xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision c42f44bd1b42cf21df3d94c738f22032fae308b4)
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] dev_flow
716  *   Pointer to device flow structure.
717  */
718 static void
719 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
720 {
721 	struct mlx5_priv *priv = dev->data->dev_private;
722 	struct rte_flow *flow = dev_flow->flow;
723 	const int mark = !!(dev_flow->handle.act_flags &
724 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
725 	const int tunnel = !!(dev_flow->handle.layers & MLX5_FLOW_LAYER_TUNNEL);
726 	unsigned int i;
727 
728 	for (i = 0; i != flow->rss.queue_num; ++i) {
729 		int idx = (*flow->rss.queue)[i];
730 		struct mlx5_rxq_ctrl *rxq_ctrl =
731 			container_of((*priv->rxqs)[idx],
732 				     struct mlx5_rxq_ctrl, rxq);
733 
734 		/*
735 		 * To support metadata register copy on Tx loopback,
736 		 * this must be always enabled (metadata may arive
737 		 * from other port - not from local flows only.
738 		 */
739 		if (priv->config.dv_flow_en &&
740 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
741 		    mlx5_flow_ext_mreg_supported(dev)) {
742 			rxq_ctrl->rxq.mark = 1;
743 			rxq_ctrl->flow_mark_n = 1;
744 		} else if (mark) {
745 			rxq_ctrl->rxq.mark = 1;
746 			rxq_ctrl->flow_mark_n++;
747 		}
748 		if (tunnel) {
749 			unsigned int j;
750 
751 			/* Increase the counter matching the flow. */
752 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
753 				if ((tunnels_info[j].tunnel &
754 				     dev_flow->handle.layers) ==
755 				    tunnels_info[j].tunnel) {
756 					rxq_ctrl->flow_tunnels_n[j]++;
757 					break;
758 				}
759 			}
760 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
761 		}
762 	}
763 }
764 
765 /**
766  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
767  *
768  * @param[in] dev
769  *   Pointer to the Ethernet device structure.
770  * @param[in] flow
771  *   Pointer to flow structure.
772  */
773 static void
774 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
775 {
776 	struct mlx5_flow *dev_flow;
777 
778 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
779 		flow_drv_rxq_flags_set(dev, dev_flow);
780 }
781 
782 /**
783  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
784  * device flow if no other flow uses it with the same kind of request.
785  *
786  * @param dev
787  *   Pointer to Ethernet device.
788  * @param[in] dev_flow
789  *   Pointer to the device flow.
790  */
791 static void
792 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
793 {
794 	struct mlx5_priv *priv = dev->data->dev_private;
795 	struct rte_flow *flow = dev_flow->flow;
796 	const int mark = !!(dev_flow->handle.act_flags &
797 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
798 	const int tunnel = !!(dev_flow->handle.layers & MLX5_FLOW_LAYER_TUNNEL);
799 	unsigned int i;
800 
801 	MLX5_ASSERT(dev->data->dev_started);
802 	for (i = 0; i != flow->rss.queue_num; ++i) {
803 		int idx = (*flow->rss.queue)[i];
804 		struct mlx5_rxq_ctrl *rxq_ctrl =
805 			container_of((*priv->rxqs)[idx],
806 				     struct mlx5_rxq_ctrl, rxq);
807 
808 		if (priv->config.dv_flow_en &&
809 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
810 		    mlx5_flow_ext_mreg_supported(dev)) {
811 			rxq_ctrl->rxq.mark = 1;
812 			rxq_ctrl->flow_mark_n = 1;
813 		} else if (mark) {
814 			rxq_ctrl->flow_mark_n--;
815 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
816 		}
817 		if (tunnel) {
818 			unsigned int j;
819 
820 			/* Decrease the counter matching the flow. */
821 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
822 				if ((tunnels_info[j].tunnel &
823 				     dev_flow->handle.layers) ==
824 				    tunnels_info[j].tunnel) {
825 					rxq_ctrl->flow_tunnels_n[j]--;
826 					break;
827 				}
828 			}
829 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
830 		}
831 	}
832 }
833 
834 /**
835  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
836  * @p flow if no other flow uses it with the same kind of request.
837  *
838  * @param dev
839  *   Pointer to Ethernet device.
840  * @param[in] flow
841  *   Pointer to the flow.
842  */
843 static void
844 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
845 {
846 	struct mlx5_flow *dev_flow;
847 
848 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
849 		flow_drv_rxq_flags_trim(dev, dev_flow);
850 }
851 
852 /**
853  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
854  *
855  * @param dev
856  *   Pointer to Ethernet device.
857  */
858 static void
859 flow_rxq_flags_clear(struct rte_eth_dev *dev)
860 {
861 	struct mlx5_priv *priv = dev->data->dev_private;
862 	unsigned int i;
863 
864 	for (i = 0; i != priv->rxqs_n; ++i) {
865 		struct mlx5_rxq_ctrl *rxq_ctrl;
866 		unsigned int j;
867 
868 		if (!(*priv->rxqs)[i])
869 			continue;
870 		rxq_ctrl = container_of((*priv->rxqs)[i],
871 					struct mlx5_rxq_ctrl, rxq);
872 		rxq_ctrl->flow_mark_n = 0;
873 		rxq_ctrl->rxq.mark = 0;
874 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
875 			rxq_ctrl->flow_tunnels_n[j] = 0;
876 		rxq_ctrl->rxq.tunnel = 0;
877 	}
878 }
879 
880 /*
881  * return a pointer to the desired action in the list of actions.
882  *
883  * @param[in] actions
884  *   The list of actions to search the action in.
885  * @param[in] action
886  *   The action to find.
887  *
888  * @return
889  *   Pointer to the action in the list, if found. NULL otherwise.
890  */
891 const struct rte_flow_action *
892 mlx5_flow_find_action(const struct rte_flow_action *actions,
893 		      enum rte_flow_action_type action)
894 {
895 	if (actions == NULL)
896 		return NULL;
897 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
898 		if (actions->type == action)
899 			return actions;
900 	return NULL;
901 }
902 
903 /*
904  * Validate the flag action.
905  *
906  * @param[in] action_flags
907  *   Bit-fields that holds the actions detected until now.
908  * @param[in] attr
909  *   Attributes of flow that includes this action.
910  * @param[out] error
911  *   Pointer to error structure.
912  *
913  * @return
914  *   0 on success, a negative errno value otherwise and rte_errno is set.
915  */
916 int
917 mlx5_flow_validate_action_flag(uint64_t action_flags,
918 			       const struct rte_flow_attr *attr,
919 			       struct rte_flow_error *error)
920 {
921 	if (action_flags & MLX5_FLOW_ACTION_MARK)
922 		return rte_flow_error_set(error, EINVAL,
923 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
924 					  "can't mark and flag in same flow");
925 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
926 		return rte_flow_error_set(error, EINVAL,
927 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
928 					  "can't have 2 flag"
929 					  " actions in same flow");
930 	if (attr->egress)
931 		return rte_flow_error_set(error, ENOTSUP,
932 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
933 					  "flag action not supported for "
934 					  "egress");
935 	return 0;
936 }
937 
938 /*
939  * Validate the mark action.
940  *
941  * @param[in] action
942  *   Pointer to the queue action.
943  * @param[in] action_flags
944  *   Bit-fields that holds the actions detected until now.
945  * @param[in] attr
946  *   Attributes of flow that includes this action.
947  * @param[out] error
948  *   Pointer to error structure.
949  *
950  * @return
951  *   0 on success, a negative errno value otherwise and rte_errno is set.
952  */
953 int
954 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
955 			       uint64_t action_flags,
956 			       const struct rte_flow_attr *attr,
957 			       struct rte_flow_error *error)
958 {
959 	const struct rte_flow_action_mark *mark = action->conf;
960 
961 	if (!mark)
962 		return rte_flow_error_set(error, EINVAL,
963 					  RTE_FLOW_ERROR_TYPE_ACTION,
964 					  action,
965 					  "configuration cannot be null");
966 	if (mark->id >= MLX5_FLOW_MARK_MAX)
967 		return rte_flow_error_set(error, EINVAL,
968 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
969 					  &mark->id,
970 					  "mark id must in 0 <= id < "
971 					  RTE_STR(MLX5_FLOW_MARK_MAX));
972 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
973 		return rte_flow_error_set(error, EINVAL,
974 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
975 					  "can't flag and mark in same flow");
976 	if (action_flags & MLX5_FLOW_ACTION_MARK)
977 		return rte_flow_error_set(error, EINVAL,
978 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
979 					  "can't have 2 mark actions in same"
980 					  " flow");
981 	if (attr->egress)
982 		return rte_flow_error_set(error, ENOTSUP,
983 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
984 					  "mark action not supported for "
985 					  "egress");
986 	return 0;
987 }
988 
989 /*
990  * Validate the drop action.
991  *
992  * @param[in] action_flags
993  *   Bit-fields that holds the actions detected until now.
994  * @param[in] attr
995  *   Attributes of flow that includes this action.
996  * @param[out] error
997  *   Pointer to error structure.
998  *
999  * @return
1000  *   0 on success, a negative errno value otherwise and rte_errno is set.
1001  */
1002 int
1003 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1004 			       const struct rte_flow_attr *attr,
1005 			       struct rte_flow_error *error)
1006 {
1007 	if (attr->egress)
1008 		return rte_flow_error_set(error, ENOTSUP,
1009 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1010 					  "drop action not supported for "
1011 					  "egress");
1012 	return 0;
1013 }
1014 
1015 /*
1016  * Validate the queue action.
1017  *
1018  * @param[in] action
1019  *   Pointer to the queue action.
1020  * @param[in] action_flags
1021  *   Bit-fields that holds the actions detected until now.
1022  * @param[in] dev
1023  *   Pointer to the Ethernet device structure.
1024  * @param[in] attr
1025  *   Attributes of flow that includes this action.
1026  * @param[out] error
1027  *   Pointer to error structure.
1028  *
1029  * @return
1030  *   0 on success, a negative errno value otherwise and rte_errno is set.
1031  */
1032 int
1033 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1034 				uint64_t action_flags,
1035 				struct rte_eth_dev *dev,
1036 				const struct rte_flow_attr *attr,
1037 				struct rte_flow_error *error)
1038 {
1039 	struct mlx5_priv *priv = dev->data->dev_private;
1040 	const struct rte_flow_action_queue *queue = action->conf;
1041 
1042 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1043 		return rte_flow_error_set(error, EINVAL,
1044 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1045 					  "can't have 2 fate actions in"
1046 					  " same flow");
1047 	if (!priv->rxqs_n)
1048 		return rte_flow_error_set(error, EINVAL,
1049 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1050 					  NULL, "No Rx queues configured");
1051 	if (queue->index >= priv->rxqs_n)
1052 		return rte_flow_error_set(error, EINVAL,
1053 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1054 					  &queue->index,
1055 					  "queue index out of range");
1056 	if (!(*priv->rxqs)[queue->index])
1057 		return rte_flow_error_set(error, EINVAL,
1058 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1059 					  &queue->index,
1060 					  "queue is not configured");
1061 	if (attr->egress)
1062 		return rte_flow_error_set(error, ENOTSUP,
1063 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1064 					  "queue action not supported for "
1065 					  "egress");
1066 	return 0;
1067 }
1068 
1069 /*
1070  * Validate the rss action.
1071  *
1072  * @param[in] action
1073  *   Pointer to the queue action.
1074  * @param[in] action_flags
1075  *   Bit-fields that holds the actions detected until now.
1076  * @param[in] dev
1077  *   Pointer to the Ethernet device structure.
1078  * @param[in] attr
1079  *   Attributes of flow that includes this action.
1080  * @param[in] item_flags
1081  *   Items that were detected.
1082  * @param[out] error
1083  *   Pointer to error structure.
1084  *
1085  * @return
1086  *   0 on success, a negative errno value otherwise and rte_errno is set.
1087  */
1088 int
1089 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1090 			      uint64_t action_flags,
1091 			      struct rte_eth_dev *dev,
1092 			      const struct rte_flow_attr *attr,
1093 			      uint64_t item_flags,
1094 			      struct rte_flow_error *error)
1095 {
1096 	struct mlx5_priv *priv = dev->data->dev_private;
1097 	const struct rte_flow_action_rss *rss = action->conf;
1098 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1099 	unsigned int i;
1100 
1101 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1102 		return rte_flow_error_set(error, EINVAL,
1103 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1104 					  "can't have 2 fate actions"
1105 					  " in same flow");
1106 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1107 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1108 		return rte_flow_error_set(error, ENOTSUP,
1109 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1110 					  &rss->func,
1111 					  "RSS hash function not supported");
1112 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1113 	if (rss->level > 2)
1114 #else
1115 	if (rss->level > 1)
1116 #endif
1117 		return rte_flow_error_set(error, ENOTSUP,
1118 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1119 					  &rss->level,
1120 					  "tunnel RSS is not supported");
1121 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1122 	if (rss->key_len == 0 && rss->key != NULL)
1123 		return rte_flow_error_set(error, ENOTSUP,
1124 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1125 					  &rss->key_len,
1126 					  "RSS hash key length 0");
1127 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1128 		return rte_flow_error_set(error, ENOTSUP,
1129 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1130 					  &rss->key_len,
1131 					  "RSS hash key too small");
1132 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1133 		return rte_flow_error_set(error, ENOTSUP,
1134 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1135 					  &rss->key_len,
1136 					  "RSS hash key too large");
1137 	if (rss->queue_num > priv->config.ind_table_max_size)
1138 		return rte_flow_error_set(error, ENOTSUP,
1139 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1140 					  &rss->queue_num,
1141 					  "number of queues too large");
1142 	if (rss->types & MLX5_RSS_HF_MASK)
1143 		return rte_flow_error_set(error, ENOTSUP,
1144 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1145 					  &rss->types,
1146 					  "some RSS protocols are not"
1147 					  " supported");
1148 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1149 	    !(rss->types & ETH_RSS_IP))
1150 		return rte_flow_error_set(error, EINVAL,
1151 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1152 					  "L3 partial RSS requested but L3 RSS"
1153 					  " type not specified");
1154 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1155 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1156 		return rte_flow_error_set(error, EINVAL,
1157 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1158 					  "L4 partial RSS requested but L4 RSS"
1159 					  " type not specified");
1160 	if (!priv->rxqs_n)
1161 		return rte_flow_error_set(error, EINVAL,
1162 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1163 					  NULL, "No Rx queues configured");
1164 	if (!rss->queue_num)
1165 		return rte_flow_error_set(error, EINVAL,
1166 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1167 					  NULL, "No queues configured");
1168 	for (i = 0; i != rss->queue_num; ++i) {
1169 		if (rss->queue[i] >= priv->rxqs_n)
1170 			return rte_flow_error_set
1171 				(error, EINVAL,
1172 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1173 				 &rss->queue[i], "queue index out of range");
1174 		if (!(*priv->rxqs)[rss->queue[i]])
1175 			return rte_flow_error_set
1176 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1177 				 &rss->queue[i], "queue is not configured");
1178 	}
1179 	if (attr->egress)
1180 		return rte_flow_error_set(error, ENOTSUP,
1181 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1182 					  "rss action not supported for "
1183 					  "egress");
1184 	if (rss->level > 1 &&  !tunnel)
1185 		return rte_flow_error_set(error, EINVAL,
1186 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1187 					  "inner RSS is not supported for "
1188 					  "non-tunnel flows");
1189 	return 0;
1190 }
1191 
1192 /*
1193  * Validate the count action.
1194  *
1195  * @param[in] dev
1196  *   Pointer to the Ethernet device structure.
1197  * @param[in] attr
1198  *   Attributes of flow that includes this action.
1199  * @param[out] error
1200  *   Pointer to error structure.
1201  *
1202  * @return
1203  *   0 on success, a negative errno value otherwise and rte_errno is set.
1204  */
1205 int
1206 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1207 				const struct rte_flow_attr *attr,
1208 				struct rte_flow_error *error)
1209 {
1210 	if (attr->egress)
1211 		return rte_flow_error_set(error, ENOTSUP,
1212 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1213 					  "count action not supported for "
1214 					  "egress");
1215 	return 0;
1216 }
1217 
1218 /**
1219  * Verify the @p attributes will be correctly understood by the NIC and store
1220  * them in the @p flow if everything is correct.
1221  *
1222  * @param[in] dev
1223  *   Pointer to the Ethernet device structure.
1224  * @param[in] attributes
1225  *   Pointer to flow attributes
1226  * @param[out] error
1227  *   Pointer to error structure.
1228  *
1229  * @return
1230  *   0 on success, a negative errno value otherwise and rte_errno is set.
1231  */
1232 int
1233 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1234 			      const struct rte_flow_attr *attributes,
1235 			      struct rte_flow_error *error)
1236 {
1237 	struct mlx5_priv *priv = dev->data->dev_private;
1238 	uint32_t priority_max = priv->config.flow_prio - 1;
1239 
1240 	if (attributes->group)
1241 		return rte_flow_error_set(error, ENOTSUP,
1242 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1243 					  NULL, "groups is not supported");
1244 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1245 	    attributes->priority >= priority_max)
1246 		return rte_flow_error_set(error, ENOTSUP,
1247 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1248 					  NULL, "priority out of range");
1249 	if (attributes->egress)
1250 		return rte_flow_error_set(error, ENOTSUP,
1251 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1252 					  "egress is not supported");
1253 	if (attributes->transfer && !priv->config.dv_esw_en)
1254 		return rte_flow_error_set(error, ENOTSUP,
1255 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1256 					  NULL, "transfer is not supported");
1257 	if (!attributes->ingress)
1258 		return rte_flow_error_set(error, EINVAL,
1259 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1260 					  NULL,
1261 					  "ingress attribute is mandatory");
1262 	return 0;
1263 }
1264 
1265 /**
1266  * Validate ICMP6 item.
1267  *
1268  * @param[in] item
1269  *   Item specification.
1270  * @param[in] item_flags
1271  *   Bit-fields that holds the items detected until now.
1272  * @param[out] error
1273  *   Pointer to error structure.
1274  *
1275  * @return
1276  *   0 on success, a negative errno value otherwise and rte_errno is set.
1277  */
1278 int
1279 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1280 			       uint64_t item_flags,
1281 			       uint8_t target_protocol,
1282 			       struct rte_flow_error *error)
1283 {
1284 	const struct rte_flow_item_icmp6 *mask = item->mask;
1285 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1286 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1287 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1288 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1289 				      MLX5_FLOW_LAYER_OUTER_L4;
1290 	int ret;
1291 
1292 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1293 		return rte_flow_error_set(error, EINVAL,
1294 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1295 					  "protocol filtering not compatible"
1296 					  " with ICMP6 layer");
1297 	if (!(item_flags & l3m))
1298 		return rte_flow_error_set(error, EINVAL,
1299 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1300 					  "IPv6 is mandatory to filter on"
1301 					  " ICMP6");
1302 	if (item_flags & l4m)
1303 		return rte_flow_error_set(error, EINVAL,
1304 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1305 					  "multiple L4 layers not supported");
1306 	if (!mask)
1307 		mask = &rte_flow_item_icmp6_mask;
1308 	ret = mlx5_flow_item_acceptable
1309 		(item, (const uint8_t *)mask,
1310 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1311 		 sizeof(struct rte_flow_item_icmp6), error);
1312 	if (ret < 0)
1313 		return ret;
1314 	return 0;
1315 }
1316 
1317 /**
1318  * Validate ICMP item.
1319  *
1320  * @param[in] item
1321  *   Item specification.
1322  * @param[in] item_flags
1323  *   Bit-fields that holds the items detected until now.
1324  * @param[out] error
1325  *   Pointer to error structure.
1326  *
1327  * @return
1328  *   0 on success, a negative errno value otherwise and rte_errno is set.
1329  */
1330 int
1331 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1332 			     uint64_t item_flags,
1333 			     uint8_t target_protocol,
1334 			     struct rte_flow_error *error)
1335 {
1336 	const struct rte_flow_item_icmp *mask = item->mask;
1337 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1338 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1339 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1340 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1341 				      MLX5_FLOW_LAYER_OUTER_L4;
1342 	int ret;
1343 
1344 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1345 		return rte_flow_error_set(error, EINVAL,
1346 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1347 					  "protocol filtering not compatible"
1348 					  " with ICMP layer");
1349 	if (!(item_flags & l3m))
1350 		return rte_flow_error_set(error, EINVAL,
1351 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1352 					  "IPv4 is mandatory to filter"
1353 					  " on ICMP");
1354 	if (item_flags & l4m)
1355 		return rte_flow_error_set(error, EINVAL,
1356 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1357 					  "multiple L4 layers not supported");
1358 	if (!mask)
1359 		mask = &rte_flow_item_icmp_mask;
1360 	ret = mlx5_flow_item_acceptable
1361 		(item, (const uint8_t *)mask,
1362 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1363 		 sizeof(struct rte_flow_item_icmp), error);
1364 	if (ret < 0)
1365 		return ret;
1366 	return 0;
1367 }
1368 
1369 /**
1370  * Validate Ethernet item.
