",
5398 		 dev->data->port_id);
5399 	return -ENOTSUP;
5400 }
5401 
5402 #define MLX5_POOL_QUERY_FREQ_US 1000000
5403 
5404 /**
5405  * Set the periodic procedure for triggering asynchronous batch queries for all
5406  * the counter pools.
5407  *
5408  * @param[in] sh
5409  *   Pointer to mlx5_ibv_shared object.
5410  */
5411 void
5412 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5413 {
5414 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5415 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5416 	uint32_t us;
5417 
5418 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5419 	pools_n += rte_atomic16_read(&cont->n_valid);
5420 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5421 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5422 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5423 		sh->cmng.query_thread_on = 0;
5424 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5425 	} else {
5426 		sh->cmng.query_thread_on = 1;
5427 	}
5428 }
5429 
5430 /**
5431  * The periodic procedure for triggering asynchronous batch queries for all the
5432  * counter pools. This function is probably called by the host thread.
5433  *
5434  * @param[in] arg
5435  *   The parameter for the alarm process.
5436  */
5437 void
5438 mlx5_flow_query_alarm(void *arg)
5439 {
5440 	struct mlx5_ibv_shared *sh = arg;
5441 	struct mlx5_devx_obj *dcs;
5442 	uint16_t offset;
5443 	int ret;
5444 	uint8_t batch = sh->cmng.batch;
5445 	uint16_t pool_index = sh->cmng.pool_index;
5446 	struct mlx5_pools_container *cont;
5447 	struct mlx5_pools_container *mcont;
5448 	struct mlx5_flow_counter_pool *pool;
5449 
5450 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5451 		goto set_alarm;
5452 next_container:
5453 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5454 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5455 	/* Check if resize was done and need to flip a container. */
5456 	if (cont != mcont) {
5457 		if (cont->pools) {
5458 			/* Clean the old container. */
5459 			rte_free(cont->pools);
5460 			memset(cont, 0, sizeof(*cont));
5461 		}
5462 		rte_cio_wmb();
5463 		 /* Flip the host container. */
5464 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5465 		cont = mcont;
5466 	}
5467 	if (!cont->pools) {
5468 		/* 2 empty containers case is unexpected. */
5469 		if (unlikely(batch != sh->cmng.batch))
5470 			goto set_alarm;
5471 		batch ^= 0x1;
5472 		pool_index = 0;
5473 		goto next_container;
5474 	}
5475 	pool = cont->pools[pool_index];
5476 	if (pool->raw_hw)
5477 		/* There is a pool query in progress. */
5478 		goto set_alarm;
5479 	pool->raw_hw =
5480 		LIST_FIRST(&sh->cmng.free_stat_raws);
5481 	if (!pool->raw_hw)
5482 		/* No free counter statistics raw memory. */
5483 		goto set_alarm;
5484 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5485 							      (&pool->a64_dcs);
5486 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5487 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5488 					       offset, NULL, NULL,
5489 					       pool->raw_hw->mem_mng->dm->id,
5490 					       (void *)(uintptr_t)
5491 					       (pool->raw_hw->data + offset),
5492 					       sh->devx_comp,
5493 					       (uint64_t)(uintptr_t)pool);
5494 	if (ret) {
5495 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5496 			" %d", pool->min_dcs->id);
5497 		pool->raw_hw = NULL;
5498 		goto set_alarm;
5499 	}
5500 	pool->raw_hw->min_dcs_id = dcs->id;
5501 	LIST_REMOVE(pool->raw_hw, next);
5502 	sh->cmng.pending_queries++;
5503 	pool_index++;
5504 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5505 		batch ^= 0x1;
5506 		pool_index = 0;
5507 	}
5508 set_alarm:
5509 	sh->cmng.batch = batch;
5510 	sh->cmng.pool_index = pool_index;
5511 	mlx5_set_query_alarm(sh);
5512 }
5513 
5514 /**
5515  * Handler for the HW respond about ready values from an asynchronous batch
5516  * query. This function is probably called by the host thread.
5517  *
5518  * @param[in] sh
5519  *   The pointer to the shared IB device context.
5520  * @param[in] async_id
5521  *   The Devx async ID.
5522  * @param[in] status
5523  *   The status of the completion.
