xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 9cd9d3e702fba4700539c1a2eddac13dd14ecf70)
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_glue.h>
31 #include <mlx5_devx_cmds.h>
32 #include <mlx5_prm.h>
33 
34 #include "mlx5_defs.h"
35 #include "mlx5.h"
36 #include "mlx5_flow.h"
37 #include "mlx5_rxtx.h"
38 
39 /* Dev ops structure defined in mlx5.c */
40 extern const struct eth_dev_ops mlx5_dev_ops;
41 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
42 
43 /** Device flow drivers. */
44 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
45 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
46 #endif
47 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
48 
49 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
50 
51 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
52 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
53 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
54 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
55 #endif
56 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
57 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
58 };
59 
60 enum mlx5_expansion {
61 	MLX5_EXPANSION_ROOT,
62 	MLX5_EXPANSION_ROOT_OUTER,
63 	MLX5_EXPANSION_ROOT_ETH_VLAN,
64 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
65 	MLX5_EXPANSION_OUTER_ETH,
66 	MLX5_EXPANSION_OUTER_ETH_VLAN,
67 	MLX5_EXPANSION_OUTER_VLAN,
68 	MLX5_EXPANSION_OUTER_IPV4,
69 	MLX5_EXPANSION_OUTER_IPV4_UDP,
70 	MLX5_EXPANSION_OUTER_IPV4_TCP,
71 	MLX5_EXPANSION_OUTER_IPV6,
72 	MLX5_EXPANSION_OUTER_IPV6_UDP,
73 	MLX5_EXPANSION_OUTER_IPV6_TCP,
74 	MLX5_EXPANSION_VXLAN,
75 	MLX5_EXPANSION_VXLAN_GPE,
76 	MLX5_EXPANSION_GRE,
77 	MLX5_EXPANSION_MPLS,
78 	MLX5_EXPANSION_ETH,
79 	MLX5_EXPANSION_ETH_VLAN,
80 	MLX5_EXPANSION_VLAN,
81 	MLX5_EXPANSION_IPV4,
82 	MLX5_EXPANSION_IPV4_UDP,
83 	MLX5_EXPANSION_IPV4_TCP,
84 	MLX5_EXPANSION_IPV6,
85 	MLX5_EXPANSION_IPV6_UDP,
86 	MLX5_EXPANSION_IPV6_TCP,
87 };
88 
89 /** Supported expansion of items. */
90 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
91 	[MLX5_EXPANSION_ROOT] = {
92 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
93 						 MLX5_EXPANSION_IPV4,
94 						 MLX5_EXPANSION_IPV6),
95 		.type = RTE_FLOW_ITEM_TYPE_END,
96 	},
97 	[MLX5_EXPANSION_ROOT_OUTER] = {
98 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
99 						 MLX5_EXPANSION_OUTER_IPV4,
100 						 MLX5_EXPANSION_OUTER_IPV6),
101 		.type = RTE_FLOW_ITEM_TYPE_END,
102 	},
103 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
104 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
105 		.type = RTE_FLOW_ITEM_TYPE_END,
106 	},
107 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
108 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
109 		.type = RTE_FLOW_ITEM_TYPE_END,
110 	},
111 	[MLX5_EXPANSION_OUTER_ETH] = {
112 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
113 						 MLX5_EXPANSION_OUTER_IPV6,
114 						 MLX5_EXPANSION_MPLS),
115 		.type = RTE_FLOW_ITEM_TYPE_ETH,
116 		.rss_types = 0,
117 	},
118 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
119 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
120 		.type = RTE_FLOW_ITEM_TYPE_ETH,
121 		.rss_types = 0,
122 	},
123 	[MLX5_EXPANSION_OUTER_VLAN] = {
124 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
125 						 MLX5_EXPANSION_OUTER_IPV6),
126 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
127 	},
128 	[MLX5_EXPANSION_OUTER_IPV4] = {
129 		.next = RTE_FLOW_EXPAND_RSS_NEXT
130 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
131 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
132 			 MLX5_EXPANSION_GRE,
133 			 MLX5_EXPANSION_IPV4,
134 			 MLX5_EXPANSION_IPV6),
135 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
136 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
137 			ETH_RSS_NONFRAG_IPV4_OTHER,
138 	},
139 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
140 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
141 						 MLX5_EXPANSION_VXLAN_GPE),
142 		.type = RTE_FLOW_ITEM_TYPE_UDP,
143 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
144 	},
145 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
146 		.type = RTE_FLOW_ITEM_TYPE_TCP,
147 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
148 	},
149 	[MLX5_EXPANSION_OUTER_IPV6] = {
150 		.next = RTE_FLOW_EXPAND_RSS_NEXT
151 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
152 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
153 			 MLX5_EXPANSION_IPV4,
154 			 MLX5_EXPANSION_IPV6),
155 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
156 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
157 			ETH_RSS_NONFRAG_IPV6_OTHER,
158 	},
159 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
160 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
161 						 MLX5_EXPANSION_VXLAN_GPE),
162 		.type = RTE_FLOW_ITEM_TYPE_UDP,
163 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
164 	},
165 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
166 		.type = RTE_FLOW_ITEM_TYPE_TCP,
167 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
168 	},
169 	[MLX5_EXPANSION_VXLAN] = {
170 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
171 						 MLX5_EXPANSION_IPV4,
172 						 MLX5_EXPANSION_IPV6),
173 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
174 	},
175 	[MLX5_EXPANSION_VXLAN_GPE] = {
176 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
177 						 MLX5_EXPANSION_IPV4,
178 						 MLX5_EXPANSION_IPV6),
179 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
180 	},
181 	[MLX5_EXPANSION_GRE] = {
182 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
183 		.type = RTE_FLOW_ITEM_TYPE_GRE,
184 	},
185 	[MLX5_EXPANSION_MPLS] = {
186 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
187 						 MLX5_EXPANSION_IPV6),
188 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
189 	},
190 	[MLX5_EXPANSION_ETH] = {
191 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
192 						 MLX5_EXPANSION_IPV6),
193 		.type = RTE_FLOW_ITEM_TYPE_ETH,
194 	},
195 	[MLX5_EXPANSION_ETH_VLAN] = {
196 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
197 		.type = RTE_FLOW_ITEM_TYPE_ETH,
198 	},
199 	[MLX5_EXPANSION_VLAN] = {
200 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
201 						 MLX5_EXPANSION_IPV6),
202 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
203 	},
204 	[MLX5_EXPANSION_IPV4] = {
205 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
206 						 MLX5_EXPANSION_IPV4_TCP),
207 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
208 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
209 			ETH_RSS_NONFRAG_IPV4_OTHER,
210 	},
211 	[MLX5_EXPANSION_IPV4_UDP] = {
212 		.type = RTE_FLOW_ITEM_TYPE_UDP,
213 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
214 	},
215 	[MLX5_EXPANSION_IPV4_TCP] = {
216 		.type = RTE_FLOW_ITEM_TYPE_TCP,
217 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
218 	},
219 	[MLX5_EXPANSION_IPV6] = {
220 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
221 						 MLX5_EXPANSION_IPV6_TCP),
222 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
223 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
224 			ETH_RSS_NONFRAG_IPV6_OTHER,
225 	},
226 	[MLX5_EXPANSION_IPV6_UDP] = {
227 		.type = RTE_FLOW_ITEM_TYPE_UDP,
228 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
229 	},
230 	[MLX5_EXPANSION_IPV6_TCP] = {
231 		.type = RTE_FLOW_ITEM_TYPE_TCP,
232 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
233 	},
234 };
235 
236 static const struct rte_flow_ops mlx5_flow_ops = {
237 	.validate = mlx5_flow_validate,
238 	.create = mlx5_flow_create,
239 	.destroy = mlx5_flow_destroy,
240 	.flush = mlx5_flow_flush,
241 	.isolate = mlx5_flow_isolate,
242 	.query = mlx5_flow_query,
243 	.dev_dump = mlx5_flow_dev_dump,
244 };
245 
246 /* Convert FDIR request to Generic flow. */
247 struct mlx5_fdir {
248 	struct rte_flow_attr attr;
249 	struct rte_flow_item items[4];
250 	struct rte_flow_item_eth l2;
251 	struct rte_flow_item_eth l2_mask;
252 	union {
253 		struct rte_flow_item_ipv4 ipv4;
254 		struct rte_flow_item_ipv6 ipv6;
255 	} l3;
256 	union {
257 		struct rte_flow_item_ipv4 ipv4;
258 		struct rte_flow_item_ipv6 ipv6;
259 	} l3_mask;
260 	union {
261 		struct rte_flow_item_udp udp;
262 		struct rte_flow_item_tcp tcp;
263 	} l4;
264 	union {
265 		struct rte_flow_item_udp udp;
266 		struct rte_flow_item_tcp tcp;
267 	} l4_mask;
268 	struct rte_flow_action actions[2];
269 	struct rte_flow_action_queue queue;
270 };
271 
272 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
273 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
274 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
275 };
276 
277 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
278 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
279 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
280 	{ 9, 10, 11 }, { 12, 13, 14 },
281 };
282 
283 /* Tunnel information. */
284 struct mlx5_flow_tunnel_info {
285 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
286 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
287 };
288 
289 static struct mlx5_flow_tunnel_info tunnels_info[] = {
290 	{
291 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
292 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
293 	},
294 	{
295 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
296 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
297 	},
298 	{
299 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
300 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
301 	},
302 	{
303 		.tunnel = MLX5_FLOW_LAYER_GRE,
304 		.ptype = RTE_PTYPE_TUNNEL_GRE,
305 	},
306 	{
307 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
308 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
309 	},
310 	{
311 		.tunnel = MLX5_FLOW_LAYER_MPLS,
312 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
313 	},
314 	{
315 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
316 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
317 	},
318 	{
319 		.tunnel = MLX5_FLOW_LAYER_IPIP,
320 		.ptype = RTE_PTYPE_TUNNEL_IP,
321 	},
322 	{
323 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
324 		.ptype = RTE_PTYPE_TUNNEL_IP,
325 	},
326 	{
327 		.tunnel = MLX5_FLOW_LAYER_GTP,
328 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
329 	},
330 };
331 
332 /**
333  * Translate tag ID to register.
334  *
335  * @param[in] dev
336  *   Pointer to the Ethernet device structure.
337  * @param[in] feature
338  *   The feature that request the register.
339  * @param[in] id
340  *   The request register ID.
341  * @param[out] error
342  *   Error description in case of any.
343  *
344  * @return
345  *   The request register on success, a negative errno
346  *   value otherwise and rte_errno is set.
347  */
348 int
349 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
350 		     enum mlx5_feature_name feature,
351 		     uint32_t id,
352 		     struct rte_flow_error *error)
353 {
354 	struct mlx5_priv *priv = dev->data->dev_private;
355 	struct mlx5_dev_config *config = &priv->config;
356 	enum modify_reg start_reg;
357 	bool skip_mtr_reg = false;
358 
359 	switch (feature) {
360 	case MLX5_HAIRPIN_RX:
361 		return REG_B;
362 	case MLX5_HAIRPIN_TX:
363 		return REG_A;
364 	case MLX5_METADATA_RX:
365 		switch (config->dv_xmeta_en) {
366 		case MLX5_XMETA_MODE_LEGACY:
367 			return REG_B;
368 		case MLX5_XMETA_MODE_META16:
369 			return REG_C_0;
370 		case MLX5_XMETA_MODE_META32:
371 			return REG_C_1;
372 		}
373 		break;
374 	case MLX5_METADATA_TX:
375 		return REG_A;
376 	case MLX5_METADATA_FDB:
377 		switch (config->dv_xmeta_en) {
378 		case MLX5_XMETA_MODE_LEGACY:
379 			return REG_NONE;
380 		case MLX5_XMETA_MODE_META16:
381 			return REG_C_0;
382 		case MLX5_XMETA_MODE_META32:
383 			return REG_C_1;
384 		}
385 		break;
386 	case MLX5_FLOW_MARK:
387 		switch (config->dv_xmeta_en) {
388 		case MLX5_XMETA_MODE_LEGACY:
389 			return REG_NONE;
390 		case MLX5_XMETA_MODE_META16:
391 			return REG_C_1;
392 		case MLX5_XMETA_MODE_META32:
393 			return REG_C_0;
394 		}
395 		break;
396 	case MLX5_MTR_SFX:
397 		/*
398 		 * If meter color and flow match share one register, flow match
399 		 * should use the meter color register for match.
400 		 */
401 		if (priv->mtr_reg_share)
402 			return priv->mtr_color_reg;
403 		else
404 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
405 			       REG_C_3;
406 	case MLX5_MTR_COLOR:
407 		MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
408 		return priv->mtr_color_reg;
409 	case MLX5_COPY_MARK:
410 		/*
411 		 * Metadata COPY_MARK register using is in meter suffix sub
412 		 * flow while with meter. It's safe to share the same register.
413 		 */
414 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
415 	case MLX5_APP_TAG:
416 		/*
417 		 * If meter is enable, it will engage the register for color
418 		 * match and flow match. If meter color match is not using the
419 		 * REG_C_2, need to skip the REG_C_x be used by meter color
420 		 * match.
421 		 * If meter is disable, free to use all available registers.
422 		 */
423 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
424 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
425 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
426 		if (id > (REG_C_7 - start_reg))
427 			return rte_flow_error_set(error, EINVAL,
428 						  RTE_FLOW_ERROR_TYPE_ITEM,
429 						  NULL, "invalid tag id");
430 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
431 			return rte_flow_error_set(error, ENOTSUP,
432 						  RTE_FLOW_ERROR_TYPE_ITEM,
433 						  NULL, "unsupported tag id");
434 		/*
435 		 * This case means meter is using the REG_C_x great than 2.
436 		 * Take care not to conflict with meter color REG_C_x.
437 		 * If the available index REG_C_y >= REG_C_x, skip the
438 		 * color register.
439 		 */
440 		if (skip_mtr_reg && config->flow_mreg_c
441 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
442 			if (config->flow_mreg_c
443 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
444 				return config->flow_mreg_c
445 					       [id + 1 + start_reg - REG_C_0];
446 			return rte_flow_error_set(error, ENOTSUP,
447 						  RTE_FLOW_ERROR_TYPE_ITEM,
448 						  NULL, "unsupported tag id");
449 		}
450 		return config->flow_mreg_c[id + start_reg - REG_C_0];
451 	}
452 	MLX5_ASSERT(false);
453 	return rte_flow_error_set(error, EINVAL,
454 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
455 				  NULL, "invalid feature name");
456 }
457 
458 /**
459  * Check extensive flow metadata register support.
460  *
461  * @param dev
462  *   Pointer to rte_eth_dev structure.
463  *
464  * @return
465  *   True if device supports extensive flow metadata register, otherwise false.
466  */
467 bool
468 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
469 {
470 	struct mlx5_priv *priv = dev->data->dev_private;
471 	struct mlx5_dev_config *config = &priv->config;
472 
473 	/*
474 	 * Having available reg_c can be regarded inclusively as supporting
475 	 * extensive flow metadata register, which could mean,
476 	 * - metadata register copy action by modify header.
477 	 * - 16 modify header actions is supported.
478 	 * - reg_c's are preserved across different domain (FDB and NIC) on
479 	 *   packet loopback by flow lookup miss.
480 	 */
481 	return config->flow_mreg_c[2] != REG_NONE;
482 }
483 
484 /**
485  * Discover the maximum number of priority available.
486  *
487  * @param[in] dev
488  *   Pointer to the Ethernet device structure.
489  *
490  * @return
491  *   number of supported flow priority on success, a negative errno
492  *   value otherwise and rte_errno is set.
493  */
494 int
495 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
496 {
497 	struct mlx5_priv *priv = dev->data->dev_private;
498 	struct {
499 		struct ibv_flow_attr attr;
500 		struct ibv_flow_spec_eth eth;
501 		struct ibv_flow_spec_action_drop drop;
502 	} flow_attr = {
503 		.attr = {
504 			.num_of_specs = 2,
505 			.port = (uint8_t)priv->ibv_port,
506 		},
507 		.eth = {
508 			.type = IBV_FLOW_SPEC_ETH,
509 			.size = sizeof(struct ibv_flow_spec_eth),
510 		},
511 		.drop = {
512 			.size = sizeof(struct ibv_flow_spec_action_drop),
513 			.type = IBV_FLOW_SPEC_ACTION_DROP,
514 		},
515 	};
516 	struct ibv_flow *flow;
517 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
518 	uint16_t vprio[] = { 8, 16 };
519 	int i;
520 	int priority = 0;
521 
522 	if (!drop) {
523 		rte_errno = ENOTSUP;
524 		return -rte_errno;
525 	}
526 	for (i = 0; i != RTE_DIM(vprio); i++) {
527 		flow_attr.attr.priority = vprio[i] - 1;
528 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
529 		if (!flow)
530 			break;
531 		claim_zero(mlx5_glue->destroy_flow(flow));
532 		priority = vprio[i];
533 	}
534 	mlx5_hrxq_drop_release(dev);
535 	switch (priority) {
536 	case 8:
537 		priority = RTE_DIM(priority_map_3);
538 		break;
539 	case 16:
540 		priority = RTE_DIM(priority_map_5);
541 		break;
542 	default:
543 		rte_errno = ENOTSUP;
544 		DRV_LOG(ERR,
545 			"port %u verbs maximum priority: %d expected 8/16",
546 			dev->data->port_id, priority);
547 		return -rte_errno;
548 	}
549 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
550 		dev->data->port_id, priority);
551 	return priority;
552 }
553 
554 /**
555  * Adjust flow priority based on the highest layer and the request priority.
556  *
557  * @param[in] dev
558  *   Pointer to the Ethernet device structure.
559  * @param[in] priority
560  *   The rule base priority.
561  * @param[in] subpriority
562  *   The priority based on the items.
563  *
564  * @return
565  *   The new priority.
566  */
567 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
568 				   uint32_t subpriority)
569 {
570 	uint32_t res = 0;
571 	struct mlx5_priv *priv = dev->data->dev_private;
572 
573 	switch (priv->config.flow_prio) {
574 	case RTE_DIM(priority_map_3):
575 		res = priority_map_3[priority][subpriority];
576 		break;
577 	case RTE_DIM(priority_map_5):
578 		res = priority_map_5[priority][subpriority];
579 		break;
580 	}
581 	return  res;
582 }
583 
584 /**
585  * Verify the @p item specifications (spec, last, mask) are compatible with the
586  * NIC capabilities.
587  *
588  * @param[in] item
589  *   Item specification.
590  * @param[in] mask
591  *   @p item->mask or flow default bit-masks.
592  * @param[in] nic_mask
593  *   Bit-masks covering supported fields by the NIC to compare with user mask.
594  * @param[in] size
595  *   Bit-masks size in bytes.
596  * @param[out] error
597  *   Pointer to error structure.
598  *
599  * @return
600  *   0 on success, a negative errno value otherwise and rte_errno is set.
601  */
602 int
603 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
604 			  const uint8_t *mask,
605 			  const uint8_t *nic_mask,
606 			  unsigned int size,
607 			  struct rte_flow_error *error)
608 {
609 	unsigned int i;
610 
611 	MLX5_ASSERT(nic_mask);
612 	for (i = 0; i < size; ++i)
613 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
614 			return rte_flow_error_set(error, ENOTSUP,
615 						  RTE_FLOW_ERROR_TYPE_ITEM,
616 						  item,
617 						  "mask enables non supported"
618 						  " bits");
619 	if (!item->spec && (item->mask || item->last))
620 		return rte_flow_error_set(error, EINVAL,
621 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
622 					  "mask/last without a spec is not"
623 					  " supported");
624 	if (item->spec && item->last) {
625 		uint8_t spec[size];
626 		uint8_t last[size];
627 		unsigned int i;
628 		int ret;
629 
630 		for (i = 0; i < size; ++i) {
631 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
632 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
633 		}
634 		ret = memcmp(spec, last, size);
635 		if (ret != 0)
636 			return rte_flow_error_set(error, EINVAL,
637 						  RTE_FLOW_ERROR_TYPE_ITEM,
638 						  item,
639 						  "range is not valid");
640 	}
641 	return 0;
642 }
643 
644 /**
645  * Adjust the hash fields according to the @p flow information.
646  *
647  * @param[in] dev_flow.
648  *   Pointer to the mlx5_flow.
649  * @param[in] tunnel
650  *   1 when the hash field is for a tunnel item.
651  * @param[in] layer_types
652  *   ETH_RSS_* types.
653  * @param[in] hash_fields
654  *   Item hash fields.
655  *
656  * @return
657  *   The hash fields that should be used.
658  */
659 uint64_t
660 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
661 			    int tunnel __rte_unused, uint64_t layer_types,
662 			    uint64_t hash_fields)
663 {
664 	struct rte_flow *flow = dev_flow->flow;
665 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
666 	int rss_request_inner = flow->rss.level >= 2;
667 
668 	/* Check RSS hash level for tunnel. */
669 	if (tunnel && rss_request_inner)
670 		hash_fields |= IBV_RX_HASH_INNER;
671 	else if (tunnel || rss_request_inner)
672 		return 0;
673 #endif
674 	/* Check if requested layer matches RSS hash fields. */
675 	if (!(flow->rss.types & layer_types))
676 		return 0;
677 	return hash_fields;
678 }
679 
680 /**
681  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
682  * if several tunnel rules are used on this queue, the tunnel ptype will be
683  * cleared.
684  *
685  * @param rxq_ctrl
686  *   Rx queue to update.
687  */
688 static void
689 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
690 {
691 	unsigned int i;
692 	uint32_t tunnel_ptype = 0;
693 
694 	/* Look up for the ptype to use. */
695 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
696 		if (!rxq_ctrl->flow_tunnels_n[i])
697 			continue;
698 		if (!tunnel_ptype) {
699 			tunnel_ptype = tunnels_info[i].ptype;
700 		} else {
701 			tunnel_ptype = 0;
702 			break;
703 		}
704 	}
705 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
706 }
707 
708 /**
709  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
710  * flow.
711  *
712  * @param[in] dev
713  *   Pointer to the Ethernet device structure.
714  * @param[in] dev_flow
715  *   Pointer to device flow structure.
716  */
717 static void
718 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
719 {
720 	struct mlx5_priv *priv = dev->data->dev_private;
721 	struct rte_flow *flow = dev_flow->flow;
722 	const int mark = !!(dev_flow->actions &
723 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
724 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
725 	unsigned int i;
726 
727 	for (i = 0; i != flow->rss.queue_num; ++i) {
728 		int idx = (*flow->rss.queue)[i];
729 		struct mlx5_rxq_ctrl *rxq_ctrl =
730 			container_of((*priv->rxqs)[idx],
731 				     struct mlx5_rxq_ctrl, rxq);
732 
733 		/*
734 		 * To support metadata register copy on Tx loopback,
735 		 * this must be always enabled (metadata may arive
736 		 * from other port - not from local flows only.