1371  *
1372  * @param[in] item
1373  *   Item specification.
1374  * @param[in] item_flags
1375  *   Bit-fields that holds the items detected until now.
1376  * @param[out] error
1377  *   Pointer to error structure.
1378  *
1379  * @return
1380  *   0 on success, a negative errno value otherwise and rte_errno is set.
1381  */
1382 int
1383 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1384 			    uint64_t item_flags,
1385 			    struct rte_flow_error *error)
1386 {
1387 	const struct rte_flow_item_eth *mask = item->mask;
1388 	const struct rte_flow_item_eth nic_mask = {
1389 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1390 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1391 		.type = RTE_BE16(0xffff),
1392 	};
1393 	int ret;
1394 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1395 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1396 				       MLX5_FLOW_LAYER_OUTER_L2;
1397 
1398 	if (item_flags & ethm)
1399 		return rte_flow_error_set(error, ENOTSUP,
1400 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1401 					  "multiple L2 layers not supported");
1402 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1403 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1404 		return rte_flow_error_set(error, EINVAL,
1405 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1406 					  "L2 layer should not follow "
1407 					  "L3 layers");
1408 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1409 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1410 		return rte_flow_error_set(error, EINVAL,
1411 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1412 					  "L2 layer should not follow VLAN");
1413 	if (!mask)
1414 		mask = &rte_flow_item_eth_mask;
1415 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1416 					(const uint8_t *)&nic_mask,
1417 					sizeof(struct rte_flow_item_eth),
1418 					error);
1419 	return ret;
1420 }
1421 
1422 /**
1423  * Validate VLAN item.
1424  *
1425  * @param[in] item
1426  *   Item specification.
1427  * @param[in] item_flags
1428  *   Bit-fields that holds the items detected until now.
1429  * @param[in] dev
1430  *   Ethernet device flow is being created on.
1431  * @param[out] error
1432  *   Pointer to error structure.
1433  *
1434  * @return
1435  *   0 on success, a negative errno value otherwise and rte_errno is set.
1436  */
1437 int
1438 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1439 			     uint64_t item_flags,
1440 			     struct rte_eth_dev *dev,
1441 			     struct rte_flow_error *error)
1442 {
1443 	const struct rte_flow_item_vlan *spec = item->spec;
1444 	const struct rte_flow_item_vlan *mask = item->mask;
1445 	const struct rte_flow_item_vlan nic_mask = {
1446 		.tci = RTE_BE16(UINT16_MAX),
1447 		.inner_type = RTE_BE16(UINT16_MAX),
1448 	};
1449 	uint16_t vlan_tag = 0;
1450 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1451 	int ret;
1452 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1453 					MLX5_FLOW_LAYER_INNER_L4) :
1454 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1455 					MLX5_FLOW_LAYER_OUTER_L4);
1456 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1457 					MLX5_FLOW_LAYER_OUTER_VLAN;
1458 
1459 	if (item_flags & vlanm)
1460 		return rte_flow_error_set(error, EINVAL,
1461 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1462 					  "multiple VLAN layers not supported");
1463 	else if ((item_flags & l34m) != 0)
1464 		return rte_flow_error_set(error, EINVAL,
1465 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1466 					  "VLAN cannot follow L3/L4 layer");
1467 	if (!mask)
1468 		mask = &rte_flow_item_vlan_mask;
1469 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1470 					(const uint8_t *)&nic_mask,
1471 					sizeof(struct rte_flow_item_vlan),
1472 					error);
1473 	if (ret)
1474 		return ret;
1475 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1476 		struct mlx5_priv *priv = dev->data->dev_private;
1477 
1478 		if (priv->vmwa_context) {
1479 			/*
1480 			 * Non-NULL context means we have a virtual machine
1481 			 * and SR-IOV enabled, we have to create VLAN interface
1482 			 * to make hypervisor to setup E-Switch vport
1483 			 * context correctly. We avoid creating the multiple
1484 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1485 			 */
1486 			return rte_flow_error_set(error, EINVAL,
1487 						  RTE_FLOW_ERROR_TYPE_ITEM,
1488 						  item,
1489 						  "VLAN tag mask is not"
1490 						  " supported in virtual"
1491 						  " environment");
1492 		}
1493 	}
1494 	if (spec) {
1495 		vlan_tag = spec->tci;
1496 		vlan_tag &= mask->tci;
1497 	}
1498 	/*
1499 	 * From verbs perspective an empty VLAN is equivalent
1500 	 * to a packet without VLAN layer.
1501 	 */
1502 	if (!vlan_tag)
1503 		return rte_flow_error_set(error, EINVAL,
1504 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1505 					  item->spec,
1506 					  "VLAN cannot be empty");
1507 	return 0;
1508 }
1509 
1510 /**
1511  * Validate IPV4 item.
1512  *
1513  * @param[in] item
1514  *   Item specification.
1515  * @param[in] item_flags
1516  *   Bit-fields that holds the items detected until now.
1517  * @param[in] acc_mask
1518  *   Acceptable mask, if NULL default internal default mask
1519  *   will be used to check whether item fields are supported.
1520  * @param[out] error
1521  *   Pointer to error structure.
1522  *
1523  * @return
1524  *   0 on success, a negative errno value otherwise and rte_errno is set.
1525  */
1526 int
1527 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1528 			     uint64_t item_flags,
1529 			     uint64_t last_item,
1530 			     uint16_t ether_type,
1531 			     const struct rte_flow_item_ipv4 *acc_mask,
1532 			     struct rte_flow_error *error)
1533 {
1534 	const struct rte_flow_item_ipv4 *mask = item->mask;
1535 	const struct rte_flow_item_ipv4 *spec = item->spec;
1536 	const struct rte_flow_item_ipv4 nic_mask = {
1537 		.hdr = {
1538 			.src_addr = RTE_BE32(0xffffffff),
1539 			.dst_addr = RTE_BE32(0xffffffff),
1540 			.type_of_service = 0xff,
1541 			.next_proto_id = 0xff,
1542 		},
1543 	};
1544 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1545 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1546 				      MLX5_FLOW_LAYER_OUTER_L3;
1547 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1548 				      MLX5_FLOW_LAYER_OUTER_L4;
1549 	int ret;
1550 	uint8_t next_proto = 0xFF;
1551 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1552 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1553 				  MLX5_FLOW_LAYER_INNER_VLAN);
1554 
1555 	if ((last_item & l2_vlan) && ether_type &&
1556 	    ether_type != RTE_ETHER_TYPE_IPV4)
1557 		return rte_flow_error_set(error, EINVAL,
1558 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1559 					  "IPv4 cannot follow L2/VLAN layer "
1560 					  "which ether type is not IPv4");
1561 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1562 		if (mask && spec)
1563 			next_proto = mask->hdr.next_proto_id &
1564 				     spec->hdr.next_proto_id;
1565 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1566 			return rte_flow_error_set(error, EINVAL,
1567 						  RTE_FLOW_ERROR_TYPE_ITEM,
1568 						  item,
1569 						  "multiple tunnel "
1570 						  "not supported");
1571 	}
1572 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1573 		return rte_flow_error_set(error, EINVAL,
1574 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1575 					  "wrong tunnel type - IPv6 specified "
1576 					  "but IPv4 item provided");
1577 	if (item_flags & l3m)
1578 		return rte_flow_error_set(error, ENOTSUP,
1579 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1580 					  "multiple L3 layers not supported");
1581 	else if (item_flags & l4m)
1582 		return rte_flow_error_set(error, EINVAL,
1583 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1584 					  "L3 cannot follow an L4 layer.");
1585 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1586 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1587 		return rte_flow_error_set(error, EINVAL,
1588 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1589 					  "L3 cannot follow an NVGRE layer.");
1590 	if (!mask)
1591 		mask = &rte_flow_item_ipv4_mask;
1592 	else if (mask->hdr.next_proto_id != 0 &&
1593 		 mask->hdr.next_proto_id != 0xff)
1594 		return rte_flow_error_set(error, EINVAL,
1595 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1596 					  "partial mask is not supported"
1597 					  " for protocol");
1598 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1599 					acc_mask ? (const uint8_t *)acc_mask
1600 						 : (const uint8_t *)&nic_mask,
1601 					sizeof(struct rte_flow_item_ipv4),
1602 					error);
1603 	if (ret < 0)
1604 		return ret;
1605 	return 0;
1606 }
1607 
1608 /**
1609  * Validate IPV6 item.
1610  *
1611  * @param[in] item
1612  *   Item specification.
1613  * @param[in] item_flags
1614  *   Bit-fields that holds the items detected until now.
1615  * @param[in] acc_mask
1616  *   Acceptable mask, if NULL default internal default mask
1617  *   will be used to check whether item fields are supported.
1618  * @param[out] error
1619  *   Pointer to error structure.
1620  *
1621  * @return
1622  *   0 on success, a negative errno value otherwise and rte_errno is set.
1623  */
1624 int
1625 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1626 			     uint64_t item_flags,
1627 			     uint64_t last_item,
1628 			     uint16_t ether_type,
1629 			     const struct rte_flow_item_ipv6 *acc_mask,
1630 			     struct rte_flow_error *error)
1631 {
1632 	const struct rte_flow_item_ipv6 *mask = item->mask;
1633 	const struct rte_flow_item_ipv6 *spec = item->spec;
1634 	const struct rte_flow_item_ipv6 nic_mask = {
1635 		.hdr = {
1636 			.src_addr =
1637 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1638 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1639 			.dst_addr =
1640 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1641 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1642 			.vtc_flow = RTE_BE32(0xffffffff),
1643 			.proto = 0xff,
1644 		},
1645 	};
1646 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1647 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1648 				      MLX5_FLOW_LAYER_OUTER_L3;
1649 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1650 				      MLX5_FLOW_LAYER_OUTER_L4;
1651 	int ret;
1652 	uint8_t next_proto = 0xFF;
1653 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1654 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1655 				  MLX5_FLOW_LAYER_INNER_VLAN);
1656 
1657 	if ((last_item & l2_vlan) && ether_type &&
1658 	    ether_type != RTE_ETHER_TYPE_IPV6)
1659 		return rte_flow_error_set(error, EINVAL,
1660 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1661 					  "IPv6 cannot follow L2/VLAN layer "
1662 					  "which ether type is not IPv6");
1663 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1664 		if (mask && spec)
1665 			next_proto = mask->hdr.proto & spec->hdr.proto;
1666 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1667 			return rte_flow_error_set(error, EINVAL,
1668 						  RTE_FLOW_ERROR_TYPE_ITEM,
1669 						  item,
1670 						  "multiple tunnel "
1671 						  "not supported");
1672 	}
1673 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1674 		return rte_flow_error_set(error, EINVAL,
1675 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1676 					  "wrong tunnel type - IPv4 specified "
1677 					  "but IPv6 item provided");
1678 	if (item_flags & l3m)
1679 		return rte_flow_error_set(error, ENOTSUP,
1680 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1681 					  "multiple L3 layers not supported");
1682 	else if (item_flags & l4m)
1683 		return rte_flow_error_set(error, EINVAL,
1684 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1685 					  "L3 cannot follow an L4 layer.");
1686 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1687 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1688 		return rte_flow_error_set(error, EINVAL,
1689 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1690 					  "L3 cannot follow an NVGRE layer.");
1691 	if (!mask)
1692 		mask = &rte_flow_item_ipv6_mask;
1693 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1694 					acc_mask ? (const uint8_t *)acc_mask
1695 						 : (const uint8_t *)&nic_mask,
1696 					sizeof(struct rte_flow_item_ipv6),
1697 					error);
1698 	if (ret < 0)
1699 		return ret;
1700 	return 0;
1701 }
1702 
1703 /**
1704  * Validate UDP item.
1705  *
1706  * @param[in] item
1707  *   Item specification.
1708  * @param[in] item_flags
1709  *   Bit-fields that holds the items detected until now.
1710  * @param[in] target_protocol
1711  *   The next protocol in the previous item.
1712  * @param[in] flow_mask
1713  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1714  * @param[out] error
1715  *   Pointer to error structure.
1716  *
1717  * @return
1718  *   0 on success, a negative errno value otherwise and rte_errno is set.
1719  */
1720 int
1721 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1722 			    uint64_t item_flags,
1723 			    uint8_t target_protocol,
1724 			    struct rte_flow_error *error)
1725 {
1726 	const struct rte_flow_item_udp *mask = item->mask;
1727 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1728 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1729 				      MLX5_FLOW_LAYER_OUTER_L3;
1730 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1731 				      MLX5_FLOW_LAYER_OUTER_L4;
1732 	int ret;
1733 
1734 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1735 		return rte_flow_error_set(error, EINVAL,
1736 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1737 					  "protocol filtering not compatible"
1738 					  " with UDP layer");
1739 	if (!(item_flags & l3m))
1740 		return rte_flow_error_set(error, EINVAL,
1741 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1742 					  "L3 is mandatory to filter on L4");
1743 	if (item_flags & l4m)
1744 		return rte_flow_error_set(error, EINVAL,
1745 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1746 					  "multiple L4 layers not supported");
1747 	if (!mask)
1748 		mask = &rte_flow_item_udp_mask;
1749 	ret = mlx5_flow_item_acceptable
1750 		(item, (const uint8_t *)mask,
1751 		 (const uint8_t *)&rte_flow_item_udp_mask,
1752 		 sizeof(struct rte_flow_item_udp), error);
1753 	if (ret < 0)
1754 		return ret;
1755 	return 0;
1756 }
1757 
1758 /**
1759  * Validate TCP item.
1760  *
1761  * @param[in] item
1762  *   Item specification.
1763  * @param[in] item_flags
1764  *   Bit-fields that holds the items detected until now.
1765  * @param[in] target_protocol
1766  *   The next protocol in the previous item.
1767  * @param[out] error
1768  *   Pointer to error structure.
1769  *
1770  * @return
1771  *   0 on success, a negative errno value otherwise and rte_errno is set.
1772  */
1773 int
1774 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1775 			    uint64_t item_flags,
1776 			    uint8_t target_protocol,
1777 			    const struct rte_flow_item_tcp *flow_mask,
1778 			    struct rte_flow_error *error)
1779 {
1780 	const struct rte_flow_item_tcp *mask = item->mask;
1781 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1782 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1783 				      MLX5_FLOW_LAYER_OUTER_L3;
1784 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1785 				      MLX5_FLOW_LAYER_OUTER_L4;
1786 	int ret;
1787 
1788 	MLX5_ASSERT(flow_mask);
1789 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1790 		return rte_flow_error_set(error, EINVAL,
1791 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1792 					  "protocol filtering not compatible"
1793 					  " with TCP layer");
1794 	if (!(item_flags & l3m))
1795 		return rte_flow_error_set(error, EINVAL,
1796 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1797 					  "L3 is mandatory to filter on L4");
1798 	if (item_flags & l4m)
1799 		return rte_flow_error_set(error, EINVAL,
1800 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1801 					  "multiple L4 layers not supported");
1802 	if (!mask)
1803 		mask = &rte_flow_item_tcp_mask;
1804 	ret = mlx5_flow_item_acceptable
1805 		(item, (const uint8_t *)mask,
1806 		 (const uint8_t *)flow_mask,
1807 		 sizeof(struct rte_flow_item_tcp), error);
1808 	if (ret < 0)
1809 		return ret;
1810 	return 0;
1811 }
1812 
1813 /**
1814  * Validate VXLAN item.
1815  *
1816  * @param[in] item
1817  *   Item specification.
1818  * @param[in] item_flags
1819  *   Bit-fields that holds the items detected until now.
1820  * @param[in] target_protocol
1821  *   The next protocol in the previous item.
1822  * @param[out] error
1823  *   Pointer to error structure.
1824  *
1825  * @return
1826  *   0 on success, a negative errno value otherwise and rte_errno is set.
1827  */
1828 int
1829 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1830 			      uint64_t item_flags,
1831 			      struct rte_flow_error *error)
1832 {
1833 	const struct rte_flow_item_vxlan *spec = item->spec;
1834 	const struct rte_flow_item_vxlan *mask = item->mask;
1835 	int ret;
1836 	union vni {
1837 		uint32_t vlan_id;
1838 		uint8_t vni[4];
1839 	} id = { .vlan_id = 0, };
1840 
1841 
1842 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1843 		return rte_flow_error_set(error, ENOTSUP,
1844 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1845 					  "multiple tunnel layers not"
1846 					  " supported");
1847 	/*
1848 	 * Verify only UDPv4 is present as defined in
1849 	 * https://tools.ietf.org/html/rfc7348
1850 	 */
1851 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1852 		return rte_flow_error_set(error, EINVAL,
1853 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1854 					  "no outer UDP layer found");
1855 	if (!mask)
1856 		mask = &rte_flow_item_vxlan_mask;
1857 	ret = mlx5_flow_item_acceptable
1858 		(item, (const uint8_t *)mask,
1859 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1860 		 sizeof(struct rte_flow_item_vxlan),
1861 		 error);
1862 	if (ret < 0)
1863 		return ret;
1864 	if (spec) {
1865 		memcpy(&id.vni[1], spec->vni, 3);
1866 		memcpy(&id.vni[1], mask->vni, 3);
1867 	}
1868 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1869 		return rte_flow_error_set(error, ENOTSUP,
1870 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1871 					  "VXLAN tunnel must be fully defined");
1872 	return 0;
1873 }
1874 
1875 /**
1876  * Validate VXLAN_GPE item.
1877  *
1878  * @param[in] item
1879  *   Item specification.
1880  * @param[in] item_flags
1881  *   Bit-fields that holds the items detected until now.
1882  * @param[in] priv
1883  *   Pointer to the private data structure.
1884  * @param[in] target_protocol
1885  *   The next protocol in the previous item.
1886  * @param[out] error
1887  *   Pointer to error structure.
1888  *
1889  * @return
1890  *   0 on success, a negative errno value otherwise and rte_errno is set.
1891  */
1892 int
1893 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1894 				  uint64_t item_flags,
1895 				  struct rte_eth_dev *dev,
1896 				  struct rte_flow_error *error)
1897 {
1898 	struct mlx5_priv *priv = dev->data->dev_private;
1899 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1900 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1901 	int ret;
1902 	union vni {
1903 		uint32_t vlan_id;
1904 		uint8_t vni[4];
1905 	} id = { .vlan_id = 0, };
1906 
1907 	if (!priv->config.l3_vxlan_en)
1908 		return rte_flow_error_set(error, ENOTSUP,
1909 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1910 					  "L3 VXLAN is not enabled by device"
1911 					  " parameter and/or not configured in"
1912 					  " firmware");
1913 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1914 		return rte_flow_error_set(error, ENOTSUP,
1915 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1916 					  "multiple tunnel layers not"
1917 					  " supported");
1918 	/*
1919 	 * Verify only UDPv4 is present as defined in
1920 	 * https://tools.ietf.org/html/rfc7348
1921 	 */
1922 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1923 		return rte_flow_error_set(error, EINVAL,
1924 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1925 					  "no outer UDP layer found");
1926 	if (!mask)
1927 		mask = &rte_flow_item_vxlan_gpe_mask;
1928 	ret = mlx5_flow_item_acceptable
1929 		(item, (const uint8_t *)mask,
1930 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1931 		 sizeof(struct rte_flow_item_vxlan_gpe),
1932 		 error);
1933 	if (ret < 0)
1934 		return ret;
1935 	if (spec) {
1936 		if (spec->protocol)
1937 			return rte_flow_error_set(error, ENOTSUP,
1938 						  RTE_FLOW_ERROR_TYPE_ITEM,
1939 						  item,
1940 						  "VxLAN-GPE protocol"
1941 						  " not supported");
1942 		memcpy(&id.vni[1], spec->vni, 3);
1943 		memcpy(&id.vni[1], mask->vni, 3);
1944 	}
1945 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1946 		return rte_flow_error_set(error, ENOTSUP,
1947 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1948 					  "VXLAN-GPE tunnel must be fully"
1949 					  " defined");
1950 	return 0;
1951 }
1952 /**
1953  * Validate GRE Key item.
