xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision b88341ca35fc8eda9cd4bf2e796865544788f035)
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 #include <stdbool.h>
12 
13 /* Verbs header. */
14 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
15 #ifdef PEDANTIC
16 #pragma GCC diagnostic ignored "-Wpedantic"
17 #endif
18 #include <infiniband/verbs.h>
19 #ifdef PEDANTIC
20 #pragma GCC diagnostic error "-Wpedantic"
21 #endif
22 
23 #include <rte_common.h>
24 #include <rte_ether.h>
25 #include <rte_ethdev_driver.h>
26 #include <rte_flow.h>
27 #include <rte_flow_driver.h>
28 #include <rte_malloc.h>
29 #include <rte_ip.h>
30 
31 #include <mlx5_glue.h>
32 #include <mlx5_devx_cmds.h>
33 #include <mlx5_prm.h>
34 
35 #include "mlx5_defs.h"
36 #include "mlx5.h"
37 #include "mlx5_flow.h"
38 #include "mlx5_rxtx.h"
39 
40 /* Dev ops structure defined in mlx5.c */
41 extern const struct eth_dev_ops mlx5_dev_ops;
42 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
43 
44 /** Device flow drivers. */
45 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
46 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
47 #endif
48 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
49 
50 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
51 
52 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
53 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
54 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
55 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
56 #endif
57 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
58 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
59 };
60 
61 enum mlx5_expansion {
62 	MLX5_EXPANSION_ROOT,
63 	MLX5_EXPANSION_ROOT_OUTER,
64 	MLX5_EXPANSION_ROOT_ETH_VLAN,
65 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
66 	MLX5_EXPANSION_OUTER_ETH,
67 	MLX5_EXPANSION_OUTER_ETH_VLAN,
68 	MLX5_EXPANSION_OUTER_VLAN,
69 	MLX5_EXPANSION_OUTER_IPV4,
70 	MLX5_EXPANSION_OUTER_IPV4_UDP,
71 	MLX5_EXPANSION_OUTER_IPV4_TCP,
72 	MLX5_EXPANSION_OUTER_IPV6,
73 	MLX5_EXPANSION_OUTER_IPV6_UDP,
74 	MLX5_EXPANSION_OUTER_IPV6_TCP,
75 	MLX5_EXPANSION_VXLAN,
76 	MLX5_EXPANSION_VXLAN_GPE,
77 	MLX5_EXPANSION_GRE,
78 	MLX5_EXPANSION_MPLS,
79 	MLX5_EXPANSION_ETH,
80 	MLX5_EXPANSION_ETH_VLAN,
81 	MLX5_EXPANSION_VLAN,
82 	MLX5_EXPANSION_IPV4,
83 	MLX5_EXPANSION_IPV4_UDP,
84 	MLX5_EXPANSION_IPV4_TCP,
85 	MLX5_EXPANSION_IPV6,
86 	MLX5_EXPANSION_IPV6_UDP,
87 	MLX5_EXPANSION_IPV6_TCP,
88 };
89 
90 /** Supported expansion of items. */
91 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
92 	[MLX5_EXPANSION_ROOT] = {
93 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
94 						 MLX5_EXPANSION_IPV4,
95 						 MLX5_EXPANSION_IPV6),
96 		.type = RTE_FLOW_ITEM_TYPE_END,
97 	},
98 	[MLX5_EXPANSION_ROOT_OUTER] = {
99 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
100 						 MLX5_EXPANSION_OUTER_IPV4,
101 						 MLX5_EXPANSION_OUTER_IPV6),
102 		.type = RTE_FLOW_ITEM_TYPE_END,
103 	},
104 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
105 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
106 		.type = RTE_FLOW_ITEM_TYPE_END,
107 	},
108 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
109 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
110 		.type = RTE_FLOW_ITEM_TYPE_END,
111 	},
112 	[MLX5_EXPANSION_OUTER_ETH] = {
113 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
114 						 MLX5_EXPANSION_OUTER_IPV6,
115 						 MLX5_EXPANSION_MPLS),
116 		.type = RTE_FLOW_ITEM_TYPE_ETH,
117 		.rss_types = 0,
118 	},
119 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
120 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
121 		.type = RTE_FLOW_ITEM_TYPE_ETH,
122 		.rss_types = 0,
123 	},
124 	[MLX5_EXPANSION_OUTER_VLAN] = {
125 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
126 						 MLX5_EXPANSION_OUTER_IPV6),
127 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
128 	},
129 	[MLX5_EXPANSION_OUTER_IPV4] = {
130 		.next = RTE_FLOW_EXPAND_RSS_NEXT
131 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
132 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
133 			 MLX5_EXPANSION_GRE,
134 			 MLX5_EXPANSION_IPV4,
135 			 MLX5_EXPANSION_IPV6),
136 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
137 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
138 			ETH_RSS_NONFRAG_IPV4_OTHER,
139 	},
140 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
141 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
142 						 MLX5_EXPANSION_VXLAN_GPE),
143 		.type = RTE_FLOW_ITEM_TYPE_UDP,
144 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
145 	},
146 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
147 		.type = RTE_FLOW_ITEM_TYPE_TCP,
148 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
149 	},
150 	[MLX5_EXPANSION_OUTER_IPV6] = {
151 		.next = RTE_FLOW_EXPAND_RSS_NEXT
152 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
153 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
154 			 MLX5_EXPANSION_IPV4,
155 			 MLX5_EXPANSION_IPV6),
156 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
157 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
158 			ETH_RSS_NONFRAG_IPV6_OTHER,
159 	},
160 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
161 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
162 						 MLX5_EXPANSION_VXLAN_GPE),
163 		.type = RTE_FLOW_ITEM_TYPE_UDP,
164 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
165 	},
166 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
167 		.type = RTE_FLOW_ITEM_TYPE_TCP,
168 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
169 	},
170 	[MLX5_EXPANSION_VXLAN] = {
171 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
172 						 MLX5_EXPANSION_IPV4,
173 						 MLX5_EXPANSION_IPV6),
174 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
175 	},
176 	[MLX5_EXPANSION_VXLAN_GPE] = {
177 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
178 						 MLX5_EXPANSION_IPV4,
179 						 MLX5_EXPANSION_IPV6),
180 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
181 	},
182 	[MLX5_EXPANSION_GRE] = {
183 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
184 		.type = RTE_FLOW_ITEM_TYPE_GRE,
185 	},
186 	[MLX5_EXPANSION_MPLS] = {
187 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 						 MLX5_EXPANSION_IPV6),
189 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
190 	},
191 	[MLX5_EXPANSION_ETH] = {
192 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
193 						 MLX5_EXPANSION_IPV6),
194 		.type = RTE_FLOW_ITEM_TYPE_ETH,
195 	},
196 	[MLX5_EXPANSION_ETH_VLAN] = {
197 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
198 		.type = RTE_FLOW_ITEM_TYPE_ETH,
199 	},
200 	[MLX5_EXPANSION_VLAN] = {
201 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
202 						 MLX5_EXPANSION_IPV6),
203 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
204 	},
205 	[MLX5_EXPANSION_IPV4] = {
206 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
207 						 MLX5_EXPANSION_IPV4_TCP),
208 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
209 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
210 			ETH_RSS_NONFRAG_IPV4_OTHER,
211 	},
212 	[MLX5_EXPANSION_IPV4_UDP] = {
213 		.type = RTE_FLOW_ITEM_TYPE_UDP,
214 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
215 	},
216 	[MLX5_EXPANSION_IPV4_TCP] = {
217 		.type = RTE_FLOW_ITEM_TYPE_TCP,
218 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
219 	},
220 	[MLX5_EXPANSION_IPV6] = {
221 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
222 						 MLX5_EXPANSION_IPV6_TCP),
223 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
224 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
225 			ETH_RSS_NONFRAG_IPV6_OTHER,
226 	},
227 	[MLX5_EXPANSION_IPV6_UDP] = {
228 		.type = RTE_FLOW_ITEM_TYPE_UDP,
229 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
230 	},
231 	[MLX5_EXPANSION_IPV6_TCP] = {
232 		.type = RTE_FLOW_ITEM_TYPE_TCP,
233 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
234 	},
235 };
236 
237 static const struct rte_flow_ops mlx5_flow_ops = {
238 	.validate = mlx5_flow_validate,
239 	.create = mlx5_flow_create,
240 	.destroy = mlx5_flow_destroy,
241 	.flush = mlx5_flow_flush,
242 	.isolate = mlx5_flow_isolate,
243 	.query = mlx5_flow_query,
244 	.dev_dump = mlx5_flow_dev_dump,
245 };
246 
247 /* Convert FDIR request to Generic flow. */
248 struct mlx5_fdir {
249 	struct rte_flow_attr attr;
250 	struct rte_flow_item items[4];
251 	struct rte_flow_item_eth l2;
252 	struct rte_flow_item_eth l2_mask;
253 	union {
254 		struct rte_flow_item_ipv4 ipv4;
255 		struct rte_flow_item_ipv6 ipv6;
256 	} l3;
257 	union {
258 		struct rte_flow_item_ipv4 ipv4;
259 		struct rte_flow_item_ipv6 ipv6;
260 	} l3_mask;
261 	union {
262 		struct rte_flow_item_udp udp;
263 		struct rte_flow_item_tcp tcp;
264 	} l4;
265 	union {
266 		struct rte_flow_item_udp udp;
267 		struct rte_flow_item_tcp tcp;
268 	} l4_mask;
269 	struct rte_flow_action actions[2];
270 	struct rte_flow_action_queue queue;
271 };
272 
273 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
274 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
275 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
276 };
277 
278 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
279 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
280 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
281 	{ 9, 10, 11 }, { 12, 13, 14 },
282 };
283 
284 /* Tunnel information. */
285 struct mlx5_flow_tunnel_info {
286 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
287 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
288 };
289 
290 static struct mlx5_flow_tunnel_info tunnels_info[] = {
291 	{
292 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
293 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
294 	},
295 	{
296 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
297 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
298 	},
299 	{
300 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
301 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
302 	},
303 	{
304 		.tunnel = MLX5_FLOW_LAYER_GRE,
305 		.ptype = RTE_PTYPE_TUNNEL_GRE,
306 	},
307 	{
308 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
309 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
310 	},
311 	{
312 		.tunnel = MLX5_FLOW_LAYER_MPLS,
313 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
314 	},
315 	{
316 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
317 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
318 	},
319 	{
320 		.tunnel = MLX5_FLOW_LAYER_IPIP,
321 		.ptype = RTE_PTYPE_TUNNEL_IP,
322 	},
323 	{
324 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
325 		.ptype = RTE_PTYPE_TUNNEL_IP,
326 	},
327 	{
328 		.tunnel = MLX5_FLOW_LAYER_GTP,
329 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
330 	},
331 };
332 
333 /**
334  * Translate tag ID to register.
335  *
336  * @param[in] dev
337  *   Pointer to the Ethernet device structure.
338  * @param[in] feature
339  *   The feature that request the register.
340  * @param[in] id
341  *   The request register ID.
342  * @param[out] error
343  *   Error description in case of any.
344  *
345  * @return
346  *   The request register on success, a negative errno
347  *   value otherwise and rte_errno is set.
348  */
349 int
350 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
351 		     enum mlx5_feature_name feature,
352 		     uint32_t id,
353 		     struct rte_flow_error *error)
354 {
355 	struct mlx5_priv *priv = dev->data->dev_private;
356 	struct mlx5_dev_config *config = &priv->config;
357 	enum modify_reg start_reg;
358 	bool skip_mtr_reg = false;
359 
360 	switch (feature) {
361 	case MLX5_HAIRPIN_RX:
362 		return REG_B;
363 	case MLX5_HAIRPIN_TX:
364 		return REG_A;
365 	case MLX5_METADATA_RX:
366 		switch (config->dv_xmeta_en) {
367 		case MLX5_XMETA_MODE_LEGACY:
368 			return REG_B;
369 		case MLX5_XMETA_MODE_META16:
370 			return REG_C_0;
371 		case MLX5_XMETA_MODE_META32:
372 			return REG_C_1;
373 		}
374 		break;
375 	case MLX5_METADATA_TX:
376 		return REG_A;
377 	case MLX5_METADATA_FDB:
378 		switch (config->dv_xmeta_en) {
379 		case MLX5_XMETA_MODE_LEGACY:
380 			return REG_NONE;
381 		case MLX5_XMETA_MODE_META16:
382 			return REG_C_0;
383 		case MLX5_XMETA_MODE_META32:
384 			return REG_C_1;
385 		}
386 		break;
387 	case MLX5_FLOW_MARK:
388 		switch (config->dv_xmeta_en) {
389 		case MLX5_XMETA_MODE_LEGACY:
390 			return REG_NONE;
391 		case MLX5_XMETA_MODE_META16:
392 			return REG_C_1;
393 		case MLX5_XMETA_MODE_META32:
394 			return REG_C_0;
395 		}
396 		break;
397 	case MLX5_MTR_SFX:
398 		/*
399 		 * If meter color and flow match share one register, flow match
400 		 * should use the meter color register for match.
401 		 */
402 		if (priv->mtr_reg_share)
403 			return priv->mtr_color_reg;
404 		else
405 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
406 			       REG_C_3;
407 	case MLX5_MTR_COLOR:
408 		MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
409 		return priv->mtr_color_reg;
410 	case MLX5_COPY_MARK:
411 		/*
412 		 * Metadata COPY_MARK register using is in meter suffix sub
413 		 * flow while with meter. It's safe to share the same register.
414 		 */
415 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
416 	case MLX5_APP_TAG:
417 		/*
418 		 * If meter is enable, it will engage the register for color
419 		 * match and flow match. If meter color match is not using the
420 		 * REG_C_2, need to skip the REG_C_x be used by meter color
421 		 * match.
422 		 * If meter is disable, free to use all available registers.
423 		 */
424 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
425 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
426 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
427 		if (id > (REG_C_7 - start_reg))
428 			return rte_flow_error_set(error, EINVAL,
429 						  RTE_FLOW_ERROR_TYPE_ITEM,
430 						  NULL, "invalid tag id");
431 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
432 			return rte_flow_error_set(error, ENOTSUP,
433 						  RTE_FLOW_ERROR_TYPE_ITEM,
434 						  NULL, "unsupported tag id");
435 		/*
436 		 * This case means meter is using the REG_C_x great than 2.
437 		 * Take care not to conflict with meter color REG_C_x.
438 		 * If the available index REG_C_y >= REG_C_x, skip the
439 		 * color register.
440 		 */
441 		if (skip_mtr_reg && config->flow_mreg_c
442 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
443 			if (config->flow_mreg_c
444 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
445 				return config->flow_mreg_c
446 					       [id + 1 + start_reg - REG_C_0];
447 			return rte_flow_error_set(error, ENOTSUP,
448 						  RTE_FLOW_ERROR_TYPE_ITEM,
449 						  NULL, "unsupported tag id");
450 		}
451 		return config->flow_mreg_c[id + start_reg - REG_C_0];
452 	}
453 	MLX5_ASSERT(false);
454 	return rte_flow_error_set(error, EINVAL,
455 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
456 				  NULL, "invalid feature name");
457 }
458 
459 /**
460  * Check extensive flow metadata register support.
461  *
462  * @param dev
463  *   Pointer to rte_eth_dev structure.
464  *
465  * @return
466  *   True if device supports extensive flow metadata register, otherwise false.
467  */
468 bool
469 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
470 {
471 	struct mlx5_priv *priv = dev->data->dev_private;
472 	struct mlx5_dev_config *config = &priv->config;
473 
474 	/*
475 	 * Having available reg_c can be regarded inclusively as supporting
476 	 * extensive flow metadata register, which could mean,
477 	 * - metadata register copy action by modify header.
478 	 * - 16 modify header actions is supported.
479 	 * - reg_c's are preserved across different domain (FDB and NIC) on
480 	 *   packet loopback by flow lookup miss.
481 	 */
482 	return config->flow_mreg_c[2] != REG_NONE;
483 }
484 
485 /**
486  * Discover the maximum number of priority available.
487  *
488  * @param[in] dev
489  *   Pointer to the Ethernet device structure.
490  *
491  * @return
492  *   number of supported flow priority on success, a negative errno
493  *   value otherwise and rte_errno is set.
494  */
495 int
496 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
497 {
498 	struct mlx5_priv *priv = dev->data->dev_private;
499 	struct {
500 		struct ibv_flow_attr attr;
501 		struct ibv_flow_spec_eth eth;
502 		struct ibv_flow_spec_action_drop drop;
503 	} flow_attr = {
504 		.attr = {
505 			.num_of_specs = 2,
506 			.port = (uint8_t)priv->ibv_port,
507 		},
508 		.eth = {
509 			.type = IBV_FLOW_SPEC_ETH,
510 			.size = sizeof(struct ibv_flow_spec_eth),
511 		},
512 		.drop = {
513 			.size = sizeof(struct ibv_flow_spec_action_drop),
514 			.type = IBV_FLOW_SPEC_ACTION_DROP,
515 		},
516 	};
517 	struct ibv_flow *flow;
518 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
519 	uint16_t vprio[] = { 8, 16 };
520 	int i;
521 	int priority = 0;
522 
523 	if (!drop) {
524 		rte_errno = ENOTSUP;
525 		return -rte_errno;
526 	}
527 	for (i = 0; i != RTE_DIM(vprio); i++) {
528 		flow_attr.attr.priority = vprio[i] - 1;
529 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
530 		if (!flow)
531 			break;
532 		claim_zero(mlx5_glue->destroy_flow(flow));
533 		priority = vprio[i];
534 	}
535 	mlx5_hrxq_drop_release(dev);
536 	switch (priority) {
537 	case 8:
538 		priority = RTE_DIM(priority_map_3);
539 		break;
540 	case 16:
541 		priority = RTE_DIM(priority_map_5);
542 		break;
543 	default:
544 		rte_errno = ENOTSUP;
545 		DRV_LOG(ERR,
546 			"port %u verbs maximum priority: %d expected 8/16",
547 			dev->data->port_id, priority);
548 		return -rte_errno;
549 	}
550 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
551 		dev->data->port_id, priority);
552 	return priority;
553 }
554 
555 /**
556  * Adjust flow priority based on the highest layer and the request priority.
557  *
558  * @param[in] dev
559  *   Pointer to the Ethernet device structure.
560  * @param[in] priority
561  *   The rule base priority.
562  * @param[in] subpriority
563  *   The priority based on the items.
564  *
565  * @return
566  *   The new priority.
567  */
568 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
569 				   uint32_t subpriority)
570 {
571 	uint32_t res = 0;
572 	struct mlx5_priv *priv = dev->data->dev_private;
573 
574 	switch (priv->config.flow_prio) {
575 	case RTE_DIM(priority_map_3):
576 		res = priority_map_3[priority][subpriority];
577 		break;
578 	case RTE_DIM(priority_map_5):
579 		res = priority_map_5[priority][subpriority];
580 		break;
581 	}
582 	return  res;
583 }
584 
585 /**
586  * Verify the @p item specifications (spec, last, mask) are compatible with the
587  * NIC capabilities.
588  *
589  * @param[in] item
590  *   Item specification.
591  * @param[in] mask
592  *   @p item->mask or flow default bit-masks.
593  * @param[in] nic_mask
594  *   Bit-masks covering supported fields by the NIC to compare with user mask.
595  * @param[in] size
596  *   Bit-masks size in bytes.
597  * @param[out] error
598  *   Pointer to error structure.
599  *
600  * @return
601  *   0 on success, a negative errno value otherwise and rte_errno is set.
602  */
603 int
604 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
605 			  const uint8_t *mask,
606 			  const uint8_t *nic_mask,
607 			  unsigned int size,
608 			  struct rte_flow_error *error)
609 {
610 	unsigned int i;
611 
612 	MLX5_ASSERT(nic_mask);
613 	for (i = 0; i < size; ++i)
614 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
615 			return rte_flow_error_set(error, ENOTSUP,
616 						  RTE_FLOW_ERROR_TYPE_ITEM,
617 						  item,
618 						  "mask enables non supported"
619 						  " bits");
620 	if (!item->spec && (item->mask || item->last))
621 		return rte_flow_error_set(error, EINVAL,
622 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
623 					  "mask/last without a spec is not"
624 					  " supported");
625 	if (item->spec && item->last) {
626 		uint8_t spec[size];
627 		uint8_t last[size];
628 		unsigned int i;
629 		int ret;
630 
631 		for (i = 0; i < size; ++i) {
632 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
633 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
634 		}
635 		ret = memcmp(spec, last, size);
636 		if (ret != 0)
637 			return rte_flow_error_set(error, EINVAL,
638 						  RTE_FLOW_ERROR_TYPE_ITEM,
639 						  item,
640 						  "range is not valid");
641 	}
642 	return 0;
643 }
644 
645 /**
646  * Adjust the hash fields according to the @p flow information.
647  *
648  * @param[in] dev_flow.
649  *   Pointer to the mlx5_flow.
650  * @param[in] tunnel
651  *   1 when the hash field is for a tunnel item.
652  * @param[in] layer_types
653  *   ETH_RSS_* types.
654  * @param[in] hash_fields
655  *   Item hash fields.
656  *
657  * @return
658  *   The hash fields that should be used.
659  */
660 uint64_t
661 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
662 			    int tunnel __rte_unused, uint64_t layer_types,
663 			    uint64_t hash_fields)
664 {
665 	struct rte_flow *flow = dev_flow->flow;
666 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
667 	int rss_request_inner = flow->rss.level >= 2;
668 
669 	/* Check RSS hash level for tunnel. */
670 	if (tunnel && rss_request_inner)
671 		hash_fields |= IBV_RX_HASH_INNER;
672 	else if (tunnel || rss_request_inner)
673 		return 0;
674 #endif
675 	/* Check if requested layer matches RSS hash fields. */
676 	if (!(flow->rss.types & layer_types))
677 		return 0;
678 	return hash_fields;
679 }
680 
681 /**
682  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
683  * if several tunnel rules are used on this queue, the tunnel ptype will be
684  * cleared.
685  *
686  * @param rxq_ctrl
687  *   Rx queue to update.
688  */
689 static void
690 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
691 {
692 	unsigned int i;
693 	uint32_t tunnel_ptype = 0;
694 
695 	/* Look up for the ptype to use. */
696 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
697 		if (!rxq_ctrl->flow_tunnels_n[i])
698 			continue;
699 		if (!tunnel_ptype) {
700 			tunnel_ptype = tunnels_info[i].ptype;
701 		} else {
702 			tunnel_ptype = 0;
703 			break;
704 		}
705 	}
706 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
707 }
708 
709 /**
710  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
711  * flow.
712  *
713  * @param[in] dev
714  *   Pointer to the Ethernet device structure.
715  * @param[in] flow
716  *   Pointer to flow structure.
717  * @param[in] dev_handle
718  *   Pointer to device flow handle structure.
719  */
720 static void
721 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow,
722 		       struct mlx5_flow_handle *dev_handle)
723 {
724 	struct mlx5_priv *priv = dev->data->dev_private;
725 	const int mark = !!(dev_handle->act_flags &
726 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
727 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
728 	unsigned int i;
729 
730 	for (i = 0; i != flow->rss.queue_num; ++i) {
731 		int idx = (*flow->rss.queue)[i];
732 		struct mlx5_rxq_ctrl *rxq_ctrl =
733 			container_of((*priv->rxqs)[idx],
734 				     struct mlx5_rxq_ctrl, rxq);
735 
736 		/*
737 		 * To support metadata register copy on Tx loopback,
738 		 * this must be always enabled (metadata may arive
739 		 * from other port - not from local flows only.