737 		 */
738 		if (priv->config.dv_flow_en &&
739 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
740 		    mlx5_flow_ext_mreg_supported(dev)) {
741 			rxq_ctrl->rxq.mark = 1;
742 			rxq_ctrl->flow_mark_n = 1;
743 		} else if (mark) {
744 			rxq_ctrl->rxq.mark = 1;
745 			rxq_ctrl->flow_mark_n++;
746 		}
747 		if (tunnel) {
748 			unsigned int j;
749 
750 			/* Increase the counter matching the flow. */
751 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
752 				if ((tunnels_info[j].tunnel &
753 				     dev_flow->layers) ==
754 				    tunnels_info[j].tunnel) {
755 					rxq_ctrl->flow_tunnels_n[j]++;
756 					break;
757 				}
758 			}
759 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
760 		}
761 	}
762 }
763 
764 /**
765  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
766  *
767  * @param[in] dev
768  *   Pointer to the Ethernet device structure.
769  * @param[in] flow
770  *   Pointer to flow structure.
771  */
772 static void
773 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
774 {
775 	struct mlx5_flow *dev_flow;
776 
777 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
778 		flow_drv_rxq_flags_set(dev, dev_flow);
779 }
780 
781 /**
782  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
783  * device flow if no other flow uses it with the same kind of request.
784  *
785  * @param dev
786  *   Pointer to Ethernet device.
787  * @param[in] dev_flow
788  *   Pointer to the device flow.
789  */
790 static void
791 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
792 {
793 	struct mlx5_priv *priv = dev->data->dev_private;
794 	struct rte_flow *flow = dev_flow->flow;
795 	const int mark = !!(dev_flow->actions &
796 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
797 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
798 	unsigned int i;
799 
800 	MLX5_ASSERT(dev->data->dev_started);
801 	for (i = 0; i != flow->rss.queue_num; ++i) {
802 		int idx = (*flow->rss.queue)[i];
803 		struct mlx5_rxq_ctrl *rxq_ctrl =
804 			container_of((*priv->rxqs)[idx],
805 				     struct mlx5_rxq_ctrl, rxq);
806 
807 		if (priv->config.dv_flow_en &&
808 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
809 		    mlx5_flow_ext_mreg_supported(dev)) {
810 			rxq_ctrl->rxq.mark = 1;
811 			rxq_ctrl->flow_mark_n = 1;
812 		} else if (mark) {
813 			rxq_ctrl->flow_mark_n--;
814 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
815 		}
816 		if (tunnel) {
817 			unsigned int j;
818 
819 			/* Decrease the counter matching the flow. */
820 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
821 				if ((tunnels_info[j].tunnel &
822 				     dev_flow->layers) ==
823 				    tunnels_info[j].tunnel) {
824 					rxq_ctrl->flow_tunnels_n[j]--;
825 					break;
826 				}
827 			}
828 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
829 		}
830 	}
831 }
832 
833 /**
834  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
835  * @p flow if no other flow uses it with the same kind of request.
836  *
837  * @param dev
838  *   Pointer to Ethernet device.
839  * @param[in] flow
840  *   Pointer to the flow.
841  */
842 static void
843 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
844 {
845 	struct mlx5_flow *dev_flow;
846 
847 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
848 		flow_drv_rxq_flags_trim(dev, dev_flow);
849 }
850 
851 /**
852  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
853  *
854  * @param dev
855  *   Pointer to Ethernet device.
856  */
857 static void
858 flow_rxq_flags_clear(struct rte_eth_dev *dev)
859 {
860 	struct mlx5_priv *priv = dev->data->dev_private;
861 	unsigned int i;
862 
863 	for (i = 0; i != priv->rxqs_n; ++i) {
864 		struct mlx5_rxq_ctrl *rxq_ctrl;
865 		unsigned int j;
866 
867 		if (!(*priv->rxqs)[i])
868 			continue;
869 		rxq_ctrl = container_of((*priv->rxqs)[i],
870 					struct mlx5_rxq_ctrl, rxq);
871 		rxq_ctrl->flow_mark_n = 0;
872 		rxq_ctrl->rxq.mark = 0;
873 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
874 			rxq_ctrl->flow_tunnels_n[j] = 0;
875 		rxq_ctrl->rxq.tunnel = 0;
876 	}
877 }
878 
879 /*
880  * return a pointer to the desired action in the list of actions.
881  *
882  * @param[in] actions
883  *   The list of actions to search the action in.
884  * @param[in] action
885  *   The action to find.
886  *
887  * @return
888  *   Pointer to the action in the list, if found. NULL otherwise.
889  */
890 const struct rte_flow_action *
891 mlx5_flow_find_action(const struct rte_flow_action *actions,
892 		      enum rte_flow_action_type action)
893 {
894 	if (actions == NULL)
895 		return NULL;
896 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
897 		if (actions->type == action)
898 			return actions;
899 	return NULL;
900 }
901 
902 /*
903  * Validate the flag action.
904  *
905  * @param[in] action_flags
906  *   Bit-fields that holds the actions detected until now.
907  * @param[in] attr
908  *   Attributes of flow that includes this action.
909  * @param[out] error
910  *   Pointer to error structure.
911  *
912  * @return
913  *   0 on success, a negative errno value otherwise and rte_errno is set.
914  */
915 int
916 mlx5_flow_validate_action_flag(uint64_t action_flags,
917 			       const struct rte_flow_attr *attr,
918 			       struct rte_flow_error *error)
919 {
920 	if (action_flags & MLX5_FLOW_ACTION_MARK)
921 		return rte_flow_error_set(error, EINVAL,
922 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
923 					  "can't mark and flag in same flow");
924 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
925 		return rte_flow_error_set(error, EINVAL,
926 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
927 					  "can't have 2 flag"
928 					  " actions in same flow");
929 	if (attr->egress)
930 		return rte_flow_error_set(error, ENOTSUP,
931 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
932 					  "flag action not supported for "
933 					  "egress");
934 	return 0;
935 }
936 
937 /*
938  * Validate the mark action.
939  *
940  * @param[in] action
941  *   Pointer to the queue action.
942  * @param[in] action_flags
943  *   Bit-fields that holds the actions detected until now.
944  * @param[in] attr
945  *   Attributes of flow that includes this action.
946  * @param[out] error
947  *   Pointer to error structure.
948  *
949  * @return
950  *   0 on success, a negative errno value otherwise and rte_errno is set.
951  */
952 int
953 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
954 			       uint64_t action_flags,
955 			       const struct rte_flow_attr *attr,
956 			       struct rte_flow_error *error)
957 {
958 	const struct rte_flow_action_mark *mark = action->conf;
959 
960 	if (!mark)
961 		return rte_flow_error_set(error, EINVAL,
962 					  RTE_FLOW_ERROR_TYPE_ACTION,
963 					  action,
964 					  "configuration cannot be null");
965 	if (mark->id >= MLX5_FLOW_MARK_MAX)
966 		return rte_flow_error_set(error, EINVAL,
967 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
968 					  &mark->id,
969 					  "mark id must in 0 <= id < "
970 					  RTE_STR(MLX5_FLOW_MARK_MAX));
971 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
972 		return rte_flow_error_set(error, EINVAL,
973 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
974 					  "can't flag and mark in same flow");
975 	if (action_flags & MLX5_FLOW_ACTION_MARK)
976 		return rte_flow_error_set(error, EINVAL,
977 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
978 					  "can't have 2 mark actions in same"
979 					  " flow");
980 	if (attr->egress)
981 		return rte_flow_error_set(error, ENOTSUP,
982 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
983 					  "mark action not supported for "
984 					  "egress");
985 	return 0;
986 }
987 
988 /*
989  * Validate the drop action.
990  *
991  * @param[in] action_flags
992  *   Bit-fields that holds the actions detected until now.
993  * @param[in] attr
994  *   Attributes of flow that includes this action.
995  * @param[out] error
996  *   Pointer to error structure.
997  *
998  * @return
999  *   0 on success, a negative errno value otherwise and rte_errno is set.
1000  */
1001 int
1002 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1003 			       const struct rte_flow_attr *attr,
1004 			       struct rte_flow_error *error)
1005 {
1006 	if (attr->egress)
1007 		return rte_flow_error_set(error, ENOTSUP,
1008 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1009 					  "drop action not supported for "
1010 					  "egress");
1011 	return 0;
1012 }
1013 
1014 /*
1015  * Validate the queue action.
1016  *
1017  * @param[in] action
1018  *   Pointer to the queue action.
1019  * @param[in] action_flags
1020  *   Bit-fields that holds the actions detected until now.
1021  * @param[in] dev
1022  *   Pointer to the Ethernet device structure.
1023  * @param[in] attr
1024  *   Attributes of flow that includes this action.
1025  * @param[out] error
1026  *   Pointer to error structure.
1027  *
1028  * @return
1029  *   0 on success, a negative errno value otherwise and rte_errno is set.
1030  */
1031 int
1032 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1033 				uint64_t action_flags,
1034 				struct rte_eth_dev *dev,
1035 				const struct rte_flow_attr *attr,
1036 				struct rte_flow_error *error)
1037 {
1038 	struct mlx5_priv *priv = dev->data->dev_private;
1039 	const struct rte_flow_action_queue *queue = action->conf;
1040 
1041 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1042 		return rte_flow_error_set(error, EINVAL,
1043 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1044 					  "can't have 2 fate actions in"
1045 					  " same flow");
1046 	if (!priv->rxqs_n)
1047 		return rte_flow_error_set(error, EINVAL,
1048 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1049 					  NULL, "No Rx queues configured");
1050 	if (queue->index >= priv->rxqs_n)
1051 		return rte_flow_error_set(error, EINVAL,
1052 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1053 					  &queue->index,
1054 					  "queue index out of range");
1055 	if (!(*priv->rxqs)[queue->index])
1056 		return rte_flow_error_set(error, EINVAL,
1057 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1058 					  &queue->index,
1059 					  "queue is not configured");
1060 	if (attr->egress)
1061 		return rte_flow_error_set(error, ENOTSUP,
1062 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1063 					  "queue action not supported for "
1064 					  "egress");
1065 	return 0;
1066 }
1067 
1068 /*
1069  * Validate the rss action.
1070  *
1071  * @param[in] action
1072  *   Pointer to the queue action.
1073  * @param[in] action_flags
1074  *   Bit-fields that holds the actions detected until now.
1075  * @param[in] dev
1076  *   Pointer to the Ethernet device structure.
1077  * @param[in] attr
1078  *   Attributes of flow that includes this action.
1079  * @param[in] item_flags
1080  *   Items that were detected.
1081  * @param[out] error
1082  *   Pointer to error structure.
1083  *
1084  * @return
1085  *   0 on success, a negative errno value otherwise and rte_errno is set.
1086  */
1087 int
1088 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1089 			      uint64_t action_flags,
1090 			      struct rte_eth_dev *dev,
1091 			      const struct rte_flow_attr *attr,
1092 			      uint64_t item_flags,
1093 			      struct rte_flow_error *error)
1094 {
1095 	struct mlx5_priv *priv = dev->data->dev_private;
1096 	const struct rte_flow_action_rss *rss = action->conf;
1097 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1098 	unsigned int i;
1099 
1100 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1101 		return rte_flow_error_set(error, EINVAL,
1102 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1103 					  "can't have 2 fate actions"
1104 					  " in same flow");
1105 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1106 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1107 		return rte_flow_error_set(error, ENOTSUP,
1108 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1109 					  &rss->func,
1110 					  "RSS hash function not supported");
1111 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1112 	if (rss->level > 2)
1113 #else
1114 	if (rss->level > 1)
1115 #endif
1116 		return rte_flow_error_set(error, ENOTSUP,
1117 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1118 					  &rss->level,
1119 					  "tunnel RSS is not supported");
1120 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1121 	if (rss->key_len == 0 && rss->key != NULL)
1122 		return rte_flow_error_set(error, ENOTSUP,
1123 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1124 					  &rss->key_len,
1125 					  "RSS hash key length 0");
1126 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1127 		return rte_flow_error_set(error, ENOTSUP,
1128 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1129 					  &rss->key_len,
1130 					  "RSS hash key too small");
1131 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1132 		return rte_flow_error_set(error, ENOTSUP,
1133 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1134 					  &rss->key_len,
1135 					  "RSS hash key too large");
1136 	if (rss->queue_num > priv->config.ind_table_max_size)
1137 		return rte_flow_error_set(error, ENOTSUP,
1138 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1139 					  &rss->queue_num,
1140 					  "number of queues too large");
1141 	if (rss->types & MLX5_RSS_HF_MASK)
1142 		return rte_flow_error_set(error, ENOTSUP,
1143 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1144 					  &rss->types,
1145 					  "some RSS protocols are not"
1146 					  " supported");
1147 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1148 	    !(rss->types & ETH_RSS_IP))
1149 		return rte_flow_error_set(error, EINVAL,
1150 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1151 					  "L3 partial RSS requested but L3 RSS"
1152 					  " type not specified");
1153 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1154 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1155 		return rte_flow_error_set(error, EINVAL,
1156 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1157 					  "L4 partial RSS requested but L4 RSS"
1158 					  " type not specified");
1159 	if (!priv->rxqs_n)
1160 		return rte_flow_error_set(error, EINVAL,
1161 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1162 					  NULL, "No Rx queues configured");
1163 	if (!rss->queue_num)
1164 		return rte_flow_error_set(error, EINVAL,
1165 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1166 					  NULL, "No queues configured");
1167 	for (i = 0; i != rss->queue_num; ++i) {
1168 		if (rss->queue[i] >= priv->rxqs_n)
1169 			return rte_flow_error_set
1170 				(error, EINVAL,
1171 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1172 				 &rss->queue[i], "queue index out of range");
1173 		if (!(*priv->rxqs)[rss->queue[i]])
1174 			return rte_flow_error_set
1175 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1176 				 &rss->queue[i], "queue is not configured");
1177 	}
1178 	if (attr->egress)
1179 		return rte_flow_error_set(error, ENOTSUP,
1180 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1181 					  "rss action not supported for "
1182 					  "egress");
1183 	if (rss->level > 1 &&  !tunnel)
1184 		return rte_flow_error_set(error, EINVAL,
1185 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1186 					  "inner RSS is not supported for "
1187 					  "non-tunnel flows");
1188 	return 0;
1189 }
1190 
1191 /*
1192  * Validate the count action.
1193  *
1194  * @param[in] dev
1195  *   Pointer to the Ethernet device structure.
1196  * @param[in] attr
1197  *   Attributes of flow that includes this action.
1198  * @param[out] error
1199  *   Pointer to error structure.
1200  *
1201  * @return
1202  *   0 on success, a negative errno value otherwise and rte_errno is set.
1203  */
1204 int
1205 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1206 				const struct rte_flow_attr *attr,
1207 				struct rte_flow_error *error)
1208 {
1209 	if (attr->egress)
1210 		return rte_flow_error_set(error, ENOTSUP,
1211 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1212 					  "count action not supported for "
1213 					  "egress");
1214 	return 0;
1215 }
1216 
1217 /**
1218  * Verify the @p attributes will be correctly understood by the NIC and store
1219  * them in the @p flow if everything is correct.
1220  *
1221  * @param[in] dev
1222  *   Pointer to the Ethernet device structure.
1223  * @param[in] attributes
1224  *   Pointer to flow attributes
1225  * @param[out] error
1226  *   Pointer to error structure.
1227  *
1228  * @return
1229  *   0 on success, a negative errno value otherwise and rte_errno is set.
1230  */
1231 int
1232 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1233 			      const struct rte_flow_attr *attributes,
1234 			      struct rte_flow_error *error)
1235 {
1236 	struct mlx5_priv *priv = dev->data->dev_private;
1237 	uint32_t priority_max = priv->config.flow_prio - 1;
1238 
1239 	if (attributes->group)
1240 		return rte_flow_error_set(error, ENOTSUP,
1241 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1242 					  NULL, "groups is not supported");
1243 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1244 	    attributes->priority >= priority_max)
1245 		return rte_flow_error_set(error, ENOTSUP,
1246 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1247 					  NULL, "priority out of range");
1248 	if (attributes->egress)
1249 		return rte_flow_error_set(error, ENOTSUP,
1250 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1251 					  "egress is not supported");
1252 	if (attributes->transfer && !priv->config.dv_esw_en)
1253 		return rte_flow_error_set(error, ENOTSUP,
1254 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1255 					  NULL, "transfer is not supported");
1256 	if (!attributes->ingress)
1257 		return rte_flow_error_set(error, EINVAL,
1258 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1259 					  NULL,
1260 					  "ingress attribute is mandatory");
1261 	return 0;
1262 }
1263 
1264 /**
1265  * Validate ICMP6 item.
1266  *
1267  * @param[in] item
1268  *   Item specification.
1269  * @param[in] item_flags
1270  *   Bit-fields that holds the items detected until now.
1271  * @param[out] error
1272  *   Pointer to error structure.
1273  *
1274  * @return
1275  *   0 on success, a negative errno value otherwise and rte_errno is set.
1276  */
1277 int
1278 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1279 			       uint64_t item_flags,
1280 			       uint8_t target_protocol,
1281 			       struct rte_flow_error *error)
1282 {
1283 	const struct rte_flow_item_icmp6 *mask = item->mask;
1284 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1285 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1286 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1287 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1288 				      MLX5_FLOW_LAYER_OUTER_L4;
1289 	int ret;
1290 
1291 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1292 		return rte_flow_error_set(error, EINVAL,
1293 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1294 					  "protocol filtering not compatible"
1295 					  " with ICMP6 layer");
1296 	if (!(item_flags & l3m))
1297 		return rte_flow_error_set(error, EINVAL,
1298 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1299 					  "IPv6 is mandatory to filter on"
1300 					  " ICMP6");
1301 	if (item_flags & l4m)
1302 		return rte_flow_error_set(error, EINVAL,
1303 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1304 					  "multiple L4 layers not supported");
1305 	if (!mask)
1306 		mask = &rte_flow_item_icmp6_mask;
1307 	ret = mlx5_flow_item_acceptable
1308 		(item, (const uint8_t *)mask,
1309 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1310 		 sizeof(struct rte_flow_item_icmp6), error);
1311 	if (ret < 0)
1312 		return ret;
1313 	return 0;
1314 }
1315 
1316 /**
1317  * Validate ICMP item.
1318  *
1319  * @param[in] item
1320  *   Item specification.
1321  * @param[in] item_flags
1322  *   Bit-fields that holds the items detected until now.
1323  * @param[out] error
1324  *   Pointer to error structure.
1325  *
1326  * @return
1327  *   0 on success, a negative errno value otherwise and rte_errno is set.
1328  */
1329 int
1330 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1331 			     uint64_t item_flags,
1332 			     uint8_t target_protocol,
1333 			     struct rte_flow_error *error)
1334 {
1335 	const struct rte_flow_item_icmp *mask = item->mask;
1336 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1337 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1338 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1339 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1340 				      MLX5_FLOW_LAYER_OUTER_L4;
1341 	int ret;
1342 
1343 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1344 		return rte_flow_error_set(error, EINVAL,
1345 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1346 					  "protocol filtering not compatible"
1347 					  " with ICMP layer");
1348 	if (!(item_flags & l3m))
1349 		return rte_flow_error_set(error, EINVAL,
1350 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1351 					  "IPv4 is mandatory to filter"
1352 					  " on ICMP");
1353 	if (item_flags & l4m)
1354 		return rte_flow_error_set(error, EINVAL,
1355 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1356 					  "multiple L4 layers not supported");
1357 	if (!mask)
1358 		mask = &rte_flow_item_icmp_mask;
1359 	ret = mlx5_flow_item_acceptable
1360 		(item, (const uint8_t *)mask,
1361 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1362 		 sizeof(struct rte_flow_item_icmp), error);
1363 	if (ret < 0)
1364 		return ret;
1365 	return 0;
1366 }
1367 
1368 /**
1369  * Validate Ethernet item.
1370  *
1371  * @param[in] item
1372  *   Item specification.
1373  * @param[in] item_flags
1374  *   Bit-fields that holds the items detected until now.
1375  * @param[out] error
1376  *   Pointer to error structure.
1377  *
1378  * @return
1379  *   0 on success, a negative errno value otherwise and rte_errno is set.
1380  */
1381 int
1382 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1383 			    uint64_t item_flags,
1384 			    struct rte_flow_error *error)
1385 {
1386 	const struct rte_flow_item_eth *mask = item->mask;
1387 	const struct rte_flow_item_eth nic_mask = {
1388 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1389 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1390 		.type = RTE_BE16(0xffff),
1391 	};
1392 	int ret;
1393 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1394 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1395 				       MLX5_FLOW_LAYER_OUTER_L2;
1396 
1397 	if (item_flags & ethm)
1398 		return rte_flow_error_set(error, ENOTSUP,
1399 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1400 					  "multiple L2 layers not supported");
1401 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1402 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1403 		return rte_flow_error_set(error, EINVAL,
1404 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1405 					  "L2 layer should not follow "
1406 					  "L3 layers");
1407 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1408 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1409 		return rte_flow_error_set(error, EINVAL,
1410 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1411 					  "L2 layer should not follow VLAN");
1412 	if (!mask)
1413 		mask = &rte_flow_item_eth_mask;
1414 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1415 					(const uint8_t *)&nic_mask,
1416 					sizeof(struct rte_flow_item_eth),
1417 					error);
1418 	return ret;
1419 }
1420 
1421 /**
1422  * Validate VLAN item.
1423  *
1424  * @param[in] item
1425  *   Item specification.
1426  * @param[in] item_flags
1427  *   Bit-fields that holds the items detected until now.
1428  * @param[in] dev
1429  *   Ethernet device flow is being created on.
1430  * @param[out] error
1431  *   Pointer to error structure.
1432  *
1433  * @return
1434  *   0 on success, a negative errno value otherwise and rte_errno is set.
1435  */
1436 int
1437 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1438 			     uint64_t item_flags,
1439 			     struct rte_eth_dev *dev,
1440 			     struct rte_flow_error *error)
1441 {
1442 	const struct rte_flow_item_vlan *spec = item->spec;
1443 	const struct rte_flow_item_vlan *mask = item->mask;
1444 	const struct rte_flow_item_vlan nic_mask = {
1445 		.tci = RTE_BE16(UINT16_MAX),
1446 		.inner_type = RTE_BE16(UINT16_MAX),
1447 	};
1448 	uint16_t vlan_tag = 0;
1449 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1450 	int ret;
1451 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1452 					MLX5_FLOW_LAYER_INNER_L4) :
1453 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1454 					MLX5_FLOW_LAYER_OUTER_L4);
1455 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1456 					MLX5_FLOW_LAYER_OUTER_VLAN;
1457 
1458 	if (item_flags & vlanm)
1459 		return rte_flow_error_set(error, EINVAL,
1460 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1461 					  "multiple VLAN layers not supported");
1462 	else if ((item_flags & l34m) != 0)
1463 		return rte_flow_error_set(error, EINVAL,
1464 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1465 					  "VLAN cannot follow L3/L4 layer");
1466 	if (!mask)
1467 		mask = &rte_flow_item_vlan_mask;
1468 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1469 					(const uint8_t *)&nic_mask,
1470 					sizeof(struct rte_flow_item_vlan),
1471 					error);
1472 	if (ret)
1473 		return ret;
1474 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1475 		struct mlx5_priv *priv = dev->data->dev_private;
1476 
1477 		if (priv->vmwa_context) {
1478 			/*
1479 			 * Non-NULL context means we have a virtual machine
1480 			 * and SR-IOV enabled, we have to create VLAN interface
1481 			 * to make hypervisor to setup E-Switch vport
1482 			 * context correctly. We avoid creating the multiple
1483 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1484 			 */
1485 			return rte_flow_error_set(error, EINVAL,
1486 						  RTE_FLOW_ERROR_TYPE_ITEM,
1487 						  item,
1488 						  "VLAN tag mask is not"
1489 						  " supported in virtual"
1490 						  " environment");
1491 		}
1492 	}
1493 	if (spec) {
1494 		vlan_tag = spec->tci;
1495 		vlan_tag &= mask->tci;
1496 	}
1497 	/*
1498 	 * From verbs perspective an empty VLAN is equivalent
1499 	 * to a packet without VLAN layer.