1954  *
1955  * @param[in] item
1956  *   Item specification.
1957  * @param[in] item_flags
1958  *   Bit flags to mark detected items.
1959  * @param[in] gre_item
1960  *   Pointer to gre_item
1961  * @param[out] error
1962  *   Pointer to error structure.
1963  *
1964  * @return
1965  *   0 on success, a negative errno value otherwise and rte_errno is set.
1966  */
1967 int
1968 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1969 				uint64_t item_flags,
1970 				const struct rte_flow_item *gre_item,
1971 				struct rte_flow_error *error)
1972 {
1973 	const rte_be32_t *mask = item->mask;
1974 	int ret = 0;
1975 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1976 	const struct rte_flow_item_gre *gre_spec;
1977 	const struct rte_flow_item_gre *gre_mask;
1978 
1979 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1980 		return rte_flow_error_set(error, ENOTSUP,
1981 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1982 					  "Multiple GRE key not support");
1983 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1984 		return rte_flow_error_set(error, ENOTSUP,
1985 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1986 					  "No preceding GRE header");
1987 	if (item_flags & MLX5_FLOW_LAYER_INNER)
1988 		return rte_flow_error_set(error, ENOTSUP,
1989 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1990 					  "GRE key following a wrong item");
1991 	gre_mask = gre_item->mask;
1992 	if (!gre_mask)
1993 		gre_mask = &rte_flow_item_gre_mask;
1994 	gre_spec = gre_item->spec;
1995 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
1996 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
1997 		return rte_flow_error_set(error, EINVAL,
1998 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1999 					  "Key bit must be on");
2000 
2001 	if (!mask)
2002 		mask = &gre_key_default_mask;
2003 	ret = mlx5_flow_item_acceptable
2004 		(item, (const uint8_t *)mask,
2005 		 (const uint8_t *)&gre_key_default_mask,
2006 		 sizeof(rte_be32_t), error);
2007 	return ret;
2008 }
2009 
2010 /**
2011  * Validate GRE item.
2012  *
2013  * @param[in] item
2014  *   Item specification.
2015  * @param[in] item_flags
2016  *   Bit flags to mark detected items.
2017  * @param[in] target_protocol
2018  *   The next protocol in the previous item.
2019  * @param[out] error
2020  *   Pointer to error structure.
2021  *
2022  * @return
2023  *   0 on success, a negative errno value otherwise and rte_errno is set.
2024  */
2025 int
2026 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2027 			    uint64_t item_flags,
2028 			    uint8_t target_protocol,
2029 			    struct rte_flow_error *error)
2030 {
2031 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2032 	const struct rte_flow_item_gre *mask = item->mask;
2033 	int ret;
2034 	const struct rte_flow_item_gre nic_mask = {
2035 		.c_rsvd0_ver = RTE_BE16(0xB000),
2036 		.protocol = RTE_BE16(UINT16_MAX),
2037 	};
2038 
2039 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2040 		return rte_flow_error_set(error, EINVAL,
2041 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2042 					  "protocol filtering not compatible"
2043 					  " with this GRE layer");
2044 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2045 		return rte_flow_error_set(error, ENOTSUP,
2046 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2047 					  "multiple tunnel layers not"
2048 					  " supported");
2049 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2050 		return rte_flow_error_set(error, ENOTSUP,
2051 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2052 					  "L3 Layer is missing");
2053 	if (!mask)
2054 		mask = &rte_flow_item_gre_mask;
2055 	ret = mlx5_flow_item_acceptable
2056 		(item, (const uint8_t *)mask,
2057 		 (const uint8_t *)&nic_mask,
2058 		 sizeof(struct rte_flow_item_gre), error);
2059 	if (ret < 0)
2060 		return ret;
2061 #ifndef HAVE_MLX5DV_DR
2062 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2063 	if (spec && (spec->protocol & mask->protocol))
2064 		return rte_flow_error_set(error, ENOTSUP,
2065 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2066 					  "without MPLS support the"
2067 					  " specification cannot be used for"
2068 					  " filtering");
2069 #endif
2070 #endif
2071 	return 0;
2072 }
2073 
2074 /**
2075  * Validate Geneve item.
2076  *
2077  * @param[in] item
2078  *   Item specification.
2079  * @param[in] itemFlags
2080  *   Bit-fields that holds the items detected until now.
2081  * @param[in] enPriv
2082  *   Pointer to the private data structure.
2083  * @param[out] error
2084  *   Pointer to error structure.
2085  *
2086  * @return
2087  *   0 on success, a negative errno value otherwise and rte_errno is set.
2088  */
2089 
2090 int
2091 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2092 			       uint64_t item_flags,
2093 			       struct rte_eth_dev *dev,
2094 			       struct rte_flow_error *error)
2095 {
2096 	struct mlx5_priv *priv = dev->data->dev_private;
2097 	const struct rte_flow_item_geneve *spec = item->spec;
2098 	const struct rte_flow_item_geneve *mask = item->mask;
2099 	int ret;
2100 	uint16_t gbhdr;
2101 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2102 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2103 	const struct rte_flow_item_geneve nic_mask = {
2104 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2105 		.vni = "\xff\xff\xff",
2106 		.protocol = RTE_BE16(UINT16_MAX),
2107 	};
2108 
2109 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2110 		return rte_flow_error_set(error, ENOTSUP,
2111 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2112 					  "L3 Geneve is not enabled by device"
2113 					  " parameter and/or not configured in"
2114 					  " firmware");
2115 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2116 		return rte_flow_error_set(error, ENOTSUP,
2117 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2118 					  "multiple tunnel layers not"
2119 					  " supported");
2120 	/*
2121 	 * Verify only UDPv4 is present as defined in
2122 	 * https://tools.ietf.org/html/rfc7348
2123 	 */
2124 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2125 		return rte_flow_error_set(error, EINVAL,
2126 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2127 					  "no outer UDP layer found");
2128 	if (!mask)
2129 		mask = &rte_flow_item_geneve_mask;
2130 	ret = mlx5_flow_item_acceptable
2131 				  (item, (const uint8_t *)mask,
2132 				   (const uint8_t *)&nic_mask,
2133 				   sizeof(struct rte_flow_item_geneve), error);
2134 	if (ret)
2135 		return ret;
2136 	if (spec) {
2137 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2138 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2139 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2140 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2141 			return rte_flow_error_set(error, ENOTSUP,
2142 						  RTE_FLOW_ERROR_TYPE_ITEM,
2143 						  item,
2144 						  "Geneve protocol unsupported"
2145 						  " fields are being used");
2146 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2147 			return rte_flow_error_set
2148 					(error, ENOTSUP,
2149 					 RTE_FLOW_ERROR_TYPE_ITEM,
2150 					 item,
2151 					 "Unsupported Geneve options length");
2152 	}
2153 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2154 		return rte_flow_error_set
2155 				    (error, ENOTSUP,
2156 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2157 				     "Geneve tunnel must be fully defined");
2158 	return 0;
2159 }
2160 
2161 /**
2162  * Validate MPLS item.
2163  *
2164  * @param[in] dev
2165  *   Pointer to the rte_eth_dev structure.
2166  * @param[in] item
2167  *   Item specification.
2168  * @param[in] item_flags
2169  *   Bit-fields that holds the items detected until now.
2170  * @param[in] prev_layer
2171  *   The protocol layer indicated in previous item.
2172  * @param[out] error
2173  *   Pointer to error structure.
2174  *
2175  * @return
2176  *   0 on success, a negative errno value otherwise and rte_errno is set.
2177  */
2178 int
2179 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2180 			     const struct rte_flow_item *item __rte_unused,
2181 			     uint64_t item_flags __rte_unused,
2182 			     uint64_t prev_layer __rte_unused,
2183 			     struct rte_flow_error *error)
2184 {
2185 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2186 	const struct rte_flow_item_mpls *mask = item->mask;
2187 	struct mlx5_priv *priv = dev->data->dev_private;
2188 	int ret;
2189 
2190 	if (!priv->config.mpls_en)
2191 		return rte_flow_error_set(error, ENOTSUP,
2192 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2193 					  "MPLS not supported or"
2194 					  " disabled in firmware"
2195 					  " configuration.");
2196 	/* MPLS over IP, UDP, GRE is allowed */
2197 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2198 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2199 			    MLX5_FLOW_LAYER_GRE)))
2200 		return rte_flow_error_set(error, EINVAL,
2201 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2202 					  "protocol filtering not compatible"
2203 					  " with MPLS layer");
2204 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2205 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2206 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2207 		return rte_flow_error_set(error, ENOTSUP,
2208 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2209 					  "multiple tunnel layers not"
2210 					  " supported");
2211 	if (!mask)
2212 		mask = &rte_flow_item_mpls_mask;
2213 	ret = mlx5_flow_item_acceptable
2214 		(item, (const uint8_t *)mask,
2215 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2216 		 sizeof(struct rte_flow_item_mpls), error);
2217 	if (ret < 0)
2218 		return ret;
2219 	return 0;
2220 #endif
2221 	return rte_flow_error_set(error, ENOTSUP,
2222 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2223 				  "MPLS is not supported by Verbs, please"
2224 				  " update.");
2225 }
2226 
2227 /**
2228  * Validate NVGRE item.
2229  *
2230  * @param[in] item
2231  *   Item specification.
2232  * @param[in] item_flags
2233  *   Bit flags to mark detected items.
2234  * @param[in] target_protocol
2235  *   The next protocol in the previous item.
2236  * @param[out] error
2237  *   Pointer to error structure.
2238  *
2239  * @return
2240  *   0 on success, a negative errno value otherwise and rte_errno is set.
2241  */
2242 int
2243 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2244 			      uint64_t item_flags,
2245 			      uint8_t target_protocol,
2246 			      struct rte_flow_error *error)
2247 {
2248 	const struct rte_flow_item_nvgre *mask = item->mask;
2249 	int ret;
2250 
2251 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2252 		return rte_flow_error_set(error, EINVAL,
2253 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2254 					  "protocol filtering not compatible"
2255 					  " with this GRE layer");
2256 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2257 		return rte_flow_error_set(error, ENOTSUP,
2258 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2259 					  "multiple tunnel layers not"
2260 					  " supported");
2261 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2262 		return rte_flow_error_set(error, ENOTSUP,
2263 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2264 					  "L3 Layer is missing");
2265 	if (!mask)
2266 		mask = &rte_flow_item_nvgre_mask;
2267 	ret = mlx5_flow_item_acceptable
2268 		(item, (const uint8_t *)mask,
2269 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2270 		 sizeof(struct rte_flow_item_nvgre), error);
2271 	if (ret < 0)
2272 		return ret;
2273 	return 0;
2274 }
2275 
2276 /* Allocate unique ID for the split Q/RSS subflows. */
2277 static uint32_t
2278 flow_qrss_get_id(struct rte_eth_dev *dev)
2279 {
2280 	struct mlx5_priv *priv = dev->data->dev_private;
2281 	uint32_t qrss_id, ret;
2282 
2283 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2284 	if (ret)
2285 		return 0;
2286 	MLX5_ASSERT(qrss_id);
2287 	return qrss_id;
2288 }
2289 
2290 /* Free unique ID for the split Q/RSS subflows. */
2291 static void
2292 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2293 {
2294 	struct mlx5_priv *priv = dev->data->dev_private;
2295 
2296 	if (qrss_id)
2297 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2298 }
2299 
2300 /**
2301  * Release resource related QUEUE/RSS action split.
2302  *
2303  * @param dev
2304  *   Pointer to Ethernet device.
2305  * @param flow
2306  *   Flow to release id's from.
2307  */
2308 static void
2309 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2310 			     struct rte_flow *flow)
2311 {
2312 	struct mlx5_flow *dev_flow;
2313 
2314 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2315 		if (dev_flow->handle.qrss_id)
2316 			flow_qrss_free_id(dev, dev_flow->handle.qrss_id);
2317 }
2318 
2319 static int
2320 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2321 		   const struct rte_flow_attr *attr __rte_unused,
2322 		   const struct rte_flow_item items[] __rte_unused,
2323 		   const struct rte_flow_action actions[] __rte_unused,
2324 		   bool external __rte_unused,
2325 		   struct rte_flow_error *error)
2326 {
2327 	return rte_flow_error_set(error, ENOTSUP,
2328 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2329 }
2330 
2331 static struct mlx5_flow *
2332 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2333 		  const struct rte_flow_item items[] __rte_unused,
2334 		  const struct rte_flow_action actions[] __rte_unused,
2335 		  struct rte_flow_error *error)
2336 {
2337 	rte_flow_error_set(error, ENOTSUP,
2338 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2339 	return NULL;
2340 }
2341 
2342 static int
2343 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2344 		    struct mlx5_flow *dev_flow __rte_unused,
2345 		    const struct rte_flow_attr *attr __rte_unused,
2346 		    const struct rte_flow_item items[] __rte_unused,
2347 		    const struct rte_flow_action actions[] __rte_unused,
2348 		    struct rte_flow_error *error)
2349 {
2350 	return rte_flow_error_set(error, ENOTSUP,
2351 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2352 }
2353 
2354 static int
2355 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2356 		struct rte_flow *flow __rte_unused,
2357 		struct rte_flow_error *error)
2358 {
2359 	return rte_flow_error_set(error, ENOTSUP,
2360 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2361 }
2362 
2363 static void
2364 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2365 		 struct rte_flow *flow __rte_unused)
2366 {
2367 }
2368 
2369 static void
2370 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2371 		  struct rte_flow *flow __rte_unused)
2372 {
2373 }
2374 
2375 static int
2376 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2377 		struct rte_flow *flow __rte_unused,
2378 		const struct rte_flow_action *actions __rte_unused,
2379 		void *data __rte_unused,
2380 		struct rte_flow_error *error)
2381 {
2382 	return rte_flow_error_set(error, ENOTSUP,
2383 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2384 }
2385 
2386 /* Void driver to protect from null pointer reference. */
2387 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2388 	.validate = flow_null_validate,
2389 	.prepare = flow_null_prepare,
2390 	.translate = flow_null_translate,
2391 	.apply = flow_null_apply,
2392 	.remove = flow_null_remove,
2393 	.destroy = flow_null_destroy,
2394 	.query = flow_null_query,
2395 };
2396 
2397 /**
2398  * Select flow driver type according to flow attributes and device
2399  * configuration.
2400  *
2401  * @param[in] dev
2402  *   Pointer to the dev structure.
2403  * @param[in] attr
2404  *   Pointer to the flow attributes.
2405  *
2406  * @return
2407  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2408  */
2409 static enum mlx5_flow_drv_type
2410 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2411 {
2412 	struct mlx5_priv *priv = dev->data->dev_private;
2413 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2414 
2415 	if (attr->transfer && priv->config.dv_esw_en)
2416 		type = MLX5_FLOW_TYPE_DV;
2417 	if (!attr->transfer)
2418 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2419 						 MLX5_FLOW_TYPE_VERBS;
2420 	return type;
2421 }
2422 
2423 #define flow_get_drv_ops(type) flow_drv_ops[type]
2424 
2425 /**
2426  * Flow driver validation API. This abstracts calling driver specific functions.
2427  * The type of flow driver is determined according to flow attributes.
2428  *
2429  * @param[in] dev
2430  *   Pointer to the dev structure.
2431  * @param[in] attr
2432  *   Pointer to the flow attributes.
2433  * @param[in] items
2434  *   Pointer to the list of items.
2435  * @param[in] actions
2436  *   Pointer to the list of actions.
2437  * @param[in] external
2438  *   This flow rule is created by request external to PMD.
2439  * @param[out] error
2440  *   Pointer to the error structure.
2441  *
2442  * @return
2443  *   0 on success, a negative errno value otherwise and rte_errno is set.
2444  */
2445 static inline int
2446 flow_drv_validate(struct rte_eth_dev *dev,
2447 		  const struct rte_flow_attr *attr,
2448 		  const struct rte_flow_item items[],
2449 		  const struct rte_flow_action actions[],
2450 		  bool external, struct rte_flow_error *error)
2451 {
2452 	const struct mlx5_flow_driver_ops *fops;
2453 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2454 
2455 	fops = flow_get_drv_ops(type);
2456 	return fops->validate(dev, attr, items, actions, external, error);
2457 }
2458 
2459 /**
2460  * Flow driver preparation API. This abstracts calling driver specific
2461  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2462  * calculates the size of memory required for device flow, allocates the memory,
2463  * initializes the device flow and returns the pointer.
2464  *
2465  * @note
2466  *   This function initializes device flow structure such as dv or verbs in
2467  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2468  *   rest. For example, adding returning device flow to flow->dev_flow list and
2469  *   setting backward reference to the flow should be done out of this function.
2470  *   layers field is not filled either.
2471  *
2472  * @param[in] attr
2473  *   Pointer to the flow attributes.
2474  * @param[in] items
2475  *   Pointer to the list of items.
2476  * @param[in] actions
2477  *   Pointer to the list of actions.
2478  * @param[out] error
2479  *   Pointer to the error structure.
2480  *
2481  * @return
2482  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2483  */
2484 static inline struct mlx5_flow *
2485 flow_drv_prepare(const struct rte_flow *flow,
2486 		 const struct rte_flow_attr *attr,
2487 		 const struct rte_flow_item items[],
2488 		 const struct rte_flow_action actions[],
2489 		 struct rte_flow_error *error)
2490 {
2491 	const struct mlx5_flow_driver_ops *fops;
2492 	enum mlx5_flow_drv_type type = flow->drv_type;
2493 
2494 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2495 	fops = flow_get_drv_ops(type);
2496 	return fops->prepare(attr, items, actions, error);
2497 }
2498 
2499 /**
2500  * Flow driver translation API. This abstracts calling driver specific
2501  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2502  * translates a generic flow into a driver flow. flow_drv_prepare() must
2503  * precede.
2504  *
2505  * @note
2506  *   dev_flow->layers could be filled as a result of parsing during translation
2507  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2508  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2509  *   flow->actions could be overwritten even though all the expanded dev_flows
2510  *   have the same actions.
2511  *
2512  * @param[in] dev
2513  *   Pointer to the rte dev structure.
2514  * @param[in, out] dev_flow
2515  *   Pointer to the mlx5 flow.
2516  * @param[in] attr
2517  *   Pointer to the flow attributes.
2518  * @param[in] items
2519  *   Pointer to the list of items.
2520  * @param[in] actions
2521  *   Pointer to the list of actions.
2522  * @param[out] error
2523  *   Pointer to the error structure.
2524  *
2525  * @return
2526  *   0 on success, a negative errno value otherwise and rte_errno is set.
2527  */
2528 static inline int
2529 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2530 		   const struct rte_flow_attr *attr,
2531 		   const struct rte_flow_item items[],
2532 		   const struct rte_flow_action actions[],
2533 		   struct rte_flow_error *error)
2534 {
2535 	const struct mlx5_flow_driver_ops *fops;
2536 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2537 
2538 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2539 	fops = flow_get_drv_ops(type);
2540 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2541 }
2542 
2543 /**
2544  * Flow driver apply API. This abstracts calling driver specific functions.
2545  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2546  * translated driver flows on to device. flow_drv_translate() must precede.
2547  *
2548  * @param[in] dev
2549  *   Pointer to Ethernet device structure.
2550  * @param[in, out] flow
2551  *   Pointer to flow structure.
2552  * @param[out] error
2553  *   Pointer to error structure.
2554  *
2555  * @return
2556  *   0 on success, a negative errno value otherwise and rte_errno is set.
2557  */
2558 static inline int
2559 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2560 	       struct rte_flow_error *error)
2561 {
2562 	const struct mlx5_flow_driver_ops *fops;
2563 	enum mlx5_flow_drv_type type = flow->drv_type;
2564 
2565 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2566 	fops = flow_get_drv_ops(type);
2567 	return fops->apply(dev, flow, error);
2568 }
2569 
2570 /**
2571  * Flow driver remove API. This abstracts calling driver specific functions.
2572  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2573  * on device. All the resources of the flow should be freed by calling
2574  * flow_drv_destroy().
2575  *
2576  * @param[in] dev
2577  *   Pointer to Ethernet device.
2578  * @param[in, out] flow
2579  *   Pointer to flow structure.
2580  */
2581 static inline void
2582 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2583 {
2584 	const struct mlx5_flow_driver_ops *fops;
2585 	enum mlx5_flow_drv_type type = flow->drv_type;
2586 
2587 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2588 	fops = flow_get_drv_ops(type);
2589 	fops->remove(dev, flow);
2590 }
2591 
2592 /**
2593  * Flow driver destroy API. This abstracts calling driver specific functions.