740 		 */
741 		if (priv->config.dv_flow_en &&
742 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
743 		    mlx5_flow_ext_mreg_supported(dev)) {
744 			rxq_ctrl->rxq.mark = 1;
745 			rxq_ctrl->flow_mark_n = 1;
746 		} else if (mark) {
747 			rxq_ctrl->rxq.mark = 1;
748 			rxq_ctrl->flow_mark_n++;
749 		}
750 		if (tunnel) {
751 			unsigned int j;
752 
753 			/* Increase the counter matching the flow. */
754 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
755 				if ((tunnels_info[j].tunnel &
756 				     dev_handle->layers) ==
757 				    tunnels_info[j].tunnel) {
758 					rxq_ctrl->flow_tunnels_n[j]++;
759 					break;
760 				}
761 			}
762 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
763 		}
764 	}
765 }
766 
767 /**
768  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
769  *
770  * @param[in] dev
771  *   Pointer to the Ethernet device structure.
772  * @param[in] flow
773  *   Pointer to flow structure.
774  */
775 static void
776 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
777 {
778 	struct mlx5_priv *priv = dev->data->dev_private;
779 	uint32_t handle_idx;
780 	struct mlx5_flow_handle *dev_handle;
781 
782 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
783 		       handle_idx, dev_handle, next)
784 		flow_drv_rxq_flags_set(dev, flow, dev_handle);
785 }
786 
787 /**
788  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
789  * device flow if no other flow uses it with the same kind of request.
790  *
791  * @param dev
792  *   Pointer to Ethernet device.
793  * @param[in] flow
794  *   Pointer to flow structure.
795  * @param[in] dev_handle
796  *   Pointer to the device flow handle structure.
797  */
798 static void
799 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow,
800 			struct mlx5_flow_handle *dev_handle)
801 {
802 	struct mlx5_priv *priv = dev->data->dev_private;
803 	const int mark = !!(dev_handle->act_flags &
804 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
805 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
806 	unsigned int i;
807 
808 	MLX5_ASSERT(dev->data->dev_started);
809 	for (i = 0; i != flow->rss.queue_num; ++i) {
810 		int idx = (*flow->rss.queue)[i];
811 		struct mlx5_rxq_ctrl *rxq_ctrl =
812 			container_of((*priv->rxqs)[idx],
813 				     struct mlx5_rxq_ctrl, rxq);
814 
815 		if (priv->config.dv_flow_en &&
816 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
817 		    mlx5_flow_ext_mreg_supported(dev)) {
818 			rxq_ctrl->rxq.mark = 1;
819 			rxq_ctrl->flow_mark_n = 1;
820 		} else if (mark) {
821 			rxq_ctrl->flow_mark_n--;
822 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
823 		}
824 		if (tunnel) {
825 			unsigned int j;
826 
827 			/* Decrease the counter matching the flow. */
828 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
829 				if ((tunnels_info[j].tunnel &
830 				     dev_handle->layers) ==
831 				    tunnels_info[j].tunnel) {
832 					rxq_ctrl->flow_tunnels_n[j]--;
833 					break;
834 				}
835 			}
836 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
837 		}
838 	}
839 }
840 
841 /**
842  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
843  * @p flow if no other flow uses it with the same kind of request.
844  *
845  * @param dev
846  *   Pointer to Ethernet device.
847  * @param[in] flow
848  *   Pointer to the flow.
849  */
850 static void
851 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
852 {
853 	struct mlx5_priv *priv = dev->data->dev_private;
854 	uint32_t handle_idx;
855 	struct mlx5_flow_handle *dev_handle;
856 
857 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
858 		       handle_idx, dev_handle, next)
859 		flow_drv_rxq_flags_trim(dev, flow, dev_handle);
860 }
861 
862 /**
863  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
864  *
865  * @param dev
866  *   Pointer to Ethernet device.
867  */
868 static void
869 flow_rxq_flags_clear(struct rte_eth_dev *dev)
870 {
871 	struct mlx5_priv *priv = dev->data->dev_private;
872 	unsigned int i;
873 
874 	for (i = 0; i != priv->rxqs_n; ++i) {
875 		struct mlx5_rxq_ctrl *rxq_ctrl;
876 		unsigned int j;
877 
878 		if (!(*priv->rxqs)[i])
879 			continue;
880 		rxq_ctrl = container_of((*priv->rxqs)[i],
881 					struct mlx5_rxq_ctrl, rxq);
882 		rxq_ctrl->flow_mark_n = 0;
883 		rxq_ctrl->rxq.mark = 0;
884 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
885 			rxq_ctrl->flow_tunnels_n[j] = 0;
886 		rxq_ctrl->rxq.tunnel = 0;
887 	}
888 }
889 
890 /*
891  * return a pointer to the desired action in the list of actions.
892  *
893  * @param[in] actions
894  *   The list of actions to search the action in.
895  * @param[in] action
896  *   The action to find.
897  *
898  * @return
899  *   Pointer to the action in the list, if found. NULL otherwise.
900  */
901 const struct rte_flow_action *
902 mlx5_flow_find_action(const struct rte_flow_action *actions,
903 		      enum rte_flow_action_type action)
904 {
905 	if (actions == NULL)
906 		return NULL;
907 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
908 		if (actions->type == action)
909 			return actions;
910 	return NULL;
911 }
912 
913 /*
914  * Validate the flag action.
915  *
916  * @param[in] action_flags
917  *   Bit-fields that holds the actions detected until now.
918  * @param[in] attr
919  *   Attributes of flow that includes this action.
920  * @param[out] error
921  *   Pointer to error structure.
922  *
923  * @return
924  *   0 on success, a negative errno value otherwise and rte_errno is set.
925  */
926 int
927 mlx5_flow_validate_action_flag(uint64_t action_flags,
928 			       const struct rte_flow_attr *attr,
929 			       struct rte_flow_error *error)
930 {
931 	if (action_flags & MLX5_FLOW_ACTION_MARK)
932 		return rte_flow_error_set(error, EINVAL,
933 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
934 					  "can't mark and flag in same flow");
935 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
936 		return rte_flow_error_set(error, EINVAL,
937 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
938 					  "can't have 2 flag"
939 					  " actions in same flow");
940 	if (attr->egress)
941 		return rte_flow_error_set(error, ENOTSUP,
942 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
943 					  "flag action not supported for "
944 					  "egress");
945 	return 0;
946 }
947 
948 /*
949  * Validate the mark action.
950  *
951  * @param[in] action
952  *   Pointer to the queue action.
953  * @param[in] action_flags
954  *   Bit-fields that holds the actions detected until now.
955  * @param[in] attr
956  *   Attributes of flow that includes this action.
957  * @param[out] error
958  *   Pointer to error structure.
959  *
960  * @return
961  *   0 on success, a negative errno value otherwise and rte_errno is set.
962  */
963 int
964 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
965 			       uint64_t action_flags,
966 			       const struct rte_flow_attr *attr,
967 			       struct rte_flow_error *error)
968 {
969 	const struct rte_flow_action_mark *mark = action->conf;
970 
971 	if (!mark)
972 		return rte_flow_error_set(error, EINVAL,
973 					  RTE_FLOW_ERROR_TYPE_ACTION,
974 					  action,
975 					  "configuration cannot be null");
976 	if (mark->id >= MLX5_FLOW_MARK_MAX)
977 		return rte_flow_error_set(error, EINVAL,
978 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
979 					  &mark->id,
980 					  "mark id must in 0 <= id < "
981 					  RTE_STR(MLX5_FLOW_MARK_MAX));
982 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
983 		return rte_flow_error_set(error, EINVAL,
984 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
985 					  "can't flag and mark in same flow");
986 	if (action_flags & MLX5_FLOW_ACTION_MARK)
987 		return rte_flow_error_set(error, EINVAL,
988 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
989 					  "can't have 2 mark actions in same"
990 					  " flow");
991 	if (attr->egress)
992 		return rte_flow_error_set(error, ENOTSUP,
993 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
994 					  "mark action not supported for "
995 					  "egress");
996 	return 0;
997 }
998 
999 /*
1000  * Validate the drop action.
1001  *
1002  * @param[in] action_flags
1003  *   Bit-fields that holds the actions detected until now.
1004  * @param[in] attr
1005  *   Attributes of flow that includes this action.
1006  * @param[out] error
1007  *   Pointer to error structure.
1008  *
1009  * @return
1010  *   0 on success, a negative errno value otherwise and rte_errno is set.
1011  */
1012 int
1013 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1014 			       const struct rte_flow_attr *attr,
1015 			       struct rte_flow_error *error)
1016 {
1017 	if (attr->egress)
1018 		return rte_flow_error_set(error, ENOTSUP,
1019 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1020 					  "drop action not supported for "
1021 					  "egress");
1022 	return 0;
1023 }
1024 
1025 /*
1026  * Validate the queue action.
1027  *
1028  * @param[in] action
1029  *   Pointer to the queue action.
1030  * @param[in] action_flags
1031  *   Bit-fields that holds the actions detected until now.
1032  * @param[in] dev
1033  *   Pointer to the Ethernet device structure.
1034  * @param[in] attr
1035  *   Attributes of flow that includes this action.
1036  * @param[out] error
1037  *   Pointer to error structure.
1038  *
1039  * @return
1040  *   0 on success, a negative errno value otherwise and rte_errno is set.
1041  */
1042 int
1043 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1044 				uint64_t action_flags,
1045 				struct rte_eth_dev *dev,
1046 				const struct rte_flow_attr *attr,
1047 				struct rte_flow_error *error)
1048 {
1049 	struct mlx5_priv *priv = dev->data->dev_private;
1050 	const struct rte_flow_action_queue *queue = action->conf;
1051 
1052 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1053 		return rte_flow_error_set(error, EINVAL,
1054 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1055 					  "can't have 2 fate actions in"
1056 					  " same flow");
1057 	if (!priv->rxqs_n)
1058 		return rte_flow_error_set(error, EINVAL,
1059 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1060 					  NULL, "No Rx queues configured");
1061 	if (queue->index >= priv->rxqs_n)
1062 		return rte_flow_error_set(error, EINVAL,
1063 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1064 					  &queue->index,
1065 					  "queue index out of range");
1066 	if (!(*priv->rxqs)[queue->index])
1067 		return rte_flow_error_set(error, EINVAL,
1068 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1069 					  &queue->index,
1070 					  "queue is not configured");
1071 	if (attr->egress)
1072 		return rte_flow_error_set(error, ENOTSUP,
1073 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1074 					  "queue action not supported for "
1075 					  "egress");
1076 	return 0;
1077 }
1078 
1079 /*
1080  * Validate the rss action.
1081  *
1082  * @param[in] action
1083  *   Pointer to the queue action.
1084  * @param[in] action_flags
1085  *   Bit-fields that holds the actions detected until now.
1086  * @param[in] dev
1087  *   Pointer to the Ethernet device structure.
1088  * @param[in] attr
1089  *   Attributes of flow that includes this action.
1090  * @param[in] item_flags
1091  *   Items that were detected.
1092  * @param[out] error
1093  *   Pointer to error structure.
1094  *
1095  * @return
1096  *   0 on success, a negative errno value otherwise and rte_errno is set.
1097  */
1098 int
1099 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1100 			      uint64_t action_flags,
1101 			      struct rte_eth_dev *dev,
1102 			      const struct rte_flow_attr *attr,
1103 			      uint64_t item_flags,
1104 			      struct rte_flow_error *error)
1105 {
1106 	struct mlx5_priv *priv = dev->data->dev_private;
1107 	const struct rte_flow_action_rss *rss = action->conf;
1108 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1109 	unsigned int i;
1110 
1111 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1112 		return rte_flow_error_set(error, EINVAL,
1113 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1114 					  "can't have 2 fate actions"
1115 					  " in same flow");
1116 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1117 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1118 		return rte_flow_error_set(error, ENOTSUP,
1119 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1120 					  &rss->func,
1121 					  "RSS hash function not supported");
1122 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1123 	if (rss->level > 2)
1124 #else
1125 	if (rss->level > 1)
1126 #endif
1127 		return rte_flow_error_set(error, ENOTSUP,
1128 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1129 					  &rss->level,
1130 					  "tunnel RSS is not supported");
1131 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1132 	if (rss->key_len == 0 && rss->key != NULL)
1133 		return rte_flow_error_set(error, ENOTSUP,
1134 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1135 					  &rss->key_len,
1136 					  "RSS hash key length 0");
1137 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1138 		return rte_flow_error_set(error, ENOTSUP,
1139 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1140 					  &rss->key_len,
1141 					  "RSS hash key too small");
1142 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1143 		return rte_flow_error_set(error, ENOTSUP,
1144 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1145 					  &rss->key_len,
1146 					  "RSS hash key too large");
1147 	if (rss->queue_num > priv->config.ind_table_max_size)
1148 		return rte_flow_error_set(error, ENOTSUP,
1149 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1150 					  &rss->queue_num,
1151 					  "number of queues too large");
1152 	if (rss->types & MLX5_RSS_HF_MASK)
1153 		return rte_flow_error_set(error, ENOTSUP,
1154 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1155 					  &rss->types,
1156 					  "some RSS protocols are not"
1157 					  " supported");
1158 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1159 	    !(rss->types & ETH_RSS_IP))
1160 		return rte_flow_error_set(error, EINVAL,
1161 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1162 					  "L3 partial RSS requested but L3 RSS"
1163 					  " type not specified");
1164 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1165 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1166 		return rte_flow_error_set(error, EINVAL,
1167 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1168 					  "L4 partial RSS requested but L4 RSS"
1169 					  " type not specified");
1170 	if (!priv->rxqs_n)
1171 		return rte_flow_error_set(error, EINVAL,
1172 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1173 					  NULL, "No Rx queues configured");
1174 	if (!rss->queue_num)
1175 		return rte_flow_error_set(error, EINVAL,
1176 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1177 					  NULL, "No queues configured");
1178 	for (i = 0; i != rss->queue_num; ++i) {
1179 		if (rss->queue[i] >= priv->rxqs_n)
1180 			return rte_flow_error_set
1181 				(error, EINVAL,
1182 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1183 				 &rss->queue[i], "queue index out of range");
1184 		if (!(*priv->rxqs)[rss->queue[i]])
1185 			return rte_flow_error_set
1186 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1187 				 &rss->queue[i], "queue is not configured");
1188 	}
1189 	if (attr->egress)
1190 		return rte_flow_error_set(error, ENOTSUP,
1191 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1192 					  "rss action not supported for "
1193 					  "egress");
1194 	if (rss->level > 1 &&  !tunnel)
1195 		return rte_flow_error_set(error, EINVAL,
1196 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1197 					  "inner RSS is not supported for "
1198 					  "non-tunnel flows");
1199 	return 0;
1200 }
1201 
1202 /*
1203  * Validate the count action.
1204  *
1205  * @param[in] dev
1206  *   Pointer to the Ethernet device structure.
1207  * @param[in] attr
1208  *   Attributes of flow that includes this action.
1209  * @param[out] error
1210  *   Pointer to error structure.
1211  *
1212  * @return
1213  *   0 on success, a negative errno value otherwise and rte_errno is set.
1214  */
1215 int
1216 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1217 				const struct rte_flow_attr *attr,
1218 				struct rte_flow_error *error)
1219 {
1220 	if (attr->egress)
1221 		return rte_flow_error_set(error, ENOTSUP,
1222 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1223 					  "count action not supported for "
1224 					  "egress");
1225 	return 0;
1226 }
1227 
1228 /**
1229  * Verify the @p attributes will be correctly understood by the NIC and store
1230  * them in the @p flow if everything is correct.
1231  *
1232  * @param[in] dev
1233  *   Pointer to the Ethernet device structure.
1234  * @param[in] attributes
1235  *   Pointer to flow attributes
1236  * @param[out] error
1237  *   Pointer to error structure.
1238  *
1239  * @return
1240  *   0 on success, a negative errno value otherwise and rte_errno is set.
1241  */
1242 int
1243 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1244 			      const struct rte_flow_attr *attributes,
1245 			      struct rte_flow_error *error)
1246 {
1247 	struct mlx5_priv *priv = dev->data->dev_private;
1248 	uint32_t priority_max = priv->config.flow_prio - 1;
1249 
1250 	if (attributes->group)
1251 		return rte_flow_error_set(error, ENOTSUP,
1252 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1253 					  NULL, "groups is not supported");
1254 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1255 	    attributes->priority >= priority_max)
1256 		return rte_flow_error_set(error, ENOTSUP,
1257 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1258 					  NULL, "priority out of range");
1259 	if (attributes->egress)
1260 		return rte_flow_error_set(error, ENOTSUP,
1261 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1262 					  "egress is not supported");
1263 	if (attributes->transfer && !priv->config.dv_esw_en)
1264 		return rte_flow_error_set(error, ENOTSUP,
1265 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1266 					  NULL, "transfer is not supported");
1267 	if (!attributes->ingress)
1268 		return rte_flow_error_set(error, EINVAL,
1269 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1270 					  NULL,
1271 					  "ingress attribute is mandatory");
1272 	return 0;
1273 }
1274 
1275 /**
1276  * Validate ICMP6 item.
1277  *
1278  * @param[in] item
1279  *   Item specification.
1280  * @param[in] item_flags
1281  *   Bit-fields that holds the items detected until now.
1282  * @param[out] error
1283  *   Pointer to error structure.
1284  *
1285  * @return
1286  *   0 on success, a negative errno value otherwise and rte_errno is set.
1287  */
1288 int
1289 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1290 			       uint64_t item_flags,
1291 			       uint8_t target_protocol,
1292 			       struct rte_flow_error *error)
1293 {
1294 	const struct rte_flow_item_icmp6 *mask = item->mask;
1295 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1296 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1297 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1298 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1299 				      MLX5_FLOW_LAYER_OUTER_L4;
1300 	int ret;
1301 
1302 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1303 		return rte_flow_error_set(error, EINVAL,
1304 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1305 					  "protocol filtering not compatible"
1306 					  " with ICMP6 layer");
1307 	if (!(item_flags & l3m))
1308 		return rte_flow_error_set(error, EINVAL,
1309 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1310 					  "IPv6 is mandatory to filter on"
1311 					  " ICMP6");
1312 	if (item_flags & l4m)
1313 		return rte_flow_error_set(error, EINVAL,
1314 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1315 					  "multiple L4 layers not supported");
1316 	if (!mask)
1317 		mask = &rte_flow_item_icmp6_mask;
1318 	ret = mlx5_flow_item_acceptable
1319 		(item, (const uint8_t *)mask,
1320 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1321 		 sizeof(struct rte_flow_item_icmp6), error);
1322 	if (ret < 0)
1323 		return ret;
1324 	return 0;
1325 }
1326 
1327 /**
1328  * Validate ICMP item.
1329  *
1330  * @param[in] item
1331  *   Item specification.
1332  * @param[in] item_flags
1333  *   Bit-fields that holds the items detected until now.
1334  * @param[out] error
1335  *   Pointer to error structure.
1336  *
1337  * @return
1338  *   0 on success, a negative errno value otherwise and rte_errno is set.
1339  */
1340 int
1341 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1342 			     uint64_t item_flags,
1343 			     uint8_t target_protocol,
1344 			     struct rte_flow_error *error)
1345 {
1346 	const struct rte_flow_item_icmp *mask = item->mask;
1347 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1348 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1349 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1350 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1351 				      MLX5_FLOW_LAYER_OUTER_L4;
1352 	int ret;
1353 
1354 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1355 		return rte_flow_error_set(error, EINVAL,
1356 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1357 					  "protocol filtering not compatible"
1358 					  " with ICMP layer");
1359 	if (!(item_flags & l3m))
1360 		return rte_flow_error_set(error, EINVAL,
1361 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1362 					  "IPv4 is mandatory to filter"
1363 					  " on ICMP");
1364 	if (item_flags & l4m)
1365 		return rte_flow_error_set(error, EINVAL,
1366 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1367 					  "multiple L4 layers not supported");
1368 	if (!mask)
1369 		mask = &rte_flow_item_icmp_mask;
1370 	ret = mlx5_flow_item_acceptable
1371 		(item, (const uint8_t *)mask,
1372 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1373 		 sizeof(struct rte_flow_item_icmp), error);
1374 	if (ret < 0)
1375 		return ret;
1376 	return 0;
1377 }
1378 
1379 /**
1380  * Validate Ethernet item.
1381  *
1382  * @param[in] item
1383  *   Item specification.
1384  * @param[in] item_flags
1385  *   Bit-fields that holds the items detected until now.
1386  * @param[out] error
1387  *   Pointer to error structure.
1388  *
1389  * @return
1390  *   0 on success, a negative errno value otherwise and rte_errno is set.
1391  */
1392 int
1393 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1394 			    uint64_t item_flags,
1395 			    struct rte_flow_error *error)
1396 {
1397 	const struct rte_flow_item_eth *mask = item->mask;
1398 	const struct rte_flow_item_eth nic_mask = {
1399 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1400 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1401 		.type = RTE_BE16(0xffff),
1402 	};
1403 	int ret;
1404 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1405 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1406 				       MLX5_FLOW_LAYER_OUTER_L2;
1407 
1408 	if (item_flags & ethm)
1409 		return rte_flow_error_set(error, ENOTSUP,
1410 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1411 					  "multiple L2 layers not supported");
1412 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1413 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1414 		return rte_flow_error_set(error, EINVAL,
1415 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1416 					  "L2 layer should not follow "
1417 					  "L3 layers");
1418 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1419 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1420 		return rte_flow_error_set(error, EINVAL,
1421 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1422 					  "L2 layer should not follow VLAN");
1423 	if (!mask)
1424 		mask = &rte_flow_item_eth_mask;
1425 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1426 					(const uint8_t *)&nic_mask,
1427 					sizeof(struct rte_flow_item_eth),
1428 					error);
1429 	return ret;
1430 }
1431 
1432 /**
1433  * Validate VLAN item.
1434  *
1435  * @param[in] item
1436  *   Item specification.
1437  * @param[in] item_flags
1438  *   Bit-fields that holds the items detected until now.
1439  * @param[in] dev
1440  *   Ethernet device flow is being created on.
1441  * @param[out] error
1442  *   Pointer to error structure.
1443  *
1444  * @return
1445  *   0 on success, a negative errno value otherwise and rte_errno is set.
1446  */
1447 int
1448 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1449 			     uint64_t item_flags,
1450 			     struct rte_eth_dev *dev,
1451 			     struct rte_flow_error *error)
1452 {
1453 	const struct rte_flow_item_vlan *spec = item->spec;
1454 	const struct rte_flow_item_vlan *mask = item->mask;
1455 	const struct rte_flow_item_vlan nic_mask = {
1456 		.tci = RTE_BE16(UINT16_MAX),
1457 		.inner_type = RTE_BE16(UINT16_MAX),
1458 	};
1459 	uint16_t vlan_tag = 0;
1460 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1461 	int ret;
1462 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1463 					MLX5_FLOW_LAYER_INNER_L4) :
1464 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1465 					MLX5_FLOW_LAYER_OUTER_L4);
1466 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1467 					MLX5_FLOW_LAYER_OUTER_VLAN;
1468 
1469 	if (item_flags & vlanm)
1470 		return rte_flow_error_set(error, EINVAL,
1471 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1472 					  "multiple VLAN layers not supported");
1473 	else if ((item_flags & l34m) != 0)
1474 		return rte_flow_error_set(error, EINVAL,
1475 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1476 					  "VLAN cannot follow L3/L4 layer");
1477 	if (!mask)
1478 		mask = &rte_flow_item_vlan_mask;
1479 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1480 					(const uint8_t *)&nic_mask,
1481 					sizeof(struct rte_flow_item_vlan),
1482 					error);
1483 	if (ret)
1484 		return ret;
1485 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1486 		struct mlx5_priv *priv = dev->data->dev_private;
1487 
1488 		if (priv->vmwa_context) {
1489 			/*
1490 			 * Non-NULL context means we have a virtual machine
1491 			 * and SR-IOV enabled, we have to create VLAN interface
1492 			 * to make hypervisor to setup E-Switch vport
1493 			 * context correctly. We avoid creating the multiple
1494 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1495 			 */
1496 			return rte_flow_error_set(error, EINVAL,
1497 						  RTE_FLOW_ERROR_TYPE_ITEM,
1498 						  item,
1499 						  "VLAN tag mask is not"
1500 						  " supported in virtual"
1501 						  " environment");
1502 		}
1503 	}
1504 	if (spec) {
1505 		vlan_tag = spec->tci;
1506 		vlan_tag &= mask->tci;
1507 	}
1508 	/*
1509 	 * From verbs perspective an empty VLAN is equivalent
1510 	 * to a packet without VLAN layer.