1500 	 */
1501 	if (!vlan_tag)
1502 		return rte_flow_error_set(error, EINVAL,
1503 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1504 					  item->spec,
1505 					  "VLAN cannot be empty");
1506 	return 0;
1507 }
1508 
1509 /**
1510  * Validate IPV4 item.
1511  *
1512  * @param[in] item
1513  *   Item specification.
1514  * @param[in] item_flags
1515  *   Bit-fields that holds the items detected until now.
1516  * @param[in] acc_mask
1517  *   Acceptable mask, if NULL default internal default mask
1518  *   will be used to check whether item fields are supported.
1519  * @param[out] error
1520  *   Pointer to error structure.
1521  *
1522  * @return
1523  *   0 on success, a negative errno value otherwise and rte_errno is set.
1524  */
1525 int
1526 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1527 			     uint64_t item_flags,
1528 			     uint64_t last_item,
1529 			     uint16_t ether_type,
1530 			     const struct rte_flow_item_ipv4 *acc_mask,
1531 			     struct rte_flow_error *error)
1532 {
1533 	const struct rte_flow_item_ipv4 *mask = item->mask;
1534 	const struct rte_flow_item_ipv4 *spec = item->spec;
1535 	const struct rte_flow_item_ipv4 nic_mask = {
1536 		.hdr = {
1537 			.src_addr = RTE_BE32(0xffffffff),
1538 			.dst_addr = RTE_BE32(0xffffffff),
1539 			.type_of_service = 0xff,
1540 			.next_proto_id = 0xff,
1541 		},
1542 	};
1543 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1544 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1545 				      MLX5_FLOW_LAYER_OUTER_L3;
1546 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1547 				      MLX5_FLOW_LAYER_OUTER_L4;
1548 	int ret;
1549 	uint8_t next_proto = 0xFF;
1550 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1551 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1552 				  MLX5_FLOW_LAYER_INNER_VLAN);
1553 
1554 	if ((last_item & l2_vlan) && ether_type &&
1555 	    ether_type != RTE_ETHER_TYPE_IPV4)
1556 		return rte_flow_error_set(error, EINVAL,
1557 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1558 					  "IPv4 cannot follow L2/VLAN layer "
1559 					  "which ether type is not IPv4");
1560 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1561 		if (mask && spec)
1562 			next_proto = mask->hdr.next_proto_id &
1563 				     spec->hdr.next_proto_id;
1564 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1565 			return rte_flow_error_set(error, EINVAL,
1566 						  RTE_FLOW_ERROR_TYPE_ITEM,
1567 						  item,
1568 						  "multiple tunnel "
1569 						  "not supported");
1570 	}
1571 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1572 		return rte_flow_error_set(error, EINVAL,
1573 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1574 					  "wrong tunnel type - IPv6 specified "
1575 					  "but IPv4 item provided");
1576 	if (item_flags & l3m)
1577 		return rte_flow_error_set(error, ENOTSUP,
1578 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1579 					  "multiple L3 layers not supported");
1580 	else if (item_flags & l4m)
1581 		return rte_flow_error_set(error, EINVAL,
1582 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1583 					  "L3 cannot follow an L4 layer.");
1584 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1585 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1586 		return rte_flow_error_set(error, EINVAL,
1587 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1588 					  "L3 cannot follow an NVGRE layer.");
1589 	if (!mask)
1590 		mask = &rte_flow_item_ipv4_mask;
1591 	else if (mask->hdr.next_proto_id != 0 &&
1592 		 mask->hdr.next_proto_id != 0xff)
1593 		return rte_flow_error_set(error, EINVAL,
1594 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1595 					  "partial mask is not supported"
1596 					  " for protocol");
1597 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1598 					acc_mask ? (const uint8_t *)acc_mask
1599 						 : (const uint8_t *)&nic_mask,
1600 					sizeof(struct rte_flow_item_ipv4),
1601 					error);
1602 	if (ret < 0)
1603 		return ret;
1604 	return 0;
1605 }
1606 
1607 /**
1608  * Validate IPV6 item.
1609  *
1610  * @param[in] item
1611  *   Item specification.
1612  * @param[in] item_flags
1613  *   Bit-fields that holds the items detected until now.
1614  * @param[in] acc_mask
1615  *   Acceptable mask, if NULL default internal default mask
1616  *   will be used to check whether item fields are supported.
1617  * @param[out] error
1618  *   Pointer to error structure.
1619  *
1620  * @return
1621  *   0 on success, a negative errno value otherwise and rte_errno is set.
1622  */
1623 int
1624 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1625 			     uint64_t item_flags,
1626 			     uint64_t last_item,
1627 			     uint16_t ether_type,
1628 			     const struct rte_flow_item_ipv6 *acc_mask,
1629 			     struct rte_flow_error *error)
1630 {
1631 	const struct rte_flow_item_ipv6 *mask = item->mask;
1632 	const struct rte_flow_item_ipv6 *spec = item->spec;
1633 	const struct rte_flow_item_ipv6 nic_mask = {
1634 		.hdr = {
1635 			.src_addr =
1636 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1637 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1638 			.dst_addr =
1639 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1640 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1641 			.vtc_flow = RTE_BE32(0xffffffff),
1642 			.proto = 0xff,
1643 		},
1644 	};
1645 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1646 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1647 				      MLX5_FLOW_LAYER_OUTER_L3;
1648 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1649 				      MLX5_FLOW_LAYER_OUTER_L4;
1650 	int ret;
1651 	uint8_t next_proto = 0xFF;
1652 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1653 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1654 				  MLX5_FLOW_LAYER_INNER_VLAN);
1655 
1656 	if ((last_item & l2_vlan) && ether_type &&
1657 	    ether_type != RTE_ETHER_TYPE_IPV6)
1658 		return rte_flow_error_set(error, EINVAL,
1659 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1660 					  "IPv6 cannot follow L2/VLAN layer "
1661 					  "which ether type is not IPv6");
1662 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1663 		if (mask && spec)
1664 			next_proto = mask->hdr.proto & spec->hdr.proto;
1665 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1666 			return rte_flow_error_set(error, EINVAL,
1667 						  RTE_FLOW_ERROR_TYPE_ITEM,
1668 						  item,
1669 						  "multiple tunnel "
1670 						  "not supported");
1671 	}
1672 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1673 		return rte_flow_error_set(error, EINVAL,
1674 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1675 					  "wrong tunnel type - IPv4 specified "
1676 					  "but IPv6 item provided");
1677 	if (item_flags & l3m)
1678 		return rte_flow_error_set(error, ENOTSUP,
1679 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1680 					  "multiple L3 layers not supported");
1681 	else if (item_flags & l4m)
1682 		return rte_flow_error_set(error, EINVAL,
1683 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1684 					  "L3 cannot follow an L4 layer.");
1685 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1686 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1687 		return rte_flow_error_set(error, EINVAL,
1688 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1689 					  "L3 cannot follow an NVGRE layer.");
1690 	if (!mask)
1691 		mask = &rte_flow_item_ipv6_mask;
1692 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1693 					acc_mask ? (const uint8_t *)acc_mask
1694 						 : (const uint8_t *)&nic_mask,
1695 					sizeof(struct rte_flow_item_ipv6),
1696 					error);
1697 	if (ret < 0)
1698 		return ret;
1699 	return 0;
1700 }
1701 
1702 /**
1703  * Validate UDP item.
1704  *
1705  * @param[in] item
1706  *   Item specification.
1707  * @param[in] item_flags
1708  *   Bit-fields that holds the items detected until now.
1709  * @param[in] target_protocol
1710  *   The next protocol in the previous item.
1711  * @param[in] flow_mask
1712  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1713  * @param[out] error
1714  *   Pointer to error structure.
1715  *
1716  * @return
1717  *   0 on success, a negative errno value otherwise and rte_errno is set.
1718  */
1719 int
1720 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1721 			    uint64_t item_flags,
1722 			    uint8_t target_protocol,
1723 			    struct rte_flow_error *error)
1724 {
1725 	const struct rte_flow_item_udp *mask = item->mask;
1726 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1727 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1728 				      MLX5_FLOW_LAYER_OUTER_L3;
1729 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1730 				      MLX5_FLOW_LAYER_OUTER_L4;
1731 	int ret;
1732 
1733 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1734 		return rte_flow_error_set(error, EINVAL,
1735 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1736 					  "protocol filtering not compatible"
1737 					  " with UDP layer");
1738 	if (!(item_flags & l3m))
1739 		return rte_flow_error_set(error, EINVAL,
1740 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1741 					  "L3 is mandatory to filter on L4");
1742 	if (item_flags & l4m)
1743 		return rte_flow_error_set(error, EINVAL,
1744 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1745 					  "multiple L4 layers not supported");
1746 	if (!mask)
1747 		mask = &rte_flow_item_udp_mask;
1748 	ret = mlx5_flow_item_acceptable
1749 		(item, (const uint8_t *)mask,
1750 		 (const uint8_t *)&rte_flow_item_udp_mask,
1751 		 sizeof(struct rte_flow_item_udp), error);
1752 	if (ret < 0)
1753 		return ret;
1754 	return 0;
1755 }
1756 
1757 /**
1758  * Validate TCP item.
1759  *
1760  * @param[in] item
1761  *   Item specification.
1762  * @param[in] item_flags
1763  *   Bit-fields that holds the items detected until now.
1764  * @param[in] target_protocol
1765  *   The next protocol in the previous item.
1766  * @param[out] error
1767  *   Pointer to error structure.
1768  *
1769  * @return
1770  *   0 on success, a negative errno value otherwise and rte_errno is set.
1771  */
1772 int
1773 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1774 			    uint64_t item_flags,
1775 			    uint8_t target_protocol,
1776 			    const struct rte_flow_item_tcp *flow_mask,
1777 			    struct rte_flow_error *error)
1778 {
1779 	const struct rte_flow_item_tcp *mask = item->mask;
1780 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1781 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1782 				      MLX5_FLOW_LAYER_OUTER_L3;
1783 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1784 				      MLX5_FLOW_LAYER_OUTER_L4;
1785 	int ret;
1786 
1787 	MLX5_ASSERT(flow_mask);
1788 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1789 		return rte_flow_error_set(error, EINVAL,
1790 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1791 					  "protocol filtering not compatible"
1792 					  " with TCP layer");
1793 	if (!(item_flags & l3m))
1794 		return rte_flow_error_set(error, EINVAL,
1795 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1796 					  "L3 is mandatory to filter on L4");
1797 	if (item_flags & l4m)
1798 		return rte_flow_error_set(error, EINVAL,
1799 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1800 					  "multiple L4 layers not supported");
1801 	if (!mask)
1802 		mask = &rte_flow_item_tcp_mask;
1803 	ret = mlx5_flow_item_acceptable
1804 		(item, (const uint8_t *)mask,
1805 		 (const uint8_t *)flow_mask,
1806 		 sizeof(struct rte_flow_item_tcp), error);
1807 	if (ret < 0)
1808 		return ret;
1809 	return 0;
1810 }
1811 
1812 /**
1813  * Validate VXLAN item.
1814  *
1815  * @param[in] item
1816  *   Item specification.
1817  * @param[in] item_flags
1818  *   Bit-fields that holds the items detected until now.
1819  * @param[in] target_protocol
1820  *   The next protocol in the previous item.
1821  * @param[out] error
1822  *   Pointer to error structure.
1823  *
1824  * @return
1825  *   0 on success, a negative errno value otherwise and rte_errno is set.
1826  */
1827 int
1828 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1829 			      uint64_t item_flags,
1830 			      struct rte_flow_error *error)
1831 {
1832 	const struct rte_flow_item_vxlan *spec = item->spec;
1833 	const struct rte_flow_item_vxlan *mask = item->mask;
1834 	int ret;
1835 	union vni {
1836 		uint32_t vlan_id;
1837 		uint8_t vni[4];
1838 	} id = { .vlan_id = 0, };
1839 	uint32_t vlan_id = 0;
1840 
1841 
1842 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1843 		return rte_flow_error_set(error, ENOTSUP,
1844 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1845 					  "multiple tunnel layers not"
1846 					  " supported");
1847 	/*
1848 	 * Verify only UDPv4 is present as defined in
1849 	 * https://tools.ietf.org/html/rfc7348
1850 	 */
1851 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1852 		return rte_flow_error_set(error, EINVAL,
1853 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1854 					  "no outer UDP layer found");
1855 	if (!mask)
1856 		mask = &rte_flow_item_vxlan_mask;
1857 	ret = mlx5_flow_item_acceptable
1858 		(item, (const uint8_t *)mask,
1859 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1860 		 sizeof(struct rte_flow_item_vxlan),
1861 		 error);
1862 	if (ret < 0)
1863 		return ret;
1864 	if (spec) {
1865 		memcpy(&id.vni[1], spec->vni, 3);
1866 		vlan_id = id.vlan_id;
1867 		memcpy(&id.vni[1], mask->vni, 3);
1868 		vlan_id &= id.vlan_id;
1869 	}
1870 	/*
1871 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1872 	 * only this layer is defined in the Verbs specification it is
1873 	 * interpreted as wildcard and all packets will match this
1874 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1875 	 * udp), all packets matching the layers before will also
1876 	 * match this rule.  To avoid such situation, VNI 0 is
1877 	 * currently refused.
1878 	 */
1879 	if (!vlan_id)
1880 		return rte_flow_error_set(error, ENOTSUP,
1881 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1882 					  "VXLAN vni cannot be 0");
1883 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1884 		return rte_flow_error_set(error, ENOTSUP,
1885 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1886 					  "VXLAN tunnel must be fully defined");
1887 	return 0;
1888 }
1889 
1890 /**
1891  * Validate VXLAN_GPE item.
1892  *
1893  * @param[in] item
1894  *   Item specification.
1895  * @param[in] item_flags
1896  *   Bit-fields that holds the items detected until now.
1897  * @param[in] priv
1898  *   Pointer to the private data structure.
1899  * @param[in] target_protocol
1900  *   The next protocol in the previous item.
1901  * @param[out] error
1902  *   Pointer to error structure.
1903  *
1904  * @return
1905  *   0 on success, a negative errno value otherwise and rte_errno is set.
1906  */
1907 int
1908 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1909 				  uint64_t item_flags,
1910 				  struct rte_eth_dev *dev,
1911 				  struct rte_flow_error *error)
1912 {
1913 	struct mlx5_priv *priv = dev->data->dev_private;
1914 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1915 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1916 	int ret;
1917 	union vni {
1918 		uint32_t vlan_id;
1919 		uint8_t vni[4];
1920 	} id = { .vlan_id = 0, };
1921 	uint32_t vlan_id = 0;
1922 
1923 	if (!priv->config.l3_vxlan_en)
1924 		return rte_flow_error_set(error, ENOTSUP,
1925 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1926 					  "L3 VXLAN is not enabled by device"
1927 					  " parameter and/or not configured in"
1928 					  " firmware");
1929 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1930 		return rte_flow_error_set(error, ENOTSUP,
1931 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1932 					  "multiple tunnel layers not"
1933 					  " supported");
1934 	/*
1935 	 * Verify only UDPv4 is present as defined in
1936 	 * https://tools.ietf.org/html/rfc7348
1937 	 */
1938 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1939 		return rte_flow_error_set(error, EINVAL,
1940 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1941 					  "no outer UDP layer found");
1942 	if (!mask)
1943 		mask = &rte_flow_item_vxlan_gpe_mask;
1944 	ret = mlx5_flow_item_acceptable
1945 		(item, (const uint8_t *)mask,
1946 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1947 		 sizeof(struct rte_flow_item_vxlan_gpe),
1948 		 error);
1949 	if (ret < 0)
1950 		return ret;
1951 	if (spec) {
1952 		if (spec->protocol)
1953 			return rte_flow_error_set(error, ENOTSUP,
1954 						  RTE_FLOW_ERROR_TYPE_ITEM,
1955 						  item,
1956 						  "VxLAN-GPE protocol"
1957 						  " not supported");
1958 		memcpy(&id.vni[1], spec->vni, 3);
1959 		vlan_id = id.vlan_id;
1960 		memcpy(&id.vni[1], mask->vni, 3);
1961 		vlan_id &= id.vlan_id;
1962 	}
1963 	/*
1964 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1965 	 * layer is defined in the Verbs specification it is interpreted as
1966 	 * wildcard and all packets will match this rule, if it follows a full
1967 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1968 	 * before will also match this rule.  To avoid such situation, VNI 0
1969 	 * is currently refused.
1970 	 */
1971 	if (!vlan_id)
1972 		return rte_flow_error_set(error, ENOTSUP,
1973 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1974 					  "VXLAN-GPE vni cannot be 0");
1975 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1976 		return rte_flow_error_set(error, ENOTSUP,
1977 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1978 					  "VXLAN-GPE tunnel must be fully"
1979 					  " defined");
1980 	return 0;
1981 }
1982 /**
1983  * Validate GRE Key item.
1984  *
1985  * @param[in] item
1986  *   Item specification.
1987  * @param[in] item_flags
1988  *   Bit flags to mark detected items.
1989  * @param[in] gre_item
1990  *   Pointer to gre_item
1991  * @param[out] error
1992  *   Pointer to error structure.
1993  *
1994  * @return
1995  *   0 on success, a negative errno value otherwise and rte_errno is set.
1996  */
1997 int
1998 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1999 				uint64_t item_flags,
2000 				const struct rte_flow_item *gre_item,
2001 				struct rte_flow_error *error)
2002 {
2003 	const rte_be32_t *mask = item->mask;
2004 	int ret = 0;
2005 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2006 	const struct rte_flow_item_gre *gre_spec;
2007 	const struct rte_flow_item_gre *gre_mask;
2008 
2009 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2010 		return rte_flow_error_set(error, ENOTSUP,
2011 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2012 					  "Multiple GRE key not support");
2013 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2014 		return rte_flow_error_set(error, ENOTSUP,
2015 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2016 					  "No preceding GRE header");
2017 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2018 		return rte_flow_error_set(error, ENOTSUP,
2019 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2020 					  "GRE key following a wrong item");
2021 	gre_mask = gre_item->mask;
2022 	if (!gre_mask)
2023 		gre_mask = &rte_flow_item_gre_mask;
2024 	gre_spec = gre_item->spec;
2025 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2026 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2027 		return rte_flow_error_set(error, EINVAL,
2028 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2029 					  "Key bit must be on");
2030 
2031 	if (!mask)
2032 		mask = &gre_key_default_mask;
2033 	ret = mlx5_flow_item_acceptable
2034 		(item, (const uint8_t *)mask,
2035 		 (const uint8_t *)&gre_key_default_mask,
2036 		 sizeof(rte_be32_t), error);
2037 	return ret;
2038 }
2039 
2040 /**
2041  * Validate GRE item.
2042  *
2043  * @param[in] item
2044  *   Item specification.
2045  * @param[in] item_flags
2046  *   Bit flags to mark detected items.
2047  * @param[in] target_protocol
2048  *   The next protocol in the previous item.
2049  * @param[out] error
2050  *   Pointer to error structure.
2051  *
2052  * @return
2053  *   0 on success, a negative errno value otherwise and rte_errno is set.
2054  */
2055 int
2056 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2057 			    uint64_t item_flags,
2058 			    uint8_t target_protocol,
2059 			    struct rte_flow_error *error)
2060 {
2061 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2062 	const struct rte_flow_item_gre *mask = item->mask;
2063 	int ret;
2064 	const struct rte_flow_item_gre nic_mask = {
2065 		.c_rsvd0_ver = RTE_BE16(0xB000),
2066 		.protocol = RTE_BE16(UINT16_MAX),
2067 	};
2068 
2069 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2070 		return rte_flow_error_set(error, EINVAL,
2071 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2072 					  "protocol filtering not compatible"
2073 					  " with this GRE layer");
2074 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2075 		return rte_flow_error_set(error, ENOTSUP,
2076 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2077 					  "multiple tunnel layers not"
2078 					  " supported");
2079 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2080 		return rte_flow_error_set(error, ENOTSUP,
2081 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2082 					  "L3 Layer is missing");
2083 	if (!mask)
2084 		mask = &rte_flow_item_gre_mask;
2085 	ret = mlx5_flow_item_acceptable
2086 		(item, (const uint8_t *)mask,
2087 		 (const uint8_t *)&nic_mask,
2088 		 sizeof(struct rte_flow_item_gre), error);
2089 	if (ret < 0)
2090 		return ret;
2091 #ifndef HAVE_MLX5DV_DR
2092 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2093 	if (spec && (spec->protocol & mask->protocol))
2094 		return rte_flow_error_set(error, ENOTSUP,
2095 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2096 					  "without MPLS support the"
2097 					  " specification cannot be used for"
2098 					  " filtering");
2099 #endif
2100 #endif
2101 	return 0;
2102 }
2103 
2104 /**
2105  * Validate Geneve item.
2106  *
2107  * @param[in] item
2108  *   Item specification.
2109  * @param[in] itemFlags
2110  *   Bit-fields that holds the items detected until now.
2111  * @param[in] enPriv
2112  *   Pointer to the private data structure.
2113  * @param[out] error
2114  *   Pointer to error structure.
2115  *
2116  * @return
2117  *   0 on success, a negative errno value otherwise and rte_errno is set.