2594  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2595  * on device and releases resources of the flow.
2596  *
2597  * @param[in] dev
2598  *   Pointer to Ethernet device.
2599  * @param[in, out] flow
2600  *   Pointer to flow structure.
2601  */
2602 static inline void
2603 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2604 {
2605 	const struct mlx5_flow_driver_ops *fops;
2606 	enum mlx5_flow_drv_type type = flow->drv_type;
2607 
2608 	flow_mreg_split_qrss_release(dev, flow);
2609 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2610 	fops = flow_get_drv_ops(type);
2611 	fops->destroy(dev, flow);
2612 }
2613 
2614 /**
2615  * Validate a flow supported by the NIC.
2616  *
2617  * @see rte_flow_validate()
2618  * @see rte_flow_ops
2619  */
2620 int
2621 mlx5_flow_validate(struct rte_eth_dev *dev,
2622 		   const struct rte_flow_attr *attr,
2623 		   const struct rte_flow_item items[],
2624 		   const struct rte_flow_action actions[],
2625 		   struct rte_flow_error *error)
2626 {
2627 	int ret;
2628 
2629 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2630 	if (ret < 0)
2631 		return ret;
2632 	return 0;
2633 }
2634 
2635 /**
2636  * Get RSS action from the action list.
2637  *
2638  * @param[in] actions
2639  *   Pointer to the list of actions.
2640  *
2641  * @return
2642  *   Pointer to the RSS action if exist, else return NULL.
2643  */
2644 static const struct rte_flow_action_rss*
2645 flow_get_rss_action(const struct rte_flow_action actions[])
2646 {
2647 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2648 		switch (actions->type) {
2649 		case RTE_FLOW_ACTION_TYPE_RSS:
2650 			return (const struct rte_flow_action_rss *)
2651 			       actions->conf;
2652 		default:
2653 			break;
2654 		}
2655 	}
2656 	return NULL;
2657 }
2658 
2659 static unsigned int
2660 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2661 {
2662 	const struct rte_flow_item *item;
2663 	unsigned int has_vlan = 0;
2664 
2665 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2666 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2667 			has_vlan = 1;
2668 			break;
2669 		}
2670 	}
2671 	if (has_vlan)
2672 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2673 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2674 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2675 			       MLX5_EXPANSION_ROOT_OUTER;
2676 }
2677 
2678 /**
2679  *  Get layer flags from the prefix flow.
2680  *
2681  *  Some flows may be split to several subflows, the prefix subflow gets the
2682  *  match items and the suffix sub flow gets the actions.
2683  *  Some actions need the user defined match item flags to get the detail for
2684  *  the action.
2685  *  This function helps the suffix flow to get the item layer flags from prefix
2686  *  subflow.
2687  *
2688  * @param[in] dev_flow
2689  *   Pointer the created preifx subflow.
2690  *
2691  * @return
2692  *   The layers get from prefix subflow.
2693  */
2694 static inline uint64_t
2695 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2696 {
2697 	uint64_t layers = 0;
2698 
2699 	/*
2700 	 * Layers bits could be localization, but usually the compiler will
2701 	 * help to do the optimization work for source code.
2702 	 * If no decap actions, use the layers directly.
2703 	 */
2704 	if (!(dev_flow->handle.act_flags & MLX5_FLOW_ACTION_DECAP))
2705 		return dev_flow->handle.layers;
2706 	/* Convert L3 layers with decap action. */
2707 	if (dev_flow->handle.layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2708 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2709 	else if (dev_flow->handle.layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2710 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2711 	/* Convert L4 layers with decap action.  */
2712 	if (dev_flow->handle.layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2713 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2714 	else if (dev_flow->handle.layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2715 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2716 	return layers;
2717 }
2718 
2719 /**
2720  * Get metadata split action information.
2721  *
2722  * @param[in] actions
2723  *   Pointer to the list of actions.
2724  * @param[out] qrss
2725  *   Pointer to the return pointer.
2726  * @param[out] qrss_type
2727  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2728  *   if no QUEUE/RSS is found.
2729  * @param[out] encap_idx
2730  *   Pointer to the index of the encap action if exists, otherwise the last
2731  *   action index.
2732  *
2733  * @return
2734  *   Total number of actions.
2735  */
2736 static int
2737 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2738 				       const struct rte_flow_action **qrss,
2739 				       int *encap_idx)
2740 {
2741 	const struct rte_flow_action_raw_encap *raw_encap;
2742 	int actions_n = 0;
2743 	int raw_decap_idx = -1;
2744 
2745 	*encap_idx = -1;
2746 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2747 		switch (actions->type) {
2748 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2749 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2750 			*encap_idx = actions_n;
2751 			break;
2752 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2753 			raw_decap_idx = actions_n;
2754 			break;
2755 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2756 			raw_encap = actions->conf;
2757 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2758 				*encap_idx = raw_decap_idx != -1 ?
2759 						      raw_decap_idx : actions_n;
2760 			break;
2761 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2762 		case RTE_FLOW_ACTION_TYPE_RSS:
2763 			*qrss = actions;
2764 			break;
2765 		default:
2766 			break;
2767 		}
2768 		actions_n++;
2769 	}
2770 	if (*encap_idx == -1)
2771 		*encap_idx = actions_n;
2772 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2773 	return actions_n + 1;
2774 }
2775 
2776 /**
2777  * Check meter action from the action list.
2778  *
2779  * @param[in] actions
2780  *   Pointer to the list of actions.
2781  * @param[out] mtr
2782  *   Pointer to the meter exist flag.
2783  *
2784  * @return
2785  *   Total number of actions.
2786  */
2787 static int
2788 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2789 {
2790 	int actions_n = 0;
2791 
2792 	MLX5_ASSERT(mtr);
2793 	*mtr = 0;
2794 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2795 		switch (actions->type) {
2796 		case RTE_FLOW_ACTION_TYPE_METER:
2797 			*mtr = 1;
2798 			break;
2799 		default:
2800 			break;
2801 		}
2802 		actions_n++;
2803 	}
2804 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2805 	return actions_n + 1;
2806 }
2807 
2808 /**
2809  * Check if the flow should be splited due to hairpin.
2810  * The reason for the split is that in current HW we can't
2811  * support encap on Rx, so if a flow have encap we move it
2812  * to Tx.
2813  *
2814  * @param dev
2815  *   Pointer to Ethernet device.
2816  * @param[in] attr
2817  *   Flow rule attributes.
2818  * @param[in] actions
2819  *   Associated actions (list terminated by the END action).
2820  *
2821  * @return
2822  *   > 0 the number of actions and the flow should be split,
2823  *   0 when no split required.
2824  */
2825 static int
2826 flow_check_hairpin_split(struct rte_eth_dev *dev,
2827 			 const struct rte_flow_attr *attr,
2828 			 const struct rte_flow_action actions[])
2829 {
2830 	int queue_action = 0;
2831 	int action_n = 0;
2832 	int encap = 0;
2833 	const struct rte_flow_action_queue *queue;
2834 	const struct rte_flow_action_rss *rss;
2835 	const struct rte_flow_action_raw_encap *raw_encap;
2836 
2837 	if (!attr->ingress)
2838 		return 0;
2839 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2840 		switch (actions->type) {
2841 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2842 			queue = actions->conf;
2843 			if (queue == NULL)
2844 				return 0;
2845 			if (mlx5_rxq_get_type(dev, queue->index) !=
2846 			    MLX5_RXQ_TYPE_HAIRPIN)
2847 				return 0;
2848 			queue_action = 1;
2849 			action_n++;
2850 			break;
2851 		case RTE_FLOW_ACTION_TYPE_RSS:
2852 			rss = actions->conf;
2853 			if (rss == NULL || rss->queue_num == 0)
2854 				return 0;
2855 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2856 			    MLX5_RXQ_TYPE_HAIRPIN)
2857 				return 0;
2858 			queue_action = 1;
2859 			action_n++;
2860 			break;
2861 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2862 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2863 			encap = 1;
2864 			action_n++;
2865 			break;
2866 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2867 			raw_encap = actions->conf;
2868 			if (raw_encap->size >
2869 			    (sizeof(struct rte_flow_item_eth) +
2870 			     sizeof(struct rte_flow_item_ipv4)))
2871 				encap = 1;
2872 			action_n++;
2873 			break;
2874 		default:
2875 			action_n++;
2876 			break;
2877 		}
2878 	}
2879 	if (encap == 1 && queue_action)
2880 		return action_n;
2881 	return 0;
2882 }
2883 
2884 /* Declare flow create/destroy prototype in advance. */
2885 static struct rte_flow *
2886 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2887 		 const struct rte_flow_attr *attr,
2888 		 const struct rte_flow_item items[],
2889 		 const struct rte_flow_action actions[],
2890 		 bool external, struct rte_flow_error *error);
2891 
2892 static void
2893 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2894 		  struct rte_flow *flow);
2895 
2896 /**
2897  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2898  *
2899  * As mark_id is unique, if there's already a registered flow for the mark_id,
2900  * return by increasing the reference counter of the resource. Otherwise, create
2901  * the resource (mcp_res) and flow.
2902  *
2903  * Flow looks like,
2904  *   - If ingress port is ANY and reg_c[1] is mark_id,
2905  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2906  *
2907  * For default flow (zero mark_id), flow is like,
2908  *   - If ingress port is ANY,
2909  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2910  *
2911  * @param dev
2912  *   Pointer to Ethernet device.
2913  * @param mark_id
2914  *   ID of MARK action, zero means default flow for META.
2915  * @param[out] error
2916  *   Perform verbose error reporting if not NULL.
2917  *
2918  * @return
2919  *   Associated resource on success, NULL otherwise and rte_errno is set.
2920  */
2921 static struct mlx5_flow_mreg_copy_resource *
2922 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2923 			  struct rte_flow_error *error)
2924 {
2925 	struct mlx5_priv *priv = dev->data->dev_private;
2926 	struct rte_flow_attr attr = {
2927 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2928 		.ingress = 1,
2929 	};
2930 	struct mlx5_rte_flow_item_tag tag_spec = {
2931 		.data = mark_id,
2932 	};
2933 	struct rte_flow_item items[] = {
2934 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2935 	};
2936 	struct rte_flow_action_mark ftag = {
2937 		.id = mark_id,
2938 	};
2939 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2940 		.dst = REG_B,
2941 		.src = 0,
2942 	};
2943 	struct rte_flow_action_jump jump = {
2944 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2945 	};
2946 	struct rte_flow_action actions[] = {
2947 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2948 	};
2949 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2950 	int ret;
2951 
2952 	/* Fill the register fileds in the flow. */
2953 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2954 	if (ret < 0)
2955 		return NULL;
2956 	tag_spec.id = ret;
2957 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2958 	if (ret < 0)
2959 		return NULL;
2960 	cp_mreg.src = ret;
2961 	/* Check if already registered. */
2962 	MLX5_ASSERT(priv->mreg_cp_tbl);
2963 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2964 	if (mcp_res) {
2965 		/* For non-default rule. */
2966 		if (mark_id != MLX5_DEFAULT_COPY_ID)
2967 			mcp_res->refcnt++;
2968 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2969 			    mcp_res->refcnt == 1);
2970 		return mcp_res;
2971 	}
2972 	/* Provide the full width of FLAG specific value. */
2973 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2974 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2975 	/* Build a new flow. */
2976 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
2977 		items[0] = (struct rte_flow_item){
2978 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2979 			.spec = &tag_spec,
2980 		};
2981 		items[1] = (struct rte_flow_item){
2982 			.type = RTE_FLOW_ITEM_TYPE_END,
2983 		};
2984 		actions[0] = (struct rte_flow_action){
2985 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2986 			.conf = &ftag,
2987 		};
2988 		actions[1] = (struct rte_flow_action){
2989 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2990 			.conf = &cp_mreg,
2991 		};
2992 		actions[2] = (struct rte_flow_action){
2993 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2994 			.conf = &jump,
2995 		};
2996 		actions[3] = (struct rte_flow_action){
2997 			.type = RTE_FLOW_ACTION_TYPE_END,
2998 		};
2999 	} else {
3000 		/* Default rule, wildcard match. */
3001 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3002 		items[0] = (struct rte_flow_item){
3003 			.type = RTE_FLOW_ITEM_TYPE_END,
3004 		};
3005 		actions[0] = (struct rte_flow_action){
3006 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3007 			.conf = &cp_mreg,
3008 		};
3009 		actions[1] = (struct rte_flow_action){
3010 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3011 			.conf = &jump,
3012 		};
3013 		actions[2] = (struct rte_flow_action){
3014 			.type = RTE_FLOW_ACTION_TYPE_END,
3015 		};
3016 	}
3017 	/* Build a new entry. */
3018 	mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3019 	if (!mcp_res) {
3020 		rte_errno = ENOMEM;
3021 		return NULL;
3022 	}
3023 	/*
3024 	 * The copy Flows are not included in any list. There
3025 	 * ones are referenced from other Flows and can not
3026 	 * be applied, removed, deleted in ardbitrary order
3027 	 * by list traversing.
3028 	 */
3029 	mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3030 					 actions, false, error);
3031 	if (!mcp_res->flow)
3032 		goto error;
3033 	mcp_res->refcnt++;
3034 	mcp_res->hlist_ent.key = mark_id;
3035 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3036 				&mcp_res->hlist_ent);
3037 	MLX5_ASSERT(!ret);
3038 	if (ret)
3039 		goto error;
3040 	return mcp_res;
3041 error:
3042 	if (mcp_res->flow)
3043 		flow_list_destroy(dev, NULL, mcp_res->flow);
3044 	rte_free(mcp_res);
3045 	return NULL;
3046 }
3047 
3048 /**
3049  * Release flow in RX_CP_TBL.
3050  *
3051  * @param dev
3052  *   Pointer to Ethernet device.
3053  * @flow
3054  *   Parent flow for wich copying is provided.
3055  */
3056 static void
3057 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3058 			  struct rte_flow *flow)
3059 {
3060 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3061 	struct mlx5_priv *priv = dev->data->dev_private;
3062 
3063 	if (!mcp_res || !priv->mreg_cp_tbl)
3064 		return;
3065 	if (flow->copy_applied) {
3066 		MLX5_ASSERT(mcp_res->appcnt);
3067 		flow->copy_applied = 0;
3068 		--mcp_res->appcnt;
3069 		if (!mcp_res->appcnt)
3070 			flow_drv_remove(dev, mcp_res->flow);
3071 	}
3072 	/*
3073 	 * We do not check availability of metadata registers here,
3074 	 * because copy resources are not allocated in this case.
3075 	 */
3076 	if (--mcp_res->refcnt)
3077 		return;
3078 	MLX5_ASSERT(mcp_res->flow);
3079 	flow_list_destroy(dev, NULL, mcp_res->flow);
3080 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3081 	rte_free(mcp_res);
3082 	flow->mreg_copy = NULL;
3083 }
3084 
3085 /**
3086  * Start flow in RX_CP_TBL.
3087  *
3088  * @param dev
3089  *   Pointer to Ethernet device.
3090  * @flow
3091  *   Parent flow for wich copying is provided.
3092  *
3093  * @return
3094  *   0 on success, a negative errno value otherwise and rte_errno is set.
3095  */
3096 static int
3097 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3098 			    struct rte_flow *flow)
3099 {
3100 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3101 	int ret;
3102 
3103 	if (!mcp_res || flow->copy_applied)
3104 		return 0;
3105 	if (!mcp_res->appcnt) {
3106 		ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3107 		if (ret)
3108 			return ret;
3109 	}
3110 	++mcp_res->appcnt;
3111 	flow->copy_applied = 1;
3112 	return 0;
3113 }
3114 
3115 /**
3116  * Stop flow in RX_CP_TBL.
3117  *
3118  * @param dev
3119  *   Pointer to Ethernet device.
3120  * @flow
3121  *   Parent flow for wich copying is provided.
3122  */
3123 static void
3124 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3125 			   struct rte_flow *flow)
3126 {
3127 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3128 
3129 	if (!mcp_res || !flow->copy_applied)
3130 		return;
3131 	MLX5_ASSERT(mcp_res->appcnt);
3132 	--mcp_res->appcnt;
3133 	flow->copy_applied = 0;
3134 	if (!mcp_res->appcnt)
3135 		flow_drv_remove(dev, mcp_res->flow);
3136 }
3137 
3138 /**
3139  * Remove the default copy action from RX_CP_TBL.
3140  *
3141  * @param dev
3142  *   Pointer to Ethernet device.
3143  */
3144 static void
3145 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3146 {
3147 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3148 	struct mlx5_priv *priv = dev->data->dev_private;
3149 
3150 	/* Check if default flow is registered. */
3151 	if (!priv->mreg_cp_tbl)
3152 		return;
3153 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3154 					    MLX5_DEFAULT_COPY_ID);
3155 	if (!mcp_res)
3156 		return;
3157 	MLX5_ASSERT(mcp_res->flow);
3158 	flow_list_destroy(dev, NULL, mcp_res->flow);
3159 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3160 	rte_free(mcp_res);
3161 }
3162 
3163 /**
3164  * Add the default copy action in in RX_CP_TBL.
3165  *
3166  * @param dev
3167  *   Pointer to Ethernet device.
3168  * @param[out] error
3169  *   Perform verbose error reporting if not NULL.
3170  *
3171  * @return
3172  *   0 for success, negative value otherwise and rte_errno is set.
3173  */
3174 static int
3175 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3176 				  struct rte_flow_error *error)
3177 {
3178 	struct mlx5_priv *priv = dev->data->dev_private;
3179 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3180 
3181 	/* Check whether extensive metadata feature is engaged. */
3182 	if (!priv->config.dv_flow_en ||
3183 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3184 	    !mlx5_flow_ext_mreg_supported(dev) ||
3185 	    !priv->sh->dv_regc0_mask)
3186 		return 0;
3187 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3188 	if (!mcp_res)
3189 		return -rte_errno;
3190 	return 0;
3191 }
3192 
3193 /**
3194  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3195  *
3196  * All the flow having Q/RSS action should be split by
3197  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3198  * performs the following,
3199  *   - CQE->flow_tag := reg_c[1] (MARK)
3200  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3201  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3202  * but there should be a flow per each MARK ID set by MARK action.
3203  *
3204  * For the aforementioned reason, if there's a MARK action in flow's action
3205  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3206  * the MARK ID to CQE's flow_tag like,
3207  *   - If reg_c[1] is mark_id,
3208  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3209  *
3210  * For SET_META action which stores value in reg_c[0], as the destination is
3211  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3212  * MARK ID means the default flow. The default flow looks like,
3213  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3214  *
3215  * @param dev
3216  *   Pointer to Ethernet device.
3217  * @param flow
3218  *   Pointer to flow structure.
3219  * @param[in] actions
3220  *   Pointer to the list of actions.
3221  * @param[out] error
3222  *   Perform verbose error reporting if not NULL.
3223  *
3224  * @return
3225  *   0 on success, negative value otherwise and rte_errno is set.
3226  */
3227 static int
3228 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3229 			    struct rte_flow *flow,
3230 			    const struct rte_flow_action *actions,
3231 			    struct rte_flow_error *error)
3232 {
3233 	struct mlx5_priv *priv = dev->data->dev_private;
3234 	struct mlx5_dev_config *config = &priv->config;
3235 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3236 	const struct rte_flow_action_mark *mark;
3237 
3238 	/* Check whether extensive metadata feature is engaged. */
3239 	if (!config->dv_flow_en ||
3240 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3241 	    !mlx5_flow_ext_mreg_supported(dev) ||
3242 	    !priv->sh->dv_regc0_mask)
3243 		return 0;
3244 	/* Find MARK action. */
3245 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3246 		switch (actions->type) {
3247 		case RTE_FLOW_ACTION_TYPE_FLAG:
3248 			mcp_res = flow_mreg_add_copy_action
3249 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3250 			if (!mcp_res)
3251 				return -rte_errno;
3252 			flow->mreg_copy = mcp_res;
3253 			if (dev->data->dev_started) {
3254 				mcp_res->appcnt++;
3255 				flow->copy_applied = 1;
3256 			}
3257 			return 0;
3258 		case RTE_FLOW_ACTION_TYPE_MARK:
3259 			mark = (const struct rte_flow_action_mark *)
3260 				actions->conf;
3261 			mcp_res =
3262 				flow_mreg_add_copy_action(dev, mark->id, error);
3263 			if (!mcp_res)
3264 				return -rte_errno;
3265 			flow->mreg_copy = mcp_res;
3266 			if (dev->data->dev_started) {
3267 				mcp_res->appcnt++;
3268 				flow->copy_applied = 1;
3269 			}
3270 			return 0;
3271 		default:
3272 			break;
3273 		}
3274 	}
3275 	return 0;
3276 }
3277 
3278 #define MLX5_MAX_SPLIT_ACTIONS 24
3279 #define MLX5_MAX_SPLIT_ITEMS 24
3280 
3281 /**
3282  * Split the hairpin flow.