1511 	 */
1512 	if (!vlan_tag)
1513 		return rte_flow_error_set(error, EINVAL,
1514 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1515 					  item->spec,
1516 					  "VLAN cannot be empty");
1517 	return 0;
1518 }
1519 
1520 /**
1521  * Validate IPV4 item.
1522  *
1523  * @param[in] item
1524  *   Item specification.
1525  * @param[in] item_flags
1526  *   Bit-fields that holds the items detected until now.
1527  * @param[in] acc_mask
1528  *   Acceptable mask, if NULL default internal default mask
1529  *   will be used to check whether item fields are supported.
1530  * @param[out] error
1531  *   Pointer to error structure.
1532  *
1533  * @return
1534  *   0 on success, a negative errno value otherwise and rte_errno is set.
1535  */
1536 int
1537 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1538 			     uint64_t item_flags,
1539 			     uint64_t last_item,
1540 			     uint16_t ether_type,
1541 			     const struct rte_flow_item_ipv4 *acc_mask,
1542 			     struct rte_flow_error *error)
1543 {
1544 	const struct rte_flow_item_ipv4 *mask = item->mask;
1545 	const struct rte_flow_item_ipv4 *spec = item->spec;
1546 	const struct rte_flow_item_ipv4 nic_mask = {
1547 		.hdr = {
1548 			.src_addr = RTE_BE32(0xffffffff),
1549 			.dst_addr = RTE_BE32(0xffffffff),
1550 			.type_of_service = 0xff,
1551 			.next_proto_id = 0xff,
1552 		},
1553 	};
1554 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1555 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1556 				      MLX5_FLOW_LAYER_OUTER_L3;
1557 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1558 				      MLX5_FLOW_LAYER_OUTER_L4;
1559 	int ret;
1560 	uint8_t next_proto = 0xFF;
1561 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1562 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1563 				  MLX5_FLOW_LAYER_INNER_VLAN);
1564 
1565 	if ((last_item & l2_vlan) && ether_type &&
1566 	    ether_type != RTE_ETHER_TYPE_IPV4)
1567 		return rte_flow_error_set(error, EINVAL,
1568 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1569 					  "IPv4 cannot follow L2/VLAN layer "
1570 					  "which ether type is not IPv4");
1571 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1572 		if (mask && spec)
1573 			next_proto = mask->hdr.next_proto_id &
1574 				     spec->hdr.next_proto_id;
1575 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1576 			return rte_flow_error_set(error, EINVAL,
1577 						  RTE_FLOW_ERROR_TYPE_ITEM,
1578 						  item,
1579 						  "multiple tunnel "
1580 						  "not supported");
1581 	}
1582 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1583 		return rte_flow_error_set(error, EINVAL,
1584 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1585 					  "wrong tunnel type - IPv6 specified "
1586 					  "but IPv4 item provided");
1587 	if (item_flags & l3m)
1588 		return rte_flow_error_set(error, ENOTSUP,
1589 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1590 					  "multiple L3 layers not supported");
1591 	else if (item_flags & l4m)
1592 		return rte_flow_error_set(error, EINVAL,
1593 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1594 					  "L3 cannot follow an L4 layer.");
1595 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1596 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1597 		return rte_flow_error_set(error, EINVAL,
1598 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1599 					  "L3 cannot follow an NVGRE layer.");
1600 	if (!mask)
1601 		mask = &rte_flow_item_ipv4_mask;
1602 	else if (mask->hdr.next_proto_id != 0 &&
1603 		 mask->hdr.next_proto_id != 0xff)
1604 		return rte_flow_error_set(error, EINVAL,
1605 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1606 					  "partial mask is not supported"
1607 					  " for protocol");
1608 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1609 					acc_mask ? (const uint8_t *)acc_mask
1610 						 : (const uint8_t *)&nic_mask,
1611 					sizeof(struct rte_flow_item_ipv4),
1612 					error);
1613 	if (ret < 0)
1614 		return ret;
1615 	return 0;
1616 }
1617 
1618 /**
1619  * Validate IPV6 item.
1620  *
1621  * @param[in] item
1622  *   Item specification.
1623  * @param[in] item_flags
1624  *   Bit-fields that holds the items detected until now.
1625  * @param[in] acc_mask
1626  *   Acceptable mask, if NULL default internal default mask
1627  *   will be used to check whether item fields are supported.
1628  * @param[out] error
1629  *   Pointer to error structure.
1630  *
1631  * @return
1632  *   0 on success, a negative errno value otherwise and rte_errno is set.
1633  */
1634 int
1635 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1636 			     uint64_t item_flags,
1637 			     uint64_t last_item,
1638 			     uint16_t ether_type,
1639 			     const struct rte_flow_item_ipv6 *acc_mask,
1640 			     struct rte_flow_error *error)
1641 {
1642 	const struct rte_flow_item_ipv6 *mask = item->mask;
1643 	const struct rte_flow_item_ipv6 *spec = item->spec;
1644 	const struct rte_flow_item_ipv6 nic_mask = {
1645 		.hdr = {
1646 			.src_addr =
1647 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1648 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1649 			.dst_addr =
1650 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1651 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1652 			.vtc_flow = RTE_BE32(0xffffffff),
1653 			.proto = 0xff,
1654 		},
1655 	};
1656 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1657 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1658 				      MLX5_FLOW_LAYER_OUTER_L3;
1659 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1660 				      MLX5_FLOW_LAYER_OUTER_L4;
1661 	int ret;
1662 	uint8_t next_proto = 0xFF;
1663 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1664 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1665 				  MLX5_FLOW_LAYER_INNER_VLAN);
1666 
1667 	if ((last_item & l2_vlan) && ether_type &&
1668 	    ether_type != RTE_ETHER_TYPE_IPV6)
1669 		return rte_flow_error_set(error, EINVAL,
1670 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1671 					  "IPv6 cannot follow L2/VLAN layer "
1672 					  "which ether type is not IPv6");
1673 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1674 		if (mask && spec)
1675 			next_proto = mask->hdr.proto & spec->hdr.proto;
1676 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1677 			return rte_flow_error_set(error, EINVAL,
1678 						  RTE_FLOW_ERROR_TYPE_ITEM,
1679 						  item,
1680 						  "multiple tunnel "
1681 						  "not supported");
1682 	}
1683 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1684 		return rte_flow_error_set(error, EINVAL,
1685 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1686 					  "wrong tunnel type - IPv4 specified "
1687 					  "but IPv6 item provided");
1688 	if (item_flags & l3m)
1689 		return rte_flow_error_set(error, ENOTSUP,
1690 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1691 					  "multiple L3 layers not supported");
1692 	else if (item_flags & l4m)
1693 		return rte_flow_error_set(error, EINVAL,
1694 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1695 					  "L3 cannot follow an L4 layer.");
1696 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1697 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1698 		return rte_flow_error_set(error, EINVAL,
1699 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1700 					  "L3 cannot follow an NVGRE layer.");
1701 	if (!mask)
1702 		mask = &rte_flow_item_ipv6_mask;
1703 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1704 					acc_mask ? (const uint8_t *)acc_mask
1705 						 : (const uint8_t *)&nic_mask,
1706 					sizeof(struct rte_flow_item_ipv6),
1707 					error);
1708 	if (ret < 0)
1709 		return ret;
1710 	return 0;
1711 }
1712 
1713 /**
1714  * Validate UDP item.
1715  *
1716  * @param[in] item
1717  *   Item specification.
1718  * @param[in] item_flags
1719  *   Bit-fields that holds the items detected until now.
1720  * @param[in] target_protocol
1721  *   The next protocol in the previous item.
1722  * @param[in] flow_mask
1723  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1724  * @param[out] error
1725  *   Pointer to error structure.
1726  *
1727  * @return
1728  *   0 on success, a negative errno value otherwise and rte_errno is set.
1729  */
1730 int
1731 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1732 			    uint64_t item_flags,
1733 			    uint8_t target_protocol,
1734 			    struct rte_flow_error *error)
1735 {
1736 	const struct rte_flow_item_udp *mask = item->mask;
1737 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1738 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1739 				      MLX5_FLOW_LAYER_OUTER_L3;
1740 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1741 				      MLX5_FLOW_LAYER_OUTER_L4;
1742 	int ret;
1743 
1744 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1745 		return rte_flow_error_set(error, EINVAL,
1746 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1747 					  "protocol filtering not compatible"
1748 					  " with UDP layer");
1749 	if (!(item_flags & l3m))
1750 		return rte_flow_error_set(error, EINVAL,
1751 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1752 					  "L3 is mandatory to filter on L4");
1753 	if (item_flags & l4m)
1754 		return rte_flow_error_set(error, EINVAL,
1755 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1756 					  "multiple L4 layers not supported");
1757 	if (!mask)
1758 		mask = &rte_flow_item_udp_mask;
1759 	ret = mlx5_flow_item_acceptable
1760 		(item, (const uint8_t *)mask,
1761 		 (const uint8_t *)&rte_flow_item_udp_mask,
1762 		 sizeof(struct rte_flow_item_udp), error);
1763 	if (ret < 0)
1764 		return ret;
1765 	return 0;
1766 }
1767 
1768 /**
1769  * Validate TCP item.
1770  *
1771  * @param[in] item
1772  *   Item specification.
1773  * @param[in] item_flags
1774  *   Bit-fields that holds the items detected until now.
1775  * @param[in] target_protocol
1776  *   The next protocol in the previous item.
1777  * @param[out] error
1778  *   Pointer to error structure.
1779  *
1780  * @return
1781  *   0 on success, a negative errno value otherwise and rte_errno is set.
1782  */
1783 int
1784 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1785 			    uint64_t item_flags,
1786 			    uint8_t target_protocol,
1787 			    const struct rte_flow_item_tcp *flow_mask,
1788 			    struct rte_flow_error *error)
1789 {
1790 	const struct rte_flow_item_tcp *mask = item->mask;
1791 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1792 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1793 				      MLX5_FLOW_LAYER_OUTER_L3;
1794 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1795 				      MLX5_FLOW_LAYER_OUTER_L4;
1796 	int ret;
1797 
1798 	MLX5_ASSERT(flow_mask);
1799 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1800 		return rte_flow_error_set(error, EINVAL,
1801 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1802 					  "protocol filtering not compatible"
1803 					  " with TCP layer");
1804 	if (!(item_flags & l3m))
1805 		return rte_flow_error_set(error, EINVAL,
1806 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1807 					  "L3 is mandatory to filter on L4");
1808 	if (item_flags & l4m)
1809 		return rte_flow_error_set(error, EINVAL,
1810 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1811 					  "multiple L4 layers not supported");
1812 	if (!mask)
1813 		mask = &rte_flow_item_tcp_mask;
1814 	ret = mlx5_flow_item_acceptable
1815 		(item, (const uint8_t *)mask,
1816 		 (const uint8_t *)flow_mask,
1817 		 sizeof(struct rte_flow_item_tcp), error);
1818 	if (ret < 0)
1819 		return ret;
1820 	return 0;
1821 }
1822 
1823 /**
1824  * Validate VXLAN item.
1825  *
1826  * @param[in] item
1827  *   Item specification.
1828  * @param[in] item_flags
1829  *   Bit-fields that holds the items detected until now.
1830  * @param[in] target_protocol
1831  *   The next protocol in the previous item.
1832  * @param[out] error
1833  *   Pointer to error structure.
1834  *
1835  * @return
1836  *   0 on success, a negative errno value otherwise and rte_errno is set.
1837  */
1838 int
1839 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1840 			      uint64_t item_flags,
1841 			      struct rte_flow_error *error)
1842 {
1843 	const struct rte_flow_item_vxlan *spec = item->spec;
1844 	const struct rte_flow_item_vxlan *mask = item->mask;
1845 	int ret;
1846 	union vni {
1847 		uint32_t vlan_id;
1848 		uint8_t vni[4];
1849 	} id = { .vlan_id = 0, };
1850 
1851 
1852 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1853 		return rte_flow_error_set(error, ENOTSUP,
1854 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1855 					  "multiple tunnel layers not"
1856 					  " supported");
1857 	/*
1858 	 * Verify only UDPv4 is present as defined in
1859 	 * https://tools.ietf.org/html/rfc7348
1860 	 */
1861 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1862 		return rte_flow_error_set(error, EINVAL,
1863 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1864 					  "no outer UDP layer found");
1865 	if (!mask)
1866 		mask = &rte_flow_item_vxlan_mask;
1867 	ret = mlx5_flow_item_acceptable
1868 		(item, (const uint8_t *)mask,
1869 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1870 		 sizeof(struct rte_flow_item_vxlan),
1871 		 error);
1872 	if (ret < 0)
1873 		return ret;
1874 	if (spec) {
1875 		memcpy(&id.vni[1], spec->vni, 3);
1876 		memcpy(&id.vni[1], mask->vni, 3);
1877 	}
1878 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1879 		return rte_flow_error_set(error, ENOTSUP,
1880 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1881 					  "VXLAN tunnel must be fully defined");
1882 	return 0;
1883 }
1884 
1885 /**
1886  * Validate VXLAN_GPE item.
1887  *
1888  * @param[in] item
1889  *   Item specification.
1890  * @param[in] item_flags
1891  *   Bit-fields that holds the items detected until now.
1892  * @param[in] priv
1893  *   Pointer to the private data structure.
1894  * @param[in] target_protocol
1895  *   The next protocol in the previous item.
1896  * @param[out] error
1897  *   Pointer to error structure.
1898  *
1899  * @return
1900  *   0 on success, a negative errno value otherwise and rte_errno is set.
1901  */
1902 int
1903 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1904 				  uint64_t item_flags,
1905 				  struct rte_eth_dev *dev,
1906 				  struct rte_flow_error *error)
1907 {
1908 	struct mlx5_priv *priv = dev->data->dev_private;
1909 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1910 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1911 	int ret;
1912 	union vni {
1913 		uint32_t vlan_id;
1914 		uint8_t vni[4];
1915 	} id = { .vlan_id = 0, };
1916 
1917 	if (!priv->config.l3_vxlan_en)
1918 		return rte_flow_error_set(error, ENOTSUP,
1919 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1920 					  "L3 VXLAN is not enabled by device"
1921 					  " parameter and/or not configured in"
1922 					  " firmware");
1923 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1924 		return rte_flow_error_set(error, ENOTSUP,
1925 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1926 					  "multiple tunnel layers not"
1927 					  " supported");
1928 	/*
1929 	 * Verify only UDPv4 is present as defined in
1930 	 * https://tools.ietf.org/html/rfc7348
1931 	 */
1932 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1933 		return rte_flow_error_set(error, EINVAL,
1934 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1935 					  "no outer UDP layer found");
1936 	if (!mask)
1937 		mask = &rte_flow_item_vxlan_gpe_mask;
1938 	ret = mlx5_flow_item_acceptable
1939 		(item, (const uint8_t *)mask,
1940 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1941 		 sizeof(struct rte_flow_item_vxlan_gpe),
1942 		 error);
1943 	if (ret < 0)
1944 		return ret;
1945 	if (spec) {
1946 		if (spec->protocol)
1947 			return rte_flow_error_set(error, ENOTSUP,
1948 						  RTE_FLOW_ERROR_TYPE_ITEM,
1949 						  item,
1950 						  "VxLAN-GPE protocol"
1951 						  " not supported");
1952 		memcpy(&id.vni[1], spec->vni, 3);
1953 		memcpy(&id.vni[1], mask->vni, 3);
1954 	}
1955 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1956 		return rte_flow_error_set(error, ENOTSUP,
1957 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1958 					  "VXLAN-GPE tunnel must be fully"
1959 					  " defined");
1960 	return 0;
1961 }
1962 /**
1963  * Validate GRE Key item.
1964  *
1965  * @param[in] item
1966  *   Item specification.
1967  * @param[in] item_flags
1968  *   Bit flags to mark detected items.
1969  * @param[in] gre_item
1970  *   Pointer to gre_item
1971  * @param[out] error
1972  *   Pointer to error structure.
1973  *
1974  * @return
1975  *   0 on success, a negative errno value otherwise and rte_errno is set.
1976  */
1977 int
1978 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1979 				uint64_t item_flags,
1980 				const struct rte_flow_item *gre_item,
1981 				struct rte_flow_error *error)
1982 {
1983 	const rte_be32_t *mask = item->mask;
1984 	int ret = 0;
1985 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1986 	const struct rte_flow_item_gre *gre_spec;
1987 	const struct rte_flow_item_gre *gre_mask;
1988 
1989 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1990 		return rte_flow_error_set(error, ENOTSUP,
1991 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1992 					  "Multiple GRE key not support");
1993 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1994 		return rte_flow_error_set(error, ENOTSUP,
1995 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1996 					  "No preceding GRE header");
1997 	if (item_flags & MLX5_FLOW_LAYER_INNER)
1998 		return rte_flow_error_set(error, ENOTSUP,
1999 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2000 					  "GRE key following a wrong item");
2001 	gre_mask = gre_item->mask;
2002 	if (!gre_mask)
2003 		gre_mask = &rte_flow_item_gre_mask;
2004 	gre_spec = gre_item->spec;
2005 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2006 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2007 		return rte_flow_error_set(error, EINVAL,
2008 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2009 					  "Key bit must be on");
2010 
2011 	if (!mask)
2012 		mask = &gre_key_default_mask;
2013 	ret = mlx5_flow_item_acceptable
2014 		(item, (const uint8_t *)mask,
2015 		 (const uint8_t *)&gre_key_default_mask,
2016 		 sizeof(rte_be32_t), error);
2017 	return ret;
2018 }
2019 
2020 /**
2021  * Validate GRE item.
2022  *
2023  * @param[in] item
2024  *   Item specification.
2025  * @param[in] item_flags
2026  *   Bit flags to mark detected items.
2027  * @param[in] target_protocol
2028  *   The next protocol in the previous item.
2029  * @param[out] error
2030  *   Pointer to error structure.
2031  *
2032  * @return
2033  *   0 on success, a negative errno value otherwise and rte_errno is set.
2034  */
2035 int
2036 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2037 			    uint64_t item_flags,
2038 			    uint8_t target_protocol,
2039 			    struct rte_flow_error *error)
2040 {
2041 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2042 	const struct rte_flow_item_gre *mask = item->mask;
2043 	int ret;
2044 	const struct rte_flow_item_gre nic_mask = {
2045 		.c_rsvd0_ver = RTE_BE16(0xB000),
2046 		.protocol = RTE_BE16(UINT16_MAX),
2047 	};
2048 
2049 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2050 		return rte_flow_error_set(error, EINVAL,
2051 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2052 					  "protocol filtering not compatible"
2053 					  " with this GRE layer");
2054 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2055 		return rte_flow_error_set(error, ENOTSUP,
2056 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2057 					  "multiple tunnel layers not"
2058 					  " supported");
2059 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2060 		return rte_flow_error_set(error, ENOTSUP,
2061 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2062 					  "L3 Layer is missing");
2063 	if (!mask)
2064 		mask = &rte_flow_item_gre_mask;
2065 	ret = mlx5_flow_item_acceptable
2066 		(item, (const uint8_t *)mask,
2067 		 (const uint8_t *)&nic_mask,
2068 		 sizeof(struct rte_flow_item_gre), error);
2069 	if (ret < 0)
2070 		return ret;
2071 #ifndef HAVE_MLX5DV_DR
2072 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2073 	if (spec && (spec->protocol & mask->protocol))
2074 		return rte_flow_error_set(error, ENOTSUP,
2075 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2076 					  "without MPLS support the"
2077 					  " specification cannot be used for"
2078 					  " filtering");
2079 #endif
2080 #endif
2081 	return 0;
2082 }
2083 
2084 /**
2085  * Validate Geneve item.
2086  *
2087  * @param[in] item
2088  *   Item specification.
2089  * @param[in] itemFlags
2090  *   Bit-fields that holds the items detected until now.
2091  * @param[in] enPriv
2092  *   Pointer to the private data structure.
2093  * @param[out] error
2094  *   Pointer to error structure.
2095  *
2096  * @return
2097  *   0 on success, a negative errno value otherwise and rte_errno is set.
2098  */
2099 
2100 int
2101 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2102 			       uint64_t item_flags,
2103 			       struct rte_eth_dev *dev,
2104 			       struct rte_flow_error *error)
2105 {
2106 	struct mlx5_priv *priv = dev->data->dev_private;
2107 	const struct rte_flow_item_geneve *spec = item->spec;
2108 	const struct rte_flow_item_geneve *mask = item->mask;
2109 	int ret;
2110 	uint16_t gbhdr;
2111 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2112 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2113 	const struct rte_flow_item_geneve nic_mask = {
2114 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2115 		.vni = "\xff\xff\xff",
2116 		.protocol = RTE_BE16(UINT16_MAX),
2117 	};
2118 
2119 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2120 		return rte_flow_error_set(error, ENOTSUP,
2121 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2122 					  "L3 Geneve is not enabled by device"
2123 					  " parameter and/or not configured in"
2124 					  " firmware");
2125 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2126 		return rte_flow_error_set(error, ENOTSUP,
2127 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2128 					  "multiple tunnel layers not"
2129 					  " supported");
2130 	/*
2131 	 * Verify only UDPv4 is present as defined in
2132 	 * https://tools.ietf.org/html/rfc7348
2133 	 */
2134 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2135 		return rte_flow_error_set(error, EINVAL,
2136 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2137 					  "no outer UDP layer found");
2138 	if (!mask)
2139 		mask = &rte_flow_item_geneve_mask;
2140 	ret = mlx5_flow_item_acceptable
2141 				  (item, (const uint8_t *)mask,
2142 				   (const uint8_t *)&nic_mask,
2143 				   sizeof(struct rte_flow_item_geneve), error);
2144 	if (ret)
2145 		return ret;
2146 	if (spec) {
2147 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2148 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2149 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2150 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2151 			return rte_flow_error_set(error, ENOTSUP,
2152 						  RTE_FLOW_ERROR_TYPE_ITEM,
2153 						  item,
2154 						  "Geneve protocol unsupported"
2155 						  " fields are being used");
2156 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2157 			return rte_flow_error_set
2158 					(error, ENOTSUP,
2159 					 RTE_FLOW_ERROR_TYPE_ITEM,
2160 					 item,
2161 					 "Unsupported Geneve options length");
2162 	}
2163 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2164 		return rte_flow_error_set
2165 				    (error, ENOTSUP,
2166 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2167 				     "Geneve tunnel must be fully defined");
2168 	return 0;
2169 }
2170 
2171 /**
2172  * Validate MPLS item.
2173  *
2174  * @param[in] dev
2175  *   Pointer to the rte_eth_dev structure.
2176  * @param[in] item
2177  *   Item specification.
2178  * @param[in] item_flags
2179  *   Bit-fields that holds the items detected until now.
2180  * @param[in] prev_layer
2181  *   The protocol layer indicated in previous item.
2182  * @param[out] error
2183  *   Pointer to error structure.