2118  */
2119 
2120 int
2121 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2122 			       uint64_t item_flags,
2123 			       struct rte_eth_dev *dev,
2124 			       struct rte_flow_error *error)
2125 {
2126 	struct mlx5_priv *priv = dev->data->dev_private;
2127 	const struct rte_flow_item_geneve *spec = item->spec;
2128 	const struct rte_flow_item_geneve *mask = item->mask;
2129 	int ret;
2130 	uint16_t gbhdr;
2131 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2132 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2133 	const struct rte_flow_item_geneve nic_mask = {
2134 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2135 		.vni = "\xff\xff\xff",
2136 		.protocol = RTE_BE16(UINT16_MAX),
2137 	};
2138 
2139 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2140 		return rte_flow_error_set(error, ENOTSUP,
2141 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2142 					  "L3 Geneve is not enabled by device"
2143 					  " parameter and/or not configured in"
2144 					  " firmware");
2145 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2146 		return rte_flow_error_set(error, ENOTSUP,
2147 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2148 					  "multiple tunnel layers not"
2149 					  " supported");
2150 	/*
2151 	 * Verify only UDPv4 is present as defined in
2152 	 * https://tools.ietf.org/html/rfc7348
2153 	 */
2154 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2155 		return rte_flow_error_set(error, EINVAL,
2156 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2157 					  "no outer UDP layer found");
2158 	if (!mask)
2159 		mask = &rte_flow_item_geneve_mask;
2160 	ret = mlx5_flow_item_acceptable
2161 				  (item, (const uint8_t *)mask,
2162 				   (const uint8_t *)&nic_mask,
2163 				   sizeof(struct rte_flow_item_geneve), error);
2164 	if (ret)
2165 		return ret;
2166 	if (spec) {
2167 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2168 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2169 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2170 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2171 			return rte_flow_error_set(error, ENOTSUP,
2172 						  RTE_FLOW_ERROR_TYPE_ITEM,
2173 						  item,
2174 						  "Geneve protocol unsupported"
2175 						  " fields are being used");
2176 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2177 			return rte_flow_error_set
2178 					(error, ENOTSUP,
2179 					 RTE_FLOW_ERROR_TYPE_ITEM,
2180 					 item,
2181 					 "Unsupported Geneve options length");
2182 	}
2183 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2184 		return rte_flow_error_set
2185 				    (error, ENOTSUP,
2186 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2187 				     "Geneve tunnel must be fully defined");
2188 	return 0;
2189 }
2190 
2191 /**
2192  * Validate MPLS item.
2193  *
2194  * @param[in] dev
2195  *   Pointer to the rte_eth_dev structure.
2196  * @param[in] item
2197  *   Item specification.
2198  * @param[in] item_flags
2199  *   Bit-fields that holds the items detected until now.
2200  * @param[in] prev_layer
2201  *   The protocol layer indicated in previous item.
2202  * @param[out] error
2203  *   Pointer to error structure.
2204  *
2205  * @return
2206  *   0 on success, a negative errno value otherwise and rte_errno is set.
2207  */
2208 int
2209 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2210 			     const struct rte_flow_item *item __rte_unused,
2211 			     uint64_t item_flags __rte_unused,
2212 			     uint64_t prev_layer __rte_unused,
2213 			     struct rte_flow_error *error)
2214 {
2215 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2216 	const struct rte_flow_item_mpls *mask = item->mask;
2217 	struct mlx5_priv *priv = dev->data->dev_private;
2218 	int ret;
2219 
2220 	if (!priv->config.mpls_en)
2221 		return rte_flow_error_set(error, ENOTSUP,
2222 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2223 					  "MPLS not supported or"
2224 					  " disabled in firmware"
2225 					  " configuration.");
2226 	/* MPLS over IP, UDP, GRE is allowed */
2227 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2228 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2229 			    MLX5_FLOW_LAYER_GRE)))
2230 		return rte_flow_error_set(error, EINVAL,
2231 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2232 					  "protocol filtering not compatible"
2233 					  " with MPLS layer");
2234 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2235 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2236 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2237 		return rte_flow_error_set(error, ENOTSUP,
2238 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2239 					  "multiple tunnel layers not"
2240 					  " supported");
2241 	if (!mask)
2242 		mask = &rte_flow_item_mpls_mask;
2243 	ret = mlx5_flow_item_acceptable
2244 		(item, (const uint8_t *)mask,
2245 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2246 		 sizeof(struct rte_flow_item_mpls), error);
2247 	if (ret < 0)
2248 		return ret;
2249 	return 0;
2250 #endif
2251 	return rte_flow_error_set(error, ENOTSUP,
2252 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2253 				  "MPLS is not supported by Verbs, please"
2254 				  " update.");
2255 }
2256 
2257 /**
2258  * Validate NVGRE item.
2259  *
2260  * @param[in] item
2261  *   Item specification.
2262  * @param[in] item_flags
2263  *   Bit flags to mark detected items.
2264  * @param[in] target_protocol
2265  *   The next protocol in the previous item.
2266  * @param[out] error
2267  *   Pointer to error structure.
2268  *
2269  * @return
2270  *   0 on success, a negative errno value otherwise and rte_errno is set.
2271  */
2272 int
2273 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2274 			      uint64_t item_flags,
2275 			      uint8_t target_protocol,
2276 			      struct rte_flow_error *error)
2277 {
2278 	const struct rte_flow_item_nvgre *mask = item->mask;
2279 	int ret;
2280 
2281 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2282 		return rte_flow_error_set(error, EINVAL,
2283 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2284 					  "protocol filtering not compatible"
2285 					  " with this GRE layer");
2286 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2287 		return rte_flow_error_set(error, ENOTSUP,
2288 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2289 					  "multiple tunnel layers not"
2290 					  " supported");
2291 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2292 		return rte_flow_error_set(error, ENOTSUP,
2293 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2294 					  "L3 Layer is missing");
2295 	if (!mask)
2296 		mask = &rte_flow_item_nvgre_mask;
2297 	ret = mlx5_flow_item_acceptable
2298 		(item, (const uint8_t *)mask,
2299 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2300 		 sizeof(struct rte_flow_item_nvgre), error);
2301 	if (ret < 0)
2302 		return ret;
2303 	return 0;
2304 }
2305 
2306 /* Allocate unique ID for the split Q/RSS subflows. */
2307 static uint32_t
2308 flow_qrss_get_id(struct rte_eth_dev *dev)
2309 {
2310 	struct mlx5_priv *priv = dev->data->dev_private;
2311 	uint32_t qrss_id, ret;
2312 
2313 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2314 	if (ret)
2315 		return 0;
2316 	MLX5_ASSERT(qrss_id);
2317 	return qrss_id;
2318 }
2319 
2320 /* Free unique ID for the split Q/RSS subflows. */
2321 static void
2322 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2323 {
2324 	struct mlx5_priv *priv = dev->data->dev_private;
2325 
2326 	if (qrss_id)
2327 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2328 }
2329 
2330 /**
2331  * Release resource related QUEUE/RSS action split.
2332  *
2333  * @param dev
2334  *   Pointer to Ethernet device.
2335  * @param flow
2336  *   Flow to release id's from.
2337  */
2338 static void
2339 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2340 			     struct rte_flow *flow)
2341 {
2342 	struct mlx5_flow *dev_flow;
2343 
2344 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2345 		if (dev_flow->qrss_id)
2346 			flow_qrss_free_id(dev, dev_flow->qrss_id);
2347 }
2348 
2349 static int
2350 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2351 		   const struct rte_flow_attr *attr __rte_unused,
2352 		   const struct rte_flow_item items[] __rte_unused,
2353 		   const struct rte_flow_action actions[] __rte_unused,
2354 		   bool external __rte_unused,
2355 		   struct rte_flow_error *error)
2356 {
2357 	return rte_flow_error_set(error, ENOTSUP,
2358 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2359 }
2360 
2361 static struct mlx5_flow *
2362 flow_null_prepare(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 	rte_flow_error_set(error, ENOTSUP,
2368 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2369 	return NULL;
2370 }
2371 
2372 static int
2373 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2374 		    struct mlx5_flow *dev_flow __rte_unused,
2375 		    const struct rte_flow_attr *attr __rte_unused,
2376 		    const struct rte_flow_item items[] __rte_unused,
2377 		    const struct rte_flow_action actions[] __rte_unused,
2378 		    struct rte_flow_error *error)
2379 {
2380 	return rte_flow_error_set(error, ENOTSUP,
2381 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2382 }
2383 
2384 static int
2385 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2386 		struct rte_flow *flow __rte_unused,
2387 		struct rte_flow_error *error)
2388 {
2389 	return rte_flow_error_set(error, ENOTSUP,
2390 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2391 }
2392 
2393 static void
2394 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2395 		 struct rte_flow *flow __rte_unused)
2396 {
2397 }
2398 
2399 static void
2400 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2401 		  struct rte_flow *flow __rte_unused)
2402 {
2403 }
2404 
2405 static int
2406 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2407 		struct rte_flow *flow __rte_unused,
2408 		const struct rte_flow_action *actions __rte_unused,
2409 		void *data __rte_unused,
2410 		struct rte_flow_error *error)
2411 {
2412 	return rte_flow_error_set(error, ENOTSUP,
2413 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2414 }
2415 
2416 /* Void driver to protect from null pointer reference. */
2417 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2418 	.validate = flow_null_validate,
2419 	.prepare = flow_null_prepare,
2420 	.translate = flow_null_translate,
2421 	.apply = flow_null_apply,
2422 	.remove = flow_null_remove,
2423 	.destroy = flow_null_destroy,
2424 	.query = flow_null_query,
2425 };
2426 
2427 /**
2428  * Select flow driver type according to flow attributes and device
2429  * configuration.
2430  *
2431  * @param[in] dev
2432  *   Pointer to the dev structure.
2433  * @param[in] attr
2434  *   Pointer to the flow attributes.
2435  *
2436  * @return
2437  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2438  */
2439 static enum mlx5_flow_drv_type
2440 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2441 {
2442 	struct mlx5_priv *priv = dev->data->dev_private;
2443 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2444 
2445 	if (attr->transfer && priv->config.dv_esw_en)
2446 		type = MLX5_FLOW_TYPE_DV;
2447 	if (!attr->transfer)
2448 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2449 						 MLX5_FLOW_TYPE_VERBS;
2450 	return type;
2451 }
2452 
2453 #define flow_get_drv_ops(type) flow_drv_ops[type]
2454 
2455 /**
2456  * Flow driver validation API. This abstracts calling driver specific functions.
2457  * The type of flow driver is determined according to flow attributes.
2458  *
2459  * @param[in] dev
2460  *   Pointer to the dev structure.
2461  * @param[in] attr
2462  *   Pointer to the flow attributes.
2463  * @param[in] items
2464  *   Pointer to the list of items.
2465  * @param[in] actions
2466  *   Pointer to the list of actions.
2467  * @param[in] external
2468  *   This flow rule is created by request external to PMD.
2469  * @param[out] error
2470  *   Pointer to the error structure.
2471  *
2472  * @return
2473  *   0 on success, a negative errno value otherwise and rte_errno is set.
2474  */
2475 static inline int
2476 flow_drv_validate(struct rte_eth_dev *dev,
2477 		  const struct rte_flow_attr *attr,
2478 		  const struct rte_flow_item items[],
2479 		  const struct rte_flow_action actions[],
2480 		  bool external, struct rte_flow_error *error)
2481 {
2482 	const struct mlx5_flow_driver_ops *fops;
2483 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2484 
2485 	fops = flow_get_drv_ops(type);
2486 	return fops->validate(dev, attr, items, actions, external, error);
2487 }
2488 
2489 /**
2490  * Flow driver preparation API. This abstracts calling driver specific
2491  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2492  * calculates the size of memory required for device flow, allocates the memory,
2493  * initializes the device flow and returns the pointer.
2494  *
2495  * @note
2496  *   This function initializes device flow structure such as dv or verbs in
2497  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2498  *   rest. For example, adding returning device flow to flow->dev_flow list and
2499  *   setting backward reference to the flow should be done out of this function.
2500  *   layers field is not filled either.
2501  *
2502  * @param[in] attr
2503  *   Pointer to the flow attributes.
2504  * @param[in] items
2505  *   Pointer to the list of items.
2506  * @param[in] actions
2507  *   Pointer to the list of actions.
2508  * @param[out] error
2509  *   Pointer to the error structure.
2510  *
2511  * @return
2512  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2513  */
2514 static inline struct mlx5_flow *
2515 flow_drv_prepare(const struct rte_flow *flow,
2516 		 const struct rte_flow_attr *attr,
2517 		 const struct rte_flow_item items[],
2518 		 const struct rte_flow_action actions[],
2519 		 struct rte_flow_error *error)
2520 {
2521 	const struct mlx5_flow_driver_ops *fops;
2522 	enum mlx5_flow_drv_type type = flow->drv_type;
2523 
2524 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2525 	fops = flow_get_drv_ops(type);
2526 	return fops->prepare(attr, items, actions, error);
2527 }
2528 
2529 /**
2530  * Flow driver translation API. This abstracts calling driver specific
2531  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2532  * translates a generic flow into a driver flow. flow_drv_prepare() must
2533  * precede.
2534  *
2535  * @note
2536  *   dev_flow->layers could be filled as a result of parsing during translation
2537  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2538  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2539  *   flow->actions could be overwritten even though all the expanded dev_flows
2540  *   have the same actions.
2541  *
2542  * @param[in] dev
2543  *   Pointer to the rte dev structure.
2544  * @param[in, out] dev_flow
2545  *   Pointer to the mlx5 flow.
2546  * @param[in] attr
2547  *   Pointer to the flow attributes.
2548  * @param[in] items
2549  *   Pointer to the list of items.
2550  * @param[in] actions
2551  *   Pointer to the list of actions.
2552  * @param[out] error
2553  *   Pointer to the error structure.
2554  *
2555  * @return
2556  *   0 on success, a negative errno value otherwise and rte_errno is set.
2557  */
2558 static inline int
2559 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2560 		   const struct rte_flow_attr *attr,
2561 		   const struct rte_flow_item items[],
2562 		   const struct rte_flow_action actions[],
2563 		   struct rte_flow_error *error)
2564 {
2565 	const struct mlx5_flow_driver_ops *fops;
2566 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2567 
2568 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2569 	fops = flow_get_drv_ops(type);
2570 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2571 }
2572 
2573 /**
2574  * Flow driver apply API. This abstracts calling driver specific functions.
2575  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2576  * translated driver flows on to device. flow_drv_translate() must precede.
2577  *
2578  * @param[in] dev
2579  *   Pointer to Ethernet device structure.
2580  * @param[in, out] flow
2581  *   Pointer to flow structure.
2582  * @param[out] error
2583  *   Pointer to error structure.
2584  *
2585  * @return
2586  *   0 on success, a negative errno value otherwise and rte_errno is set.
2587  */
2588 static inline int
2589 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2590 	       struct rte_flow_error *error)
2591 {
2592 	const struct mlx5_flow_driver_ops *fops;
2593 	enum mlx5_flow_drv_type type = flow->drv_type;
2594 
2595 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2596 	fops = flow_get_drv_ops(type);
2597 	return fops->apply(dev, flow, error);
2598 }
2599 
2600 /**
2601  * Flow driver remove API. This abstracts calling driver specific functions.
2602  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2603  * on device. All the resources of the flow should be freed by calling
2604  * flow_drv_destroy().
2605  *
2606  * @param[in] dev
2607  *   Pointer to Ethernet device.
2608  * @param[in, out] flow
2609  *   Pointer to flow structure.
2610  */
2611 static inline void
2612 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2613 {
2614 	const struct mlx5_flow_driver_ops *fops;
2615 	enum mlx5_flow_drv_type type = flow->drv_type;
2616 
2617 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2618 	fops = flow_get_drv_ops(type);
2619 	fops->remove(dev, flow);
2620 }
2621 
2622 /**
2623  * Flow driver destroy API. This abstracts calling driver specific functions.
2624  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2625  * on device and releases resources of the flow.
2626  *
2627  * @param[in] dev
2628  *   Pointer to Ethernet device.
2629  * @param[in, out] flow
2630  *   Pointer to flow structure.
2631  */
2632 static inline void
2633 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2634 {
2635 	const struct mlx5_flow_driver_ops *fops;
2636 	enum mlx5_flow_drv_type type = flow->drv_type;
2637 
2638 	flow_mreg_split_qrss_release(dev, flow);
2639 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2640 	fops = flow_get_drv_ops(type);
2641 	fops->destroy(dev, flow);
2642 }
2643 
2644 /**
2645  * Validate a flow supported by the NIC.
2646  *
2647  * @see rte_flow_validate()
2648  * @see rte_flow_ops
2649  */
2650 int
2651 mlx5_flow_validate(struct rte_eth_dev *dev,
2652 		   const struct rte_flow_attr *attr,
2653 		   const struct rte_flow_item items[],
2654 		   const struct rte_flow_action actions[],
2655 		   struct rte_flow_error *error)
2656 {
2657 	int ret;
2658 
2659 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2660 	if (ret < 0)
2661 		return ret;
2662 	return 0;
2663 }
2664 
2665 /**
2666  * Get RSS action from the action list.
2667  *
2668  * @param[in] actions
2669  *   Pointer to the list of actions.
2670  *
2671  * @return
2672  *   Pointer to the RSS action if exist, else return NULL.
2673  */
2674 static const struct rte_flow_action_rss*
2675 flow_get_rss_action(const struct rte_flow_action actions[])
2676 {
2677 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2678 		switch (actions->type) {
2679 		case RTE_FLOW_ACTION_TYPE_RSS:
2680 			return (const struct rte_flow_action_rss *)
2681 			       actions->conf;
2682 		default:
2683 			break;
2684 		}
2685 	}
2686 	return NULL;
2687 }
2688 
2689 static unsigned int
2690 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2691 {
2692 	const struct rte_flow_item *item;
2693 	unsigned int has_vlan = 0;
2694 
2695 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2696 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2697 			has_vlan = 1;
2698 			break;
2699 		}
2700 	}
2701 	if (has_vlan)
2702 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2703 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2704 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2705 			       MLX5_EXPANSION_ROOT_OUTER;
2706 }
2707 
2708 /**
2709  *  Get layer flags from the prefix flow.
2710  *
2711  *  Some flows may be split to several subflows, the prefix subflow gets the
2712  *  match items and the suffix sub flow gets the actions.
2713  *  Some actions need the user defined match item flags to get the detail for
2714  *  the action.
2715  *  This function helps the suffix flow to get the item layer flags from prefix
2716  *  subflow.
2717  *
2718  * @param[in] dev_flow
2719  *   Pointer the created preifx subflow.
2720  *
2721  * @return
2722  *   The layers get from prefix subflow.
2723  */
2724 static inline uint64_t
2725 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2726 {
2727 	uint64_t layers = 0;
2728 
2729 	/* If no decap actions, use the layers directly. */
2730 	if (!(dev_flow->actions & MLX5_FLOW_ACTION_DECAP))
2731 		return dev_flow->layers;
2732 	/* Convert L3 layers with decap action. */
2733 	if (dev_flow->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2734 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2735 	else if (dev_flow->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2736 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2737 	/* Convert L4 layers with decap action.  */
2738 	if (dev_flow->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2739 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2740 	else if (dev_flow->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2741 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2742 	return layers;
2743 }
2744 
2745 /**
2746  * Get metadata split action information.
2747  *
2748  * @param[in] actions
2749  *   Pointer to the list of actions.
2750  * @param[out] qrss
2751  *   Pointer to the return pointer.
2752  * @param[out] qrss_type
2753  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2754  *   if no QUEUE/RSS is found.
2755  * @param[out] encap_idx
2756  *   Pointer to the index of the encap action if exists, otherwise the last
2757  *   action index.
2758  *
2759  * @return
2760  *   Total number of actions.
2761  */
2762 static int
2763 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2764 				       const struct rte_flow_action **qrss,
2765 				       int *encap_idx)
2766 {
2767 	const struct rte_flow_action_raw_encap *raw_encap;
2768 	int actions_n = 0;
2769 	int raw_decap_idx = -1;
2770 
2771 	*encap_idx = -1;
2772 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2773 		switch (actions->type) {
2774 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2775 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2776 			*encap_idx = actions_n;
2777 			break;
2778 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2779 			raw_decap_idx = actions_n;
2780 			break;
2781 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2782 			raw_encap = actions->conf;
2783 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2784 				*encap_idx = raw_decap_idx != -1 ?
2785 						      raw_decap_idx : actions_n;
2786 			break;
2787 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2788 		case RTE_FLOW_ACTION_TYPE_RSS:
2789 			*qrss = actions;
2790 			break;
2791 		default:
2792 			break;
2793 		}
2794 		actions_n++;
2795 	}
2796 	if (*encap_idx == -1)
2797 		*encap_idx = actions_n;
2798 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2799 	return actions_n + 1;
2800 }
2801 
2802 /**
2803  * Check meter action from the action list.
2804  *
2805  * @param[in] actions
2806  *   Pointer to the list of actions.
2807  * @param[out] mtr
2808  *   Pointer to the meter exist flag.
2809  *
2810  * @return
2811  *   Total number of actions.
2812  */
2813 static int
2814 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2815 {
2816 	int actions_n = 0;
2817 
2818 	MLX5_ASSERT(mtr);
2819 	*mtr = 0;
2820 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2821 		switch (actions->type) {
2822 		case RTE_FLOW_ACTION_TYPE_METER:
2823 			*mtr = 1;
2824 			break;
2825 		default:
2826 			break;
2827 		}
2828 		actions_n++;
2829 	}
2830 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2831 	return actions_n + 1;
2832 }
2833 
2834 /**
2835  * Check if the flow should be splited due to hairpin.
2836  * The reason for the split is that in current HW we can't
2837  * support encap on Rx, so if a flow have encap we move it
2838  * to Tx.
2839  *
2840  * @param dev
2841  *   Pointer to Ethernet device.
2842  * @param[in] attr
2843  *   Flow rule attributes.
2844  * @param[in] actions
2845  *   Associated actions (list terminated by the END action).
2846  *
2847  * @return
2848  *   > 0 the number of actions and the flow should be split,
2849  *   0 when no split required.
2850  */
2851 static int
2852 flow_check_hairpin_split(struct rte_eth_dev *dev,
2853 			 const struct rte_flow_attr *attr,
2854 			 const struct rte_flow_action actions[])
2855 {
2856 	int queue_action = 0;
2857 	int action_n = 0;
2858 	int encap = 0;
2859 	const struct rte_flow_action_queue *queue;
2860 	const struct rte_flow_action_rss *rss;
2861 	const struct rte_flow_action_raw_encap *raw_encap;
2862 
2863 	if (!attr->ingress)
2864 		return 0;
2865 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2866 		switch (actions->type) {
2867 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2868 			queue = actions->conf;
2869 			if (queue == NULL)
2870 				return 0;
2871 			if (mlx5_rxq_get_type(dev, queue->index) !=
2872 			    MLX5_RXQ_TYPE_HAIRPIN)
2873 				return 0;
2874 			queue_action = 1;
2875 			action_n++;
2876 			break;
2877 		case RTE_FLOW_ACTION_TYPE_RSS:
2878 			rss = actions->conf;
2879 			if (rss == NULL || rss->queue_num == 0)
2880 				return 0;
2881 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2882 			    MLX5_RXQ_TYPE_HAIRPIN)
2883 				return 0;
2884 			queue_action = 1;
2885 			action_n++;
2886 			break;
2887 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2888 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2889 			encap = 1;
2890 			action_n++;
2891 			break;
2892 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2893 			raw_encap = actions->conf;
2894 			if (raw_encap->size >
2895 			    (sizeof(struct rte_flow_item_eth) +
2896 			     sizeof(struct rte_flow_item_ipv4)))
2897 				encap = 1;
2898 			action_n++;
2899 			break;
2900 		default:
2901 			action_n++;
2902 			break;
2903 		}
2904 	}
2905 	if (encap == 1 && queue_action)
2906 		return action_n;
2907 	return 0;
2908 }
2909 
2910 /* Declare flow create/destroy prototype in advance. */
2911 static struct rte_flow *
2912 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2913 		 const struct rte_flow_attr *attr,
2914 		 const struct rte_flow_item items[],
2915 		 const struct rte_flow_action actions[],
2916 		 bool external, struct rte_flow_error *error);
2917 
2918 static void
2919 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2920 		  struct rte_flow *flow);
2921 
2922 /**
2923  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2924  *
2925  * As mark_id is unique, if there's already a registered flow for the mark_id,
2926  * return by increasing the reference counter of the resource. Otherwise, create
2927  * the resource (mcp_res) and flow.