3283  * Since HW can't support encap on Rx we move the encap to Tx.
3284  * If the count action is after the encap then we also
3285  * move the count action. in this case the count will also measure
3286  * the outer bytes.
3287  *
3288  * @param dev
3289  *   Pointer to Ethernet device.
3290  * @param[in] actions
3291  *   Associated actions (list terminated by the END action).
3292  * @param[out] actions_rx
3293  *   Rx flow actions.
3294  * @param[out] actions_tx
3295  *   Tx flow actions..
3296  * @param[out] pattern_tx
3297  *   The pattern items for the Tx flow.
3298  * @param[out] flow_id
3299  *   The flow ID connected to this flow.
3300  *
3301  * @return
3302  *   0 on success.
3303  */
3304 static int
3305 flow_hairpin_split(struct rte_eth_dev *dev,
3306 		   const struct rte_flow_action actions[],
3307 		   struct rte_flow_action actions_rx[],
3308 		   struct rte_flow_action actions_tx[],
3309 		   struct rte_flow_item pattern_tx[],
3310 		   uint32_t *flow_id)
3311 {
3312 	struct mlx5_priv *priv = dev->data->dev_private;
3313 	const struct rte_flow_action_raw_encap *raw_encap;
3314 	const struct rte_flow_action_raw_decap *raw_decap;
3315 	struct mlx5_rte_flow_action_set_tag *set_tag;
3316 	struct rte_flow_action *tag_action;
3317 	struct mlx5_rte_flow_item_tag *tag_item;
3318 	struct rte_flow_item *item;
3319 	char *addr;
3320 	int encap = 0;
3321 
3322 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3323 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3324 		switch (actions->type) {
3325 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3326 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3327 			rte_memcpy(actions_tx, actions,
3328 			       sizeof(struct rte_flow_action));
3329 			actions_tx++;
3330 			break;
3331 		case RTE_FLOW_ACTION_TYPE_COUNT:
3332 			if (encap) {
3333 				rte_memcpy(actions_tx, actions,
3334 					   sizeof(struct rte_flow_action));
3335 				actions_tx++;
3336 			} else {
3337 				rte_memcpy(actions_rx, actions,
3338 					   sizeof(struct rte_flow_action));
3339 				actions_rx++;
3340 			}
3341 			break;
3342 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3343 			raw_encap = actions->conf;
3344 			if (raw_encap->size >
3345 			    (sizeof(struct rte_flow_item_eth) +
3346 			     sizeof(struct rte_flow_item_ipv4))) {
3347 				memcpy(actions_tx, actions,
3348 				       sizeof(struct rte_flow_action));
3349 				actions_tx++;
3350 				encap = 1;
3351 			} else {
3352 				rte_memcpy(actions_rx, actions,
3353 					   sizeof(struct rte_flow_action));
3354 				actions_rx++;
3355 			}
3356 			break;
3357 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3358 			raw_decap = actions->conf;
3359 			if (raw_decap->size <
3360 			    (sizeof(struct rte_flow_item_eth) +
3361 			     sizeof(struct rte_flow_item_ipv4))) {
3362 				memcpy(actions_tx, actions,
3363 				       sizeof(struct rte_flow_action));
3364 				actions_tx++;
3365 			} else {
3366 				rte_memcpy(actions_rx, actions,
3367 					   sizeof(struct rte_flow_action));
3368 				actions_rx++;
3369 			}
3370 			break;
3371 		default:
3372 			rte_memcpy(actions_rx, actions,
3373 				   sizeof(struct rte_flow_action));
3374 			actions_rx++;
3375 			break;
3376 		}
3377 	}
3378 	/* Add set meta action and end action for the Rx flow. */
3379 	tag_action = actions_rx;
3380 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3381 	actions_rx++;
3382 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3383 	actions_rx++;
3384 	set_tag = (void *)actions_rx;
3385 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3386 	MLX5_ASSERT(set_tag->id > REG_NONE);
3387 	set_tag->data = *flow_id;
3388 	tag_action->conf = set_tag;
3389 	/* Create Tx item list. */
3390 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3391 	addr = (void *)&pattern_tx[2];
3392 	item = pattern_tx;
3393 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3394 	tag_item = (void *)addr;
3395 	tag_item->data = *flow_id;
3396 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3397 	MLX5_ASSERT(set_tag->id > REG_NONE);
3398 	item->spec = tag_item;
3399 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3400 	tag_item = (void *)addr;
3401 	tag_item->data = UINT32_MAX;
3402 	tag_item->id = UINT16_MAX;
3403 	item->mask = tag_item;
3404 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3405 	item->last = NULL;
3406 	item++;
3407 	item->type = RTE_FLOW_ITEM_TYPE_END;
3408 	return 0;
3409 }
3410 
3411 /**
3412  * The last stage of splitting chain, just creates the subflow
3413  * without any modification.
3414  *
3415  * @param dev
3416  *   Pointer to Ethernet device.
3417  * @param[in] flow
3418  *   Parent flow structure pointer.
3419  * @param[in, out] sub_flow
3420  *   Pointer to return the created subflow, may be NULL.
3421  * @param[in] prefix_layers
3422  *   Prefix subflow layers, may be 0.
3423  * @param[in] attr
3424  *   Flow rule attributes.
3425  * @param[in] items
3426  *   Pattern specification (list terminated by the END pattern item).
3427  * @param[in] actions
3428  *   Associated actions (list terminated by the END action).
3429  * @param[in] external
3430  *   This flow rule is created by request external to PMD.
3431  * @param[out] error
3432  *   Perform verbose error reporting if not NULL.
3433  * @return
3434  *   0 on success, negative value otherwise
3435  */
3436 static int
3437 flow_create_split_inner(struct rte_eth_dev *dev,
3438 			struct rte_flow *flow,
3439 			struct mlx5_flow **sub_flow,
3440 			uint64_t prefix_layers,
3441 			const struct rte_flow_attr *attr,
3442 			const struct rte_flow_item items[],
3443 			const struct rte_flow_action actions[],
3444 			bool external, struct rte_flow_error *error)
3445 {
3446 	struct mlx5_flow *dev_flow;
3447 
3448 	dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3449 	if (!dev_flow)
3450 		return -rte_errno;
3451 	dev_flow->flow = flow;
3452 	dev_flow->external = external;
3453 	/* Subflow object was created, we must include one in the list. */
3454 	LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3455 	/*
3456 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3457 	 * flow may need some user defined item layer flags.
3458 	 */
3459 	if (prefix_layers)
3460 		dev_flow->handle.layers = prefix_layers;
3461 	if (sub_flow)
3462 		*sub_flow = dev_flow;
3463 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3464 }
3465 
3466 /**
3467  * Split the meter flow.
3468  *
3469  * As meter flow will split to three sub flow, other than meter
3470  * action, the other actions make sense to only meter accepts
3471  * the packet. If it need to be dropped, no other additional
3472  * actions should be take.
3473  *
3474  * One kind of special action which decapsulates the L3 tunnel
3475  * header will be in the prefix sub flow, as not to take the
3476  * L3 tunnel header into account.
3477  *
3478  * @param dev
3479  *   Pointer to Ethernet device.
3480  * @param[in] items
3481  *   Pattern specification (list terminated by the END pattern item).
3482  * @param[out] sfx_items
3483  *   Suffix flow match items (list terminated by the END pattern item).
3484  * @param[in] actions
3485  *   Associated actions (list terminated by the END action).
3486  * @param[out] actions_sfx
3487  *   Suffix flow actions.
3488  * @param[out] actions_pre
3489  *   Prefix flow actions.
3490  * @param[out] pattern_sfx
3491  *   The pattern items for the suffix flow.
3492  * @param[out] tag_sfx
3493  *   Pointer to suffix flow tag.
3494  *
3495  * @return
3496  *   0 on success.
3497  */
3498 static int
3499 flow_meter_split_prep(struct rte_eth_dev *dev,
3500 		 const struct rte_flow_item items[],
3501 		 struct rte_flow_item sfx_items[],
3502 		 const struct rte_flow_action actions[],
3503 		 struct rte_flow_action actions_sfx[],
3504 		 struct rte_flow_action actions_pre[])
3505 {
3506 	struct rte_flow_action *tag_action = NULL;
3507 	struct rte_flow_item *tag_item;
3508 	struct mlx5_rte_flow_action_set_tag *set_tag;
3509 	struct rte_flow_error error;
3510 	const struct rte_flow_action_raw_encap *raw_encap;
3511 	const struct rte_flow_action_raw_decap *raw_decap;
3512 	struct mlx5_rte_flow_item_tag *tag_spec;
3513 	struct mlx5_rte_flow_item_tag *tag_mask;
3514 	uint32_t tag_id;
3515 	bool copy_vlan = false;
3516 
3517 	/* Prepare the actions for prefix and suffix flow. */
3518 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3519 		struct rte_flow_action **action_cur = NULL;
3520 
3521 		switch (actions->type) {
3522 		case RTE_FLOW_ACTION_TYPE_METER:
3523 			/* Add the extra tag action first. */
3524 			tag_action = actions_pre;
3525 			tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3526 			actions_pre++;
3527 			action_cur = &actions_pre;
3528 			break;
3529 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3530 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3531 			action_cur = &actions_pre;
3532 			break;
3533 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3534 			raw_encap = actions->conf;
3535 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3536 				action_cur = &actions_pre;
3537 			break;
3538 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3539 			raw_decap = actions->conf;
3540 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3541 				action_cur = &actions_pre;
3542 			break;
3543 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3544 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3545 			copy_vlan = true;
3546 			break;
3547 		default:
3548 			break;
3549 		}
3550 		if (!action_cur)
3551 			action_cur = &actions_sfx;
3552 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3553 		(*action_cur)++;
3554 	}
3555 	/* Add end action to the actions. */
3556 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3557 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3558 	actions_pre++;
3559 	/* Set the tag. */
3560 	set_tag = (void *)actions_pre;
3561 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3562 	/*
3563 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3564 	 */
3565 	tag_id = flow_qrss_get_id(dev);
3566 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3567 	assert(tag_action);
3568 	tag_action->conf = set_tag;
3569 	/* Prepare the suffix subflow items. */
3570 	tag_item = sfx_items++;
3571 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3572 		int item_type = items->type;
3573 
3574 		switch (item_type) {
3575 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3576 			memcpy(sfx_items, items, sizeof(*sfx_items));
3577 			sfx_items++;
3578 			break;
3579 		case RTE_FLOW_ITEM_TYPE_VLAN:
3580 			if (copy_vlan) {
3581 				memcpy(sfx_items, items, sizeof(*sfx_items));
3582 				/*
3583 				 * Convert to internal match item, it is used
3584 				 * for vlan push and set vid.
3585 				 */
3586 				sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3587 				sfx_items++;
3588 			}
3589 			break;
3590 		default:
3591 			break;
3592 		}
3593 	}
3594 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3595 	sfx_items++;
3596 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3597 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3598 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3599 	tag_mask = tag_spec + 1;
3600 	tag_mask->data = 0xffffff00;
3601 	tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3602 	tag_item->spec = tag_spec;
3603 	tag_item->last = NULL;
3604 	tag_item->mask = tag_mask;
3605 	return tag_id;
3606 }
3607 
3608 /**
3609  * Split action list having QUEUE/RSS for metadata register copy.
3610  *
3611  * Once Q/RSS action is detected in user's action list, the flow action
3612  * should be split in order to copy metadata registers, which will happen in
3613  * RX_CP_TBL like,
3614  *   - CQE->flow_tag := reg_c[1] (MARK)
3615  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3616  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3617  * This is because the last action of each flow must be a terminal action
3618  * (QUEUE, RSS or DROP).
3619  *
3620  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3621  * stored and kept in the mlx5_flow structure per each sub_flow.
3622  *
3623  * The Q/RSS action is replaced with,
3624  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3625  * And the following JUMP action is added at the end,
3626  *   - JUMP, to RX_CP_TBL.
3627  *
3628  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3629  * flow_create_split_metadata() routine. The flow will look like,
3630  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3631  *
3632  * @param dev
3633  *   Pointer to Ethernet device.
3634  * @param[out] split_actions
3635  *   Pointer to store split actions to jump to CP_TBL.
3636  * @param[in] actions
3637  *   Pointer to the list of original flow actions.
3638  * @param[in] qrss
3639  *   Pointer to the Q/RSS action.
3640  * @param[in] actions_n
3641  *   Number of original actions.
3642  * @param[out] error
3643  *   Perform verbose error reporting if not NULL.
3644  *
3645  * @return
3646  *   non-zero unique flow_id on success, otherwise 0 and
3647  *   error/rte_error are set.
3648  */
3649 static uint32_t
3650 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3651 			  struct rte_flow_action *split_actions,
3652 			  const struct rte_flow_action *actions,
3653 			  const struct rte_flow_action *qrss,
3654 			  int actions_n, struct rte_flow_error *error)
3655 {
3656 	struct mlx5_rte_flow_action_set_tag *set_tag;
3657 	struct rte_flow_action_jump *jump;
3658 	const int qrss_idx = qrss - actions;
3659 	uint32_t flow_id = 0;
3660 	int ret = 0;
3661 
3662 	/*
3663 	 * Given actions will be split
3664 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3665 	 * - Add jump to mreg CP_TBL.
3666 	 * As a result, there will be one more action.
3667 	 */
3668 	++actions_n;
3669 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3670 	set_tag = (void *)(split_actions + actions_n);
3671 	/*
3672 	 * If tag action is not set to void(it means we are not the meter
3673 	 * suffix flow), add the tag action. Since meter suffix flow already
3674 	 * has the tag added.
3675 	 */
3676 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3677 		/*
3678 		 * Allocate the new subflow ID. This one is unique within
3679 		 * device and not shared with representors. Otherwise,
3680 		 * we would have to resolve multi-thread access synch
3681 		 * issue. Each flow on the shared device is appended
3682 		 * with source vport identifier, so the resulting
3683 		 * flows will be unique in the shared (by master and
3684 		 * representors) domain even if they have coinciding
3685 		 * IDs.
3686 		 */
3687 		flow_id = flow_qrss_get_id(dev);
3688 		if (!flow_id)
3689 			return rte_flow_error_set(error, ENOMEM,
3690 						  RTE_FLOW_ERROR_TYPE_ACTION,
3691 						  NULL, "can't allocate id "
3692 						  "for split Q/RSS subflow");
3693 		/* Internal SET_TAG action to set flow ID. */
3694 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3695 			.data = flow_id,
3696 		};
3697 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3698 		if (ret < 0)
3699 			return ret;
3700 		set_tag->id = ret;
3701 		/* Construct new actions array. */
3702 		/* Replace QUEUE/RSS action. */
3703 		split_actions[qrss_idx] = (struct rte_flow_action){
3704 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3705 			.conf = set_tag,
3706 		};
3707 	}
3708 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3709 	jump = (void *)(set_tag + 1);
3710 	*jump = (struct rte_flow_action_jump){
3711 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3712 	};
3713 	split_actions[actions_n - 2] = (struct rte_flow_action){
3714 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3715 		.conf = jump,
3716 	};
3717 	split_actions[actions_n - 1] = (struct rte_flow_action){
3718 		.type = RTE_FLOW_ACTION_TYPE_END,
3719 	};
3720 	return flow_id;
3721 }
3722 
3723 /**
3724  * Extend the given action list for Tx metadata copy.
3725  *
3726  * Copy the given action list to the ext_actions and add flow metadata register
3727  * copy action in order to copy reg_a set by WQE to reg_c[0].
3728  *
3729  * @param[out] ext_actions
3730  *   Pointer to the extended action list.
3731  * @param[in] actions
3732  *   Pointer to the list of actions.
3733  * @param[in] actions_n
3734  *   Number of actions in the list.
3735  * @param[out] error
3736  *   Perform verbose error reporting if not NULL.
3737  * @param[in] encap_idx
3738  *   The encap action inndex.
3739  *
3740  * @return
3741  *   0 on success, negative value otherwise
3742  */
3743 static int
3744 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3745 		       struct rte_flow_action *ext_actions,
3746 		       const struct rte_flow_action *actions,
3747 		       int actions_n, struct rte_flow_error *error,
3748 		       int encap_idx)
3749 {
3750 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3751 		(struct mlx5_flow_action_copy_mreg *)
3752 			(ext_actions + actions_n + 1);
3753 	int ret;
3754 
3755 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3756 	if (ret < 0)
3757 		return ret;
3758 	cp_mreg->dst = ret;
3759 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3760 	if (ret < 0)
3761 		return ret;
3762 	cp_mreg->src = ret;
3763 	if (encap_idx != 0)
3764 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3765 	if (encap_idx == actions_n - 1) {
3766 		ext_actions[actions_n - 1] = (struct rte_flow_action){
3767 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3768 			.conf = cp_mreg,
3769 		};
3770 		ext_actions[actions_n] = (struct rte_flow_action){
3771 			.type = RTE_FLOW_ACTION_TYPE_END,
3772 		};
3773 	} else {
3774 		ext_actions[encap_idx] = (struct rte_flow_action){
3775 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3776 			.conf = cp_mreg,
3777 		};
3778 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3779 				sizeof(*ext_actions) * (actions_n - encap_idx));
3780 	}
3781 	return 0;
3782 }
3783 
3784 /**
3785  * The splitting for metadata feature.
3786  *
3787  * - Q/RSS action on NIC Rx should be split in order to pass by
3788  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3789  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3790  *
3791  * - All the actions on NIC Tx should have a mreg copy action to
3792  *   copy reg_a from WQE to reg_c[0].
3793  *
3794  * @param dev
3795  *   Pointer to Ethernet device.
3796  * @param[in] flow
3797  *   Parent flow structure pointer.
3798  * @param[in] prefix_layers
3799  *   Prefix flow layer flags.
3800  * @param[in] attr
3801  *   Flow rule attributes.
3802  * @param[in] items
3803  *   Pattern specification (list terminated by the END pattern item).
3804  * @param[in] actions
3805  *   Associated actions (list terminated by the END action).
3806  * @param[in] external
3807  *   This flow rule is created by request external to PMD.
3808  * @param[out] error
3809  *   Perform verbose error reporting if not NULL.
3810  * @return
3811  *   0 on success, negative value otherwise
3812  */
3813 static int
3814 flow_create_split_metadata(struct rte_eth_dev *dev,
3815 			   struct rte_flow *flow,
3816 			   uint64_t prefix_layers,
3817 			   const struct rte_flow_attr *attr,
3818 			   const struct rte_flow_item items[],
3819 			   const struct rte_flow_action actions[],
3820 			   bool external, struct rte_flow_error *error)
3821 {
3822 	struct mlx5_priv *priv = dev->data->dev_private;
3823 	struct mlx5_dev_config *config = &priv->config;
3824 	const struct rte_flow_action *qrss = NULL;
3825 	struct rte_flow_action *ext_actions = NULL;
3826 	struct mlx5_flow *dev_flow = NULL;
3827 	uint32_t qrss_id = 0;
3828 	int mtr_sfx = 0;
3829 	size_t act_size;
3830 	int actions_n;
3831 	int encap_idx;
3832 	int ret;
3833 
3834 	/* Check whether extensive metadata feature is engaged. */
3835 	if (!config->dv_flow_en ||
3836 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3837 	    !mlx5_flow_ext_mreg_supported(dev))
3838 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3839 					       attr, items, actions, external,
3840 					       error);
3841 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3842 							   &encap_idx);
3843 	if (qrss) {
3844 		/* Exclude hairpin flows from splitting. */
3845 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3846 			const struct rte_flow_action_queue *queue;
3847 
3848 			queue = qrss->conf;
3849 			if (mlx5_rxq_get_type(dev, queue->index) ==
3850 			    MLX5_RXQ_TYPE_HAIRPIN)
3851 				qrss = NULL;
3852 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3853 			const struct rte_flow_action_rss *rss;
3854 
3855 			rss = qrss->conf;
3856 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3857 			    MLX5_RXQ_TYPE_HAIRPIN)
3858 				qrss = NULL;
3859 		}
3860 	}
3861 	if (qrss) {
3862 		/* Check if it is in meter suffix table. */
3863 		mtr_sfx = attr->group == (attr->transfer ?