2184  *
2185  * @return
2186  *   0 on success, a negative errno value otherwise and rte_errno is set.
2187  */
2188 int
2189 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2190 			     const struct rte_flow_item *item __rte_unused,
2191 			     uint64_t item_flags __rte_unused,
2192 			     uint64_t prev_layer __rte_unused,
2193 			     struct rte_flow_error *error)
2194 {
2195 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2196 	const struct rte_flow_item_mpls *mask = item->mask;
2197 	struct mlx5_priv *priv = dev->data->dev_private;
2198 	int ret;
2199 
2200 	if (!priv->config.mpls_en)
2201 		return rte_flow_error_set(error, ENOTSUP,
2202 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2203 					  "MPLS not supported or"
2204 					  " disabled in firmware"
2205 					  " configuration.");
2206 	/* MPLS over IP, UDP, GRE is allowed */
2207 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2208 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2209 			    MLX5_FLOW_LAYER_GRE)))
2210 		return rte_flow_error_set(error, EINVAL,
2211 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2212 					  "protocol filtering not compatible"
2213 					  " with MPLS layer");
2214 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2215 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2216 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2217 		return rte_flow_error_set(error, ENOTSUP,
2218 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2219 					  "multiple tunnel layers not"
2220 					  " supported");
2221 	if (!mask)
2222 		mask = &rte_flow_item_mpls_mask;
2223 	ret = mlx5_flow_item_acceptable
2224 		(item, (const uint8_t *)mask,
2225 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2226 		 sizeof(struct rte_flow_item_mpls), error);
2227 	if (ret < 0)
2228 		return ret;
2229 	return 0;
2230 #endif
2231 	return rte_flow_error_set(error, ENOTSUP,
2232 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2233 				  "MPLS is not supported by Verbs, please"
2234 				  " update.");
2235 }
2236 
2237 /**
2238  * Validate NVGRE item.
2239  *
2240  * @param[in] item
2241  *   Item specification.
2242  * @param[in] item_flags
2243  *   Bit flags to mark detected items.
2244  * @param[in] target_protocol
2245  *   The next protocol in the previous item.
2246  * @param[out] error
2247  *   Pointer to error structure.
2248  *
2249  * @return
2250  *   0 on success, a negative errno value otherwise and rte_errno is set.
2251  */
2252 int
2253 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2254 			      uint64_t item_flags,
2255 			      uint8_t target_protocol,
2256 			      struct rte_flow_error *error)
2257 {
2258 	const struct rte_flow_item_nvgre *mask = item->mask;
2259 	int ret;
2260 
2261 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2262 		return rte_flow_error_set(error, EINVAL,
2263 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2264 					  "protocol filtering not compatible"
2265 					  " with this GRE layer");
2266 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2267 		return rte_flow_error_set(error, ENOTSUP,
2268 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2269 					  "multiple tunnel layers not"
2270 					  " supported");
2271 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2272 		return rte_flow_error_set(error, ENOTSUP,
2273 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2274 					  "L3 Layer is missing");
2275 	if (!mask)
2276 		mask = &rte_flow_item_nvgre_mask;
2277 	ret = mlx5_flow_item_acceptable
2278 		(item, (const uint8_t *)mask,
2279 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2280 		 sizeof(struct rte_flow_item_nvgre), error);
2281 	if (ret < 0)
2282 		return ret;
2283 	return 0;
2284 }
2285 
2286 /* Allocate unique ID for the split Q/RSS subflows. */
2287 static uint32_t
2288 flow_qrss_get_id(struct rte_eth_dev *dev)
2289 {
2290 	struct mlx5_priv *priv = dev->data->dev_private;
2291 	uint32_t qrss_id, ret;
2292 
2293 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2294 	if (ret)
2295 		return 0;
2296 	MLX5_ASSERT(qrss_id);
2297 	return qrss_id;
2298 }
2299 
2300 /* Free unique ID for the split Q/RSS subflows. */
2301 static void
2302 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2303 {
2304 	struct mlx5_priv *priv = dev->data->dev_private;
2305 
2306 	if (qrss_id)
2307 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2308 }
2309 
2310 /**
2311  * Release resource related QUEUE/RSS action split.
2312  *
2313  * @param dev
2314  *   Pointer to Ethernet device.
2315  * @param flow
2316  *   Flow to release id's from.
2317  */
2318 static void
2319 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2320 			     struct rte_flow *flow)
2321 {
2322 	struct mlx5_priv *priv = dev->data->dev_private;
2323 	uint32_t handle_idx;
2324 	struct mlx5_flow_handle *dev_handle;
2325 
2326 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2327 		       handle_idx, dev_handle, next)
2328 		if (dev_handle->qrss_id)
2329 			flow_qrss_free_id(dev, dev_handle->qrss_id);
2330 }
2331 
2332 static int
2333 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2334 		   const struct rte_flow_attr *attr __rte_unused,
2335 		   const struct rte_flow_item items[] __rte_unused,
2336 		   const struct rte_flow_action actions[] __rte_unused,
2337 		   bool external __rte_unused,
2338 		   struct rte_flow_error *error)
2339 {
2340 	return rte_flow_error_set(error, ENOTSUP,
2341 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2342 }
2343 
2344 static struct mlx5_flow *
2345 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2346 		  const struct rte_flow_attr *attr __rte_unused,
2347 		  const struct rte_flow_item items[] __rte_unused,
2348 		  const struct rte_flow_action actions[] __rte_unused,
2349 		  struct rte_flow_error *error)
2350 {
2351 	rte_flow_error_set(error, ENOTSUP,
2352 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2353 	return NULL;
2354 }
2355 
2356 static int
2357 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2358 		    struct mlx5_flow *dev_flow __rte_unused,
2359 		    const struct rte_flow_attr *attr __rte_unused,
2360 		    const struct rte_flow_item items[] __rte_unused,
2361 		    const struct rte_flow_action actions[] __rte_unused,
2362 		    struct rte_flow_error *error)
2363 {
2364 	return rte_flow_error_set(error, ENOTSUP,
2365 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2366 }
2367 
2368 static int
2369 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2370 		struct rte_flow *flow __rte_unused,
2371 		struct rte_flow_error *error)
2372 {
2373 	return rte_flow_error_set(error, ENOTSUP,
2374 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2375 }
2376 
2377 static void
2378 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2379 		 struct rte_flow *flow __rte_unused)
2380 {
2381 }
2382 
2383 static void
2384 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2385 		  struct rte_flow *flow __rte_unused)
2386 {
2387 }
2388 
2389 static int
2390 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2391 		struct rte_flow *flow __rte_unused,
2392 		const struct rte_flow_action *actions __rte_unused,
2393 		void *data __rte_unused,
2394 		struct rte_flow_error *error)
2395 {
2396 	return rte_flow_error_set(error, ENOTSUP,
2397 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2398 }
2399 
2400 /* Void driver to protect from null pointer reference. */
2401 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2402 	.validate = flow_null_validate,
2403 	.prepare = flow_null_prepare,
2404 	.translate = flow_null_translate,
2405 	.apply = flow_null_apply,
2406 	.remove = flow_null_remove,
2407 	.destroy = flow_null_destroy,
2408 	.query = flow_null_query,
2409 };
2410 
2411 /**
2412  * Select flow driver type according to flow attributes and device
2413  * configuration.
2414  *
2415  * @param[in] dev
2416  *   Pointer to the dev structure.
2417  * @param[in] attr
2418  *   Pointer to the flow attributes.
2419  *
2420  * @return
2421  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2422  */
2423 static enum mlx5_flow_drv_type
2424 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2425 {
2426 	struct mlx5_priv *priv = dev->data->dev_private;
2427 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2428 
2429 	if (attr->transfer && priv->config.dv_esw_en)
2430 		type = MLX5_FLOW_TYPE_DV;
2431 	if (!attr->transfer)
2432 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2433 						 MLX5_FLOW_TYPE_VERBS;
2434 	return type;
2435 }
2436 
2437 #define flow_get_drv_ops(type) flow_drv_ops[type]
2438 
2439 /**
2440  * Flow driver validation API. This abstracts calling driver specific functions.
2441  * The type of flow driver is determined according to flow attributes.
2442  *
2443  * @param[in] dev
2444  *   Pointer to the dev structure.
2445  * @param[in] attr
2446  *   Pointer to the flow attributes.
2447  * @param[in] items
2448  *   Pointer to the list of items.
2449  * @param[in] actions
2450  *   Pointer to the list of actions.
2451  * @param[in] external
2452  *   This flow rule is created by request external to PMD.
2453  * @param[out] error
2454  *   Pointer to the error structure.
2455  *
2456  * @return
2457  *   0 on success, a negative errno value otherwise and rte_errno is set.
2458  */
2459 static inline int
2460 flow_drv_validate(struct rte_eth_dev *dev,
2461 		  const struct rte_flow_attr *attr,
2462 		  const struct rte_flow_item items[],
2463 		  const struct rte_flow_action actions[],
2464 		  bool external, struct rte_flow_error *error)
2465 {
2466 	const struct mlx5_flow_driver_ops *fops;
2467 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2468 
2469 	fops = flow_get_drv_ops(type);
2470 	return fops->validate(dev, attr, items, actions, external, error);
2471 }
2472 
2473 /**
2474  * Flow driver preparation API. This abstracts calling driver specific
2475  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2476  * calculates the size of memory required for device flow, allocates the memory,
2477  * initializes the device flow and returns the pointer.
2478  *
2479  * @note
2480  *   This function initializes device flow structure such as dv or verbs in
2481  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2482  *   rest. For example, adding returning device flow to flow->dev_flow list and
2483  *   setting backward reference to the flow should be done out of this function.
2484  *   layers field is not filled either.
2485  *
2486  * @param[in] dev
2487  *   Pointer to the dev structure.
2488  * @param[in] attr
2489  *   Pointer to the flow attributes.
2490  * @param[in] items
2491  *   Pointer to the list of items.
2492  * @param[in] actions
2493  *   Pointer to the list of actions.
2494  * @param[out] error
2495  *   Pointer to the error structure.
2496  *
2497  * @return
2498  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2499  */
2500 static inline struct mlx5_flow *
2501 flow_drv_prepare(struct rte_eth_dev *dev,
2502 		 const struct rte_flow *flow,
2503 		 const struct rte_flow_attr *attr,
2504 		 const struct rte_flow_item items[],
2505 		 const struct rte_flow_action actions[],
2506 		 struct rte_flow_error *error)
2507 {
2508 	const struct mlx5_flow_driver_ops *fops;
2509 	enum mlx5_flow_drv_type type = flow->drv_type;
2510 
2511 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2512 	fops = flow_get_drv_ops(type);
2513 	return fops->prepare(dev, attr, items, actions, error);
2514 }
2515 
2516 /**
2517  * Flow driver translation API. This abstracts calling driver specific
2518  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2519  * translates a generic flow into a driver flow. flow_drv_prepare() must
2520  * precede.
2521  *
2522  * @note
2523  *   dev_flow->layers could be filled as a result of parsing during translation
2524  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2525  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2526  *   flow->actions could be overwritten even though all the expanded dev_flows
2527  *   have the same actions.
2528  *
2529  * @param[in] dev
2530  *   Pointer to the rte dev structure.
2531  * @param[in, out] dev_flow
2532  *   Pointer to the mlx5 flow.
2533  * @param[in] attr
2534  *   Pointer to the flow attributes.
2535  * @param[in] items
2536  *   Pointer to the list of items.
2537  * @param[in] actions
2538  *   Pointer to the list of actions.
2539  * @param[out] error
2540  *   Pointer to the error structure.
2541  *
2542  * @return
2543  *   0 on success, a negative errno value otherwise and rte_errno is set.
2544  */
2545 static inline int
2546 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2547 		   const struct rte_flow_attr *attr,
2548 		   const struct rte_flow_item items[],
2549 		   const struct rte_flow_action actions[],
2550 		   struct rte_flow_error *error)
2551 {
2552 	const struct mlx5_flow_driver_ops *fops;
2553 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2554 
2555 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2556 	fops = flow_get_drv_ops(type);
2557 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2558 }
2559 
2560 /**
2561  * Flow driver apply API. This abstracts calling driver specific functions.
2562  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2563  * translated driver flows on to device. flow_drv_translate() must precede.
2564  *
2565  * @param[in] dev
2566  *   Pointer to Ethernet device structure.
2567  * @param[in, out] flow
2568  *   Pointer to flow structure.
2569  * @param[out] error
2570  *   Pointer to error structure.
2571  *
2572  * @return
2573  *   0 on success, a negative errno value otherwise and rte_errno is set.
2574  */
2575 static inline int
2576 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2577 	       struct rte_flow_error *error)
2578 {
2579 	const struct mlx5_flow_driver_ops *fops;
2580 	enum mlx5_flow_drv_type type = flow->drv_type;
2581 
2582 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2583 	fops = flow_get_drv_ops(type);
2584 	return fops->apply(dev, flow, error);
2585 }
2586 
2587 /**
2588  * Flow driver remove API. This abstracts calling driver specific functions.
2589  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2590  * on device. All the resources of the flow should be freed by calling
2591  * flow_drv_destroy().
2592  *
2593  * @param[in] dev
2594  *   Pointer to Ethernet device.
2595  * @param[in, out] flow
2596  *   Pointer to flow structure.
2597  */
2598 static inline void
2599 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2600 {
2601 	const struct mlx5_flow_driver_ops *fops;
2602 	enum mlx5_flow_drv_type type = flow->drv_type;
2603 
2604 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2605 	fops = flow_get_drv_ops(type);
2606 	fops->remove(dev, flow);
2607 }
2608 
2609 /**
2610  * Flow driver destroy API. This abstracts calling driver specific functions.
2611  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2612  * on device and releases resources of the flow.
2613  *
2614  * @param[in] dev
2615  *   Pointer to Ethernet device.
2616  * @param[in, out] flow
2617  *   Pointer to flow structure.
2618  */
2619 static inline void
2620 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2621 {
2622 	const struct mlx5_flow_driver_ops *fops;
2623 	enum mlx5_flow_drv_type type = flow->drv_type;
2624 
2625 	flow_mreg_split_qrss_release(dev, flow);
2626 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2627 	fops = flow_get_drv_ops(type);
2628 	fops->destroy(dev, flow);
2629 }
2630 
2631 /**
2632  * Validate a flow supported by the NIC.
2633  *
2634  * @see rte_flow_validate()
2635  * @see rte_flow_ops
2636  */
2637 int
2638 mlx5_flow_validate(struct rte_eth_dev *dev,
2639 		   const struct rte_flow_attr *attr,
2640 		   const struct rte_flow_item items[],
2641 		   const struct rte_flow_action actions[],
2642 		   struct rte_flow_error *error)
2643 {
2644 	int ret;
2645 
2646 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2647 	if (ret < 0)
2648 		return ret;
2649 	return 0;
2650 }
2651 
2652 /**
2653  * Get RSS action from the action list.
2654  *
2655  * @param[in] actions
2656  *   Pointer to the list of actions.
2657  *
2658  * @return
2659  *   Pointer to the RSS action if exist, else return NULL.
2660  */
2661 static const struct rte_flow_action_rss*
2662 flow_get_rss_action(const struct rte_flow_action actions[])
2663 {
2664 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2665 		switch (actions->type) {
2666 		case RTE_FLOW_ACTION_TYPE_RSS:
2667 			return (const struct rte_flow_action_rss *)
2668 			       actions->conf;
2669 		default:
2670 			break;
2671 		}
2672 	}
2673 	return NULL;
2674 }
2675 
2676 static unsigned int
2677 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2678 {
2679 	const struct rte_flow_item *item;
2680 	unsigned int has_vlan = 0;
2681 
2682 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2683 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2684 			has_vlan = 1;
2685 			break;
2686 		}
2687 	}
2688 	if (has_vlan)
2689 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2690 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2691 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2692 			       MLX5_EXPANSION_ROOT_OUTER;
2693 }
2694 
2695 /**
2696  *  Get layer flags from the prefix flow.
2697  *
2698  *  Some flows may be split to several subflows, the prefix subflow gets the
2699  *  match items and the suffix sub flow gets the actions.
2700  *  Some actions need the user defined match item flags to get the detail for
2701  *  the action.
2702  *  This function helps the suffix flow to get the item layer flags from prefix
2703  *  subflow.
2704  *
2705  * @param[in] dev_flow
2706  *   Pointer the created preifx subflow.
2707  *
2708  * @return
2709  *   The layers get from prefix subflow.
2710  */
2711 static inline uint64_t
2712 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2713 {
2714 	uint64_t layers = 0;
2715 
2716 	/*
2717 	 * Layers bits could be localization, but usually the compiler will
2718 	 * help to do the optimization work for source code.
2719 	 * If no decap actions, use the layers directly.
2720 	 */
2721 	if (!(dev_flow->handle->act_flags & MLX5_FLOW_ACTION_DECAP))
2722 		return dev_flow->handle->layers;
2723 	/* Convert L3 layers with decap action. */
2724 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2725 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2726 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2727 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2728 	/* Convert L4 layers with decap action.  */
2729 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2730 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2731 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2732 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2733 	return layers;
2734 }
2735 
2736 /**
2737  * Get metadata split action information.
2738  *
2739  * @param[in] actions
2740  *   Pointer to the list of actions.
2741  * @param[out] qrss
2742  *   Pointer to the return pointer.
2743  * @param[out] qrss_type
2744  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2745  *   if no QUEUE/RSS is found.
2746  * @param[out] encap_idx
2747  *   Pointer to the index of the encap action if exists, otherwise the last
2748  *   action index.
2749  *
2750  * @return
2751  *   Total number of actions.
2752  */
2753 static int
2754 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2755 				       const struct rte_flow_action **qrss,
2756 				       int *encap_idx)
2757 {
2758 	const struct rte_flow_action_raw_encap *raw_encap;
2759 	int actions_n = 0;
2760 	int raw_decap_idx = -1;
2761 
2762 	*encap_idx = -1;
2763 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2764 		switch (actions->type) {
2765 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2766 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2767 			*encap_idx = actions_n;
2768 			break;
2769 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2770 			raw_decap_idx = actions_n;
2771 			break;
2772 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2773 			raw_encap = actions->conf;
2774 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2775 				*encap_idx = raw_decap_idx != -1 ?
2776 						      raw_decap_idx : actions_n;
2777 			break;
2778 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2779 		case RTE_FLOW_ACTION_TYPE_RSS:
2780 			*qrss = actions;
2781 			break;
2782 		default:
2783 			break;
2784 		}
2785 		actions_n++;
2786 	}
2787 	if (*encap_idx == -1)
2788 		*encap_idx = actions_n;
2789 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2790 	return actions_n + 1;
2791 }
2792 
2793 /**
2794  * Check meter action from the action list.
2795  *
2796  * @param[in] actions
2797  *   Pointer to the list of actions.
2798  * @param[out] mtr
2799  *   Pointer to the meter exist flag.
2800  *
2801  * @return
2802  *   Total number of actions.
2803  */
2804 static int
2805 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2806 {
2807 	int actions_n = 0;
2808 
2809 	MLX5_ASSERT(mtr);
2810 	*mtr = 0;
2811 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2812 		switch (actions->type) {
2813 		case RTE_FLOW_ACTION_TYPE_METER:
2814 			*mtr = 1;
2815 			break;
2816 		default:
2817 			break;
2818 		}
2819 		actions_n++;
2820 	}
2821 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2822 	return actions_n + 1;
2823 }
2824 
2825 /**
2826  * Check if the flow should be splited due to hairpin.
2827  * The reason for the split is that in current HW we can't
2828  * support encap on Rx, so if a flow have encap we move it
2829  * to Tx.
2830  *
2831  * @param dev
2832  *   Pointer to Ethernet device.
2833  * @param[in] attr
2834  *   Flow rule attributes.
2835  * @param[in] actions
2836  *   Associated actions (list terminated by the END action).
2837  *
2838  * @return
2839  *   > 0 the number of actions and the flow should be split,
2840  *   0 when no split required.
2841  */
2842 static int
2843 flow_check_hairpin_split(struct rte_eth_dev *dev,
2844 			 const struct rte_flow_attr *attr,
2845 			 const struct rte_flow_action actions[])
2846 {
2847 	int queue_action = 0;
2848 	int action_n = 0;
2849 	int encap = 0;
2850 	const struct rte_flow_action_queue *queue;
2851 	const struct rte_flow_action_rss *rss;
2852 	const struct rte_flow_action_raw_encap *raw_encap;
2853 
2854 	if (!attr->ingress)
2855 		return 0;
2856 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2857 		switch (actions->type) {
2858 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2859 			queue = actions->conf;
2860 			if (queue == NULL)
2861 				return 0;
2862 			if (mlx5_rxq_get_type(dev, queue->index) !=
2863 			    MLX5_RXQ_TYPE_HAIRPIN)
2864 				return 0;
2865 			queue_action = 1;
2866 			action_n++;
2867 			break;
2868 		case RTE_FLOW_ACTION_TYPE_RSS:
2869 			rss = actions->conf;
2870 			if (rss == NULL || rss->queue_num == 0)
2871 				return 0;
2872 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2873 			    MLX5_RXQ_TYPE_HAIRPIN)
2874 				return 0;
2875 			queue_action = 1;
2876 			action_n++;
2877 			break;
2878 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2879 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2880 			encap = 1;
2881 			action_n++;
2882 			break;
2883 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2884 			raw_encap = actions->conf;
2885 			if (raw_encap->size >
2886 			    (sizeof(struct rte_flow_item_eth) +
2887 			     sizeof(struct rte_flow_item_ipv4)))
2888 				encap = 1;
2889 			action_n++;
2890 			break;
2891 		default:
2892 			action_n++;
2893 			break;
2894 		}
2895 	}
2896 	if (encap == 1 && queue_action)
2897 		return action_n;
2898 	return 0;
2899 }
2900 
2901 /* Declare flow create/destroy prototype in advance. */
2902 static struct rte_flow *
2903 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2904 		 const struct rte_flow_attr *attr,
2905 		 const struct rte_flow_item items[],
2906 		 const struct rte_flow_action actions[],
2907 		 bool external, struct rte_flow_error *error);
2908 
2909 static void
2910 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2911 		  struct rte_flow *flow);
2912 
2913 /**
2914  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2915  *
2916  * As mark_id is unique, if there's already a registered flow for the mark_id,
2917  * return by increasing the reference counter of the resource. Otherwise, create
2918  * the resource (mcp_res) and flow.
2919  *
2920  * Flow looks like,
2921  *   - If ingress port is ANY and reg_c[1] is mark_id,
2922  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2923  *
2924  * For default flow (zero mark_id), flow is like,
2925  *   - If ingress port is ANY,
2926  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2927  *
2928  * @param dev
2929  *   Pointer to Ethernet device.
2930  * @param mark_id
2931  *   ID of MARK action, zero means default flow for META.
2932  * @param[out] error
2933  *   Perform verbose error reporting if not NULL.
2934  *
2935  * @return
2936  *   Associated resource on success, NULL otherwise and rte_errno is set.