2928  *
2929  * Flow looks like,
2930  *   - If ingress port is ANY and reg_c[1] is mark_id,
2931  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2932  *
2933  * For default flow (zero mark_id), flow is like,
2934  *   - If ingress port is ANY,
2935  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2936  *
2937  * @param dev
2938  *   Pointer to Ethernet device.
2939  * @param mark_id
2940  *   ID of MARK action, zero means default flow for META.
2941  * @param[out] error
2942  *   Perform verbose error reporting if not NULL.
2943  *
2944  * @return
2945  *   Associated resource on success, NULL otherwise and rte_errno is set.
2946  */
2947 static struct mlx5_flow_mreg_copy_resource *
2948 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2949 			  struct rte_flow_error *error)
2950 {
2951 	struct mlx5_priv *priv = dev->data->dev_private;
2952 	struct rte_flow_attr attr = {
2953 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2954 		.ingress = 1,
2955 	};
2956 	struct mlx5_rte_flow_item_tag tag_spec = {
2957 		.data = mark_id,
2958 	};
2959 	struct rte_flow_item items[] = {
2960 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2961 	};
2962 	struct rte_flow_action_mark ftag = {
2963 		.id = mark_id,
2964 	};
2965 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2966 		.dst = REG_B,
2967 		.src = 0,
2968 	};
2969 	struct rte_flow_action_jump jump = {
2970 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2971 	};
2972 	struct rte_flow_action actions[] = {
2973 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2974 	};
2975 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2976 	int ret;
2977 
2978 	/* Fill the register fileds in the flow. */
2979 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2980 	if (ret < 0)
2981 		return NULL;
2982 	tag_spec.id = ret;
2983 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2984 	if (ret < 0)
2985 		return NULL;
2986 	cp_mreg.src = ret;
2987 	/* Check if already registered. */
2988 	MLX5_ASSERT(priv->mreg_cp_tbl);
2989 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2990 	if (mcp_res) {
2991 		/* For non-default rule. */
2992 		if (mark_id != MLX5_DEFAULT_COPY_ID)
2993 			mcp_res->refcnt++;
2994 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2995 			    mcp_res->refcnt == 1);
2996 		return mcp_res;
2997 	}
2998 	/* Provide the full width of FLAG specific value. */
2999 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3000 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3001 	/* Build a new flow. */
3002 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3003 		items[0] = (struct rte_flow_item){
3004 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3005 			.spec = &tag_spec,
3006 		};
3007 		items[1] = (struct rte_flow_item){
3008 			.type = RTE_FLOW_ITEM_TYPE_END,
3009 		};
3010 		actions[0] = (struct rte_flow_action){
3011 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3012 			.conf = &ftag,
3013 		};
3014 		actions[1] = (struct rte_flow_action){
3015 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3016 			.conf = &cp_mreg,
3017 		};
3018 		actions[2] = (struct rte_flow_action){
3019 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3020 			.conf = &jump,
3021 		};
3022 		actions[3] = (struct rte_flow_action){
3023 			.type = RTE_FLOW_ACTION_TYPE_END,
3024 		};
3025 	} else {
3026 		/* Default rule, wildcard match. */
3027 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3028 		items[0] = (struct rte_flow_item){
3029 			.type = RTE_FLOW_ITEM_TYPE_END,
3030 		};
3031 		actions[0] = (struct rte_flow_action){
3032 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3033 			.conf = &cp_mreg,
3034 		};
3035 		actions[1] = (struct rte_flow_action){
3036 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3037 			.conf = &jump,
3038 		};
3039 		actions[2] = (struct rte_flow_action){
3040 			.type = RTE_FLOW_ACTION_TYPE_END,
3041 		};
3042 	}
3043 	/* Build a new entry. */
3044 	mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3045 	if (!mcp_res) {
3046 		rte_errno = ENOMEM;
3047 		return NULL;
3048 	}
3049 	/*
3050 	 * The copy Flows are not included in any list. There
3051 	 * ones are referenced from other Flows and can not
3052 	 * be applied, removed, deleted in ardbitrary order
3053 	 * by list traversing.
3054 	 */
3055 	mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3056 					 actions, false, error);
3057 	if (!mcp_res->flow)
3058 		goto error;
3059 	mcp_res->refcnt++;
3060 	mcp_res->hlist_ent.key = mark_id;
3061 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3062 				&mcp_res->hlist_ent);
3063 	MLX5_ASSERT(!ret);
3064 	if (ret)
3065 		goto error;
3066 	return mcp_res;
3067 error:
3068 	if (mcp_res->flow)
3069 		flow_list_destroy(dev, NULL, mcp_res->flow);
3070 	rte_free(mcp_res);
3071 	return NULL;
3072 }
3073 
3074 /**
3075  * Release flow in RX_CP_TBL.
3076  *
3077  * @param dev
3078  *   Pointer to Ethernet device.
3079  * @flow
3080  *   Parent flow for wich copying is provided.
3081  */
3082 static void
3083 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3084 			  struct rte_flow *flow)
3085 {
3086 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3087 	struct mlx5_priv *priv = dev->data->dev_private;
3088 
3089 	if (!mcp_res || !priv->mreg_cp_tbl)
3090 		return;
3091 	if (flow->copy_applied) {
3092 		MLX5_ASSERT(mcp_res->appcnt);
3093 		flow->copy_applied = 0;
3094 		--mcp_res->appcnt;
3095 		if (!mcp_res->appcnt)
3096 			flow_drv_remove(dev, mcp_res->flow);
3097 	}
3098 	/*
3099 	 * We do not check availability of metadata registers here,
3100 	 * because copy resources are not allocated in this case.
3101 	 */
3102 	if (--mcp_res->refcnt)
3103 		return;
3104 	MLX5_ASSERT(mcp_res->flow);
3105 	flow_list_destroy(dev, NULL, mcp_res->flow);
3106 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3107 	rte_free(mcp_res);
3108 	flow->mreg_copy = NULL;
3109 }
3110 
3111 /**
3112  * Start flow in RX_CP_TBL.
3113  *
3114  * @param dev
3115  *   Pointer to Ethernet device.
3116  * @flow
3117  *   Parent flow for wich copying is provided.
3118  *
3119  * @return
3120  *   0 on success, a negative errno value otherwise and rte_errno is set.
3121  */
3122 static int
3123 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3124 			    struct rte_flow *flow)
3125 {
3126 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3127 	int ret;
3128 
3129 	if (!mcp_res || flow->copy_applied)
3130 		return 0;
3131 	if (!mcp_res->appcnt) {
3132 		ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3133 		if (ret)
3134 			return ret;
3135 	}
3136 	++mcp_res->appcnt;
3137 	flow->copy_applied = 1;
3138 	return 0;
3139 }
3140 
3141 /**
3142  * Stop flow in RX_CP_TBL.
3143  *
3144  * @param dev
3145  *   Pointer to Ethernet device.
3146  * @flow
3147  *   Parent flow for wich copying is provided.
3148  */
3149 static void
3150 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3151 			   struct rte_flow *flow)
3152 {
3153 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3154 
3155 	if (!mcp_res || !flow->copy_applied)
3156 		return;
3157 	MLX5_ASSERT(mcp_res->appcnt);
3158 	--mcp_res->appcnt;
3159 	flow->copy_applied = 0;
3160 	if (!mcp_res->appcnt)
3161 		flow_drv_remove(dev, mcp_res->flow);
3162 }
3163 
3164 /**
3165  * Remove the default copy action from RX_CP_TBL.
3166  *
3167  * @param dev
3168  *   Pointer to Ethernet device.
3169  */
3170 static void
3171 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3172 {
3173 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3174 	struct mlx5_priv *priv = dev->data->dev_private;
3175 
3176 	/* Check if default flow is registered. */
3177 	if (!priv->mreg_cp_tbl)
3178 		return;
3179 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3180 					    MLX5_DEFAULT_COPY_ID);
3181 	if (!mcp_res)
3182 		return;
3183 	MLX5_ASSERT(mcp_res->flow);
3184 	flow_list_destroy(dev, NULL, mcp_res->flow);
3185 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3186 	rte_free(mcp_res);
3187 }
3188 
3189 /**
3190  * Add the default copy action in in RX_CP_TBL.
3191  *
3192  * @param dev
3193  *   Pointer to Ethernet device.
3194  * @param[out] error
3195  *   Perform verbose error reporting if not NULL.
3196  *
3197  * @return
3198  *   0 for success, negative value otherwise and rte_errno is set.
3199  */
3200 static int
3201 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3202 				  struct rte_flow_error *error)
3203 {
3204 	struct mlx5_priv *priv = dev->data->dev_private;
3205 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3206 
3207 	/* Check whether extensive metadata feature is engaged. */
3208 	if (!priv->config.dv_flow_en ||
3209 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3210 	    !mlx5_flow_ext_mreg_supported(dev) ||
3211 	    !priv->sh->dv_regc0_mask)
3212 		return 0;
3213 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3214 	if (!mcp_res)
3215 		return -rte_errno;
3216 	return 0;
3217 }
3218 
3219 /**
3220  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3221  *
3222  * All the flow having Q/RSS action should be split by
3223  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3224  * performs the following,
3225  *   - CQE->flow_tag := reg_c[1] (MARK)
3226  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3227  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3228  * but there should be a flow per each MARK ID set by MARK action.
3229  *
3230  * For the aforementioned reason, if there's a MARK action in flow's action
3231  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3232  * the MARK ID to CQE's flow_tag like,
3233  *   - If reg_c[1] is mark_id,
3234  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3235  *
3236  * For SET_META action which stores value in reg_c[0], as the destination is
3237  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3238  * MARK ID means the default flow. The default flow looks like,
3239  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3240  *
3241  * @param dev
3242  *   Pointer to Ethernet device.
3243  * @param flow
3244  *   Pointer to flow structure.
3245  * @param[in] actions
3246  *   Pointer to the list of actions.
3247  * @param[out] error
3248  *   Perform verbose error reporting if not NULL.
3249  *
3250  * @return
3251  *   0 on success, negative value otherwise and rte_errno is set.
3252  */
3253 static int
3254 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3255 			    struct rte_flow *flow,
3256 			    const struct rte_flow_action *actions,
3257 			    struct rte_flow_error *error)
3258 {
3259 	struct mlx5_priv *priv = dev->data->dev_private;
3260 	struct mlx5_dev_config *config = &priv->config;
3261 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3262 	const struct rte_flow_action_mark *mark;
3263 
3264 	/* Check whether extensive metadata feature is engaged. */
3265 	if (!config->dv_flow_en ||
3266 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3267 	    !mlx5_flow_ext_mreg_supported(dev) ||
3268 	    !priv->sh->dv_regc0_mask)
3269 		return 0;
3270 	/* Find MARK action. */
3271 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3272 		switch (actions->type) {
3273 		case RTE_FLOW_ACTION_TYPE_FLAG:
3274 			mcp_res = flow_mreg_add_copy_action
3275 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3276 			if (!mcp_res)
3277 				return -rte_errno;
3278 			flow->mreg_copy = mcp_res;
3279 			if (dev->data->dev_started) {
3280 				mcp_res->appcnt++;
3281 				flow->copy_applied = 1;
3282 			}
3283 			return 0;
3284 		case RTE_FLOW_ACTION_TYPE_MARK:
3285 			mark = (const struct rte_flow_action_mark *)
3286 				actions->conf;
3287 			mcp_res =
3288 				flow_mreg_add_copy_action(dev, mark->id, error);
3289 			if (!mcp_res)
3290 				return -rte_errno;
3291 			flow->mreg_copy = mcp_res;
3292 			if (dev->data->dev_started) {
3293 				mcp_res->appcnt++;
3294 				flow->copy_applied = 1;
3295 			}
3296 			return 0;
3297 		default:
3298 			break;
3299 		}
3300 	}
3301 	return 0;
3302 }
3303 
3304 #define MLX5_MAX_SPLIT_ACTIONS 24
3305 #define MLX5_MAX_SPLIT_ITEMS 24
3306 
3307 /**
3308  * Split the hairpin flow.
3309  * Since HW can't support encap on Rx we move the encap to Tx.
3310  * If the count action is after the encap then we also
3311  * move the count action. in this case the count will also measure
3312  * the outer bytes.
3313  *
3314  * @param dev
3315  *   Pointer to Ethernet device.
3316  * @param[in] actions
3317  *   Associated actions (list terminated by the END action).
3318  * @param[out] actions_rx
3319  *   Rx flow actions.
3320  * @param[out] actions_tx
3321  *   Tx flow actions..
3322  * @param[out] pattern_tx
3323  *   The pattern items for the Tx flow.
3324  * @param[out] flow_id
3325  *   The flow ID connected to this flow.
3326  *
3327  * @return
3328  *   0 on success.
3329  */
3330 static int
3331 flow_hairpin_split(struct rte_eth_dev *dev,
3332 		   const struct rte_flow_action actions[],
3333 		   struct rte_flow_action actions_rx[],
3334 		   struct rte_flow_action actions_tx[],
3335 		   struct rte_flow_item pattern_tx[],
3336 		   uint32_t *flow_id)
3337 {
3338 	struct mlx5_priv *priv = dev->data->dev_private;
3339 	const struct rte_flow_action_raw_encap *raw_encap;
3340 	const struct rte_flow_action_raw_decap *raw_decap;
3341 	struct mlx5_rte_flow_action_set_tag *set_tag;
3342 	struct rte_flow_action *tag_action;
3343 	struct mlx5_rte_flow_item_tag *tag_item;
3344 	struct rte_flow_item *item;
3345 	char *addr;
3346 	int encap = 0;
3347 
3348 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3349 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3350 		switch (actions->type) {
3351 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3352 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3353 			rte_memcpy(actions_tx, actions,
3354 			       sizeof(struct rte_flow_action));
3355 			actions_tx++;
3356 			break;
3357 		case RTE_FLOW_ACTION_TYPE_COUNT:
3358 			if (encap) {
3359 				rte_memcpy(actions_tx, actions,
3360 					   sizeof(struct rte_flow_action));
3361 				actions_tx++;
3362 			} else {
3363 				rte_memcpy(actions_rx, actions,
3364 					   sizeof(struct rte_flow_action));
3365 				actions_rx++;
3366 			}
3367 			break;
3368 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3369 			raw_encap = actions->conf;
3370 			if (raw_encap->size >
3371 			    (sizeof(struct rte_flow_item_eth) +
3372 			     sizeof(struct rte_flow_item_ipv4))) {
3373 				memcpy(actions_tx, actions,
3374 				       sizeof(struct rte_flow_action));
3375 				actions_tx++;
3376 				encap = 1;
3377 			} else {
3378 				rte_memcpy(actions_rx, actions,
3379 					   sizeof(struct rte_flow_action));
3380 				actions_rx++;
3381 			}
3382 			break;
3383 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3384 			raw_decap = actions->conf;
3385 			if (raw_decap->size <
3386 			    (sizeof(struct rte_flow_item_eth) +
3387 			     sizeof(struct rte_flow_item_ipv4))) {
3388 				memcpy(actions_tx, actions,
3389 				       sizeof(struct rte_flow_action));
3390 				actions_tx++;
3391 			} else {
3392 				rte_memcpy(actions_rx, actions,
3393 					   sizeof(struct rte_flow_action));
3394 				actions_rx++;
3395 			}
3396 			break;
3397 		default:
3398 			rte_memcpy(actions_rx, actions,
3399 				   sizeof(struct rte_flow_action));
3400 			actions_rx++;
3401 			break;
3402 		}
3403 	}
3404 	/* Add set meta action and end action for the Rx flow. */
3405 	tag_action = actions_rx;
3406 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3407 	actions_rx++;
3408 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3409 	actions_rx++;
3410 	set_tag = (void *)actions_rx;
3411 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3412 	MLX5_ASSERT(set_tag->id > REG_NONE);
3413 	set_tag->data = *flow_id;
3414 	tag_action->conf = set_tag;
3415 	/* Create Tx item list. */
3416 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3417 	addr = (void *)&pattern_tx[2];
3418 	item = pattern_tx;
3419 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3420 	tag_item = (void *)addr;
3421 	tag_item->data = *flow_id;
3422 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3423 	MLX5_ASSERT(set_tag->id > REG_NONE);
3424 	item->spec = tag_item;
3425 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3426 	tag_item = (void *)addr;
3427 	tag_item->data = UINT32_MAX;
3428 	tag_item->id = UINT16_MAX;
3429 	item->mask = tag_item;
3430 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3431 	item->last = NULL;
3432 	item++;
3433 	item->type = RTE_FLOW_ITEM_TYPE_END;
3434 	return 0;
3435 }
3436 
3437 /**
3438  * The last stage of splitting chain, just creates the subflow
3439  * without any modification.
3440  *
3441  * @param dev
3442  *   Pointer to Ethernet device.
3443  * @param[in] flow
3444  *   Parent flow structure pointer.
3445  * @param[in, out] sub_flow
3446  *   Pointer to return the created subflow, may be NULL.
3447  * @param[in] prefix_layers
3448  *   Prefix subflow layers, may be 0.
3449  * @param[in] attr
3450  *   Flow rule attributes.
3451  * @param[in] items
3452  *   Pattern specification (list terminated by the END pattern item).
3453  * @param[in] actions
3454  *   Associated actions (list terminated by the END action).
3455  * @param[in] external
3456  *   This flow rule is created by request external to PMD.
3457  * @param[out] error
3458  *   Perform verbose error reporting if not NULL.
3459  * @return
3460  *   0 on success, negative value otherwise
3461  */
3462 static int
3463 flow_create_split_inner(struct rte_eth_dev *dev,
3464 			struct rte_flow *flow,
3465 			struct mlx5_flow **sub_flow,
3466 			uint64_t prefix_layers,
3467 			const struct rte_flow_attr *attr,
3468 			const struct rte_flow_item items[],
3469 			const struct rte_flow_action actions[],
3470 			bool external, struct rte_flow_error *error)
3471 {
3472 	struct mlx5_flow *dev_flow;
3473 
3474 	dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3475 	if (!dev_flow)
3476 		return -rte_errno;
3477 	dev_flow->flow = flow;
3478 	dev_flow->external = external;
3479 	/* Subflow object was created, we must include one in the list. */
3480 	LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3481 	/*
3482 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3483 	 * flow may need some user defined item layer flags.
3484 	 */
3485 	if (prefix_layers)
3486 		dev_flow->layers = prefix_layers;
3487 	if (sub_flow)
3488 		*sub_flow = dev_flow;
3489 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3490 }
3491 
3492 /**
3493  * Split the meter flow.
3494  *
3495  * As meter flow will split to three sub flow, other than meter
3496  * action, the other actions make sense to only meter accepts
3497  * the packet. If it need to be dropped, no other additional
3498  * actions should be take.
3499  *
3500  * One kind of special action which decapsulates the L3 tunnel
3501  * header will be in the prefix sub flow, as not to take the
3502  * L3 tunnel header into account.
3503  *
3504  * @param dev
3505  *   Pointer to Ethernet device.
3506  * @param[in] items
3507  *   Pattern specification (list terminated by the END pattern item).
3508  * @param[out] sfx_items
3509  *   Suffix flow match items (list terminated by the END pattern item).
3510  * @param[in] actions
3511  *   Associated actions (list terminated by the END action).
3512  * @param[out] actions_sfx
3513  *   Suffix flow actions.
3514  * @param[out] actions_pre
3515  *   Prefix flow actions.
3516  * @param[out] pattern_sfx
3517  *   The pattern items for the suffix flow.
3518  * @param[out] tag_sfx
3519  *   Pointer to suffix flow tag.
3520  *
3521  * @return
3522  *   0 on success.
3523  */
3524 static int
3525 flow_meter_split_prep(struct rte_eth_dev *dev,
3526 		 const struct rte_flow_item items[],
3527 		 struct rte_flow_item sfx_items[],
3528 		 const struct rte_flow_action actions[],
3529 		 struct rte_flow_action actions_sfx[],
3530 		 struct rte_flow_action actions_pre[])
3531 {
3532 	struct rte_flow_action *tag_action = NULL;
3533 	struct rte_flow_item *tag_item;
3534 	struct mlx5_rte_flow_action_set_tag *set_tag;
3535 	struct rte_flow_error error;
3536 	const struct rte_flow_action_raw_encap *raw_encap;
3537 	const struct rte_flow_action_raw_decap *raw_decap;
3538 	struct mlx5_rte_flow_item_tag *tag_spec;
3539 	struct mlx5_rte_flow_item_tag *tag_mask;
3540 	uint32_t tag_id;
3541 	bool copy_vlan = false;
3542 
3543 	/* Prepare the actions for prefix and suffix flow. */
3544 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3545 		struct rte_flow_action **action_cur = NULL;
3546 
3547 		switch (actions->type) {
3548 		case RTE_FLOW_ACTION_TYPE_METER:
3549 			/* Add the extra tag action first. */
3550 			tag_action = actions_pre;
3551 			tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3552 			actions_pre++;
3553 			action_cur = &actions_pre;
3554 			break;
3555 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3556 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3557 			action_cur = &actions_pre;
3558 			break;
3559 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3560 			raw_encap = actions->conf;
3561 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3562 				action_cur = &actions_pre;
3563 			break;
3564 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3565 			raw_decap = actions->conf;
3566 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3567 				action_cur = &actions_pre;
3568 			break;
3569 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3570 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3571 			copy_vlan = true;
3572 			break;
3573 		default:
3574 			break;
3575 		}
3576 		if (!action_cur)
3577 			action_cur = &actions_sfx;
3578 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3579 		(*action_cur)++;
3580 	}
3581 	/* Add end action to the actions. */
3582 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3583 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3584 	actions_pre++;
3585 	/* Set the tag. */
3586 	set_tag = (void *)actions_pre;
3587 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3588 	/*
3589 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3590 	 */
3591 	tag_id = flow_qrss_get_id(dev);
3592 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3593 	assert(tag_action);
3594 	tag_action->conf = set_tag;
3595 	/* Prepare the suffix subflow items. */
3596 	tag_item = sfx_items++;
3597 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3598 		int item_type = items->type;
3599 
3600 		switch (item_type) {
3601 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3602 			memcpy(sfx_items, items, sizeof(*sfx_items));
3603 			sfx_items++;
3604 			break;
3605 		case RTE_FLOW_ITEM_TYPE_VLAN:
3606 			if (copy_vlan) {
3607 				memcpy(sfx_items, items, sizeof(*sfx_items));
3608 				/*
3609 				 * Convert to internal match item, it is used
3610 				 * for vlan push and set vid.