5524  */
5525 void
5526 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5527 				  uint64_t async_id, int status)
5528 {
5529 	struct mlx5_flow_counter_pool *pool =
5530 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5531 	struct mlx5_counter_stats_raw *raw_to_free;
5532 
5533 	if (unlikely(status)) {
5534 		raw_to_free = pool->raw_hw;
5535 	} else {
5536 		raw_to_free = pool->raw;
5537 		rte_spinlock_lock(&pool->sl);
5538 		pool->raw = pool->raw_hw;
5539 		rte_spinlock_unlock(&pool->sl);
5540 		rte_atomic64_add(&pool->query_gen, 1);
5541 		/* Be sure the new raw counters data is updated in memory. */
5542 		rte_cio_wmb();
5543 	}
5544 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5545 	pool->raw_hw = NULL;
5546 	sh->cmng.pending_queries--;
5547 }
5548 
5549 /**
5550  * Translate the rte_flow group index to HW table value.
5551  *
5552  * @param[in] attributes
5553  *   Pointer to flow attributes
5554  * @param[in] external
5555  *   Value is part of flow rule created by request external to PMD.
5556  * @param[in] group
5557  *   rte_flow group index value.
5558  * @param[out] table
5559  *   HW table value.
5560  * @param[out] error
5561  *   Pointer to error structure.
5562  *
5563  * @return
5564  *   0 on success, a negative errno value otherwise and rte_errno is set.
5565  */
5566 int
5567 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5568 			 uint32_t group, uint32_t *table,
5569 			 struct rte_flow_error *error)
5570 {
5571 	if (attributes->transfer && external) {
5572 		if (group == UINT32_MAX)
5573 			return rte_flow_error_set
5574 						(error, EINVAL,
5575 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5576 						 NULL,
5577 						 "group index not supported");
5578 		*table = group + 1;
5579 	} else {
5580 		*table = group;
5581 	}
5582 	return 0;
5583 }
5584 
5585 /**
5586  * Discover availability of metadata reg_c's.
5587  *
5588  * Iteratively use test flows to check availability.
5589  *
5590  * @param[in] dev
5591  *   Pointer to the Ethernet device structure.
5592  *
5593  * @return
5594  *   0 on success, a negative errno value otherwise and rte_errno is set.
5595  */
5596 int
5597 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5598 {
5599 	struct mlx5_priv *priv = dev->data->dev_private;
5600 	struct mlx5_dev_config *config = &priv->config;
5601 	enum modify_reg idx;
5602 	int n = 0;
5603 
5604 	/* reg_c[0] and reg_c[1] are reserved. */
5605 	config->flow_mreg_c[n++] = REG_C_0;
5606 	config->flow_mreg_c[n++] = REG_C_1;
5607 	/* Discover availability of other reg_c's. */
5608 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5609 		struct rte_flow_attr attr = {
5610 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5611 			.priority = MLX5_FLOW_PRIO_RSVD,
5612 			.ingress = 1,
5613 		};
5614 		struct rte_flow_item items[] = {
5615 			[0] = {
5616 				.type = RTE_FLOW_ITEM_TYPE_END,
5617 			},
5618 		};
5619 		struct rte_flow_action actions[] = {
5620 			[0] = {
5621 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5622 				.conf = &(struct mlx5_flow_action_copy_mreg){
5623 					.src = REG_C_1,
5624 					.dst = idx,
5625 				},
5626 			},
5627 			[1] = {
5628 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5629 				.conf = &(struct rte_flow_action_jump){
5630 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5631 				},
5632 			},
5633 			[2] = {
5634 				.type = RTE_FLOW_ACTION_TYPE_END,
5635 			},
5636 		};
5637 		struct rte_flow *flow;
5638 		struct rte_flow_error error;
5639 
5640 		if (!config->dv_flow_en)
5641 			break;
5642 		/* Create internal flow, validation skips copy action. */
5643 		flow = flow_list_create(dev, NULL, &attr, items,
5644 					actions, false, &error);
5645 		if (!flow)
5646 			continue;
5647 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5648 			config->flow_mreg_c[n++] = idx;
5649 		flow_list_destroy(dev, NULL, flow);
5650 	}
5651 	for (; n < MLX5_MREG_C_NUM; ++n)
5652 		config->flow_mreg_c[n] = REG_NONE;
5653 	return 0;
5654 }
5655