2937  */
2938 static struct mlx5_flow_mreg_copy_resource *
2939 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2940 			  struct rte_flow_error *error)
2941 {
2942 	struct mlx5_priv *priv = dev->data->dev_private;
2943 	struct rte_flow_attr attr = {
2944 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2945 		.ingress = 1,
2946 	};
2947 	struct mlx5_rte_flow_item_tag tag_spec = {
2948 		.data = mark_id,
2949 	};
2950 	struct rte_flow_item items[] = {
2951 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2952 	};
2953 	struct rte_flow_action_mark ftag = {
2954 		.id = mark_id,
2955 	};
2956 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2957 		.dst = REG_B,
2958 		.src = 0,
2959 	};
2960 	struct rte_flow_action_jump jump = {
2961 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2962 	};
2963 	struct rte_flow_action actions[] = {
2964 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2965 	};
2966 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2967 	int ret;
2968 
2969 	/* Fill the register fileds in the flow. */
2970 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2971 	if (ret < 0)
2972 		return NULL;
2973 	tag_spec.id = ret;
2974 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2975 	if (ret < 0)
2976 		return NULL;
2977 	cp_mreg.src = ret;
2978 	/* Check if already registered. */
2979 	MLX5_ASSERT(priv->mreg_cp_tbl);
2980 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2981 	if (mcp_res) {
2982 		/* For non-default rule. */
2983 		if (mark_id != MLX5_DEFAULT_COPY_ID)
2984 			mcp_res->refcnt++;
2985 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2986 			    mcp_res->refcnt == 1);
2987 		return mcp_res;
2988 	}
2989 	/* Provide the full width of FLAG specific value. */
2990 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2991 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2992 	/* Build a new flow. */
2993 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
2994 		items[0] = (struct rte_flow_item){
2995 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2996 			.spec = &tag_spec,
2997 		};
2998 		items[1] = (struct rte_flow_item){
2999 			.type = RTE_FLOW_ITEM_TYPE_END,
3000 		};
3001 		actions[0] = (struct rte_flow_action){
3002 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3003 			.conf = &ftag,
3004 		};
3005 		actions[1] = (struct rte_flow_action){
3006 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3007 			.conf = &cp_mreg,
3008 		};
3009 		actions[2] = (struct rte_flow_action){
3010 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3011 			.conf = &jump,
3012 		};
3013 		actions[3] = (struct rte_flow_action){
3014 			.type = RTE_FLOW_ACTION_TYPE_END,
3015 		};
3016 	} else {
3017 		/* Default rule, wildcard match. */
3018 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3019 		items[0] = (struct rte_flow_item){
3020 			.type = RTE_FLOW_ITEM_TYPE_END,
3021 		};
3022 		actions[0] = (struct rte_flow_action){
3023 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3024 			.conf = &cp_mreg,
3025 		};
3026 		actions[1] = (struct rte_flow_action){
3027 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3028 			.conf = &jump,
3029 		};
3030 		actions[2] = (struct rte_flow_action){
3031 			.type = RTE_FLOW_ACTION_TYPE_END,
3032 		};
3033 	}
3034 	/* Build a new entry. */
3035 	mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3036 	if (!mcp_res) {
3037 		rte_errno = ENOMEM;
3038 		return NULL;
3039 	}
3040 	/*
3041 	 * The copy Flows are not included in any list. There
3042 	 * ones are referenced from other Flows and can not
3043 	 * be applied, removed, deleted in ardbitrary order
3044 	 * by list traversing.
3045 	 */
3046 	mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3047 					 actions, false, error);
3048 	if (!mcp_res->flow)
3049 		goto error;
3050 	mcp_res->refcnt++;
3051 	mcp_res->hlist_ent.key = mark_id;
3052 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3053 				&mcp_res->hlist_ent);
3054 	MLX5_ASSERT(!ret);
3055 	if (ret)
3056 		goto error;
3057 	return mcp_res;
3058 error:
3059 	if (mcp_res->flow)
3060 		flow_list_destroy(dev, NULL, mcp_res->flow);
3061 	rte_free(mcp_res);
3062 	return NULL;
3063 }
3064 
3065 /**
3066  * Release flow in RX_CP_TBL.
3067  *
3068  * @param dev
3069  *   Pointer to Ethernet device.
3070  * @flow
3071  *   Parent flow for wich copying is provided.
3072  */
3073 static void
3074 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3075 			  struct rte_flow *flow)
3076 {
3077 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3078 	struct mlx5_priv *priv = dev->data->dev_private;
3079 
3080 	if (!mcp_res || !priv->mreg_cp_tbl)
3081 		return;
3082 	if (flow->copy_applied) {
3083 		MLX5_ASSERT(mcp_res->appcnt);
3084 		flow->copy_applied = 0;
3085 		--mcp_res->appcnt;
3086 		if (!mcp_res->appcnt)
3087 			flow_drv_remove(dev, mcp_res->flow);
3088 	}
3089 	/*
3090 	 * We do not check availability of metadata registers here,
3091 	 * because copy resources are not allocated in this case.
3092 	 */
3093 	if (--mcp_res->refcnt)
3094 		return;
3095 	MLX5_ASSERT(mcp_res->flow);
3096 	flow_list_destroy(dev, NULL, mcp_res->flow);
3097 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3098 	rte_free(mcp_res);
3099 	flow->mreg_copy = NULL;
3100 }
3101 
3102 /**
3103  * Start flow in RX_CP_TBL.
3104  *
3105  * @param dev
3106  *   Pointer to Ethernet device.
3107  * @flow
3108  *   Parent flow for wich copying is provided.
3109  *
3110  * @return
3111  *   0 on success, a negative errno value otherwise and rte_errno is set.
3112  */
3113 static int
3114 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3115 			    struct rte_flow *flow)
3116 {
3117 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3118 	int ret;
3119 
3120 	if (!mcp_res || flow->copy_applied)
3121 		return 0;
3122 	if (!mcp_res->appcnt) {
3123 		ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3124 		if (ret)
3125 			return ret;
3126 	}
3127 	++mcp_res->appcnt;
3128 	flow->copy_applied = 1;
3129 	return 0;
3130 }
3131 
3132 /**
3133  * Stop flow in RX_CP_TBL.
3134  *
3135  * @param dev
3136  *   Pointer to Ethernet device.
3137  * @flow
3138  *   Parent flow for wich copying is provided.
3139  */
3140 static void
3141 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3142 			   struct rte_flow *flow)
3143 {
3144 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3145 
3146 	if (!mcp_res || !flow->copy_applied)
3147 		return;
3148 	MLX5_ASSERT(mcp_res->appcnt);
3149 	--mcp_res->appcnt;
3150 	flow->copy_applied = 0;
3151 	if (!mcp_res->appcnt)
3152 		flow_drv_remove(dev, mcp_res->flow);
3153 }
3154 
3155 /**
3156  * Remove the default copy action from RX_CP_TBL.
3157  *
3158  * @param dev
3159  *   Pointer to Ethernet device.
3160  */
3161 static void
3162 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3163 {
3164 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3165 	struct mlx5_priv *priv = dev->data->dev_private;
3166 
3167 	/* Check if default flow is registered. */
3168 	if (!priv->mreg_cp_tbl)
3169 		return;
3170 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3171 					    MLX5_DEFAULT_COPY_ID);
3172 	if (!mcp_res)
3173 		return;
3174 	MLX5_ASSERT(mcp_res->flow);
3175 	flow_list_destroy(dev, NULL, mcp_res->flow);
3176 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3177 	rte_free(mcp_res);
3178 }
3179 
3180 /**
3181  * Add the default copy action in in RX_CP_TBL.
3182  *
3183  * @param dev
3184  *   Pointer to Ethernet device.
3185  * @param[out] error
3186  *   Perform verbose error reporting if not NULL.
3187  *
3188  * @return
3189  *   0 for success, negative value otherwise and rte_errno is set.
3190  */
3191 static int
3192 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3193 				  struct rte_flow_error *error)
3194 {
3195 	struct mlx5_priv *priv = dev->data->dev_private;
3196 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3197 
3198 	/* Check whether extensive metadata feature is engaged. */
3199 	if (!priv->config.dv_flow_en ||
3200 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3201 	    !mlx5_flow_ext_mreg_supported(dev) ||
3202 	    !priv->sh->dv_regc0_mask)
3203 		return 0;
3204 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3205 	if (!mcp_res)
3206 		return -rte_errno;
3207 	return 0;
3208 }
3209 
3210 /**
3211  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3212  *
3213  * All the flow having Q/RSS action should be split by
3214  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3215  * performs the following,
3216  *   - CQE->flow_tag := reg_c[1] (MARK)
3217  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3218  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3219  * but there should be a flow per each MARK ID set by MARK action.
3220  *
3221  * For the aforementioned reason, if there's a MARK action in flow's action
3222  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3223  * the MARK ID to CQE's flow_tag like,
3224  *   - If reg_c[1] is mark_id,
3225  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3226  *
3227  * For SET_META action which stores value in reg_c[0], as the destination is
3228  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3229  * MARK ID means the default flow. The default flow looks like,
3230  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3231  *
3232  * @param dev
3233  *   Pointer to Ethernet device.
3234  * @param flow
3235  *   Pointer to flow structure.
3236  * @param[in] actions
3237  *   Pointer to the list of actions.
3238  * @param[out] error
3239  *   Perform verbose error reporting if not NULL.
3240  *
3241  * @return
3242  *   0 on success, negative value otherwise and rte_errno is set.
3243  */
3244 static int
3245 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3246 			    struct rte_flow *flow,
3247 			    const struct rte_flow_action *actions,
3248 			    struct rte_flow_error *error)
3249 {
3250 	struct mlx5_priv *priv = dev->data->dev_private;
3251 	struct mlx5_dev_config *config = &priv->config;
3252 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3253 	const struct rte_flow_action_mark *mark;
3254 
3255 	/* Check whether extensive metadata feature is engaged. */
3256 	if (!config->dv_flow_en ||
3257 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3258 	    !mlx5_flow_ext_mreg_supported(dev) ||
3259 	    !priv->sh->dv_regc0_mask)
3260 		return 0;
3261 	/* Find MARK action. */
3262 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3263 		switch (actions->type) {
3264 		case RTE_FLOW_ACTION_TYPE_FLAG:
3265 			mcp_res = flow_mreg_add_copy_action
3266 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3267 			if (!mcp_res)
3268 				return -rte_errno;
3269 			flow->mreg_copy = mcp_res;
3270 			if (dev->data->dev_started) {
3271 				mcp_res->appcnt++;
3272 				flow->copy_applied = 1;
3273 			}
3274 			return 0;
3275 		case RTE_FLOW_ACTION_TYPE_MARK:
3276 			mark = (const struct rte_flow_action_mark *)
3277 				actions->conf;
3278 			mcp_res =
3279 				flow_mreg_add_copy_action(dev, mark->id, error);
3280 			if (!mcp_res)
3281 				return -rte_errno;
3282 			flow->mreg_copy = mcp_res;
3283 			if (dev->data->dev_started) {
3284 				mcp_res->appcnt++;
3285 				flow->copy_applied = 1;
3286 			}
3287 			return 0;
3288 		default:
3289 			break;
3290 		}
3291 	}
3292 	return 0;
3293 }
3294 
3295 #define MLX5_MAX_SPLIT_ACTIONS 24
3296 #define MLX5_MAX_SPLIT_ITEMS 24
3297 
3298 /**
3299  * Split the hairpin flow.
3300  * Since HW can't support encap on Rx we move the encap to Tx.
3301  * If the count action is after the encap then we also
3302  * move the count action. in this case the count will also measure
3303  * the outer bytes.
3304  *
3305  * @param dev
3306  *   Pointer to Ethernet device.
3307  * @param[in] actions
3308  *   Associated actions (list terminated by the END action).
3309  * @param[out] actions_rx
3310  *   Rx flow actions.
3311  * @param[out] actions_tx
3312  *   Tx flow actions..
3313  * @param[out] pattern_tx
3314  *   The pattern items for the Tx flow.
3315  * @param[out] flow_id
3316  *   The flow ID connected to this flow.
3317  *
3318  * @return
3319  *   0 on success.
3320  */
3321 static int
3322 flow_hairpin_split(struct rte_eth_dev *dev,
3323 		   const struct rte_flow_action actions[],
3324 		   struct rte_flow_action actions_rx[],
3325 		   struct rte_flow_action actions_tx[],
3326 		   struct rte_flow_item pattern_tx[],
3327 		   uint32_t *flow_id)
3328 {
3329 	struct mlx5_priv *priv = dev->data->dev_private;
3330 	const struct rte_flow_action_raw_encap *raw_encap;
3331 	const struct rte_flow_action_raw_decap *raw_decap;
3332 	struct mlx5_rte_flow_action_set_tag *set_tag;
3333 	struct rte_flow_action *tag_action;
3334 	struct mlx5_rte_flow_item_tag *tag_item;
3335 	struct rte_flow_item *item;
3336 	char *addr;
3337 	int encap = 0;
3338 
3339 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3340 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3341 		switch (actions->type) {
3342 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3343 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3344 			rte_memcpy(actions_tx, actions,
3345 			       sizeof(struct rte_flow_action));
3346 			actions_tx++;
3347 			break;
3348 		case RTE_FLOW_ACTION_TYPE_COUNT:
3349 			if (encap) {
3350 				rte_memcpy(actions_tx, actions,
3351 					   sizeof(struct rte_flow_action));
3352 				actions_tx++;
3353 			} else {
3354 				rte_memcpy(actions_rx, actions,
3355 					   sizeof(struct rte_flow_action));
3356 				actions_rx++;
3357 			}
3358 			break;
3359 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3360 			raw_encap = actions->conf;
3361 			if (raw_encap->size >
3362 			    (sizeof(struct rte_flow_item_eth) +
3363 			     sizeof(struct rte_flow_item_ipv4))) {
3364 				memcpy(actions_tx, actions,
3365 				       sizeof(struct rte_flow_action));
3366 				actions_tx++;
3367 				encap = 1;
3368 			} else {
3369 				rte_memcpy(actions_rx, actions,
3370 					   sizeof(struct rte_flow_action));
3371 				actions_rx++;
3372 			}
3373 			break;
3374 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3375 			raw_decap = actions->conf;
3376 			if (raw_decap->size <
3377 			    (sizeof(struct rte_flow_item_eth) +
3378 			     sizeof(struct rte_flow_item_ipv4))) {
3379 				memcpy(actions_tx, actions,
3380 				       sizeof(struct rte_flow_action));
3381 				actions_tx++;
3382 			} else {
3383 				rte_memcpy(actions_rx, actions,
3384 					   sizeof(struct rte_flow_action));
3385 				actions_rx++;
3386 			}
3387 			break;
3388 		default:
3389 			rte_memcpy(actions_rx, actions,
3390 				   sizeof(struct rte_flow_action));
3391 			actions_rx++;
3392 			break;
3393 		}
3394 	}
3395 	/* Add set meta action and end action for the Rx flow. */
3396 	tag_action = actions_rx;
3397 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3398 	actions_rx++;
3399 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3400 	actions_rx++;
3401 	set_tag = (void *)actions_rx;
3402 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3403 	MLX5_ASSERT(set_tag->id > REG_NONE);
3404 	set_tag->data = *flow_id;
3405 	tag_action->conf = set_tag;
3406 	/* Create Tx item list. */
3407 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3408 	addr = (void *)&pattern_tx[2];
3409 	item = pattern_tx;
3410 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3411 	tag_item = (void *)addr;
3412 	tag_item->data = *flow_id;
3413 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3414 	MLX5_ASSERT(set_tag->id > REG_NONE);
3415 	item->spec = tag_item;
3416 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3417 	tag_item = (void *)addr;
3418 	tag_item->data = UINT32_MAX;
3419 	tag_item->id = UINT16_MAX;
3420 	item->mask = tag_item;
3421 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3422 	item->last = NULL;
3423 	item++;
3424 	item->type = RTE_FLOW_ITEM_TYPE_END;
3425 	return 0;
3426 }
3427 
3428 /**
3429  * The last stage of splitting chain, just creates the subflow
3430  * without any modification.
3431  *
3432  * @param[in] dev
3433  *   Pointer to Ethernet device.
3434  * @param[in] flow
3435  *   Parent flow structure pointer.
3436  * @param[in, out] sub_flow
3437  *   Pointer to return the created subflow, may be NULL.
3438  * @param[in] prefix_layers
3439  *   Prefix subflow layers, may be 0.
3440  * @param[in] attr
3441  *   Flow rule attributes.
3442  * @param[in] items
3443  *   Pattern specification (list terminated by the END pattern item).
3444  * @param[in] actions
3445  *   Associated actions (list terminated by the END action).
3446  * @param[in] external
3447  *   This flow rule is created by request external to PMD.
3448  * @param[out] error
3449  *   Perform verbose error reporting if not NULL.
3450  * @return
3451  *   0 on success, negative value otherwise
3452  */
3453 static int
3454 flow_create_split_inner(struct rte_eth_dev *dev,
3455 			struct rte_flow *flow,
3456 			struct mlx5_flow **sub_flow,
3457 			uint64_t prefix_layers,
3458 			const struct rte_flow_attr *attr,
3459 			const struct rte_flow_item items[],
3460 			const struct rte_flow_action actions[],
3461 			bool external, struct rte_flow_error *error)
3462 {
3463 	struct mlx5_flow *dev_flow;
3464 
3465 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions, error);
3466 	if (!dev_flow)
3467 		return -rte_errno;
3468 	dev_flow->flow = flow;
3469 	dev_flow->external = external;
3470 	/* Subflow object was created, we must include one in the list. */
3471 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3472 		      dev_flow->handle, next);
3473 	/*
3474 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3475 	 * flow may need some user defined item layer flags.
3476 	 */
3477 	if (prefix_layers)
3478 		dev_flow->handle->layers = prefix_layers;
3479 	if (sub_flow)
3480 		*sub_flow = dev_flow;
3481 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3482 }
3483 
3484 /**
3485  * Split the meter flow.
3486  *
3487  * As meter flow will split to three sub flow, other than meter
3488  * action, the other actions make sense to only meter accepts
3489  * the packet. If it need to be dropped, no other additional
3490  * actions should be take.
3491  *
3492  * One kind of special action which decapsulates the L3 tunnel
3493  * header will be in the prefix sub flow, as not to take the
3494  * L3 tunnel header into account.
3495  *
3496  * @param dev
3497  *   Pointer to Ethernet device.
3498  * @param[in] items
3499  *   Pattern specification (list terminated by the END pattern item).
3500  * @param[out] sfx_items
3501  *   Suffix flow match items (list terminated by the END pattern item).
3502  * @param[in] actions
3503  *   Associated actions (list terminated by the END action).
3504  * @param[out] actions_sfx
3505  *   Suffix flow actions.
3506  * @param[out] actions_pre
3507  *   Prefix flow actions.
3508  * @param[out] pattern_sfx
3509  *   The pattern items for the suffix flow.
3510  * @param[out] tag_sfx
3511  *   Pointer to suffix flow tag.
3512  *
3513  * @return
3514  *   0 on success.
3515  */
3516 static int
3517 flow_meter_split_prep(struct rte_eth_dev *dev,
3518 		 const struct rte_flow_item items[],
3519 		 struct rte_flow_item sfx_items[],
3520 		 const struct rte_flow_action actions[],
3521 		 struct rte_flow_action actions_sfx[],
3522 		 struct rte_flow_action actions_pre[])
3523 {
3524 	struct rte_flow_action *tag_action = NULL;
3525 	struct rte_flow_item *tag_item;
3526 	struct mlx5_rte_flow_action_set_tag *set_tag;
3527 	struct rte_flow_error error;
3528 	const struct rte_flow_action_raw_encap *raw_encap;
3529 	const struct rte_flow_action_raw_decap *raw_decap;
3530 	struct mlx5_rte_flow_item_tag *tag_spec;
3531 	struct mlx5_rte_flow_item_tag *tag_mask;
3532 	uint32_t tag_id;
3533 	bool copy_vlan = false;
3534 
3535 	/* Prepare the actions for prefix and suffix flow. */
3536 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3537 		struct rte_flow_action **action_cur = NULL;
3538 
3539 		switch (actions->type) {
3540 		case RTE_FLOW_ACTION_TYPE_METER:
3541 			/* Add the extra tag action first. */
3542 			tag_action = actions_pre;
3543 			tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3544 			actions_pre++;
3545 			action_cur = &actions_pre;
3546 			break;
3547 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3548 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3549 			action_cur = &actions_pre;
3550 			break;
3551 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3552 			raw_encap = actions->conf;
3553 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3554 				action_cur = &actions_pre;
3555 			break;
3556 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3557 			raw_decap = actions->conf;
3558 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3559 				action_cur = &actions_pre;
3560 			break;
3561 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3562 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3563 			copy_vlan = true;
3564 			break;
3565 		default:
3566 			break;
3567 		}
3568 		if (!action_cur)
3569 			action_cur = &actions_sfx;
3570 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3571 		(*action_cur)++;
3572 	}
3573 	/* Add end action to the actions. */
3574 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3575 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3576 	actions_pre++;
3577 	/* Set the tag. */
3578 	set_tag = (void *)actions_pre;
3579 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3580 	/*
3581 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3582 	 */
3583 	tag_id = flow_qrss_get_id(dev);
3584 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3585 	assert(tag_action);
3586 	tag_action->conf = set_tag;
3587 	/* Prepare the suffix subflow items. */
3588 	tag_item = sfx_items++;
3589 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3590 		int item_type = items->type;
3591 
3592 		switch (item_type) {
3593 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3594 			memcpy(sfx_items, items, sizeof(*sfx_items));
3595 			sfx_items++;
3596 			break;
3597 		case RTE_FLOW_ITEM_TYPE_VLAN:
3598 			if (copy_vlan) {
3599 				memcpy(sfx_items, items, sizeof(*sfx_items));
3600 				/*
3601 				 * Convert to internal match item, it is used
3602 				 * for vlan push and set vid.
3603 				 */
3604 				sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3605 				sfx_items++;
3606 			}
3607 			break;
3608 		default:
3609 			break;
3610 		}
3611 	}
3612 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3613 	sfx_items++;
3614 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3615 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3616 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3617 	tag_mask = tag_spec + 1;
3618 	tag_mask->data = 0xffffff00;
3619 	tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3620 	tag_item->spec = tag_spec;
3621 	tag_item->last = NULL;
3622 	tag_item->mask = tag_mask;
3623 	return tag_id;
3624 }
3625 
3626 /**
3627  * Split action list having QUEUE/RSS for metadata register copy.
3628  *
3629  * Once Q/RSS action is detected in user's action list, the flow action
3630  * should be split in order to copy metadata registers, which will happen in
3631  * RX_CP_TBL like,
3632  *   - CQE->flow_tag := reg_c[1] (MARK)
3633  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3634  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3635  * This is because the last action of each flow must be a terminal action
3636  * (QUEUE, RSS or DROP).
3637  *
3638  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3639  * stored and kept in the mlx5_flow structure per each sub_flow.
3640  *
3641  * The Q/RSS action is replaced with,
3642  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3643  * And the following JUMP action is added at the end,
3644  *   - JUMP, to RX_CP_TBL.
3645  *
3646  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3647  * flow_create_split_metadata() routine. The flow will look like,
3648  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3649  *
3650  * @param dev
3651  *   Pointer to Ethernet device.
3652  * @param[out] split_actions
3653  *   Pointer to store split actions to jump to CP_TBL.
3654  * @param[in] actions
3655  *   Pointer to the list of original flow actions.
3656  * @param[in] qrss
3657  *   Pointer to the Q/RSS action.
3658  * @param[in] actions_n
3659  *   Number of original actions.
3660  * @param[out] error
3661  *   Perform verbose error reporting if not NULL.
3662  *
3663  * @return
3664  *   non-zero unique flow_id on success, otherwise 0 and
3665  *   error/rte_error are set.
3666  */
3667 static uint32_t
3668 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3669 			  struct rte_flow_action *split_actions,
3670 			  const struct rte_flow_action *actions,
3671 			  const struct rte_flow_action *qrss,
3672 			  int actions_n, struct rte_flow_error *error)
3673 {
3674 	struct mlx5_rte_flow_action_set_tag *set_tag;
3675 	struct rte_flow_action_jump *jump;
3676 	const int qrss_idx = qrss - actions;
3677 	uint32_t flow_id = 0;
3678 	int ret = 0;
3679 
3680 	/*
3681 	 * Given actions will be split
3682 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3683 	 * - Add jump to mreg CP_TBL.