3864 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3865 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3866 		/*
3867 		 * Q/RSS action on NIC Rx should be split in order to pass by
3868 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3869 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3870 		 */
3871 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3872 			   sizeof(struct rte_flow_action_set_tag) +
3873 			   sizeof(struct rte_flow_action_jump);
3874 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3875 		if (!ext_actions)
3876 			return rte_flow_error_set(error, ENOMEM,
3877 						  RTE_FLOW_ERROR_TYPE_ACTION,
3878 						  NULL, "no memory to split "
3879 						  "metadata flow");
3880 		/*
3881 		 * If we are the suffix flow of meter, tag already exist.
3882 		 * Set the tag action to void.
3883 		 */
3884 		if (mtr_sfx)
3885 			ext_actions[qrss - actions].type =
3886 						RTE_FLOW_ACTION_TYPE_VOID;
3887 		else
3888 			ext_actions[qrss - actions].type =
3889 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3890 		/*
3891 		 * Create the new actions list with removed Q/RSS action
3892 		 * and appended set tag and jump to register copy table
3893 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3894 		 * in advance, because it is needed for set tag action.
3895 		 */
3896 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3897 						    qrss, actions_n, error);
3898 		if (!mtr_sfx && !qrss_id) {
3899 			ret = -rte_errno;
3900 			goto exit;
3901 		}
3902 	} else if (attr->egress && !attr->transfer) {
3903 		/*
3904 		 * All the actions on NIC Tx should have a metadata register
3905 		 * copy action to copy reg_a from WQE to reg_c[meta]
3906 		 */
3907 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3908 			   sizeof(struct mlx5_flow_action_copy_mreg);
3909 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3910 		if (!ext_actions)
3911 			return rte_flow_error_set(error, ENOMEM,
3912 						  RTE_FLOW_ERROR_TYPE_ACTION,
3913 						  NULL, "no memory to split "
3914 						  "metadata flow");
3915 		/* Create the action list appended with copy register. */
3916 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3917 					     actions_n, error, encap_idx);
3918 		if (ret < 0)
3919 			goto exit;
3920 	}
3921 	/* Add the unmodified original or prefix subflow. */
3922 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
3923 				      items, ext_actions ? ext_actions :
3924 				      actions, external, error);
3925 	if (ret < 0)
3926 		goto exit;
3927 	MLX5_ASSERT(dev_flow);
3928 	if (qrss) {
3929 		const struct rte_flow_attr q_attr = {
3930 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3931 			.ingress = 1,
3932 		};
3933 		/* Internal PMD action to set register. */
3934 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3935 			.data = qrss_id,
3936 			.id = 0,
3937 		};
3938 		struct rte_flow_item q_items[] = {
3939 			{
3940 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3941 				.spec = &q_tag_spec,
3942 				.last = NULL,
3943 				.mask = NULL,
3944 			},
3945 			{
3946 				.type = RTE_FLOW_ITEM_TYPE_END,
3947 			},
3948 		};
3949 		struct rte_flow_action q_actions[] = {
3950 			{
3951 				.type = qrss->type,
3952 				.conf = qrss->conf,
3953 			},
3954 			{
3955 				.type = RTE_FLOW_ACTION_TYPE_END,
3956 			},
3957 		};
3958 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
3959 
3960 		/*
3961 		 * Configure the tag item only if there is no meter subflow.
3962 		 * Since tag is already marked in the meter suffix subflow
3963 		 * we can just use the meter suffix items as is.
3964 		 */
3965 		if (qrss_id) {
3966 			/* Not meter subflow. */
3967 			MLX5_ASSERT(!mtr_sfx);
3968 			/*
3969 			 * Put unique id in prefix flow due to it is destroyed
3970 			 * after suffix flow and id will be freed after there
3971 			 * is no actual flows with this id and identifier
3972 			 * reallocation becomes possible (for example, for
3973 			 * other flows in other threads).
3974 			 */
3975 			dev_flow->handle.qrss_id = qrss_id;
3976 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3977 						   error);
3978 			if (ret < 0)
3979 				goto exit;
3980 			q_tag_spec.id = ret;
3981 		}
3982 		dev_flow = NULL;
3983 		/* Add suffix subflow to execute Q/RSS. */
3984 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
3985 					      &q_attr, mtr_sfx ? items :
3986 					      q_items, q_actions,
3987 					      external, error);
3988 		if (ret < 0)
3989 			goto exit;
3990 		/* qrss ID should be freed if failed. */
3991 		qrss_id = 0;
3992 		MLX5_ASSERT(dev_flow);
3993 	}
3994 
3995 exit:
3996 	/*
3997 	 * We do not destroy the partially created sub_flows in case of error.
3998 	 * These ones are included into parent flow list and will be destroyed
3999 	 * by flow_drv_destroy.
4000 	 */
4001 	flow_qrss_free_id(dev, qrss_id);
4002 	rte_free(ext_actions);
4003 	return ret;
4004 }
4005 
4006 /**
4007  * The splitting for meter feature.
4008  *
4009  * - The meter flow will be split to two flows as prefix and
4010  *   suffix flow. The packets make sense only it pass the prefix
4011  *   meter action.
4012  *
4013  * - Reg_C_5 is used for the packet to match betweend prefix and
4014  *   suffix flow.
4015  *
4016  * @param dev
4017  *   Pointer to Ethernet device.
4018  * @param[in] flow
4019  *   Parent flow structure pointer.
4020  * @param[in] attr
4021  *   Flow rule attributes.
4022  * @param[in] items
4023  *   Pattern specification (list terminated by the END pattern item).
4024  * @param[in] actions
4025  *   Associated actions (list terminated by the END action).
4026  * @param[in] external
4027  *   This flow rule is created by request external to PMD.
4028  * @param[out] error
4029  *   Perform verbose error reporting if not NULL.
4030  * @return
4031  *   0 on success, negative value otherwise
4032  */
4033 static int
4034 flow_create_split_meter(struct rte_eth_dev *dev,
4035 			   struct rte_flow *flow,
4036 			   const struct rte_flow_attr *attr,
4037 			   const struct rte_flow_item items[],
4038 			   const struct rte_flow_action actions[],
4039 			   bool external, struct rte_flow_error *error)
4040 {
4041 	struct mlx5_priv *priv = dev->data->dev_private;
4042 	struct rte_flow_action *sfx_actions = NULL;
4043 	struct rte_flow_action *pre_actions = NULL;
4044 	struct rte_flow_item *sfx_items = NULL;
4045 	struct mlx5_flow *dev_flow = NULL;
4046 	struct rte_flow_attr sfx_attr = *attr;
4047 	uint32_t mtr = 0;
4048 	uint32_t mtr_tag_id = 0;
4049 	size_t act_size;
4050 	size_t item_size;
4051 	int actions_n = 0;
4052 	int ret;
4053 
4054 	if (priv->mtr_en)
4055 		actions_n = flow_check_meter_action(actions, &mtr);
4056 	if (mtr) {
4057 		/* The five prefix actions: meter, decap, encap, tag, end. */
4058 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4059 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4060 		/* tag, vlan, port id, end. */
4061 #define METER_SUFFIX_ITEM 4
4062 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4063 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4064 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4065 		if (!sfx_actions)
4066 			return rte_flow_error_set(error, ENOMEM,
4067 						  RTE_FLOW_ERROR_TYPE_ACTION,
4068 						  NULL, "no memory to split "
4069 						  "meter flow");
4070 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4071 			     act_size);
4072 		pre_actions = sfx_actions + actions_n;
4073 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4074 						   actions, sfx_actions,
4075 						   pre_actions);
4076 		if (!mtr_tag_id) {
4077 			ret = -rte_errno;
4078 			goto exit;
4079 		}
4080 		/* Add the prefix subflow. */
4081 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4082 					      items, pre_actions, external,
4083 					      error);
4084 		if (ret) {
4085 			ret = -rte_errno;
4086 			goto exit;
4087 		}
4088 		dev_flow->handle.mtr_flow_id = mtr_tag_id;
4089 		/* Setting the sfx group atrr. */
4090 		sfx_attr.group = sfx_attr.transfer ?
4091 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4092 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4093 	}
4094 	/* Add the prefix subflow. */
4095 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4096 					 flow_get_prefix_layer_flags(dev_flow) :
4097 					 0, &sfx_attr,
4098 					 sfx_items ? sfx_items : items,
4099 					 sfx_actions ? sfx_actions : actions,
4100 					 external, error);
4101 exit:
4102 	if (sfx_actions)
4103 		rte_free(sfx_actions);
4104 	return ret;
4105 }
4106 
4107 /**
4108  * Split the flow to subflow set. The splitters might be linked
4109  * in the chain, like this:
4110  * flow_create_split_outer() calls:
4111  *   flow_create_split_meter() calls:
4112  *     flow_create_split_metadata(meter_subflow_0) calls:
4113  *       flow_create_split_inner(metadata_subflow_0)
4114  *       flow_create_split_inner(metadata_subflow_1)
4115  *       flow_create_split_inner(metadata_subflow_2)
4116  *     flow_create_split_metadata(meter_subflow_1) calls:
4117  *       flow_create_split_inner(metadata_subflow_0)
4118  *       flow_create_split_inner(metadata_subflow_1)
4119  *       flow_create_split_inner(metadata_subflow_2)
4120  *
4121  * This provide flexible way to add new levels of flow splitting.
4122  * The all of successfully created subflows are included to the
4123  * parent flow dev_flow list.
4124  *
4125  * @param dev
4126  *   Pointer to Ethernet device.
4127  * @param[in] flow
4128  *   Parent flow structure pointer.
4129  * @param[in] attr
4130  *   Flow rule attributes.
4131  * @param[in] items
4132  *   Pattern specification (list terminated by the END pattern item).
4133  * @param[in] actions
4134  *   Associated actions (list terminated by the END action).
4135  * @param[in] external
4136  *   This flow rule is created by request external to PMD.
4137  * @param[out] error
4138  *   Perform verbose error reporting if not NULL.
4139  * @return
4140  *   0 on success, negative value otherwise
4141  */
4142 static int
4143 flow_create_split_outer(struct rte_eth_dev *dev,
4144 			struct rte_flow *flow,
4145 			const struct rte_flow_attr *attr,
4146 			const struct rte_flow_item items[],
4147 			const struct rte_flow_action actions[],
4148 			bool external, struct rte_flow_error *error)
4149 {
4150 	int ret;
4151 
4152 	ret = flow_create_split_meter(dev, flow, attr, items,
4153 					 actions, external, error);
4154 	MLX5_ASSERT(ret <= 0);
4155 	return ret;
4156 }
4157 
4158 /**
4159  * Create a flow and add it to @p list.
4160  *
4161  * @param dev
4162  *   Pointer to Ethernet device.
4163  * @param list
4164  *   Pointer to a TAILQ flow list. If this parameter NULL,
4165  *   no list insertion occurred, flow is just created,
4166  *   this is caller's responsibility to track the
4167  *   created flow.
4168  * @param[in] attr
4169  *   Flow rule attributes.
4170  * @param[in] items
4171  *   Pattern specification (list terminated by the END pattern item).
4172  * @param[in] actions
4173  *   Associated actions (list terminated by the END action).
4174  * @param[in] external
4175  *   This flow rule is created by request external to PMD.
4176  * @param[out] error
4177  *   Perform verbose error reporting if not NULL.
4178  *
4179  * @return
4180  *   A flow on success, NULL otherwise and rte_errno is set.
4181  */
4182 static struct rte_flow *
4183 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4184 		 const struct rte_flow_attr *attr,
4185 		 const struct rte_flow_item items[],
4186 		 const struct rte_flow_action actions[],
4187 		 bool external, struct rte_flow_error *error)
4188 {
4189 	struct mlx5_priv *priv = dev->data->dev_private;
4190 	struct rte_flow *flow = NULL;
4191 	struct mlx5_flow *dev_flow;
4192 	const struct rte_flow_action_rss *rss;
4193 	union {
4194 		struct rte_flow_expand_rss buf;
4195 		uint8_t buffer[2048];
4196 	} expand_buffer;
4197 	union {
4198 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4199 		uint8_t buffer[2048];
4200 	} actions_rx;
4201 	union {
4202 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4203 		uint8_t buffer[2048];
4204 	} actions_hairpin_tx;
4205 	union {
4206 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4207 		uint8_t buffer[2048];
4208 	} items_tx;
4209 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4210 	const struct rte_flow_action *p_actions_rx = actions;
4211 	uint32_t i;
4212 	uint32_t flow_size;
4213 	int hairpin_flow = 0;
4214 	uint32_t hairpin_id = 0;
4215 	struct rte_flow_attr attr_tx = { .priority = 0 };
4216 	int ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4217 				    error);
4218 
4219 	if (ret < 0)
4220 		return NULL;
4221 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4222 	if (hairpin_flow > 0) {
4223 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4224 			rte_errno = EINVAL;
4225 			return NULL;
4226 		}
4227 		flow_hairpin_split(dev, actions, actions_rx.actions,
4228 				   actions_hairpin_tx.actions, items_tx.items,
4229 				   &hairpin_id);
4230 		p_actions_rx = actions_rx.actions;
4231 	}
4232 	flow_size = sizeof(struct rte_flow);
4233 	rss = flow_get_rss_action(p_actions_rx);
4234 	if (rss)
4235 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4236 					    sizeof(void *));
4237 	else
4238 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4239 	flow = rte_calloc(__func__, 1, flow_size, 0);
4240 	if (!flow) {
4241 		rte_errno = ENOMEM;
4242 		goto error_before_flow;
4243 	}
4244 	flow->drv_type = flow_get_drv_type(dev, attr);
4245 	if (hairpin_id != 0)
4246 		flow->hairpin_flow_id = hairpin_id;
4247 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4248 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4249 	flow->rss.queue = (void *)(flow + 1);
4250 	if (rss) {
4251 		/*
4252 		 * The following information is required by
4253 		 * mlx5_flow_hashfields_adjust() in advance.
4254 		 */
4255 		flow->rss.level = rss->level;
4256 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4257 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4258 	}
4259 	LIST_INIT(&flow->dev_flows);
4260 	if (rss && rss->types) {
4261 		unsigned int graph_root;
4262 
4263 		graph_root = find_graph_root(items, rss->level);
4264 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4265 					  items, rss->types,
4266 					  mlx5_support_expansion,
4267 					  graph_root);
4268 		MLX5_ASSERT(ret > 0 &&
4269 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4270 	} else {
4271 		buf->entries = 1;
4272 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4273 	}
4274 	for (i = 0; i < buf->entries; ++i) {
4275 		/*
4276 		 * The splitter may create multiple dev_flows,
4277 		 * depending on configuration. In the simplest
4278 		 * case it just creates unmodified original flow.
4279 		 */
4280 		ret = flow_create_split_outer(dev, flow, attr,
4281 					      buf->entry[i].pattern,
4282 					      p_actions_rx, external,
4283 					      error);
4284 		if (ret < 0)
4285 			goto error;
4286 	}
4287 	/* Create the tx flow. */
4288 	if (hairpin_flow) {
4289 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4290 		attr_tx.ingress = 0;
4291 		attr_tx.egress = 1;
4292 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4293 					    actions_hairpin_tx.actions, error);
4294 		if (!dev_flow)
4295 			goto error;
4296 		dev_flow->flow = flow;
4297 		dev_flow->external = 0;
4298 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4299 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4300 					 items_tx.items,
4301 					 actions_hairpin_tx.actions, error);
4302 		if (ret < 0)
4303 			goto error;
4304 	}
4305 	/*
4306 	 * Update the metadata register copy table. If extensive
4307 	 * metadata feature is enabled and registers are supported
4308 	 * we might create the extra rte_flow for each unique
4309 	 * MARK/FLAG action ID.
4310 	 *
4311 	 * The table is updated for ingress Flows only, because
4312 	 * the egress Flows belong to the different device and
4313 	 * copy table should be updated in peer NIC Rx domain.
4314 	 */
4315 	if (attr->ingress &&
4316 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4317 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4318 		if (ret)
4319 			goto error;
4320 	}
4321 	if (dev->data->dev_started) {
4322 		ret = flow_drv_apply(dev, flow, error);
4323 		if (ret < 0)
4324 			goto error;
4325 	}
4326 	if (list)
4327 		TAILQ_INSERT_TAIL(list, flow, next);
4328 	flow_rxq_flags_set(dev, flow);
4329 	return flow;
4330 error_before_flow:
4331 	if (hairpin_id)
4332 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4333 				     hairpin_id);
4334 	return NULL;
4335 error:
4336 	MLX5_ASSERT(flow);
4337 	flow_mreg_del_copy_action(dev, flow);
4338 	ret = rte_errno; /* Save rte_errno before cleanup. */
4339 	if (flow->hairpin_flow_id)
4340 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4341 				     flow->hairpin_flow_id);
4342 	MLX5_ASSERT(flow);
4343 	flow_drv_destroy(dev, flow);
4344 	rte_free(flow);
4345 	rte_errno = ret; /* Restore rte_errno. */
4346 	return NULL;
4347 }
4348 
4349 /**
4350  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4351  * incoming packets to table 1.
4352  *
4353  * Other flow rules, requested for group n, will be created in
4354  * e-switch table n+1.
4355  * Jump action to e-switch group n will be created to group n+1.
4356  *
4357  * Used when working in switchdev mode, to utilise advantages of table 1
4358  * and above.
4359  *
4360  * @param dev
4361  *   Pointer to Ethernet device.
4362  *
4363  * @return
4364  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4365  */
4366 struct rte_flow *
4367 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4368 {
4369 	const struct rte_flow_attr attr = {
4370 		.group = 0,
4371 		.priority = 0,
4372 		.ingress = 1,
4373 		.egress = 0,
4374 		.transfer = 1,
4375 	};
4376 	const struct rte_flow_item pattern = {
4377 		.type = RTE_FLOW_ITEM_TYPE_END,
4378 	};
4379 	struct rte_flow_action_jump jump = {
4380 		.group = 1,
4381 	};
4382 	const struct rte_flow_action actions[] = {
4383 		{
4384 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4385 			.conf = &jump,
4386 		},
4387 		{
4388 			.type = RTE_FLOW_ACTION_TYPE_END,
4389 		},
4390 	};
4391 	struct mlx5_priv *priv = dev->data->dev_private;
4392 	struct rte_flow_error error;
4393 
4394 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4395 				actions, false, &error);
4396 }
4397 
4398 /**
4399  * Create a flow.
4400  *
4401  * @see rte_flow_create()
4402  * @see rte_flow_ops
4403  */
4404 struct rte_flow *
4405 mlx5_flow_create(struct rte_eth_dev *dev,
4406 		 const struct rte_flow_attr *attr,
4407 		 const struct rte_flow_item items[],
4408 		 const struct rte_flow_action actions[],
4409 		 struct rte_flow_error *error)
4410 {
4411 	struct mlx5_priv *priv = dev->data->dev_private;
4412 
4413 	return flow_list_create(dev, &priv->flows,
4414 				attr, items, actions, true, error);
4415 }
4416 
4417 /**
4418  * Destroy a flow in a list.
4419  *
4420  * @param dev
4421  *   Pointer to Ethernet device.
4422  * @param list
4423  *   Pointer to a TAILQ flow list. If this parameter NULL,
4424  *   there is no flow removal from the list.
4425  * @param[in] flow
4426  *   Flow to destroy.
4427  */
4428 static void
4429 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4430 		  struct rte_flow *flow)
4431 {
4432 	struct mlx5_priv *priv = dev->data->dev_private;
4433 
4434 	/*
4435 	 * Update RX queue flags only if port is started, otherwise it is
4436 	 * already clean.
4437 	 */
4438 	if (dev->data->dev_started)
4439 		flow_rxq_flags_trim(dev, flow);
4440 	if (flow->hairpin_flow_id)
4441 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4442 				     flow->hairpin_flow_id);
4443 	flow_drv_destroy(dev, flow);
4444 	if (list)
4445 		TAILQ_REMOVE(list, flow, next);
4446 	flow_mreg_del_copy_action(dev, flow);
4447 	rte_free(flow->fdir);
4448 	rte_free(flow);
4449 }
4450 
4451 /**
4452  * Destroy all flows.
4453  *
4454  * @param dev
4455  *   Pointer to Ethernet device.
4456  * @param list
4457  *   Pointer to a TAILQ flow list.
4458  * @param active
4459  *   If flushing is called avtively.