3611 				 */
3612 				sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3613 				sfx_items++;
3614 			}
3615 			break;
3616 		default:
3617 			break;
3618 		}
3619 	}
3620 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3621 	sfx_items++;
3622 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3623 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3624 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3625 	tag_mask = tag_spec + 1;
3626 	tag_mask->data = 0xffffff00;
3627 	tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3628 	tag_item->spec = tag_spec;
3629 	tag_item->last = NULL;
3630 	tag_item->mask = tag_mask;
3631 	return tag_id;
3632 }
3633 
3634 /**
3635  * Split action list having QUEUE/RSS for metadata register copy.
3636  *
3637  * Once Q/RSS action is detected in user's action list, the flow action
3638  * should be split in order to copy metadata registers, which will happen in
3639  * RX_CP_TBL like,
3640  *   - CQE->flow_tag := reg_c[1] (MARK)
3641  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3642  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3643  * This is because the last action of each flow must be a terminal action
3644  * (QUEUE, RSS or DROP).
3645  *
3646  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3647  * stored and kept in the mlx5_flow structure per each sub_flow.
3648  *
3649  * The Q/RSS action is replaced with,
3650  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3651  * And the following JUMP action is added at the end,
3652  *   - JUMP, to RX_CP_TBL.
3653  *
3654  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3655  * flow_create_split_metadata() routine. The flow will look like,
3656  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3657  *
3658  * @param dev
3659  *   Pointer to Ethernet device.
3660  * @param[out] split_actions
3661  *   Pointer to store split actions to jump to CP_TBL.
3662  * @param[in] actions
3663  *   Pointer to the list of original flow actions.
3664  * @param[in] qrss
3665  *   Pointer to the Q/RSS action.
3666  * @param[in] actions_n
3667  *   Number of original actions.
3668  * @param[out] error
3669  *   Perform verbose error reporting if not NULL.
3670  *
3671  * @return
3672  *   non-zero unique flow_id on success, otherwise 0 and
3673  *   error/rte_error are set.
3674  */
3675 static uint32_t
3676 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3677 			  struct rte_flow_action *split_actions,
3678 			  const struct rte_flow_action *actions,
3679 			  const struct rte_flow_action *qrss,
3680 			  int actions_n, struct rte_flow_error *error)
3681 {
3682 	struct mlx5_rte_flow_action_set_tag *set_tag;
3683 	struct rte_flow_action_jump *jump;
3684 	const int qrss_idx = qrss - actions;
3685 	uint32_t flow_id = 0;
3686 	int ret = 0;
3687 
3688 	/*
3689 	 * Given actions will be split
3690 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3691 	 * - Add jump to mreg CP_TBL.
3692 	 * As a result, there will be one more action.
3693 	 */
3694 	++actions_n;
3695 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3696 	set_tag = (void *)(split_actions + actions_n);
3697 	/*
3698 	 * If tag action is not set to void(it means we are not the meter
3699 	 * suffix flow), add the tag action. Since meter suffix flow already
3700 	 * has the tag added.
3701 	 */
3702 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3703 		/*
3704 		 * Allocate the new subflow ID. This one is unique within
3705 		 * device and not shared with representors. Otherwise,
3706 		 * we would have to resolve multi-thread access synch
3707 		 * issue. Each flow on the shared device is appended
3708 		 * with source vport identifier, so the resulting
3709 		 * flows will be unique in the shared (by master and
3710 		 * representors) domain even if they have coinciding
3711 		 * IDs.
3712 		 */
3713 		flow_id = flow_qrss_get_id(dev);
3714 		if (!flow_id)
3715 			return rte_flow_error_set(error, ENOMEM,
3716 						  RTE_FLOW_ERROR_TYPE_ACTION,
3717 						  NULL, "can't allocate id "
3718 						  "for split Q/RSS subflow");
3719 		/* Internal SET_TAG action to set flow ID. */
3720 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3721 			.data = flow_id,
3722 		};
3723 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3724 		if (ret < 0)
3725 			return ret;
3726 		set_tag->id = ret;
3727 		/* Construct new actions array. */
3728 		/* Replace QUEUE/RSS action. */
3729 		split_actions[qrss_idx] = (struct rte_flow_action){
3730 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3731 			.conf = set_tag,
3732 		};
3733 	}
3734 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3735 	jump = (void *)(set_tag + 1);
3736 	*jump = (struct rte_flow_action_jump){
3737 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3738 	};
3739 	split_actions[actions_n - 2] = (struct rte_flow_action){
3740 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3741 		.conf = jump,
3742 	};
3743 	split_actions[actions_n - 1] = (struct rte_flow_action){
3744 		.type = RTE_FLOW_ACTION_TYPE_END,
3745 	};
3746 	return flow_id;
3747 }
3748 
3749 /**
3750  * Extend the given action list for Tx metadata copy.
3751  *
3752  * Copy the given action list to the ext_actions and add flow metadata register
3753  * copy action in order to copy reg_a set by WQE to reg_c[0].
3754  *
3755  * @param[out] ext_actions
3756  *   Pointer to the extended action list.
3757  * @param[in] actions
3758  *   Pointer to the list of actions.
3759  * @param[in] actions_n
3760  *   Number of actions in the list.
3761  * @param[out] error
3762  *   Perform verbose error reporting if not NULL.
3763  * @param[in] encap_idx
3764  *   The encap action inndex.
3765  *
3766  * @return
3767  *   0 on success, negative value otherwise
3768  */
3769 static int
3770 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3771 		       struct rte_flow_action *ext_actions,
3772 		       const struct rte_flow_action *actions,
3773 		       int actions_n, struct rte_flow_error *error,
3774 		       int encap_idx)
3775 {
3776 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3777 		(struct mlx5_flow_action_copy_mreg *)
3778 			(ext_actions + actions_n + 1);
3779 	int ret;
3780 
3781 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3782 	if (ret < 0)
3783 		return ret;
3784 	cp_mreg->dst = ret;
3785 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3786 	if (ret < 0)
3787 		return ret;
3788 	cp_mreg->src = ret;
3789 	if (encap_idx != 0)
3790 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3791 	if (encap_idx == actions_n - 1) {
3792 		ext_actions[actions_n - 1] = (struct rte_flow_action){
3793 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3794 			.conf = cp_mreg,
3795 		};
3796 		ext_actions[actions_n] = (struct rte_flow_action){
3797 			.type = RTE_FLOW_ACTION_TYPE_END,
3798 		};
3799 	} else {
3800 		ext_actions[encap_idx] = (struct rte_flow_action){
3801 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3802 			.conf = cp_mreg,
3803 		};
3804 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3805 				sizeof(*ext_actions) * (actions_n - encap_idx));
3806 	}
3807 	return 0;
3808 }
3809 
3810 /**
3811  * The splitting for metadata feature.
3812  *
3813  * - Q/RSS action on NIC Rx should be split in order to pass by
3814  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3815  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3816  *
3817  * - All the actions on NIC Tx should have a mreg copy action to
3818  *   copy reg_a from WQE to reg_c[0].
3819  *
3820  * @param dev
3821  *   Pointer to Ethernet device.
3822  * @param[in] flow
3823  *   Parent flow structure pointer.
3824  * @param[in] prefix_layers
3825  *   Prefix flow layer flags.
3826  * @param[in] attr
3827  *   Flow rule attributes.
3828  * @param[in] items
3829  *   Pattern specification (list terminated by the END pattern item).
3830  * @param[in] actions
3831  *   Associated actions (list terminated by the END action).
3832  * @param[in] external
3833  *   This flow rule is created by request external to PMD.
3834  * @param[out] error
3835  *   Perform verbose error reporting if not NULL.
3836  * @return
3837  *   0 on success, negative value otherwise
3838  */
3839 static int
3840 flow_create_split_metadata(struct rte_eth_dev *dev,
3841 			   struct rte_flow *flow,
3842 			   uint64_t prefix_layers,
3843 			   const struct rte_flow_attr *attr,
3844 			   const struct rte_flow_item items[],
3845 			   const struct rte_flow_action actions[],
3846 			   bool external, struct rte_flow_error *error)
3847 {
3848 	struct mlx5_priv *priv = dev->data->dev_private;
3849 	struct mlx5_dev_config *config = &priv->config;
3850 	const struct rte_flow_action *qrss = NULL;
3851 	struct rte_flow_action *ext_actions = NULL;
3852 	struct mlx5_flow *dev_flow = NULL;
3853 	uint32_t qrss_id = 0;
3854 	int mtr_sfx = 0;
3855 	size_t act_size;
3856 	int actions_n;
3857 	int encap_idx;
3858 	int ret;
3859 
3860 	/* Check whether extensive metadata feature is engaged. */
3861 	if (!config->dv_flow_en ||
3862 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3863 	    !mlx5_flow_ext_mreg_supported(dev))
3864 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3865 					       attr, items, actions, external,
3866 					       error);
3867 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3868 							   &encap_idx);
3869 	if (qrss) {
3870 		/* Exclude hairpin flows from splitting. */
3871 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3872 			const struct rte_flow_action_queue *queue;
3873 
3874 			queue = qrss->conf;
3875 			if (mlx5_rxq_get_type(dev, queue->index) ==
3876 			    MLX5_RXQ_TYPE_HAIRPIN)
3877 				qrss = NULL;
3878 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3879 			const struct rte_flow_action_rss *rss;
3880 
3881 			rss = qrss->conf;
3882 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3883 			    MLX5_RXQ_TYPE_HAIRPIN)
3884 				qrss = NULL;
3885 		}
3886 	}
3887 	if (qrss) {
3888 		/* Check if it is in meter suffix table. */
3889 		mtr_sfx = attr->group == (attr->transfer ?
3890 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3891 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3892 		/*
3893 		 * Q/RSS action on NIC Rx should be split in order to pass by
3894 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3895 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3896 		 */
3897 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3898 			   sizeof(struct rte_flow_action_set_tag) +
3899 			   sizeof(struct rte_flow_action_jump);
3900 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3901 		if (!ext_actions)
3902 			return rte_flow_error_set(error, ENOMEM,
3903 						  RTE_FLOW_ERROR_TYPE_ACTION,
3904 						  NULL, "no memory to split "
3905 						  "metadata flow");
3906 		/*
3907 		 * If we are the suffix flow of meter, tag already exist.
3908 		 * Set the tag action to void.
3909 		 */
3910 		if (mtr_sfx)
3911 			ext_actions[qrss - actions].type =
3912 						RTE_FLOW_ACTION_TYPE_VOID;
3913 		else
3914 			ext_actions[qrss - actions].type =
3915 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3916 		/*
3917 		 * Create the new actions list with removed Q/RSS action
3918 		 * and appended set tag and jump to register copy table
3919 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3920 		 * in advance, because it is needed for set tag action.
3921 		 */
3922 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3923 						    qrss, actions_n, error);
3924 		if (!mtr_sfx && !qrss_id) {
3925 			ret = -rte_errno;
3926 			goto exit;
3927 		}
3928 	} else if (attr->egress && !attr->transfer) {
3929 		/*
3930 		 * All the actions on NIC Tx should have a metadata register
3931 		 * copy action to copy reg_a from WQE to reg_c[meta]
3932 		 */
3933 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3934 			   sizeof(struct mlx5_flow_action_copy_mreg);
3935 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3936 		if (!ext_actions)
3937 			return rte_flow_error_set(error, ENOMEM,
3938 						  RTE_FLOW_ERROR_TYPE_ACTION,
3939 						  NULL, "no memory to split "
3940 						  "metadata flow");
3941 		/* Create the action list appended with copy register. */
3942 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3943 					     actions_n, error, encap_idx);
3944 		if (ret < 0)
3945 			goto exit;
3946 	}
3947 	/* Add the unmodified original or prefix subflow. */
3948 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
3949 				      items, ext_actions ? ext_actions :
3950 				      actions, external, error);
3951 	if (ret < 0)
3952 		goto exit;
3953 	MLX5_ASSERT(dev_flow);
3954 	if (qrss) {
3955 		const struct rte_flow_attr q_attr = {
3956 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3957 			.ingress = 1,
3958 		};
3959 		/* Internal PMD action to set register. */
3960 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3961 			.data = qrss_id,
3962 			.id = 0,
3963 		};
3964 		struct rte_flow_item q_items[] = {
3965 			{
3966 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3967 				.spec = &q_tag_spec,
3968 				.last = NULL,
3969 				.mask = NULL,
3970 			},
3971 			{
3972 				.type = RTE_FLOW_ITEM_TYPE_END,
3973 			},
3974 		};
3975 		struct rte_flow_action q_actions[] = {
3976 			{
3977 				.type = qrss->type,
3978 				.conf = qrss->conf,
3979 			},
3980 			{
3981 				.type = RTE_FLOW_ACTION_TYPE_END,
3982 			},
3983 		};
3984 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
3985 
3986 		/*
3987 		 * Configure the tag item only if there is no meter subflow.
3988 		 * Since tag is already marked in the meter suffix subflow
3989 		 * we can just use the meter suffix items as is.
3990 		 */
3991 		if (qrss_id) {
3992 			/* Not meter subflow. */
3993 			MLX5_ASSERT(!mtr_sfx);
3994 			/*
3995 			 * Put unique id in prefix flow due to it is destroyed
3996 			 * after suffix flow and id will be freed after there
3997 			 * is no actual flows with this id and identifier
3998 			 * reallocation becomes possible (for example, for
3999 			 * other flows in other threads).
4000 			 */
4001 			dev_flow->qrss_id = qrss_id;
4002 			qrss_id = 0;
4003 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4004 						   error);
4005 			if (ret < 0)
4006 				goto exit;
4007 			q_tag_spec.id = ret;
4008 		}
4009 		dev_flow = NULL;
4010 		/* Add suffix subflow to execute Q/RSS. */
4011 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4012 					      &q_attr, mtr_sfx ? items :
4013 					      q_items, q_actions,
4014 					      external, error);
4015 		if (ret < 0)
4016 			goto exit;
4017 		MLX5_ASSERT(dev_flow);
4018 	}
4019 
4020 exit:
4021 	/*
4022 	 * We do not destroy the partially created sub_flows in case of error.
4023 	 * These ones are included into parent flow list and will be destroyed
4024 	 * by flow_drv_destroy.
4025 	 */
4026 	flow_qrss_free_id(dev, qrss_id);
4027 	rte_free(ext_actions);
4028 	return ret;
4029 }
4030 
4031 /**
4032  * The splitting for meter feature.
4033  *
4034  * - The meter flow will be split to two flows as prefix and
4035  *   suffix flow. The packets make sense only it pass the prefix
4036  *   meter action.
4037  *
4038  * - Reg_C_5 is used for the packet to match betweend prefix and
4039  *   suffix flow.
4040  *
4041  * @param dev
4042  *   Pointer to Ethernet device.
4043  * @param[in] flow
4044  *   Parent flow structure pointer.
4045  * @param[in] attr
4046  *   Flow rule attributes.
4047  * @param[in] items
4048  *   Pattern specification (list terminated by the END pattern item).
4049  * @param[in] actions
4050  *   Associated actions (list terminated by the END action).
4051  * @param[in] external
4052  *   This flow rule is created by request external to PMD.
4053  * @param[out] error
4054  *   Perform verbose error reporting if not NULL.
4055  * @return
4056  *   0 on success, negative value otherwise
4057  */
4058 static int
4059 flow_create_split_meter(struct rte_eth_dev *dev,
4060 			   struct rte_flow *flow,
4061 			   const struct rte_flow_attr *attr,
4062 			   const struct rte_flow_item items[],
4063 			   const struct rte_flow_action actions[],
4064 			   bool external, struct rte_flow_error *error)
4065 {
4066 	struct mlx5_priv *priv = dev->data->dev_private;
4067 	struct rte_flow_action *sfx_actions = NULL;
4068 	struct rte_flow_action *pre_actions = NULL;
4069 	struct rte_flow_item *sfx_items = NULL;
4070 	struct mlx5_flow *dev_flow = NULL;
4071 	struct rte_flow_attr sfx_attr = *attr;
4072 	uint32_t mtr = 0;
4073 	uint32_t mtr_tag_id = 0;
4074 	size_t act_size;
4075 	size_t item_size;
4076 	int actions_n = 0;
4077 	int ret;
4078 
4079 	if (priv->mtr_en)
4080 		actions_n = flow_check_meter_action(actions, &mtr);
4081 	if (mtr) {
4082 		/* The five prefix actions: meter, decap, encap, tag, end. */
4083 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4084 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4085 		/* tag, vlan, port id, end. */
4086 #define METER_SUFFIX_ITEM 4
4087 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4088 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4089 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4090 		if (!sfx_actions)
4091 			return rte_flow_error_set(error, ENOMEM,
4092 						  RTE_FLOW_ERROR_TYPE_ACTION,
4093 						  NULL, "no memory to split "
4094 						  "meter flow");
4095 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4096 			     act_size);
4097 		pre_actions = sfx_actions + actions_n;
4098 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4099 						   actions, sfx_actions,
4100 						   pre_actions);
4101 		if (!mtr_tag_id) {
4102 			ret = -rte_errno;
4103 			goto exit;
4104 		}
4105 		/* Add the prefix subflow. */
4106 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4107 					      items, pre_actions, external,
4108 					      error);
4109 		if (ret) {
4110 			ret = -rte_errno;
4111 			goto exit;
4112 		}
4113 		dev_flow->mtr_flow_id = mtr_tag_id;
4114 		/* Setting the sfx group atrr. */
4115 		sfx_attr.group = sfx_attr.transfer ?
4116 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4117 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4118 	}
4119 	/* Add the prefix subflow. */
4120 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4121 					 flow_get_prefix_layer_flags(dev_flow) :
4122 					 0, &sfx_attr,
4123 					 sfx_items ? sfx_items : items,
4124 					 sfx_actions ? sfx_actions : actions,
4125 					 external, error);
4126 exit:
4127 	if (sfx_actions)
4128 		rte_free(sfx_actions);
4129 	return ret;
4130 }
4131 
4132 /**
4133  * Split the flow to subflow set. The splitters might be linked
4134  * in the chain, like this:
4135  * flow_create_split_outer() calls:
4136  *   flow_create_split_meter() calls:
4137  *     flow_create_split_metadata(meter_subflow_0) calls:
4138  *       flow_create_split_inner(metadata_subflow_0)
4139  *       flow_create_split_inner(metadata_subflow_1)
4140  *       flow_create_split_inner(metadata_subflow_2)
4141  *     flow_create_split_metadata(meter_subflow_1) calls:
4142  *       flow_create_split_inner(metadata_subflow_0)
4143  *       flow_create_split_inner(metadata_subflow_1)
4144  *       flow_create_split_inner(metadata_subflow_2)
4145  *
4146  * This provide flexible way to add new levels of flow splitting.
4147  * The all of successfully created subflows are included to the
4148  * parent flow dev_flow list.
4149  *
4150  * @param dev
4151  *   Pointer to Ethernet device.
4152  * @param[in] flow
4153  *   Parent flow structure pointer.
4154  * @param[in] attr
4155  *   Flow rule attributes.
4156  * @param[in] items
4157  *   Pattern specification (list terminated by the END pattern item).
4158  * @param[in] actions
4159  *   Associated actions (list terminated by the END action).
4160  * @param[in] external
4161  *   This flow rule is created by request external to PMD.
4162  * @param[out] error
4163  *   Perform verbose error reporting if not NULL.
4164  * @return
4165  *   0 on success, negative value otherwise
4166  */
4167 static int
4168 flow_create_split_outer(struct rte_eth_dev *dev,
4169 			struct rte_flow *flow,
4170 			const struct rte_flow_attr *attr,
4171 			const struct rte_flow_item items[],
4172 			const struct rte_flow_action actions[],
4173 			bool external, struct rte_flow_error *error)
4174 {
4175 	int ret;
4176 
4177 	ret = flow_create_split_meter(dev, flow, attr, items,
4178 					 actions, external, error);
4179 	MLX5_ASSERT(ret <= 0);
4180 	return ret;
4181 }
4182 
4183 /**
4184  * Create a flow and add it to @p list.
4185  *
4186  * @param dev
4187  *   Pointer to Ethernet device.
4188  * @param list
4189  *   Pointer to a TAILQ flow list. If this parameter NULL,
4190  *   no list insertion occurred, flow is just created,
4191  *   this is caller's responsibility to track the
4192  *   created flow.
4193  * @param[in] attr
4194  *   Flow rule attributes.
4195  * @param[in] items
4196  *   Pattern specification (list terminated by the END pattern item).
4197  * @param[in] actions
4198  *   Associated actions (list terminated by the END action).
4199  * @param[in] external
4200  *   This flow rule is created by request external to PMD.
4201  * @param[out] error
4202  *   Perform verbose error reporting if not NULL.
4203  *
4204  * @return
4205  *   A flow on success, NULL otherwise and rte_errno is set.
4206  */
4207 static struct rte_flow *
4208 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4209 		 const struct rte_flow_attr *attr,
4210 		 const struct rte_flow_item items[],
4211 		 const struct rte_flow_action actions[],
4212 		 bool external, struct rte_flow_error *error)
4213 {
4214 	struct mlx5_priv *priv = dev->data->dev_private;
4215 	struct rte_flow *flow = NULL;
4216 	struct mlx5_flow *dev_flow;
4217 	const struct rte_flow_action_rss *rss;
4218 	union {
4219 		struct rte_flow_expand_rss buf;
4220 		uint8_t buffer[2048];
4221 	} expand_buffer;
4222 	union {
4223 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4224 		uint8_t buffer[2048];
4225 	} actions_rx;
4226 	union {
4227 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4228 		uint8_t buffer[2048];
4229 	} actions_hairpin_tx;
4230 	union {
4231 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4232 		uint8_t buffer[2048];
4233 	} items_tx;
4234 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4235 	const struct rte_flow_action *p_actions_rx = actions;
4236 	uint32_t i;
4237 	uint32_t flow_size;
4238 	int hairpin_flow = 0;
4239 	uint32_t hairpin_id = 0;
4240 	struct rte_flow_attr attr_tx = { .priority = 0 };
4241 	int ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4242 				    error);
4243 
4244 	if (ret < 0)
4245 		return NULL;
4246 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4247 	if (hairpin_flow > 0) {
4248 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4249 			rte_errno = EINVAL;
4250 			return NULL;
4251 		}
4252 		flow_hairpin_split(dev, actions, actions_rx.actions,
4253 				   actions_hairpin_tx.actions, items_tx.items,
4254 				   &hairpin_id);
4255 		p_actions_rx = actions_rx.actions;
4256 	}
4257 	flow_size = sizeof(struct rte_flow);
4258 	rss = flow_get_rss_action(p_actions_rx);
4259 	if (rss)
4260 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4261 					    sizeof(void *));
4262 	else
4263 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4264 	flow = rte_calloc(__func__, 1, flow_size, 0);
4265 	if (!flow) {
4266 		rte_errno = ENOMEM;
4267 		goto error_before_flow;
4268 	}
4269 	flow->drv_type = flow_get_drv_type(dev, attr);
4270 	if (hairpin_id != 0)
4271 		flow->hairpin_flow_id = hairpin_id;
4272 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4273 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4274 	flow->rss.queue = (void *)(flow + 1);
4275 	if (rss) {
4276 		/*
4277 		 * The following information is required by
4278 		 * mlx5_flow_hashfields_adjust() in advance.