3684 	 * As a result, there will be one more action.
3685 	 */
3686 	++actions_n;
3687 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3688 	set_tag = (void *)(split_actions + actions_n);
3689 	/*
3690 	 * If tag action is not set to void(it means we are not the meter
3691 	 * suffix flow), add the tag action. Since meter suffix flow already
3692 	 * has the tag added.
3693 	 */
3694 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3695 		/*
3696 		 * Allocate the new subflow ID. This one is unique within
3697 		 * device and not shared with representors. Otherwise,
3698 		 * we would have to resolve multi-thread access synch
3699 		 * issue. Each flow on the shared device is appended
3700 		 * with source vport identifier, so the resulting
3701 		 * flows will be unique in the shared (by master and
3702 		 * representors) domain even if they have coinciding
3703 		 * IDs.
3704 		 */
3705 		flow_id = flow_qrss_get_id(dev);
3706 		if (!flow_id)
3707 			return rte_flow_error_set(error, ENOMEM,
3708 						  RTE_FLOW_ERROR_TYPE_ACTION,
3709 						  NULL, "can't allocate id "
3710 						  "for split Q/RSS subflow");
3711 		/* Internal SET_TAG action to set flow ID. */
3712 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3713 			.data = flow_id,
3714 		};
3715 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3716 		if (ret < 0)
3717 			return ret;
3718 		set_tag->id = ret;
3719 		/* Construct new actions array. */
3720 		/* Replace QUEUE/RSS action. */
3721 		split_actions[qrss_idx] = (struct rte_flow_action){
3722 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3723 			.conf = set_tag,
3724 		};
3725 	}
3726 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3727 	jump = (void *)(set_tag + 1);
3728 	*jump = (struct rte_flow_action_jump){
3729 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3730 	};
3731 	split_actions[actions_n - 2] = (struct rte_flow_action){
3732 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3733 		.conf = jump,
3734 	};
3735 	split_actions[actions_n - 1] = (struct rte_flow_action){
3736 		.type = RTE_FLOW_ACTION_TYPE_END,
3737 	};
3738 	return flow_id;
3739 }
3740 
3741 /**
3742  * Extend the given action list for Tx metadata copy.
3743  *
3744  * Copy the given action list to the ext_actions and add flow metadata register
3745  * copy action in order to copy reg_a set by WQE to reg_c[0].
3746  *
3747  * @param[out] ext_actions
3748  *   Pointer to the extended action list.
3749  * @param[in] actions
3750  *   Pointer to the list of actions.
3751  * @param[in] actions_n
3752  *   Number of actions in the list.
3753  * @param[out] error
3754  *   Perform verbose error reporting if not NULL.
3755  * @param[in] encap_idx
3756  *   The encap action inndex.
3757  *
3758  * @return
3759  *   0 on success, negative value otherwise
3760  */
3761 static int
3762 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3763 		       struct rte_flow_action *ext_actions,
3764 		       const struct rte_flow_action *actions,
3765 		       int actions_n, struct rte_flow_error *error,
3766 		       int encap_idx)
3767 {
3768 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3769 		(struct mlx5_flow_action_copy_mreg *)
3770 			(ext_actions + actions_n + 1);
3771 	int ret;
3772 
3773 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3774 	if (ret < 0)
3775 		return ret;
3776 	cp_mreg->dst = ret;
3777 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3778 	if (ret < 0)
3779 		return ret;
3780 	cp_mreg->src = ret;
3781 	if (encap_idx != 0)
3782 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3783 	if (encap_idx == actions_n - 1) {
3784 		ext_actions[actions_n - 1] = (struct rte_flow_action){
3785 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3786 			.conf = cp_mreg,
3787 		};
3788 		ext_actions[actions_n] = (struct rte_flow_action){
3789 			.type = RTE_FLOW_ACTION_TYPE_END,
3790 		};
3791 	} else {
3792 		ext_actions[encap_idx] = (struct rte_flow_action){
3793 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3794 			.conf = cp_mreg,
3795 		};
3796 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3797 				sizeof(*ext_actions) * (actions_n - encap_idx));
3798 	}
3799 	return 0;
3800 }
3801 
3802 /**
3803  * The splitting for metadata feature.
3804  *
3805  * - Q/RSS action on NIC Rx should be split in order to pass by
3806  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3807  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3808  *
3809  * - All the actions on NIC Tx should have a mreg copy action to
3810  *   copy reg_a from WQE to reg_c[0].
3811  *
3812  * @param dev
3813  *   Pointer to Ethernet device.
3814  * @param[in] flow
3815  *   Parent flow structure pointer.
3816  * @param[in] prefix_layers
3817  *   Prefix flow layer flags.
3818  * @param[in] attr
3819  *   Flow rule attributes.
3820  * @param[in] items
3821  *   Pattern specification (list terminated by the END pattern item).
3822  * @param[in] actions
3823  *   Associated actions (list terminated by the END action).
3824  * @param[in] external
3825  *   This flow rule is created by request external to PMD.
3826  * @param[out] error
3827  *   Perform verbose error reporting if not NULL.
3828  * @return
3829  *   0 on success, negative value otherwise
3830  */
3831 static int
3832 flow_create_split_metadata(struct rte_eth_dev *dev,
3833 			   struct rte_flow *flow,
3834 			   uint64_t prefix_layers,
3835 			   const struct rte_flow_attr *attr,
3836 			   const struct rte_flow_item items[],
3837 			   const struct rte_flow_action actions[],
3838 			   bool external, struct rte_flow_error *error)
3839 {
3840 	struct mlx5_priv *priv = dev->data->dev_private;
3841 	struct mlx5_dev_config *config = &priv->config;
3842 	const struct rte_flow_action *qrss = NULL;
3843 	struct rte_flow_action *ext_actions = NULL;
3844 	struct mlx5_flow *dev_flow = NULL;
3845 	uint32_t qrss_id = 0;
3846 	int mtr_sfx = 0;
3847 	size_t act_size;
3848 	int actions_n;
3849 	int encap_idx;
3850 	int ret;
3851 
3852 	/* Check whether extensive metadata feature is engaged. */
3853 	if (!config->dv_flow_en ||
3854 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3855 	    !mlx5_flow_ext_mreg_supported(dev))
3856 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3857 					       attr, items, actions, external,
3858 					       error);
3859 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3860 							   &encap_idx);
3861 	if (qrss) {
3862 		/* Exclude hairpin flows from splitting. */
3863 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3864 			const struct rte_flow_action_queue *queue;
3865 
3866 			queue = qrss->conf;
3867 			if (mlx5_rxq_get_type(dev, queue->index) ==
3868 			    MLX5_RXQ_TYPE_HAIRPIN)
3869 				qrss = NULL;
3870 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3871 			const struct rte_flow_action_rss *rss;
3872 
3873 			rss = qrss->conf;
3874 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3875 			    MLX5_RXQ_TYPE_HAIRPIN)
3876 				qrss = NULL;
3877 		}
3878 	}
3879 	if (qrss) {
3880 		/* Check if it is in meter suffix table. */
3881 		mtr_sfx = attr->group == (attr->transfer ?
3882 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3883 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3884 		/*
3885 		 * Q/RSS action on NIC Rx should be split in order to pass by
3886 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3887 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3888 		 */
3889 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3890 			   sizeof(struct rte_flow_action_set_tag) +
3891 			   sizeof(struct rte_flow_action_jump);
3892 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3893 		if (!ext_actions)
3894 			return rte_flow_error_set(error, ENOMEM,
3895 						  RTE_FLOW_ERROR_TYPE_ACTION,
3896 						  NULL, "no memory to split "
3897 						  "metadata flow");
3898 		/*
3899 		 * If we are the suffix flow of meter, tag already exist.
3900 		 * Set the tag action to void.
3901 		 */
3902 		if (mtr_sfx)
3903 			ext_actions[qrss - actions].type =
3904 						RTE_FLOW_ACTION_TYPE_VOID;
3905 		else
3906 			ext_actions[qrss - actions].type =
3907 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3908 		/*
3909 		 * Create the new actions list with removed Q/RSS action
3910 		 * and appended set tag and jump to register copy table
3911 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3912 		 * in advance, because it is needed for set tag action.
3913 		 */
3914 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3915 						    qrss, actions_n, error);
3916 		if (!mtr_sfx && !qrss_id) {
3917 			ret = -rte_errno;
3918 			goto exit;
3919 		}
3920 	} else if (attr->egress && !attr->transfer) {
3921 		/*
3922 		 * All the actions on NIC Tx should have a metadata register
3923 		 * copy action to copy reg_a from WQE to reg_c[meta]
3924 		 */
3925 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3926 			   sizeof(struct mlx5_flow_action_copy_mreg);
3927 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3928 		if (!ext_actions)
3929 			return rte_flow_error_set(error, ENOMEM,
3930 						  RTE_FLOW_ERROR_TYPE_ACTION,
3931 						  NULL, "no memory to split "
3932 						  "metadata flow");
3933 		/* Create the action list appended with copy register. */
3934 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3935 					     actions_n, error, encap_idx);
3936 		if (ret < 0)
3937 			goto exit;
3938 	}
3939 	/* Add the unmodified original or prefix subflow. */
3940 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
3941 				      items, ext_actions ? ext_actions :
3942 				      actions, external, error);
3943 	if (ret < 0)
3944 		goto exit;
3945 	MLX5_ASSERT(dev_flow);
3946 	if (qrss) {
3947 		const struct rte_flow_attr q_attr = {
3948 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3949 			.ingress = 1,
3950 		};
3951 		/* Internal PMD action to set register. */
3952 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3953 			.data = qrss_id,
3954 			.id = 0,
3955 		};
3956 		struct rte_flow_item q_items[] = {
3957 			{
3958 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3959 				.spec = &q_tag_spec,
3960 				.last = NULL,
3961 				.mask = NULL,
3962 			},
3963 			{
3964 				.type = RTE_FLOW_ITEM_TYPE_END,
3965 			},
3966 		};
3967 		struct rte_flow_action q_actions[] = {
3968 			{
3969 				.type = qrss->type,
3970 				.conf = qrss->conf,
3971 			},
3972 			{
3973 				.type = RTE_FLOW_ACTION_TYPE_END,
3974 			},
3975 		};
3976 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
3977 
3978 		/*
3979 		 * Configure the tag item only if there is no meter subflow.
3980 		 * Since tag is already marked in the meter suffix subflow
3981 		 * we can just use the meter suffix items as is.
3982 		 */
3983 		if (qrss_id) {
3984 			/* Not meter subflow. */
3985 			MLX5_ASSERT(!mtr_sfx);
3986 			/*
3987 			 * Put unique id in prefix flow due to it is destroyed
3988 			 * after suffix flow and id will be freed after there
3989 			 * is no actual flows with this id and identifier
3990 			 * reallocation becomes possible (for example, for
3991 			 * other flows in other threads).
3992 			 */
3993 			dev_flow->handle->qrss_id = qrss_id;
3994 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3995 						   error);
3996 			if (ret < 0)
3997 				goto exit;
3998 			q_tag_spec.id = ret;
3999 		}
4000 		dev_flow = NULL;
4001 		/* Add suffix subflow to execute Q/RSS. */
4002 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4003 					      &q_attr, mtr_sfx ? items :
4004 					      q_items, q_actions,
4005 					      external, error);
4006 		if (ret < 0)
4007 			goto exit;
4008 		/* qrss ID should be freed if failed. */
4009 		qrss_id = 0;
4010 		MLX5_ASSERT(dev_flow);
4011 	}
4012 
4013 exit:
4014 	/*
4015 	 * We do not destroy the partially created sub_flows in case of error.
4016 	 * These ones are included into parent flow list and will be destroyed
4017 	 * by flow_drv_destroy.
4018 	 */
4019 	flow_qrss_free_id(dev, qrss_id);
4020 	rte_free(ext_actions);
4021 	return ret;
4022 }
4023 
4024 /**
4025  * The splitting for meter feature.
4026  *
4027  * - The meter flow will be split to two flows as prefix and
4028  *   suffix flow. The packets make sense only it pass the prefix
4029  *   meter action.
4030  *
4031  * - Reg_C_5 is used for the packet to match betweend prefix and
4032  *   suffix flow.
4033  *
4034  * @param dev
4035  *   Pointer to Ethernet device.
4036  * @param[in] flow
4037  *   Parent flow structure pointer.
4038  * @param[in] attr
4039  *   Flow rule attributes.
4040  * @param[in] items
4041  *   Pattern specification (list terminated by the END pattern item).
4042  * @param[in] actions
4043  *   Associated actions (list terminated by the END action).
4044  * @param[in] external
4045  *   This flow rule is created by request external to PMD.
4046  * @param[out] error
4047  *   Perform verbose error reporting if not NULL.
4048  * @return
4049  *   0 on success, negative value otherwise
4050  */
4051 static int
4052 flow_create_split_meter(struct rte_eth_dev *dev,
4053 			   struct rte_flow *flow,
4054 			   const struct rte_flow_attr *attr,
4055 			   const struct rte_flow_item items[],
4056 			   const struct rte_flow_action actions[],
4057 			   bool external, struct rte_flow_error *error)
4058 {
4059 	struct mlx5_priv *priv = dev->data->dev_private;
4060 	struct rte_flow_action *sfx_actions = NULL;
4061 	struct rte_flow_action *pre_actions = NULL;
4062 	struct rte_flow_item *sfx_items = NULL;
4063 	struct mlx5_flow *dev_flow = NULL;
4064 	struct rte_flow_attr sfx_attr = *attr;
4065 	uint32_t mtr = 0;
4066 	uint32_t mtr_tag_id = 0;
4067 	size_t act_size;
4068 	size_t item_size;
4069 	int actions_n = 0;
4070 	int ret;
4071 
4072 	if (priv->mtr_en)
4073 		actions_n = flow_check_meter_action(actions, &mtr);
4074 	if (mtr) {
4075 		/* The five prefix actions: meter, decap, encap, tag, end. */
4076 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4077 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4078 		/* tag, vlan, port id, end. */
4079 #define METER_SUFFIX_ITEM 4
4080 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4081 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4082 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4083 		if (!sfx_actions)
4084 			return rte_flow_error_set(error, ENOMEM,
4085 						  RTE_FLOW_ERROR_TYPE_ACTION,
4086 						  NULL, "no memory to split "
4087 						  "meter flow");
4088 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4089 			     act_size);
4090 		pre_actions = sfx_actions + actions_n;
4091 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4092 						   actions, sfx_actions,
4093 						   pre_actions);
4094 		if (!mtr_tag_id) {
4095 			ret = -rte_errno;
4096 			goto exit;
4097 		}
4098 		/* Add the prefix subflow. */
4099 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4100 					      items, pre_actions, external,
4101 					      error);
4102 		if (ret) {
4103 			ret = -rte_errno;
4104 			goto exit;
4105 		}
4106 		dev_flow->handle->mtr_flow_id = mtr_tag_id;
4107 		/* Setting the sfx group atrr. */
4108 		sfx_attr.group = sfx_attr.transfer ?
4109 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4110 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4111 	}
4112 	/* Add the prefix subflow. */
4113 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4114 					 flow_get_prefix_layer_flags(dev_flow) :
4115 					 0, &sfx_attr,
4116 					 sfx_items ? sfx_items : items,
4117 					 sfx_actions ? sfx_actions : actions,
4118 					 external, error);
4119 exit:
4120 	if (sfx_actions)
4121 		rte_free(sfx_actions);
4122 	return ret;
4123 }
4124 
4125 /**
4126  * Split the flow to subflow set. The splitters might be linked
4127  * in the chain, like this:
4128  * flow_create_split_outer() calls:
4129  *   flow_create_split_meter() calls:
4130  *     flow_create_split_metadata(meter_subflow_0) calls:
4131  *       flow_create_split_inner(metadata_subflow_0)
4132  *       flow_create_split_inner(metadata_subflow_1)
4133  *       flow_create_split_inner(metadata_subflow_2)
4134  *     flow_create_split_metadata(meter_subflow_1) calls:
4135  *       flow_create_split_inner(metadata_subflow_0)
4136  *       flow_create_split_inner(metadata_subflow_1)
4137  *       flow_create_split_inner(metadata_subflow_2)
4138  *
4139  * This provide flexible way to add new levels of flow splitting.
4140  * The all of successfully created subflows are included to the
4141  * parent flow dev_flow list.
4142  *
4143  * @param dev
4144  *   Pointer to Ethernet device.
4145  * @param[in] flow
4146  *   Parent flow structure pointer.
4147  * @param[in] attr
4148  *   Flow rule attributes.
4149  * @param[in] items
4150  *   Pattern specification (list terminated by the END pattern item).
4151  * @param[in] actions
4152  *   Associated actions (list terminated by the END action).
4153  * @param[in] external
4154  *   This flow rule is created by request external to PMD.
4155  * @param[out] error
4156  *   Perform verbose error reporting if not NULL.
4157  * @return
4158  *   0 on success, negative value otherwise
4159  */
4160 static int
4161 flow_create_split_outer(struct rte_eth_dev *dev,
4162 			struct rte_flow *flow,
4163 			const struct rte_flow_attr *attr,
4164 			const struct rte_flow_item items[],
4165 			const struct rte_flow_action actions[],
4166 			bool external, struct rte_flow_error *error)
4167 {
4168 	int ret;
4169 
4170 	ret = flow_create_split_meter(dev, flow, attr, items,
4171 					 actions, external, error);
4172 	MLX5_ASSERT(ret <= 0);
4173 	return ret;
4174 }
4175 
4176 /**
4177  * Create a flow and add it to @p list.
4178  *
4179  * @param dev
4180  *   Pointer to Ethernet device.
4181  * @param list
4182  *   Pointer to a TAILQ flow list. If this parameter NULL,
4183  *   no list insertion occurred, flow is just created,
4184  *   this is caller's responsibility to track the
4185  *   created flow.
4186  * @param[in] attr
4187  *   Flow rule attributes.
4188  * @param[in] items
4189  *   Pattern specification (list terminated by the END pattern item).
4190  * @param[in] actions
4191  *   Associated actions (list terminated by the END action).
4192  * @param[in] external
4193  *   This flow rule is created by request external to PMD.
4194  * @param[out] error
4195  *   Perform verbose error reporting if not NULL.
4196  *
4197  * @return
4198  *   A flow on success, NULL otherwise and rte_errno is set.
4199  */
4200 static struct rte_flow *
4201 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4202 		 const struct rte_flow_attr *attr,
4203 		 const struct rte_flow_item items[],
4204 		 const struct rte_flow_action actions[],
4205 		 bool external, struct rte_flow_error *error)
4206 {
4207 	struct mlx5_priv *priv = dev->data->dev_private;
4208 	struct rte_flow *flow = NULL;
4209 	struct mlx5_flow *dev_flow;
4210 	const struct rte_flow_action_rss *rss;
4211 	union {
4212 		struct rte_flow_expand_rss buf;
4213 		uint8_t buffer[2048];
4214 	} expand_buffer;
4215 	union {
4216 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4217 		uint8_t buffer[2048];
4218 	} actions_rx;
4219 	union {
4220 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4221 		uint8_t buffer[2048];
4222 	} actions_hairpin_tx;
4223 	union {
4224 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4225 		uint8_t buffer[2048];
4226 	} items_tx;
4227 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4228 	const struct rte_flow_action *p_actions_rx = actions;
4229 	uint32_t i;
4230 	uint32_t flow_size;
4231 	int hairpin_flow = 0;
4232 	uint32_t hairpin_id = 0;
4233 	struct rte_flow_attr attr_tx = { .priority = 0 };
4234 	int ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4235 				    error);
4236 
4237 	if (ret < 0)
4238 		return NULL;
4239 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4240 	if (hairpin_flow > 0) {
4241 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4242 			rte_errno = EINVAL;
4243 			return NULL;
4244 		}
4245 		flow_hairpin_split(dev, actions, actions_rx.actions,
4246 				   actions_hairpin_tx.actions, items_tx.items,
4247 				   &hairpin_id);
4248 		p_actions_rx = actions_rx.actions;
4249 	}
4250 	flow_size = sizeof(struct rte_flow);
4251 	rss = flow_get_rss_action(p_actions_rx);
4252 	if (rss)
4253 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4254 					    sizeof(void *));
4255 	else
4256 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4257 	flow = rte_calloc(__func__, 1, flow_size, 0);
4258 	if (!flow) {
4259 		rte_errno = ENOMEM;
4260 		goto error_before_flow;
4261 	}
4262 	flow->drv_type = flow_get_drv_type(dev, attr);
4263 	if (hairpin_id != 0)
4264 		flow->hairpin_flow_id = hairpin_id;
4265 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4266 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4267 	flow->rss.queue = (void *)(flow + 1);
4268 	if (rss) {
4269 		/*
4270 		 * The following information is required by
4271 		 * mlx5_flow_hashfields_adjust() in advance.
4272 		 */
4273 		flow->rss.level = rss->level;
4274 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4275 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4276 	}
4277 	flow->dev_handles = 0;
4278 	if (rss && rss->types) {
4279 		unsigned int graph_root;
4280 
4281 		graph_root = find_graph_root(items, rss->level);
4282 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4283 					  items, rss->types,
4284 					  mlx5_support_expansion,
4285 					  graph_root);
4286 		MLX5_ASSERT(ret > 0 &&
4287 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4288 	} else {
4289 		buf->entries = 1;
4290 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4291 	}
4292 	/*
4293 	 * Record the start index when there is a nested call. All sub-flows
4294 	 * need to be translated before another calling.
4295 	 * No need to use ping-pong buffer to save memory here.
4296 	 */
4297 	if (priv->flow_idx) {
4298 		MLX5_ASSERT(!priv->flow_nested_idx);
4299 		priv->flow_nested_idx = priv->flow_idx;
4300 	}
4301 	for (i = 0; i < buf->entries; ++i) {
4302 		/*
4303 		 * The splitter may create multiple dev_flows,
4304 		 * depending on configuration. In the simplest
4305 		 * case it just creates unmodified original flow.
4306 		 */
4307 		ret = flow_create_split_outer(dev, flow, attr,
4308 					      buf->entry[i].pattern,
4309 					      p_actions_rx, external,
4310 					      error);
4311 		if (ret < 0)
4312 			goto error;
4313 	}
4314 	/* Create the tx flow. */
4315 	if (hairpin_flow) {
4316 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4317 		attr_tx.ingress = 0;
4318 		attr_tx.egress = 1;
4319 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4320 					    actions_hairpin_tx.actions, error);
4321 		if (!dev_flow)
4322 			goto error;
4323 		dev_flow->flow = flow;
4324 		dev_flow->external = 0;
4325 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4326 			      dev_flow->handle, next);
4327 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4328 					 items_tx.items,
4329 					 actions_hairpin_tx.actions, error);
4330 		if (ret < 0)
4331 			goto error;
4332 	}
4333 	/*
4334 	 * Update the metadata register copy table. If extensive
4335 	 * metadata feature is enabled and registers are supported
4336 	 * we might create the extra rte_flow for each unique
4337 	 * MARK/FLAG action ID.
4338 	 *
4339 	 * The table is updated for ingress Flows only, because
4340 	 * the egress Flows belong to the different device and
4341 	 * copy table should be updated in peer NIC Rx domain.
4342 	 */
4343 	if (attr->ingress &&
4344 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4345 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4346 		if (ret)
4347 			goto error;
4348 	}
4349 	/*
4350 	 * If the flow is external (from application) OR device is started, then
4351 	 * the flow will be applied immediately.