4460  */
4461 void
4462 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list,
4463 		     bool active)
4464 {
4465 	uint32_t num_flushed = 0;
4466 
4467 	while (!TAILQ_EMPTY(list)) {
4468 		struct rte_flow *flow;
4469 
4470 		flow = TAILQ_FIRST(list);
4471 		flow_list_destroy(dev, list, flow);
4472 		num_flushed++;
4473 	}
4474 	if (active) {
4475 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4476 			dev->data->port_id, num_flushed);
4477 	}
4478 }
4479 
4480 /**
4481  * Remove all flows.
4482  *
4483  * @param dev
4484  *   Pointer to Ethernet device.
4485  * @param list
4486  *   Pointer to a TAILQ flow list.
4487  */
4488 void
4489 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4490 {
4491 	struct rte_flow *flow;
4492 
4493 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4494 		flow_drv_remove(dev, flow);
4495 		flow_mreg_stop_copy_action(dev, flow);
4496 	}
4497 	flow_mreg_del_default_copy_action(dev);
4498 	flow_rxq_flags_clear(dev);
4499 }
4500 
4501 /**
4502  * Add all flows.
4503  *
4504  * @param dev
4505  *   Pointer to Ethernet device.
4506  * @param list
4507  *   Pointer to a TAILQ flow list.
4508  *
4509  * @return
4510  *   0 on success, a negative errno value otherwise and rte_errno is set.
4511  */
4512 int
4513 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4514 {
4515 	struct rte_flow *flow;
4516 	struct rte_flow_error error;
4517 	int ret = 0;
4518 
4519 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4520 	ret = flow_mreg_add_default_copy_action(dev, &error);
4521 	if (ret < 0)
4522 		return -rte_errno;
4523 	/* Apply Flows created by application. */
4524 	TAILQ_FOREACH(flow, list, next) {
4525 		ret = flow_mreg_start_copy_action(dev, flow);
4526 		if (ret < 0)
4527 			goto error;
4528 		ret = flow_drv_apply(dev, flow, &error);
4529 		if (ret < 0)
4530 			goto error;
4531 		flow_rxq_flags_set(dev, flow);
4532 	}
4533 	return 0;
4534 error:
4535 	ret = rte_errno; /* Save rte_errno before cleanup. */
4536 	mlx5_flow_stop(dev, list);
4537 	rte_errno = ret; /* Restore rte_errno. */
4538 	return -rte_errno;
4539 }
4540 
4541 /**
4542  * Stop all default actions for flows.
4543  *
4544  * @param dev
4545  *   Pointer to Ethernet device.
4546  * @param list
4547  *   Pointer to a TAILQ flow list.
4548  */
4549 void
4550 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4551 {
4552 	flow_mreg_del_default_copy_action(dev);
4553 }
4554 
4555 /**
4556  * Start all default actions for flows.
4557  *
4558  * @param dev
4559  *   Pointer to Ethernet device.
4560  * @return
4561  *   0 on success, a negative errno value otherwise and rte_errno is set.
4562  */
4563 int
4564 mlx5_flow_start_default(struct rte_eth_dev *dev)
4565 {
4566 	struct rte_flow_error error;
4567 
4568 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4569 	return flow_mreg_add_default_copy_action(dev, &error);
4570 }
4571 
4572 /**
4573  * Verify the flow list is empty
4574  *
4575  * @param dev
4576  *  Pointer to Ethernet device.
4577  *
4578  * @return the number of flows not released.
4579  */
4580 int
4581 mlx5_flow_verify(struct rte_eth_dev *dev)
4582 {
4583 	struct mlx5_priv *priv = dev->data->dev_private;
4584 	struct rte_flow *flow;
4585 	int ret = 0;
4586 
4587 	TAILQ_FOREACH(flow, &priv->flows, next) {
4588 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4589 			dev->data->port_id, (void *)flow);
4590 		++ret;
4591 	}
4592 	return ret;
4593 }
4594 
4595 /**
4596  * Enable default hairpin egress flow.
4597  *
4598  * @param dev
4599  *   Pointer to Ethernet device.
4600  * @param queue
4601  *   The queue index.
4602  *
4603  * @return
4604  *   0 on success, a negative errno value otherwise and rte_errno is set.
4605  */
4606 int
4607 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4608 			    uint32_t queue)
4609 {
4610 	struct mlx5_priv *priv = dev->data->dev_private;
4611 	const struct rte_flow_attr attr = {
4612 		.egress = 1,
4613 		.priority = 0,
4614 	};
4615 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4616 		.queue = queue,
4617 	};
4618 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4619 		.queue = UINT32_MAX,
4620 	};
4621 	struct rte_flow_item items[] = {
4622 		{
4623 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4624 			.spec = &queue_spec,
4625 			.last = NULL,
4626 			.mask = &queue_mask,
4627 		},
4628 		{
4629 			.type = RTE_FLOW_ITEM_TYPE_END,
4630 		},
4631 	};
4632 	struct rte_flow_action_jump jump = {
4633 		.group = MLX5_HAIRPIN_TX_TABLE,
4634 	};
4635 	struct rte_flow_action actions[2];
4636 	struct rte_flow *flow;
4637 	struct rte_flow_error error;
4638 
4639 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4640 	actions[0].conf = &jump;
4641 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4642 	flow = flow_list_create(dev, &priv->ctrl_flows,
4643 				&attr, items, actions, false, &error);
4644 	if (!flow) {
4645 		DRV_LOG(DEBUG,
4646 			"Failed to create ctrl flow: rte_errno(%d),"
4647 			" type(%d), message(%s)",
4648 			rte_errno, error.type,
4649 			error.message ? error.message : " (no stated reason)");
4650 		return -rte_errno;
4651 	}
4652 	return 0;
4653 }
4654 
4655 /**
4656  * Enable a control flow configured from the control plane.
4657  *
4658  * @param dev
4659  *   Pointer to Ethernet device.
4660  * @param eth_spec
4661  *   An Ethernet flow spec to apply.
4662  * @param eth_mask
4663  *   An Ethernet flow mask to apply.
4664  * @param vlan_spec
4665  *   A VLAN flow spec to apply.
4666  * @param vlan_mask
4667  *   A VLAN flow mask to apply.
4668  *
4669  * @return
4670  *   0 on success, a negative errno value otherwise and rte_errno is set.
4671  */
4672 int
4673 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4674 		    struct rte_flow_item_eth *eth_spec,
4675 		    struct rte_flow_item_eth *eth_mask,
4676 		    struct rte_flow_item_vlan *vlan_spec,
4677 		    struct rte_flow_item_vlan *vlan_mask)
4678 {
4679 	struct mlx5_priv *priv = dev->data->dev_private;
4680 	const struct rte_flow_attr attr = {
4681 		.ingress = 1,
4682 		.priority = MLX5_FLOW_PRIO_RSVD,
4683 	};
4684 	struct rte_flow_item items[] = {
4685 		{
4686 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4687 			.spec = eth_spec,
4688 			.last = NULL,
4689 			.mask = eth_mask,
4690 		},
4691 		{
4692 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4693 					      RTE_FLOW_ITEM_TYPE_END,
4694 			.spec = vlan_spec,
4695 			.last = NULL,
4696 			.mask = vlan_mask,
4697 		},
4698 		{
4699 			.type = RTE_FLOW_ITEM_TYPE_END,
4700 		},
4701 	};
4702 	uint16_t queue[priv->reta_idx_n];
4703 	struct rte_flow_action_rss action_rss = {
4704 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4705 		.level = 0,
4706 		.types = priv->rss_conf.rss_hf,
4707 		.key_len = priv->rss_conf.rss_key_len,
4708 		.queue_num = priv->reta_idx_n,
4709 		.key = priv->rss_conf.rss_key,
4710 		.queue = queue,
4711 	};
4712 	struct rte_flow_action actions[] = {
4713 		{
4714 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4715 			.conf = &action_rss,
4716 		},
4717 		{
4718 			.type = RTE_FLOW_ACTION_TYPE_END,
4719 		},
4720 	};
4721 	struct rte_flow *flow;
4722 	struct rte_flow_error error;
4723 	unsigned int i;
4724 
4725 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4726 		return 0;
4727 	}
4728 	for (i = 0; i != priv->reta_idx_n; ++i)
4729 		queue[i] = (*priv->reta_idx)[i];
4730 	flow = flow_list_create(dev, &priv->ctrl_flows,
4731 				&attr, items, actions, false, &error);
4732 	if (!flow)
4733 		return -rte_errno;
4734 	return 0;
4735 }
4736 
4737 /**
4738  * Enable a flow control configured from the control plane.
4739  *
4740  * @param dev
4741  *   Pointer to Ethernet device.
4742  * @param eth_spec
4743  *   An Ethernet flow spec to apply.
4744  * @param eth_mask
4745  *   An Ethernet flow mask to apply.
4746  *
4747  * @return
4748  *   0 on success, a negative errno value otherwise and rte_errno is set.
4749  */
4750 int
4751 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4752 	       struct rte_flow_item_eth *eth_spec,
4753 	       struct rte_flow_item_eth *eth_mask)
4754 {
4755 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4756 }
4757 
4758 /**
4759  * Destroy a flow.
4760  *
4761  * @see rte_flow_destroy()
4762  * @see rte_flow_ops
4763  */
4764 int
4765 mlx5_flow_destroy(struct rte_eth_dev *dev,
4766 		  struct rte_flow *flow,
4767 		  struct rte_flow_error *error __rte_unused)
4768 {
4769 	struct mlx5_priv *priv = dev->data->dev_private;
4770 
4771 	flow_list_destroy(dev, &priv->flows, flow);
4772 	return 0;
4773 }
4774 
4775 /**
4776  * Destroy all flows.
4777  *
4778  * @see rte_flow_flush()
4779  * @see rte_flow_ops
4780  */
4781 int
4782 mlx5_flow_flush(struct rte_eth_dev *dev,
4783 		struct rte_flow_error *error __rte_unused)
4784 {
4785 	struct mlx5_priv *priv = dev->data->dev_private;
4786 
4787 	mlx5_flow_list_flush(dev, &priv->flows, false);
4788 	return 0;
4789 }
4790 
4791 /**
4792  * Isolated mode.
4793  *
4794  * @see rte_flow_isolate()
4795  * @see rte_flow_ops
4796  */
4797 int
4798 mlx5_flow_isolate(struct rte_eth_dev *dev,
4799 		  int enable,
4800 		  struct rte_flow_error *error)
4801 {
4802 	struct mlx5_priv *priv = dev->data->dev_private;
4803 
4804 	if (dev->data->dev_started) {
4805 		rte_flow_error_set(error, EBUSY,
4806 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4807 				   NULL,
4808 				   "port must be stopped first");
4809 		return -rte_errno;
4810 	}
4811 	priv->isolated = !!enable;
4812 	if (enable)
4813 		dev->dev_ops = &mlx5_dev_ops_isolate;
4814 	else
4815 		dev->dev_ops = &mlx5_dev_ops;
4816 	return 0;
4817 }
4818 
4819 /**
4820  * Query a flow.
4821  *
4822  * @see rte_flow_query()
4823  * @see rte_flow_ops
4824  */
4825 static int
4826 flow_drv_query(struct rte_eth_dev *dev,
4827 	       struct rte_flow *flow,
4828 	       const struct rte_flow_action *actions,
4829 	       void *data,
4830 	       struct rte_flow_error *error)
4831 {
4832 	const struct mlx5_flow_driver_ops *fops;
4833 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4834 
4835 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4836 	fops = flow_get_drv_ops(ftype);
4837 
4838 	return fops->query(dev, flow, actions, data, error);
4839 }
4840 
4841 /**
4842  * Query a flow.
4843  *
4844  * @see rte_flow_query()
4845  * @see rte_flow_ops
4846  */
4847 int
4848 mlx5_flow_query(struct rte_eth_dev *dev,
4849 		struct rte_flow *flow,
4850 		const struct rte_flow_action *actions,
4851 		void *data,
4852 		struct rte_flow_error *error)
4853 {
4854 	int ret;
4855 
4856 	ret = flow_drv_query(dev, flow, actions, data, error);
4857 	if (ret < 0)
4858 		return ret;
4859 	return 0;
4860 }
4861 
4862 /**
4863  * Convert a flow director filter to a generic flow.
4864  *
4865  * @param dev
4866  *   Pointer to Ethernet device.
4867  * @param fdir_filter
4868  *   Flow director filter to add.
4869  * @param attributes
4870  *   Generic flow parameters structure.
4871  *
4872  * @return
4873  *   0 on success, a negative errno value otherwise and rte_errno is set.
4874  */
4875 static int
4876 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4877 			 const struct rte_eth_fdir_filter *fdir_filter,
4878 			 struct mlx5_fdir *attributes)
4879 {
4880 	struct mlx5_priv *priv = dev->data->dev_private;
4881 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4882 	const struct rte_eth_fdir_masks *mask =
4883 		&dev->data->dev_conf.fdir_conf.mask;
4884 
4885 	/* Validate queue number. */
4886 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4887 		DRV_LOG(ERR, "port %u invalid queue number %d",
4888 			dev->data->port_id, fdir_filter->action.rx_queue);
4889 		rte_errno = EINVAL;
4890 		return -rte_errno;
4891 	}
4892 	attributes->attr.ingress = 1;
4893 	attributes->items[0] = (struct rte_flow_item) {
4894 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4895 		.spec = &attributes->l2,
4896 		.mask = &attributes->l2_mask,
4897 	};
4898 	switch (fdir_filter->action.behavior) {
4899 	case RTE_ETH_FDIR_ACCEPT:
4900 		attributes->actions[0] = (struct rte_flow_action){
4901 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4902 			.conf = &attributes->queue,
4903 		};
4904 		break;
4905 	case RTE_ETH_FDIR_REJECT:
4906 		attributes->actions[0] = (struct rte_flow_action){
4907 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4908 		};
4909 		break;
4910 	default:
4911 		DRV_LOG(ERR, "port %u invalid behavior %d",
4912 			dev->data->port_id,
4913 			fdir_filter->action.behavior);
4914 		rte_errno = ENOTSUP;
4915 		return -rte_errno;
4916 	}
4917 	attributes->queue.index = fdir_filter->action.rx_queue;
4918 	/* Handle L3. */
4919 	switch (fdir_filter->input.flow_type) {
4920 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4921 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4922 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4923 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4924 			.src_addr = input->flow.ip4_flow.src_ip,
4925 			.dst_addr = input->flow.ip4_flow.dst_ip,
4926 			.time_to_live = input->flow.ip4_flow.ttl,
4927 			.type_of_service = input->flow.ip4_flow.tos,
4928 		};
4929 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4930 			.src_addr = mask->ipv4_mask.src_ip,
4931 			.dst_addr = mask->ipv4_mask.dst_ip,
4932 			.time_to_live = mask->ipv4_mask.ttl,
4933 			.type_of_service = mask->ipv4_mask.tos,
4934 			.next_proto_id = mask->ipv4_mask.proto,
4935 		};
4936 		attributes->items[1] = (struct rte_flow_item){
4937 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
4938 			.spec = &attributes->l3,
4939 			.mask = &attributes->l3_mask,
4940 		};
4941 		break;
4942 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4943 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4944 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4945 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4946 			.hop_limits = input->flow.ipv6_flow.hop_limits,
4947 			.proto = input->flow.ipv6_flow.proto,
4948 		};
4949 
4950 		memcpy(attributes->l3.ipv6.hdr.src_addr,
4951 		       input->flow.ipv6_flow.src_ip,
4952 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4953 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
4954 		       input->flow.ipv6_flow.dst_ip,
4955 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4956 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4957 		       mask->ipv6_mask.src_ip,
4958 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4959 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4960 		       mask->ipv6_mask.dst_ip,
4961 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4962 		attributes->items[1] = (struct rte_flow_item){
4963 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
4964 			.spec = &attributes->l3,
4965 			.mask = &attributes->l3_mask,
4966 		};
4967 		break;
4968 	default:
4969 		DRV_LOG(ERR, "port %u invalid flow type%d",
4970 			dev->data->port_id, fdir_filter->input.flow_type);
4971 		rte_errno = ENOTSUP;
4972 		return -rte_errno;
4973 	}
4974 	/* Handle L4. */
4975 	switch (fdir_filter->input.flow_type) {
4976 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4977 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4978 			.src_port = input->flow.udp4_flow.src_port,
4979 			.dst_port = input->flow.udp4_flow.dst_port,
4980 		};
4981 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4982 			.src_port = mask->src_port_mask,
4983 			.dst_port = mask->dst_port_mask,
4984 		};
4985 		attributes->items[2] = (struct rte_flow_item){
4986 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4987 			.spec = &attributes->l4,
4988 			.mask = &attributes->l4_mask,
4989 		};
4990 		break;
4991 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4992 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4993 			.src_port = input->flow.tcp4_flow.src_port,
4994 			.dst_port = input->flow.tcp4_flow.dst_port,
4995 		};
4996 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4997 			.src_port = mask->src_port_mask,
4998 			.dst_port = mask->dst_port_mask,
4999 		};
5000 		attributes->items[2] = (struct rte_flow_item){
5001 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5002 			.spec = &attributes->l4,
5003 			.mask = &attributes->l4_mask,
5004 		};
5005 		break;
5006 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5007 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5008 			.src_port = input->flow.udp6_flow.src_port,
5009 			.dst_port = input->flow.udp6_flow.dst_port,
5010 		};
5011 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5012 			.src_port = mask->src_port_mask,
5013 			.dst_port = mask->dst_port_mask,
5014 		};
5015 		attributes->items[2] = (struct rte_flow_item){
5016 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5017 			.spec = &attributes->l4,
5018 			.mask = &attributes->l4_mask,
5019 		};
5020 		break;
5021 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5022 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5023 			.src_port = input->flow.tcp6_flow.src_port,
5024 			.dst_port = input->flow.tcp6_flow.dst_port,
5025 		};
5026 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5027 			.src_port = mask->src_port_mask,
5028 			.dst_port = mask->dst_port_mask,
5029 		};
5030 		attributes->items[2] = (struct rte_flow_item){
5031 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5032 			.spec = &attributes->l4,
5033 			.mask = &attributes->l4_mask,
5034 		};
5035 		break;
5036 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5037 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5038 		break;
5039 	default:
5040 		DRV_LOG(ERR, "port %u invalid flow type%d",
5041 			dev->data->port_id, fdir_filter->input.flow_type);
5042 		rte_errno = ENOTSUP;
5043 		return -rte_errno;
5044 	}
5045 	return 0;
5046 }
5047 
5048 #define FLOW_FDIR_CMP(f1, f2, fld) \
5049 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5050 
5051 /**
5052  * Compare two FDIR flows. If items and actions are identical, the two flows are
5053  * regarded as same.
5054  *
5055  * @param dev
5056  *   Pointer to Ethernet device.
5057  * @param f1
5058  *   FDIR flow to compare.
5059  * @param f2
5060  *   FDIR flow to compare.
5061  *
5062  * @return
5063  *   Zero on match, 1 otherwise.
5064  */
5065 static int
5066 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5067 {
5068 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5069 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5070 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5071 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5072 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5073 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5074 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5075 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5076 		return 1;
5077 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5078 	    FLOW_FDIR_CMP(f1, f2, queue))
5079 		return 1;
5080 	return 0;
5081 }
5082 
5083 /**
5084  * Search device flow list to find out a matched FDIR flow.
5085  *
5086  * @param dev
5087  *   Pointer to Ethernet device.
5088  * @param fdir_flow
5089  *   FDIR flow to lookup.
5090  *
5091  * @return
5092  *   Pointer of flow if found, NULL otherwise.
5093  */
5094 static struct rte_flow *
5095 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5096 {
5097 	struct mlx5_priv *priv = dev->data->dev_private;
5098 	struct rte_flow *flow = NULL;
5099 
5100 	MLX5_ASSERT(fdir_flow);
5101 	TAILQ_FOREACH(flow, &priv->flows, next) {
5102 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5103 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5104 				dev->data->port_id, (void *)flow);
5105 			break;
5106 		}
5107 	}
5108 	return flow;
5109 }
5110 
5111 /**
5112  * Add new flow director filter and store it in list.
5113  *
5114  * @param dev
5115  *   Pointer to Ethernet device.
5116  * @param fdir_filter
5117  *   Flow director filter to add.
5118  *
5119  * @return
5120  *   0 on success, a negative errno value otherwise and rte_errno is set.