4279 		 */
4280 		flow->rss.level = rss->level;
4281 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4282 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4283 	}
4284 	LIST_INIT(&flow->dev_flows);
4285 	if (rss && rss->types) {
4286 		unsigned int graph_root;
4287 
4288 		graph_root = find_graph_root(items, rss->level);
4289 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4290 					  items, rss->types,
4291 					  mlx5_support_expansion,
4292 					  graph_root);
4293 		MLX5_ASSERT(ret > 0 &&
4294 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4295 	} else {
4296 		buf->entries = 1;
4297 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4298 	}
4299 	for (i = 0; i < buf->entries; ++i) {
4300 		/*
4301 		 * The splitter may create multiple dev_flows,
4302 		 * depending on configuration. In the simplest
4303 		 * case it just creates unmodified original flow.
4304 		 */
4305 		ret = flow_create_split_outer(dev, flow, attr,
4306 					      buf->entry[i].pattern,
4307 					      p_actions_rx, external,
4308 					      error);
4309 		if (ret < 0)
4310 			goto error;
4311 	}
4312 	/* Create the tx flow. */
4313 	if (hairpin_flow) {
4314 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4315 		attr_tx.ingress = 0;
4316 		attr_tx.egress = 1;
4317 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4318 					    actions_hairpin_tx.actions, error);
4319 		if (!dev_flow)
4320 			goto error;
4321 		dev_flow->flow = flow;
4322 		dev_flow->external = 0;
4323 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4324 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4325 					 items_tx.items,
4326 					 actions_hairpin_tx.actions, error);
4327 		if (ret < 0)
4328 			goto error;
4329 	}
4330 	/*
4331 	 * Update the metadata register copy table. If extensive
4332 	 * metadata feature is enabled and registers are supported
4333 	 * we might create the extra rte_flow for each unique
4334 	 * MARK/FLAG action ID.
4335 	 *
4336 	 * The table is updated for ingress Flows only, because
4337 	 * the egress Flows belong to the different device and
4338 	 * copy table should be updated in peer NIC Rx domain.
4339 	 */
4340 	if (attr->ingress &&
4341 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4342 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4343 		if (ret)
4344 			goto error;
4345 	}
4346 	if (dev->data->dev_started) {
4347 		ret = flow_drv_apply(dev, flow, error);
4348 		if (ret < 0)
4349 			goto error;
4350 	}
4351 	if (list)
4352 		TAILQ_INSERT_TAIL(list, flow, next);
4353 	flow_rxq_flags_set(dev, flow);
4354 	return flow;
4355 error_before_flow:
4356 	if (hairpin_id)
4357 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4358 				     hairpin_id);
4359 	return NULL;
4360 error:
4361 	MLX5_ASSERT(flow);
4362 	flow_mreg_del_copy_action(dev, flow);
4363 	ret = rte_errno; /* Save rte_errno before cleanup. */
4364 	if (flow->hairpin_flow_id)
4365 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4366 				     flow->hairpin_flow_id);
4367 	MLX5_ASSERT(flow);
4368 	flow_drv_destroy(dev, flow);
4369 	rte_free(flow);
4370 	rte_errno = ret; /* Restore rte_errno. */
4371 	return NULL;
4372 }
4373 
4374 /**
4375  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4376  * incoming packets to table 1.
4377  *
4378  * Other flow rules, requested for group n, will be created in
4379  * e-switch table n+1.
4380  * Jump action to e-switch group n will be created to group n+1.
4381  *
4382  * Used when working in switchdev mode, to utilise advantages of table 1
4383  * and above.
4384  *
4385  * @param dev
4386  *   Pointer to Ethernet device.
4387  *
4388  * @return
4389  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4390  */
4391 struct rte_flow *
4392 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4393 {
4394 	const struct rte_flow_attr attr = {
4395 		.group = 0,
4396 		.priority = 0,
4397 		.ingress = 1,
4398 		.egress = 0,
4399 		.transfer = 1,
4400 	};
4401 	const struct rte_flow_item pattern = {
4402 		.type = RTE_FLOW_ITEM_TYPE_END,
4403 	};
4404 	struct rte_flow_action_jump jump = {
4405 		.group = 1,
4406 	};
4407 	const struct rte_flow_action actions[] = {
4408 		{
4409 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4410 			.conf = &jump,
4411 		},
4412 		{
4413 			.type = RTE_FLOW_ACTION_TYPE_END,
4414 		},
4415 	};
4416 	struct mlx5_priv *priv = dev->data->dev_private;
4417 	struct rte_flow_error error;
4418 
4419 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4420 				actions, false, &error);
4421 }
4422 
4423 /**
4424  * Create a flow.
4425  *
4426  * @see rte_flow_create()
4427  * @see rte_flow_ops
4428  */
4429 struct rte_flow *
4430 mlx5_flow_create(struct rte_eth_dev *dev,
4431 		 const struct rte_flow_attr *attr,
4432 		 const struct rte_flow_item items[],
4433 		 const struct rte_flow_action actions[],
4434 		 struct rte_flow_error *error)
4435 {
4436 	struct mlx5_priv *priv = dev->data->dev_private;
4437 
4438 	return flow_list_create(dev, &priv->flows,
4439 				attr, items, actions, true, error);
4440 }
4441 
4442 /**
4443  * Destroy a flow in a list.
4444  *
4445  * @param dev
4446  *   Pointer to Ethernet device.
4447  * @param list
4448  *   Pointer to a TAILQ flow list. If this parameter NULL,
4449  *   there is no flow removal from the list.
4450  * @param[in] flow
4451  *   Flow to destroy.
4452  */
4453 static void
4454 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4455 		  struct rte_flow *flow)
4456 {
4457 	struct mlx5_priv *priv = dev->data->dev_private;
4458 
4459 	/*
4460 	 * Update RX queue flags only if port is started, otherwise it is
4461 	 * already clean.
4462 	 */
4463 	if (dev->data->dev_started)
4464 		flow_rxq_flags_trim(dev, flow);
4465 	if (flow->hairpin_flow_id)
4466 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4467 				     flow->hairpin_flow_id);
4468 	flow_drv_destroy(dev, flow);
4469 	if (list)
4470 		TAILQ_REMOVE(list, flow, next);
4471 	flow_mreg_del_copy_action(dev, flow);
4472 	rte_free(flow->fdir);
4473 	rte_free(flow);
4474 }
4475 
4476 /**
4477  * Destroy all flows.
4478  *
4479  * @param dev
4480  *   Pointer to Ethernet device.
4481  * @param list
4482  *   Pointer to a TAILQ flow list.
4483  */
4484 void
4485 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4486 {
4487 	while (!TAILQ_EMPTY(list)) {
4488 		struct rte_flow *flow;
4489 
4490 		flow = TAILQ_FIRST(list);
4491 		flow_list_destroy(dev, list, flow);
4492 	}
4493 }
4494 
4495 /**
4496  * Remove all flows.
4497  *
4498  * @param dev
4499  *   Pointer to Ethernet device.
4500  * @param list
4501  *   Pointer to a TAILQ flow list.
4502  */
4503 void
4504 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4505 {
4506 	struct rte_flow *flow;
4507 
4508 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4509 		flow_drv_remove(dev, flow);
4510 		flow_mreg_stop_copy_action(dev, flow);
4511 	}
4512 	flow_mreg_del_default_copy_action(dev);
4513 	flow_rxq_flags_clear(dev);
4514 }
4515 
4516 /**
4517  * Add all flows.
4518  *
4519  * @param dev
4520  *   Pointer to Ethernet device.
4521  * @param list
4522  *   Pointer to a TAILQ flow list.
4523  *
4524  * @return
4525  *   0 on success, a negative errno value otherwise and rte_errno is set.
4526  */
4527 int
4528 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4529 {
4530 	struct rte_flow *flow;
4531 	struct rte_flow_error error;
4532 	int ret = 0;
4533 
4534 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4535 	ret = flow_mreg_add_default_copy_action(dev, &error);
4536 	if (ret < 0)
4537 		return -rte_errno;
4538 	/* Apply Flows created by application. */
4539 	TAILQ_FOREACH(flow, list, next) {
4540 		ret = flow_mreg_start_copy_action(dev, flow);
4541 		if (ret < 0)
4542 			goto error;
4543 		ret = flow_drv_apply(dev, flow, &error);
4544 		if (ret < 0)
4545 			goto error;
4546 		flow_rxq_flags_set(dev, flow);
4547 	}
4548 	return 0;
4549 error:
4550 	ret = rte_errno; /* Save rte_errno before cleanup. */
4551 	mlx5_flow_stop(dev, list);
4552 	rte_errno = ret; /* Restore rte_errno. */
4553 	return -rte_errno;
4554 }
4555 
4556 /**
4557  * Verify the flow list is empty
4558  *
4559  * @param dev
4560  *  Pointer to Ethernet device.
4561  *
4562  * @return the number of flows not released.
4563  */
4564 int
4565 mlx5_flow_verify(struct rte_eth_dev *dev)
4566 {
4567 	struct mlx5_priv *priv = dev->data->dev_private;
4568 	struct rte_flow *flow;
4569 	int ret = 0;
4570 
4571 	TAILQ_FOREACH(flow, &priv->flows, next) {
4572 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4573 			dev->data->port_id, (void *)flow);
4574 		++ret;
4575 	}
4576 	return ret;
4577 }
4578 
4579 /**
4580  * Enable default hairpin egress flow.
4581  *
4582  * @param dev
4583  *   Pointer to Ethernet device.
4584  * @param queue
4585  *   The queue index.
4586  *
4587  * @return
4588  *   0 on success, a negative errno value otherwise and rte_errno is set.
4589  */
4590 int
4591 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4592 			    uint32_t queue)
4593 {
4594 	struct mlx5_priv *priv = dev->data->dev_private;
4595 	const struct rte_flow_attr attr = {
4596 		.egress = 1,
4597 		.priority = 0,
4598 	};
4599 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4600 		.queue = queue,
4601 	};
4602 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4603 		.queue = UINT32_MAX,
4604 	};
4605 	struct rte_flow_item items[] = {
4606 		{
4607 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4608 			.spec = &queue_spec,
4609 			.last = NULL,
4610 			.mask = &queue_mask,
4611 		},
4612 		{
4613 			.type = RTE_FLOW_ITEM_TYPE_END,
4614 		},
4615 	};
4616 	struct rte_flow_action_jump jump = {
4617 		.group = MLX5_HAIRPIN_TX_TABLE,
4618 	};
4619 	struct rte_flow_action actions[2];
4620 	struct rte_flow *flow;
4621 	struct rte_flow_error error;
4622 
4623 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4624 	actions[0].conf = &jump;
4625 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4626 	flow = flow_list_create(dev, &priv->ctrl_flows,
4627 				&attr, items, actions, false, &error);
4628 	if (!flow) {
4629 		DRV_LOG(DEBUG,
4630 			"Failed to create ctrl flow: rte_errno(%d),"
4631 			" type(%d), message(%s)",
4632 			rte_errno, error.type,
4633 			error.message ? error.message : " (no stated reason)");
4634 		return -rte_errno;
4635 	}
4636 	return 0;
4637 }
4638 
4639 /**
4640  * Enable a control flow configured from the control plane.
4641  *
4642  * @param dev
4643  *   Pointer to Ethernet device.
4644  * @param eth_spec
4645  *   An Ethernet flow spec to apply.
4646  * @param eth_mask
4647  *   An Ethernet flow mask to apply.
4648  * @param vlan_spec
4649  *   A VLAN flow spec to apply.
4650  * @param vlan_mask
4651  *   A VLAN flow mask to apply.
4652  *
4653  * @return
4654  *   0 on success, a negative errno value otherwise and rte_errno is set.
4655  */
4656 int
4657 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4658 		    struct rte_flow_item_eth *eth_spec,
4659 		    struct rte_flow_item_eth *eth_mask,
4660 		    struct rte_flow_item_vlan *vlan_spec,
4661 		    struct rte_flow_item_vlan *vlan_mask)
4662 {
4663 	struct mlx5_priv *priv = dev->data->dev_private;
4664 	const struct rte_flow_attr attr = {
4665 		.ingress = 1,
4666 		.priority = MLX5_FLOW_PRIO_RSVD,
4667 	};
4668 	struct rte_flow_item items[] = {
4669 		{
4670 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4671 			.spec = eth_spec,
4672 			.last = NULL,
4673 			.mask = eth_mask,
4674 		},
4675 		{
4676 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4677 					      RTE_FLOW_ITEM_TYPE_END,
4678 			.spec = vlan_spec,
4679 			.last = NULL,
4680 			.mask = vlan_mask,
4681 		},
4682 		{
4683 			.type = RTE_FLOW_ITEM_TYPE_END,
4684 		},
4685 	};
4686 	uint16_t queue[priv->reta_idx_n];
4687 	struct rte_flow_action_rss action_rss = {
4688 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4689 		.level = 0,
4690 		.types = priv->rss_conf.rss_hf,
4691 		.key_len = priv->rss_conf.rss_key_len,
4692 		.queue_num = priv->reta_idx_n,
4693 		.key = priv->rss_conf.rss_key,
4694 		.queue = queue,
4695 	};
4696 	struct rte_flow_action actions[] = {
4697 		{
4698 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4699 			.conf = &action_rss,
4700 		},
4701 		{
4702 			.type = RTE_FLOW_ACTION_TYPE_END,
4703 		},
4704 	};
4705 	struct rte_flow *flow;
4706 	struct rte_flow_error error;
4707 	unsigned int i;
4708 
4709 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4710 		return 0;
4711 	}
4712 	for (i = 0; i != priv->reta_idx_n; ++i)
4713 		queue[i] = (*priv->reta_idx)[i];
4714 	flow = flow_list_create(dev, &priv->ctrl_flows,
4715 				&attr, items, actions, false, &error);
4716 	if (!flow)
4717 		return -rte_errno;
4718 	return 0;
4719 }
4720 
4721 /**
4722  * Enable a flow control configured from the control plane.
4723  *
4724  * @param dev
4725  *   Pointer to Ethernet device.
4726  * @param eth_spec
4727  *   An Ethernet flow spec to apply.
4728  * @param eth_mask
4729  *   An Ethernet flow mask to apply.
4730  *
4731  * @return
4732  *   0 on success, a negative errno value otherwise and rte_errno is set.
4733  */
4734 int
4735 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4736 	       struct rte_flow_item_eth *eth_spec,
4737 	       struct rte_flow_item_eth *eth_mask)
4738 {
4739 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4740 }
4741 
4742 /**
4743  * Destroy a flow.
4744  *
4745  * @see rte_flow_destroy()
4746  * @see rte_flow_ops
4747  */
4748 int
4749 mlx5_flow_destroy(struct rte_eth_dev *dev,
4750 		  struct rte_flow *flow,
4751 		  struct rte_flow_error *error __rte_unused)
4752 {
4753 	struct mlx5_priv *priv = dev->data->dev_private;
4754 
4755 	flow_list_destroy(dev, &priv->flows, flow);
4756 	return 0;
4757 }
4758 
4759 /**
4760  * Destroy all flows.
4761  *
4762  * @see rte_flow_flush()
4763  * @see rte_flow_ops
4764  */
4765 int
4766 mlx5_flow_flush(struct rte_eth_dev *dev,
4767 		struct rte_flow_error *error __rte_unused)
4768 {
4769 	struct mlx5_priv *priv = dev->data->dev_private;
4770 
4771 	mlx5_flow_list_flush(dev, &priv->flows);
4772 	return 0;
4773 }
4774 
4775 /**
4776  * Isolated mode.
4777  *
4778  * @see rte_flow_isolate()
4779  * @see rte_flow_ops
4780  */
4781 int
4782 mlx5_flow_isolate(struct rte_eth_dev *dev,
4783 		  int enable,
4784 		  struct rte_flow_error *error)
4785 {
4786 	struct mlx5_priv *priv = dev->data->dev_private;
4787 
4788 	if (dev->data->dev_started) {
4789 		rte_flow_error_set(error, EBUSY,
4790 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4791 				   NULL,
4792 				   "port must be stopped first");
4793 		return -rte_errno;
4794 	}
4795 	priv->isolated = !!enable;
4796 	if (enable)
4797 		dev->dev_ops = &mlx5_dev_ops_isolate;
4798 	else
4799 		dev->dev_ops = &mlx5_dev_ops;
4800 	return 0;
4801 }
4802 
4803 /**
4804  * Query a flow.
4805  *
4806  * @see rte_flow_query()
4807  * @see rte_flow_ops
4808  */
4809 static int
4810 flow_drv_query(struct rte_eth_dev *dev,
4811 	       struct rte_flow *flow,
4812 	       const struct rte_flow_action *actions,
4813 	       void *data,
4814 	       struct rte_flow_error *error)
4815 {
4816 	const struct mlx5_flow_driver_ops *fops;
4817 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4818 
4819 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4820 	fops = flow_get_drv_ops(ftype);
4821 
4822 	return fops->query(dev, flow, actions, data, error);
4823 }
4824 
4825 /**
4826  * Query a flow.
4827  *
4828  * @see rte_flow_query()
4829  * @see rte_flow_ops
4830  */
4831 int
4832 mlx5_flow_query(struct rte_eth_dev *dev,
4833 		struct rte_flow *flow,
4834 		const struct rte_flow_action *actions,
4835 		void *data,
4836 		struct rte_flow_error *error)
4837 {
4838 	int ret;
4839 
4840 	ret = flow_drv_query(dev, flow, actions, data, error);
4841 	if (ret < 0)
4842 		return ret;
4843 	return 0;
4844 }
4845 
4846 /**
4847  * Convert a flow director filter to a generic flow.
4848  *
4849  * @param dev
4850  *   Pointer to Ethernet device.
4851  * @param fdir_filter
4852  *   Flow director filter to add.
4853  * @param attributes
4854  *   Generic flow parameters structure.
4855  *
4856  * @return
4857  *   0 on success, a negative errno value otherwise and rte_errno is set.
4858  */
4859 static int
4860 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4861 			 const struct rte_eth_fdir_filter *fdir_filter,
4862 			 struct mlx5_fdir *attributes)
4863 {
4864 	struct mlx5_priv *priv = dev->data->dev_private;
4865 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4866 	const struct rte_eth_fdir_masks *mask =
4867 		&dev->data->dev_conf.fdir_conf.mask;
4868 
4869 	/* Validate queue number. */
4870 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4871 		DRV_LOG(ERR, "port %u invalid queue number %d",
4872 			dev->data->port_id, fdir_filter->action.rx_queue);
4873 		rte_errno = EINVAL;
4874 		return -rte_errno;
4875 	}
4876 	attributes->attr.ingress = 1;
4877 	attributes->items[0] = (struct rte_flow_item) {
4878 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4879 		.spec = &attributes->l2,
4880 		.mask = &attributes->l2_mask,
4881 	};
4882 	switch (fdir_filter->action.behavior) {
4883 	case RTE_ETH_FDIR_ACCEPT:
4884 		attributes->actions[0] = (struct rte_flow_action){
4885 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4886 			.conf = &attributes->queue,
4887 		};
4888 		break;
4889 	case RTE_ETH_FDIR_REJECT:
4890 		attributes->actions[0] = (struct rte_flow_action){
4891 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4892 		};
4893 		break;
4894 	default:
4895 		DRV_LOG(ERR, "port %u invalid behavior %d",
4896 			dev->data->port_id,
4897 			fdir_filter->action.behavior);
4898 		rte_errno = ENOTSUP;
4899 		return -rte_errno;
4900 	}
4901 	attributes->queue.index = fdir_filter->action.rx_queue;
4902 	/* Handle L3. */
4903 	switch (fdir_filter->input.flow_type) {
4904 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4905 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4906 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4907 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4908 			.src_addr = input->flow.ip4_flow.src_ip,
4909 			.dst_addr = input->flow.ip4_flow.dst_ip,
4910 			.time_to_live = input->flow.ip4_flow.ttl,
4911 			.type_of_service = input->flow.ip4_flow.tos,
4912 		};
4913 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4914 			.src_addr = mask->ipv4_mask.src_ip,
4915 			.dst_addr = mask->ipv4_mask.dst_ip,
4916 			.time_to_live = mask->ipv4_mask.ttl,
4917 			.type_of_service = mask->ipv4_mask.tos,
4918 			.next_proto_id = mask->ipv4_mask.proto,
4919 		};
4920 		attributes->items[1] = (struct rte_flow_item){
4921 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
4922 			.spec = &attributes->l3,
4923 			.mask = &attributes->l3_mask,
4924 		};
4925 		break;
4926 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4927 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4928 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4929 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4930 			.hop_limits = input->flow.ipv6_flow.hop_limits,
4931 			.proto = input->flow.ipv6_flow.proto,
4932 		};
4933 
4934 		memcpy(attributes->l3.ipv6.hdr.src_addr,
4935 		       input->flow.ipv6_flow.src_ip,
4936 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4937 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
4938 		       input->flow.ipv6_flow.dst_ip,
4939 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4940 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4941 		       mask->ipv6_mask.src_ip,
4942 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4943 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4944 		       mask->ipv6_mask.dst_ip,
4945 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4946 		attributes->items[1] = (struct rte_flow_item){
4947 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
4948 			.spec = &attributes->l3,
4949 			.mask = &attributes->l3_mask,
4950 		};
4951 		break;
4952 	default:
4953 		DRV_LOG(ERR, "port %u invalid flow type%d",
4954 			dev->data->port_id, fdir_filter->input.flow_type);
4955 		rte_errno = ENOTSUP;
4956 		return -rte_errno;
4957 	}
4958 	/* Handle L4. */
4959 	switch (fdir_filter->input.flow_type) {
4960 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4961 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4962 			.src_port = input->flow.udp4_flow.src_port,
4963 			.dst_port = input->flow.udp4_flow.dst_port,
4964 		};
4965 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4966 			.src_port = mask->src_port_mask,
4967 			.dst_port = mask->dst_port_mask,
4968 		};
4969 		attributes->items[2] = (struct rte_flow_item){
4970 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4971 			.spec = &attributes->l4,
4972 			.mask = &attributes->l4_mask,
4973 		};
4974 		break;
4975 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4976 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4977 			.src_port = input->flow.tcp4_flow.src_port,
4978 			.dst_port = input->flow.tcp4_flow.dst_port,
4979 		};
4980 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4981 			.src_port = mask->src_port_mask,
4982 			.dst_port = mask->dst_port_mask,
4983 		};
4984 		attributes->items[2] = (struct rte_flow_item){
4985 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4986 			.spec = &attributes->l4,
4987 			.mask = &attributes->l4_mask,
4988 		};
4989 		break;
4990 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4991 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4992 			.src_port = input->flow.udp6_flow.src_port,
4993 			.dst_port = input->flow.udp6_flow.dst_port,
4994 		};
4995 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4996 			.src_port = mask->src_port_mask,
4997 			.dst_port = mask->dst_port_mask,
4998 		};
4999 		attributes->items[2] = (struct rte_flow_item){
5000 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5001 			.spec = &attributes->l4,
5002 			.mask = &attributes->l4_mask,
5003 		};
5004 		break;
5005 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5006 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5007 			.src_port = input->flow.tcp6_flow.src_port,
5008 			.dst_port = input->flow.tcp6_flow.dst_port,
5009 		};
5010 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5011 			.src_port = mask->src_port_mask,
5012 			.dst_port = mask->dst_port_mask,
5013 		};
5014 		attributes->items[2] = (struct rte_flow_item){
5015 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5016 			.spec = &attributes->l4,
5017 			.mask = &attributes->l4_mask,
5018 		};
5019 		break;
5020 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5021 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5022 		break;
5023 	default:
5024 		DRV_LOG(ERR, "port %u invalid flow type%d",
5025 			dev->data->port_id, fdir_filter->input.flow_type);
5026 		rte_errno = ENOTSUP;
5027 		return -rte_errno;
5028 	}
5029 	return 0;
5030 }
5031 
5032 #define FLOW_FDIR_CMP(f1, f2, fld) \
5033 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5034 
5035 /**
5036  * Compare two FDIR flows. If items and actions are identical, the two flows are
5037  * regarded as same.