4352 	 */
4353 	if (external || dev->data->dev_started) {
4354 		ret = flow_drv_apply(dev, flow, error);
4355 		if (ret < 0)
4356 			goto error;
4357 	}
4358 	if (list)
4359 		TAILQ_INSERT_TAIL(list, flow, next);
4360 	flow_rxq_flags_set(dev, flow);
4361 	/* Nested flow creation index recovery. */
4362 	priv->flow_idx = priv->flow_nested_idx;
4363 	if (priv->flow_nested_idx)
4364 		priv->flow_nested_idx = 0;
4365 	return flow;
4366 error:
4367 	MLX5_ASSERT(flow);
4368 	ret = rte_errno; /* Save rte_errno before cleanup. */
4369 	flow_mreg_del_copy_action(dev, flow);
4370 	flow_drv_destroy(dev, flow);
4371 	rte_free(flow);
4372 	rte_errno = ret; /* Restore rte_errno. */
4373 error_before_flow:
4374 	ret = rte_errno;
4375 	if (hairpin_id)
4376 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4377 				     hairpin_id);
4378 	rte_errno = ret;
4379 	priv->flow_idx = priv->flow_nested_idx;
4380 	if (priv->flow_nested_idx)
4381 		priv->flow_nested_idx = 0;
4382 	return NULL;
4383 }
4384 
4385 /**
4386  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4387  * incoming packets to table 1.
4388  *
4389  * Other flow rules, requested for group n, will be created in
4390  * e-switch table n+1.
4391  * Jump action to e-switch group n will be created to group n+1.
4392  *
4393  * Used when working in switchdev mode, to utilise advantages of table 1
4394  * and above.
4395  *
4396  * @param dev
4397  *   Pointer to Ethernet device.
4398  *
4399  * @return
4400  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4401  */
4402 struct rte_flow *
4403 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4404 {
4405 	const struct rte_flow_attr attr = {
4406 		.group = 0,
4407 		.priority = 0,
4408 		.ingress = 1,
4409 		.egress = 0,
4410 		.transfer = 1,
4411 	};
4412 	const struct rte_flow_item pattern = {
4413 		.type = RTE_FLOW_ITEM_TYPE_END,
4414 	};
4415 	struct rte_flow_action_jump jump = {
4416 		.group = 1,
4417 	};
4418 	const struct rte_flow_action actions[] = {
4419 		{
4420 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4421 			.conf = &jump,
4422 		},
4423 		{
4424 			.type = RTE_FLOW_ACTION_TYPE_END,
4425 		},
4426 	};
4427 	struct mlx5_priv *priv = dev->data->dev_private;
4428 	struct rte_flow_error error;
4429 
4430 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4431 				actions, false, &error);
4432 }
4433 
4434 /**
4435  * Create a flow.
4436  *
4437  * @see rte_flow_create()
4438  * @see rte_flow_ops
4439  */
4440 struct rte_flow *
4441 mlx5_flow_create(struct rte_eth_dev *dev,
4442 		 const struct rte_flow_attr *attr,
4443 		 const struct rte_flow_item items[],
4444 		 const struct rte_flow_action actions[],
4445 		 struct rte_flow_error *error)
4446 {
4447 	struct mlx5_priv *priv = dev->data->dev_private;
4448 
4449 	/*
4450 	 * If the device is not started yet, it is not allowed to created a
4451 	 * flow from application. PMD default flows and traffic control flows
4452 	 * are not affected.
4453 	 */
4454 	if (unlikely(!dev->data->dev_started)) {
4455 		rte_errno = ENODEV;
4456 		DRV_LOG(DEBUG, "port %u is not started when "
4457 			"inserting a flow", dev->data->port_id);
4458 		return NULL;
4459 	}
4460 	return flow_list_create(dev, &priv->flows,
4461 				attr, items, actions, true, error);
4462 }
4463 
4464 /**
4465  * Destroy a flow in a list.
4466  *
4467  * @param dev
4468  *   Pointer to Ethernet device.
4469  * @param list
4470  *   Pointer to a TAILQ flow list. If this parameter NULL,
4471  *   there is no flow removal from the list.
4472  * @param[in] flow
4473  *   Flow to destroy.
4474  */
4475 static void
4476 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4477 		  struct rte_flow *flow)
4478 {
4479 	struct mlx5_priv *priv = dev->data->dev_private;
4480 
4481 	/*
4482 	 * Update RX queue flags only if port is started, otherwise it is
4483 	 * already clean.
4484 	 */
4485 	if (dev->data->dev_started)
4486 		flow_rxq_flags_trim(dev, flow);
4487 	if (flow->hairpin_flow_id)
4488 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4489 				     flow->hairpin_flow_id);
4490 	flow_drv_destroy(dev, flow);
4491 	if (list)
4492 		TAILQ_REMOVE(list, flow, next);
4493 	flow_mreg_del_copy_action(dev, flow);
4494 	rte_free(flow->fdir);
4495 	rte_free(flow);
4496 }
4497 
4498 /**
4499  * Destroy all flows.
4500  *
4501  * @param dev
4502  *   Pointer to Ethernet device.
4503  * @param list
4504  *   Pointer to a TAILQ flow list.
4505  * @param active
4506  *   If flushing is called avtively.
4507  */
4508 void
4509 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list,
4510 		     bool active)
4511 {
4512 	uint32_t num_flushed = 0;
4513 
4514 	while (!TAILQ_EMPTY(list)) {
4515 		struct rte_flow *flow;
4516 
4517 		flow = TAILQ_FIRST(list);
4518 		flow_list_destroy(dev, list, flow);
4519 		num_flushed++;
4520 	}
4521 	if (active) {
4522 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4523 			dev->data->port_id, num_flushed);
4524 	}
4525 }
4526 
4527 /**
4528  * Remove all flows.
4529  *
4530  * @param dev
4531  *   Pointer to Ethernet device.
4532  * @param list
4533  *   Pointer to a TAILQ flow list.
4534  */
4535 void
4536 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4537 {
4538 	struct rte_flow *flow;
4539 
4540 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4541 		flow_drv_remove(dev, flow);
4542 		flow_mreg_stop_copy_action(dev, flow);
4543 	}
4544 	flow_mreg_del_default_copy_action(dev);
4545 	flow_rxq_flags_clear(dev);
4546 }
4547 
4548 /**
4549  * Add all flows.
4550  *
4551  * @param dev
4552  *   Pointer to Ethernet device.
4553  * @param list
4554  *   Pointer to a TAILQ flow list.
4555  *
4556  * @return
4557  *   0 on success, a negative errno value otherwise and rte_errno is set.
4558  */
4559 int
4560 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4561 {
4562 	struct rte_flow *flow;
4563 	struct rte_flow_error error;
4564 	int ret = 0;
4565 
4566 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4567 	ret = flow_mreg_add_default_copy_action(dev, &error);
4568 	if (ret < 0)
4569 		return -rte_errno;
4570 	/* Apply Flows created by application. */
4571 	TAILQ_FOREACH(flow, list, next) {
4572 		ret = flow_mreg_start_copy_action(dev, flow);
4573 		if (ret < 0)
4574 			goto error;
4575 		ret = flow_drv_apply(dev, flow, &error);
4576 		if (ret < 0)
4577 			goto error;
4578 		flow_rxq_flags_set(dev, flow);
4579 	}
4580 	return 0;
4581 error:
4582 	ret = rte_errno; /* Save rte_errno before cleanup. */
4583 	mlx5_flow_stop(dev, list);
4584 	rte_errno = ret; /* Restore rte_errno. */
4585 	return -rte_errno;
4586 }
4587 
4588 /**
4589  * Stop all default actions for flows.
4590  *
4591  * @param dev
4592  *   Pointer to Ethernet device.
4593  */
4594 void
4595 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4596 {
4597 	flow_mreg_del_default_copy_action(dev);
4598 }
4599 
4600 /**
4601  * Start all default actions for flows.
4602  *
4603  * @param dev
4604  *   Pointer to Ethernet device.
4605  * @return
4606  *   0 on success, a negative errno value otherwise and rte_errno is set.
4607  */
4608 int
4609 mlx5_flow_start_default(struct rte_eth_dev *dev)
4610 {
4611 	struct rte_flow_error error;
4612 
4613 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4614 	return flow_mreg_add_default_copy_action(dev, &error);
4615 }
4616 
4617 /**
4618  * Allocate intermediate resources for flow creation.
4619  *
4620  * @param dev
4621  *   Pointer to Ethernet device.
4622  */
4623 void
4624 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4625 {
4626 	struct mlx5_priv *priv = dev->data->dev_private;
4627 
4628 	if (!priv->inter_flows)
4629 		priv->inter_flows = rte_calloc(__func__, MLX5_NUM_MAX_DEV_FLOWS,
4630 					       sizeof(struct mlx5_flow), 0);
4631 	/* Reset the index. */
4632 	priv->flow_idx = 0;
4633 	priv->flow_nested_idx = 0;
4634 }
4635 
4636 /**
4637  * Free intermediate resources for flows.
4638  *
4639  * @param dev
4640  *   Pointer to Ethernet device.
4641  */
4642 void
4643 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4644 {
4645 	struct mlx5_priv *priv = dev->data->dev_private;
4646 
4647 	rte_free(priv->inter_flows);
4648 	priv->inter_flows = NULL;
4649 }
4650 
4651 /**
4652  * Verify the flow list is empty
4653  *
4654  * @param dev
4655  *  Pointer to Ethernet device.
4656  *
4657  * @return the number of flows not released.
4658  */
4659 int
4660 mlx5_flow_verify(struct rte_eth_dev *dev)
4661 {
4662 	struct mlx5_priv *priv = dev->data->dev_private;
4663 	struct rte_flow *flow;
4664 	int ret = 0;
4665 
4666 	TAILQ_FOREACH(flow, &priv->flows, next) {
4667 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4668 			dev->data->port_id, (void *)flow);
4669 		++ret;
4670 	}
4671 	return ret;
4672 }
4673 
4674 /**
4675  * Enable default hairpin egress flow.
4676  *
4677  * @param dev
4678  *   Pointer to Ethernet device.
4679  * @param queue
4680  *   The queue index.
4681  *
4682  * @return
4683  *   0 on success, a negative errno value otherwise and rte_errno is set.
4684  */
4685 int
4686 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4687 			    uint32_t queue)
4688 {
4689 	struct mlx5_priv *priv = dev->data->dev_private;
4690 	const struct rte_flow_attr attr = {
4691 		.egress = 1,
4692 		.priority = 0,
4693 	};
4694 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4695 		.queue = queue,
4696 	};
4697 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4698 		.queue = UINT32_MAX,
4699 	};
4700 	struct rte_flow_item items[] = {
4701 		{
4702 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4703 			.spec = &queue_spec,
4704 			.last = NULL,
4705 			.mask = &queue_mask,
4706 		},
4707 		{
4708 			.type = RTE_FLOW_ITEM_TYPE_END,
4709 		},
4710 	};
4711 	struct rte_flow_action_jump jump = {
4712 		.group = MLX5_HAIRPIN_TX_TABLE,
4713 	};
4714 	struct rte_flow_action actions[2];
4715 	struct rte_flow *flow;
4716 	struct rte_flow_error error;
4717 
4718 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4719 	actions[0].conf = &jump;
4720 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4721 	flow = flow_list_create(dev, &priv->ctrl_flows,
4722 				&attr, items, actions, false, &error);
4723 	if (!flow) {
4724 		DRV_LOG(DEBUG,
4725 			"Failed to create ctrl flow: rte_errno(%d),"
4726 			" type(%d), message(%s)",
4727 			rte_errno, error.type,
4728 			error.message ? error.message : " (no stated reason)");
4729 		return -rte_errno;
4730 	}
4731 	return 0;
4732 }
4733 
4734 /**
4735  * Enable a control flow configured from the control plane.
4736  *
4737  * @param dev
4738  *   Pointer to Ethernet device.
4739  * @param eth_spec
4740  *   An Ethernet flow spec to apply.
4741  * @param eth_mask
4742  *   An Ethernet flow mask to apply.
4743  * @param vlan_spec
4744  *   A VLAN flow spec to apply.
4745  * @param vlan_mask
4746  *   A VLAN flow mask to apply.
4747  *
4748  * @return
4749  *   0 on success, a negative errno value otherwise and rte_errno is set.
4750  */
4751 int
4752 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4753 		    struct rte_flow_item_eth *eth_spec,
4754 		    struct rte_flow_item_eth *eth_mask,
4755 		    struct rte_flow_item_vlan *vlan_spec,
4756 		    struct rte_flow_item_vlan *vlan_mask)
4757 {
4758 	struct mlx5_priv *priv = dev->data->dev_private;
4759 	const struct rte_flow_attr attr = {
4760 		.ingress = 1,
4761 		.priority = MLX5_FLOW_PRIO_RSVD,
4762 	};
4763 	struct rte_flow_item items[] = {
4764 		{
4765 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4766 			.spec = eth_spec,
4767 			.last = NULL,
4768 			.mask = eth_mask,
4769 		},
4770 		{
4771 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4772 					      RTE_FLOW_ITEM_TYPE_END,
4773 			.spec = vlan_spec,
4774 			.last = NULL,
4775 			.mask = vlan_mask,
4776 		},
4777 		{
4778 			.type = RTE_FLOW_ITEM_TYPE_END,
4779 		},
4780 	};
4781 	uint16_t queue[priv->reta_idx_n];
4782 	struct rte_flow_action_rss action_rss = {
4783 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4784 		.level = 0,
4785 		.types = priv->rss_conf.rss_hf,
4786 		.key_len = priv->rss_conf.rss_key_len,
4787 		.queue_num = priv->reta_idx_n,
4788 		.key = priv->rss_conf.rss_key,
4789 		.queue = queue,
4790 	};
4791 	struct rte_flow_action actions[] = {
4792 		{
4793 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4794 			.conf = &action_rss,
4795 		},
4796 		{
4797 			.type = RTE_FLOW_ACTION_TYPE_END,
4798 		},
4799 	};
4800 	struct rte_flow *flow;
4801 	struct rte_flow_error error;
4802 	unsigned int i;
4803 
4804 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4805 		return 0;
4806 	}
4807 	for (i = 0; i != priv->reta_idx_n; ++i)
4808 		queue[i] = (*priv->reta_idx)[i];
4809 	flow = flow_list_create(dev, &priv->ctrl_flows,
4810 				&attr, items, actions, false, &error);
4811 	if (!flow)
4812 		return -rte_errno;
4813 	return 0;
4814 }
4815 
4816 /**
4817  * Enable a flow control configured from the control plane.
4818  *
4819  * @param dev
4820  *   Pointer to Ethernet device.
4821  * @param eth_spec
4822  *   An Ethernet flow spec to apply.
4823  * @param eth_mask
4824  *   An Ethernet flow mask to apply.
4825  *
4826  * @return
4827  *   0 on success, a negative errno value otherwise and rte_errno is set.
4828  */
4829 int
4830 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4831 	       struct rte_flow_item_eth *eth_spec,
4832 	       struct rte_flow_item_eth *eth_mask)
4833 {
4834 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4835 }
4836 
4837 /**
4838  * Destroy a flow.
4839  *
4840  * @see rte_flow_destroy()
4841  * @see rte_flow_ops
4842  */
4843 int
4844 mlx5_flow_destroy(struct rte_eth_dev *dev,
4845 		  struct rte_flow *flow,
4846 		  struct rte_flow_error *error __rte_unused)
4847 {
4848 	struct mlx5_priv *priv = dev->data->dev_private;
4849 
4850 	flow_list_destroy(dev, &priv->flows, flow);
4851 	return 0;
4852 }
4853 
4854 /**
4855  * Destroy all flows.
4856  *
4857  * @see rte_flow_flush()
4858  * @see rte_flow_ops
4859  */
4860 int
4861 mlx5_flow_flush(struct rte_eth_dev *dev,
4862 		struct rte_flow_error *error __rte_unused)
4863 {
4864 	struct mlx5_priv *priv = dev->data->dev_private;
4865 
4866 	mlx5_flow_list_flush(dev, &priv->flows, false);
4867 	return 0;
4868 }
4869 
4870 /**
4871  * Isolated mode.
4872  *
4873  * @see rte_flow_isolate()
4874  * @see rte_flow_ops
4875  */
4876 int
4877 mlx5_flow_isolate(struct rte_eth_dev *dev,
4878 		  int enable,
4879 		  struct rte_flow_error *error)
4880 {
4881 	struct mlx5_priv *priv = dev->data->dev_private;
4882 
4883 	if (dev->data->dev_started) {
4884 		rte_flow_error_set(error, EBUSY,
4885 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4886 				   NULL,
4887 				   "port must be stopped first");
4888 		return -rte_errno;
4889 	}
4890 	priv->isolated = !!enable;
4891 	if (enable)
4892 		dev->dev_ops = &mlx5_dev_ops_isolate;
4893 	else
4894 		dev->dev_ops = &mlx5_dev_ops;
4895 	return 0;
4896 }
4897 
4898 /**
4899  * Query a flow.
4900  *
4901  * @see rte_flow_query()
4902  * @see rte_flow_ops
4903  */
4904 static int
4905 flow_drv_query(struct rte_eth_dev *dev,
4906 	       struct rte_flow *flow,
4907 	       const struct rte_flow_action *actions,
4908 	       void *data,
4909 	       struct rte_flow_error *error)
4910 {
4911 	const struct mlx5_flow_driver_ops *fops;
4912 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4913 
4914 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4915 	fops = flow_get_drv_ops(ftype);
4916 
4917 	return fops->query(dev, flow, actions, data, error);
4918 }
4919 
4920 /**
4921  * Query a flow.
4922  *
4923  * @see rte_flow_query()
4924  * @see rte_flow_ops
4925  */
4926 int
4927 mlx5_flow_query(struct rte_eth_dev *dev,
4928 		struct rte_flow *flow,
4929 		const struct rte_flow_action *actions,
4930 		void *data,
4931 		struct rte_flow_error *error)
4932 {
4933 	int ret;
4934 
4935 	ret = flow_drv_query(dev, flow, actions, data, error);
4936 	if (ret < 0)
4937 		return ret;
4938 	return 0;
4939 }
4940 
4941 /**
4942  * Convert a flow director filter to a generic flow.
4943  *
4944  * @param dev
4945  *   Pointer to Ethernet device.
4946  * @param fdir_filter
4947  *   Flow director filter to add.
4948  * @param attributes
4949  *   Generic flow parameters structure.
4950  *
4951  * @return
4952  *   0 on success, a negative errno value otherwise and rte_errno is set.
4953  */
4954 static int
4955 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4956 			 const struct rte_eth_fdir_filter *fdir_filter,
4957 			 struct mlx5_fdir *attributes)
4958 {
4959 	struct mlx5_priv *priv = dev->data->dev_private;
4960 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4961 	const struct rte_eth_fdir_masks *mask =
4962 		&dev->data->dev_conf.fdir_conf.mask;
4963 
4964 	/* Validate queue number. */
4965 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4966 		DRV_LOG(ERR, "port %u invalid queue number %d",
4967 			dev->data->port_id, fdir_filter->action.rx_queue);
4968 		rte_errno = EINVAL;
4969 		return -rte_errno;
4970 	}
4971 	attributes->attr.ingress = 1;
4972 	attributes->items[0] = (struct rte_flow_item) {
4973 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4974 		.spec = &attributes->l2,
4975 		.mask = &attributes->l2_mask,
4976 	};
4977 	switch (fdir_filter->action.behavior) {
4978 	case RTE_ETH_FDIR_ACCEPT:
4979 		attributes->actions[0] = (struct rte_flow_action){
4980 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4981 			.conf = &attributes->queue,
4982 		};
4983 		break;
4984 	case RTE_ETH_FDIR_REJECT:
4985 		attributes->actions[0] = (struct rte_flow_action){
4986 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4987 		};
4988 		break;
4989 	default:
4990 		DRV_LOG(ERR, "port %u invalid behavior %d",
4991 			dev->data->port_id,
4992 			fdir_filter->action.behavior);
4993 		rte_errno = ENOTSUP;
4994 		return -rte_errno;
4995 	}
4996 	attributes->queue.index = fdir_filter->action.rx_queue;
4997 	/* Handle L3. */
4998 	switch (fdir_filter->input.flow_type) {
4999 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5000 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5001 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5002 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5003 			.src_addr = input->flow.ip4_flow.src_ip,
5004 			.dst_addr = input->flow.ip4_flow.dst_ip,
5005 			.time_to_live = input->flow.ip4_flow.ttl,
5006 			.type_of_service = input->flow.ip4_flow.tos,
5007 		};
5008 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5009 			.src_addr = mask->ipv4_mask.src_ip,
5010 			.dst_addr = mask->ipv4_mask.dst_ip,
5011 			.time_to_live = mask->ipv4_mask.ttl,
5012 			.type_of_service = mask->ipv4_mask.tos,
5013 			.next_proto_id = mask->ipv4_mask.proto,
5014 		};
5015 		attributes->items[1] = (struct rte_flow_item){
5016 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5017 			.spec = &attributes->l3,
5018 			.mask = &attributes->l3_mask,
5019 		};
5020 		break;
5021 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5022 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5023 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5024 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5025 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5026 			.proto = input->flow.ipv6_flow.proto,
5027 		};
5028 
5029 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5030 		       input->flow.ipv6_flow.src_ip,
5031 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5032 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5033 		       input->flow.ipv6_flow.dst_ip,
5034 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5035 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5036 		       mask->ipv6_mask.src_ip,
5037 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5038 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5039 		       mask->ipv6_mask.dst_ip,
5040 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5041 		attributes->items[1] = (struct rte_flow_item){
5042 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5043 			.spec = &attributes->l3,
5044 			.mask = &attributes->l3_mask,
5045 		};
5046 		break;
5047 	default:
5048 		DRV_LOG(ERR, "port %u invalid flow type%d",
5049 			dev->data->port_id, fdir_filter->input.flow_type);
5050 		rte_errno = ENOTSUP;
5051 		return -rte_errno;
5052 	}
5053 	/* Handle L4. */
5054 	switch (fdir_filter->input.flow_type) {
5055 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5056 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5057 			.src_port = input->flow.udp4_flow.src_port,
5058 			.dst_port = input->flow.udp4_flow.dst_port,
5059 		};
5060 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5061 			.src_port = mask->src_port_mask,
5062 			.dst_port = mask->dst_port_mask,
5063 		};
5064 		attributes->items[2] = (struct rte_flow_item){
5065 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5066 			.spec = &attributes->l4,
5067 			.mask = &attributes->l4_mask,
5068 		};
5069 		break;
5070 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5071 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5072 			.src_port = input->flow.tcp4_flow.src_port,
5073 			.dst_port = input->flow.tcp4_flow.dst_port,
5074 		};
5075 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5076 			.src_port = mask->src_port_mask,
5077 			.dst_port = mask->dst_port_mask,
5078 		};
5079 		attributes->items[2] = (struct rte_flow_item){
5080 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5081 			.spec = &attributes->l4,
5082 			.mask = &attributes->l4_mask,
5083 		};
5084 		break;
5085 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5086 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5087 			.src_port = input->flow.udp6_flow.src_port,
5088 			.dst_port = input->flow.udp6_flow.dst_port,
5089 		};
5090 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5091 			.src_port = mask->src_port_mask,
5092 			.dst_port = mask->dst_port_mask,
5093 		};
5094 		attributes->items[2] = (struct rte_flow_item){
5095 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5096 			.spec = &attributes->l4,
5097 			.mask = &attributes->l4_mask,
5098 		};
5099 		break;
5100 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5101 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5102 			.src_port = input->flow.tcp6_flow.src_port,
5103 			.dst_port = input->flow.tcp6_flow.dst_port,
5104 		};
5105 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5106 			.src_port = mask->src_port_mask,
5107 			.dst_port = mask->dst_port_mask,
5108 		};
5109 		attributes->items[2] = (struct rte_flow_item){
5110 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5111 			.spec = &attributes->l4,
5112 			.mask = &attributes->l4_mask,
5113 		};
5114 		break;
5115 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5116 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5117 		break;
5118 	default:
5119 		DRV_LOG(ERR, "port %u invalid flow type%d",
5120 			dev->data->port_id, fdir_filter->input.flow_type);
5121 		rte_errno = ENOTSUP;
5122 		return -rte_errno;
5123 	}
5124 	return 0;
5125 }
5126 
5127 #define FLOW_FDIR_CMP(f1, f2, fld) \
5128 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5129 
5130 /**
5131  * Compare two FDIR flows. If items and actions are identical, the two flows are
5132  * regarded as same.