5121  */
5122 static int
5123 flow_fdir_filter_add(struct rte_eth_dev *dev,
5124 		     const struct rte_eth_fdir_filter *fdir_filter)
5125 {
5126 	struct mlx5_priv *priv = dev->data->dev_private;
5127 	struct mlx5_fdir *fdir_flow;
5128 	struct rte_flow *flow;
5129 	int ret;
5130 
5131 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5132 	if (!fdir_flow) {
5133 		rte_errno = ENOMEM;
5134 		return -rte_errno;
5135 	}
5136 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5137 	if (ret)
5138 		goto error;
5139 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5140 	if (flow) {
5141 		rte_errno = EEXIST;
5142 		goto error;
5143 	}
5144 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5145 				fdir_flow->items, fdir_flow->actions, true,
5146 				NULL);
5147 	if (!flow)
5148 		goto error;
5149 	MLX5_ASSERT(!flow->fdir);
5150 	flow->fdir = fdir_flow;
5151 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5152 		dev->data->port_id, (void *)flow);
5153 	return 0;
5154 error:
5155 	rte_free(fdir_flow);
5156 	return -rte_errno;
5157 }
5158 
5159 /**
5160  * Delete specific filter.
5161  *
5162  * @param dev
5163  *   Pointer to Ethernet device.
5164  * @param fdir_filter
5165  *   Filter to be deleted.
5166  *
5167  * @return
5168  *   0 on success, a negative errno value otherwise and rte_errno is set.
5169  */
5170 static int
5171 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5172 			const struct rte_eth_fdir_filter *fdir_filter)
5173 {
5174 	struct mlx5_priv *priv = dev->data->dev_private;
5175 	struct rte_flow *flow;
5176 	struct mlx5_fdir fdir_flow = {
5177 		.attr.group = 0,
5178 	};
5179 	int ret;
5180 
5181 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5182 	if (ret)
5183 		return -rte_errno;
5184 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5185 	if (!flow) {
5186 		rte_errno = ENOENT;
5187 		return -rte_errno;
5188 	}
5189 	flow_list_destroy(dev, &priv->flows, flow);
5190 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5191 		dev->data->port_id, (void *)flow);
5192 	return 0;
5193 }
5194 
5195 /**
5196  * Update queue for specific filter.
5197  *
5198  * @param dev
5199  *   Pointer to Ethernet device.
5200  * @param fdir_filter
5201  *   Filter to be updated.
5202  *
5203  * @return
5204  *   0 on success, a negative errno value otherwise and rte_errno is set.
5205  */
5206 static int
5207 flow_fdir_filter_update(struct rte_eth_dev *dev,
5208 			const struct rte_eth_fdir_filter *fdir_filter)
5209 {
5210 	int ret;
5211 
5212 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5213 	if (ret)
5214 		return ret;
5215 	return flow_fdir_filter_add(dev, fdir_filter);
5216 }
5217 
5218 /**
5219  * Flush all filters.
5220  *
5221  * @param dev
5222  *   Pointer to Ethernet device.
5223  */
5224 static void
5225 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5226 {
5227 	struct mlx5_priv *priv = dev->data->dev_private;
5228 
5229 	mlx5_flow_list_flush(dev, &priv->flows, false);
5230 }
5231 
5232 /**
5233  * Get flow director information.
5234  *
5235  * @param dev
5236  *   Pointer to Ethernet device.
5237  * @param[out] fdir_info
5238  *   Resulting flow director information.
5239  */
5240 static void
5241 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5242 {
5243 	struct rte_eth_fdir_masks *mask =
5244 		&dev->data->dev_conf.fdir_conf.mask;
5245 
5246 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5247 	fdir_info->guarant_spc = 0;
5248 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5249 	fdir_info->max_flexpayload = 0;
5250 	fdir_info->flow_types_mask[0] = 0;
5251 	fdir_info->flex_payload_unit = 0;
5252 	fdir_info->max_flex_payload_segment_num = 0;
5253 	fdir_info->flex_payload_limit = 0;
5254 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5255 }
5256 
5257 /**
5258  * Deal with flow director operations.
5259  *
5260  * @param dev
5261  *   Pointer to Ethernet device.
5262  * @param filter_op
5263  *   Operation to perform.
5264  * @param arg
5265  *   Pointer to operation-specific structure.
5266  *
5267  * @return
5268  *   0 on success, a negative errno value otherwise and rte_errno is set.
5269  */
5270 static int
5271 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5272 		    void *arg)
5273 {
5274 	enum rte_fdir_mode fdir_mode =
5275 		dev->data->dev_conf.fdir_conf.mode;
5276 
5277 	if (filter_op == RTE_ETH_FILTER_NOP)
5278 		return 0;
5279 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5280 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5281 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5282 			dev->data->port_id, fdir_mode);
5283 		rte_errno = EINVAL;
5284 		return -rte_errno;
5285 	}
5286 	switch (filter_op) {
5287 	case RTE_ETH_FILTER_ADD:
5288 		return flow_fdir_filter_add(dev, arg);
5289 	case RTE_ETH_FILTER_UPDATE:
5290 		return flow_fdir_filter_update(dev, arg);
5291 	case RTE_ETH_FILTER_DELETE:
5292 		return flow_fdir_filter_delete(dev, arg);
5293 	case RTE_ETH_FILTER_FLUSH:
5294 		flow_fdir_filter_flush(dev);
5295 		break;
5296 	case RTE_ETH_FILTER_INFO:
5297 		flow_fdir_info_get(dev, arg);
5298 		break;
5299 	default:
5300 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5301 			dev->data->port_id, filter_op);
5302 		rte_errno = EINVAL;
5303 		return -rte_errno;
5304 	}
5305 	return 0;
5306 }
5307 
5308 /**
5309  * Manage filter operations.
5310  *
5311  * @param dev
5312  *   Pointer to Ethernet device structure.
5313  * @param filter_type
5314  *   Filter type.
5315  * @param filter_op
5316  *   Operation to perform.
5317  * @param arg
5318  *   Pointer to operation-specific structure.
5319  *
5320  * @return
5321  *   0 on success, a negative errno value otherwise and rte_errno is set.
5322  */
5323 int
5324 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5325 		     enum rte_filter_type filter_type,
5326 		     enum rte_filter_op filter_op,
5327 		     void *arg)
5328 {
5329 	switch (filter_type) {
5330 	case RTE_ETH_FILTER_GENERIC:
5331 		if (filter_op != RTE_ETH_FILTER_GET) {
5332 			rte_errno = EINVAL;
5333 			return -rte_errno;
5334 		}
5335 		*(const void **)arg = &mlx5_flow_ops;
5336 		return 0;
5337 	case RTE_ETH_FILTER_FDIR:
5338 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5339 	default:
5340 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5341 			dev->data->port_id, filter_type);
5342 		rte_errno = ENOTSUP;
5343 		return -rte_errno;
5344 	}
5345 	return 0;
5346 }
5347 
5348 /**
5349  * Create the needed meter and suffix tables.
5350  *
5351  * @param[in] dev
5352  *   Pointer to Ethernet device.
5353  * @param[in] fm
5354  *   Pointer to the flow meter.
5355  *
5356  * @return
5357  *   Pointer to table set on success, NULL otherwise.
5358  */
5359 struct mlx5_meter_domains_infos *
5360 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5361 			  const struct mlx5_flow_meter *fm)
5362 {
5363 	const struct mlx5_flow_driver_ops *fops;
5364 
5365 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5366 	return fops->create_mtr_tbls(dev, fm);
5367 }
5368 
5369 /**
5370  * Destroy the meter table set.
5371  *
5372  * @param[in] dev
5373  *   Pointer to Ethernet device.
5374  * @param[in] tbl
5375  *   Pointer to the meter table set.
5376  *
5377  * @return
5378  *   0 on success.
5379  */
5380 int
5381 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5382 			   struct mlx5_meter_domains_infos *tbls)
5383 {
5384 	const struct mlx5_flow_driver_ops *fops;
5385 
5386 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5387 	return fops->destroy_mtr_tbls(dev, tbls);
5388 }
5389 
5390 /**
5391  * Create policer rules.
5392  *
5393  * @param[in] dev
5394  *   Pointer to Ethernet device.
5395  * @param[in] fm
5396  *   Pointer to flow meter structure.
5397  * @param[in] attr
5398  *   Pointer to flow attributes.
5399  *
5400  * @return
5401  *   0 on success, -1 otherwise.
5402  */
5403 int
5404 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5405 			       struct mlx5_flow_meter *fm,
5406 			       const struct rte_flow_attr *attr)
5407 {
5408 	const struct mlx5_flow_driver_ops *fops;
5409 
5410 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5411 	return fops->create_policer_rules(dev, fm, attr);
5412 }
5413 
5414 /**
5415  * Destroy policer rules.
5416  *
5417  * @param[in] fm
5418  *   Pointer to flow meter structure.
5419  * @param[in] attr
5420  *   Pointer to flow attributes.
5421  *
5422  * @return
5423  *   0 on success, -1 otherwise.
5424  */
5425 int
5426 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5427 				struct mlx5_flow_meter *fm,
5428 				const struct rte_flow_attr *attr)
5429 {
5430 	const struct mlx5_flow_driver_ops *fops;
5431 
5432 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5433 	return fops->destroy_policer_rules(dev, fm, attr);
5434 }
5435 
5436 /**
5437  * Allocate a counter.
5438  *
5439  * @param[in] dev
5440  *   Pointer to Ethernet device structure.
5441  *
5442  * @return
5443  *   Pointer to allocated counter  on success, NULL otherwise.
5444  */
5445 struct mlx5_flow_counter *
5446 mlx5_counter_alloc(struct rte_eth_dev *dev)
5447 {
5448 	const struct mlx5_flow_driver_ops *fops;
5449 	struct rte_flow_attr attr = { .transfer = 0 };
5450 
5451 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5452 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5453 		return fops->counter_alloc(dev);
5454 	}
5455 	DRV_LOG(ERR,
5456 		"port %u counter allocate is not supported.",
5457 		 dev->data->port_id);
5458 	return NULL;
5459 }
5460 
5461 /**
5462  * Free a counter.
5463  *
5464  * @param[in] dev
5465  *   Pointer to Ethernet device structure.
5466  * @param[in] cnt
5467  *   Pointer to counter to be free.
5468  */
5469 void
5470 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5471 {
5472 	const struct mlx5_flow_driver_ops *fops;
5473 	struct rte_flow_attr attr = { .transfer = 0 };
5474 
5475 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5476 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5477 		fops->counter_free(dev, cnt);
5478 		return;
5479 	}
5480 	DRV_LOG(ERR,
5481 		"port %u counter free is not supported.",
5482 		 dev->data->port_id);
5483 }
5484 
5485 /**
5486  * Query counter statistics.
5487  *
5488  * @param[in] dev
5489  *   Pointer to Ethernet device structure.
5490  * @param[in] cnt
5491  *   Pointer to counter to query.
5492  * @param[in] clear
5493  *   Set to clear counter statistics.
5494  * @param[out] pkts
5495  *   The counter hits packets number to save.
5496  * @param[out] bytes
5497  *   The counter hits bytes number to save.
5498  *
5499  * @return
5500  *   0 on success, a negative errno value otherwise.
5501  */
5502 int
5503 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5504 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5505 {
5506 	const struct mlx5_flow_driver_ops *fops;
5507 	struct rte_flow_attr attr = { .transfer = 0 };
5508 
5509 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5510 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5511 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5512 	}
5513 	DRV_LOG(ERR,
5514 		"port %u counter query is not supported.",
5515 		 dev->data->port_id);
5516 	return -ENOTSUP;
5517 }
5518 
5519 #define MLX5_POOL_QUERY_FREQ_US 1000000
5520 
5521 /**
5522  * Set the periodic procedure for triggering asynchronous batch queries for all
5523  * the counter pools.
5524  *
5525  * @param[in] sh
5526  *   Pointer to mlx5_ibv_shared object.
5527  */
5528 void
5529 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5530 {
5531 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5532 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5533 	uint32_t us;
5534 
5535 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5536 	pools_n += rte_atomic16_read(&cont->n_valid);
5537 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5538 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5539 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5540 		sh->cmng.query_thread_on = 0;
5541 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5542 	} else {
5543 		sh->cmng.query_thread_on = 1;
5544 	}
5545 }
5546 
5547 /**
5548  * The periodic procedure for triggering asynchronous batch queries for all the
5549  * counter pools. This function is probably called by the host thread.
5550  *
5551  * @param[in] arg
5552  *   The parameter for the alarm process.
5553  */
5554 void
5555 mlx5_flow_query_alarm(void *arg)
5556 {
5557 	struct mlx5_ibv_shared *sh = arg;
5558 	struct mlx5_devx_obj *dcs;
5559 	uint16_t offset;
5560 	int ret;
5561 	uint8_t batch = sh->cmng.batch;
5562 	uint16_t pool_index = sh->cmng.pool_index;
5563 	struct mlx5_pools_container *cont;
5564 	struct mlx5_pools_container *mcont;
5565 	struct mlx5_flow_counter_pool *pool;
5566 
5567 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5568 		goto set_alarm;
5569 next_container:
5570 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5571 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5572 	/* Check if resize was done and need to flip a container. */
5573 	if (cont != mcont) {
5574 		if (cont->pools) {
5575 			/* Clean the old container. */
5576 			rte_free(cont->pools);
5577 			memset(cont, 0, sizeof(*cont));
5578 		}
5579 		rte_cio_wmb();
5580 		 /* Flip the host container. */
5581 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5582 		cont = mcont;
5583 	}
5584 	if (!cont->pools) {
5585 		/* 2 empty containers case is unexpected. */
5586 		if (unlikely(batch != sh->cmng.batch))
5587 			goto set_alarm;
5588 		batch ^= 0x1;
5589 		pool_index = 0;
5590 		goto next_container;
5591 	}
5592 	pool = cont->pools[pool_index];
5593 	if (pool->raw_hw)
5594 		/* There is a pool query in progress. */
5595 		goto set_alarm;
5596 	pool->raw_hw =
5597 		LIST_FIRST(&sh->cmng.free_stat_raws);
5598 	if (!pool->raw_hw)
5599 		/* No free counter statistics raw memory. */
5600 		goto set_alarm;
5601 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5602 							      (&pool->a64_dcs);
5603 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5604 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5605 					       offset, NULL, NULL,
5606 					       pool->raw_hw->mem_mng->dm->id,
5607 					       (void *)(uintptr_t)
5608 					       (pool->raw_hw->data + offset),
5609 					       sh->devx_comp,
5610 					       (uint64_t)(uintptr_t)pool);
5611 	if (ret) {
5612 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5613 			" %d", pool->min_dcs->id);
5614 		pool->raw_hw = NULL;
5615 		goto set_alarm;
5616 	}
5617 	pool->raw_hw->min_dcs_id = dcs->id;
5618 	LIST_REMOVE(pool->raw_hw, next);
5619 	sh->cmng.pending_queries++;
5620 	pool_index++;
5621 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5622 		batch ^= 0x1;
5623 		pool_index = 0;
5624 	}
5625 set_alarm:
5626 	sh->cmng.batch = batch;
5627 	sh->cmng.pool_index = pool_index;
5628 	mlx5_set_query_alarm(sh);
5629 }
5630 
5631 /**
5632  * Handler for the HW respond about ready values from an asynchronous batch
5633  * query. This function is probably called by the host thread.
5634  *
5635  * @param[in] sh
5636  *   The pointer to the shared IB device context.
5637  * @param[in] async_id
5638  *   The Devx async ID.
5639  * @param[in] status
5640  *   The status of the completion.
5641  */
5642 void
5643 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5644 				  uint64_t async_id, int status)
5645 {
5646 	struct mlx5_flow_counter_pool *pool =
5647 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5648 	struct mlx5_counter_stats_raw *raw_to_free;
5649 
5650 	if (unlikely(status)) {
5651 		raw_to_free = pool->raw_hw;
5652 	} else {
5653 		raw_to_free = pool->raw;
5654 		rte_spinlock_lock(&pool->sl);
5655 		pool->raw = pool->raw_hw;
5656 		rte_spinlock_unlock(&pool->sl);
5657 		rte_atomic64_add(&pool->query_gen, 1);
5658 		/* Be sure the new raw counters data is updated in memory. */
5659 		rte_cio_wmb();
5660 	}
5661 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5662 	pool->raw_hw = NULL;
5663 	sh->cmng.pending_queries--;
5664 }
5665 
5666 /**
5667  * Translate the rte_flow group index to HW table value.
5668  *
5669  * @param[in] attributes
5670  *   Pointer to flow attributes
5671  * @param[in] external
5672  *   Value is part of flow rule created by request external to PMD.
5673  * @param[in] group
5674  *   rte_flow group index value.
5675  * @param[out] fdb_def_rule
5676  *   Whether fdb jump to table 1 is configured.
5677  * @param[out] table
5678  *   HW table value.
5679  * @param[out] error
5680  *   Pointer to error structure.
5681  *
5682  * @return
5683  *   0 on success, a negative errno value otherwise and rte_errno is set.
5684  */
5685 int
5686 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5687 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
5688 			 struct rte_flow_error *error)
5689 {
5690 	if (attributes->transfer && external && fdb_def_rule) {
5691 		if (group == UINT32_MAX)
5692 			return rte_flow_error_set
5693 						(error, EINVAL,
5694 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5695 						 NULL,
5696 						 "group index not supported");
5697 		*table = group + 1;
5698 	} else {
5699 		*table = group;
5700 	}
5701 	return 0;
5702 }
5703 
5704 /**
5705  * Discover availability of metadata reg_c's.
5706  *
5707  * Iteratively use test flows to check availability.
5708  *
5709  * @param[in] dev
5710  *   Pointer to the Ethernet device structure.
5711  *
5712  * @return
5713  *   0 on success, a negative errno value otherwise and rte_errno is set.
5714  */
5715 int
5716 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5717 {
5718 	struct mlx5_priv *priv = dev->data->dev_private;
5719 	struct mlx5_dev_config *config = &priv->config;
5720 	enum modify_reg idx;
5721 	int n = 0;
5722 
5723 	/* reg_c[0] and reg_c[1] are reserved. */
5724 	config->flow_mreg_c[n++] = REG_C_0;
5725 	config->flow_mreg_c[n++] = REG_C_1;
5726 	/* Discover availability of other reg_c's. */
5727 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5728 		struct rte_flow_attr attr = {
5729 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5730 			.priority = MLX5_FLOW_PRIO_RSVD,
5731 			.ingress = 1,
5732 		};
5733 		struct rte_flow_item items[] = {
5734 			[0] = {
5735 				.type = RTE_FLOW_ITEM_TYPE_END,
5736 			},
5737 		};
5738 		struct rte_flow_action actions[] = {
5739 			[0] = {
5740 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5741 				.conf = &(struct mlx5_flow_action_copy_mreg){
5742 					.src = REG_C_1,
5743 					.dst = idx,
5744 				},
5745 			},
5746 			[1] = {
5747 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5748 				.conf = &(struct rte_flow_action_jump){
5749 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5750 				},
5751 			},
5752 			[2] = {
5753 				.type = RTE_FLOW_ACTION_TYPE_END,
5754 			},
5755 		};
5756 		struct rte_flow *flow;
5757 		struct rte_flow_error error;
5758 
5759 		if (!config->dv_flow_en)
5760 			break;
5761 		/* Create internal flow, validation skips copy action. */
5762 		flow = flow_list_create(dev, NULL, &attr, items,
5763 					actions, false, &error);
5764 		if (!flow)
5765 			continue;
5766 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5767 			config->flow_mreg_c[n++] = idx;
5768 		flow_list_destroy(dev, NULL, flow);
5769 	}
5770 	for (; n < MLX5_MREG_C_NUM; ++n)
5771 		config->flow_mreg_c[n] = REG_NONE;
5772 	return 0;
5773 }
5774 
5775 /**
5776  * Dump flow raw hw data to file
5777  *
5778  * @param[in] dev
5779  *    The pointer to Ethernet device.
5780  * @param[in] file
5781  *   A pointer to a file for output.
5782  * @param[out] error
5783  *   Perform verbose error reporting if not NULL. PMDs initialize this
5784  *   structure in case of error only.
5785  * @return
5786  *   0 on success, a nagative value otherwise.
5787  */
5788 int
5789 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5790 		   FILE *file,
5791 		   struct rte_flow_error *error __rte_unused)
5792 {
5793 	struct mlx5_priv *priv = dev->data->dev_private;
5794 	struct mlx5_ibv_shared *sh = priv->sh;
5795 
5796 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5797 				       sh->tx_domain, file);
5798 }
5799