5038  *
5039  * @param dev
5040  *   Pointer to Ethernet device.
5041  * @param f1
5042  *   FDIR flow to compare.
5043  * @param f2
5044  *   FDIR flow to compare.
5045  *
5046  * @return
5047  *   Zero on match, 1 otherwise.
5048  */
5049 static int
5050 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5051 {
5052 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5053 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5054 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5055 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5056 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5057 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5058 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5059 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5060 		return 1;
5061 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5062 	    FLOW_FDIR_CMP(f1, f2, queue))
5063 		return 1;
5064 	return 0;
5065 }
5066 
5067 /**
5068  * Search device flow list to find out a matched FDIR flow.
5069  *
5070  * @param dev
5071  *   Pointer to Ethernet device.
5072  * @param fdir_flow
5073  *   FDIR flow to lookup.
5074  *
5075  * @return
5076  *   Pointer of flow if found, NULL otherwise.
5077  */
5078 static struct rte_flow *
5079 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5080 {
5081 	struct mlx5_priv *priv = dev->data->dev_private;
5082 	struct rte_flow *flow = NULL;
5083 
5084 	MLX5_ASSERT(fdir_flow);
5085 	TAILQ_FOREACH(flow, &priv->flows, next) {
5086 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5087 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5088 				dev->data->port_id, (void *)flow);
5089 			break;
5090 		}
5091 	}
5092 	return flow;
5093 }
5094 
5095 /**
5096  * Add new flow director filter and store it in list.
5097  *
5098  * @param dev
5099  *   Pointer to Ethernet device.
5100  * @param fdir_filter
5101  *   Flow director filter to add.
5102  *
5103  * @return
5104  *   0 on success, a negative errno value otherwise and rte_errno is set.
5105  */
5106 static int
5107 flow_fdir_filter_add(struct rte_eth_dev *dev,
5108 		     const struct rte_eth_fdir_filter *fdir_filter)
5109 {
5110 	struct mlx5_priv *priv = dev->data->dev_private;
5111 	struct mlx5_fdir *fdir_flow;
5112 	struct rte_flow *flow;
5113 	int ret;
5114 
5115 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5116 	if (!fdir_flow) {
5117 		rte_errno = ENOMEM;
5118 		return -rte_errno;
5119 	}
5120 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5121 	if (ret)
5122 		goto error;
5123 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5124 	if (flow) {
5125 		rte_errno = EEXIST;
5126 		goto error;
5127 	}
5128 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5129 				fdir_flow->items, fdir_flow->actions, true,
5130 				NULL);
5131 	if (!flow)
5132 		goto error;
5133 	MLX5_ASSERT(!flow->fdir);
5134 	flow->fdir = fdir_flow;
5135 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5136 		dev->data->port_id, (void *)flow);
5137 	return 0;
5138 error:
5139 	rte_free(fdir_flow);
5140 	return -rte_errno;
5141 }
5142 
5143 /**
5144  * Delete specific filter.
5145  *
5146  * @param dev
5147  *   Pointer to Ethernet device.
5148  * @param fdir_filter
5149  *   Filter to be deleted.
5150  *
5151  * @return
5152  *   0 on success, a negative errno value otherwise and rte_errno is set.
5153  */
5154 static int
5155 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5156 			const struct rte_eth_fdir_filter *fdir_filter)
5157 {
5158 	struct mlx5_priv *priv = dev->data->dev_private;
5159 	struct rte_flow *flow;
5160 	struct mlx5_fdir fdir_flow = {
5161 		.attr.group = 0,
5162 	};
5163 	int ret;
5164 
5165 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5166 	if (ret)
5167 		return -rte_errno;
5168 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5169 	if (!flow) {
5170 		rte_errno = ENOENT;
5171 		return -rte_errno;
5172 	}
5173 	flow_list_destroy(dev, &priv->flows, flow);
5174 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5175 		dev->data->port_id, (void *)flow);
5176 	return 0;
5177 }
5178 
5179 /**
5180  * Update queue for specific filter.
5181  *
5182  * @param dev
5183  *   Pointer to Ethernet device.
5184  * @param fdir_filter
5185  *   Filter to be updated.
5186  *
5187  * @return
5188  *   0 on success, a negative errno value otherwise and rte_errno is set.
5189  */
5190 static int
5191 flow_fdir_filter_update(struct rte_eth_dev *dev,
5192 			const struct rte_eth_fdir_filter *fdir_filter)
5193 {
5194 	int ret;
5195 
5196 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5197 	if (ret)
5198 		return ret;
5199 	return flow_fdir_filter_add(dev, fdir_filter);
5200 }
5201 
5202 /**
5203  * Flush all filters.
5204  *
5205  * @param dev
5206  *   Pointer to Ethernet device.
5207  */
5208 static void
5209 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5210 {
5211 	struct mlx5_priv *priv = dev->data->dev_private;
5212 
5213 	mlx5_flow_list_flush(dev, &priv->flows);
5214 }
5215 
5216 /**
5217  * Get flow director information.
5218  *
5219  * @param dev
5220  *   Pointer to Ethernet device.
5221  * @param[out] fdir_info
5222  *   Resulting flow director information.
5223  */
5224 static void
5225 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5226 {
5227 	struct rte_eth_fdir_masks *mask =
5228 		&dev->data->dev_conf.fdir_conf.mask;
5229 
5230 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5231 	fdir_info->guarant_spc = 0;
5232 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5233 	fdir_info->max_flexpayload = 0;
5234 	fdir_info->flow_types_mask[0] = 0;
5235 	fdir_info->flex_payload_unit = 0;
5236 	fdir_info->max_flex_payload_segment_num = 0;
5237 	fdir_info->flex_payload_limit = 0;
5238 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5239 }
5240 
5241 /**
5242  * Deal with flow director operations.
5243  *
5244  * @param dev
5245  *   Pointer to Ethernet device.
5246  * @param filter_op
5247  *   Operation to perform.
5248  * @param arg
5249  *   Pointer to operation-specific structure.
5250  *
5251  * @return
5252  *   0 on success, a negative errno value otherwise and rte_errno is set.
5253  */
5254 static int
5255 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5256 		    void *arg)
5257 {
5258 	enum rte_fdir_mode fdir_mode =
5259 		dev->data->dev_conf.fdir_conf.mode;
5260 
5261 	if (filter_op == RTE_ETH_FILTER_NOP)
5262 		return 0;
5263 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5264 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5265 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5266 			dev->data->port_id, fdir_mode);
5267 		rte_errno = EINVAL;
5268 		return -rte_errno;
5269 	}
5270 	switch (filter_op) {
5271 	case RTE_ETH_FILTER_ADD:
5272 		return flow_fdir_filter_add(dev, arg);
5273 	case RTE_ETH_FILTER_UPDATE:
5274 		return flow_fdir_filter_update(dev, arg);
5275 	case RTE_ETH_FILTER_DELETE:
5276 		return flow_fdir_filter_delete(dev, arg);
5277 	case RTE_ETH_FILTER_FLUSH:
5278 		flow_fdir_filter_flush(dev);
5279 		break;
5280 	case RTE_ETH_FILTER_INFO:
5281 		flow_fdir_info_get(dev, arg);
5282 		break;
5283 	default:
5284 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5285 			dev->data->port_id, filter_op);
5286 		rte_errno = EINVAL;
5287 		return -rte_errno;
5288 	}
5289 	return 0;
5290 }
5291 
5292 /**
5293  * Manage filter operations.
5294  *
5295  * @param dev
5296  *   Pointer to Ethernet device structure.
5297  * @param filter_type
5298  *   Filter type.
5299  * @param filter_op
5300  *   Operation to perform.
5301  * @param arg
5302  *   Pointer to operation-specific structure.
5303  *
5304  * @return
5305  *   0 on success, a negative errno value otherwise and rte_errno is set.
5306  */
5307 int
5308 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5309 		     enum rte_filter_type filter_type,
5310 		     enum rte_filter_op filter_op,
5311 		     void *arg)
5312 {
5313 	switch (filter_type) {
5314 	case RTE_ETH_FILTER_GENERIC:
5315 		if (filter_op != RTE_ETH_FILTER_GET) {
5316 			rte_errno = EINVAL;
5317 			return -rte_errno;
5318 		}
5319 		*(const void **)arg = &mlx5_flow_ops;
5320 		return 0;
5321 	case RTE_ETH_FILTER_FDIR:
5322 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5323 	default:
5324 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5325 			dev->data->port_id, filter_type);
5326 		rte_errno = ENOTSUP;
5327 		return -rte_errno;
5328 	}
5329 	return 0;
5330 }
5331 
5332 /**
5333  * Create the needed meter and suffix tables.
5334  *
5335  * @param[in] dev
5336  *   Pointer to Ethernet device.
5337  * @param[in] fm
5338  *   Pointer to the flow meter.
5339  *
5340  * @return
5341  *   Pointer to table set on success, NULL otherwise.
5342  */
5343 struct mlx5_meter_domains_infos *
5344 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5345 			  const struct mlx5_flow_meter *fm)
5346 {
5347 	const struct mlx5_flow_driver_ops *fops;
5348 
5349 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5350 	return fops->create_mtr_tbls(dev, fm);
5351 }
5352 
5353 /**
5354  * Destroy the meter table set.
5355  *
5356  * @param[in] dev
5357  *   Pointer to Ethernet device.
5358  * @param[in] tbl
5359  *   Pointer to the meter table set.
5360  *
5361  * @return
5362  *   0 on success.
5363  */
5364 int
5365 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5366 			   struct mlx5_meter_domains_infos *tbls)
5367 {
5368 	const struct mlx5_flow_driver_ops *fops;
5369 
5370 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5371 	return fops->destroy_mtr_tbls(dev, tbls);
5372 }
5373 
5374 /**
5375  * Create policer rules.
5376  *
5377  * @param[in] dev
5378  *   Pointer to Ethernet device.
5379  * @param[in] fm
5380  *   Pointer to flow meter structure.
5381  * @param[in] attr
5382  *   Pointer to flow attributes.
5383  *
5384  * @return
5385  *   0 on success, -1 otherwise.
5386  */
5387 int
5388 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5389 			       struct mlx5_flow_meter *fm,
5390 			       const struct rte_flow_attr *attr)
5391 {
5392 	const struct mlx5_flow_driver_ops *fops;
5393 
5394 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5395 	return fops->create_policer_rules(dev, fm, attr);
5396 }
5397 
5398 /**
5399  * Destroy policer rules.
5400  *
5401  * @param[in] fm
5402  *   Pointer to flow meter structure.
5403  * @param[in] attr
5404  *   Pointer to flow attributes.
5405  *
5406  * @return
5407  *   0 on success, -1 otherwise.
5408  */
5409 int
5410 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5411 				struct mlx5_flow_meter *fm,
5412 				const struct rte_flow_attr *attr)
5413 {
5414 	const struct mlx5_flow_driver_ops *fops;
5415 
5416 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5417 	return fops->destroy_policer_rules(dev, fm, attr);
5418 }
5419 
5420 /**
5421  * Allocate a counter.
5422  *
5423  * @param[in] dev
5424  *   Pointer to Ethernet device structure.
5425  *
5426  * @return
5427  *   Pointer to allocated counter  on success, NULL otherwise.
5428  */
5429 struct mlx5_flow_counter *
5430 mlx5_counter_alloc(struct rte_eth_dev *dev)
5431 {
5432 	const struct mlx5_flow_driver_ops *fops;
5433 	struct rte_flow_attr attr = { .transfer = 0 };
5434 
5435 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5436 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5437 		return fops->counter_alloc(dev);
5438 	}
5439 	DRV_LOG(ERR,
5440 		"port %u counter allocate is not supported.",
5441 		 dev->data->port_id);
5442 	return NULL;
5443 }
5444 
5445 /**
5446  * Free a counter.
5447  *
5448  * @param[in] dev
5449  *   Pointer to Ethernet device structure.
5450  * @param[in] cnt
5451  *   Pointer to counter to be free.
5452  */
5453 void
5454 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5455 {
5456 	const struct mlx5_flow_driver_ops *fops;
5457 	struct rte_flow_attr attr = { .transfer = 0 };
5458 
5459 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5460 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5461 		fops->counter_free(dev, cnt);
5462 		return;
5463 	}
5464 	DRV_LOG(ERR,
5465 		"port %u counter free is not supported.",
5466 		 dev->data->port_id);
5467 }
5468 
5469 /**
5470  * Query counter statistics.
5471  *
5472  * @param[in] dev
5473  *   Pointer to Ethernet device structure.
5474  * @param[in] cnt
5475  *   Pointer to counter to query.
5476  * @param[in] clear
5477  *   Set to clear counter statistics.
5478  * @param[out] pkts
5479  *   The counter hits packets number to save.
5480  * @param[out] bytes
5481  *   The counter hits bytes number to save.
5482  *
5483  * @return
5484  *   0 on success, a negative errno value otherwise.
5485  */
5486 int
5487 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5488 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5489 {
5490 	const struct mlx5_flow_driver_ops *fops;
5491 	struct rte_flow_attr attr = { .transfer = 0 };
5492 
5493 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5494 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5495 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5496 	}
5497 	DRV_LOG(ERR,
5498 		"port %u counter query is not supported.",
5499 		 dev->data->port_id);
5500 	return -ENOTSUP;
5501 }
5502 
5503 #define MLX5_POOL_QUERY_FREQ_US 1000000
5504 
5505 /**
5506  * Set the periodic procedure for triggering asynchronous batch queries for all
5507  * the counter pools.
5508  *
5509  * @param[in] sh
5510  *   Pointer to mlx5_ibv_shared object.
5511  */
5512 void
5513 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5514 {
5515 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5516 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5517 	uint32_t us;
5518 
5519 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5520 	pools_n += rte_atomic16_read(&cont->n_valid);
5521 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5522 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5523 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5524 		sh->cmng.query_thread_on = 0;
5525 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5526 	} else {
5527 		sh->cmng.query_thread_on = 1;
5528 	}
5529 }
5530 
5531 /**
5532  * The periodic procedure for triggering asynchronous batch queries for all the
5533  * counter pools. This function is probably called by the host thread.
5534  *
5535  * @param[in] arg
5536  *   The parameter for the alarm process.
5537  */
5538 void
5539 mlx5_flow_query_alarm(void *arg)
5540 {
5541 	struct mlx5_ibv_shared *sh = arg;
5542 	struct mlx5_devx_obj *dcs;
5543 	uint16_t offset;
5544 	int ret;
5545 	uint8_t batch = sh->cmng.batch;
5546 	uint16_t pool_index = sh->cmng.pool_index;
5547 	struct mlx5_pools_container *cont;
5548 	struct mlx5_pools_container *mcont;
5549 	struct mlx5_flow_counter_pool *pool;
5550 
5551 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5552 		goto set_alarm;
5553 next_container:
5554 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5555 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5556 	/* Check if resize was done and need to flip a container. */
5557 	if (cont != mcont) {
5558 		if (cont->pools) {
5559 			/* Clean the old container. */
5560 			rte_free(cont->pools);
5561 			memset(cont, 0, sizeof(*cont));
5562 		}
5563 		rte_cio_wmb();
5564 		 /* Flip the host container. */
5565 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5566 		cont = mcont;
5567 	}
5568 	if (!cont->pools) {
5569 		/* 2 empty containers case is unexpected. */
5570 		if (unlikely(batch != sh->cmng.batch))
5571 			goto set_alarm;
5572 		batch ^= 0x1;
5573 		pool_index = 0;
5574 		goto next_container;
5575 	}
5576 	pool = cont->pools[pool_index];
5577 	if (pool->raw_hw)
5578 		/* There is a pool query in progress. */
5579 		goto set_alarm;
5580 	pool->raw_hw =
5581 		LIST_FIRST(&sh->cmng.free_stat_raws);
5582 	if (!pool->raw_hw)
5583 		/* No free counter statistics raw memory. */
5584 		goto set_alarm;
5585 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5586 							      (&pool->a64_dcs);
5587 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5588 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5589 					       offset, NULL, NULL,
5590 					       pool->raw_hw->mem_mng->dm->id,
5591 					       (void *)(uintptr_t)
5592 					       (pool->raw_hw->data + offset),
5593 					       sh->devx_comp,
5594 					       (uint64_t)(uintptr_t)pool);
5595 	if (ret) {
5596 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5597 			" %d", pool->min_dcs->id);
5598 		pool->raw_hw = NULL;
5599 		goto set_alarm;
5600 	}
5601 	pool->raw_hw->min_dcs_id = dcs->id;
5602 	LIST_REMOVE(pool->raw_hw, next);
5603 	sh->cmng.pending_queries++;
5604 	pool_index++;
5605 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5606 		batch ^= 0x1;
5607 		pool_index = 0;
5608 	}
5609 set_alarm:
5610 	sh->cmng.batch = batch;
5611 	sh->cmng.pool_index = pool_index;
5612 	mlx5_set_query_alarm(sh);
5613 }
5614 
5615 /**
5616  * Handler for the HW respond about ready values from an asynchronous batch
5617  * query. This function is probably called by the host thread.
5618  *
5619  * @param[in] sh
5620  *   The pointer to the shared IB device context.
5621  * @param[in] async_id
5622  *   The Devx async ID.
5623  * @param[in] status
5624  *   The status of the completion.
5625  */
5626 void
5627 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5628 				  uint64_t async_id, int status)
5629 {
5630 	struct mlx5_flow_counter_pool *pool =
5631 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5632 	struct mlx5_counter_stats_raw *raw_to_free;
5633 
5634 	if (unlikely(status)) {
5635 		raw_to_free = pool->raw_hw;
5636 	} else {
5637 		raw_to_free = pool->raw;
5638 		rte_spinlock_lock(&pool->sl);
5639 		pool->raw = pool->raw_hw;
5640 		rte_spinlock_unlock(&pool->sl);
5641 		rte_atomic64_add(&pool->query_gen, 1);
5642 		/* Be sure the new raw counters data is updated in memory. */
5643 		rte_cio_wmb();
5644 	}
5645 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5646 	pool->raw_hw = NULL;
5647 	sh->cmng.pending_queries--;
5648 }
5649 
5650 /**
5651  * Translate the rte_flow group index to HW table value.
5652  *
5653  * @param[in] attributes
5654  *   Pointer to flow attributes
5655  * @param[in] external
5656  *   Value is part of flow rule created by request external to PMD.
5657  * @param[in] group
5658  *   rte_flow group index value.
5659  * @param[out] fdb_def_rule
5660  *   Whether fdb jump to table 1 is configured.
5661  * @param[out] table
5662  *   HW table value.
5663  * @param[out] error
5664  *   Pointer to error structure.
5665  *
5666  * @return
5667  *   0 on success, a negative errno value otherwise and rte_errno is set.
5668  */
5669 int
5670 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5671 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
5672 			 struct rte_flow_error *error)
5673 {
5674 	if (attributes->transfer && external && fdb_def_rule) {
5675 		if (group == UINT32_MAX)
5676 			return rte_flow_error_set
5677 						(error, EINVAL,
5678 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5679 						 NULL,
5680 						 "group index not supported");
5681 		*table = group + 1;
5682 	} else {
5683 		*table = group;
5684 	}
5685 	return 0;
5686 }
5687 
5688 /**
5689  * Discover availability of metadata reg_c's.
5690  *
5691  * Iteratively use test flows to check availability.
5692  *
5693  * @param[in] dev
5694  *   Pointer to the Ethernet device structure.
5695  *
5696  * @return
5697  *   0 on success, a negative errno value otherwise and rte_errno is set.
5698  */
5699 int
5700 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5701 {
5702 	struct mlx5_priv *priv = dev->data->dev_private;
5703 	struct mlx5_dev_config *config = &priv->config;
5704 	enum modify_reg idx;
5705 	int n = 0;
5706 
5707 	/* reg_c[0] and reg_c[1] are reserved. */
5708 	config->flow_mreg_c[n++] = REG_C_0;
5709 	config->flow_mreg_c[n++] = REG_C_1;
5710 	/* Discover availability of other reg_c's. */
5711 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5712 		struct rte_flow_attr attr = {
5713 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5714 			.priority = MLX5_FLOW_PRIO_RSVD,
5715 			.ingress = 1,
5716 		};
5717 		struct rte_flow_item items[] = {
5718 			[0] = {
5719 				.type = RTE_FLOW_ITEM_TYPE_END,
5720 			},
5721 		};
5722 		struct rte_flow_action actions[] = {
5723 			[0] = {
5724 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5725 				.conf = &(struct mlx5_flow_action_copy_mreg){
5726 					.src = REG_C_1,
5727 					.dst = idx,
5728 				},
5729 			},
5730 			[1] = {
5731 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5732 				.conf = &(struct rte_flow_action_jump){
5733 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5734 				},
5735 			},
5736 			[2] = {
5737 				.type = RTE_FLOW_ACTION_TYPE_END,
5738 			},
5739 		};
5740 		struct rte_flow *flow;
5741 		struct rte_flow_error error;
5742 
5743 		if (!config->dv_flow_en)
5744 			break;
5745 		/* Create internal flow, validation skips copy action. */
5746 		flow = flow_list_create(dev, NULL, &attr, items,
5747 					actions, false, &error);
5748 		if (!flow)
5749 			continue;
5750 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5751 			config->flow_mreg_c[n++] = idx;
5752 		flow_list_destroy(dev, NULL, flow);
5753 	}
5754 	for (; n < MLX5_MREG_C_NUM; ++n)
5755 		config->flow_mreg_c[n] = REG_NONE;
5756 	return 0;
5757 }
5758 
5759 /**
5760  * Dump flow raw hw data to file
5761  *
5762  * @param[in] dev
5763  *    The pointer to Ethernet device.
5764  * @param[in] file
5765  *   A pointer to a file for output.
5766  * @param[out] error
5767  *   Perform verbose error reporting if not NULL. PMDs initialize this
5768  *   structure in case of error only.
5769  * @return
5770  *   0 on success, a nagative value otherwise.
5771  */
5772 int
5773 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5774 		   FILE *file,
5775 		   struct rte_flow_error *error __rte_unused)
5776 {
5777 	struct mlx5_priv *priv = dev->data->dev_private;
5778 	struct mlx5_ibv_shared *sh = priv->sh;
5779 
5780 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5781 				       sh->tx_domain, file);
5782 }
5783