5133  *
5134  * @param dev
5135  *   Pointer to Ethernet device.
5136  * @param f1
5137  *   FDIR flow to compare.
5138  * @param f2
5139  *   FDIR flow to compare.
5140  *
5141  * @return
5142  *   Zero on match, 1 otherwise.
5143  */
5144 static int
5145 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5146 {
5147 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5148 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5149 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5150 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5151 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5152 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5153 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5154 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5155 		return 1;
5156 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5157 	    FLOW_FDIR_CMP(f1, f2, queue))
5158 		return 1;
5159 	return 0;
5160 }
5161 
5162 /**
5163  * Search device flow list to find out a matched FDIR flow.
5164  *
5165  * @param dev
5166  *   Pointer to Ethernet device.
5167  * @param fdir_flow
5168  *   FDIR flow to lookup.
5169  *
5170  * @return
5171  *   Pointer of flow if found, NULL otherwise.
5172  */
5173 static struct rte_flow *
5174 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5175 {
5176 	struct mlx5_priv *priv = dev->data->dev_private;
5177 	struct rte_flow *flow = NULL;
5178 
5179 	MLX5_ASSERT(fdir_flow);
5180 	TAILQ_FOREACH(flow, &priv->flows, next) {
5181 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5182 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5183 				dev->data->port_id, (void *)flow);
5184 			break;
5185 		}
5186 	}
5187 	return flow;
5188 }
5189 
5190 /**
5191  * Add new flow director filter and store it in list.
5192  *
5193  * @param dev
5194  *   Pointer to Ethernet device.
5195  * @param fdir_filter
5196  *   Flow director filter to add.
5197  *
5198  * @return
5199  *   0 on success, a negative errno value otherwise and rte_errno is set.
5200  */
5201 static int
5202 flow_fdir_filter_add(struct rte_eth_dev *dev,
5203 		     const struct rte_eth_fdir_filter *fdir_filter)
5204 {
5205 	struct mlx5_priv *priv = dev->data->dev_private;
5206 	struct mlx5_fdir *fdir_flow;
5207 	struct rte_flow *flow;
5208 	int ret;
5209 
5210 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5211 	if (!fdir_flow) {
5212 		rte_errno = ENOMEM;
5213 		return -rte_errno;
5214 	}
5215 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5216 	if (ret)
5217 		goto error;
5218 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5219 	if (flow) {
5220 		rte_errno = EEXIST;
5221 		goto error;
5222 	}
5223 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5224 				fdir_flow->items, fdir_flow->actions, true,
5225 				NULL);
5226 	if (!flow)
5227 		goto error;
5228 	MLX5_ASSERT(!flow->fdir);
5229 	flow->fdir = fdir_flow;
5230 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5231 		dev->data->port_id, (void *)flow);
5232 	return 0;
5233 error:
5234 	rte_free(fdir_flow);
5235 	return -rte_errno;
5236 }
5237 
5238 /**
5239  * Delete specific filter.
5240  *
5241  * @param dev
5242  *   Pointer to Ethernet device.
5243  * @param fdir_filter
5244  *   Filter to be deleted.
5245  *
5246  * @return
5247  *   0 on success, a negative errno value otherwise and rte_errno is set.
5248  */
5249 static int
5250 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5251 			const struct rte_eth_fdir_filter *fdir_filter)
5252 {
5253 	struct mlx5_priv *priv = dev->data->dev_private;
5254 	struct rte_flow *flow;
5255 	struct mlx5_fdir fdir_flow = {
5256 		.attr.group = 0,
5257 	};
5258 	int ret;
5259 
5260 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5261 	if (ret)
5262 		return -rte_errno;
5263 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5264 	if (!flow) {
5265 		rte_errno = ENOENT;
5266 		return -rte_errno;
5267 	}
5268 	flow_list_destroy(dev, &priv->flows, flow);
5269 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5270 		dev->data->port_id, (void *)flow);
5271 	return 0;
5272 }
5273 
5274 /**
5275  * Update queue for specific filter.
5276  *
5277  * @param dev
5278  *   Pointer to Ethernet device.
5279  * @param fdir_filter
5280  *   Filter to be updated.
5281  *
5282  * @return
5283  *   0 on success, a negative errno value otherwise and rte_errno is set.
5284  */
5285 static int
5286 flow_fdir_filter_update(struct rte_eth_dev *dev,
5287 			const struct rte_eth_fdir_filter *fdir_filter)
5288 {
5289 	int ret;
5290 
5291 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5292 	if (ret)
5293 		return ret;
5294 	return flow_fdir_filter_add(dev, fdir_filter);
5295 }
5296 
5297 /**
5298  * Flush all filters.
5299  *
5300  * @param dev
5301  *   Pointer to Ethernet device.
5302  */
5303 static void
5304 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5305 {
5306 	struct mlx5_priv *priv = dev->data->dev_private;
5307 
5308 	mlx5_flow_list_flush(dev, &priv->flows, false);
5309 }
5310 
5311 /**
5312  * Get flow director information.
5313  *
5314  * @param dev
5315  *   Pointer to Ethernet device.
5316  * @param[out] fdir_info
5317  *   Resulting flow director information.
5318  */
5319 static void
5320 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5321 {
5322 	struct rte_eth_fdir_masks *mask =
5323 		&dev->data->dev_conf.fdir_conf.mask;
5324 
5325 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5326 	fdir_info->guarant_spc = 0;
5327 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5328 	fdir_info->max_flexpayload = 0;
5329 	fdir_info->flow_types_mask[0] = 0;
5330 	fdir_info->flex_payload_unit = 0;
5331 	fdir_info->max_flex_payload_segment_num = 0;
5332 	fdir_info->flex_payload_limit = 0;
5333 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5334 }
5335 
5336 /**
5337  * Deal with flow director operations.
5338  *
5339  * @param dev
5340  *   Pointer to Ethernet device.
5341  * @param filter_op
5342  *   Operation to perform.
5343  * @param arg
5344  *   Pointer to operation-specific structure.
5345  *
5346  * @return
5347  *   0 on success, a negative errno value otherwise and rte_errno is set.
5348  */
5349 static int
5350 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5351 		    void *arg)
5352 {
5353 	enum rte_fdir_mode fdir_mode =
5354 		dev->data->dev_conf.fdir_conf.mode;
5355 
5356 	if (filter_op == RTE_ETH_FILTER_NOP)
5357 		return 0;
5358 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5359 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5360 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5361 			dev->data->port_id, fdir_mode);
5362 		rte_errno = EINVAL;
5363 		return -rte_errno;
5364 	}
5365 	switch (filter_op) {
5366 	case RTE_ETH_FILTER_ADD:
5367 		return flow_fdir_filter_add(dev, arg);
5368 	case RTE_ETH_FILTER_UPDATE:
5369 		return flow_fdir_filter_update(dev, arg);
5370 	case RTE_ETH_FILTER_DELETE:
5371 		return flow_fdir_filter_delete(dev, arg);
5372 	case RTE_ETH_FILTER_FLUSH:
5373 		flow_fdir_filter_flush(dev);
5374 		break;
5375 	case RTE_ETH_FILTER_INFO:
5376 		flow_fdir_info_get(dev, arg);
5377 		break;
5378 	default:
5379 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5380 			dev->data->port_id, filter_op);
5381 		rte_errno = EINVAL;
5382 		return -rte_errno;
5383 	}
5384 	return 0;
5385 }
5386 
5387 /**
5388  * Manage filter operations.
5389  *
5390  * @param dev
5391  *   Pointer to Ethernet device structure.
5392  * @param filter_type
5393  *   Filter type.
5394  * @param filter_op
5395  *   Operation to perform.
5396  * @param arg
5397  *   Pointer to operation-specific structure.
5398  *
5399  * @return
5400  *   0 on success, a negative errno value otherwise and rte_errno is set.
5401  */
5402 int
5403 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5404 		     enum rte_filter_type filter_type,
5405 		     enum rte_filter_op filter_op,
5406 		     void *arg)
5407 {
5408 	switch (filter_type) {
5409 	case RTE_ETH_FILTER_GENERIC:
5410 		if (filter_op != RTE_ETH_FILTER_GET) {
5411 			rte_errno = EINVAL;
5412 			return -rte_errno;
5413 		}
5414 		*(const void **)arg = &mlx5_flow_ops;
5415 		return 0;
5416 	case RTE_ETH_FILTER_FDIR:
5417 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5418 	default:
5419 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5420 			dev->data->port_id, filter_type);
5421 		rte_errno = ENOTSUP;
5422 		return -rte_errno;
5423 	}
5424 	return 0;
5425 }
5426 
5427 /**
5428  * Create the needed meter and suffix tables.
5429  *
5430  * @param[in] dev
5431  *   Pointer to Ethernet device.
5432  * @param[in] fm
5433  *   Pointer to the flow meter.
5434  *
5435  * @return
5436  *   Pointer to table set on success, NULL otherwise.
5437  */
5438 struct mlx5_meter_domains_infos *
5439 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5440 			  const struct mlx5_flow_meter *fm)
5441 {
5442 	const struct mlx5_flow_driver_ops *fops;
5443 
5444 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5445 	return fops->create_mtr_tbls(dev, fm);
5446 }
5447 
5448 /**
5449  * Destroy the meter table set.
5450  *
5451  * @param[in] dev
5452  *   Pointer to Ethernet device.
5453  * @param[in] tbl
5454  *   Pointer to the meter table set.
5455  *
5456  * @return
5457  *   0 on success.
5458  */
5459 int
5460 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5461 			   struct mlx5_meter_domains_infos *tbls)
5462 {
5463 	const struct mlx5_flow_driver_ops *fops;
5464 
5465 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5466 	return fops->destroy_mtr_tbls(dev, tbls);
5467 }
5468 
5469 /**
5470  * Create policer rules.
5471  *
5472  * @param[in] dev
5473  *   Pointer to Ethernet device.
5474  * @param[in] fm
5475  *   Pointer to flow meter structure.
5476  * @param[in] attr
5477  *   Pointer to flow attributes.
5478  *
5479  * @return
5480  *   0 on success, -1 otherwise.
5481  */
5482 int
5483 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5484 			       struct mlx5_flow_meter *fm,
5485 			       const struct rte_flow_attr *attr)
5486 {
5487 	const struct mlx5_flow_driver_ops *fops;
5488 
5489 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5490 	return fops->create_policer_rules(dev, fm, attr);
5491 }
5492 
5493 /**
5494  * Destroy policer rules.
5495  *
5496  * @param[in] fm
5497  *   Pointer to flow meter structure.
5498  * @param[in] attr
5499  *   Pointer to flow attributes.
5500  *
5501  * @return
5502  *   0 on success, -1 otherwise.
5503  */
5504 int
5505 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5506 				struct mlx5_flow_meter *fm,
5507 				const struct rte_flow_attr *attr)
5508 {
5509 	const struct mlx5_flow_driver_ops *fops;
5510 
5511 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5512 	return fops->destroy_policer_rules(dev, fm, attr);
5513 }
5514 
5515 /**
5516  * Allocate a counter.
5517  *
5518  * @param[in] dev
5519  *   Pointer to Ethernet device structure.
5520  *
5521  * @return
5522  *   Index to allocated counter  on success, 0 otherwise.
5523  */
5524 uint32_t
5525 mlx5_counter_alloc(struct rte_eth_dev *dev)
5526 {
5527 	const struct mlx5_flow_driver_ops *fops;
5528 	struct rte_flow_attr attr = { .transfer = 0 };
5529 
5530 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5531 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5532 		return fops->counter_alloc(dev);
5533 	}
5534 	DRV_LOG(ERR,
5535 		"port %u counter allocate is not supported.",
5536 		 dev->data->port_id);
5537 	return 0;
5538 }
5539 
5540 /**
5541  * Free a counter.
5542  *
5543  * @param[in] dev
5544  *   Pointer to Ethernet device structure.
5545  * @param[in] cnt
5546  *   Index to counter to be free.
5547  */
5548 void
5549 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5550 {
5551 	const struct mlx5_flow_driver_ops *fops;
5552 	struct rte_flow_attr attr = { .transfer = 0 };
5553 
5554 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5555 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5556 		fops->counter_free(dev, cnt);
5557 		return;
5558 	}
5559 	DRV_LOG(ERR,
5560 		"port %u counter free is not supported.",
5561 		 dev->data->port_id);
5562 }
5563 
5564 /**
5565  * Query counter statistics.
5566  *
5567  * @param[in] dev
5568  *   Pointer to Ethernet device structure.
5569  * @param[in] cnt
5570  *   Index to counter to query.
5571  * @param[in] clear
5572  *   Set to clear counter statistics.
5573  * @param[out] pkts
5574  *   The counter hits packets number to save.
5575  * @param[out] bytes
5576  *   The counter hits bytes number to save.
5577  *
5578  * @return
5579  *   0 on success, a negative errno value otherwise.
5580  */
5581 int
5582 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5583 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5584 {
5585 	const struct mlx5_flow_driver_ops *fops;
5586 	struct rte_flow_attr attr = { .transfer = 0 };
5587 
5588 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5589 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5590 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5591 	}
5592 	DRV_LOG(ERR,
5593 		"port %u counter query is not supported.",
5594 		 dev->data->port_id);
5595 	return -ENOTSUP;
5596 }
5597 
5598 #define MLX5_POOL_QUERY_FREQ_US 1000000
5599 
5600 /**
5601  * Set the periodic procedure for triggering asynchronous batch queries for all
5602  * the counter pools.
5603  *
5604  * @param[in] sh
5605  *   Pointer to mlx5_ibv_shared object.
5606  */
5607 void
5608 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5609 {
5610 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5611 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5612 	uint32_t us;
5613 
5614 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5615 	pools_n += rte_atomic16_read(&cont->n_valid);
5616 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5617 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5618 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5619 		sh->cmng.query_thread_on = 0;
5620 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5621 	} else {
5622 		sh->cmng.query_thread_on = 1;
5623 	}
5624 }
5625 
5626 /**
5627  * The periodic procedure for triggering asynchronous batch queries for all the
5628  * counter pools. This function is probably called by the host thread.
5629  *
5630  * @param[in] arg
5631  *   The parameter for the alarm process.
5632  */
5633 void
5634 mlx5_flow_query_alarm(void *arg)
5635 {
5636 	struct mlx5_ibv_shared *sh = arg;
5637 	struct mlx5_devx_obj *dcs;
5638 	uint16_t offset;
5639 	int ret;
5640 	uint8_t batch = sh->cmng.batch;
5641 	uint16_t pool_index = sh->cmng.pool_index;
5642 	struct mlx5_pools_container *cont;
5643 	struct mlx5_pools_container *mcont;
5644 	struct mlx5_flow_counter_pool *pool;
5645 
5646 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5647 		goto set_alarm;
5648 next_container:
5649 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5650 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5651 	/* Check if resize was done and need to flip a container. */
5652 	if (cont != mcont) {
5653 		if (cont->pools) {
5654 			/* Clean the old container. */
5655 			rte_free(cont->pools);
5656 			memset(cont, 0, sizeof(*cont));
5657 		}
5658 		rte_cio_wmb();
5659 		 /* Flip the host container. */
5660 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5661 		cont = mcont;
5662 	}
5663 	if (!cont->pools) {
5664 		/* 2 empty containers case is unexpected. */
5665 		if (unlikely(batch != sh->cmng.batch))
5666 			goto set_alarm;
5667 		batch ^= 0x1;
5668 		pool_index = 0;
5669 		goto next_container;
5670 	}
5671 	pool = cont->pools[pool_index];
5672 	if (pool->raw_hw)
5673 		/* There is a pool query in progress. */
5674 		goto set_alarm;
5675 	pool->raw_hw =
5676 		LIST_FIRST(&sh->cmng.free_stat_raws);
5677 	if (!pool->raw_hw)
5678 		/* No free counter statistics raw memory. */
5679 		goto set_alarm;
5680 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5681 							      (&pool->a64_dcs);
5682 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5683 	/*
5684 	 * Identify the counters released between query trigger and query
5685 	 * handle more effiecntly. The counter released in this gap period
5686 	 * should wait for a new round of query as the new arrived packets
5687 	 * will not be taken into account.
5688 	 */
5689 	rte_atomic64_add(&pool->start_query_gen, 1);
5690 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5691 					       offset, NULL, NULL,
5692 					       pool->raw_hw->mem_mng->dm->id,
5693 					       (void *)(uintptr_t)
5694 					       (pool->raw_hw->data + offset),
5695 					       sh->devx_comp,
5696 					       (uint64_t)(uintptr_t)pool);
5697 	if (ret) {
5698 		rte_atomic64_sub(&pool->start_query_gen, 1);
5699 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5700 			" %d", pool->min_dcs->id);
5701 		pool->raw_hw = NULL;
5702 		goto set_alarm;
5703 	}
5704 	pool->raw_hw->min_dcs_id = dcs->id;
5705 	LIST_REMOVE(pool->raw_hw, next);
5706 	sh->cmng.pending_queries++;
5707 	pool_index++;
5708 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5709 		batch ^= 0x1;
5710 		pool_index = 0;
5711 	}
5712 set_alarm:
5713 	sh->cmng.batch = batch;
5714 	sh->cmng.pool_index = pool_index;
5715 	mlx5_set_query_alarm(sh);
5716 }
5717 
5718 /**
5719  * Handler for the HW respond about ready values from an asynchronous batch
5720  * query. This function is probably called by the host thread.
5721  *
5722  * @param[in] sh
5723  *   The pointer to the shared IB device context.
5724  * @param[in] async_id
5725  *   The Devx async ID.
5726  * @param[in] status
5727  *   The status of the completion.
5728  */
5729 void
5730 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5731 				  uint64_t async_id, int status)
5732 {
5733 	struct mlx5_flow_counter_pool *pool =
5734 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5735 	struct mlx5_counter_stats_raw *raw_to_free;
5736 
5737 	if (unlikely(status)) {
5738 		rte_atomic64_sub(&pool->start_query_gen, 1);
5739 		raw_to_free = pool->raw_hw;
5740 	} else {
5741 		raw_to_free = pool->raw;
5742 		rte_spinlock_lock(&pool->sl);
5743 		pool->raw = pool->raw_hw;
5744 		rte_spinlock_unlock(&pool->sl);
5745 		MLX5_ASSERT(rte_atomic64_read(&pool->end_query_gen) + 1 ==
5746 			    rte_atomic64_read(&pool->start_query_gen));
5747 		rte_atomic64_set(&pool->end_query_gen,
5748 				 rte_atomic64_read(&pool->start_query_gen));
5749 		/* Be sure the new raw counters data is updated in memory. */
5750 		rte_cio_wmb();
5751 	}
5752 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5753 	pool->raw_hw = NULL;
5754 	sh->cmng.pending_queries--;
5755 }
5756 
5757 /**
5758  * Translate the rte_flow group index to HW table value.
5759  *
5760  * @param[in] attributes
5761  *   Pointer to flow attributes
5762  * @param[in] external
5763  *   Value is part of flow rule created by request external to PMD.
5764  * @param[in] group
5765  *   rte_flow group index value.
5766  * @param[out] fdb_def_rule
5767  *   Whether fdb jump to table 1 is configured.
5768  * @param[out] table
5769  *   HW table value.
5770  * @param[out] error
5771  *   Pointer to error structure.
5772  *
5773  * @return
5774  *   0 on success, a negative errno value otherwise and rte_errno is set.
5775  */
5776 int
5777 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5778 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
5779 			 struct rte_flow_error *error)
5780 {
5781 	if (attributes->transfer && external && fdb_def_rule) {
5782 		if (group == UINT32_MAX)
5783 			return rte_flow_error_set
5784 						(error, EINVAL,
5785 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5786 						 NULL,
5787 						 "group index not supported");
5788 		*table = group + 1;
5789 	} else {
5790 		*table = group;
5791 	}
5792 	return 0;
5793 }
5794 
5795 /**
5796  * Discover availability of metadata reg_c's.
5797  *
5798  * Iteratively use test flows to check availability.
5799  *
5800  * @param[in] dev
5801  *   Pointer to the Ethernet device structure.
5802  *
5803  * @return
5804  *   0 on success, a negative errno value otherwise and rte_errno is set.
5805  */
5806 int
5807 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5808 {
5809 	struct mlx5_priv *priv = dev->data->dev_private;
5810 	struct mlx5_dev_config *config = &priv->config;
5811 	enum modify_reg idx;
5812 	int n = 0;
5813 
5814 	/* reg_c[0] and reg_c[1] are reserved. */
5815 	config->flow_mreg_c[n++] = REG_C_0;
5816 	config->flow_mreg_c[n++] = REG_C_1;
5817 	/* Discover availability of other reg_c's. */
5818 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5819 		struct rte_flow_attr attr = {
5820 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5821 			.priority = MLX5_FLOW_PRIO_RSVD,
5822 			.ingress = 1,
5823 		};
5824 		struct rte_flow_item items[] = {
5825 			[0] = {
5826 				.type = RTE_FLOW_ITEM_TYPE_END,
5827 			},
5828 		};
5829 		struct rte_flow_action actions[] = {
5830 			[0] = {
5831 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5832 				.conf = &(struct mlx5_flow_action_copy_mreg){
5833 					.src = REG_C_1,
5834 					.dst = idx,
5835 				},
5836 			},
5837 			[1] = {
5838 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5839 				.conf = &(struct rte_flow_action_jump){
5840 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5841 				},
5842 			},
5843 			[2] = {
5844 				.type = RTE_FLOW_ACTION_TYPE_END,
5845 			},
5846 		};
5847 		struct rte_flow *flow;
5848 		struct rte_flow_error error;
5849 
5850 		if (!config->dv_flow_en)
5851 			break;
5852 		/* Create internal flow, validation skips copy action. */
5853 		flow = flow_list_create(dev, NULL, &attr, items,
5854 					actions, false, &error);
5855 		if (!flow)
5856 			continue;
5857 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5858 			config->flow_mreg_c[n++] = idx;
5859 		flow_list_destroy(dev, NULL, flow);
5860 	}
5861 	for (; n < MLX5_MREG_C_NUM; ++n)
5862 		config->flow_mreg_c[n] = REG_NONE;
5863 	return 0;
5864 }
5865 
5866 /**
5867  * Dump flow raw hw data to file
5868  *
5869  * @param[in] dev
5870  *    The pointer to Ethernet device.
5871  * @param[in] file
5872  *   A pointer to a file for output.
5873  * @param[out] error
5874  *   Perform verbose error reporting if not NULL. PMDs initialize this
5875  *   structure in case of error only.
5876  * @return
5877  *   0 on success, a nagative value otherwise.
5878  */
5879 int
5880 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5881 		   FILE *file,
5882 		   struct rte_flow_error *error __rte_unused)
5883 {
5884 	struct mlx5_priv *priv = dev->data->dev_private;
5885 	struct mlx5_ibv_shared *sh = priv->sh;
5886 
5887 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5888 				       sh->tx_domain, file);
5889 }
5890